commit.c 33.6 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0+
2
/*
3
 * linux/fs/jbd2/commit.c
4 5 6 7 8 9 10 11 12 13 14
 *
 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
 *
 * Copyright 1998 Red Hat corp --- All Rights Reserved
 *
 * Journal commit routines for the generic filesystem journaling code;
 * part of the ext2fs journaling system.
 */

#include <linux/time.h>
#include <linux/fs.h>
15
#include <linux/jbd2.h>
16 17 18 19
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
20
#include <linux/jiffies.h>
21
#include <linux/crc32.h>
22 23
#include <linux/writeback.h>
#include <linux/backing-dev.h>
24
#include <linux/bio.h>
25
#include <linux/blkdev.h>
26
#include <linux/bitops.h>
27
#include <trace/events/jbd2.h>
28 29

/*
30
 * IO end handler for temporary buffer_heads handling writes to the journal.
31 32 33
 */
static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
{
34 35
	struct buffer_head *orig_bh = bh->b_private;

36 37 38 39 40
	BUFFER_TRACE(bh, "");
	if (uptodate)
		set_buffer_uptodate(bh);
	else
		clear_buffer_uptodate(bh);
41 42
	if (orig_bh) {
		clear_bit_unlock(BH_Shadow, &orig_bh->b_state);
43
		smp_mb__after_atomic();
44 45
		wake_up_bit(&orig_bh->b_state, BH_Shadow);
	}
46 47 48 49
	unlock_buffer(bh);
}

/*
50 51
 * When an ext4 file is truncated, it is possible that some pages are not
 * successfully freed, because they are attached to a committing transaction.
52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77
 * After the transaction commits, these pages are left on the LRU, with no
 * ->mapping, and with attached buffers.  These pages are trivially reclaimable
 * by the VM, but their apparent absence upsets the VM accounting, and it makes
 * the numbers in /proc/meminfo look odd.
 *
 * So here, we have a buffer which has just come off the forget list.  Look to
 * see if we can strip all buffers from the backing page.
 *
 * Called under lock_journal(), and possibly under journal_datalist_lock.  The
 * caller provided us with a ref against the buffer, and we drop that here.
 */
static void release_buffer_page(struct buffer_head *bh)
{
	struct page *page;

	if (buffer_dirty(bh))
		goto nope;
	if (atomic_read(&bh->b_count) != 1)
		goto nope;
	page = bh->b_page;
	if (!page)
		goto nope;
	if (page->mapping)
		goto nope;

	/* OK, it's a truncated page */
N
Nick Piggin 已提交
78
	if (!trylock_page(page))
79 80
		goto nope;

81
	get_page(page);
82 83 84
	__brelse(bh);
	try_to_free_buffers(page);
	unlock_page(page);
85
	put_page(page);
86 87 88 89 90 91
	return;

nope:
	__brelse(bh);
}

92
static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
D
Darrick J. Wong 已提交
93 94 95 96
{
	struct commit_header *h;
	__u32 csum;

97
	if (!jbd2_journal_has_csum_v2or3(j))
D
Darrick J. Wong 已提交
98 99
		return;

100
	h = (struct commit_header *)(bh->b_data);
D
Darrick J. Wong 已提交
101 102 103
	h->h_chksum_type = 0;
	h->h_chksum_size = 0;
	h->h_chksum[0] = 0;
104
	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
D
Darrick J. Wong 已提交
105 106 107
	h->h_chksum[0] = cpu_to_be32(csum);
}

108 109
/*
 * Done it all: now submit the commit record.  We should have
110 111 112 113 114 115
 * cleaned up our previous buffers by now, so if we are in abort
 * mode we can now just skip the rest of the journal write
 * entirely.
 *
 * Returns 1 if the journal needs to be aborted or 0 on success
 */
116 117 118 119
static int journal_submit_commit_record(journal_t *journal,
					transaction_t *commit_transaction,
					struct buffer_head **cbh,
					__u32 crc32_sum)
120
{
121
	struct commit_header *tmp;
122
	struct buffer_head *bh;
123
	int ret;
124
	struct timespec64 now;
125

126 127
	*cbh = NULL;

128 129 130
	if (is_journal_aborted(journal))
		return 0;

131 132
	bh = jbd2_journal_get_descriptor_buffer(commit_transaction,
						JBD2_COMMIT_BLOCK);
133
	if (!bh)
134 135
		return 1;

136
	tmp = (struct commit_header *)bh->b_data;
137
	ktime_get_coarse_real_ts64(&now);
138 139
	tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
	tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
140

141
	if (jbd2_has_feature_checksum(journal)) {
142 143 144
		tmp->h_chksum_type 	= JBD2_CRC32_CHKSUM;
		tmp->h_chksum_size 	= JBD2_CRC32_CHKSUM_SIZE;
		tmp->h_chksum[0] 	= cpu_to_be32(crc32_sum);
145
	}
146
	jbd2_commit_block_csum_set(journal, bh);
147

148
	BUFFER_TRACE(bh, "submit commit block");
149
	lock_buffer(bh);
150
	clear_buffer_dirty(bh);
151 152 153 154
	set_buffer_uptodate(bh);
	bh->b_end_io = journal_end_buffer_io_sync;

	if (journal->j_flags & JBD2_BARRIER &&
155
	    !jbd2_has_feature_async_commit(journal))
156 157
		ret = submit_bh(REQ_OP_WRITE,
			REQ_SYNC | REQ_PREFLUSH | REQ_FUA, bh);
158
	else
159
		ret = submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
160

161 162 163 164 165 166 167 168
	*cbh = bh;
	return ret;
}

/*
 * This function along with journal_submit_commit_record
 * allows to write the commit record asynchronously.
 */
169 170
static int journal_wait_on_commit_record(journal_t *journal,
					 struct buffer_head *bh)
171 172 173 174 175
{
	int ret = 0;

	clear_buffer_dirty(bh);
	wait_on_buffer(bh);
176

177 178 179 180 181
	if (unlikely(!buffer_uptodate(bh)))
		ret = -EIO;
	put_bh(bh);            /* One for getblk() */

	return ret;
182 183
}

184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203
/*
 * write the filemap data using writepage() address_space_operations.
 * We don't do block allocation here even for delalloc. We don't
 * use writepages() because with dealyed allocation we may be doing
 * block allocation in writepages().
 */
static int journal_submit_inode_data_buffers(struct address_space *mapping)
{
	int ret;
	struct writeback_control wbc = {
		.sync_mode =  WB_SYNC_ALL,
		.nr_to_write = mapping->nrpages * 2,
		.range_start = 0,
		.range_end = i_size_read(mapping->host),
	};

	ret = generic_writepages(mapping, &wbc);
	return ret;
}

204 205 206 207 208 209 210 211
/*
 * Submit all the data buffers of inode associated with the transaction to
 * disk.
 *
 * We are in a committing transaction. Therefore no new inode can be added to
 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
 * operate on from being released while we write out pages.
 */
212
static int journal_submit_data_buffers(journal_t *journal,
213 214 215 216 217 218 219 220
		transaction_t *commit_transaction)
{
	struct jbd2_inode *jinode;
	int err, ret = 0;
	struct address_space *mapping;

	spin_lock(&journal->j_list_lock);
	list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
221 222
		if (!(jinode->i_flags & JI_WRITE_DATA))
			continue;
223
		mapping = jinode->i_vfs_inode->i_mapping;
224
		jinode->i_flags |= JI_COMMIT_RUNNING;
225
		spin_unlock(&journal->j_list_lock);
226 227 228 229 230 231
		/*
		 * submit the inode data buffers. We use writepage
		 * instead of writepages. Because writepages can do
		 * block allocation  with delalloc. We need to write
		 * only allocated blocks here.
		 */
232
		trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
233
		err = journal_submit_inode_data_buffers(mapping);
234 235 236 237
		if (!ret)
			ret = err;
		spin_lock(&journal->j_list_lock);
		J_ASSERT(jinode->i_transaction == commit_transaction);
238 239
		jinode->i_flags &= ~JI_COMMIT_RUNNING;
		smp_mb();
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
		wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
	}
	spin_unlock(&journal->j_list_lock);
	return ret;
}

/*
 * Wait for data submitted for writeout, refile inodes to proper
 * transaction if needed.
 *
 */
static int journal_finish_inode_data_buffers(journal_t *journal,
		transaction_t *commit_transaction)
{
	struct jbd2_inode *jinode, *next_i;
	int err, ret = 0;

257
	/* For locking, see the comment in journal_submit_data_buffers() */
258 259
	spin_lock(&journal->j_list_lock);
	list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
260 261
		if (!(jinode->i_flags & JI_WAIT_DATA))
			continue;
262
		jinode->i_flags |= JI_COMMIT_RUNNING;
263
		spin_unlock(&journal->j_list_lock);
264 265 266 267
		err = filemap_fdatawait_keep_errors(
				jinode->i_vfs_inode->i_mapping);
		if (!ret)
			ret = err;
268
		spin_lock(&journal->j_list_lock);
269 270
		jinode->i_flags &= ~JI_COMMIT_RUNNING;
		smp_mb();
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
		wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
	}

	/* Now refile inode to proper lists */
	list_for_each_entry_safe(jinode, next_i,
				 &commit_transaction->t_inode_list, i_list) {
		list_del(&jinode->i_list);
		if (jinode->i_next_transaction) {
			jinode->i_transaction = jinode->i_next_transaction;
			jinode->i_next_transaction = NULL;
			list_add(&jinode->i_list,
				&jinode->i_transaction->t_inode_list);
		} else {
			jinode->i_transaction = NULL;
		}
	}
	spin_unlock(&journal->j_list_lock);

	return ret;
}

292 293 294 295 296 297
static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
{
	struct page *page = bh->b_page;
	char *addr;
	__u32 checksum;

298
	addr = kmap_atomic(page);
299 300
	checksum = crc32_be(crc32_sum,
		(void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
301
	kunmap_atomic(addr);
302 303 304 305

	return checksum;
}

306
static void write_tag_block(journal_t *j, journal_block_tag_t *tag,
307
				   unsigned long long block)
Z
Zach Brown 已提交
308 309
{
	tag->t_blocknr = cpu_to_be32(block & (u32)~0);
310
	if (jbd2_has_feature_64bit(j))
Z
Zach Brown 已提交
311 312 313
		tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
}

314 315 316
static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
				    struct buffer_head *bh, __u32 sequence)
{
317
	journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
318 319
	struct page *page = bh->b_page;
	__u8 *addr;
320
	__u32 csum32;
321
	__be32 seq;
322

323
	if (!jbd2_journal_has_csum_v2or3(j))
324 325
		return;

326
	seq = cpu_to_be32(sequence);
327
	addr = kmap_atomic(page);
328
	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
329 330
	csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
			     bh->b_size);
331
	kunmap_atomic(addr);
332

333
	if (jbd2_has_feature_csum3(j))
334 335 336
		tag3->t_checksum = cpu_to_be32(csum32);
	else
		tag->t_checksum = cpu_to_be16(csum32);
337
}
338
/*
339
 * jbd2_journal_commit_transaction
340 341 342 343
 *
 * The primary function for committing a transaction to the log.  This
 * function is called by the journal thread to begin a complete commit.
 */
344
void jbd2_journal_commit_transaction(journal_t *journal)
345
{
346
	struct transaction_stats_s stats;
347
	transaction_t *commit_transaction;
348 349
	struct journal_head *jh;
	struct buffer_head *descriptor;
350 351 352 353
	struct buffer_head **wbuf = journal->j_wbuf;
	int bufs;
	int flags;
	int err;
354
	unsigned long long blocknr;
J
Josef Bacik 已提交
355 356
	ktime_t start_time;
	u64 commit_time;
357 358 359 360 361
	char *tagp = NULL;
	journal_block_tag_t *tag = NULL;
	int space_left = 0;
	int first_tag = 0;
	int tag_flag;
362
	int i;
Z
Zach Brown 已提交
363
	int tag_bytes = journal_tag_bytes(journal);
364 365
	struct buffer_head *cbh = NULL; /* For transactional checksums */
	__u32 crc32_sum = ~0;
366
	struct blk_plug plug;
367 368 369 370
	/* Tail of the journal */
	unsigned long first_block;
	tid_t first_tid;
	int update_tail;
371
	int csum_size = 0;
372
	LIST_HEAD(io_bufs);
373
	LIST_HEAD(log_bufs);
374

375
	if (jbd2_journal_has_csum_v2or3(journal))
376
		csum_size = sizeof(struct jbd2_journal_block_tail);
377 378 379 380 381 382

	/*
	 * First job: lock down the current transaction and wait for
	 * all outstanding updates to complete.
	 */

383 384
	/* Do we need to erase the effects of a prior jbd2_journal_flush? */
	if (journal->j_flags & JBD2_FLUSHED) {
385
		jbd_debug(3, "super block updated\n");
386
		mutex_lock_io(&journal->j_checkpoint_mutex);
387 388 389 390 391 392 393 394 395
		/*
		 * We hold j_checkpoint_mutex so tail cannot change under us.
		 * We don't need any special data guarantees for writing sb
		 * since journal is empty and it is ok for write to be
		 * flushed only with transaction commit.
		 */
		jbd2_journal_update_sb_log_tail(journal,
						journal->j_tail_sequence,
						journal->j_tail,
396
						REQ_SYNC);
397
		mutex_unlock(&journal->j_checkpoint_mutex);
398 399 400 401 402 403 404 405 406
	} else {
		jbd_debug(3, "superblock not updated\n");
	}

	J_ASSERT(journal->j_running_transaction != NULL);
	J_ASSERT(journal->j_committing_transaction == NULL);

	commit_transaction = journal->j_running_transaction;

407
	trace_jbd2_start_commit(journal, commit_transaction);
E
Eryu Guan 已提交
408
	jbd_debug(1, "JBD2: starting commit of transaction %d\n",
409 410
			commit_transaction->t_tid);

411
	write_lock(&journal->j_state_lock);
412
	J_ASSERT(commit_transaction->t_state == T_RUNNING);
413 414
	commit_transaction->t_state = T_LOCKED;

415
	trace_jbd2_commit_locking(journal, commit_transaction);
416
	stats.run.rs_wait = commit_transaction->t_max_wait;
417
	stats.run.rs_request_delay = 0;
418
	stats.run.rs_locked = jiffies;
419 420 421 422
	if (commit_transaction->t_requested)
		stats.run.rs_request_delay =
			jbd2_time_diff(commit_transaction->t_requested,
				       stats.run.rs_locked);
423 424
	stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
					      stats.run.rs_locked);
425

426
	spin_lock(&commit_transaction->t_handle_lock);
427
	while (atomic_read(&commit_transaction->t_updates)) {
428 429 430 431
		DEFINE_WAIT(wait);

		prepare_to_wait(&journal->j_wait_updates, &wait,
					TASK_UNINTERRUPTIBLE);
432
		if (atomic_read(&commit_transaction->t_updates)) {
433
			spin_unlock(&commit_transaction->t_handle_lock);
434
			write_unlock(&journal->j_state_lock);
435
			schedule();
436
			write_lock(&journal->j_state_lock);
437 438 439 440 441 442
			spin_lock(&commit_transaction->t_handle_lock);
		}
		finish_wait(&journal->j_wait_updates, &wait);
	}
	spin_unlock(&commit_transaction->t_handle_lock);

443
	J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
444 445 446 447 448 449 450
			journal->j_max_transaction_buffers);

	/*
	 * First thing we are allowed to do is to discard any remaining
	 * BJ_Reserved buffers.  Note, it is _not_ permissible to assume
	 * that there are no such buffers: if a large filesystem
	 * operation like a truncate needs to split itself over multiple
451
	 * transactions, then it may try to do a jbd2_journal_restart() while
452 453 454 455 456 457 458
	 * there are still BJ_Reserved buffers outstanding.  These must
	 * be released cleanly from the current transaction.
	 *
	 * In this case, the filesystem must still reserve write access
	 * again before modifying the buffer in the new transaction, but
	 * we do not require it to remember exactly which old buffers it
	 * has reserved.  This is consistent with the existing behaviour
459
	 * that multiple jbd2_journal_get_write_access() calls to the same
L
Lucas De Marchi 已提交
460
	 * buffer are perfectly permissible.
461 462 463 464 465
	 */
	while (commit_transaction->t_reserved_list) {
		jh = commit_transaction->t_reserved_list;
		JBUFFER_TRACE(jh, "reserved, unused: refile");
		/*
466
		 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
467 468 469 470 471 472
		 * leave undo-committed data.
		 */
		if (jh->b_committed_data) {
			struct buffer_head *bh = jh2bh(jh);

			jbd_lock_bh_state(bh);
M
Mingming Cao 已提交
473
			jbd2_free(jh->b_committed_data, bh->b_size);
474 475 476
			jh->b_committed_data = NULL;
			jbd_unlock_bh_state(bh);
		}
477
		jbd2_journal_refile_buffer(journal, jh);
478 479 480 481 482 483 484 485
	}

	/*
	 * Now try to drop any written-back buffers from the journal's
	 * checkpoint lists.  We do this *before* commit because it potentially
	 * frees some memory
	 */
	spin_lock(&journal->j_list_lock);
486
	__jbd2_journal_clean_checkpoint_list(journal, false);
487 488
	spin_unlock(&journal->j_list_lock);

E
Eryu Guan 已提交
489
	jbd_debug(3, "JBD2: commit phase 1\n");
490

491 492 493 494 495 496
	/*
	 * Clear revoked flag to reflect there is no revoked buffers
	 * in the next transaction which is going to be started.
	 */
	jbd2_clear_buffer_revoked_flags(journal);

497 498 499
	/*
	 * Switch to a new revoke table.
	 */
500
	jbd2_journal_switch_revoke_table(journal);
501

J
Jan Kara 已提交
502 503 504 505 506 507
	/*
	 * Reserved credits cannot be claimed anymore, free them
	 */
	atomic_sub(atomic_read(&journal->j_reserved_credits),
		   &commit_transaction->t_outstanding_credits);

508
	trace_jbd2_commit_flushing(journal, commit_transaction);
509 510 511
	stats.run.rs_flushing = jiffies;
	stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
					     stats.run.rs_flushing);
512

513 514 515
	commit_transaction->t_state = T_FLUSH;
	journal->j_committing_transaction = commit_transaction;
	journal->j_running_transaction = NULL;
J
Josef Bacik 已提交
516
	start_time = ktime_get();
517 518
	commit_transaction->t_log_start = journal->j_head;
	wake_up(&journal->j_wait_transaction_locked);
519
	write_unlock(&journal->j_state_lock);
520

521
	jbd_debug(3, "JBD2: commit phase 2a\n");
522 523 524 525 526

	/*
	 * Now start flushing things to disk, in the order they appear
	 * on the transaction lists.  Data blocks go first.
	 */
527
	err = journal_submit_data_buffers(journal, commit_transaction);
528
	if (err)
529
		jbd2_journal_abort(journal, err);
530

531
	blk_start_plug(&plug);
532
	jbd2_journal_write_revoke_records(commit_transaction, &log_bufs);
533

534
	jbd_debug(3, "JBD2: commit phase 2b\n");
535 536 537 538 539 540

	/*
	 * Way to go: we have now written out all of the data for a
	 * transaction!  Now comes the tricky part: we need to write out
	 * metadata.  Loop over the transaction's entire buffer list:
	 */
541
	write_lock(&journal->j_state_lock);
542
	commit_transaction->t_state = T_COMMIT;
543
	write_unlock(&journal->j_state_lock);
544

545
	trace_jbd2_commit_logging(journal, commit_transaction);
546 547 548
	stats.run.rs_logging = jiffies;
	stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
					       stats.run.rs_logging);
549 550
	stats.run.rs_blocks =
		atomic_read(&commit_transaction->t_outstanding_credits);
551
	stats.run.rs_blocks_logged = 0;
552

553
	J_ASSERT(commit_transaction->t_nr_buffers <=
554
		 atomic_read(&commit_transaction->t_outstanding_credits));
555

556
	err = 0;
557
	bufs = 0;
558
	descriptor = NULL;
559 560 561 562 563 564 565
	while (commit_transaction->t_buffers) {

		/* Find the next buffer to be journaled... */

		jh = commit_transaction->t_buffers;

		/* If we're in abort mode, we just un-journal the buffer and
566
		   release it. */
567 568

		if (is_journal_aborted(journal)) {
569
			clear_buffer_jbddirty(jh2bh(jh));
570
			JBUFFER_TRACE(jh, "journal is aborting: refile");
J
Joel Becker 已提交
571 572 573 574
			jbd2_buffer_abort_trigger(jh,
						  jh->b_frozen_data ?
						  jh->b_frozen_triggers :
						  jh->b_triggers);
575
			jbd2_journal_refile_buffer(journal, jh);
576 577 578 579 580 581 582 583 584 585 586 587 588 589 590
			/* If that was the last one, we need to clean up
			 * any descriptor buffers which may have been
			 * already allocated, even if we are now
			 * aborting. */
			if (!commit_transaction->t_buffers)
				goto start_journal_io;
			continue;
		}

		/* Make sure we have a descriptor block in which to
		   record the metadata buffer. */

		if (!descriptor) {
			J_ASSERT (bufs == 0);

E
Eryu Guan 已提交
591
			jbd_debug(4, "JBD2: get descriptor\n");
592

593 594 595
			descriptor = jbd2_journal_get_descriptor_buffer(
							commit_transaction,
							JBD2_DESCRIPTOR_BLOCK);
596
			if (!descriptor) {
597
				jbd2_journal_abort(journal, -EIO);
598 599 600
				continue;
			}

E
Eryu Guan 已提交
601
			jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
602 603 604 605 606
				(unsigned long long)descriptor->b_blocknr,
				descriptor->b_data);
			tagp = &descriptor->b_data[sizeof(journal_header_t)];
			space_left = descriptor->b_size -
						sizeof(journal_header_t);
607
			first_tag = 1;
608 609 610
			set_buffer_jwrite(descriptor);
			set_buffer_dirty(descriptor);
			wbuf[bufs++] = descriptor;
611 612 613

			/* Record it so that we can wait for IO
                           completion later */
614 615
			BUFFER_TRACE(descriptor, "ph3: file as descriptor");
			jbd2_file_log_bh(&log_bufs, descriptor);
616 617 618 619
		}

		/* Where is the buffer to be written? */

620
		err = jbd2_journal_next_log_block(journal, &blocknr);
621 622 623 624
		/* If the block mapping failed, just abandon the buffer
		   and repeat this loop: we'll fall into the
		   refile-on-abort condition above. */
		if (err) {
625
			jbd2_journal_abort(journal, err);
626 627 628 629 630 631
			continue;
		}

		/*
		 * start_this_handle() uses t_outstanding_credits to determine
		 * the free space in the log, but this counter is changed
632
		 * by jbd2_journal_next_log_block() also.
633
		 */
634
		atomic_dec(&commit_transaction->t_outstanding_credits);
635 636 637

		/* Bump b_count to prevent truncate from stumbling over
                   the shadowed buffer!  @@@ This can go if we ever get
638
                   rid of the shadow pairing of buffers. */
639 640 641
		atomic_inc(&jh2bh(jh)->b_count);

		/*
642 643
		 * Make a temporary IO buffer with which to write it out
		 * (this will requeue the metadata buffer to BJ_Shadow).
644
		 */
645
		set_bit(BH_JWrite, &jh2bh(jh)->b_state);
646
		JBUFFER_TRACE(jh, "ph3: write metadata");
647
		flags = jbd2_journal_write_metadata_buffer(commit_transaction,
648
						jh, &wbuf[bufs], blocknr);
649 650 651 652
		if (flags < 0) {
			jbd2_journal_abort(journal, flags);
			continue;
		}
653
		jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
654 655 656 657 658 659

		/* Record the new block's tag in the current descriptor
                   buffer */

		tag_flag = 0;
		if (flags & 1)
660
			tag_flag |= JBD2_FLAG_ESCAPE;
661
		if (!first_tag)
662
			tag_flag |= JBD2_FLAG_SAME_UUID;
663 664

		tag = (journal_block_tag_t *) tagp;
665
		write_tag_block(journal, tag, jh2bh(jh)->b_blocknr);
666
		tag->t_flags = cpu_to_be16(tag_flag);
667
		jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
668
					commit_transaction->t_tid);
Z
Zach Brown 已提交
669 670
		tagp += tag_bytes;
		space_left -= tag_bytes;
671
		bufs++;
672 673 674 675 676 677 678 679 680 681 682 683 684

		if (first_tag) {
			memcpy (tagp, journal->j_uuid, 16);
			tagp += 16;
			space_left -= 16;
			first_tag = 0;
		}

		/* If there's no more to do, or if the descriptor is full,
		   let the IO rip! */

		if (bufs == journal->j_wbufsize ||
		    commit_transaction->t_buffers == NULL ||
685
		    space_left < tag_bytes + 16 + csum_size) {
686

E
Eryu Guan 已提交
687
			jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
688 689 690 691 692

			/* Write an end-of-descriptor marker before
                           submitting the IOs.  "tag" still points to
                           the last tag we set up. */

693
			tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
694

695
			jbd2_descriptor_block_csum_set(journal, descriptor);
696 697 698
start_journal_io:
			for (i = 0; i < bufs; i++) {
				struct buffer_head *bh = wbuf[i];
699 700 701
				/*
				 * Compute checksum.
				 */
702
				if (jbd2_has_feature_checksum(journal)) {
703 704 705 706
					crc32_sum =
					    jbd2_checksum_data(crc32_sum, bh);
				}

707 708 709 710
				lock_buffer(bh);
				clear_buffer_dirty(bh);
				set_buffer_uptodate(bh);
				bh->b_end_io = journal_end_buffer_io_sync;
711
				submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
712 713
			}
			cond_resched();
714
			stats.run.rs_blocks_logged += bufs;
715 716 717 718 719 720 721 722

			/* Force a new descriptor to be generated next
                           time round the loop. */
			descriptor = NULL;
			bufs = 0;
		}
	}

723 724 725 726 727 728 729 730 731 732
	err = journal_finish_inode_data_buffers(journal, commit_transaction);
	if (err) {
		printk(KERN_WARNING
			"JBD2: Detected IO errors while flushing file data "
		       "on %s\n", journal->j_devname);
		if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
			jbd2_journal_abort(journal, err);
		err = 0;
	}

733 734 735 736 737 738 739 740 741 742
	/*
	 * Get current oldest transaction in the log before we issue flush
	 * to the filesystem device. After the flush we can be sure that
	 * blocks of all older transactions are checkpointed to persistent
	 * storage and we will be safe to update journal start in the
	 * superblock with the numbers we get here.
	 */
	update_tail =
		jbd2_journal_get_log_tail(journal, &first_tid, &first_block);

743
	write_lock(&journal->j_state_lock);
744 745 746 747 748 749 750 751 752
	if (update_tail) {
		long freed = first_block - journal->j_tail;

		if (first_block < journal->j_tail)
			freed += journal->j_last - journal->j_first;
		/* Update tail only if we free significant amount of space */
		if (freed < journal->j_maxlen / 4)
			update_tail = 0;
	}
753 754 755
	J_ASSERT(commit_transaction->t_state == T_COMMIT);
	commit_transaction->t_state = T_COMMIT_DFLUSH;
	write_unlock(&journal->j_state_lock);
756

757 758 759 760 761
	/* 
	 * If the journal is not located on the file system device,
	 * then we must flush the file system device before we issue
	 * the commit record
	 */
762
	if (commit_transaction->t_need_data_flush &&
763 764
	    (journal->j_fs_dev != journal->j_dev) &&
	    (journal->j_flags & JBD2_BARRIER))
765
		blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
766

767
	/* Done it all: now write the commit record asynchronously. */
768
	if (jbd2_has_feature_async_commit(journal)) {
769 770 771 772
		err = journal_submit_commit_record(journal, commit_transaction,
						 &cbh, crc32_sum);
		if (err)
			__jbd2_journal_abort_hard(journal);
773
	}
774

775 776
	blk_finish_plug(&plug);

777 778 779 780
	/* Lo and behold: we have just managed to send a transaction to
           the log.  Before we can commit it, wait for the IO so far to
           complete.  Control buffers being written are on the
           transaction's t_log_list queue, and metadata buffers are on
781
           the io_bufs list.
782 783 784 785 786 787

	   Wait for the buffers in reverse order.  That way we are
	   less likely to be woken up until all IOs have completed, and
	   so we incur less scheduling load.
	*/

E
Eryu Guan 已提交
788
	jbd_debug(3, "JBD2: commit phase 3\n");
789

790 791 792 793
	while (!list_empty(&io_bufs)) {
		struct buffer_head *bh = list_entry(io_bufs.prev,
						    struct buffer_head,
						    b_assoc_buffers);
794

795 796
		wait_on_buffer(bh);
		cond_resched();
797 798 799

		if (unlikely(!buffer_uptodate(bh)))
			err = -EIO;
800
		jbd2_unfile_log_bh(bh);
801 802

		/*
803 804
		 * The list contains temporary buffer heads created by
		 * jbd2_journal_write_metadata_buffer().
805 806 807 808 809 810
		 */
		BUFFER_TRACE(bh, "dumping temporary bh");
		__brelse(bh);
		J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
		free_buffer_head(bh);

811
		/* We also have to refile the corresponding shadowed buffer */
812 813
		jh = commit_transaction->t_shadow_list->b_tprev;
		bh = jh2bh(jh);
814
		clear_buffer_jwrite(bh);
815
		J_ASSERT_BH(bh, buffer_jbddirty(bh));
816
		J_ASSERT_BH(bh, !buffer_shadow(bh));
817 818 819 820 821 822

		/* The metadata is now released for reuse, but we need
                   to remember it against this transaction so that when
                   we finally commit, we can do any checkpointing
                   required. */
		JBUFFER_TRACE(jh, "file as BJ_Forget");
823
		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
824 825 826 827 828 829
		JBUFFER_TRACE(jh, "brelse shadowed buffer");
		__brelse(bh);
	}

	J_ASSERT (commit_transaction->t_shadow_list == NULL);

E
Eryu Guan 已提交
830
	jbd_debug(3, "JBD2: commit phase 4\n");
831 832

	/* Here we wait for the revoke record and descriptor record buffers */
833
	while (!list_empty(&log_bufs)) {
834 835
		struct buffer_head *bh;

836 837 838
		bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
		wait_on_buffer(bh);
		cond_resched();
839 840 841 842 843 844

		if (unlikely(!buffer_uptodate(bh)))
			err = -EIO;

		BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
		clear_buffer_jwrite(bh);
845
		jbd2_unfile_log_bh(bh);
846 847 848 849
		__brelse(bh);		/* One for getblk */
		/* AKPM: bforget here */
	}

850 851 852
	if (err)
		jbd2_journal_abort(journal, err);

E
Eryu Guan 已提交
853
	jbd_debug(3, "JBD2: commit phase 5\n");
854 855 856 857
	write_lock(&journal->j_state_lock);
	J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
	commit_transaction->t_state = T_COMMIT_JFLUSH;
	write_unlock(&journal->j_state_lock);
858

859
	if (!jbd2_has_feature_async_commit(journal)) {
860 861 862 863 864
		err = journal_submit_commit_record(journal, commit_transaction,
						&cbh, crc32_sum);
		if (err)
			__jbd2_journal_abort_hard(journal);
	}
865
	if (cbh)
866
		err = journal_wait_on_commit_record(journal, cbh);
867
	if (jbd2_has_feature_async_commit(journal) &&
868
	    journal->j_flags & JBD2_BARRIER) {
869
		blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
870
	}
871 872

	if (err)
873
		jbd2_journal_abort(journal, err);
874

875 876 877 878 879 880 881 882
	/*
	 * Now disk caches for filesystem device are flushed so we are safe to
	 * erase checkpointed transactions from the log by updating journal
	 * superblock.
	 */
	if (update_tail)
		jbd2_update_log_tail(journal, first_tid, first_block);

883 884 885 886 887
	/* End of a transaction!  Finally, we can do checkpoint
           processing: any buffers committed as a result of this
           transaction can be removed from any checkpoint list it was on
           before. */

E
Eryu Guan 已提交
888
	jbd_debug(3, "JBD2: commit phase 6\n");
889

890
	J_ASSERT(list_empty(&commit_transaction->t_inode_list));
891 892 893 894 895 896 897 898 899 900 901 902 903
	J_ASSERT(commit_transaction->t_buffers == NULL);
	J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
	J_ASSERT(commit_transaction->t_shadow_list == NULL);

restart_loop:
	/*
	 * As there are other places (journal_unmap_buffer()) adding buffers
	 * to this list we have to be careful and hold the j_list_lock.
	 */
	spin_lock(&journal->j_list_lock);
	while (commit_transaction->t_forget) {
		transaction_t *cp_transaction;
		struct buffer_head *bh;
904
		int try_to_free = 0;
905 906 907 908

		jh = commit_transaction->t_forget;
		spin_unlock(&journal->j_list_lock);
		bh = jh2bh(jh);
909 910 911 912 913
		/*
		 * Get a reference so that bh cannot be freed before we are
		 * done with it.
		 */
		get_bh(bh);
914
		jbd_lock_bh_state(bh);
915
		J_ASSERT_JH(jh,	jh->b_transaction == commit_transaction);
916 917 918 919 920 921 922 923 924 925

		/*
		 * If there is undo-protected committed data against
		 * this buffer, then we can remove it now.  If it is a
		 * buffer needing such protection, the old frozen_data
		 * field now points to a committed version of the
		 * buffer, so rotate that field to the new committed
		 * data.
		 *
		 * Otherwise, we can just throw away the frozen data now.
J
Joel Becker 已提交
926 927 928
		 *
		 * We also know that the frozen data has already fired
		 * its triggers if they exist, so we can clear that too.
929 930
		 */
		if (jh->b_committed_data) {
M
Mingming Cao 已提交
931
			jbd2_free(jh->b_committed_data, bh->b_size);
932 933 934 935
			jh->b_committed_data = NULL;
			if (jh->b_frozen_data) {
				jh->b_committed_data = jh->b_frozen_data;
				jh->b_frozen_data = NULL;
J
Joel Becker 已提交
936
				jh->b_frozen_triggers = NULL;
937 938
			}
		} else if (jh->b_frozen_data) {
M
Mingming Cao 已提交
939
			jbd2_free(jh->b_frozen_data, bh->b_size);
940
			jh->b_frozen_data = NULL;
J
Joel Becker 已提交
941
			jh->b_frozen_triggers = NULL;
942 943 944 945 946 947
		}

		spin_lock(&journal->j_list_lock);
		cp_transaction = jh->b_cp_transaction;
		if (cp_transaction) {
			JBUFFER_TRACE(jh, "remove from old cp transaction");
948
			cp_transaction->t_chp_stats.cs_dropped++;
949
			__jbd2_journal_remove_checkpoint(jh);
950 951 952 953
		}

		/* Only re-checkpoint the buffer_head if it is marked
		 * dirty.  If the buffer was added to the BJ_Forget list
954
		 * by jbd2_journal_forget, it may no longer be dirty and
955 956 957
		 * there's no point in keeping a checkpoint record for
		 * it. */

958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986
		/*
		* A buffer which has been freed while still being journaled by
		* a previous transaction.
		*/
		if (buffer_freed(bh)) {
			/*
			 * If the running transaction is the one containing
			 * "add to orphan" operation (b_next_transaction !=
			 * NULL), we have to wait for that transaction to
			 * commit before we can really get rid of the buffer.
			 * So just clear b_modified to not confuse transaction
			 * credit accounting and refile the buffer to
			 * BJ_Forget of the running transaction. If the just
			 * committed transaction contains "add to orphan"
			 * operation, we can completely invalidate the buffer
			 * now. We are rather through in that since the
			 * buffer may be still accessible when blocksize <
			 * pagesize and it is attached to the last partial
			 * page.
			 */
			jh->b_modified = 0;
			if (!jh->b_next_transaction) {
				clear_buffer_freed(bh);
				clear_buffer_jbddirty(bh);
				clear_buffer_mapped(bh);
				clear_buffer_new(bh);
				clear_buffer_req(bh);
				bh->b_bdev = NULL;
			}
987 988 989 990
		}

		if (buffer_jbddirty(bh)) {
			JBUFFER_TRACE(jh, "add to new checkpointing trans");
991
			__jbd2_journal_insert_checkpoint(jh, commit_transaction);
992 993
			if (is_journal_aborted(journal))
				clear_buffer_jbddirty(bh);
994 995
		} else {
			J_ASSERT_BH(bh, !buffer_dirty(bh));
996 997
			/*
			 * The buffer on BJ_Forget list and not jbddirty means
998 999 1000 1001 1002
			 * it has been freed by this transaction and hence it
			 * could not have been reallocated until this
			 * transaction has committed. *BUT* it could be
			 * reallocated once we have written all the data to
			 * disk and before we process the buffer on BJ_Forget
1003 1004 1005 1006
			 * list.
			 */
			if (!jh->b_next_transaction)
				try_to_free = 1;
1007
		}
1008 1009 1010 1011 1012 1013 1014
		JBUFFER_TRACE(jh, "refile or unfile buffer");
		__jbd2_journal_refile_buffer(jh);
		jbd_unlock_bh_state(bh);
		if (try_to_free)
			release_buffer_page(bh);	/* Drops bh reference */
		else
			__brelse(bh);
1015 1016 1017 1018
		cond_resched_lock(&journal->j_list_lock);
	}
	spin_unlock(&journal->j_list_lock);
	/*
1019 1020 1021 1022
	 * This is a bit sleazy.  We use j_list_lock to protect transition
	 * of a transaction into T_FINISHED state and calling
	 * __jbd2_journal_drop_transaction(). Otherwise we could race with
	 * other checkpointing code processing the transaction...
1023
	 */
1024
	write_lock(&journal->j_state_lock);
1025 1026 1027 1028 1029 1030 1031
	spin_lock(&journal->j_list_lock);
	/*
	 * Now recheck if some buffers did not get attached to the transaction
	 * while the lock was dropped...
	 */
	if (commit_transaction->t_forget) {
		spin_unlock(&journal->j_list_lock);
1032
		write_unlock(&journal->j_state_lock);
1033 1034 1035
		goto restart_loop;
	}

1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
	/* Add the transaction to the checkpoint list
	 * __journal_remove_checkpoint() can not destroy transaction
	 * under us because it is not marked as T_FINISHED yet */
	if (journal->j_checkpoint_transactions == NULL) {
		journal->j_checkpoint_transactions = commit_transaction;
		commit_transaction->t_cpnext = commit_transaction;
		commit_transaction->t_cpprev = commit_transaction;
	} else {
		commit_transaction->t_cpnext =
			journal->j_checkpoint_transactions;
		commit_transaction->t_cpprev =
			commit_transaction->t_cpnext->t_cpprev;
		commit_transaction->t_cpnext->t_cpprev =
			commit_transaction;
		commit_transaction->t_cpprev->t_cpnext =
				commit_transaction;
	}
	spin_unlock(&journal->j_list_lock);

1055 1056
	/* Done with this transaction! */

E
Eryu Guan 已提交
1057
	jbd_debug(3, "JBD2: commit phase 7\n");
1058

1059
	J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1060

1061
	commit_transaction->t_start = jiffies;
1062 1063
	stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
					      commit_transaction->t_start);
1064 1065

	/*
1066
	 * File the transaction statistics
1067 1068
	 */
	stats.ts_tid = commit_transaction->t_tid;
1069 1070
	stats.run.rs_handle_count =
		atomic_read(&commit_transaction->t_handle_count);
1071 1072
	trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
			     commit_transaction->t_tid, &stats.run);
1073
	stats.ts_requested = (commit_transaction->t_requested) ? 1 : 0;
1074

1075
	commit_transaction->t_state = T_COMMIT_CALLBACK;
1076 1077 1078
	J_ASSERT(commit_transaction == journal->j_committing_transaction);
	journal->j_commit_sequence = commit_transaction->t_tid;
	journal->j_committing_transaction = NULL;
J
Josef Bacik 已提交
1079
	commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1080

J
Josef Bacik 已提交
1081 1082 1083 1084 1085 1086 1087 1088 1089
	/*
	 * weight the commit time higher than the average time so we don't
	 * react too strongly to vast changes in the commit time
	 */
	if (likely(journal->j_average_commit_time))
		journal->j_average_commit_time = (commit_time +
				journal->j_average_commit_time*3) / 4;
	else
		journal->j_average_commit_time = commit_time;
1090

1091
	write_unlock(&journal->j_state_lock);
1092

1093 1094 1095
	if (journal->j_commit_callback)
		journal->j_commit_callback(journal, commit_transaction);

1096
	trace_jbd2_end_commit(journal, commit_transaction);
E
Eryu Guan 已提交
1097
	jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1098 1099
		  journal->j_commit_sequence, journal->j_tail_sequence);

1100 1101 1102
	write_lock(&journal->j_state_lock);
	spin_lock(&journal->j_list_lock);
	commit_transaction->t_state = T_FINISHED;
1103
	/* Check if the transaction can be dropped now that we are finished */
1104 1105 1106 1107 1108 1109 1110
	if (commit_transaction->t_checkpoint_list == NULL &&
	    commit_transaction->t_checkpoint_io_list == NULL) {
		__jbd2_journal_drop_transaction(journal, commit_transaction);
		jbd2_journal_free_transaction(commit_transaction);
	}
	spin_unlock(&journal->j_list_lock);
	write_unlock(&journal->j_state_lock);
1111
	wake_up(&journal->j_wait_done_commit);
1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128

	/*
	 * Calculate overall stats
	 */
	spin_lock(&journal->j_history_lock);
	journal->j_stats.ts_tid++;
	journal->j_stats.ts_requested += stats.ts_requested;
	journal->j_stats.run.rs_wait += stats.run.rs_wait;
	journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
	journal->j_stats.run.rs_running += stats.run.rs_running;
	journal->j_stats.run.rs_locked += stats.run.rs_locked;
	journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
	journal->j_stats.run.rs_logging += stats.run.rs_logging;
	journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
	journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
	journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
	spin_unlock(&journal->j_history_lock);
1129
}