commit.c 34.8 KB
Newer Older
1
/*
2
 * linux/fs/jbd2/commit.c
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
 *
 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
 *
 * Copyright 1998 Red Hat corp --- All Rights Reserved
 *
 * This file is part of the Linux kernel and is made available under
 * the terms of the GNU General Public License, version 2, or at your
 * option, any later version, incorporated herein by reference.
 *
 * Journal commit routines for the generic filesystem journaling code;
 * part of the ext2fs journaling system.
 */

#include <linux/time.h>
#include <linux/fs.h>
18
#include <linux/jbd2.h>
19 20 21 22
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
23
#include <linux/jiffies.h>
24
#include <linux/crc32.h>
25 26
#include <linux/writeback.h>
#include <linux/backing-dev.h>
27
#include <linux/bio.h>
28
#include <linux/blkdev.h>
29
#include <linux/bitops.h>
30
#include <trace/events/jbd2.h>
31 32 33 34 35 36 37 38 39 40 41 42 43 44 45

/*
 * Default IO end handler for temporary BJ_IO buffer_heads.
 */
static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
{
	BUFFER_TRACE(bh, "");
	if (uptodate)
		set_buffer_uptodate(bh);
	else
		clear_buffer_uptodate(bh);
	unlock_buffer(bh);
}

/*
46 47
 * When an ext4 file is truncated, it is possible that some pages are not
 * successfully freed, because they are attached to a committing transaction.
48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
 * 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
74
	if (!trylock_page(page))
75 76 77 78 79 80 81 82 83 84 85 86 87
		goto nope;

	page_cache_get(page);
	__brelse(bh);
	try_to_free_buffers(page);
	unlock_page(page);
	page_cache_release(page);
	return;

nope:
	__brelse(bh);
}

88
static void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
D
Darrick J. Wong 已提交
89 90 91 92 93 94 95
{
	struct commit_header *h;
	__u32 csum;

	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
		return;

96
	h = (struct commit_header *)(bh->b_data);
D
Darrick J. Wong 已提交
97 98 99
	h->h_chksum_type = 0;
	h->h_chksum_size = 0;
	h->h_chksum[0] = 0;
100
	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
D
Darrick J. Wong 已提交
101 102 103
	h->h_chksum[0] = cpu_to_be32(csum);
}

104 105
/*
 * Done it all: now submit the commit record.  We should have
106 107 108 109 110 111
 * 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
 */
112 113 114 115
static int journal_submit_commit_record(journal_t *journal,
					transaction_t *commit_transaction,
					struct buffer_head **cbh,
					__u32 crc32_sum)
116
{
117
	struct commit_header *tmp;
118
	struct buffer_head *bh;
119
	int ret;
120
	struct timespec now = current_kernel_time();
121

122 123
	*cbh = NULL;

124 125 126
	if (is_journal_aborted(journal))
		return 0;

127 128
	bh = jbd2_journal_get_descriptor_buffer(journal);
	if (!bh)
129 130
		return 1;

131 132 133 134
	tmp = (struct commit_header *)bh->b_data;
	tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
	tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
	tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
135 136
	tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
	tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
137 138 139 140 141 142

	if (JBD2_HAS_COMPAT_FEATURE(journal,
				    JBD2_FEATURE_COMPAT_CHECKSUM)) {
		tmp->h_chksum_type 	= JBD2_CRC32_CHKSUM;
		tmp->h_chksum_size 	= JBD2_CRC32_CHKSUM_SIZE;
		tmp->h_chksum[0] 	= cpu_to_be32(crc32_sum);
143
	}
144
	jbd2_commit_block_csum_set(journal, bh);
145

146
	BUFFER_TRACE(bh, "submit commit block");
147
	lock_buffer(bh);
148
	clear_buffer_dirty(bh);
149 150 151 152
	set_buffer_uptodate(bh);
	bh->b_end_io = journal_end_buffer_io_sync;

	if (journal->j_flags & JBD2_BARRIER &&
153
	    !JBD2_HAS_INCOMPAT_FEATURE(journal,
154
				       JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
J
Jens Axboe 已提交
155
		ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh);
156
	else
J
Jens Axboe 已提交
157
		ret = submit_bh(WRITE_SYNC, bh);
158

159 160 161 162 163 164 165 166
	*cbh = bh;
	return ret;
}

/*
 * This function along with journal_submit_commit_record
 * allows to write the commit record asynchronously.
 */
167 168
static int journal_wait_on_commit_record(journal_t *journal,
					 struct buffer_head *bh)
169 170 171 172 173
{
	int ret = 0;

	clear_buffer_dirty(bh);
	wait_on_buffer(bh);
174

175 176 177 178 179
	if (unlikely(!buffer_uptodate(bh)))
		ret = -EIO;
	put_bh(bh);            /* One for getblk() */

	return ret;
180 181
}

182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
/*
 * 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;
}

202 203 204 205 206 207 208 209
/*
 * 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.
 */
210
static int journal_submit_data_buffers(journal_t *journal,
211 212 213 214 215 216 217 218 219
		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) {
		mapping = jinode->i_vfs_inode->i_mapping;
220
		set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
221
		spin_unlock(&journal->j_list_lock);
222 223 224 225 226 227
		/*
		 * 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.
		 */
228
		trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
229
		err = journal_submit_inode_data_buffers(mapping);
230 231 232 233
		if (!ret)
			ret = err;
		spin_lock(&journal->j_list_lock);
		J_ASSERT(jinode->i_transaction == commit_transaction);
234 235
		clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
		smp_mb__after_clear_bit();
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
		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;

253
	/* For locking, see the comment in journal_submit_data_buffers() */
254 255
	spin_lock(&journal->j_list_lock);
	list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
256
		set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
257 258
		spin_unlock(&journal->j_list_lock);
		err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
259 260 261
		if (err) {
			/*
			 * Because AS_EIO is cleared by
262
			 * filemap_fdatawait_range(), set it again so
263 264 265 266 267 268 269 270
			 * that user process can get -EIO from fsync().
			 */
			set_bit(AS_EIO,
				&jinode->i_vfs_inode->i_mapping->flags);

			if (!ret)
				ret = err;
		}
271
		spin_lock(&journal->j_list_lock);
272 273
		clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
		smp_mb__after_clear_bit();
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
		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;
}

295 296 297 298 299 300
static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
{
	struct page *page = bh->b_page;
	char *addr;
	__u32 checksum;

301
	addr = kmap_atomic(page);
302 303
	checksum = crc32_be(crc32_sum,
		(void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
304
	kunmap_atomic(addr);
305 306 307 308 309

	return checksum;
}

static void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
310
				   unsigned long long block)
Z
Zach Brown 已提交
311 312
{
	tag->t_blocknr = cpu_to_be32(block & (u32)~0);
313
	if (tag_bytes > JBD2_TAG_SIZE32)
Z
Zach Brown 已提交
314 315 316
		tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
}

317
static void jbd2_descr_block_csum_set(journal_t *j,
318
				      struct buffer_head *bh)
319 320 321 322 323 324 325
{
	struct jbd2_journal_block_tail *tail;
	__u32 csum;

	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
		return;

326
	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
327 328
			sizeof(struct jbd2_journal_block_tail));
	tail->t_checksum = 0;
329
	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
330 331 332
	tail->t_checksum = cpu_to_be32(csum);
}

333 334 335 336 337
static void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
				    struct buffer_head *bh, __u32 sequence)
{
	struct page *page = bh->b_page;
	__u8 *addr;
338
	__u32 csum32;
339 340 341 342 343

	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
		return;

	sequence = cpu_to_be32(sequence);
344
	addr = kmap_atomic(page);
345 346 347 348
	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence,
			     sizeof(sequence));
	csum32 = jbd2_chksum(j, csum32, addr + offset_in_page(bh->b_data),
			     bh->b_size);
349
	kunmap_atomic(addr);
350

351 352
	/* We only have space to store the lower 16 bits of the crc32c. */
	tag->t_checksum = cpu_to_be16(csum32);
353
}
354
/*
355
 * jbd2_journal_commit_transaction
356 357 358 359
 *
 * The primary function for committing a transaction to the log.  This
 * function is called by the journal thread to begin a complete commit.
 */
360
void jbd2_journal_commit_transaction(journal_t *journal)
361
{
362
	struct transaction_stats_s stats;
363
	transaction_t *commit_transaction;
364 365
	struct journal_head *jh;
	struct buffer_head *descriptor;
366 367 368 369
	struct buffer_head **wbuf = journal->j_wbuf;
	int bufs;
	int flags;
	int err;
370
	unsigned long long blocknr;
J
Josef Bacik 已提交
371 372
	ktime_t start_time;
	u64 commit_time;
373 374 375 376 377 378
	char *tagp = NULL;
	journal_header_t *header;
	journal_block_tag_t *tag = NULL;
	int space_left = 0;
	int first_tag = 0;
	int tag_flag;
379
	int i;
Z
Zach Brown 已提交
380
	int tag_bytes = journal_tag_bytes(journal);
381 382
	struct buffer_head *cbh = NULL; /* For transactional checksums */
	__u32 crc32_sum = ~0;
383
	struct blk_plug plug;
384 385 386 387
	/* Tail of the journal */
	unsigned long first_block;
	tid_t first_tid;
	int update_tail;
388
	int csum_size = 0;
389
	LIST_HEAD(io_bufs);
390
	LIST_HEAD(log_bufs);
391 392 393

	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
		csum_size = sizeof(struct jbd2_journal_block_tail);
394 395 396 397 398 399

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

400 401
	/* Do we need to erase the effects of a prior jbd2_journal_flush? */
	if (journal->j_flags & JBD2_FLUSHED) {
402
		jbd_debug(3, "super block updated\n");
403
		mutex_lock(&journal->j_checkpoint_mutex);
404 405 406 407 408 409 410 411 412 413
		/*
		 * 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,
						WRITE_SYNC);
414
		mutex_unlock(&journal->j_checkpoint_mutex);
415 416 417 418 419 420 421 422 423 424
	} 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;
	J_ASSERT(commit_transaction->t_state == T_RUNNING);

425
	trace_jbd2_start_commit(journal, commit_transaction);
E
Eryu Guan 已提交
426
	jbd_debug(1, "JBD2: starting commit of transaction %d\n",
427 428
			commit_transaction->t_tid);

429
	write_lock(&journal->j_state_lock);
430 431
	commit_transaction->t_state = T_LOCKED;

432
	trace_jbd2_commit_locking(journal, commit_transaction);
433
	stats.run.rs_wait = commit_transaction->t_max_wait;
434
	stats.run.rs_request_delay = 0;
435
	stats.run.rs_locked = jiffies;
436 437 438 439
	if (commit_transaction->t_requested)
		stats.run.rs_request_delay =
			jbd2_time_diff(commit_transaction->t_requested,
				       stats.run.rs_locked);
440 441
	stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
					      stats.run.rs_locked);
442

443
	spin_lock(&commit_transaction->t_handle_lock);
444
	while (atomic_read(&commit_transaction->t_updates)) {
445 446 447 448
		DEFINE_WAIT(wait);

		prepare_to_wait(&journal->j_wait_updates, &wait,
					TASK_UNINTERRUPTIBLE);
449
		if (atomic_read(&commit_transaction->t_updates)) {
450
			spin_unlock(&commit_transaction->t_handle_lock);
451
			write_unlock(&journal->j_state_lock);
452
			schedule();
453
			write_lock(&journal->j_state_lock);
454 455 456 457 458 459
			spin_lock(&commit_transaction->t_handle_lock);
		}
		finish_wait(&journal->j_wait_updates, &wait);
	}
	spin_unlock(&commit_transaction->t_handle_lock);

460
	J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
461 462 463 464 465 466 467
			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
468
	 * transactions, then it may try to do a jbd2_journal_restart() while
469 470 471 472 473 474 475
	 * 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
476
	 * that multiple jbd2_journal_get_write_access() calls to the same
L
Lucas De Marchi 已提交
477
	 * buffer are perfectly permissible.
478 479 480 481 482
	 */
	while (commit_transaction->t_reserved_list) {
		jh = commit_transaction->t_reserved_list;
		JBUFFER_TRACE(jh, "reserved, unused: refile");
		/*
483
		 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
484 485 486 487 488 489
		 * 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
490
			jbd2_free(jh->b_committed_data, bh->b_size);
491 492 493
			jh->b_committed_data = NULL;
			jbd_unlock_bh_state(bh);
		}
494
		jbd2_journal_refile_buffer(journal, jh);
495 496 497 498 499 500 501 502
	}

	/*
	 * 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);
503
	__jbd2_journal_clean_checkpoint_list(journal);
504 505
	spin_unlock(&journal->j_list_lock);

E
Eryu Guan 已提交
506
	jbd_debug(3, "JBD2: commit phase 1\n");
507

508 509 510 511 512 513
	/*
	 * 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);

514 515 516
	/*
	 * Switch to a new revoke table.
	 */
517
	jbd2_journal_switch_revoke_table(journal);
518

519
	trace_jbd2_commit_flushing(journal, commit_transaction);
520 521 522
	stats.run.rs_flushing = jiffies;
	stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
					     stats.run.rs_flushing);
523

524 525 526
	commit_transaction->t_state = T_FLUSH;
	journal->j_committing_transaction = commit_transaction;
	journal->j_running_transaction = NULL;
J
Josef Bacik 已提交
527
	start_time = ktime_get();
528 529
	commit_transaction->t_log_start = journal->j_head;
	wake_up(&journal->j_wait_transaction_locked);
530
	write_unlock(&journal->j_state_lock);
531

E
Eryu Guan 已提交
532
	jbd_debug(3, "JBD2: commit phase 2\n");
533 534 535 536 537

	/*
	 * Now start flushing things to disk, in the order they appear
	 * on the transaction lists.  Data blocks go first.
	 */
538
	err = journal_submit_data_buffers(journal, commit_transaction);
539
	if (err)
540
		jbd2_journal_abort(journal, err);
541

542
	blk_start_plug(&plug);
543
	jbd2_journal_write_revoke_records(journal, commit_transaction,
544
					  &log_bufs, WRITE_SYNC);
545
	blk_finish_plug(&plug);
546

E
Eryu Guan 已提交
547
	jbd_debug(3, "JBD2: commit phase 2\n");
548 549 550 551 552 553

	/*
	 * 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:
	 */
554
	write_lock(&journal->j_state_lock);
555
	commit_transaction->t_state = T_COMMIT;
556
	write_unlock(&journal->j_state_lock);
557

558
	trace_jbd2_commit_logging(journal, commit_transaction);
559 560 561
	stats.run.rs_logging = jiffies;
	stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
					       stats.run.rs_logging);
562 563
	stats.run.rs_blocks =
		atomic_read(&commit_transaction->t_outstanding_credits);
564
	stats.run.rs_blocks_logged = 0;
565

566
	J_ASSERT(commit_transaction->t_nr_buffers <=
567
		 atomic_read(&commit_transaction->t_outstanding_credits));
568

569
	err = 0;
570
	bufs = 0;
571
	descriptor = NULL;
572
	blk_start_plug(&plug);
573 574 575 576 577 578 579
	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
580
		   release it. */
581 582

		if (is_journal_aborted(journal)) {
583
			clear_buffer_jbddirty(jh2bh(jh));
584
			JBUFFER_TRACE(jh, "journal is aborting: refile");
J
Joel Becker 已提交
585 586 587 588
			jbd2_buffer_abort_trigger(jh,
						  jh->b_frozen_data ?
						  jh->b_frozen_triggers :
						  jh->b_triggers);
589
			jbd2_journal_refile_buffer(journal, jh);
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
			/* 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
605
			jbd_debug(4, "JBD2: get descriptor\n");
606

607
			descriptor = jbd2_journal_get_descriptor_buffer(journal);
608
			if (!descriptor) {
609
				jbd2_journal_abort(journal, -EIO);
610 611 612
				continue;
			}

E
Eryu Guan 已提交
613
			jbd_debug(4, "JBD2: got buffer %llu (%p)\n",
614 615 616
				(unsigned long long)descriptor->b_blocknr,
				descriptor->b_data);
			header = (journal_header_t *)descriptor->b_data;
617 618
			header->h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
			header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK);
619 620
			header->h_sequence  = cpu_to_be32(commit_transaction->t_tid);

621 622 623
			tagp = &descriptor->b_data[sizeof(journal_header_t)];
			space_left = descriptor->b_size -
						sizeof(journal_header_t);
624
			first_tag = 1;
625 626 627
			set_buffer_jwrite(descriptor);
			set_buffer_dirty(descriptor);
			wbuf[bufs++] = descriptor;
628 629 630

			/* Record it so that we can wait for IO
                           completion later */
631 632
			BUFFER_TRACE(descriptor, "ph3: file as descriptor");
			jbd2_file_log_bh(&log_bufs, descriptor);
633 634 635 636
		}

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

637
		err = jbd2_journal_next_log_block(journal, &blocknr);
638 639 640 641
		/* 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) {
642
			jbd2_journal_abort(journal, err);
643 644 645 646 647 648
			continue;
		}

		/*
		 * start_this_handle() uses t_outstanding_credits to determine
		 * the free space in the log, but this counter is changed
649
		 * by jbd2_journal_next_log_block() also.
650
		 */
651
		atomic_dec(&commit_transaction->t_outstanding_credits);
652 653 654

		/* Bump b_count to prevent truncate from stumbling over
                   the shadowed buffer!  @@@ This can go if we ever get
655
                   rid of the shadow pairing of buffers. */
656 657 658
		atomic_inc(&jh2bh(jh)->b_count);

		/*
659 660
		 * Make a temporary IO buffer with which to write it out
		 * (this will requeue the metadata buffer to BJ_Shadow).
661
		 */
662
		set_bit(BH_JWrite, &jh2bh(jh)->b_state);
663
		JBUFFER_TRACE(jh, "ph3: write metadata");
664
		flags = jbd2_journal_write_metadata_buffer(commit_transaction,
665
						jh, &wbuf[bufs], blocknr);
666 667 668 669
		if (flags < 0) {
			jbd2_journal_abort(journal, flags);
			continue;
		}
670
		jbd2_file_log_bh(&io_bufs, wbuf[bufs]);
671 672 673 674 675 676

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

		tag_flag = 0;
		if (flags & 1)
677
			tag_flag |= JBD2_FLAG_ESCAPE;
678
		if (!first_tag)
679
			tag_flag |= JBD2_FLAG_SAME_UUID;
680 681

		tag = (journal_block_tag_t *) tagp;
Z
Zach Brown 已提交
682
		write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr);
683
		tag->t_flags = cpu_to_be16(tag_flag);
684
		jbd2_block_tag_csum_set(journal, tag, wbuf[bufs],
685
					commit_transaction->t_tid);
Z
Zach Brown 已提交
686 687
		tagp += tag_bytes;
		space_left -= tag_bytes;
688
		bufs++;
689 690 691 692 693 694 695 696 697 698 699 700 701

		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 ||
702
		    space_left < tag_bytes + 16 + csum_size) {
703

E
Eryu Guan 已提交
704
			jbd_debug(4, "JBD2: Submit %d IOs\n", bufs);
705 706 707 708 709

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

710
			tag->t_flags |= cpu_to_be16(JBD2_FLAG_LAST_TAG);
711

712
			jbd2_descr_block_csum_set(journal, descriptor);
713 714 715
start_journal_io:
			for (i = 0; i < bufs; i++) {
				struct buffer_head *bh = wbuf[i];
716 717 718 719 720 721 722 723 724
				/*
				 * Compute checksum.
				 */
				if (JBD2_HAS_COMPAT_FEATURE(journal,
					JBD2_FEATURE_COMPAT_CHECKSUM)) {
					crc32_sum =
					    jbd2_checksum_data(crc32_sum, bh);
				}

725 726 727 728
				lock_buffer(bh);
				clear_buffer_dirty(bh);
				set_buffer_uptodate(bh);
				bh->b_end_io = journal_end_buffer_io_sync;
729
				submit_bh(WRITE_SYNC, bh);
730 731
			}
			cond_resched();
732
			stats.run.rs_blocks_logged += bufs;
733 734 735 736 737 738 739 740

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

741 742 743 744 745 746 747 748 749 750
	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;
	}

751 752 753 754 755 756 757 758 759 760
	/*
	 * 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);

761
	write_lock(&journal->j_state_lock);
762 763 764 765 766 767 768 769 770
	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;
	}
771 772 773
	J_ASSERT(commit_transaction->t_state == T_COMMIT);
	commit_transaction->t_state = T_COMMIT_DFLUSH;
	write_unlock(&journal->j_state_lock);
774

775 776 777 778 779
	/* 
	 * 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
	 */
780
	if (commit_transaction->t_need_data_flush &&
781 782
	    (journal->j_fs_dev != journal->j_dev) &&
	    (journal->j_flags & JBD2_BARRIER))
783
		blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
784

785
	/* Done it all: now write the commit record asynchronously. */
786
	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
787
				      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
788 789 790 791
		err = journal_submit_commit_record(journal, commit_transaction,
						 &cbh, crc32_sum);
		if (err)
			__jbd2_journal_abort_hard(journal);
792
	}
793

794 795
	blk_finish_plug(&plug);

796 797 798 799
	/* 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
800
           the io_bufs list.
801 802 803 804 805 806

	   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
807
	jbd_debug(3, "JBD2: commit phase 3\n");
808

809 810 811 812
	while (!list_empty(&io_bufs)) {
		struct buffer_head *bh = list_entry(io_bufs.prev,
						    struct buffer_head,
						    b_assoc_buffers);
813

814 815
		wait_on_buffer(bh);
		cond_resched();
816 817 818

		if (unlikely(!buffer_uptodate(bh)))
			err = -EIO;
819
		jbd2_unfile_log_bh(bh);
820 821

		/*
822 823
		 * The list contains temporary buffer heads created by
		 * jbd2_journal_write_metadata_buffer().
824 825 826 827 828 829
		 */
		BUFFER_TRACE(bh, "dumping temporary bh");
		__brelse(bh);
		J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
		free_buffer_head(bh);

830
		/* We also have to refile the corresponding shadowed buffer */
831 832
		jh = commit_transaction->t_shadow_list->b_tprev;
		bh = jh2bh(jh);
833
		clear_buffer_jwrite(bh);
834 835 836 837 838 839 840
		J_ASSERT_BH(bh, buffer_jbddirty(bh));

		/* 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");
841
		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
842 843 844 845 846 847 848
		/*
		 * Wake up any transactions which were waiting for this IO to
		 * complete. The barrier must be here so that changes by
		 * jbd2_journal_file_buffer() take effect before wake_up_bit()
		 * does the waitqueue check.
		 */
		smp_mb();
849 850 851 852 853 854 855
		wake_up_bit(&bh->b_state, BH_Unshadow);
		JBUFFER_TRACE(jh, "brelse shadowed buffer");
		__brelse(bh);
	}

	J_ASSERT (commit_transaction->t_shadow_list == NULL);

E
Eryu Guan 已提交
856
	jbd_debug(3, "JBD2: commit phase 4\n");
857 858

	/* Here we wait for the revoke record and descriptor record buffers */
859
	while (!list_empty(&log_bufs)) {
860 861
		struct buffer_head *bh;

862 863 864
		bh = list_entry(log_bufs.prev, struct buffer_head, b_assoc_buffers);
		wait_on_buffer(bh);
		cond_resched();
865 866 867 868 869 870

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

		BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
		clear_buffer_jwrite(bh);
871
		jbd2_unfile_log_bh(bh);
872 873 874 875
		__brelse(bh);		/* One for getblk */
		/* AKPM: bforget here */
	}

876 877 878
	if (err)
		jbd2_journal_abort(journal, err);

E
Eryu Guan 已提交
879
	jbd_debug(3, "JBD2: commit phase 5\n");
880 881 882 883
	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);
884

885
	if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
886
				       JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
887 888 889 890 891
		err = journal_submit_commit_record(journal, commit_transaction,
						&cbh, crc32_sum);
		if (err)
			__jbd2_journal_abort_hard(journal);
	}
892
	if (cbh)
893
		err = journal_wait_on_commit_record(journal, cbh);
894 895 896
	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
				      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
	    journal->j_flags & JBD2_BARRIER) {
897
		blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
898
	}
899 900

	if (err)
901
		jbd2_journal_abort(journal, err);
902

903 904 905 906 907 908 909 910
	/*
	 * 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);

911 912 913 914 915
	/* 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
916
	jbd_debug(3, "JBD2: commit phase 6\n");
917

918
	J_ASSERT(list_empty(&commit_transaction->t_inode_list));
919 920 921 922 923 924 925 926 927 928 929 930 931
	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;
932
		int try_to_free = 0;
933 934 935 936

		jh = commit_transaction->t_forget;
		spin_unlock(&journal->j_list_lock);
		bh = jh2bh(jh);
937 938 939 940 941
		/*
		 * Get a reference so that bh cannot be freed before we are
		 * done with it.
		 */
		get_bh(bh);
942
		jbd_lock_bh_state(bh);
943
		J_ASSERT_JH(jh,	jh->b_transaction == commit_transaction);
944 945 946 947 948 949 950 951 952 953

		/*
		 * 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
954 955 956
		 *
		 * We also know that the frozen data has already fired
		 * its triggers if they exist, so we can clear that too.
957 958
		 */
		if (jh->b_committed_data) {
M
Mingming Cao 已提交
959
			jbd2_free(jh->b_committed_data, bh->b_size);
960 961 962 963
			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
964
				jh->b_frozen_triggers = NULL;
965 966
			}
		} else if (jh->b_frozen_data) {
M
Mingming Cao 已提交
967
			jbd2_free(jh->b_frozen_data, bh->b_size);
968
			jh->b_frozen_data = NULL;
J
Joel Becker 已提交
969
			jh->b_frozen_triggers = NULL;
970 971 972 973 974 975
		}

		spin_lock(&journal->j_list_lock);
		cp_transaction = jh->b_cp_transaction;
		if (cp_transaction) {
			JBUFFER_TRACE(jh, "remove from old cp transaction");
976
			cp_transaction->t_chp_stats.cs_dropped++;
977
			__jbd2_journal_remove_checkpoint(jh);
978 979 980 981
		}

		/* Only re-checkpoint the buffer_head if it is marked
		 * dirty.  If the buffer was added to the BJ_Forget list
982
		 * by jbd2_journal_forget, it may no longer be dirty and
983 984 985
		 * there's no point in keeping a checkpoint record for
		 * it. */

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 1011 1012 1013 1014
		/*
		* 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;
			}
1015 1016 1017 1018
		}

		if (buffer_jbddirty(bh)) {
			JBUFFER_TRACE(jh, "add to new checkpointing trans");
1019
			__jbd2_journal_insert_checkpoint(jh, commit_transaction);
1020 1021
			if (is_journal_aborted(journal))
				clear_buffer_jbddirty(bh);
1022 1023
		} else {
			J_ASSERT_BH(bh, !buffer_dirty(bh));
1024 1025
			/*
			 * The buffer on BJ_Forget list and not jbddirty means
1026 1027 1028 1029 1030
			 * 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
1031 1032 1033 1034
			 * list.
			 */
			if (!jh->b_next_transaction)
				try_to_free = 1;
1035
		}
1036 1037 1038 1039 1040 1041 1042
		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);
1043 1044 1045 1046
		cond_resched_lock(&journal->j_list_lock);
	}
	spin_unlock(&journal->j_list_lock);
	/*
1047 1048 1049 1050
	 * 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...
1051
	 */
1052
	write_lock(&journal->j_state_lock);
1053 1054 1055 1056 1057 1058 1059
	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);
1060
		write_unlock(&journal->j_state_lock);
1061 1062 1063 1064 1065
		goto restart_loop;
	}

	/* Done with this transaction! */

E
Eryu Guan 已提交
1066
	jbd_debug(3, "JBD2: commit phase 7\n");
1067

1068
	J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
1069

1070
	commit_transaction->t_start = jiffies;
1071 1072
	stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
					      commit_transaction->t_start);
1073 1074

	/*
1075
	 * File the transaction statistics
1076 1077
	 */
	stats.ts_tid = commit_transaction->t_tid;
1078 1079
	stats.run.rs_handle_count =
		atomic_read(&commit_transaction->t_handle_count);
1080 1081
	trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
			     commit_transaction->t_tid, &stats.run);
1082 1083 1084 1085

	/*
	 * Calculate overall stats
	 */
1086
	spin_lock(&journal->j_history_lock);
1087
	journal->j_stats.ts_tid++;
1088 1089
	if (commit_transaction->t_requested)
		journal->j_stats.ts_requested++;
1090
	journal->j_stats.run.rs_wait += stats.run.rs_wait;
1091
	journal->j_stats.run.rs_request_delay += stats.run.rs_request_delay;
1092 1093 1094 1095 1096 1097 1098
	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;
1099 1100
	spin_unlock(&journal->j_history_lock);

1101
	commit_transaction->t_state = T_COMMIT_CALLBACK;
1102 1103 1104
	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
1105
	commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1106

J
Josef Bacik 已提交
1107 1108 1109 1110 1111 1112 1113 1114 1115
	/*
	 * 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;
1116

1117
	write_unlock(&journal->j_state_lock);
1118

1119 1120 1121 1122
	if (journal->j_checkpoint_transactions == NULL) {
		journal->j_checkpoint_transactions = commit_transaction;
		commit_transaction->t_cpnext = commit_transaction;
		commit_transaction->t_cpprev = commit_transaction;
1123
	} else {
1124 1125 1126 1127 1128 1129 1130
		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 =
1131 1132 1133
				commit_transaction;
	}
	spin_unlock(&journal->j_list_lock);
1134 1135 1136
	/* Drop all spin_locks because commit_callback may be block.
	 * __journal_remove_checkpoint() can not destroy transaction
	 * under us because it is not marked as T_FINISHED yet */
1137 1138 1139
	if (journal->j_commit_callback)
		journal->j_commit_callback(journal, commit_transaction);

1140
	trace_jbd2_end_commit(journal, commit_transaction);
E
Eryu Guan 已提交
1141
	jbd_debug(1, "JBD2: commit %d complete, head %d\n",
1142 1143
		  journal->j_commit_sequence, journal->j_tail_sequence);

1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
	write_lock(&journal->j_state_lock);
	spin_lock(&journal->j_list_lock);
	commit_transaction->t_state = T_FINISHED;
	/* Recheck checkpoint lists after j_list_lock was dropped */
	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);
1155 1156
	wake_up(&journal->j_wait_done_commit);
}