commit.c 31.4 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
#include <asm/system.h>
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46

/*
 * 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);
}

/*
47 48
 * When an ext4 file is truncated, it is possible that some pages are not
 * successfully freed, because they are attached to a committing transaction.
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 74
 * 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
75
	if (!trylock_page(page))
76 77 78 79 80 81 82 83 84 85 86 87 88
		goto nope;

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

nope:
	__brelse(bh);
}

89 90
/*
 * Done it all: now submit the commit record.  We should have
91 92 93 94 95 96
 * 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
 */
97 98 99 100
static int journal_submit_commit_record(journal_t *journal,
					transaction_t *commit_transaction,
					struct buffer_head **cbh,
					__u32 crc32_sum)
101 102
{
	struct journal_head *descriptor;
103
	struct commit_header *tmp;
104
	struct buffer_head *bh;
105
	int ret;
106
	struct timespec now = current_kernel_time();
107

108 109
	*cbh = NULL;

110 111 112
	if (is_journal_aborted(journal))
		return 0;

113
	descriptor = jbd2_journal_get_descriptor_buffer(journal);
114 115 116 117 118
	if (!descriptor)
		return 1;

	bh = jh2bh(descriptor);

119 120 121 122
	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);
123 124
	tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
	tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
125 126 127 128 129 130

	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);
131 132
	}

133 134
	JBUFFER_TRACE(descriptor, "submit commit block");
	lock_buffer(bh);
135
	clear_buffer_dirty(bh);
136 137 138 139
	set_buffer_uptodate(bh);
	bh->b_end_io = journal_end_buffer_io_sync;

	if (journal->j_flags & JBD2_BARRIER &&
140
	    !JBD2_HAS_INCOMPAT_FEATURE(journal,
141
				       JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
J
Jens Axboe 已提交
142
		ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh);
143
	else
J
Jens Axboe 已提交
144
		ret = submit_bh(WRITE_SYNC, bh);
145

146 147 148 149 150 151 152 153
	*cbh = bh;
	return ret;
}

/*
 * This function along with journal_submit_commit_record
 * allows to write the commit record asynchronously.
 */
154 155
static int journal_wait_on_commit_record(journal_t *journal,
					 struct buffer_head *bh)
156 157 158 159 160
{
	int ret = 0;

	clear_buffer_dirty(bh);
	wait_on_buffer(bh);
161

162 163 164 165 166 167
	if (unlikely(!buffer_uptodate(bh)))
		ret = -EIO;
	put_bh(bh);            /* One for getblk() */
	jbd2_journal_put_journal_head(bh2jh(bh));

	return ret;
168 169
}

170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
/*
 * 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;
}

190 191 192 193 194 195 196 197
/*
 * 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.
 */
198
static int journal_submit_data_buffers(journal_t *journal,
199 200 201 202 203 204 205 206 207
		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;
208
		set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
209
		spin_unlock(&journal->j_list_lock);
210 211 212 213 214 215
		/*
		 * 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.
		 */
216
		trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
217
		err = journal_submit_inode_data_buffers(mapping);
218 219 220 221
		if (!ret)
			ret = err;
		spin_lock(&journal->j_list_lock);
		J_ASSERT(jinode->i_transaction == commit_transaction);
222 223
		clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
		smp_mb__after_clear_bit();
224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
		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;

241
	/* For locking, see the comment in journal_submit_data_buffers() */
242 243
	spin_lock(&journal->j_list_lock);
	list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
244
		set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
245 246
		spin_unlock(&journal->j_list_lock);
		err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
247 248 249
		if (err) {
			/*
			 * Because AS_EIO is cleared by
250
			 * filemap_fdatawait_range(), set it again so
251 252 253 254 255 256 257 258
			 * that user process can get -EIO from fsync().
			 */
			set_bit(AS_EIO,
				&jinode->i_vfs_inode->i_mapping->flags);

			if (!ret)
				ret = err;
		}
259
		spin_lock(&journal->j_list_lock);
260 261
		clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
		smp_mb__after_clear_bit();
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
		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;
}

283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
{
	struct page *page = bh->b_page;
	char *addr;
	__u32 checksum;

	addr = kmap_atomic(page, KM_USER0);
	checksum = crc32_be(crc32_sum,
		(void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
	kunmap_atomic(addr, KM_USER0);

	return checksum;
}

static void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
298
				   unsigned long long block)
Z
Zach Brown 已提交
299 300
{
	tag->t_blocknr = cpu_to_be32(block & (u32)~0);
301
	if (tag_bytes > JBD2_TAG_SIZE32)
Z
Zach Brown 已提交
302 303 304
		tag->t_blocknr_high = cpu_to_be32((block >> 31) >> 1);
}

305
/*
306
 * jbd2_journal_commit_transaction
307 308 309 310
 *
 * The primary function for committing a transaction to the log.  This
 * function is called by the journal thread to begin a complete commit.
 */
311
void jbd2_journal_commit_transaction(journal_t *journal)
312
{
313
	struct transaction_stats_s stats;
314 315 316 317 318 319
	transaction_t *commit_transaction;
	struct journal_head *jh, *new_jh, *descriptor;
	struct buffer_head **wbuf = journal->j_wbuf;
	int bufs;
	int flags;
	int err;
320
	unsigned long long blocknr;
J
Josef Bacik 已提交
321 322
	ktime_t start_time;
	u64 commit_time;
323 324 325 326 327 328
	char *tagp = NULL;
	journal_header_t *header;
	journal_block_tag_t *tag = NULL;
	int space_left = 0;
	int first_tag = 0;
	int tag_flag;
329
	int i, to_free = 0;
Z
Zach Brown 已提交
330
	int tag_bytes = journal_tag_bytes(journal);
331 332
	struct buffer_head *cbh = NULL; /* For transactional checksums */
	__u32 crc32_sum = ~0;
333
	struct blk_plug plug;
334 335 336 337 338 339

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

340 341
	/* Do we need to erase the effects of a prior jbd2_journal_flush? */
	if (journal->j_flags & JBD2_FLUSHED) {
342
		jbd_debug(3, "super block updated\n");
343
		jbd2_journal_update_superblock(journal, 1);
344 345 346 347 348 349 350 351 352 353
	} 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);

354
	trace_jbd2_start_commit(journal, commit_transaction);
355 356 357
	jbd_debug(1, "JBD: starting commit of transaction %d\n",
			commit_transaction->t_tid);

358
	write_lock(&journal->j_state_lock);
359 360
	commit_transaction->t_state = T_LOCKED;

361
	trace_jbd2_commit_locking(journal, commit_transaction);
362 363 364 365
	stats.run.rs_wait = commit_transaction->t_max_wait;
	stats.run.rs_locked = jiffies;
	stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
					      stats.run.rs_locked);
366

367
	spin_lock(&commit_transaction->t_handle_lock);
368
	while (atomic_read(&commit_transaction->t_updates)) {
369 370 371 372
		DEFINE_WAIT(wait);

		prepare_to_wait(&journal->j_wait_updates, &wait,
					TASK_UNINTERRUPTIBLE);
373
		if (atomic_read(&commit_transaction->t_updates)) {
374
			spin_unlock(&commit_transaction->t_handle_lock);
375
			write_unlock(&journal->j_state_lock);
376
			schedule();
377
			write_lock(&journal->j_state_lock);
378 379 380 381 382 383
			spin_lock(&commit_transaction->t_handle_lock);
		}
		finish_wait(&journal->j_wait_updates, &wait);
	}
	spin_unlock(&commit_transaction->t_handle_lock);

384
	J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
385 386 387 388 389 390 391
			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
392
	 * transactions, then it may try to do a jbd2_journal_restart() while
393 394 395 396 397 398 399
	 * 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
400
	 * that multiple jbd2_journal_get_write_access() calls to the same
L
Lucas De Marchi 已提交
401
	 * buffer are perfectly permissible.
402 403 404 405 406
	 */
	while (commit_transaction->t_reserved_list) {
		jh = commit_transaction->t_reserved_list;
		JBUFFER_TRACE(jh, "reserved, unused: refile");
		/*
407
		 * A jbd2_journal_get_undo_access()+jbd2_journal_release_buffer() may
408 409 410 411 412 413
		 * 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
414
			jbd2_free(jh->b_committed_data, bh->b_size);
415 416 417
			jh->b_committed_data = NULL;
			jbd_unlock_bh_state(bh);
		}
418
		jbd2_journal_refile_buffer(journal, jh);
419 420 421 422 423 424 425 426
	}

	/*
	 * 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);
427
	__jbd2_journal_clean_checkpoint_list(journal);
428 429 430 431 432 433 434
	spin_unlock(&journal->j_list_lock);

	jbd_debug (3, "JBD: commit phase 1\n");

	/*
	 * Switch to a new revoke table.
	 */
435
	jbd2_journal_switch_revoke_table(journal);
436

437
	trace_jbd2_commit_flushing(journal, commit_transaction);
438 439 440
	stats.run.rs_flushing = jiffies;
	stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
					     stats.run.rs_flushing);
441

442 443 444
	commit_transaction->t_state = T_FLUSH;
	journal->j_committing_transaction = commit_transaction;
	journal->j_running_transaction = NULL;
J
Josef Bacik 已提交
445
	start_time = ktime_get();
446 447
	commit_transaction->t_log_start = journal->j_head;
	wake_up(&journal->j_wait_transaction_locked);
448
	write_unlock(&journal->j_state_lock);
449 450 451 452 453 454 455

	jbd_debug (3, "JBD: commit phase 2\n");

	/*
	 * Now start flushing things to disk, in the order they appear
	 * on the transaction lists.  Data blocks go first.
	 */
456
	err = journal_submit_data_buffers(journal, commit_transaction);
457
	if (err)
458
		jbd2_journal_abort(journal, err);
459

460
	blk_start_plug(&plug);
461
	jbd2_journal_write_revoke_records(journal, commit_transaction,
462 463
					  WRITE_SYNC);
	blk_finish_plug(&plug);
464 465 466 467 468 469 470 471

	jbd_debug(3, "JBD: commit phase 2\n");

	/*
	 * 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:
	 */
472
	write_lock(&journal->j_state_lock);
473
	commit_transaction->t_state = T_COMMIT;
474
	write_unlock(&journal->j_state_lock);
475

476
	trace_jbd2_commit_logging(journal, commit_transaction);
477 478 479
	stats.run.rs_logging = jiffies;
	stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
					       stats.run.rs_logging);
480 481
	stats.run.rs_blocks =
		atomic_read(&commit_transaction->t_outstanding_credits);
482
	stats.run.rs_blocks_logged = 0;
483

484
	J_ASSERT(commit_transaction->t_nr_buffers <=
485
		 atomic_read(&commit_transaction->t_outstanding_credits));
486

487
	err = 0;
488 489
	descriptor = NULL;
	bufs = 0;
490
	blk_start_plug(&plug);
491 492 493 494 495 496 497
	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
498
		   release it. */
499 500

		if (is_journal_aborted(journal)) {
501
			clear_buffer_jbddirty(jh2bh(jh));
502
			JBUFFER_TRACE(jh, "journal is aborting: refile");
J
Joel Becker 已提交
503 504 505 506
			jbd2_buffer_abort_trigger(jh,
						  jh->b_frozen_data ?
						  jh->b_frozen_triggers :
						  jh->b_triggers);
507
			jbd2_journal_refile_buffer(journal, jh);
508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
			/* 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) {
			struct buffer_head *bh;

			J_ASSERT (bufs == 0);

			jbd_debug(4, "JBD: get descriptor\n");

527
			descriptor = jbd2_journal_get_descriptor_buffer(journal);
528
			if (!descriptor) {
529
				jbd2_journal_abort(journal, -EIO);
530 531 532 533 534 535 536
				continue;
			}

			bh = jh2bh(descriptor);
			jbd_debug(4, "JBD: got buffer %llu (%p)\n",
				(unsigned long long)bh->b_blocknr, bh->b_data);
			header = (journal_header_t *)&bh->b_data[0];
537 538
			header->h_magic     = cpu_to_be32(JBD2_MAGIC_NUMBER);
			header->h_blocktype = cpu_to_be32(JBD2_DESCRIPTOR_BLOCK);
539 540 541 542 543 544 545 546 547 548 549 550
			header->h_sequence  = cpu_to_be32(commit_transaction->t_tid);

			tagp = &bh->b_data[sizeof(journal_header_t)];
			space_left = bh->b_size - sizeof(journal_header_t);
			first_tag = 1;
			set_buffer_jwrite(bh);
			set_buffer_dirty(bh);
			wbuf[bufs++] = bh;

			/* Record it so that we can wait for IO
                           completion later */
			BUFFER_TRACE(bh, "ph3: file as descriptor");
551
			jbd2_journal_file_buffer(descriptor, commit_transaction,
552 553 554 555 556
					BJ_LogCtl);
		}

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

557
		err = jbd2_journal_next_log_block(journal, &blocknr);
558 559 560 561
		/* 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) {
562
			jbd2_journal_abort(journal, err);
563 564 565 566 567 568
			continue;
		}

		/*
		 * start_this_handle() uses t_outstanding_credits to determine
		 * the free space in the log, but this counter is changed
569
		 * by jbd2_journal_next_log_block() also.
570
		 */
571
		atomic_dec(&commit_transaction->t_outstanding_credits);
572 573 574 575 576 577 578 579 580 581 582 583

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

		/* Make a temporary IO buffer with which to write it out
                   (this will requeue both the metadata buffer and the
                   temporary IO buffer). new_bh goes on BJ_IO*/

		set_bit(BH_JWrite, &jh2bh(jh)->b_state);
		/*
584
		 * akpm: jbd2_journal_write_metadata_buffer() sets
585 586 587 588 589
		 * new_bh->b_transaction to commit_transaction.
		 * We need to clean this up before we release new_bh
		 * (which is of type BJ_IO)
		 */
		JBUFFER_TRACE(jh, "ph3: write metadata");
590
		flags = jbd2_journal_write_metadata_buffer(commit_transaction,
591
						      jh, &new_jh, blocknr);
592 593 594 595
		if (flags < 0) {
			jbd2_journal_abort(journal, flags);
			continue;
		}
596 597 598 599 600 601 602 603
		set_bit(BH_JWrite, &jh2bh(new_jh)->b_state);
		wbuf[bufs++] = jh2bh(new_jh);

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

		tag_flag = 0;
		if (flags & 1)
604
			tag_flag |= JBD2_FLAG_ESCAPE;
605
		if (!first_tag)
606
			tag_flag |= JBD2_FLAG_SAME_UUID;
607 608

		tag = (journal_block_tag_t *) tagp;
Z
Zach Brown 已提交
609
		write_tag_block(tag_bytes, tag, jh2bh(jh)->b_blocknr);
610
		tag->t_flags = cpu_to_be32(tag_flag);
Z
Zach Brown 已提交
611 612
		tagp += tag_bytes;
		space_left -= tag_bytes;
613 614 615 616 617 618 619 620 621 622 623 624 625

		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 ||
Z
Zach Brown 已提交
626
		    space_left < tag_bytes + 16) {
627 628 629 630 631 632 633

			jbd_debug(4, "JBD: Submit %d IOs\n", bufs);

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

634
			tag->t_flags |= cpu_to_be32(JBD2_FLAG_LAST_TAG);
635 636 637 638

start_journal_io:
			for (i = 0; i < bufs; i++) {
				struct buffer_head *bh = wbuf[i];
639 640 641 642 643 644 645 646 647
				/*
				 * Compute checksum.
				 */
				if (JBD2_HAS_COMPAT_FEATURE(journal,
					JBD2_FEATURE_COMPAT_CHECKSUM)) {
					crc32_sum =
					    jbd2_checksum_data(crc32_sum, bh);
				}

648 649 650 651
				lock_buffer(bh);
				clear_buffer_dirty(bh);
				set_buffer_uptodate(bh);
				bh->b_end_io = journal_end_buffer_io_sync;
652
				submit_bh(WRITE_SYNC, bh);
653 654
			}
			cond_resched();
655
			stats.run.rs_blocks_logged += bufs;
656 657 658 659 660 661 662 663

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

664 665 666 667 668 669 670 671 672 673
	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;
	}

674 675 676 677
	write_lock(&journal->j_state_lock);
	J_ASSERT(commit_transaction->t_state == T_COMMIT);
	commit_transaction->t_state = T_COMMIT_DFLUSH;
	write_unlock(&journal->j_state_lock);
678 679 680 681 682
	/* 
	 * 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
	 */
683
	if (commit_transaction->t_need_data_flush &&
684 685
	    (journal->j_fs_dev != journal->j_dev) &&
	    (journal->j_flags & JBD2_BARRIER))
686
		blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
687

688
	/* Done it all: now write the commit record asynchronously. */
689
	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
690
				      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
691 692 693 694
		err = journal_submit_commit_record(journal, commit_transaction,
						 &cbh, crc32_sum);
		if (err)
			__jbd2_journal_abort_hard(journal);
695
	}
696

697 698
	blk_finish_plug(&plug);

699 700 701 702 703 704 705 706 707 708 709
	/* 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
           the t_iobuf_list queue.

	   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.
	*/

710
	jbd_debug(3, "JBD: commit phase 3\n");
711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734

	/*
	 * akpm: these are BJ_IO, and j_list_lock is not needed.
	 * See __journal_try_to_free_buffer.
	 */
wait_for_iobuf:
	while (commit_transaction->t_iobuf_list != NULL) {
		struct buffer_head *bh;

		jh = commit_transaction->t_iobuf_list->b_tprev;
		bh = jh2bh(jh);
		if (buffer_locked(bh)) {
			wait_on_buffer(bh);
			goto wait_for_iobuf;
		}
		if (cond_resched())
			goto wait_for_iobuf;

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

		clear_buffer_jwrite(bh);

		JBUFFER_TRACE(jh, "ph4: unfile after journal write");
735
		jbd2_journal_unfile_buffer(journal, jh);
736 737 738

		/*
		 * ->t_iobuf_list should contain only dummy buffer_heads
739
		 * which were created by jbd2_journal_write_metadata_buffer().
740 741
		 */
		BUFFER_TRACE(bh, "dumping temporary bh");
742
		jbd2_journal_put_journal_head(jh);
743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
		__brelse(bh);
		J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
		free_buffer_head(bh);

		/* We also have to unlock and free the corresponding
                   shadowed buffer */
		jh = commit_transaction->t_shadow_list->b_tprev;
		bh = jh2bh(jh);
		clear_bit(BH_JWrite, &bh->b_state);
		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");
759
		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
760 761 762 763 764 765 766
		/*
		 * 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();
767 768 769 770 771 772 773
		wake_up_bit(&bh->b_state, BH_Unshadow);
		JBUFFER_TRACE(jh, "brelse shadowed buffer");
		__brelse(bh);
	}

	J_ASSERT (commit_transaction->t_shadow_list == NULL);

774
	jbd_debug(3, "JBD: commit phase 4\n");
775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794

	/* Here we wait for the revoke record and descriptor record buffers */
 wait_for_ctlbuf:
	while (commit_transaction->t_log_list != NULL) {
		struct buffer_head *bh;

		jh = commit_transaction->t_log_list->b_tprev;
		bh = jh2bh(jh);
		if (buffer_locked(bh)) {
			wait_on_buffer(bh);
			goto wait_for_ctlbuf;
		}
		if (cond_resched())
			goto wait_for_ctlbuf;

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

		BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
		clear_buffer_jwrite(bh);
795 796
		jbd2_journal_unfile_buffer(journal, jh);
		jbd2_journal_put_journal_head(jh);
797 798 799 800
		__brelse(bh);		/* One for getblk */
		/* AKPM: bforget here */
	}

801 802 803
	if (err)
		jbd2_journal_abort(journal, err);

804
	jbd_debug(3, "JBD: commit phase 5\n");
805 806 807 808
	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);
809

810
	if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
811
				       JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
812 813 814 815 816
		err = journal_submit_commit_record(journal, commit_transaction,
						&cbh, crc32_sum);
		if (err)
			__jbd2_journal_abort_hard(journal);
	}
817
	if (cbh)
818
		err = journal_wait_on_commit_record(journal, cbh);
819 820 821
	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
				      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
	    journal->j_flags & JBD2_BARRIER) {
822
		blkdev_issue_flush(journal->j_dev, GFP_KERNEL, NULL);
823
	}
824 825

	if (err)
826
		jbd2_journal_abort(journal, err);
827 828 829 830 831 832

	/* 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. */

833
	jbd_debug(3, "JBD: commit phase 6\n");
834

835
	J_ASSERT(list_empty(&commit_transaction->t_inode_list));
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850
	J_ASSERT(commit_transaction->t_buffers == NULL);
	J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
	J_ASSERT(commit_transaction->t_iobuf_list == NULL);
	J_ASSERT(commit_transaction->t_shadow_list == NULL);
	J_ASSERT(commit_transaction->t_log_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;
851
		int try_to_free = 0;
852 853 854 855

		jh = commit_transaction->t_forget;
		spin_unlock(&journal->j_list_lock);
		bh = jh2bh(jh);
856 857 858 859 860
		/*
		 * Get a reference so that bh cannot be freed before we are
		 * done with it.
		 */
		get_bh(bh);
861
		jbd_lock_bh_state(bh);
862
		J_ASSERT_JH(jh,	jh->b_transaction == commit_transaction);
863 864 865 866 867 868 869 870 871 872

		/*
		 * 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
873 874 875
		 *
		 * We also know that the frozen data has already fired
		 * its triggers if they exist, so we can clear that too.
876 877
		 */
		if (jh->b_committed_data) {
M
Mingming Cao 已提交
878
			jbd2_free(jh->b_committed_data, bh->b_size);
879 880 881 882
			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
883
				jh->b_frozen_triggers = NULL;
884 885
			}
		} else if (jh->b_frozen_data) {
M
Mingming Cao 已提交
886
			jbd2_free(jh->b_frozen_data, bh->b_size);
887
			jh->b_frozen_data = NULL;
J
Joel Becker 已提交
888
			jh->b_frozen_triggers = NULL;
889 890 891 892 893 894
		}

		spin_lock(&journal->j_list_lock);
		cp_transaction = jh->b_cp_transaction;
		if (cp_transaction) {
			JBUFFER_TRACE(jh, "remove from old cp transaction");
895
			cp_transaction->t_chp_stats.cs_dropped++;
896
			__jbd2_journal_remove_checkpoint(jh);
897 898 899 900
		}

		/* Only re-checkpoint the buffer_head if it is marked
		 * dirty.  If the buffer was added to the BJ_Forget list
901
		 * by jbd2_journal_forget, it may no longer be dirty and
902 903 904 905 906 907
		 * there's no point in keeping a checkpoint record for
		 * it. */

		/* A buffer which has been freed while still being
		 * journaled by a previous transaction may end up still
		 * being dirty here, but we want to avoid writing back
908 909 910 911
		 * that buffer in the future after the "add to orphan"
		 * operation been committed,  That's not only a performance
		 * gain, it also stops aliasing problems if the buffer is
		 * left behind for writeback and gets reallocated for another
912
		 * use in a different page. */
913
		if (buffer_freed(bh) && !jh->b_next_transaction) {
914 915 916 917 918 919
			clear_buffer_freed(bh);
			clear_buffer_jbddirty(bh);
		}

		if (buffer_jbddirty(bh)) {
			JBUFFER_TRACE(jh, "add to new checkpointing trans");
920
			__jbd2_journal_insert_checkpoint(jh, commit_transaction);
921 922
			if (is_journal_aborted(journal))
				clear_buffer_jbddirty(bh);
923 924
		} else {
			J_ASSERT_BH(bh, !buffer_dirty(bh));
925 926
			/*
			 * The buffer on BJ_Forget list and not jbddirty means
927 928 929 930 931
			 * 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
932 933 934 935
			 * list.
			 */
			if (!jh->b_next_transaction)
				try_to_free = 1;
936
		}
937 938 939 940 941 942 943
		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);
944 945 946 947
		cond_resched_lock(&journal->j_list_lock);
	}
	spin_unlock(&journal->j_list_lock);
	/*
948 949 950 951
	 * 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...
952
	 */
953
	write_lock(&journal->j_state_lock);
954 955 956 957 958 959 960
	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);
961
		write_unlock(&journal->j_state_lock);
962 963 964 965 966
		goto restart_loop;
	}

	/* Done with this transaction! */

967
	jbd_debug(3, "JBD: commit phase 7\n");
968

969
	J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
970

971
	commit_transaction->t_start = jiffies;
972 973
	stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
					      commit_transaction->t_start);
974 975

	/*
976
	 * File the transaction statistics
977 978
	 */
	stats.ts_tid = commit_transaction->t_tid;
979 980
	stats.run.rs_handle_count =
		atomic_read(&commit_transaction->t_handle_count);
981 982
	trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
			     commit_transaction->t_tid, &stats.run);
983 984 985 986

	/*
	 * Calculate overall stats
	 */
987
	spin_lock(&journal->j_history_lock);
988
	journal->j_stats.ts_tid++;
989 990 991 992 993 994 995 996
	journal->j_stats.run.rs_wait += stats.run.rs_wait;
	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;
997 998
	spin_unlock(&journal->j_history_lock);

999 1000 1001 1002
	commit_transaction->t_state = T_FINISHED;
	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
1003
	commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1004

J
Josef Bacik 已提交
1005 1006 1007 1008 1009 1010 1011 1012 1013
	/*
	 * 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;
1014
	write_unlock(&journal->j_state_lock);
1015

J
Jan Kara 已提交
1016 1017
	if (commit_transaction->t_checkpoint_list == NULL &&
	    commit_transaction->t_checkpoint_io_list == NULL) {
1018
		__jbd2_journal_drop_transaction(journal, commit_transaction);
1019
		to_free = 1;
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
	} else {
		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);

1038 1039 1040
	if (journal->j_commit_callback)
		journal->j_commit_callback(journal, commit_transaction);

1041
	trace_jbd2_end_commit(journal, commit_transaction);
1042 1043
	jbd_debug(1, "JBD: commit %d complete, head %d\n",
		  journal->j_commit_sequence, journal->j_tail_sequence);
1044 1045
	if (to_free)
		kfree(commit_transaction);
1046 1047 1048

	wake_up(&journal->j_wait_done_commit);
}