commit.c 27.7 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 * linux/fs/jbd/commit.c
L
Linus Torvalds 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
 *
 * 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>
#include <linux/jbd.h>
#include <linux/errno.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
22
#include <linux/bio.h>
23
#include <linux/blkdev.h>
L
Linus Torvalds 已提交
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

/*
 * 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);
}

/*
 * When an ext3-ordered file is truncated, it is possible that many pages are
40
 * not successfully freed, because they are attached to a committing transaction.
L
Linus Torvalds 已提交
41 42 43 44 45 46 47 48
 * 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.
 *
49 50
 * Called under journal->j_list_lock.  The caller provided us with a ref
 * against the buffer, and we drop that here.
L
Linus Torvalds 已提交
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
 */
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
67
	if (!trylock_page(page))
L
Linus Torvalds 已提交
68 69 70 71 72 73 74 75 76 77 78 79 80
		goto nope;

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

nope:
	__brelse(bh);
}

81 82 83 84 85 86 87 88 89 90 91 92 93
/*
 * Decrement reference counter for data buffer. If it has been marked
 * 'BH_Freed', release it and the page to which it belongs if possible.
 */
static void release_data_buffer(struct buffer_head *bh)
{
	if (buffer_freed(bh)) {
		clear_buffer_freed(bh);
		release_buffer_page(bh);
	} else
		put_bh(bh);
}

L
Linus Torvalds 已提交
94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
/*
 * Try to acquire jbd_lock_bh_state() against the buffer, when j_list_lock is
 * held.  For ranking reasons we must trylock.  If we lose, schedule away and
 * return 0.  j_list_lock is dropped in this case.
 */
static int inverted_lock(journal_t *journal, struct buffer_head *bh)
{
	if (!jbd_trylock_bh_state(bh)) {
		spin_unlock(&journal->j_list_lock);
		schedule();
		return 0;
	}
	return 1;
}

/* Done it all: now write the commit record.  We should have
 * 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
 */
static int journal_write_commit_record(journal_t *journal,
					transaction_t *commit_transaction)
{
	struct journal_head *descriptor;
	struct buffer_head *bh;
121 122
	journal_header_t *header;
	int ret;
L
Linus Torvalds 已提交
123 124 125 126 127 128 129 130 131 132

	if (is_journal_aborted(journal))
		return 0;

	descriptor = journal_get_descriptor_buffer(journal);
	if (!descriptor)
		return 1;

	bh = jh2bh(descriptor);

133 134 135 136
	header = (journal_header_t *)(bh->b_data);
	header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
	header->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
	header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
L
Linus Torvalds 已提交
137 138 139

	JBUFFER_TRACE(descriptor, "write commit block");
	set_buffer_dirty(bh);
C
Christoph Hellwig 已提交
140

141 142 143
	if (journal->j_flags & JFS_BARRIER)
		ret = __sync_dirty_buffer(bh, WRITE_SYNC | WRITE_FLUSH_FUA);
	else
L
Linus Torvalds 已提交
144
		ret = sync_dirty_buffer(bh);
C
Christoph Hellwig 已提交
145

L
Linus Torvalds 已提交
146 147 148 149 150 151
	put_bh(bh);		/* One for getblk() */
	journal_put_journal_head(descriptor);

	return (ret == -EIO);
}

152 153
static void journal_do_submit_data(struct buffer_head **wbuf, int bufs,
				   int write_op)
154 155 156 157 158 159
{
	int i;

	for (i = 0; i < bufs; i++) {
		wbuf[i]->b_end_io = end_buffer_write_sync;
		/* We use-up our safety reference in submit_bh() */
160
		submit_bh(write_op, wbuf[i]);
161 162 163 164 165 166
	}
}

/*
 *  Submit all the data buffers to disk
 */
167
static int journal_submit_data_buffers(journal_t *journal,
168 169
				       transaction_t *commit_transaction,
				       int write_op)
170 171 172 173 174 175
{
	struct journal_head *jh;
	struct buffer_head *bh;
	int locked;
	int bufs = 0;
	struct buffer_head **wbuf = journal->j_wbuf;
176
	int err = 0;
177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203

	/*
	 * Whenever we unlock the journal and sleep, things can get added
	 * onto ->t_sync_datalist, so we have to keep looping back to
	 * write_out_data until we *know* that the list is empty.
	 *
	 * Cleanup any flushed data buffers from the data list.  Even in
	 * abort mode, we want to flush this out as soon as possible.
	 */
write_out_data:
	cond_resched();
	spin_lock(&journal->j_list_lock);

	while (commit_transaction->t_sync_datalist) {
		jh = commit_transaction->t_sync_datalist;
		bh = jh2bh(jh);
		locked = 0;

		/* Get reference just to make sure buffer does not disappear
		 * when we are forced to drop various locks */
		get_bh(bh);
		/* If the buffer is dirty, we need to submit IO and hence
		 * we need the buffer lock. We try to lock the buffer without
		 * blocking. If we fail, we need to drop j_list_lock and do
		 * blocking lock_buffer().
		 */
		if (buffer_dirty(bh)) {
N
Nick Piggin 已提交
204
			if (!trylock_buffer(bh)) {
205 206 207
				BUFFER_TRACE(bh, "needs blocking lock");
				spin_unlock(&journal->j_list_lock);
				/* Write out all data to prevent deadlocks */
208
				journal_do_submit_data(wbuf, bufs, write_op);
209 210 211 212 213 214 215 216 217 218 219 220
				bufs = 0;
				lock_buffer(bh);
				spin_lock(&journal->j_list_lock);
			}
			locked = 1;
		}
		/* We have to get bh_state lock. Again out of order, sigh. */
		if (!inverted_lock(journal, bh)) {
			jbd_lock_bh_state(bh);
			spin_lock(&journal->j_list_lock);
		}
		/* Someone already cleaned up the buffer? */
221
		if (!buffer_jbd(bh) || bh2jh(bh) != jh
222 223 224 225 226 227
			|| jh->b_transaction != commit_transaction
			|| jh->b_jlist != BJ_SyncData) {
			jbd_unlock_bh_state(bh);
			if (locked)
				unlock_buffer(bh);
			BUFFER_TRACE(bh, "already cleaned up");
228
			release_data_buffer(bh);
229 230 231 232 233 234 235 236 237 238
			continue;
		}
		if (locked && test_clear_buffer_dirty(bh)) {
			BUFFER_TRACE(bh, "needs writeout, adding to array");
			wbuf[bufs++] = bh;
			__journal_file_buffer(jh, commit_transaction,
						BJ_Locked);
			jbd_unlock_bh_state(bh);
			if (bufs == journal->j_wbufsize) {
				spin_unlock(&journal->j_list_lock);
239
				journal_do_submit_data(wbuf, bufs, write_op);
240 241 242
				bufs = 0;
				goto write_out_data;
			}
243 244 245 246 247 248
		} else if (!locked && buffer_locked(bh)) {
			__journal_file_buffer(jh, commit_transaction,
						BJ_Locked);
			jbd_unlock_bh_state(bh);
			put_bh(bh);
		} else {
249
			BUFFER_TRACE(bh, "writeout complete: unfile");
250 251
			if (unlikely(!buffer_uptodate(bh)))
				err = -EIO;
252 253 254 255 256
			__journal_unfile_buffer(jh);
			jbd_unlock_bh_state(bh);
			if (locked)
				unlock_buffer(bh);
			journal_remove_journal_head(bh);
257
			/* One for our safety reference, other for
258 259
			 * journal_remove_journal_head() */
			put_bh(bh);
260
			release_data_buffer(bh);
261 262
		}

N
Nick Piggin 已提交
263
		if (need_resched() || spin_needbreak(&journal->j_list_lock)) {
264 265 266 267 268
			spin_unlock(&journal->j_list_lock);
			goto write_out_data;
		}
	}
	spin_unlock(&journal->j_list_lock);
269
	journal_do_submit_data(wbuf, bufs, write_op);
270 271

	return err;
272 273
}

L
Linus Torvalds 已提交
274 275 276 277 278 279 280 281 282 283 284 285 286 287
/*
 * journal_commit_transaction
 *
 * The primary function for committing a transaction to the log.  This
 * function is called by the journal thread to begin a complete commit.
 */
void journal_commit_transaction(journal_t *journal)
{
	transaction_t *commit_transaction;
	struct journal_head *jh, *new_jh, *descriptor;
	struct buffer_head **wbuf = journal->j_wbuf;
	int bufs;
	int flags;
	int err;
288
	unsigned int blocknr;
J
Josef Bacik 已提交
289 290
	ktime_t start_time;
	u64 commit_time;
L
Linus Torvalds 已提交
291 292 293 294 295 296 297
	char *tagp = NULL;
	journal_header_t *header;
	journal_block_tag_t *tag = NULL;
	int space_left = 0;
	int first_tag = 0;
	int tag_flag;
	int i;
298
	struct blk_plug plug;
L
Linus Torvalds 已提交
299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358

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

	/* Do we need to erase the effects of a prior journal_flush? */
	if (journal->j_flags & JFS_FLUSHED) {
		jbd_debug(3, "super block updated\n");
		journal_update_superblock(journal, 1);
	} 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);

	jbd_debug(1, "JBD: starting commit of transaction %d\n",
			commit_transaction->t_tid);

	spin_lock(&journal->j_state_lock);
	commit_transaction->t_state = T_LOCKED;

	spin_lock(&commit_transaction->t_handle_lock);
	while (commit_transaction->t_updates) {
		DEFINE_WAIT(wait);

		prepare_to_wait(&journal->j_wait_updates, &wait,
					TASK_UNINTERRUPTIBLE);
		if (commit_transaction->t_updates) {
			spin_unlock(&commit_transaction->t_handle_lock);
			spin_unlock(&journal->j_state_lock);
			schedule();
			spin_lock(&journal->j_state_lock);
			spin_lock(&commit_transaction->t_handle_lock);
		}
		finish_wait(&journal->j_wait_updates, &wait);
	}
	spin_unlock(&commit_transaction->t_handle_lock);

	J_ASSERT (commit_transaction->t_outstanding_credits <=
			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
	 * transactions, then it may try to do a journal_restart() while
	 * 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
	 * that multiple journal_get_write_access() calls to the same
L
Lucas De Marchi 已提交
359
	 * buffer are perfectly permissible.
L
Linus Torvalds 已提交
360 361 362 363 364 365 366 367 368 369 370 371
	 */
	while (commit_transaction->t_reserved_list) {
		jh = commit_transaction->t_reserved_list;
		JBUFFER_TRACE(jh, "reserved, unused: refile");
		/*
		 * A journal_get_undo_access()+journal_release_buffer() may
		 * 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
372
			jbd_free(jh->b_committed_data, bh->b_size);
J
Jesper Juhl 已提交
373
			jh->b_committed_data = NULL;
L
Linus Torvalds 已提交
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
			jbd_unlock_bh_state(bh);
		}
		journal_refile_buffer(journal, jh);
	}

	/*
	 * 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);
	__journal_clean_checkpoint_list(journal);
	spin_unlock(&journal->j_list_lock);

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

	/*
	 * Switch to a new revoke table.
	 */
	journal_switch_revoke_table(journal);

	commit_transaction->t_state = T_FLUSH;
	journal->j_committing_transaction = commit_transaction;
	journal->j_running_transaction = NULL;
J
Josef Bacik 已提交
398
	start_time = ktime_get();
L
Linus Torvalds 已提交
399 400 401 402 403 404 405 406 407 408
	commit_transaction->t_log_start = journal->j_head;
	wake_up(&journal->j_wait_transaction_locked);
	spin_unlock(&journal->j_state_lock);

	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.
	 */
409
	blk_start_plug(&plug);
410
	err = journal_submit_data_buffers(journal, commit_transaction,
411 412
					  WRITE_SYNC);
	blk_finish_plug(&plug);
L
Linus Torvalds 已提交
413 414 415 416

	/*
	 * Wait for all previously submitted IO to complete.
	 */
417
	spin_lock(&journal->j_list_lock);
L
Linus Torvalds 已提交
418 419 420 421 422 423 424 425 426 427 428
	while (commit_transaction->t_locked_list) {
		struct buffer_head *bh;

		jh = commit_transaction->t_locked_list->b_tprev;
		bh = jh2bh(jh);
		get_bh(bh);
		if (buffer_locked(bh)) {
			spin_unlock(&journal->j_list_lock);
			wait_on_buffer(bh);
			spin_lock(&journal->j_list_lock);
		}
429
		if (unlikely(!buffer_uptodate(bh))) {
N
Nick Piggin 已提交
430
			if (!trylock_page(bh->b_page)) {
431 432 433 434 435 436 437 438 439 440 441
				spin_unlock(&journal->j_list_lock);
				lock_page(bh->b_page);
				spin_lock(&journal->j_list_lock);
			}
			if (bh->b_page->mapping)
				set_bit(AS_EIO, &bh->b_page->mapping->flags);

			unlock_page(bh->b_page);
			SetPageError(bh->b_page);
			err = -EIO;
		}
L
Linus Torvalds 已提交
442 443 444 445 446
		if (!inverted_lock(journal, bh)) {
			put_bh(bh);
			spin_lock(&journal->j_list_lock);
			continue;
		}
447 448 449
		if (buffer_jbd(bh) && bh2jh(bh) == jh &&
		    jh->b_transaction == commit_transaction &&
		    jh->b_jlist == BJ_Locked) {
L
Linus Torvalds 已提交
450 451 452 453 454 455 456
			__journal_unfile_buffer(jh);
			jbd_unlock_bh_state(bh);
			journal_remove_journal_head(bh);
			put_bh(bh);
		} else {
			jbd_unlock_bh_state(bh);
		}
457
		release_data_buffer(bh);
L
Linus Torvalds 已提交
458 459 460 461
		cond_resched_lock(&journal->j_list_lock);
	}
	spin_unlock(&journal->j_list_lock);

462 463 464 465 466 467
	if (err) {
		char b[BDEVNAME_SIZE];

		printk(KERN_WARNING
			"JBD: Detected IO errors while flushing file data "
			"on %s\n", bdevname(journal->j_fs_dev, b));
468 469
		if (journal->j_flags & JFS_ABORT_ON_SYNCDATA_ERR)
			journal_abort(journal, err);
470 471
		err = 0;
	}
L
Linus Torvalds 已提交
472

473 474 475
	blk_start_plug(&plug);

	journal_write_revoke_records(journal, commit_transaction, WRITE_SYNC);
L
Linus Torvalds 已提交
476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491

	/*
	 * If we found any dirty or locked buffers, then we should have
	 * looped back up to the write_out_data label.  If there weren't
	 * any then journal_clean_data_list should have wiped the list
	 * clean by now, so check that it is in fact empty.
	 */
	J_ASSERT (commit_transaction->t_sync_datalist == NULL);

	jbd_debug (3, "JBD: commit phase 3\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:
	 */
492
	spin_lock(&journal->j_state_lock);
L
Linus Torvalds 已提交
493
	commit_transaction->t_state = T_COMMIT;
494
	spin_unlock(&journal->j_state_lock);
L
Linus Torvalds 已提交
495

496 497 498
	J_ASSERT(commit_transaction->t_nr_buffers <=
		 commit_transaction->t_outstanding_credits);

L
Linus Torvalds 已提交
499 500 501 502 503 504 505 506 507
	descriptor = NULL;
	bufs = 0;
	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
508
		   release it. */
L
Linus Torvalds 已提交
509 510

		if (is_journal_aborted(journal)) {
511
			clear_buffer_jbddirty(jh2bh(jh));
L
Linus Torvalds 已提交
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
			JBUFFER_TRACE(jh, "journal is aborting: refile");
			journal_refile_buffer(journal, jh);
			/* 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");

			descriptor = journal_get_descriptor_buffer(journal);
			if (!descriptor) {
535
				journal_abort(journal, -EIO);
L
Linus Torvalds 已提交
536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
				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];
			header->h_magic     = cpu_to_be32(JFS_MAGIC_NUMBER);
			header->h_blocktype = cpu_to_be32(JFS_DESCRIPTOR_BLOCK);
			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");
			journal_file_buffer(descriptor, commit_transaction,
					BJ_LogCtl);
		}

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

		err = journal_next_log_block(journal, &blocknr);
		/* 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) {
568
			journal_abort(journal, err);
L
Linus Torvalds 已提交
569 570 571 572 573 574 575 576 577 578 579 580 581
			continue;
		}

		/*
		 * start_this_handle() uses t_outstanding_credits to determine
		 * the free space in the log, but this counter is changed
		 * by journal_next_log_block() also.
		 */
		commit_transaction->t_outstanding_credits--;

		/* 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. */
582
		get_bh(jh2bh(jh));
L
Linus Torvalds 已提交
583 584 585 586 587

		/* 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*/

N
Namhyung Kim 已提交
588
		set_buffer_jwrite(jh2bh(jh));
L
Linus Torvalds 已提交
589 590 591 592 593 594 595 596 597
		/*
		 * akpm: journal_write_metadata_buffer() sets
		 * 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");
		flags = journal_write_metadata_buffer(commit_transaction,
						      jh, &new_jh, blocknr);
N
Namhyung Kim 已提交
598
		set_buffer_jwrite(jh2bh(new_jh));
L
Linus Torvalds 已提交
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
		wbuf[bufs++] = jh2bh(new_jh);

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

		tag_flag = 0;
		if (flags & 1)
			tag_flag |= JFS_FLAG_ESCAPE;
		if (!first_tag)
			tag_flag |= JFS_FLAG_SAME_UUID;

		tag = (journal_block_tag_t *) tagp;
		tag->t_blocknr = cpu_to_be32(jh2bh(jh)->b_blocknr);
		tag->t_flags = cpu_to_be32(tag_flag);
		tagp += sizeof(journal_block_tag_t);
		space_left -= sizeof(journal_block_tag_t);

		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 ||
		    space_left < sizeof(journal_block_tag_t) + 16) {

			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. */

			tag->t_flags |= cpu_to_be32(JFS_FLAG_LAST_TAG);

start_journal_io:
			for (i = 0; i < bufs; i++) {
				struct buffer_head *bh = wbuf[i];
				lock_buffer(bh);
				clear_buffer_dirty(bh);
				set_buffer_uptodate(bh);
				bh->b_end_io = journal_end_buffer_io_sync;
645
				submit_bh(WRITE_SYNC, bh);
L
Linus Torvalds 已提交
646 647 648 649 650 651 652 653 654 655
			}
			cond_resched();

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

656 657
	blk_finish_plug(&plug);

L
Linus Torvalds 已提交
658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709
	/* 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.
	*/

	jbd_debug(3, "JBD: commit phase 4\n");

	/*
	 * 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");
		journal_unfile_buffer(journal, jh);

		/*
		 * ->t_iobuf_list should contain only dummy buffer_heads
		 * which were created by journal_write_metadata_buffer().
		 */
		BUFFER_TRACE(bh, "dumping temporary bh");
		journal_put_journal_head(jh);
		__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);
N
Namhyung Kim 已提交
710
		clear_buffer_jwrite(bh);
L
Linus Torvalds 已提交
711 712 713 714 715 716 717 718
		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");
		journal_file_buffer(jh, commit_transaction, BJ_Forget);
719 720 721 722 723 724 725
		/*
		 * Wake up any transactions which were waiting for this
		 * IO to complete. The barrier must be here so that changes
		 * by journal_file_buffer() take effect before wake_up_bit()
		 * does the waitqueue check.
		 */
		smp_mb();
L
Linus Torvalds 已提交
726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759
		wake_up_bit(&bh->b_state, BH_Unshadow);
		JBUFFER_TRACE(jh, "brelse shadowed buffer");
		__brelse(bh);
	}

	J_ASSERT (commit_transaction->t_shadow_list == NULL);

	jbd_debug(3, "JBD: commit phase 5\n");

	/* 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);
		journal_unfile_buffer(journal, jh);
		journal_put_journal_head(jh);
		__brelse(bh);		/* One for getblk */
		/* AKPM: bforget here */
	}

760 761 762
	if (err)
		journal_abort(journal, err);

L
Linus Torvalds 已提交
763 764
	jbd_debug(3, "JBD: commit phase 6\n");

765 766 767 768 769 770
	/* All metadata is written, now write commit record and do cleanup */
	spin_lock(&journal->j_state_lock);
	J_ASSERT(commit_transaction->t_state == T_COMMIT);
	commit_transaction->t_state = T_COMMIT_RECORD;
	spin_unlock(&journal->j_state_lock);

L
Linus Torvalds 已提交
771 772 773 774
	if (journal_write_commit_record(journal, commit_transaction))
		err = -EIO;

	if (err)
775
		journal_abort(journal, err);
L
Linus Torvalds 已提交
776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791

	/* 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. */

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

	J_ASSERT(commit_transaction->t_sync_datalist == NULL);
	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:
792 793 794 795 796
	/*
	 * 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);
L
Linus Torvalds 已提交
797 798 799 800 801
	while (commit_transaction->t_forget) {
		transaction_t *cp_transaction;
		struct buffer_head *bh;

		jh = commit_transaction->t_forget;
802
		spin_unlock(&journal->j_list_lock);
L
Linus Torvalds 已提交
803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818
		bh = jh2bh(jh);
		jbd_lock_bh_state(bh);
		J_ASSERT_JH(jh,	jh->b_transaction == commit_transaction ||
			jh->b_transaction == journal->j_running_transaction);

		/*
		 * 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.
		 */
		if (jh->b_committed_data) {
M
Mingming Cao 已提交
819
			jbd_free(jh->b_committed_data, bh->b_size);
L
Linus Torvalds 已提交
820 821 822 823 824 825
			jh->b_committed_data = NULL;
			if (jh->b_frozen_data) {
				jh->b_committed_data = jh->b_frozen_data;
				jh->b_frozen_data = NULL;
			}
		} else if (jh->b_frozen_data) {
M
Mingming Cao 已提交
826
			jbd_free(jh->b_frozen_data, bh->b_size);
L
Linus Torvalds 已提交
827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845
			jh->b_frozen_data = NULL;
		}

		spin_lock(&journal->j_list_lock);
		cp_transaction = jh->b_cp_transaction;
		if (cp_transaction) {
			JBUFFER_TRACE(jh, "remove from old cp transaction");
			__journal_remove_checkpoint(jh);
		}

		/* Only re-checkpoint the buffer_head if it is marked
		 * dirty.  If the buffer was added to the BJ_Forget list
		 * by journal_forget, it may no longer be dirty and
		 * 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
846 847 848 849
		 * 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
L
Linus Torvalds 已提交
850
		 * use in a different page. */
851
		if (buffer_freed(bh) && !jh->b_next_transaction) {
L
Linus Torvalds 已提交
852 853 854 855 856 857 858
			clear_buffer_freed(bh);
			clear_buffer_jbddirty(bh);
		}

		if (buffer_jbddirty(bh)) {
			JBUFFER_TRACE(jh, "add to new checkpointing trans");
			__journal_insert_checkpoint(jh, commit_transaction);
859 860
			if (is_journal_aborted(journal))
				clear_buffer_jbddirty(bh);
L
Linus Torvalds 已提交
861 862 863 864 865
			JBUFFER_TRACE(jh, "refile for checkpoint writeback");
			__journal_refile_buffer(jh);
			jbd_unlock_bh_state(bh);
		} else {
			J_ASSERT_BH(bh, !buffer_dirty(bh));
866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881
			/* The buffer on BJ_Forget list and not jbddirty means
			 * 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
			 * list. */
			JBUFFER_TRACE(jh, "refile or unfile freed buffer");
			__journal_refile_buffer(jh);
			if (!jh->b_transaction) {
				jbd_unlock_bh_state(bh);
				 /* needs a brelse */
				journal_remove_journal_head(bh);
				release_buffer_page(bh);
			} else
				jbd_unlock_bh_state(bh);
L
Linus Torvalds 已提交
882
		}
883 884 885 886
		cond_resched_lock(&journal->j_list_lock);
	}
	spin_unlock(&journal->j_list_lock);
	/*
887 888 889 890
	 * This is a bit sleazy.  We use j_list_lock to protect transition
	 * of a transaction into T_FINISHED state and calling
	 * __journal_drop_transaction(). Otherwise we could race with
	 * other checkpointing code processing the transaction...
891 892 893 894 895 896 897 898
	 */
	spin_lock(&journal->j_state_lock);
	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) {
L
Linus Torvalds 已提交
899
		spin_unlock(&journal->j_list_lock);
900 901
		spin_unlock(&journal->j_state_lock);
		goto restart_loop;
L
Linus Torvalds 已提交
902 903 904 905 906 907
	}

	/* Done with this transaction! */

	jbd_debug(3, "JBD: commit phase 8\n");

908
	J_ASSERT(commit_transaction->t_state == T_COMMIT_RECORD);
L
Linus Torvalds 已提交
909 910 911 912 913

	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
914 915 916 917 918 919 920 921 922 923 924 925
	commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));

	/*
	 * weight the commit time higher than the average time so we don't
	 * react too strongly to vast changes in commit time
	 */
	if (likely(journal->j_average_commit_time))
		journal->j_average_commit_time = (commit_time*3 +
				journal->j_average_commit_time) / 4;
	else
		journal->j_average_commit_time = commit_time;

L
Linus Torvalds 已提交
926 927
	spin_unlock(&journal->j_state_lock);

J
Jan Kara 已提交
928 929
	if (commit_transaction->t_checkpoint_list == NULL &&
	    commit_transaction->t_checkpoint_io_list == NULL) {
L
Linus Torvalds 已提交
930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
		__journal_drop_transaction(journal, commit_transaction);
	} 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);

	jbd_debug(1, "JBD: commit %d complete, head %d\n",
		  journal->j_commit_sequence, journal->j_tail_sequence);

	wake_up(&journal->j_wait_done_commit);
}