commit.c

嵌入式系统设计与实例开发源码

/*
 * linux/fs/commit.c
 *
 * Written by Stephen C. Tweedie <[e-mail]>, 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/sched.h>
#include <linux/fs.h>
#include <linux/jbd.h>
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/locks.h>
#include <linux/smp_lock.h>

extern spinlock_t journal_datalist_lock;

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

/*
 * 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 journal_head *next_jh, *last_jh;
	struct buffer_head *wbuf[64];
	int bufs;
	int flags;
	int err;
	unsigned long blocknr;
	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;

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

	lock_journal(journal); /* Protect journal->j_running_transaction */

#ifdef COMMIT_STATS
	spin_lock(&journal_datalist_lock);
	summarise_journal_usage(journal);
	spin_unlock(&journal_datalist_lock);
#endif

	lock_kernel();
	
	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);

	commit_transaction->t_state = T_LOCKED;
	while (commit_transaction->t_updates != 0) {
		unlock_journal(journal);
		sleep_on(&journal->j_wait_updates);
		lock_journal(journal);
	}

	J_ASSERT (commit_transaction->t_outstanding_credits <=
			journal->j_max_transaction_buffers);

	/* 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");
	}

	/*
	 * 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
	 * buffer are perfectly permissable.
	 */

	while (commit_transaction->t_reserved_list) {
		jh = commit_transaction->t_reserved_list;
		JBUFFER_TRACE(jh, "reserved, unused: refile");
		journal_refile_buffer(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_datalist_lock);
	__journal_clean_checkpoint_list(journal);
	spin_unlock(&journal_datalist_lock);

	/* First part of the commit: force the revoke list out to disk.
	 * The revoke code generates its own metadata blocks on disk for this.
	 *
	 * It is important that we do this while the transaction is
	 * still locked.  Generating the revoke records should not
	 * generate any IO stalls, so this should be quick; and doing
	 * the work while we have the transaction locked means that we
	 * only ever have to maintain the revoke list for one
	 * transaction at a time.
	 */

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

	journal_write_revoke_records(journal, commit_transaction);

	/*
	 * Now that we have built the revoke records, we can start
	 * reusing the revoke list for a new running transaction.  We
	 * can now safely start committing the old transaction: time to
	 * get a new running transaction for incoming filesystem updates
	 */

	commit_transaction->t_state = T_FLUSH;

	wake_up(&journal->j_wait_transaction_locked);

	journal->j_committing_transaction = commit_transaction;
	journal->j_running_transaction = NULL;

	commit_transaction->t_log_start = journal->j_head;

	unlock_kernel();
	
	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.
	 */

	/*
	 * Whenever we unlock the journal and sleep, things can get added
	 * onto ->t_datalist, so we have to keep looping back to write_out_data
	 * until we *know* that the list is empty.
	 */
write_out_data:

	/*
	 * Cleanup any flushed data buffers from the data list.  Even in
	 * abort mode, we want to flush this out as soon as possible.
	 *
	 * We take journal_datalist_lock to protect the lists from
	 * journal_try_to_free_buffers().
	 */
	spin_lock(&journal_datalist_lock);

write_out_data_locked:
	bufs = 0;
	next_jh = commit_transaction->t_sync_datalist;
	if (next_jh == NULL)
		goto sync_datalist_empty;
	last_jh = next_jh->b_tprev;

	do {
		struct buffer_head *bh;

		jh = next_jh;
		next_jh = jh->b_tnext;
		bh = jh2bh(jh);
		if (!buffer_locked(bh)) {
			if (buffer_dirty(bh)) {
				BUFFER_TRACE(bh, "start journal writeout");
				atomic_inc(&bh->b_count);
				wbuf[bufs++] = bh;
			} else {
				BUFFER_TRACE(bh, "writeout complete: unfile");
				__journal_unfile_buffer(jh);
				jh->b_transaction = NULL;
				__journal_remove_journal_head(bh);
				refile_buffer(bh);
				__brelse(bh);
			}
		}
		if (bufs == ARRAY_SIZE(wbuf)) {
			/*
			 * Major speedup: start here on the next scan
			 */
			J_ASSERT(commit_transaction->t_sync_datalist != 0);
			commit_transaction->t_sync_datalist = jh;
			break;
		}
	} while (jh != last_jh);

	if (bufs || current->need_resched) {
		jbd_debug(2, "submit %d writes\n", bufs);
		spin_unlock(&journal_datalist_lock);
		unlock_journal(journal);
		if (bufs)
			ll_rw_block(WRITE, bufs, wbuf);
		if (current->need_resched)
			schedule();
		journal_brelse_array(wbuf, bufs);
		lock_journal(journal);
		spin_lock(&journal_datalist_lock);
		if (bufs)
			goto write_out_data_locked;
	}

	/*
	 * Wait for all previously submitted IO on the data list to complete.
	 */
	jh = commit_transaction->t_sync_datalist;
	if (jh == NULL)
		goto sync_datalist_empty;

	do {
		struct buffer_head *bh;
		jh = jh->b_tprev;	/* Wait on the last written */
		bh = jh2bh(jh);
		if (buffer_locked(bh)) {
			spin_unlock(&journal_datalist_lock);
			unlock_journal(journal);
			wait_on_buffer(bh);
			/* the journal_head may have been removed now */
			lock_journal(journal);
			goto write_out_data;
		} else if (buffer_dirty(bh)) {
			goto write_out_data_locked;
		}
	} while (jh != commit_transaction->t_sync_datalist);
	goto write_out_data_locked;

sync_datalist_empty:
	/*
	 * Wait for all the async writepage data.  As they become unlocked
	 * in end_buffer_io_async(), the only place where they can be
	 * reaped is in try_to_free_buffers(), and we're locked against
	 * that.
	 */
	while ((jh = commit_transaction->t_async_datalist)) {
		struct buffer_head *bh = jh2bh(jh);
		if (buffer_locked(bh)) {
			spin_unlock(&journal_datalist_lock);
			unlock_journal(journal);
			wait_on_buffer(bh);
			lock_journal(journal);
			spin_lock(&journal_datalist_lock);
			continue;	/* List may have changed */
		}
		if (jh->b_next_transaction) {
			/*
			 * For writepage() buffers in journalled data mode: a
			 * later transaction may want the buffer for "metadata"
			 */
			__journal_refile_buffer(jh);
		} else {
			BUFFER_TRACE(bh, "finished async writeout: unfile");
			__journal_unfile_buffer(jh);
			jh->b_transaction = NULL;
			__journal_remove_journal_head(bh);
			BUFFER_TRACE(bh, "finished async writeout: refile");
			/* It can sometimes be on BUF_LOCKED due to migration
			 * from syncdata to asyncdata */
			if (bh->b_list != BUF_CLEAN)
				refile_buffer(bh);
			__brelse(bh);
		}
	}
	spin_unlock(&journal_datalist_lock);

	/*
	 * 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);
	J_ASSERT (commit_transaction->t_async_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:
	 */
	commit_transaction->t_state = T_COMMIT;

	descriptor = 0;
	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
		   release it for background writing. */

		if (is_journal_aborted(journal)) {
			JBUFFER_TRACE(jh, "journal is aborting: refile");
			journal_refile_buffer(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) {
				__journal_abort_hard(journal);
				continue;
			}