checkpoint.c

来自「Linux Kernel 2.6.9 for OMAP1710」· C语言 代码 · 共 626 行 · 第 1/2 页

/*
 * linux/fs/checkpoint.c
 * 
 * Written by Stephen C. Tweedie <[e-mail]>, 1999
 *
 * Copyright 1999 Red Hat Software --- 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.
 *
 * Checkpoint routines for the generic filesystem journaling code.  
 * Part of the ext2fs journaling system.  
 *
 * Checkpointing is the process of ensuring that a section of the log is
 * committed fully to disk, so that that portion of the log can be
 * reused.
 */

#include <linux/time.h>
#include <linux/fs.h>
#include <linux/jbd.h>
#include <linux/errno.h>
#include <linux/slab.h>

/*
 * Unlink a buffer from a transaction. 
 *
 * Called with j_list_lock held.
 */

static inline void __buffer_unlink(struct journal_head *jh)
{
	transaction_t *transaction;

	transaction = jh->b_cp_transaction;
	jh->b_cp_transaction = NULL;

	jh->b_cpnext->b_cpprev = jh->b_cpprev;
	jh->b_cpprev->b_cpnext = jh->b_cpnext;
	if (transaction->t_checkpoint_list == jh)
		transaction->t_checkpoint_list = jh->b_cpnext;
	if (transaction->t_checkpoint_list == jh)
		transaction->t_checkpoint_list = NULL;
}

/*
 * Try to release a checkpointed buffer from its transaction.
 * Returns 1 if we released it.
 * Requires j_list_lock
 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
 */
static int __try_to_free_cp_buf(struct journal_head *jh)
{
	int ret = 0;
	struct buffer_head *bh = jh2bh(jh);

	if (jh->b_jlist == BJ_None && !buffer_locked(bh) && !buffer_dirty(bh)) {
		JBUFFER_TRACE(jh, "remove from checkpoint list");
		__journal_remove_checkpoint(jh);
		jbd_unlock_bh_state(bh);
		journal_remove_journal_head(bh);
		BUFFER_TRACE(bh, "release");
		__brelse(bh);
		ret = 1;
	} else {
		jbd_unlock_bh_state(bh);
	}
	return ret;
}

/*
 * __log_wait_for_space: wait until there is space in the journal.
 *
 * Called under j-state_lock *only*.  It will be unlocked if we have to wait
 * for a checkpoint to free up some space in the log.
 */
void __log_wait_for_space(journal_t *journal)
{
	int nblocks;
	assert_spin_locked(&journal->j_state_lock);

	nblocks = jbd_space_needed(journal);
	while (__log_space_left(journal) < nblocks) {
		if (journal->j_flags & JFS_ABORT)
			return;
		spin_unlock(&journal->j_state_lock);
		down(&journal->j_checkpoint_sem);

		/*
		 * Test again, another process may have checkpointed while we
		 * were waiting for the checkpoint lock
		 */
		spin_lock(&journal->j_state_lock);
		nblocks = jbd_space_needed(journal);
		if (__log_space_left(journal) < nblocks) {
			spin_unlock(&journal->j_state_lock);
			log_do_checkpoint(journal);
			spin_lock(&journal->j_state_lock);
		}
		up(&journal->j_checkpoint_sem);
	}
}

/*
 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
 * The caller must restart a list walk.  Wait for someone else to run
 * jbd_unlock_bh_state().
 */
static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh)
{
	get_bh(bh);
	spin_unlock(&journal->j_list_lock);
	jbd_lock_bh_state(bh);
	jbd_unlock_bh_state(bh);
	put_bh(bh);
}

/*
 * Clean up a transaction's checkpoint list.  
 *
 * We wait for any pending IO to complete and make sure any clean
 * buffers are removed from the transaction. 
 *
 * Return 1 if we performed any actions which might have destroyed the
 * checkpoint.  (journal_remove_checkpoint() deletes the transaction when
 * the last checkpoint buffer is cleansed)
 *
 * Called with j_list_lock held.
 */
static int __cleanup_transaction(journal_t *journal, transaction_t *transaction)
{
	struct journal_head *jh, *next_jh, *last_jh;
	struct buffer_head *bh;
	int ret = 0;

	assert_spin_locked(&journal->j_list_lock);
	jh = transaction->t_checkpoint_list;
	if (!jh)
		return 0;

	last_jh = jh->b_cpprev;
	next_jh = jh;
	do {
		jh = next_jh;
		bh = jh2bh(jh);
		if (buffer_locked(bh)) {
			atomic_inc(&bh->b_count);
			spin_unlock(&journal->j_list_lock);
			wait_on_buffer(bh);
			/* the journal_head may have gone by now */
			BUFFER_TRACE(bh, "brelse");
			__brelse(bh);
			goto out_return_1;
		}

		/*
		 * This is foul
		 */
		if (!jbd_trylock_bh_state(bh)) {
			jbd_sync_bh(journal, bh);
			goto out_return_1;
		}

		if (jh->b_transaction != NULL) {
			transaction_t *t = jh->b_transaction;
			tid_t tid = t->t_tid;

			spin_unlock(&journal->j_list_lock);
			jbd_unlock_bh_state(bh);
			log_start_commit(journal, tid);
			log_wait_commit(journal, tid);
			goto out_return_1;
		}

		/*
		 * AKPM: I think the buffer_jbddirty test is redundant - it
		 * shouldn't have NULL b_transaction?
		 */
		next_jh = jh->b_cpnext;
		if (!buffer_dirty(bh) && !buffer_jbddirty(bh)) {
			BUFFER_TRACE(bh, "remove from checkpoint");
			__journal_remove_checkpoint(jh);
			jbd_unlock_bh_state(bh);
			journal_remove_journal_head(bh);
			__brelse(bh);
			ret = 1;
		} else {
			jbd_unlock_bh_state(bh);
		}
		jh = next_jh;
	} while (jh != last_jh);

	return ret;
out_return_1:
	spin_lock(&journal->j_list_lock);
	return 1;
}

#define NR_BATCH	64

static void
__flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count)
{
	int i;

	spin_unlock(&journal->j_list_lock);
	ll_rw_block(WRITE, *batch_count, bhs);
	spin_lock(&journal->j_list_lock);
	for (i = 0; i < *batch_count; i++) {
		struct buffer_head *bh = bhs[i];
		clear_buffer_jwrite(bh);
		BUFFER_TRACE(bh, "brelse");
		__brelse(bh);
	}
	*batch_count = 0;
}

/*
 * Try to flush one buffer from the checkpoint list to disk.
 *
 * Return 1 if something happened which requires us to abort the current
 * scan of the checkpoint list.  
 *
 * Called with j_list_lock held.
 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
 */
static int __flush_buffer(journal_t *journal, struct journal_head *jh,
			struct buffer_head **bhs, int *batch_count,
			int *drop_count)
{
	struct buffer_head *bh = jh2bh(jh);
	int ret = 0;

	if (buffer_dirty(bh) && !buffer_locked(bh) && jh->b_jlist == BJ_None) {
		J_ASSERT_JH(jh, jh->b_transaction == NULL);

		/*
		 * Important: we are about to write the buffer, and
		 * possibly block, while still holding the journal lock.
		 * We cannot afford to let the transaction logic start
		 * messing around with this buffer before we write it to
		 * disk, as that would break recoverability.  
		 */
		BUFFER_TRACE(bh, "queue");
		get_bh(bh);
		J_ASSERT_BH(bh, !buffer_jwrite(bh));
		set_buffer_jwrite(bh);
		bhs[*batch_count] = bh;
		jbd_unlock_bh_state(bh);
		(*batch_count)++;
		if (*batch_count == NR_BATCH) {
			__flush_batch(journal, bhs, batch_count);
			ret = 1;
		}
	} else {
		int last_buffer = 0;
		if (jh->b_cpnext == jh) {
			/* We may be about to drop the transaction.  Tell the
			 * caller that the lists have changed.
			 */
			last_buffer = 1;
		}
		if (__try_to_free_cp_buf(jh)) {
			(*drop_count)++;
			ret = last_buffer;
		}
	}
	return ret;
}

/*
 * Perform an actual checkpoint.  We don't write out only enough to
 * satisfy the current blocked requests: rather we submit a reasonably
 * sized chunk of the outstanding data to disk at once for
 * efficiency.  __log_wait_for_space() will retry if we didn't free enough.
 * 
 * However, we _do_ take into account the amount requested so that once
 * the IO has been queued, we can return as soon as enough of it has
 * completed to disk.  
 *
 * The journal should be locked before calling this function.
 */
int log_do_checkpoint(journal_t *journal)
{
	int result;
	int batch_count = 0;
	struct buffer_head *bhs[NR_BATCH];

	jbd_debug(1, "Start checkpoint\n");

	/* 
	 * First thing: if there are any transactions in the log which
	 * don't need checkpointing, just eliminate them from the
	 * journal straight away.  
	 */
	result = cleanup_journal_tail(journal);
	jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
	if (result <= 0)
		return result;

	/*
	 * OK, we need to start writing disk blocks.  Try to free up a
	 * quarter of the log in a single checkpoint if we can.
	 */
	/*
	 * AKPM: check this code.  I had a feeling a while back that it
	 * degenerates into a busy loop at unmount time.
	 */
	spin_lock(&journal->j_list_lock);
	while (journal->j_checkpoint_transactions) {
		transaction_t *transaction;
		struct journal_head *jh, *last_jh, *next_jh;