ext4: retry allocations conservatively (68fd9750) · Commits · 戴 / test

Now that we no longer try to reserve metadata blocks for delayed allocations (which tended to overestimate the required number of blocks significantly), we really don't need retry allocations when the disk is very full as aggressively any more. The only time when it makes sense to retry an allocation is if we have freshly deleted blocks that will only become available after a transaction commit. And if we lose that race, it's not worth it to try more than once. Signed-off-by:

Theodore Ts'o <tytso@mit.edu>

fs/ext4/balloc.c

+7 −8

Original line number	Diff line number	Diff line
		@@ -600,21 +600,20 @@ int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
		* ext4_should_retry_alloc() is called when ENOSPC is returned, and if
		* it is profitable to retry the operation, this function will wait
		* for the current or committing transaction to complete, and then
		* return TRUE.
		*
		* if the total number of retries exceed three times, return FALSE.
		* return TRUE. We will only retry once.
		*/
		int ext4_should_retry_alloc(struct super_block sb, int retries)
		{
		if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) \|\|
		(*retries)++ > 3 \|\|
		(*retries)++ > 1 \|\|
		!EXT4_SB(sb)->s_journal)
		return 0;

		jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);

		smp_mb();
		if (EXT4_SB(sb)->s_mb_free_pending)
		if (EXT4_SB(sb)->s_mb_free_pending == 0)
		return 0;

		jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
		jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
		return 1;
		}

Admin message