Commit b409d7a0 authored by Jan Kara's avatar Jan Kara Committed by Joel Becker
Browse files

ocfs2: Fix possible deadlock when extending quota file 



In OCFS2, allocator locks rank above transaction start. Thus we
cannot extend quota file from inside a transaction less we could
deadlock.

We solve the problem by starting transaction not already in
ocfs2_acquire_dquot() but only in ocfs2_local_read_dquot() and
ocfs2_global_read_dquot() and we allocate blocks to quota files before starting
the transaction.  In case we crash, quota files will just have a few blocks
more but that's no problem since we just use them next time we extend the
quota file.

Signed-off-by: default avatarJan Kara <jack@suse.cz>
Signed-off-by: default avatarJoel Becker <joel.becker@oracle.com>
parent 8a57a9dd

fs/ocfs2/journal.h

+0 −5
Original line number Diff line number Diff line
@@ -363,11 +363,6 @@ static inline int ocfs2_quota_trans_credits(struct super_block *sb) 
	return credits;

}



/* Number of credits needed for removing quota structure from file */

int ocfs2_calc_qdel_credits(struct super_block *sb, int type);

/* Number of credits needed for initialization of new quota structure */

int ocfs2_calc_qinit_credits(struct super_block *sb, int type);



/* group extend. inode update and last group update. */

#define OCFS2_GROUP_EXTEND_CREDITS	(OCFS2_INODE_UPDATE_CREDITS + 1)

fs/ocfs2/quota_global.c

+57 −58
Original line number Diff line number Diff line
@@ -215,11 +215,7 @@ ssize_t ocfs2_quota_write(struct super_block *sb, int type, 
		loff_t rounded_end =

				ocfs2_align_bytes_to_blocks(sb, off + len);



		down_write(&OCFS2_I(gqinode)->ip_alloc_sem);

		err = ocfs2_extend_no_holes(gqinode, rounded_end, off);

		up_write(&OCFS2_I(gqinode)->ip_alloc_sem);

		if (err < 0)

			goto out;

		/* Space is already allocated in ocfs2_global_read_dquot() */

		err = ocfs2_simple_size_update(gqinode,

					       oinfo->dqi_gqi_bh,

					       rounded_end);
@@ -405,13 +401,36 @@ int ocfs2_global_write_info(struct super_block *sb, int type) 
	return err;

}



static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)

{

	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;



	/*

	 * We may need to allocate tree blocks and a leaf block but not the

	 * root block

	 */

	return oinfo->dqi_gi.dqi_qtree_depth;

}



static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)

{

	/* We modify all the allocated blocks, tree root, and info block */

	return (ocfs2_global_qinit_alloc(sb, type) + 2) *

			OCFS2_QUOTA_BLOCK_UPDATE_CREDITS;

}



/* Read in information from global quota file and acquire a reference to it.

 * dquot_acquire() has already started the transaction and locked quota file */

int ocfs2_global_read_dquot(struct dquot *dquot)

{

	int err, err2, ex = 0;

	struct ocfs2_mem_dqinfo *info =

			sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;

	struct super_block *sb = dquot->dq_sb;

	int type = dquot->dq_type;

	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;

	struct ocfs2_super *osb = OCFS2_SB(sb);

	struct inode *gqinode = info->dqi_gqinode;

	int need_alloc = ocfs2_global_qinit_alloc(sb, type);

	handle_t *handle = NULL;



	err = ocfs2_qinfo_lock(info, 0);

	if (err < 0)
@@ -422,14 +441,33 @@ int ocfs2_global_read_dquot(struct dquot *dquot) 
	OCFS2_DQUOT(dquot)->dq_use_count++;

	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;

	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;

	if (!dquot->dq_off) {	/* No real quota entry? */

		/* Upgrade to exclusive lock for allocation */

	ocfs2_qinfo_unlock(info, 0);

		err = ocfs2_qinfo_lock(info, 1);

		if (err < 0)

			goto out_qlock;



	if (!dquot->dq_off) {	/* No real quota entry? */

		ex = 1;

		/*

		 * Add blocks to quota file before we start a transaction since

		 * locking allocators ranks above a transaction start

		 */

		WARN_ON(journal_current_handle());

		down_write(&OCFS2_I(gqinode)->ip_alloc_sem);

		err = ocfs2_extend_no_holes(gqinode,

			gqinode->i_size + (need_alloc << sb->s_blocksize_bits),

			gqinode->i_size);

		up_write(&OCFS2_I(gqinode)->ip_alloc_sem);

		if (err < 0)

			goto out;

	}



	handle = ocfs2_start_trans(osb,

				   ocfs2_calc_global_qinit_credits(sb, type));

	if (IS_ERR(handle)) {

		err = PTR_ERR(handle);

		goto out;

	}

	err = ocfs2_qinfo_lock(info, ex);

	if (err < 0)

		goto out_trans;

	err = qtree_write_dquot(&info->dqi_gi, dquot);

	if (ex && info_dirty(sb_dqinfo(dquot->dq_sb, dquot->dq_type))) {

		err2 = __ocfs2_global_write_info(dquot->dq_sb, dquot->dq_type);
@@ -441,6 +479,9 @@ out_qlock: 
		ocfs2_qinfo_unlock(info, 1);

	else

		ocfs2_qinfo_unlock(info, 0);

out_trans:

	if (handle)

		ocfs2_commit_trans(osb, handle);

out:

	if (err < 0)

		mlog_errno(err);
@@ -638,24 +679,17 @@ out: 
	return status;

}



int ocfs2_calc_qdel_credits(struct super_block *sb, int type)

static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)

{

	struct ocfs2_mem_dqinfo *oinfo;

	int features[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,

				    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA };



	if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, features[type]))

		return 0;



	oinfo = sb_dqinfo(sb, type)->dqi_priv;

	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;

	/*

	 * We modify tree, leaf block, global info, local chunk header,

	 * global and local inode; OCFS2_QINFO_WRITE_CREDITS already

	 * accounts for inode update

	 */

	return oinfo->dqi_gi.dqi_qtree_depth +

	return (oinfo->dqi_gi.dqi_qtree_depth + 2) *

	       OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +

	       OCFS2_QINFO_WRITE_CREDITS +

	       2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +

	       OCFS2_INODE_UPDATE_CREDITS;

}
@@ -688,36 +722,10 @@ out: 
	return status;

}



int ocfs2_calc_qinit_credits(struct super_block *sb, int type)

{

	struct ocfs2_mem_dqinfo *oinfo;

	int features[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,

				    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA };

	struct ocfs2_dinode *lfe, *gfe;



	if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, features[type]))

		return 0;



	oinfo = sb_dqinfo(sb, type)->dqi_priv;

	gfe = (struct ocfs2_dinode *)oinfo->dqi_gqi_bh->b_data;

	lfe = (struct ocfs2_dinode *)oinfo->dqi_lqi_bh->b_data;

	/* We can extend local file + global file. In local file we

	 * can modify info, chunk header block and dquot block. In

	 * global file we can modify info, tree and leaf block */

	return ocfs2_calc_extend_credits(sb, &lfe->id2.i_list, 0) +

	       ocfs2_calc_extend_credits(sb, &gfe->id2.i_list, 0) +

	       OCFS2_LOCAL_QINFO_WRITE_CREDITS +

	       2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +

	       oinfo->dqi_gi.dqi_qtree_depth +

	       2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS;

}



static int ocfs2_acquire_dquot(struct dquot *dquot)

{

	handle_t *handle;

	struct ocfs2_mem_dqinfo *oinfo =

			sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;

	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);

	int status = 0;



	mlog_entry("id=%u, type=%d", dquot->dq_id, dquot->dq_type);
@@ -726,16 +734,7 @@ static int ocfs2_acquire_dquot(struct dquot *dquot) 
	status = ocfs2_lock_global_qf(oinfo, 1);

	if (status < 0)

		goto out;

	handle = ocfs2_start_trans(osb,

		ocfs2_calc_qinit_credits(dquot->dq_sb, dquot->dq_type));

	if (IS_ERR(handle)) {

		status = PTR_ERR(handle);

		mlog_errno(status);

		goto out_ilock;

	}

	status = dquot_acquire(dquot);

	ocfs2_commit_trans(osb, handle);

out_ilock:

	ocfs2_unlock_global_qf(oinfo, 1);

out:

	mlog_exit(status);

CRA Git | Maintained and supported by SUSTech CRA and CCSE