Commit 7fcd3efa authored by Christoph Hellwig's avatar Christoph Hellwig Committed by Darrick J. Wong
Browse files

xfs: remove filestream item xfs_inode reference 



The filestreams allocator stores an xfs_fstrm_item structure in the MRU to
cache inode number to agno mappings for a particular length of time.  Each
xfs_fstrm_item contains the internal MRU structure, an inode pointer and
agno value.

The inode pointer stored in the xfs_fstrm_item is not referenced, however,
which means the inode itself can be removed and reclaimed before the MRU
item is freed. If this occurs, xfs_fstrm_free_func() can access freed or
unrelated memory through xfs_fstrm_item->ip and crash.

The obvious solution is to grab an inode reference for xfs_fstrm_item.
The filestream mechanism only actually uses the inode pointer as a means
to access the xfs_mount, however.  Rather than add unnecessary
complexity, simplify the implementation to store an xfs_mount pointer in
struct xfs_mru_cache, and pass it to the free callback.  This also
requires updates to the tracepoint class to provide the associated data
via parameters rather than the inode and a minor hack to peek at the MRU
key to establish the inode number at free time.

Based on debugging work and an earlier patch from Brian Foster, who
also wrote most of this changelog.

Reported-by: default avatarBrian Foster <bfoster@redhat.com>
Signed-off-by: default avatarChristoph Hellwig <hch@lst.de>
Reviewed-by: default avatarBrian Foster <bfoster@redhat.com>
Reviewed-by: default avatarDarrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: default avatarDarrick J. Wong <darrick.wong@oracle.com>
parent 0612d116

fs/xfs/xfs_filestream.c

+9 −10
Original line number Diff line number Diff line
@@ -34,7 +34,6 @@ 


struct xfs_fstrm_item {

	struct xfs_mru_cache_elem	mru;

	struct xfs_inode		*ip;

	xfs_agnumber_t			ag; /* AG in use for this directory */

};
@@ -122,14 +121,15 @@ xfs_filestream_put_ag( 


static void

xfs_fstrm_free_func(

	void			*data,

	struct xfs_mru_cache_elem *mru)

{

	struct xfs_mount	*mp = data;

	struct xfs_fstrm_item	*item =

		container_of(mru, struct xfs_fstrm_item, mru);



	xfs_filestream_put_ag(item->ip->i_mount, item->ag);



	trace_xfs_filestream_free(item->ip, item->ag);

	xfs_filestream_put_ag(mp, item->ag);

	trace_xfs_filestream_free(mp, mru->key, item->ag);



	kmem_free(item);

}
@@ -165,7 +165,7 @@ xfs_filestream_pick_ag( 
	trylock = XFS_ALLOC_FLAG_TRYLOCK;



	for (nscan = 0; 1; nscan++) {

		trace_xfs_filestream_scan(ip, ag);

		trace_xfs_filestream_scan(mp, ip->i_ino, ag);



		pag = xfs_perag_get(mp, ag);
@@ -265,7 +265,6 @@ next_ag: 
		goto out_put_ag;



	item->ag = *agp;

	item->ip = ip;



	err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);

	if (err) {
@@ -333,7 +332,7 @@ xfs_filestream_lookup_ag( 
		ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;

		xfs_mru_cache_done(mp->m_filestream);



		trace_xfs_filestream_lookup(ip, ag);

		trace_xfs_filestream_lookup(mp, ip->i_ino, ag);

		goto out;

	}
@@ -399,7 +398,7 @@ xfs_filestream_new_ag( 
	 * Only free the item here so we skip over the old AG earlier.

	 */

	if (mru)

		xfs_fstrm_free_func(mru);

		xfs_fstrm_free_func(mp, mru);



	IRELE(pip);

exit:
@@ -426,8 +425,8 @@ xfs_filestream_mount( 
	 * timer tunable to within about 10 percent.  This requires at least 10

	 * groups.

	 */

	return xfs_mru_cache_create(&mp->m_filestream, xfs_fstrm_centisecs * 10,

				    10, xfs_fstrm_free_func);

	return xfs_mru_cache_create(&mp->m_filestream, mp,

			xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);

}



void

fs/xfs/xfs_mru_cache.c

+5 −3
Original line number Diff line number Diff line
@@ -112,6 +112,7 @@ struct xfs_mru_cache { 
	xfs_mru_cache_free_func_t free_func; /* Function pointer for freeing. */

	struct delayed_work	work;      /* Workqueue data for reaping.   */

	unsigned int		queued;	   /* work has been queued */

	void			*data;

};



static struct workqueue_struct	*xfs_mru_reap_wq;
@@ -259,7 +260,7 @@ _xfs_mru_cache_clear_reap_list( 


	list_for_each_entry_safe(elem, next, &tmp, list_node) {

		list_del_init(&elem->list_node);

		mru->free_func(elem);

		mru->free_func(mru->data, elem);

	}



	spin_lock(&mru->lock);
@@ -326,6 +327,7 @@ xfs_mru_cache_uninit(void) 
int

xfs_mru_cache_create(

	struct xfs_mru_cache	**mrup,

	void			*data,

	unsigned int		lifetime_ms,

	unsigned int		grp_count,

	xfs_mru_cache_free_func_t free_func)
@@ -369,7 +371,7 @@ xfs_mru_cache_create( 


	mru->grp_time  = grp_time;

	mru->free_func = free_func;



	mru->data = data;

	*mrup = mru;



exit:
@@ -492,7 +494,7 @@ xfs_mru_cache_delete( 


	elem = xfs_mru_cache_remove(mru, key);

	if (elem)

		mru->free_func(elem);

		mru->free_func(mru->data, elem);

}



/*

fs/xfs/xfs_mru_cache.h

+4 −4
Original line number Diff line number Diff line
@@ -26,12 +26,12 @@ struct xfs_mru_cache_elem { 
};



/* Function pointer type for callback to free a client's data pointer. */

typedef void (*xfs_mru_cache_free_func_t)(struct xfs_mru_cache_elem *elem);

typedef void (*xfs_mru_cache_free_func_t)(void *, struct xfs_mru_cache_elem *);



int xfs_mru_cache_init(void);

void xfs_mru_cache_uninit(void);

int xfs_mru_cache_create(struct xfs_mru_cache **mrup, unsigned int lifetime_ms,

			     unsigned int grp_count,

int xfs_mru_cache_create(struct xfs_mru_cache **mrup, void *data,

		unsigned int lifetime_ms, unsigned int grp_count,

		xfs_mru_cache_free_func_t free_func);

void xfs_mru_cache_destroy(struct xfs_mru_cache *mru);

int xfs_mru_cache_insert(struct xfs_mru_cache *mru, unsigned long key,

fs/xfs/xfs_trace.h

+7 −7
Original line number Diff line number Diff line
@@ -506,8 +506,8 @@ DEFINE_BUF_ITEM_EVENT(xfs_trans_bhold_release); 
DEFINE_BUF_ITEM_EVENT(xfs_trans_binval);



DECLARE_EVENT_CLASS(xfs_filestream_class,

	TP_PROTO(struct xfs_inode *ip, xfs_agnumber_t agno),

	TP_ARGS(ip, agno),

	TP_PROTO(struct xfs_mount *mp, xfs_ino_t ino, xfs_agnumber_t agno),

	TP_ARGS(mp, ino, agno),

	TP_STRUCT__entry(

		__field(dev_t, dev)

		__field(xfs_ino_t, ino)
@@ -515,10 +515,10 @@ DECLARE_EVENT_CLASS(xfs_filestream_class, 
		__field(int, streams)

	),

	TP_fast_assign(

		__entry->dev = VFS_I(ip)->i_sb->s_dev;

		__entry->ino = ip->i_ino;

		__entry->dev = mp->m_super->s_dev;

		__entry->ino = ino;

		__entry->agno = agno;

		__entry->streams = xfs_filestream_peek_ag(ip->i_mount, agno);

		__entry->streams = xfs_filestream_peek_ag(mp, agno);

	),

	TP_printk("dev %d:%d ino 0x%llx agno %u streams %d",

		  MAJOR(__entry->dev), MINOR(__entry->dev),
@@ -528,8 +528,8 @@ DECLARE_EVENT_CLASS(xfs_filestream_class, 
)

#define DEFINE_FILESTREAM_EVENT(name) \

DEFINE_EVENT(xfs_filestream_class, name, \

	TP_PROTO(struct xfs_inode *ip, xfs_agnumber_t agno), \

	TP_ARGS(ip, agno))

	TP_PROTO(struct xfs_mount *mp, xfs_ino_t ino, xfs_agnumber_t agno), \

	TP_ARGS(mp, ino, agno))

DEFINE_FILESTREAM_EVENT(xfs_filestream_free);

DEFINE_FILESTREAM_EVENT(xfs_filestream_lookup);

DEFINE_FILESTREAM_EVENT(xfs_filestream_scan);

CRA Git | Maintained and supported by SUSTech CRA and CCSE