OSDN Git Service

xfs: clear _XBF_PAGES from buffers when readahead page
authorDarrick J. Wong <darrick.wong@oracle.com>
Thu, 2 Feb 2017 07:56:10 +0000 (08:56 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 4 Feb 2017 08:47:12 +0000 (09:47 +0100)
commit5d44dd54bd57c6275d82d8912730c794fc8ec8ab
tree2e3725e2ec566b46489404b92219435e31622e58
parent29f96b7e9023929f8bd80b5e1f53d4e6db3c434f
xfs: clear _XBF_PAGES from buffers when readahead page

commit 2aa6ba7b5ad3189cc27f14540aa2f57f0ed8df4b upstream.

If we try to allocate memory pages to back an xfs_buf that we're trying
to read, it's possible that we'll be so short on memory that the page
allocation fails.  For a blocking read we'll just wait, but for
readahead we simply dump all the pages we've collected so far.

Unfortunately, after dumping the pages we neglect to clear the
_XBF_PAGES state, which means that the subsequent call to xfs_buf_free
thinks that b_pages still points to pages we own.  It then double-frees
the b_pages pages.

This results in screaming about negative page refcounts from the memory
manager, which xfs oughtn't be triggering.  To reproduce this case,
mount a filesystem where the size of the inodes far outweighs the
availalble memory (a ~500M inode filesystem on a VM with 300MB memory
did the trick here) and run bulkstat in parallel with other memory
eating processes to put a huge load on the system.  The "check summary"
phase of xfs_scrub also works for this purpose.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/xfs/xfs_buf.c