A cleancache backend must be put-put-get coherent, meaning that if two successive puts store AAA and BBB at the same offset, the following get must never return AAA. Tcache conforms to this rule, but only if the second put does not fail. If it does, the get will return the value stored by the first put, resulting in user memory corruption. This should not actually happen, because a file cache page is always removed from its address space after put, so that double put is impossible. However cleancache documentation states that this rule must hold, so let's conform to it to be 100% secure.
Signed-off-by: Vladimir Davydov <[email protected]> --- mm/tcache.c | 26 +++++++++++++++----------- 1 file changed, 15 insertions(+), 11 deletions(-) diff --git a/mm/tcache.c b/mm/tcache.c index e83ad05da944..5d4320b42f5d 100644 --- a/mm/tcache.c +++ b/mm/tcache.c @@ -837,23 +837,27 @@ static void tcache_cleancache_put_page(int pool_id, struct tcache_node *node; struct page *cache_page; - if (!tcache_active) - return; - - cache_page = tcache_alloc_page(); - if (!cache_page) - return; + if (tcache_active) + cache_page = tcache_alloc_page(); + else + cache_page = NULL; - copy_highpage(cache_page, page); + if (cache_page) + copy_highpage(cache_page, page); - node = tcache_get_node_and_pool(pool_id, &key, true); + node = tcache_get_node_and_pool(pool_id, &key, !!cache_page); if (node) { - /* cleancache does not care about failures */ - (void)tcache_attach_page(node, index, cache_page); + if (cache_page) + /* update the cache page */ + (void)tcache_attach_page(node, index, cache_page); + else + /* invalidate the stale cache page */ + cache_page = tcache_detach_page(node, index); tcache_put_node_and_pool(node); } - put_page(cache_page); + if (cache_page) + put_page(cache_page); } static int tcache_cleancache_get_page(int pool_id, -- 2.1.4 _______________________________________________ Devel mailing list [email protected] https://lists.openvz.org/mailman/listinfo/devel
