nfsd: don't replace page in rq_pages if it's a continuation of last page
authorJeff Layton <jlayton@kernel.org>
Fri, 17 Mar 2023 17:13:08 +0000 (13:13 -0400)
committerChuck Lever <chuck.lever@oracle.com>
Fri, 17 Mar 2023 22:18:15 +0000 (18:18 -0400)
The splice read calls nfsd_splice_actor to put the pages containing file
data into the svc_rqst->rq_pages array. It's possible however to get a
splice result that only has a partial page at the end, if (e.g.) the
filesystem hands back a short read that doesn't cover the whole page.

nfsd_splice_actor will plop the partial page into its rq_pages array and
return. Then later, when nfsd_splice_actor is called again, the
remainder of the page may end up being filled out. At this point,
nfsd_splice_actor will put the page into the array _again_ corrupting
the reply. If this is done enough times, rq_next_page will overrun the
array and corrupt the trailing fields -- the rq_respages and
rq_next_page pointers themselves.

If we've already added the page to the array in the last pass, don't add
it to the array a second time when dealing with a splice continuation.
This was originally handled properly in nfsd_splice_actor, but commit
91e23b1c3982 ("NFSD: Clean up nfsd_splice_actor()") removed the check
for it.

Fixes: 91e23b1c3982 ("NFSD: Clean up nfsd_splice_actor()")
Cc: Al Viro <viro@zeniv.linux.org.uk>
Reported-by: Dario Lesca <d.lesca@solinos.it>
Tested-by: David Critch <dcritch@redhat.com>
Link: https://bugzilla.redhat.com/show_bug.cgi?id=2150630
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
fs/nfsd/vfs.c

index ba34a31a7c7027f640949294f1a405a9f765ba45..cd0dbea335d976e699f855c546a60f46d28b0fc4 100644 (file)
@@ -941,8 +941,15 @@ nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
        struct page *last_page;
 
        last_page = page + (offset + sd->len - 1) / PAGE_SIZE;
-       for (page += offset / PAGE_SIZE; page <= last_page; page++)
+       for (page += offset / PAGE_SIZE; page <= last_page; page++) {
+               /*
+                * Skip page replacement when extending the contents
+                * of the current page.
+                */
+               if (page == *(rqstp->rq_next_page - 1))
+                       continue;
                svc_rqst_replace_page(rqstp, page);
+       }
        if (rqstp->rq_res.page_len == 0)        // first call
                rqstp->rq_res.page_base = offset % PAGE_SIZE;
        rqstp->rq_res.page_len += sd->len;