This is the mail archive of the
mailing list for the glibc project.
On Mon, 2017-06-19 at 12:04 +0100, Phil Blundell wrote:
Under conditions that I don't entirely understand yet, we seem to be
> somehow returning from gai_suspend while its waitlist entry is
> linked into requestlist->waiting.
Here's a patch that fixes bug 20874 for me, fwiw. It still passes
"make subdirs='resolv' xcheck".
OK to commit?
From 6e5dbbcfc0594dad90dc6f8b4537dba26bceb428 Mon Sep 17 00:00:00 2001
From: Phil Blundell <email@example.com>
Date: Mon, 19 Jun 2017 13:11:00 +0100
Subject: [PATCH] gai_suspend: Remove bogus check for EAI_INPROGRESS [BZ
If we added an entry to the waitlist for any request, it is important
that we remove it again before returning. Failing to do so will
cause obscure and hard-to-debug crashes because the linked list
will contain a pointer to a struct that was assigned on the stack
and has since been overwritten.
Although we check that the current "return value" of the request
is EAI_INPROGRESS before adding an entry to its waitlist, this
value may change while we sleep so we cannot assume it will still
be EAI_INPROGRESS when we come to remove the entry afterwards.
resolv/gai_suspend.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/resolv/gai_suspend.c b/resolv/gai_suspend.c
index a86bd4360d..139d636c78 100644
@@ -111,8 +111,7 @@ gai_suspend (const struct gaicb *const list, int ent,
/* Now remove the entry in the waiting list for all requests
which didn't terminate. */
for (cnt = 0; cnt < ent; ++cnt)
- if (list[cnt] != NULL && list[cnt]->__return == EAI_INPROGRESS
- && requestlist[cnt] != NULL)
+ if (list[cnt] != NULL && requestlist[cnt] != NULL)
struct waitlist **listp = &requestlist[cnt]->waiting;