Skip to contentSkip to conversation

Allow rt_sigqueueinfo and rt_tgsigqueueinfo to use si_code == SI_ASYNCIO

PATCH In review Roland Dreier 2 participants 3 messages Mar 28, 2011

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Allow rt_sigqueueinfo and rt_tgsigqueueinfo to use si_code == SI_ASYNCIO

From: Roland Dreier <[email protected]>

Commit da48524eb206 ("Prevent rt_sigqueueinfo and rt_tgsigqueueinfo from spoofing the signal code") made the check on si_code too strict:
glibc's aio implementation wants to queue signals with SI_ASYNCIO, and indeed glibc's tst-aio4 fails with the patched kernel.

Fix this by loosening the new check to allow SI_ASYNCIO as well.

---
 kernel/signal.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
kernel/signal.c +2 -2
index 324eff5..b2bfa3a 100644
@@ -2437,7 +2437,7 @@ SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
 	/* Not even root can pretend to send signals from the kernel.
 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
 	 */
-	if (info.si_code != SI_QUEUE) {
+	if (info.si_code != SI_QUEUE && info.si_code != SI_ASYNCIO) {
 		/* We used to allow any < 0 si_code */
 		WARN_ON_ONCE(info.si_code < 0);
 		return -EPERM;
@@ -2457,7 +2457,7 @@ long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
 	/* Not even root can pretend to send signals from the kernel.
 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
 	 */
-	if (info->si_code != SI_QUEUE) {
+	if (info->si_code != SI_QUEUE && info->si_code != SI_ASYNCIO) {
 		/* We used to allow any < 0 si_code */
 		WARN_ON_ONCE(info->si_code < 0);
 		return -EPERM;
#1 lore raw
Quoted 7 lines · Roland Dreier

Commit da48524eb206 ("Prevent rt_sigqueueinfo and rt_tgsigqueueinfo from spoofing the signal code") made the check on si_code too strict:
glibc's aio implementation wants to queue signals with SI_ASYNCIO, and indeed glibc's tst-aio4 fails with the patched kernel.

Fix this by loosening the new check to allow SI_ASYNCIO as well.

Heh.

kernel/signal.c SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig, · line 2440

Quoted 8 lines
kernel/signal.c +1 -1
@@ -2437,7 +2437,7 @@ SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
 	/* Not even root can pretend to send signals from the kernel.
 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
 	 */
-	if (info.si_code != SI_QUEUE) {
+	if (info.si_code != SI_QUEUE && info.si_code != SI_ASYNCIO) {

I am starting to think that "plan B" was better.

	if (info.si_code >= 0 || info.si_code == SI_TKILL)
		return -EPERM;

Although we should probably keep WARN_ON_ONCE() anyway.

Oleg.

#2 lore raw
Quoted 4 lines · Oleg Nesterov

I am starting to think that "plan B" was better.

       if (info.si_code >= 0 || info.si_code == SI_TKILL) return -EPERM;

Indeed. Grepping further in glibc finds sysdeps/unix/sysv/linux/aio_sigqueue.c:

        /* We must pass the information about the data in a siginfo_t
           value.  */
        info.si_signo = timer->event.sigev_signo;
        info.si_code = SI_TIMER;
        info.si_pid = timer->creator_pid;
        info.si_uid = getuid ();
        info.si_value = timer->event.sigev_value;

        INLINE_SYSCALL (rt_sigqueueinfo, 3, info.si_pid, info.si_signo, &info);

and sysdeps/unix/sysv/linux/gai_sigqueue.c:

  /* We must pass the information about the data in a siginfo_t value.  */
  info.si_signo = sig;
  info.si_code = SI_ASYNCNL;
  info.si_pid = caller_pid;
  info.si_uid = __getuid ();
  info.si_value = val;

return INLINE_SYSCALL (rt_sigqueueinfo, 3, info.si_pid,

                         sig, __ptrvalue (&info));

where SI_ASYNCNL is -60, which the kernel doesn't know about.

Updated patch coming in a minute.

Thanks,
Roland