fix race condition in sigqueue

this race is fundamentally due to linux's bogus requirement that
userspace, rather than kernelspace, fill in the siginfo structure. an
intervening signal handler that calls fork could cause both the parent
and child process to send signals claiming to be from the parent,
which could in turn have harmful effects depending on what the
recipient does with the signal. we simply block all signals for the
interval between getuid and sigqueue syscalls (much like what raise()
does already) to prevent the race and make the getuid/sigqueue pair
atomic.

this will be a non-issue if linux is fixed to validate the siginfo
structure or fill it in from kernelspace.
This commit is contained in:
Rich Felker 2011-07-30 21:11:31 -04:00
parent ad5881842e
commit 07827d1a82
1 changed files with 8 additions and 2 deletions

View File

@ -1,16 +1,22 @@
#include <signal.h>
#include <string.h>
#include <unistd.h>
#include <stdint.h>
#include "syscall.h"
int sigqueue(pid_t pid, int sig, const union sigval value)
{
siginfo_t si;
sigset_t set;
int r;
memset(&si, 0, sizeof si);
si.si_signo = sig;
si.si_code = SI_QUEUE;
si.si_value = value;
si.si_pid = getpid();
si.si_uid = getuid();
return syscall(SYS_rt_sigqueueinfo, pid, sig, &si);
pthread_sigmask(SIG_BLOCK, (void *)(uint64_t[1]){-1}, &set);
si.si_pid = getpid();
r = syscall(SYS_rt_sigqueueinfo, pid, sig, &si);
pthread_sigmask(SIG_SETMASK, &set, 0);
return r;
}