new attempt at making set*id() safe and robust

changing credentials in a multi-threaded program is extremely
difficult on linux because it requires synchronizing the change
between all threads, which have their own thread-local credentials on
the kernel side. this is further complicated by the fact that changing
the real uid can fail due to exceeding RLIMIT_NPROC, making it
possible that the syscall will succeed in some threads but fail in
others.

the old __rsyscall approach being replaced was robust in that it would
report failure if any one thread failed, but in this case, the program
would be left in an inconsistent state where individual threads might
have different uid. (this was not as bad as glibc, which would
sometimes even fail to report the failure entirely!)

the new approach being committed refuses to change real user id when
it cannot temporarily set the rlimit to infinity. this is completely
POSIX conformant since POSIX does not require an implementation to
allow real-user-id changes for non-privileged processes whatsoever.
still, setting the real uid can fail due to memory allocation in the
kernel, but this can only happen if there is not already a cached
object for the target user. thus, we forcibly serialize the syscalls
attempts, and fail the entire operation on the first failure. this
*should* lead to an all-or-nothing success/failure result, but it's
still fragile and highly dependent on kernel developers not breaking
things worse than they're already broken.

ideally linux will eventually add a CLONE_USERCRED flag that would
give POSIX conformant credential changes without any hacks from
userspace, and all of this code would become redundant and could be
removed ~10 years down the line when everyone has abandoned the old
broken kernels. i'm not holding my breath...

3 files changed, 113 insertions, 118 deletions

diff --git a/src/thread/__rsyscall.c b/src/thread/__rsyscall.c
deleted file mode 100644
index e885d9e7..00000000
--- a/src/thread/__rsyscall.c
+++ /dev/null
@@ -1,114 +0,0 @@
-#include "pthread_impl.h"
-
-/* "rsyscall" is a mechanism by which a thread can synchronously force all
- * other threads to perform an arbitrary syscall. It is necessary to work
- * around the non-conformant implementation of setuid() et al on Linux,
- * which affect only the calling thread and not the whole process. This
- * implementation performs some tricks with signal delivery to work around
- * the fact that it does not keep any list of threads in userspace. */
-
-static struct {
-	volatile int lock, hold, blocks, cnt;
-	unsigned long arg[6];
-	int nr;
-	int err;
-	int init;
-} rs;
-
-static void rsyscall_handler(int sig, siginfo_t *si, void *ctx)
-{
-	struct pthread *self = __pthread_self();
-	long r;
-
-	if (!rs.hold || rs.cnt == libc.threads_minus_1) return;
-
-	/* Threads which have already decremented themselves from the
-	 * thread count must not increment rs.cnt or otherwise act. */
-	if (self->dead) {
-		sigfillset(&((ucontext_t *)ctx)->uc_sigmask);
-		return;
-	}
-
-	r = __syscall(rs.nr, rs.arg[0], rs.arg[1],
-		rs.arg[2], rs.arg[3], rs.arg[4], rs.arg[5]);
-	if (r < 0) rs.err=-r;
-
-	a_inc(&rs.cnt);
-	__wake(&rs.cnt, 1, 1);
-	while(rs.hold)
-		__wait(&rs.hold, 0, 1, 1);
-	a_dec(&rs.cnt);
-	if (!rs.cnt) __wake(&rs.cnt, 1, 1);
-}
-
-int __rsyscall(int nr, long a, long b, long c, long d, long e, long f)
-{
-	int i, ret;
-	sigset_t set = { 0 };
-	struct pthread *self;
-
-	if (!libc.threads_minus_1)
-		return syscall(nr, a, b, c, d, e, f);
-
-	self = __pthread_self();
-
-	LOCK(&rs.lock);
-	while ((i=rs.blocks))
-		__wait(&rs.blocks, 0, i, 1);
-
-	__syscall(SYS_rt_sigprocmask, SIG_BLOCK, (uint64_t[]){-1}, &set, 8);
-
-	if (!rs.init) {
-		struct sigaction sa = {
-			.sa_flags = SA_SIGINFO | SA_RESTART,
-			.sa_sigaction = rsyscall_handler,
-			.sa_mask = set
-		};
-		sigfillset(&sa.sa_mask);
-		sa.sa_sigaction = rsyscall_handler;
-		__libc_sigaction(SIGSYSCALL, &sa, 0);
-		rs.init = 1;
-	}
-
-	rs.nr = nr;
-	rs.arg[0] = a; rs.arg[1] = b;
-	rs.arg[2] = c; rs.arg[3] = d;
-	rs.arg[4] = d; rs.arg[5] = f;
-	rs.err = 0;
-	rs.cnt = 0;
-	rs.hold = 1;
-
-	/* Dispatch signals until all threads respond */
-	for (i=libc.threads_minus_1; i; i--)
-		sigqueue(self->pid, SIGSYSCALL, (union sigval){0});
-	while ((i=rs.cnt) < libc.threads_minus_1) {
-		sigqueue(self->pid, SIGSYSCALL, (union sigval){0});
-		__wait(&rs.cnt, 0, i, 1);
-	}
-
-	/* Handle any lingering signals with no-op */
-	__syscall(SYS_rt_sigprocmask, SIG_SETMASK, &set, &set, 8);
-
-	/* Resume other threads' signal handlers and wait for them */
-	rs.hold = 0;
-	__wake(&rs.hold, -1, 0);
-	while((i=rs.cnt)) __wait(&rs.cnt, 0, i, 1);
-
-	if (rs.err) errno = rs.err, ret = -1;
-	else ret = syscall(nr, a, b, c, d, e, f);
-
-	UNLOCK(&rs.lock);
-	return ret;
-}
-
-void __rsyscall_lock()
-{
-	a_inc(&rs.blocks);
-	while (rs.lock) __wait(&rs.lock, 0, 1, 1);
-}
-
-void __rsyscall_unlock()
-{
-	a_dec(&rs.blocks);
-	if (rs.lock) __wake(&rs.blocks, 1, 1);
-}
diff --git a/src/thread/pthread_create.c b/src/thread/pthread_create.c
index d60c2a4d..856015ff 100644
--- a/src/thread/pthread_create.c
+++ b/src/thread/pthread_create.c
@@ -3,8 +3,8 @@
 static void dummy_0()
 {
 }
-weak_alias(dummy_0, __rsyscall_lock);
-weak_alias(dummy_0, __rsyscall_unlock);
+weak_alias(dummy_0, __synccall_lock);
+weak_alias(dummy_0, __synccall_unlock);
 weak_alias(dummy_0, __pthread_tsd_run_dtors);
 
 #ifdef __pthread_unwind_next
@@ -99,12 +99,12 @@ int pthread_create(pthread_t *res, const pthread_attr_t *attr, void *(*entry)(vo
 	new->tlsdesc[1] = (uintptr_t)new;
 	stack = (void *)((uintptr_t)new-1 & ~(uintptr_t)15);
 
-	__rsyscall_lock();
+	__synccall_lock();
 
 	a_inc(&libc.threads_minus_1);
 	ret = __uniclone(stack, start, new);
 
-	__rsyscall_unlock();
+	__synccall_unlock();
 
 	if (ret < 0) {
 		a_dec(&libc.threads_minus_1);
diff --git a/src/thread/synccall.c b/src/thread/synccall.c
new file mode 100644
index 00000000..7c4f92bf
--- /dev/null
+++ b/src/thread/synccall.c
@@ -0,0 +1,109 @@
+#include "pthread_impl.h"
+#include <semaphore.h>
+
+static struct chain {
+	struct chain *next;
+	sem_t sem, sem2;
+} *head;
+
+static void (*callback)(void *), *context;
+static int chainlen;
+static sem_t chainlock, chaindone;
+static pthread_rwlock_t lock = PTHREAD_RWLOCK_INITIALIZER;
+
+static void handler(int sig, siginfo_t *si, void *ctx)
+{
+	struct chain ch;
+	pthread_t self = __pthread_self();
+	int old_errno = errno;
+
+	if (chainlen == libc.threads_minus_1) return;
+
+	sigqueue(self->pid, SIGSYNCCALL, (union sigval){0});
+
+	/* Threads which have already decremented themselves from the
+	 * thread count must not act. Block further receipt of signals
+	 * and return. */
+	if (self->dead) {
+		memset(&((ucontext_t *)ctx)->uc_sigmask, -1, 8);
+		errno = old_errno;
+		return;
+	}
+
+	sem_init(&ch.sem, 0, 0);
+	sem_init(&ch.sem2, 0, 0);
+
+	while (sem_wait(&chainlock));
+	ch.next = head;
+	head = &ch;
+	if (++chainlen == libc.threads_minus_1) sem_post(&chaindone);
+	sem_post(&chainlock);
+
+	while (sem_wait(&ch.sem));
+	callback(context);
+	sem_post(&ch.sem2);
+	while (sem_wait(&ch.sem));
+
+	errno = old_errno;
+}
+
+void __synccall(void (*func)(void *), void *ctx)
+{
+	pthread_t self;
+	struct sigaction sa;
+	struct chain *cur, *next;
+	uint64_t oldmask;
+
+	pthread_rwlock_wrlock(&lock);
+
+	__syscall(SYS_rt_sigprocmask, SIG_BLOCK, (uint64_t[]){-1}, &oldmask, 8);
+
+	if (!libc.threads_minus_1) {
+		func(ctx);
+		return;
+	}
+
+	sem_init(&chaindone, 0, 0);
+	sem_init(&chainlock, 0, 1);
+	chainlen = 0;
+	callback = func;
+	context = ctx;
+
+	sa.sa_flags = SA_SIGINFO | SA_RESTART;
+	sa.sa_sigaction = handler;
+	sigfillset(&sa.sa_mask);
+	__libc_sigaction(SIGSYNCCALL, &sa, 0);
+
+	self = __pthread_self();
+	sigqueue(self->pid, SIGSYNCCALL, (union sigval){0});
+	while (sem_wait(&chaindone));
+
+	for (cur=head; cur; cur=cur->next) {
+		sem_post(&cur->sem);
+		while (sem_wait(&cur->sem2));
+	}
+	func(ctx);
+
+	for (cur=head; cur; cur=next) {
+		next = cur->next;
+		sem_post(&cur->sem);
+	}
+
+	sa.sa_flags = 0;
+	sa.sa_handler = SIG_IGN;
+	__libc_sigaction(SIGSYNCCALL, &sa, 0);
+
+	__syscall(SYS_rt_sigprocmask, SIG_SETMASK, &oldmask, 0, 8);
+
+	pthread_rwlock_unlock(&lock);
+}
+
+void __synccall_lock()
+{
+	pthread_rwlock_rdlock(&lock);
+}
+
+void __synccall_unlock()
+{
+	pthread_rwlock_unlock(&lock);
+}