openwrt/toolchain/musl/patches/500-0001-reorder-thread-list-unlink-in-pthread_exit-after-all.patch
Matthias Schiffer afea16b8f7
musl: restore lock skipping for mostly-singlethreaded programs, and related patches
The remainder of the patch series proposed by upstream [2] for the locking
synchronization issue [1].

[1] https://www.openwall.com/lists/musl/2020/05/22/3
[2] https://www.openwall.com/lists/musl/2020/05/22/10

Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
2020-05-26 23:45:17 +02:00

52 lines
2.1 KiB
Diff

From 4d5aa20a94a2d3fae3e69289dc23ecafbd0c16c4 Mon Sep 17 00:00:00 2001
From: Rich Felker <dalias@aerifal.cx>
Date: Fri, 22 May 2020 17:35:14 -0400
Subject: [PATCH 1/4] reorder thread list unlink in pthread_exit after all
locks
since the backend for LOCK() skips locking if single-threaded, it's
unsafe to make the process appear single-threaded before the last use
of lock.
this fixes potential unsynchronized access to a linked list via
__dl_thread_cleanup.
---
src/thread/pthread_create.c | 19 +++++++++++--------
1 file changed, 11 insertions(+), 8 deletions(-)
--- a/src/thread/pthread_create.c
+++ b/src/thread/pthread_create.c
@@ -90,14 +90,7 @@ _Noreturn void __pthread_exit(void *resu
exit(0);
}
- /* At this point we are committed to thread termination. Unlink
- * the thread from the list. This change will not be visible
- * until the lock is released, which only happens after SYS_exit
- * has been called, via the exit futex address pointing at the lock. */
- libc.threads_minus_1--;
- self->next->prev = self->prev;
- self->prev->next = self->next;
- self->prev = self->next = self;
+ /* At this point we are committed to thread termination. */
/* Process robust list in userspace to handle non-pshared mutexes
* and the detached thread case where the robust list head will
@@ -121,6 +114,16 @@ _Noreturn void __pthread_exit(void *resu
__do_orphaned_stdio_locks();
__dl_thread_cleanup();
+ /* Last, unlink thread from the list. This change will not be visible
+ * until the lock is released, which only happens after SYS_exit
+ * has been called, via the exit futex address pointing at the lock.
+ * This needs to happen after any possible calls to LOCK() that might
+ * skip locking if libc.threads_minus_1 is zero. */
+ libc.threads_minus_1--;
+ self->next->prev = self->prev;
+ self->prev->next = self->next;
+ self->prev = self->next = self;
+
/* This atomic potentially competes with a concurrent pthread_detach
* call; the loser is responsible for freeing thread resources. */
int state = a_cas(&self->detach_state, DT_JOINABLE, DT_EXITING);