diff options
| author | Jean Boussier <[email protected]> | 2024-05-29 16:46:04 +0200 |
|---|---|---|
| committer | Jean Boussier <[email protected]> | 2024-09-05 11:43:46 +0200 |
| commit | 63cbe3f6ac9feb44a2e43b1f853e2ca7e049316c () | |
| tree | dddbbbc8678332d16f9cf5e9630645e233656797 | |
| parent | 2e5680d304a9cf9a6a2ba582091af6719e839351 (diff) | |
Proof of Concept: Allow to prevent fork from happening in known fork unsafe API
[Feature #20590] For better of for worse, fork(2) remain the primary provider of parallelism in Ruby programs. Even though it's frowned uppon in many circles, and a lot of literature will simply state that only async-signal safe APIs are safe to use after `fork()`, in practice most APIs work well as long as you are careful about not forking while another thread is holding a pthread mutex. One of the APIs that is known cause fork safety issues is `getaddrinfo`. If you fork while another thread is inside `getaddrinfo`, a mutex may be left locked in the child, with no way to unlock it. I think we could reduce the impact of these problem by preventing in for the most notorious and common cases, by locking around `fork(2)` and known unsafe APIs with a read-write lock.
Notes: Merged: https://.com/ruby/ruby/pull/10864
| -rw-r--r-- | ext/socket/raddrinfo.c | 16 | ||||
| -rw-r--r-- | internal/thread.h | 5 | ||||
| -rw-r--r-- | process.c | 5 | ||||
| -rw-r--r-- | thread_none.c | 6 | ||||
| -rw-r--r-- | thread_pthread.c | 46 | ||||
| -rw-r--r-- | thread_win32.c | 6 |
6 files changed, 82 insertions, 2 deletions
@@ -327,6 +327,12 @@ nogvl_getaddrinfo(void *arg) return (void *)(VALUE)ret; } static int rb_getaddrinfo(const char *hostp, const char *portp, const struct addrinfo *hints, struct addrinfo **ai) { @@ -336,7 +342,7 @@ rb_getaddrinfo(const char *hostp, const char *portp, const struct addrinfo *hint arg.service = portp; arg.hints = hints; arg.res = ai; - return (int)(VALUE)rb_thread_call_without_gvl(nogvl_getaddrinfo, &arg, RUBY_UBF_IO, 0); } #elif GETADDRINFO_IMPL == 2 @@ -477,6 +483,12 @@ do_pthread_create(pthread_t *th, void *(*start_routine) (void *), void *arg) return ret; } static int rb_getaddrinfo(const char *hostp, const char *portp, const struct addrinfo *hints, struct addrinfo **ai) { @@ -493,7 +505,7 @@ start: } pthread_t th; - if (do_pthread_create(&th, do_getaddrinfo, arg) != 0) { int err = errno; free_getaddrinfo_arg(arg); errno = err; @@ -46,6 +46,9 @@ VALUE rb_thread_shield_wait(VALUE self); VALUE rb_thread_shield_release(VALUE self); VALUE rb_thread_shield_destroy(VALUE self); int rb_thread_to_be_killed(VALUE thread); void rb_mutex_allow_trap(VALUE self, int val); VALUE rb_uninterruptible(VALUE (*b_proc)(VALUE), VALUE data); VALUE rb_mutex_owned_p(VALUE self); @@ -64,6 +67,8 @@ void rb_notify_fd_close_wait(struct rb_io_close_wait_list *busy); RUBY_SYMBOL_EXPORT_BEGIN /* Temporary. This API will be removed (renamed). */ VALUE rb_thread_io_blocking_region(rb_blocking_function_t *func, void *data1, int fd); VALUE rb_thread_io_blocking_call(rb_blocking_function_t *func, void *data1, int fd, int events); @@ -4227,12 +4227,17 @@ rb_fork_ruby(int *status) prefork(); before_fork_ruby(); disable_child_handler_before_fork(&old); child.pid = pid = rb_fork(); child.error = err = errno; disable_child_handler_fork_parent(&old); /* yes, bad name */ after_fork_ruby(pid); /* repeat while fork failed but retryable */ @@ -326,4 +326,10 @@ rb_thread_lock_native_thread(void) return false; } #endif /* THREAD_SYSTEM_DEPENDENT_IMPLEMENTATION */ @@ -3346,6 +3346,52 @@ native_sleep(rb_thread_t *th, rb_hrtime_t *rel) RUBY_DEBUG_LOG("wakeup"); } // thread internal event hooks (only for pthread) struct rb_internal_thread_event_hook { @@ -1011,4 +1011,10 @@ rb_thread_lock_native_thread(void) return false; } #endif /* THREAD_SYSTEM_DEPENDENT_IMPLEMENTATION */ |