linux-kernel - [PATCH] timer: Provide wrappers safe for use with LOCKDEP

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 19 Oct 2017 13:28:38 -0700
From:   Kees Cook <keescook@...omium.org>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     Craig Gallek <cgallek@...gle.com>,
        Eric Dumazet <edumazet@...gle.com>,
        "David S. Miller" <davem@...emloft.net>, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: [PATCH] timer: Provide wrappers safe for use with LOCKDEP

Under LOCKDEP, the timer lock_class_key (set up in __setup_timer) needs
to be tied to the caller's context, so an inline for timer_setup()
won't work. We do, however, want to keep the inline version around for
argument type checking, though, so this provides macro wrappers in the
LOCKDEP case.

This fixes the case of different timers sharing the same LOCKDEP instance,
and producing a false positive warning:

[  580.840858] ======================================================
[  580.842299] WARNING: possible circular locking dependency detected
[  580.843684] 4.14.0-rc4+ #17 Not tainted
[  580.844554] ------------------------------------------------------
[  580.845945] swapper/9/0 is trying to acquire lock:
[  580.847024]  (slock-AF_INET){+.-.}, at: [<ffffffff84ea4c34>] tcp_write_timer+0x24/0xd0
[  580.848834]
               but task is already holding lock:
[  580.850107]  ((timer)#2){+.-.}, at: [<ffffffff846df7c0>] call_timer_fn+0x0/0x300
[  580.851663]
               which lock already depends on the new lock.

[  580.853439]
               the existing dependency chain (in reverse order) is:
[  580.855311]
               -> #1 ((timer)#2){+.-.}:
[  580.856538]        __lock_acquire+0x114d/0x11a0
[  580.857506]        lock_acquire+0xb0/0x1d0
[  580.858373]        del_timer_sync+0x3c/0xb0
[  580.859260]        inet_csk_reqsk_queue_drop+0x7f/0x1b0
...
               -> #0 (slock-AF_INET){+.-.}:
[  580.884980]        check_prev_add+0x666/0x700
[  580.885790]        __lock_acquire+0x114d/0x11a0
[  580.886575]        lock_acquire+0xb0/0x1d0
[  580.887289]        _raw_spin_lock+0x2c/0x40
[  580.888021]        tcp_write_timer+0x24/0xd0
...
[  580.900055]  Possible unsafe locking scenario:

[  580.901043]        CPU0                    CPU1
[  580.901797]        ----                    ----
[  580.902540]   lock((timer)#2);
[  580.903046]                                lock(slock-AF_INET);
[  580.904006]                                lock((timer)#2);
[  580.904915]   lock(slock-AF_INET);
[  580.905502]

In this report, del_timer_sync() is from:

	inet_csk_reqsk_queue_drop()
		reqsk_queue_unlink()
			del_timer_sync(&req->rsk_timer)

but tcp_write_timer()'s timer is attached to icsk_retransmit_timer. Both
had the same lock_class_key, since they were using timer_setup(). Switching
to a macro allows for a separate context, avoiding the false positive.

Reported-by: Craig Gallek <cgallek@...gle.com>
Suggested-by: Eric Dumazet <edumazet@...gle.com>
Signed-off-by: Kees Cook <keescook@...omium.org>
---
This means that all the trees taking conversions are at risk for harmless
LOCKDEP warnings until this is landed. Putting this in v4.14 doesn't make
sense since no one is going to rebase their trees. Also, it depends on
the timer_setup_on_stack() function that was added in -next in timers/core.
Any thoughts on the best way to deal with this?
---
 include/linux/timer.h | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/include/linux/timer.h b/include/linux/timer.h
index 10685c33e679..09950482309b 100644
--- a/include/linux/timer.h
+++ b/include/linux/timer.h
@@ -150,6 +150,7 @@ static inline void init_timer_on_stack_key(struct timer_list *timer,
 #define TIMER_DATA_TYPE		unsigned long
 #define TIMER_FUNC_TYPE		void (*)(TIMER_DATA_TYPE)
 
+#ifndef CONFIG_LOCKDEP
 static inline void timer_setup(struct timer_list *timer,
 			       void (*callback)(struct timer_list *),
 			       unsigned int flags)
@@ -165,6 +166,19 @@ static inline void timer_setup_on_stack(struct timer_list *timer,
 	__setup_timer_on_stack(timer, (TIMER_FUNC_TYPE)callback,
 			       (TIMER_DATA_TYPE)timer, flags);
 }
+#else
+/*
+ * Under LOCKDEP, the timer lock_class_key (set up in __init_timer) needs
+ * to be tied to the caller's context, so an inline (above) won't work. We
+ * do want to keep the inline for argument type checking, though.
+ */
+# define timer_setup(timer, callback, flags)				\
+		__setup_timer(timer, (TIMER_FUNC_TYPE)callback,		\
+			      (TIMER_DATA_TYPE)timer, flags)
+# define timer_setup_on_stack(timer, callback, flags)			\
+		__setup_timer_on_stack(timer, (TIMER_FUNC_TYPE)callback,\
+				       (TIMER_DATA_TYPE)timer, flags)
+#endif
 
 #define from_timer(var, callback_timer, timer_fieldname) \
 	container_of(callback_timer, typeof(*var), timer_fieldname)
-- 
2.7.4


-- 
Kees Cook
Pixel Security