kprobes: Fix suspicious RCU usage WARN in get_kprobe()
RFC PATCH 1/1 kprobes: Fix suspicious RCU usage WARN in get_kprobe()
With CONFIG_PROVE_RCU_LIST set, we may get the following WARN in the test code:
Kprobe smoke test: started
=============================
WARNING: suspicious RCU usage 5.5.0-rc1-00013-ge15bd404ed10-dirty #802 Not tainted
-----------------------------kernel/kprobes.c:329 RCU-list traversed in non-reader section!!
other info that might help us debug this:
rcu_scheduler_active = 2, debug_locks = 1
1 lock held by swapper/0/1:
#0: ffff800011bf3648 (kprobe_mutex){+.+.}, at: register_kprobe+0x94/0x5a0
stack backtrace: CPU: 1 PID: 1 Comm: swapper/0 Not tainted 5.5.0-rc1-00013-ge15bd404ed10-dirty #802 Hardware name: Huawei Taishan 2280 /D05, BIOS Hisilicon D05 IT21 Nemo 2.0 RC0 04/18/2018 Call trace: dump_backtrace+0x0/0x1a0 show_stack+0x14/0x20 dump_stack+0xe8/0x150 lockdep_rcu_suspicious+0xcc/0x110 get_kprobe+0xb8/0xc0 __get_valid_kprobe+0x18/0xc8 register_kprobe+0x9c/0x5a0 init_test_probes+0x80/0x400 init_kprobes+0x13c/0x154 do_one_initcall+0x88/0x428 kernel_init_freeable+0x21c/0x2c4 kernel_init+0x10/0x108 ret_from_fork+0x10/0x18
Kprobe smoke test: passed successfully
The code comment tells us the locking requirements:
/*
* This routine is called either:
* - under the kprobe_mutex - during kprobe_[un]register()
* OR
* - with preemption disabled - from arch/xxx/kernel/kprobes.c
*/
struct kprobe *get_kprobe(void *addr)
{
struct hlist_head *head;
struct kprobe *p;
head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
hlist_for_each_entry_rcu(p, head, hlist,And we have the kprobe_mutex held in the path of concern, so add a RCU list traversal check condition for this.
---
I sent as an RFC as I am not 100% certain that this is the right fix. It does solve my WARN.
I also assume __get_valid_kprobe() will require a similar change for similar reason.
kernel/kprobes.c +2 -1
@@ -326,7 +326,8 @@ struct kprobe *get_kprobe(void *addr)struct kprobe *p; head = &kprobe_table[hash_ptr (addr,KPROBE_HASH_BITS )]; -hlist_for_each_entry_rcu (p, head, hlist) { +hlist_for_each_entry_rcu (p, head, hlist, +mutex_is_locked (&kprobe_mutex)) {if (p->addr == addr)return p; }
Signature · 1 line
2.17.1
Masami Hiramatsu
maintainerHi John,
Thanks for your work. Actually, I already sent the same fix 2 weeks ago.
https://lore.kernel.org/lkml/157535316659.16485.11817291759382261088.stgit@devnote2/
Thank you,
Quoted 80 lines · On Fri, 20 Dec 2019 20:49:51 +0800 John Garry <[email protected]> wrote:
With
CONFIG_PROVE_RCU_LISTset, we may get the following WARN in the test code:Kprobe smoke test: started
=============================
WARNING: suspicious RCU usage 5.5.0-rc1-00013-ge15bd404ed10-dirty #802 Not tainted
-----------------------------kernel/kprobes.c:329RCU-list traversed in non-reader section!!other info that might help us debug this:
rcu_scheduler_active= 2,debug_locks= 1
1 lock held by swapper/0/1:
#0:ffff800011bf3648(kprobe_mutex){+.+.}, at:register_kprobe+0x94/0x5a0stack backtrace: CPU: 1 PID: 1 Comm: swapper/0 Not tainted 5.5.0-rc1-00013-ge15bd404ed10-dirty #802 Hardware name: Huawei Taishan 2280 /D05, BIOS Hisilicon D05 IT21 Nemo 2.0 RC0 04/18/2018 Call trace: dump_backtrace+0x0/0x1a0 show_stack+0x14/0x20 dump_stack+0xe8/0x150 lockdep_rcu_suspicious+0xcc/0x110 get_kprobe+0xb8/0xc0 __get_valid_kprobe+0x18/0xc8 register_kprobe+0x9c/0x5a0 init_test_probes+0x80/0x400 init_kprobes+0x13c/0x154 do_one_initcall+0x88/0x428 kernel_init_freeable+0x21c/0x2c4 kernel_init+0x10/0x108 ret_from_fork+0x10/0x18Kprobe smoke test: passed successfully
The code comment tells us the locking requirements:
/*
* This routine is called either:
* - under thekprobe_mutex- during kprobe_[un]register()
* OR
* - with preemption disabled - fromarch/xxx/kernel/kprobes.c
*/
struct kprobe *get_kprobe(void *addr){ struct hlist_head *head; struct kprobe *p; head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; hlist_for_each_entry_rcu(p, head, hlist,And we have the
kprobe_mutexheld in the path of concern, so add a RCU list traversal check condition for this.Signed-off-by John Garry <[email protected]>---
I sent as an RFC as I am not 100% certain that this is the right fix. It does solve my WARN.I also assume
__get_valid_kprobe()will require a similar change for similar reason.kernel/kprobes.c +2
-1@@ -326,7 +326,8 @@ struct kprobe *get_kprobe(void *addr)struct kprobe *p; head = &kprobe_table[hash_ptr (addr,KPROBE_HASH_BITS )]; -hlist_for_each_entry_rcu (p, head, hlist) { +hlist_for_each_entry_rcu (p, head, hlist, +mutex_is_locked (&kprobe_mutex)) {if (p->addr == addr)return p; }Signature · 1 line
2.17.1
Signature · 1 line
Masami Hiramatsu <[email protected]>