LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
To: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	ananth@in.ibm.com, David Miller <davem@davemloft.net>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Rusty Russell <rusty@rustcorp.com.au>
Subject: Re: [PATCH] kprobes: disable preempt for module_text_address()
Date: Wed, 05 Nov 2008 11:30:16 -0800	[thread overview]
Message-ID: <4911F448.3010701@ct.jp.nec.com> (raw)
In-Reply-To: <4910FB3D.4020805@cn.fujitsu.com>

Lai Jiangshan wrote:
> Masami Hiramatsu wrote:
>> Hi Lai,
>>
>> Lai Jiangshan wrote:
>>> __register_kprobe() may be preempted after module_text_address()
>>> but before try_module_get(), and in this interval the module may be
>>> unloaded and try_module_get(probed_mod) will access to invalid address.
>>> this patch uses preempt_disable() to protect it.
>> Thank you for your work.
>>
>> I think this is the problem of module_text_address() because it can
>> return incorrect address of struct module if a preemption happens.
>> So, I think the module_text_address() would better to call try_module_get()
>> before returning its address, or at least they should comment that caller
>> needs disabling preemption.
>>
>> struct module *module_text_address(unsigned long addr)
>> {
>> 	struct module *mod;
>>
>> 	preempt_disable(); /*
>> 			    * I also think this preemption disabling is not so useful
>> 			    * without try_module_get(), because caller have to
>> 			    * disable preemption...
>> 			    */
>> 	mod = __module_text_address(addr);
>> 	/* here, try_module_get() is needed.
>> 	 * (or commenting "caller must disable preemption!") */
>> 	preempt_enable();
>>
>> 	/*
>> 	 * !!Here!! if the preemption happened, it could return invalid "mod".
>>          * In that case, even if module_text_address() returns non-NULL,
>> 	 * the addr is no longer in any module.
>> 	 */
>> 	return mod;
>> }
>>
>> Thank you,
>>
> 
> I don't think module_text_address() are needed to modified.
> only __register_kprobe() uses module_text_address()'s return value.
> other users of module_text_address() are just for testing it's
> return value. so only __register_kprobe() needs disabling preemption
> currently.

I guess, there is another issue if the module is unloaded before
module_text_address(). p->addr is no longer valid, but in
__register_kprobe(), module_text_address() returns NULL means
it's not a module.

How about this?
--------
From: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Subject: [PATCH] kprobe: fix module checking in __register_kprobe()

__register_kprobe() may be preempted before/after module_text_address().
In this preemption, the module may be unloaded.

If the module is unloaded before module_text_address(), kprobe address
becomes invalid. If the module is unloaded after module_text_address()
and before try_module_get(), try_module_get(probe_mod) will access invalid
address.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
---
 kernel/kprobes.c |   19 +++++++++++++------
 1 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 8b57a25..71dc2e9 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -605,7 +605,7 @@ static int __kprobes __register_kprobe(struct kprobe *p,
 {
 	int ret = 0;
 	struct kprobe *old_p;
-	struct module *probed_mod;
+	struct module *probed_mod = NULL;
 	kprobe_opcode_t *addr;
 
 	addr = kprobe_addr(p);
@@ -622,20 +622,27 @@ static int __kprobes __register_kprobe(struct kprobe *p,
 	/*
 	 * Check if are we probing a module.
 	 */
-	probed_mod = module_text_address((unsigned long) p->addr);
-	if (probed_mod) {
-		struct module *calling_mod = module_text_address(called_from);
+	if (!core_kernel_text((unsigned long) p->addr)) {
+		preempt_disable();
+		probed_mod = __module_text_address((unsigned long) p->addr);
+		if (!probed_mod) {
+			preempt_enable();
+			return -EINVAL;
+		}
 		/*
 		 * We must allow modules to probe themself and in this case
 		 * avoid incrementing the module refcount, so as to allow
 		 * unloading of self probing modules.
 		 */
-		if (calling_mod && calling_mod != probed_mod) {
-			if (unlikely(!try_module_get(probed_mod)))
+		if (__module_text_address(called_from) != probed_mod) {
+			if (unlikely(!try_module_get(probed_mod))) {
+				preempt_enable();
 				return -EINVAL;
+			}
 			p->mod_refcounted = 1;
 		} else
 			probed_mod = NULL;
+		preempt_enable();
 	}
 
 	p->nmissed = 0;
-- 
1.5.6


  reply	other threads:[~2008-11-05 19:30 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-11-04  5:56 Lai Jiangshan
2008-11-04 14:28 ` Ananth N Mavinakayanahalli
2008-11-05  0:53   ` Lai Jiangshan
2008-11-05  1:27 ` Masami Hiramatsu
2008-11-05  1:47   ` Lai Jiangshan
2008-11-05 19:30     ` Hiroshi Shimamoto [this message]
2008-11-05 21:40     ` Masami Hiramatsu
2008-11-05 22:46       ` Hiroshi Shimamoto
2008-11-05 23:07         ` Masami Hiramatsu
2008-11-06  0:06           ` [PATCH] kprobes: bugfix: try_module_get even if calling_mod is NULL Masami Hiramatsu
2008-11-07  1:00             ` Andrew Morton
2008-11-07  2:28               ` Masami Hiramatsu
2008-11-07  2:54                 ` Andrew Morton
2008-11-07  4:46                   ` Ananth N Mavinakayanahalli
2008-11-06  1:06       ` [PATCH] kprobes: disable preempt for module_text_address() Lai Jiangshan
2008-11-06 15:37         ` [PATCH] kprobes: disable preempt for module_text_address() and kernel_text_address() Masami Hiramatsu
2008-11-07  0:32           ` Lai Jiangshan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4911F448.3010701@ct.jp.nec.com \
    --to=h-shimamoto@ct.jp.nec.com \
    --cc=akpm@linux-foundation.org \
    --cc=ananth@in.ibm.com \
    --cc=davem@davemloft.net \
    --cc=laijs@cn.fujitsu.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mhiramat@redhat.com \
    --cc=rusty@rustcorp.com.au \
    --subject='Re: [PATCH] kprobes: disable preempt for module_text_address()' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).