Messages in this thread Patch in this message | | | Date | Tue, 26 Aug 2008 09:22:20 -0700 | From | Arjan van de Ven <> | Subject | Re: PATCH] debug: add notifier chain debugging |
| |
On Mon, 25 Aug 2008 22:08:00 -0700 "Tony Luck" <tony.luck@intel.com> wrote:
> >> This breaks on ia64 (and pa-risc I think) where function pointers > >> don't point directly at the code, they point to a {code,data} > >> structure which is itself located in data space, not text space. > > > > is there a way to go to the actual address? I'm sure this is a bit > > more common.... (like kallsyms!) > > See dereference_function_descriptor() in lib/vsprintf.c (where it will > be clear that my memory was wrong and that PPC64 is the other > architecture that needs this). >
The patch below fixes this; Ingo, please replace the patch with this one.
From: Arjan van de Ven <arjan@linux.intel.com> Subject: [PATCH] debug: add notifier chain debugging
during some development we suspected a case where we left something in a notifier chain that was from a module that was unloaded already... and that sort of thing is rather hard to track down.
This patch adds a very simple sanity check (which isn't all that expensive) to make sure the notifier we're about to call is actually from either the kernel itself of from a still-loaded module, avoiding a hard-to-chase-down crash.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Acked-by: Tony Luck <tony.luck@intel.com> --- include/linux/kernel.h | 3 +++ kernel/extable.c | 16 ++++++++++++++++ kernel/notifier.c | 6 ++++++ lib/vsprintf.c | 2 +- 4 files changed, 26 insertions(+), 1 deletions(-)
diff --git a/include/linux/kernel.h b/include/linux/kernel.h index 2651f80..4e1366b 100644 --- a/include/linux/kernel.h +++ b/include/linux/kernel.h @@ -187,6 +187,9 @@ extern unsigned long long memparse(char *ptr, char **retptr); extern int core_kernel_text(unsigned long addr); extern int __kernel_text_address(unsigned long addr); extern int kernel_text_address(unsigned long addr); +extern int func_ptr_is_kernel_text(void *ptr); +extern void *dereference_function_descriptor(void *ptr); + struct pid; extern struct pid *session_of_pgrp(struct pid *pgrp); diff --git a/kernel/extable.c b/kernel/extable.c index a26cb2e..adf0cc9 100644 --- a/kernel/extable.c +++ b/kernel/extable.c @@ -66,3 +66,19 @@ int kernel_text_address(unsigned long addr) return 1; return module_text_address(addr) != NULL; } + +/* + * On some architectures (PPC64, IA64) function pointers + * are actually only tokens to some data that then holds the + * real function address. As a result, to find if a function + * pointer is part of the kernel text, we need to do some + * special dereferencing first. + */ +int func_ptr_is_kernel_text(void *ptr) +{ + unsigned long addr; + addr = (unsigned long) dereference_function_descriptor(ptr); + if (core_kernel_text(addr)) + return 1; + return module_text_address(addr) != NULL; +} diff --git a/kernel/notifier.c b/kernel/notifier.c index 823be11..522277c 100644 --- a/kernel/notifier.c +++ b/kernel/notifier.c @@ -82,6 +82,12 @@ static int __kprobes notifier_call_chain(struct notifier_block **nl, while (nb && nr_to_call) { next_nb = rcu_dereference(nb->next); + if (!func_ptr_is_kernel_text(nb->notifier_call)) { + WARN(1, "Invalid notifier called!"); + nb = next_nb; + continue; + } + ret = nb->notifier_call(nb, val, v); if (nr_calls) diff --git a/lib/vsprintf.c b/lib/vsprintf.c index d8d1d11..f5e5ffb 100644 --- a/lib/vsprintf.c +++ b/lib/vsprintf.c @@ -513,7 +513,7 @@ static char *string(char *buf, char *end, char *s, int field_width, int precisio return buf; } -static inline void *dereference_function_descriptor(void *ptr) +void *dereference_function_descriptor(void *ptr) { #if defined(CONFIG_IA64) || defined(CONFIG_PPC64) void *p; -- 1.5.5.1
| |