2010-12-07 04:29:29

by Ian Munsie

Subject: [PATCH 2/6] trace syscalls: Remove redundant syscall_nr checks

From: Ian Munsie <[email protected]>

With the ftrace events now checking if the syscall_nr is valid upon
initialisation, there is no need to verify it when registering and
unregistering the events, so remove the check.

Signed-off-by: Ian Munsie <[email protected]>
---
kernel/trace/trace_syscalls.c | 9 +--------
1 files changed, 1 insertions(+), 8 deletions(-)

diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
index 14f4c02..5e34b49 100644
--- a/kernel/trace/trace_syscalls.c
+++ b/kernel/trace/trace_syscalls.c
@@ -367,8 +367,6 @@ int reg_event_syscall_enter(struct ftrace_event_call *call)
int num;

num = ((struct syscall_metadata *)call->data)->syscall_nr;
- if (num < 0 || num >= NR_syscalls)
- return -ENOSYS;
mutex_lock(&syscall_trace_lock);
if (!sys_refcount_enter)
ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
@@ -385,8 +383,6 @@ void unreg_event_syscall_enter(struct ftrace_event_call *call)
int num;

num = ((struct syscall_metadata *)call->data)->syscall_nr;
- if (num < 0 || num >= NR_syscalls)
- return;
mutex_lock(&syscall_trace_lock);
sys_refcount_enter--;
clear_bit(num, enabled_enter_syscalls);
@@ -401,8 +397,7 @@ int reg_event_syscall_exit(struct ftrace_event_call *call)
int num;

num = ((struct syscall_metadata *)call->data)->syscall_nr;
- if (num < 0 || num >= NR_syscalls)
- return -ENOSYS;
+
mutex_lock(&syscall_trace_lock);
if (!sys_refcount_exit)
ret = register_trace_sys_exit(ftrace_syscall_exit, NULL);
@@ -419,8 +414,6 @@ void unreg_event_syscall_exit(struct ftrace_event_call *call)
int num;

num = ((struct syscall_metadata *)call->data)->syscall_nr;
- if (num < 0 || num >= NR_syscalls)
- return;
mutex_lock(&syscall_trace_lock);
sys_refcount_exit--;
clear_bit(num, enabled_exit_syscalls);
--
1.7.2.3

2010-12-07 14:54:45

by Steven Rostedt

Subject: Re: [PATCH 2/6] trace syscalls: Remove redundant syscall_nr checks

On Tue, 2010-12-07 at 15:29 +1100, Ian Munsie wrote:
> From: Ian Munsie <[email protected]>
>
> With the ftrace events now checking if the syscall_nr is valid upon
> initialisation, there is no need to verify it when registering and
> unregistering the events, so remove the check.

I still like to keep these checks. I don't mind redundant checks that
are in slow paths, as they may catch a bug on a change in the future.

What you could do is change these to:

if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))

-- Steve

>
> Signed-off-by: Ian Munsie <[email protected]>
> ---
> kernel/trace/trace_syscalls.c | 9 +--------
> 1 files changed, 1 insertions(+), 8 deletions(-)
>
> diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
> index 14f4c02..5e34b49 100644
> --- a/kernel/trace/trace_syscalls.c
> +++ b/kernel/trace/trace_syscalls.c
> @@ -367,8 +367,6 @@ int reg_event_syscall_enter(struct ftrace_event_call *call)
> int num;
>
> num = ((struct syscall_metadata *)call->data)->syscall_nr;
> - if (num < 0 || num >= NR_syscalls)
> - return -ENOSYS;
> mutex_lock(&syscall_trace_lock);
> if (!sys_refcount_enter)
> ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
> @@ -385,8 +383,6 @@ void unreg_event_syscall_enter(struct ftrace_event_call *call)
> int num;
>
> num = ((struct syscall_metadata *)call->data)->syscall_nr;
> - if (num < 0 || num >= NR_syscalls)
> - return;
> mutex_lock(&syscall_trace_lock);
> sys_refcount_enter--;
> clear_bit(num, enabled_enter_syscalls);
> @@ -401,8 +397,7 @@ int reg_event_syscall_exit(struct ftrace_event_call *call)
> int num;
>
> num = ((struct syscall_metadata *)call->data)->syscall_nr;
> - if (num < 0 || num >= NR_syscalls)
> - return -ENOSYS;
> +
> mutex_lock(&syscall_trace_lock);
> if (!sys_refcount_exit)
> ret = register_trace_sys_exit(ftrace_syscall_exit, NULL);
> @@ -419,8 +414,6 @@ void unreg_event_syscall_exit(struct ftrace_event_call *call)
> int num;
>
> num = ((struct syscall_metadata *)call->data)->syscall_nr;
> - if (num < 0 || num >= NR_syscalls)
> - return;
> mutex_lock(&syscall_trace_lock);
> sys_refcount_exit--;
> clear_bit(num, enabled_exit_syscalls);

2010-12-07 23:47:47

by Ian Munsie

Subject: Re: [PATCH 2/6] trace syscalls: Remove redundant syscall_nr checks

Excerpts from Steven Rostedt's message of Wed Dec 08 01:54:42 +1100 2010:
> On Tue, 2010-12-07 at 15:29 +1100, Ian Munsie wrote:
> > From: Ian Munsie <[email protected]>
> >
> > With the ftrace events now checking if the syscall_nr is valid upon
> > initialisation, there is no need to verify it when registering and
> > unregistering the events, so remove the check.
>
> I still like to keep these checks. I don't mind redundant checks that
> are in slow paths, as they may catch a bug on a change in the future.
>
> What you could do is change these to:
>
> if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))

Hi Steve,

Thanks for the feedback. Will update and resubmit.

Cheers,
-Ian