The list func (ftrace_ops_list_func) will be patched first
before the transition between old and new calls are set,
which fixed the race described in this commit `59338f75`.
While ftrace_trace_function changes from the list func to a
ftrace_ops func, like unregistering the klp_ops to leave the only
global_ops in ftrace_ops_list, the ftrace_[regs]_call will be
replaced with the list func although it already exists. So there
should be a condition to avoid this.
This patch backups saved_ftrace_func by saved_ftrace_func_old
which will be compared with the list func before trying to patch it.
Signed-off-by: Song Shuai <[email protected]>
---
kernel/trace/ftrace.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index bc921a3f7ea8..56b1a42e1937 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -2755,6 +2755,8 @@ void __weak ftrace_arch_code_modify_post_process(void)
{
}
+static ftrace_func_t saved_ftrace_func_old;
+
void ftrace_modify_all_code(int command)
{
int update = command & FTRACE_UPDATE_TRACE_FUNC;
@@ -2774,7 +2776,7 @@ void ftrace_modify_all_code(int command)
* to make sure the ops are having the right functions
* traced.
*/
- if (update) {
+ if (update && saved_ftrace_func_old != ftrace_ops_list_func) {
err = ftrace_update_ftrace_func(ftrace_ops_list_func);
if (FTRACE_WARN_ON(err))
return;
@@ -2918,6 +2920,7 @@ static void ftrace_trampoline_free(struct ftrace_ops *ops)
static void ftrace_startup_enable(int command)
{
if (saved_ftrace_func != ftrace_trace_function) {
+ saved_ftrace_func_old = saved_ftrace_func;
saved_ftrace_func = ftrace_trace_function;
command |= FTRACE_UPDATE_TRACE_FUNC;
}
@@ -3007,6 +3010,7 @@ int ftrace_shutdown(struct ftrace_ops *ops, int command)
ops->flags &= ~FTRACE_OPS_FL_ENABLED;
if (saved_ftrace_func != ftrace_trace_function) {
+ saved_ftrace_func_old = saved_ftrace_func;
saved_ftrace_func = ftrace_trace_function;
command |= FTRACE_UPDATE_TRACE_FUNC;
}
@@ -8321,6 +8325,7 @@ static void ftrace_startup_sysctl(void)
/* Force update next time */
saved_ftrace_func = NULL;
+ saved_ftrace_func_old = NULL;
/* ftrace_start_up is true if we want ftrace running */
if (ftrace_start_up) {
command = FTRACE_UPDATE_CALLS;
--
2.20.1
Steven Rostedt <[email protected]> 于2022年11月22日周二 02:34写道：
>
> On Tue, 22 Nov 2022 02:28:25 +0000
> Song Shuai <[email protected]> wrote:
>
> > Song Shuai <[email protected]> 于2022年10月26日周三 13:20写道：
> > >
> > > The list func (ftrace_ops_list_func) will be patched first
> > > before the transition between old and new calls are set,
> > > which fixed the race described in this commit `59338f75`.
> > >
> > > While ftrace_trace_function changes from the list func to a
> > > ftrace_ops func, like unregistering the klp_ops to leave the only
> > > global_ops in ftrace_ops_list, the ftrace_[regs]_call will be
> > > replaced with the list func although it already exists. So there
> > > should be a condition to avoid this.
> > >
> > > This patch backups saved_ftrace_func by saved_ftrace_func_old
> > > which will be compared with the list func before trying to patch it.
> > >
> > Ping...
>
> Thanks for the ping. I had thought I had replied to this, but I don't
> see it in my sent folder. I may have been distracted, and lost the
> message.
>
> I'll take a look at it tomorrow.
>
> -- Steve
No problem, thanks!
-- Song
On Tue, 22 Nov 2022 02:28:25 +0000
Song Shuai <[email protected]> wrote:
> Song Shuai <[email protected]> 于2022年10月26日周三 13:20写道：
> >
> > The list func (ftrace_ops_list_func) will be patched first
> > before the transition between old and new calls are set,
> > which fixed the race described in this commit `59338f75`.
> >
> > While ftrace_trace_function changes from the list func to a
> > ftrace_ops func, like unregistering the klp_ops to leave the only
> > global_ops in ftrace_ops_list, the ftrace_[regs]_call will be
> > replaced with the list func although it already exists. So there
> > should be a condition to avoid this.
> >
> > This patch backups saved_ftrace_func by saved_ftrace_func_old
> > which will be compared with the list func before trying to patch it.
> >
> Ping...
Thanks for the ping. I had thought I had replied to this, but I don't
see it in my sent folder. I may have been distracted, and lost the
message.
I'll take a look at it tomorrow.
-- Steve
Song Shuai <[email protected]> 于2022年10月26日周三 13:20写道：
>
> The list func (ftrace_ops_list_func) will be patched first
> before the transition between old and new calls are set,
> which fixed the race described in this commit `59338f75`.
>
> While ftrace_trace_function changes from the list func to a
> ftrace_ops func, like unregistering the klp_ops to leave the only
> global_ops in ftrace_ops_list, the ftrace_[regs]_call will be
> replaced with the list func although it already exists. So there
> should be a condition to avoid this.
>
> This patch backups saved_ftrace_func by saved_ftrace_func_old
> which will be compared with the list func before trying to patch it.
>
Ping...
Thanks,
Song
> Signed-off-by: Song Shuai <[email protected]>
> ---
> kernel/trace/ftrace.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index bc921a3f7ea8..56b1a42e1937 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -2755,6 +2755,8 @@ void __weak ftrace_arch_code_modify_post_process(void)
> {
> }
>
> +static ftrace_func_t saved_ftrace_func_old;
> +
> void ftrace_modify_all_code(int command)
> {
> int update = command & FTRACE_UPDATE_TRACE_FUNC;
> @@ -2774,7 +2776,7 @@ void ftrace_modify_all_code(int command)
> * to make sure the ops are having the right functions
> * traced.
> */
> - if (update) {
> + if (update && saved_ftrace_func_old != ftrace_ops_list_func) {
> err = ftrace_update_ftrace_func(ftrace_ops_list_func);
> if (FTRACE_WARN_ON(err))
> return;
> @@ -2918,6 +2920,7 @@ static void ftrace_trampoline_free(struct ftrace_ops *ops)
> static void ftrace_startup_enable(int command)
> {
> if (saved_ftrace_func != ftrace_trace_function) {
> + saved_ftrace_func_old = saved_ftrace_func;
> saved_ftrace_func = ftrace_trace_function;
> command |= FTRACE_UPDATE_TRACE_FUNC;
> }
> @@ -3007,6 +3010,7 @@ int ftrace_shutdown(struct ftrace_ops *ops, int command)
> ops->flags &= ~FTRACE_OPS_FL_ENABLED;
>
> if (saved_ftrace_func != ftrace_trace_function) {
> + saved_ftrace_func_old = saved_ftrace_func;
> saved_ftrace_func = ftrace_trace_function;
> command |= FTRACE_UPDATE_TRACE_FUNC;
> }
> @@ -8321,6 +8325,7 @@ static void ftrace_startup_sysctl(void)
>
> /* Force update next time */
> saved_ftrace_func = NULL;
> + saved_ftrace_func_old = NULL;
> /* ftrace_start_up is true if we want ftrace running */
> if (ftrace_start_up) {
> command = FTRACE_UPDATE_CALLS;
> --
> 2.20.1
>
On Tue, 22 Nov 2022 02:28:25 +0000
Song Shuai <[email protected]> wrote:
> Song Shuai <[email protected]> 于2022年10月26日周三 13:20写道：
> >
> > The list func (ftrace_ops_list_func) will be patched first
> > before the transition between old and new calls are set,
> > which fixed the race described in this commit `59338f75`.
> >
> > While ftrace_trace_function changes from the list func to a
> > ftrace_ops func, like unregistering the klp_ops to leave the only
> > global_ops in ftrace_ops_list, the ftrace_[regs]_call will be
> > replaced with the list func although it already exists. So there
> > should be a condition to avoid this.
> >
> > This patch backups saved_ftrace_func by saved_ftrace_func_old
> > which will be compared with the list func before trying to patch it.
> >
> Ping...
Wouldn't something like this be simpler and easier to manage (without
adding another variable to keep track of)?
-- Steve
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 65a5d36463e0..d04552c0c275 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -2763,6 +2763,19 @@ void __weak ftrace_arch_code_modify_post_process(void)
{
}
+static int update_ftrace_func(ftrace_func_t func)
+{
+ static ftrace_func_t save_func;
+
+ /* Avoid updating if it hasn't changed */
+ if (func == save_func)
+ return 0;
+
+ save_func = func;
+
+ return ftrace_update_ftrace_func(func);
+}
+
void ftrace_modify_all_code(int command)
{
int update = command & FTRACE_UPDATE_TRACE_FUNC;
@@ -2783,7 +2796,7 @@ void ftrace_modify_all_code(int command)
* traced.
*/
if (update) {
- err = ftrace_update_ftrace_func(ftrace_ops_list_func);
+ err = update_ftrace_func(ftrace_ops_list_func);
if (FTRACE_WARN_ON(err))
return;
}
@@ -2799,7 +2812,7 @@ void ftrace_modify_all_code(int command)
/* If irqs are disabled, we are in stop machine */
if (!irqs_disabled())
smp_call_function(ftrace_sync_ipi, NULL, 1);
- err = ftrace_update_ftrace_func(ftrace_trace_function);
+ err = update_ftrace_func(ftrace_trace_function);
if (FTRACE_WARN_ON(err))
return;
}
Steven Rostedt <[email protected]> 于2022年11月22日周二 21:32写道：
>
> On Tue, 22 Nov 2022 02:28:25 +0000
> Song Shuai <[email protected]> wrote:
>
> > Song Shuai <[email protected]> 于2022年10月26日周三 13:20写道：
> > >
> > > The list func (ftrace_ops_list_func) will be patched first
> > > before the transition between old and new calls are set,
> > > which fixed the race described in this commit `59338f75`.
> > >
> > > While ftrace_trace_function changes from the list func to a
> > > ftrace_ops func, like unregistering the klp_ops to leave the only
> > > global_ops in ftrace_ops_list, the ftrace_[regs]_call will be
> > > replaced with the list func although it already exists. So there
> > > should be a condition to avoid this.
> > >
> > > This patch backups saved_ftrace_func by saved_ftrace_func_old
> > > which will be compared with the list func before trying to patch it.
> > >
> > Ping...
>
> Wouldn't something like this be simpler and easier to manage (without
> adding another variable to keep track of)?
>
> -- Steve
>
Thanks for the corrections, this looks great to me.
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index 65a5d36463e0..d04552c0c275 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -2763,6 +2763,19 @@ void __weak ftrace_arch_code_modify_post_process(void)
> {
> }
>
> +static int update_ftrace_func(ftrace_func_t func)
> +{
> + static ftrace_func_t save_func;
> +
> + /* Avoid updating if it hasn't changed */
> + if (func == save_func)
> + return 0;
> +
> + save_func = func;
> +
> + return ftrace_update_ftrace_func(func);
> +}
> +
> void ftrace_modify_all_code(int command)
> {
> int update = command & FTRACE_UPDATE_TRACE_FUNC;
> @@ -2783,7 +2796,7 @@ void ftrace_modify_all_code(int command)
> * traced.
> */
> if (update) {
> - err = ftrace_update_ftrace_func(ftrace_ops_list_func);
> + err = update_ftrace_func(ftrace_ops_list_func);
> if (FTRACE_WARN_ON(err))
> return;
> }
> @@ -2799,7 +2812,7 @@ void ftrace_modify_all_code(int command)
> /* If irqs are disabled, we are in stop machine */
> if (!irqs_disabled())
> smp_call_function(ftrace_sync_ipi, NULL, 1);
> - err = ftrace_update_ftrace_func(ftrace_trace_function);
> + err = update_ftrace_func(ftrace_trace_function);
> if (FTRACE_WARN_ON(err))
> return;
> }
Thanks,
Song