2022-05-09 06:46:18

by Masami Hiramatsu

[permalink] [raw]
Subject: [PATCH] rethook: Reject getting a rethook if RCU is not watching

Since the rethook_recycle() will involve the call_rcu() for reclaiming
the rethook_instance, the rethook must be set up at the RCU available
context (non idle). This rethook_recycle() in the rethook trampoline
handler is inevitable, thus the RCU available check must be done before
setting the rethook trampoline.

This adds a rcu_is_watching() check in the rethook_try_get() so that
it will return NULL if it is called when !rcu_is_watching().

Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
Signed-off-by: Masami Hiramatsu <[email protected]>
---
kernel/trace/rethook.c | 9 +++++++++
1 file changed, 9 insertions(+)

diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
index b56833700d23..c69d82273ce7 100644
--- a/kernel/trace/rethook.c
+++ b/kernel/trace/rethook.c
@@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
if (unlikely(!handler))
return NULL;

+ /*
+ * This expects the caller will set up a rethook on a function entry.
+ * When the function returns, the rethook will eventually be reclaimed
+ * or released in the rethook_recycle() with call_rcu().
+ * This means the caller must be run in the RCU-availabe context.
+ */
+ if (unlikely(!rcu_is_watching()))
+ return NULL;
+
fn = freelist_try_get(&rh->pool);
if (!fn)
return NULL;



2022-05-25 08:33:20

by Steven Rostedt

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Sat, 7 May 2022 13:46:52 +0900
Masami Hiramatsu <[email protected]> wrote:

Is this expected to go through the BPF tree?

-- Steve


> Since the rethook_recycle() will involve the call_rcu() for reclaiming
> the rethook_instance, the rethook must be set up at the RCU available
> context (non idle). This rethook_recycle() in the rethook trampoline
> handler is inevitable, thus the RCU available check must be done before
> setting the rethook trampoline.
>
> This adds a rcu_is_watching() check in the rethook_try_get() so that
> it will return NULL if it is called when !rcu_is_watching().
>
> Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> Signed-off-by: Masami Hiramatsu <[email protected]>
> ---
> kernel/trace/rethook.c | 9 +++++++++
> 1 file changed, 9 insertions(+)
>
> diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> index b56833700d23..c69d82273ce7 100644
> --- a/kernel/trace/rethook.c
> +++ b/kernel/trace/rethook.c
> @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> if (unlikely(!handler))
> return NULL;
>
> + /*
> + * This expects the caller will set up a rethook on a function entry.
> + * When the function returns, the rethook will eventually be reclaimed
> + * or released in the rethook_recycle() with call_rcu().
> + * This means the caller must be run in the RCU-availabe context.
> + */
> + if (unlikely(!rcu_is_watching()))
> + return NULL;
> +
> fn = freelist_try_get(&rh->pool);
> if (!fn)
> return NULL;


2022-05-26 15:24:22

by Jiri Olsa

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Thu, May 26, 2022 at 11:25:30PM +0900, Masami Hiramatsu wrote:
> On Tue, 24 May 2022 19:23:01 -0400
> Steven Rostedt <[email protected]> wrote:
>
> > On Sat, 7 May 2022 13:46:52 +0900
> > Masami Hiramatsu <[email protected]> wrote:
> >
> > Is this expected to go through the BPF tree?
> >
>
> Yes, since rethook (fprobe) is currently used only from eBPF.
> Jiri, can you check this is good for your test case?

sure I'll test it.. can't see the original email,
perhaps I wasn't cc-ed.. but I'll find it

is this also related to tracing 'idle' functions,
as discussed in here?
https://lore.kernel.org/bpf/[email protected]/

because that's the one I can reproduce.. but I can
certainly try that with your change as well

jirka

>
> Thank you,
>
>
> > -- Steve
> >
> >
> > > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > > the rethook_instance, the rethook must be set up at the RCU available
> > > context (non idle). This rethook_recycle() in the rethook trampoline
> > > handler is inevitable, thus the RCU available check must be done before
> > > setting the rethook trampoline.
> > >
> > > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > > it will return NULL if it is called when !rcu_is_watching().
> > >
> > > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > > Signed-off-by: Masami Hiramatsu <[email protected]>
> > > ---
> > > kernel/trace/rethook.c | 9 +++++++++
> > > 1 file changed, 9 insertions(+)
> > >
> > > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > > index b56833700d23..c69d82273ce7 100644
> > > --- a/kernel/trace/rethook.c
> > > +++ b/kernel/trace/rethook.c
> > > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > > if (unlikely(!handler))
> > > return NULL;
> > >
> > > + /*
> > > + * This expects the caller will set up a rethook on a function entry.
> > > + * When the function returns, the rethook will eventually be reclaimed
> > > + * or released in the rethook_recycle() with call_rcu().
> > > + * This means the caller must be run in the RCU-availabe context.
> > > + */
> > > + if (unlikely(!rcu_is_watching()))
> > > + return NULL;
> > > +
> > > fn = freelist_try_get(&rh->pool);
> > > if (!fn)
> > > return NULL;
> >
>
>
> --
> Masami Hiramatsu (Google) <[email protected]>

2022-05-27 02:35:41

by Masami Hiramatsu

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Tue, 24 May 2022 19:23:01 -0400
Steven Rostedt <[email protected]> wrote:

> On Sat, 7 May 2022 13:46:52 +0900
> Masami Hiramatsu <[email protected]> wrote:
>
> Is this expected to go through the BPF tree?
>

Yes, since rethook (fprobe) is currently used only from eBPF.
Jiri, can you check this is good for your test case?

Thank you,


> -- Steve
>
>
> > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > the rethook_instance, the rethook must be set up at the RCU available
> > context (non idle). This rethook_recycle() in the rethook trampoline
> > handler is inevitable, thus the RCU available check must be done before
> > setting the rethook trampoline.
> >
> > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > it will return NULL if it is called when !rcu_is_watching().
> >
> > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > Signed-off-by: Masami Hiramatsu <[email protected]>
> > ---
> > kernel/trace/rethook.c | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> >
> > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > index b56833700d23..c69d82273ce7 100644
> > --- a/kernel/trace/rethook.c
> > +++ b/kernel/trace/rethook.c
> > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > if (unlikely(!handler))
> > return NULL;
> >
> > + /*
> > + * This expects the caller will set up a rethook on a function entry.
> > + * When the function returns, the rethook will eventually be reclaimed
> > + * or released in the rethook_recycle() with call_rcu().
> > + * This means the caller must be run in the RCU-availabe context.
> > + */
> > + if (unlikely(!rcu_is_watching()))
> > + return NULL;
> > +
> > fn = freelist_try_get(&rh->pool);
> > if (!fn)
> > return NULL;
>


--
Masami Hiramatsu (Google) <[email protected]>

2022-05-28 18:50:26

by Masami Hiramatsu

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Thu, 26 May 2022 16:49:26 +0200
Jiri Olsa <[email protected]> wrote:

> On Thu, May 26, 2022 at 11:25:30PM +0900, Masami Hiramatsu wrote:
> > On Tue, 24 May 2022 19:23:01 -0400
> > Steven Rostedt <[email protected]> wrote:
> >
> > > On Sat, 7 May 2022 13:46:52 +0900
> > > Masami Hiramatsu <[email protected]> wrote:
> > >
> > > Is this expected to go through the BPF tree?
> > >
> >
> > Yes, since rethook (fprobe) is currently used only from eBPF.
> > Jiri, can you check this is good for your test case?
>
> sure I'll test it.. can't see the original email,
> perhaps I wasn't cc-ed.. but I'll find it

Here it is. I Cc-ed your @kernel.org address.
https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/T/#u

>
> is this also related to tracing 'idle' functions,
> as discussed in here?
> https://lore.kernel.org/bpf/[email protected]/

Ah, yes. So this may not happen with the above patch, but for the
hardening (ensuring it is always safe), I would like to add this.

>
> because that's the one I can reproduce.. but I can
> certainly try that with your change as well

Thank you!

>
> jirka
>
> >
> > Thank you,
> >
> >
> > > -- Steve
> > >
> > >
> > > > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > > > the rethook_instance, the rethook must be set up at the RCU available
> > > > context (non idle). This rethook_recycle() in the rethook trampoline
> > > > handler is inevitable, thus the RCU available check must be done before
> > > > setting the rethook trampoline.
> > > >
> > > > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > > > it will return NULL if it is called when !rcu_is_watching().
> > > >
> > > > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > > > Signed-off-by: Masami Hiramatsu <[email protected]>
> > > > ---
> > > > kernel/trace/rethook.c | 9 +++++++++
> > > > 1 file changed, 9 insertions(+)
> > > >
> > > > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > > > index b56833700d23..c69d82273ce7 100644
> > > > --- a/kernel/trace/rethook.c
> > > > +++ b/kernel/trace/rethook.c
> > > > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > > > if (unlikely(!handler))
> > > > return NULL;
> > > >
> > > > + /*
> > > > + * This expects the caller will set up a rethook on a function entry.
> > > > + * When the function returns, the rethook will eventually be reclaimed
> > > > + * or released in the rethook_recycle() with call_rcu().
> > > > + * This means the caller must be run in the RCU-availabe context.
> > > > + */
> > > > + if (unlikely(!rcu_is_watching()))
> > > > + return NULL;
> > > > +
> > > > fn = freelist_try_get(&rh->pool);
> > > > if (!fn)
> > > > return NULL;
> > >
> >
> >
> > --
> > Masami Hiramatsu (Google) <[email protected]>


--
Masami Hiramatsu (Google) <[email protected]>

2022-05-28 19:18:53

by Masami Hiramatsu

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Sat, 28 May 2022 00:10:08 +0200
Jiri Olsa <[email protected]> wrote:

> On Fri, May 27, 2022 at 01:14:34AM +0900, Masami Hiramatsu wrote:
> > On Thu, 26 May 2022 16:49:26 +0200
> > Jiri Olsa <[email protected]> wrote:
> >
> > > On Thu, May 26, 2022 at 11:25:30PM +0900, Masami Hiramatsu wrote:
> > > > On Tue, 24 May 2022 19:23:01 -0400
> > > > Steven Rostedt <[email protected]> wrote:
> > > >
> > > > > On Sat, 7 May 2022 13:46:52 +0900
> > > > > Masami Hiramatsu <[email protected]> wrote:
> > > > >
> > > > > Is this expected to go through the BPF tree?
> > > > >
> > > >
> > > > Yes, since rethook (fprobe) is currently used only from eBPF.
> > > > Jiri, can you check this is good for your test case?
> > >
> > > sure I'll test it.. can't see the original email,
> > > perhaps I wasn't cc-ed.. but I'll find it
> >
> > Here it is. I Cc-ed your @kernel.org address.
> > https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/T/#u
> >
> > >
> > > is this also related to tracing 'idle' functions,
> > > as discussed in here?
> > > https://lore.kernel.org/bpf/[email protected]/
> >
> > Ah, yes. So this may not happen with the above patch, but for the
> > hardening (ensuring it is always safe), I would like to add this.
> >
> > >
> > > because that's the one I can reproduce.. but I can
> > > certainly try that with your change as well
> >
> > Thank you!
>
> it did not help the idle warning as expected, but I did not
> see any problems running bpf tests on top of this

Oops, right. I forgot this is only for the rethook, not protect the
fprobe handlers, since fprobe code doesn't involve the RCU code (it
depends on ftrace's check). Sorry about that.
Hmm, I need to add a test code for this issue, but that could be
solved by your noninstr patch.

Thank you,

>
> jirka
>
> >
> > >
> > > jirka
> > >
> > > >
> > > > Thank you,
> > > >
> > > >
> > > > > -- Steve
> > > > >
> > > > >
> > > > > > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > > > > > the rethook_instance, the rethook must be set up at the RCU available
> > > > > > context (non idle). This rethook_recycle() in the rethook trampoline
> > > > > > handler is inevitable, thus the RCU available check must be done before
> > > > > > setting the rethook trampoline.
> > > > > >
> > > > > > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > > > > > it will return NULL if it is called when !rcu_is_watching().
> > > > > >
> > > > > > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > > > > > Signed-off-by: Masami Hiramatsu <[email protected]>
> > > > > > ---
> > > > > > kernel/trace/rethook.c | 9 +++++++++
> > > > > > 1 file changed, 9 insertions(+)
> > > > > >
> > > > > > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > > > > > index b56833700d23..c69d82273ce7 100644
> > > > > > --- a/kernel/trace/rethook.c
> > > > > > +++ b/kernel/trace/rethook.c
> > > > > > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > > > > > if (unlikely(!handler))
> > > > > > return NULL;
> > > > > >
> > > > > > + /*
> > > > > > + * This expects the caller will set up a rethook on a function entry.
> > > > > > + * When the function returns, the rethook will eventually be reclaimed
> > > > > > + * or released in the rethook_recycle() with call_rcu().
> > > > > > + * This means the caller must be run in the RCU-availabe context.
> > > > > > + */
> > > > > > + if (unlikely(!rcu_is_watching()))
> > > > > > + return NULL;
> > > > > > +
> > > > > > fn = freelist_try_get(&rh->pool);
> > > > > > if (!fn)
> > > > > > return NULL;
> > > > >
> > > >
> > > >
> > > > --
> > > > Masami Hiramatsu (Google) <[email protected]>
> >
> >
> > --
> > Masami Hiramatsu (Google) <[email protected]>


--
Masami Hiramatsu (Google) <[email protected]>

2022-05-28 19:43:51

by Jiri Olsa

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Fri, May 27, 2022 at 01:14:34AM +0900, Masami Hiramatsu wrote:
> On Thu, 26 May 2022 16:49:26 +0200
> Jiri Olsa <[email protected]> wrote:
>
> > On Thu, May 26, 2022 at 11:25:30PM +0900, Masami Hiramatsu wrote:
> > > On Tue, 24 May 2022 19:23:01 -0400
> > > Steven Rostedt <[email protected]> wrote:
> > >
> > > > On Sat, 7 May 2022 13:46:52 +0900
> > > > Masami Hiramatsu <[email protected]> wrote:
> > > >
> > > > Is this expected to go through the BPF tree?
> > > >
> > >
> > > Yes, since rethook (fprobe) is currently used only from eBPF.
> > > Jiri, can you check this is good for your test case?
> >
> > sure I'll test it.. can't see the original email,
> > perhaps I wasn't cc-ed.. but I'll find it
>
> Here it is. I Cc-ed your @kernel.org address.
> https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/T/#u
>
> >
> > is this also related to tracing 'idle' functions,
> > as discussed in here?
> > https://lore.kernel.org/bpf/[email protected]/
>
> Ah, yes. So this may not happen with the above patch, but for the
> hardening (ensuring it is always safe), I would like to add this.
>
> >
> > because that's the one I can reproduce.. but I can
> > certainly try that with your change as well
>
> Thank you!

it did not help the idle warning as expected, but I did not
see any problems running bpf tests on top of this

jirka

>
> >
> > jirka
> >
> > >
> > > Thank you,
> > >
> > >
> > > > -- Steve
> > > >
> > > >
> > > > > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > > > > the rethook_instance, the rethook must be set up at the RCU available
> > > > > context (non idle). This rethook_recycle() in the rethook trampoline
> > > > > handler is inevitable, thus the RCU available check must be done before
> > > > > setting the rethook trampoline.
> > > > >
> > > > > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > > > > it will return NULL if it is called when !rcu_is_watching().
> > > > >
> > > > > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > > > > Signed-off-by: Masami Hiramatsu <[email protected]>
> > > > > ---
> > > > > kernel/trace/rethook.c | 9 +++++++++
> > > > > 1 file changed, 9 insertions(+)
> > > > >
> > > > > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > > > > index b56833700d23..c69d82273ce7 100644
> > > > > --- a/kernel/trace/rethook.c
> > > > > +++ b/kernel/trace/rethook.c
> > > > > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > > > > if (unlikely(!handler))
> > > > > return NULL;
> > > > >
> > > > > + /*
> > > > > + * This expects the caller will set up a rethook on a function entry.
> > > > > + * When the function returns, the rethook will eventually be reclaimed
> > > > > + * or released in the rethook_recycle() with call_rcu().
> > > > > + * This means the caller must be run in the RCU-availabe context.
> > > > > + */
> > > > > + if (unlikely(!rcu_is_watching()))
> > > > > + return NULL;
> > > > > +
> > > > > fn = freelist_try_get(&rh->pool);
> > > > > if (!fn)
> > > > > return NULL;
> > > >
> > >
> > >
> > > --
> > > Masami Hiramatsu (Google) <[email protected]>
>
>
> --
> Masami Hiramatsu (Google) <[email protected]>

2022-06-06 03:45:28

by Andrii Nakryiko

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Fri, May 27, 2022 at 6:19 PM Masami Hiramatsu <[email protected]> wrote:
>
> On Sat, 28 May 2022 00:10:08 +0200
> Jiri Olsa <[email protected]> wrote:
>
> > On Fri, May 27, 2022 at 01:14:34AM +0900, Masami Hiramatsu wrote:
> > > On Thu, 26 May 2022 16:49:26 +0200
> > > Jiri Olsa <[email protected]> wrote:
> > >
> > > > On Thu, May 26, 2022 at 11:25:30PM +0900, Masami Hiramatsu wrote:
> > > > > On Tue, 24 May 2022 19:23:01 -0400
> > > > > Steven Rostedt <[email protected]> wrote:
> > > > >
> > > > > > On Sat, 7 May 2022 13:46:52 +0900
> > > > > > Masami Hiramatsu <[email protected]> wrote:
> > > > > >
> > > > > > Is this expected to go through the BPF tree?
> > > > > >
> > > > >
> > > > > Yes, since rethook (fprobe) is currently used only from eBPF.
> > > > > Jiri, can you check this is good for your test case?
> > > >
> > > > sure I'll test it.. can't see the original email,
> > > > perhaps I wasn't cc-ed.. but I'll find it
> > >
> > > Here it is. I Cc-ed your @kernel.org address.
> > > https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/T/#u
> > >
> > > >
> > > > is this also related to tracing 'idle' functions,
> > > > as discussed in here?
> > > > https://lore.kernel.org/bpf/[email protected]/
> > >
> > > Ah, yes. So this may not happen with the above patch, but for the
> > > hardening (ensuring it is always safe), I would like to add this.
> > >
> > > >
> > > > because that's the one I can reproduce.. but I can
> > > > certainly try that with your change as well
> > >
> > > Thank you!
> >
> > it did not help the idle warning as expected, but I did not
> > see any problems running bpf tests on top of this
>
> Oops, right. I forgot this is only for the rethook, not protect the
> fprobe handlers, since fprobe code doesn't involve the RCU code (it
> depends on ftrace's check). Sorry about that.
> Hmm, I need to add a test code for this issue, but that could be
> solved by your noninstr patch.
>


Masami,

It's not clear to me, do you intend to send a new revision with some
more tests or this patch as is ready to go into bpf tree?


> Thank you,
>
> >
> > jirka
> >
> > >
> > > >
> > > > jirka
> > > >
> > > > >
> > > > > Thank you,
> > > > >
> > > > >
> > > > > > -- Steve
> > > > > >
> > > > > >
> > > > > > > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > > > > > > the rethook_instance, the rethook must be set up at the RCU available
> > > > > > > context (non idle). This rethook_recycle() in the rethook trampoline
> > > > > > > handler is inevitable, thus the RCU available check must be done before
> > > > > > > setting the rethook trampoline.
> > > > > > >
> > > > > > > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > > > > > > it will return NULL if it is called when !rcu_is_watching().
> > > > > > >
> > > > > > > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > > > > > > Signed-off-by: Masami Hiramatsu <[email protected]>
> > > > > > > ---
> > > > > > > kernel/trace/rethook.c | 9 +++++++++
> > > > > > > 1 file changed, 9 insertions(+)
> > > > > > >
> > > > > > > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > > > > > > index b56833700d23..c69d82273ce7 100644
> > > > > > > --- a/kernel/trace/rethook.c
> > > > > > > +++ b/kernel/trace/rethook.c
> > > > > > > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > > > > > > if (unlikely(!handler))
> > > > > > > return NULL;
> > > > > > >
> > > > > > > + /*
> > > > > > > + * This expects the caller will set up a rethook on a function entry.
> > > > > > > + * When the function returns, the rethook will eventually be reclaimed
> > > > > > > + * or released in the rethook_recycle() with call_rcu().
> > > > > > > + * This means the caller must be run in the RCU-availabe context.
> > > > > > > + */
> > > > > > > + if (unlikely(!rcu_is_watching()))
> > > > > > > + return NULL;
> > > > > > > +
> > > > > > > fn = freelist_try_get(&rh->pool);
> > > > > > > if (!fn)
> > > > > > > return NULL;
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > Masami Hiramatsu (Google) <[email protected]>
> > >
> > >
> > > --
> > > Masami Hiramatsu (Google) <[email protected]>
>
>
> --
> Masami Hiramatsu (Google) <[email protected]>

2022-06-06 16:22:34

by Masami Hiramatsu

[permalink] [raw]
Subject: Re: [PATCH] rethook: Reject getting a rethook if RCU is not watching

On Fri, 3 Jun 2022 12:21:19 -0700
Andrii Nakryiko <[email protected]> wrote:

> On Fri, May 27, 2022 at 6:19 PM Masami Hiramatsu <[email protected]> wrote:
> >
> > On Sat, 28 May 2022 00:10:08 +0200
> > Jiri Olsa <[email protected]> wrote:
> >
> > > On Fri, May 27, 2022 at 01:14:34AM +0900, Masami Hiramatsu wrote:
> > > > On Thu, 26 May 2022 16:49:26 +0200
> > > > Jiri Olsa <[email protected]> wrote:
> > > >
> > > > > On Thu, May 26, 2022 at 11:25:30PM +0900, Masami Hiramatsu wrote:
> > > > > > On Tue, 24 May 2022 19:23:01 -0400
> > > > > > Steven Rostedt <[email protected]> wrote:
> > > > > >
> > > > > > > On Sat, 7 May 2022 13:46:52 +0900
> > > > > > > Masami Hiramatsu <[email protected]> wrote:
> > > > > > >
> > > > > > > Is this expected to go through the BPF tree?
> > > > > > >
> > > > > >
> > > > > > Yes, since rethook (fprobe) is currently used only from eBPF.
> > > > > > Jiri, can you check this is good for your test case?
> > > > >
> > > > > sure I'll test it.. can't see the original email,
> > > > > perhaps I wasn't cc-ed.. but I'll find it
> > > >
> > > > Here it is. I Cc-ed your @kernel.org address.
> > > > https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/T/#u
> > > >
> > > > >
> > > > > is this also related to tracing 'idle' functions,
> > > > > as discussed in here?
> > > > > https://lore.kernel.org/bpf/[email protected]/
> > > >
> > > > Ah, yes. So this may not happen with the above patch, but for the
> > > > hardening (ensuring it is always safe), I would like to add this.
> > > >
> > > > >
> > > > > because that's the one I can reproduce.. but I can
> > > > > certainly try that with your change as well
> > > >
> > > > Thank you!
> > >
> > > it did not help the idle warning as expected, but I did not
> > > see any problems running bpf tests on top of this
> >
> > Oops, right. I forgot this is only for the rethook, not protect the
> > fprobe handlers, since fprobe code doesn't involve the RCU code (it
> > depends on ftrace's check). Sorry about that.
> > Hmm, I need to add a test code for this issue, but that could be
> > solved by your noninstr patch.
> >
>
>
> Masami,
>
> It's not clear to me, do you intend to send a new revision with some
> more tests or this patch as is ready to go into bpf tree?

OK, let me make a test code against this issue. This may need a raw
fprobe test code (not a test case because it depends on that we can
trace the "arch_cpu_idle()"), but that test code won't work after
the "arch_cpu_idle()" is marked as noinstr (thus the test code will
only for the kernel which doesn't have the noinstr patch).
I want to add this check for the case if someone accidentally add
a function which is not covered by RCU and that is tracable by
fprobe (ftrace).
Thus this is a kind of preventative fix.

Thank you,

>
>
> > Thank you,
> >
> > >
> > > jirka
> > >
> > > >
> > > > >
> > > > > jirka
> > > > >
> > > > > >
> > > > > > Thank you,
> > > > > >
> > > > > >
> > > > > > > -- Steve
> > > > > > >
> > > > > > >
> > > > > > > > Since the rethook_recycle() will involve the call_rcu() for reclaiming
> > > > > > > > the rethook_instance, the rethook must be set up at the RCU available
> > > > > > > > context (non idle). This rethook_recycle() in the rethook trampoline
> > > > > > > > handler is inevitable, thus the RCU available check must be done before
> > > > > > > > setting the rethook trampoline.
> > > > > > > >
> > > > > > > > This adds a rcu_is_watching() check in the rethook_try_get() so that
> > > > > > > > it will return NULL if it is called when !rcu_is_watching().
> > > > > > > >
> > > > > > > > Fixes: 54ecbe6f1ed5 ("rethook: Add a generic return hook")
> > > > > > > > Signed-off-by: Masami Hiramatsu <[email protected]>
> > > > > > > > ---
> > > > > > > > kernel/trace/rethook.c | 9 +++++++++
> > > > > > > > 1 file changed, 9 insertions(+)
> > > > > > > >
> > > > > > > > diff --git a/kernel/trace/rethook.c b/kernel/trace/rethook.c
> > > > > > > > index b56833700d23..c69d82273ce7 100644
> > > > > > > > --- a/kernel/trace/rethook.c
> > > > > > > > +++ b/kernel/trace/rethook.c
> > > > > > > > @@ -154,6 +154,15 @@ struct rethook_node *rethook_try_get(struct rethook *rh)
> > > > > > > > if (unlikely(!handler))
> > > > > > > > return NULL;
> > > > > > > >
> > > > > > > > + /*
> > > > > > > > + * This expects the caller will set up a rethook on a function entry.
> > > > > > > > + * When the function returns, the rethook will eventually be reclaimed
> > > > > > > > + * or released in the rethook_recycle() with call_rcu().
> > > > > > > > + * This means the caller must be run in the RCU-availabe context.
> > > > > > > > + */
> > > > > > > > + if (unlikely(!rcu_is_watching()))
> > > > > > > > + return NULL;
> > > > > > > > +
> > > > > > > > fn = freelist_try_get(&rh->pool);
> > > > > > > > if (!fn)
> > > > > > > > return NULL;
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > Masami Hiramatsu (Google) <[email protected]>
> > > >
> > > >
> > > > --
> > > > Masami Hiramatsu (Google) <[email protected]>
> >
> >
> > --
> > Masami Hiramatsu (Google) <[email protected]>


--
Masami Hiramatsu (Google) <[email protected]>