From: Fernando Fernandez Mancera <hidden> Date: 2024-08-13 17:32:15
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
Thank you,
quoted hunk
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
From: Fernando F. Mancera <hidden> Date: 2024-08-25 17:06:32
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Thank you!
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
On Sun, 25 Aug 2024 19:06:22 +0200
"Fernando F. Mancera" [off-list ref] wrote:
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
quoted
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
quoted
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Yes, I think this is the better behavior for users too, rather than
silently droping some arguments.
We also need to handle this error code in callsite and log an error,
and add a testcase in the syntax error testcase. But that should be done
in another series. We should fix this first.
Thank you,
Thank you!
quoted
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
From: Fernando F. Mancera <hidden> Date: 2024-08-26 14:58:52
On 26/08/2024 01:56, Masami Hiramatsu (Google) wrote:
On Sun, 25 Aug 2024 19:06:22 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
quoted
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
quoted
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Yes, I think this is the better behavior for users too, rather than
silently droping some arguments.
We also need to handle this error code in callsite and log an error,
and add a testcase in the syntax error testcase. But that should be done
in another series. We should fix this first.
Thanks I will update the patch. If you don't mind I would also like to
do the follow up patch series.
Thank you,
quoted
Thank you!
quoted
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
From: Fernando F. Mancera <hidden> Date: 2024-09-04 22:27:06
Hi,
On 26/08/2024 01:56, Masami Hiramatsu (Google) wrote:
On Sun, 25 Aug 2024 19:06:22 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
quoted
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
quoted
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Yes, I think this is the better behavior for users too, rather than
silently droping some arguments.
We also need to handle this error code in callsite and log an error,
and add a testcase in the syntax error testcase. But that should be done
in another series. We should fix this first.
Just in case this fell through the cracks, I've sent a V2 patch. In
addition I also have a series ready for the logging and syntax error
testcase in kselftest but I didn't send it because it requires this
patch to be applied.
Thank you,
Fernando.
Thank you,
quoted
Thank you!
quoted
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
Hi,
On Thu, 5 Sep 2024 00:19:16 +0200
"Fernando F. Mancera" [off-list ref] wrote:
Hi,
On 26/08/2024 01:56, Masami Hiramatsu (Google) wrote:
quoted
On Sun, 25 Aug 2024 19:06:22 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
quoted
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
quoted
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Yes, I think this is the better behavior for users too, rather than
silently droping some arguments.
We also need to handle this error code in callsite and log an error,
and add a testcase in the syntax error testcase. But that should be done
in another series. We should fix this first.
Just in case this fell through the cracks, I've sent a V2 patch. In
addition I also have a series ready for the logging and syntax error
testcase in kselftest but I didn't send it because it requires this
patch to be applied.
I think that v2 was lost in my mailbox, but I found that in patchwork.
Let me pick it.
Thanks,
Thank you,
Fernando.
quoted
Thank you,
quoted
Thank you!
quoted
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
Hi Fernando,
I found the patch from Mikel solves the same issue and his v2 patch fixes
wider situation ("$arg*" BTF argument case). So I decided to pick his patch.
https://lore.kernel.org/all/20240930202656.292869-1-mikel@mikelr.com/
Let me know if there is any case the above fix does not cover.
Thank you,
On Tue, 22 Oct 2024 14:40:45 +0900
Masami Hiramatsu (Google) [off-list ref] wrote:
Hi,
On Thu, 5 Sep 2024 00:19:16 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 26/08/2024 01:56, Masami Hiramatsu (Google) wrote:
quoted
On Sun, 25 Aug 2024 19:06:22 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
quoted
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
quoted
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Yes, I think this is the better behavior for users too, rather than
silently droping some arguments.
We also need to handle this error code in callsite and log an error,
and add a testcase in the syntax error testcase. But that should be done
in another series. We should fix this first.
Just in case this fell through the cracks, I've sent a V2 patch. In
addition I also have a series ready for the logging and syntax error
testcase in kselftest but I didn't send it because it requires this
patch to be applied.
I think that v2 was lost in my mailbox, but I found that in patchwork.
Let me pick it.
Thanks,
quoted
Thank you,
Fernando.
quoted
Thank you,
quoted
Thank you!
quoted
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;
From: Fernando F. Mancera <hidden> Date: 2024-10-23 08:48:28
Hi,
On 22/10/2024 07:40, Masami Hiramatsu (Google) wrote:
Hi,
On Thu, 5 Sep 2024 00:19:16 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 26/08/2024 01:56, Masami Hiramatsu (Google) wrote:
quoted
On Sun, 25 Aug 2024 19:06:22 +0200
"Fernando F. Mancera" [off-list ref] wrote:
quoted
Hi,
On 25/08/2024 09:41, Masami Hiramatsu (Google) wrote:
quoted
Hi,
On Tue, 13 Aug 2024 13:25:40 -0400
Fernando Fernandez Mancera [off-list ref] wrote:
quoted
When initializing trace_probes::nr_args, make sure the maximum number of
probe arguments is honored. Oherwise, we can hit a NULL pointer
dereferences in multiple situations like on traceprobe_set_print_fmt().
Link: https://bugzilla.redhat.com/2303876
Sorry for replying later. I'm not sure why but I did not found this in my mbox...
No worries!
quoted
Anyway, trace_probe_init() should return -E2BIG in this case because
it is actuall wrong value.
Can you update your patch?
I agree this is the right solution but it would mean a change in
behavior. Is that fine? Before merging 035ba76014c0, it was writing up
to the maximum number of arguments and ignoring the rest. If you confirm
it is fine to change the behavior, I will update the patch.
Yes, I think this is the better behavior for users too, rather than
silently droping some arguments.
We also need to handle this error code in callsite and log an error,
and add a testcase in the syntax error testcase. But that should be done
in another series. We should fix this first.
Just in case this fell through the cracks, I've sent a V2 patch. In
addition I also have a series ready for the logging and syntax error
testcase in kselftest but I didn't send it because it requires this
patch to be applied.
I think that v2 was lost in my mailbox, but I found that in patchwork.
Let me pick it.
Sure, let me know if you need me to rebase it. Thanks for taking a look
to it :)
Thanks,
quoted
Thank you,
Fernando.
quoted
Thank you,
quoted
Thank you!
quoted
Thank you,
quoted
Fixes: 035ba76014c0 ("tracing/probes: cleanup: Set trace_probe::nr_args at trace_probe_init")
Signed-off-by: Fernando Fernandez Mancera <redacted>
---
kernel/trace/trace_probe.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
@@ -2043,10 +2043,14 @@ int trace_probe_init(struct trace_probe *tp, const char *event,gotoerror;}-tp->nr_args=nargs;+if(nargs>MAX_TRACE_ARGS)+tp->nr_args=MAX_TRACE_ARGS;+else+tp->nr_args=nargs;+/* Make sure pointers in args[] are NULL */if(nargs)-memset(tp->args,0,sizeof(tp->args[0])*nargs);+memset(tp->args,0,sizeof(tp->args[0])*tp->nr_args);return0;