Hi Steven,
I tested it and it prevents the kernel crash I am witnessing.
As for the side-effect that Masami mentioned regarding not being able to probe
function inside the trace_kprobe.c file, I suggest we move the target
function in
its own separate compile unit so it can be compiled with the ftrace cflags.
See patch below.
Thanks
Francis
>From d5a3645bd0046f28275d6b60207958f2751c1f47 Mon Sep 17 00:00:00 2001
From: Francis Deslauriers <francis.deslauri...@efficios.com>
Date: Wed, 11 Jul 2018 12:34:22 -0400
Subject: [PATCH] selftest/ftrace: Move kprobe selftest function to separate
compile unit
Move selftest function to its own compile unit so it can be compiled
with the ftrace cflags (CC_FLAGS_FTRACE) allowing it to be probed
during the ftrace startup tests.
Signed-off-by: Francis Deslauriers <francis.deslauri...@efficios.com>
---
kernel/trace/Makefile | 5 +++++
kernel/trace/trace_kprobe.c | 12 +-----------
kernel/trace/trace_kprobe_selftest.c | 10 ++++++++++
kernel/trace/trace_kprobe_selftest.h | 7 +++++++
4 files changed, 23 insertions(+), 11 deletions(-)
create mode 100644 kernel/trace/trace_kprobe_selftest.c
create mode 100644 kernel/trace/trace_kprobe_selftest.h
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index e2538c7..e38771e 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -13,6 +13,11 @@ obj-y += trace_selftest_dynamic.o
endif
endif
+ifdef CONFIG_FTRACE_STARTUP_TEST
+CFLAGS_trace_kprobe_selftest.o = $(CC_FLAGS_FTRACE)
+obj-$(CONFIG_KPROBE_EVENTS) += trace_kprobe_selftest.o
+endif
+
# If unlikely tracing is enabled, do not trace these files
ifdef CONFIG_TRACING_BRANCHES
KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
index 952dc2a..3fe966f 100644
--- a/kernel/trace/trace_kprobe.c
+++ b/kernel/trace/trace_kprobe.c
@@ -24,6 +24,7 @@
#include <linux/error-injection.h>
#include "trace_probe.h"
+#include "trace_kprobe_selftest.h"
#define KPROBE_EVENT_SYSTEM "kprobes"
#define KRETPROBE_MAXACTIVE_MAX 4096
@@ -1560,17 +1561,6 @@ fs_initcall(init_kprobe_trace);
#ifdef CONFIG_FTRACE_STARTUP_TEST
-/*
- * The "__used" keeps gcc from removing the function symbol
- * from the kallsyms table. 'noinline' makes sure that there
- * isn't an inlined version used by the test method below
- */
-static __used __init noinline int
-kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
-{
- return a1 + a2 + a3 + a4 + a5 + a6;
-}
-
static __init struct trace_event_file *
find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
{
diff --git a/kernel/trace/trace_kprobe_selftest.c
b/kernel/trace/trace_kprobe_selftest.c
new file mode 100644
index 0000000..a3d2090
--- /dev/null
+++ b/kernel/trace/trace_kprobe_selftest.c
@@ -0,0 +1,10 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Function used during the kprobe self test. This function is in a seperate
+ * compile unit so it can be compile with CC_FLAGS_FTRACE to ensure that it
+ * can be probed by the selftests.
+ */
+int kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int
a5, int a6)
+{
+ return a1 + a2 + a3 + a4 + a5 + a6;
+}
diff --git a/kernel/trace/trace_kprobe_selftest.h
b/kernel/trace/trace_kprobe_selftest.h
new file mode 100644
index 0000000..9243d4e
--- /dev/null
+++ b/kernel/trace/trace_kprobe_selftest.h
@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Function used during the kprobe self test. This function is in a seperate
+ * compile unit so it can be compile with CC_FLAGS_FTRACE to ensure that it
+ * can be probed by the selftests.
+ */
+int kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int
a5, int a6);
--
2.7.4
Le mar. 3 juill. 2018, à 18 h 31, Steven Rostedt <rost...@goodmis.org> a écrit :
>
> Mathieu and Francis,
>
> Looking back, this thread never got further. Would Masami's patch work
> for you?
>
> -- Steve
>
>
> On Sat, 17 Mar 2018 10:22:11 +0900
> Masami Hiramatsu <mhira...@kernel.org> wrote:
>
> > On Sat, 17 Mar 2018 09:13:34 +0900
> > Masami Hiramatsu <mhira...@kernel.org> wrote:
> >
> > > On Fri, 16 Mar 2018 13:53:01 -0400 (EDT)
> > > Mathieu Desnoyers <mathieu.desnoy...@efficios.com> wrote:
> > >
> > > > ----- On Mar 16, 2018, at 12:48 PM, rostedt rost...@goodmis.org wrote:
> > > >
> > > > > On Fri, 16 Mar 2018 12:41:34 -0400
> > > > > Steven Rostedt <rost...@goodmis.org> wrote:
> > > > >
> > > > >> Yes, kprobes are dangerous. I'm not saying it shouldn't be fixed, I'm
> > > > >> saying that I don't have time to fix it now, but would be happy to
> > > > >> accept patches if someone else does so.
> > > > >
> > > > > And looking at what I replied before for the original patch. It would
> > > > > probably be a good idea to blacklist directories. Like we do with
> > > > > function tracing. We probably should black list both kernel/tracing
> > > > > and
> > > > > kernel/events from being probed.
> > > > >
> > > > > Did this come up at plumbers? You were there too, I don't remember
> > > > > discussing it there.
> > > >
> > > > I don't remember this coming up last Plumbers nor KS neither, given
> > > > that we were focused on other topics.
> > > >
> > > > Would the general approach you envision be based on emitting all code
> > > > generated by compilation of all objects under kernel/tracing and
> > > > kernel/events into a specific "nokprobes" text section of the kernel ?
> > > > Perhaps we could create a specific linker scripts for those directories,
> > > > or do you have in mind a neater way to do this ?
> > >
> > > .kprobes.text section still exists. As I pointed in previous mail, I don't
> > > think we have to put all those code into that section. But if you want,
> > > it is acceptable to have a kconfig which push most of those ftrace related
> > > code into .kprobes.text section.
> >
> > Or, we can check it by ftrace_location_range() as below patch.
> >
> > Note that as a side-effect, we can not trace functions in trace_kprobe.c
> > anymore, and this means it is hard to me to make a testcase of kprobe
> > events.
> > It was the easiest (and maintainable) way to make test cases... sigh.
> >
> > -----
> > tracing: kprobes: Prohibit probing on notrace function
> >
> > From: Masami Hiramatsu <mhira...@kernel.org>
> >
> > Prohibit kprobe-events probing on notrace function.
> > Since probing on the notrace function can cause recursive
> > event call. In most case those are just skipped, but
> > in some case it falls into infinit recursive call.
> >
> > Signed-off-by: Masami Hiramatsu <mhira...@kernel.org>
> > ---
> > kernel/trace/trace_kprobe.c | 23 +++++++++++++++++++++++
> > 1 file changed, 23 insertions(+)
> >
> > diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
> > index 5ce9b8cf7be3..7404b012e51a 100644
> > --- a/kernel/trace/trace_kprobe.c
> > +++ b/kernel/trace/trace_kprobe.c
> > @@ -487,6 +487,23 @@ disable_trace_kprobe(struct trace_kprobe *tk, struct
> > trace_event_file *file)
> > return ret;
> > }
> >
> > +#ifdef CONFIG_KPROBES_ON_FTRACE
> > +static bool within_notrace_func(struct trace_kprobe *tk)
> > +{
> > + unsigned long offset, size, addr;
> > +
> > + addr = kallsyms_lookup_name(trace_kprobe_symbol(tk));
> > + addr += trace_kprobe_offset(tk);
> > +
> > + if (!kallsyms_lookup_size_offset(addr, &size, &offset))
> > + return true; /* Out of range. */
> > +
> > + return !ftrace_location_range(addr - offset, addr - offset + size);
> > +}
> > +#else
> > +#define within_notrace_func(tk) (false)
> > +#endif
> > +
> > /* Internal register function - just handle k*probes and flags */
> > static int __register_trace_kprobe(struct trace_kprobe *tk)
> > {
> > @@ -495,6 +512,12 @@ static int __register_trace_kprobe(struct trace_kprobe
> > *tk)
> > if (trace_probe_is_registered(&tk->tp))
> > return -EINVAL;
> >
> > + if (within_notrace_func(tk)) {
> > + pr_warn("Could not probe notrace function %s\n",
> > + trace_kprobe_symbol(tk));
> > + return -EINVAL;
> > + }
> > +
> > for (i = 0; i < tk->tp.nr_args; i++)
> > traceprobe_update_arg(&tk->tp.args[i]);
> >
> >
>
--
Francis Deslauriers
Software developer
EfficiOS inc.
