2018-07-10 18:28:56

by William Cohen

Subject: [PATCH] Check jvmti_agent snprintf return value to avoid build failures with GCC-8.1.1

Newer versions of GCC perform static analysis to determine whether
string truncation is possible with functions such as snprintf and
provide a warning if truncation could occur. The make for
jvmti_agent.c uses the compiler option that treats any compiler
warnings as compiler errors. For GCC-8.1.1 in Fedora 28 this causes
the build to fail. The return value of the snprint is now checked to
ensure snprintf produced a NULL-terminated string. If the string for
the path is invalid, the code does attempt to use the string.

Signed-off-by: William Cohen <[email protected]>
---
tools/perf/jvmti/jvmti_agent.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/tools/perf/jvmti/jvmti_agent.c b/tools/perf/jvmti/jvmti_agent.c
index 0c6d1002b524..30f14eafe4b3 100644
--- a/tools/perf/jvmti/jvmti_agent.c
+++ b/tools/perf/jvmti/jvmti_agent.c
@@ -227,7 +227,7 @@ void *jvmti_open(void)
{
char dump_path[PATH_MAX];
struct jitheader header;
- int fd;
+ int retlen, fd;
FILE *fp;

init_arch_timestamp();
@@ -249,7 +249,10 @@ void *jvmti_open(void)
/*
* jitdump file name
*/
- snprintf(dump_path, PATH_MAX, "%s/jit-%i.dump", jit_path, getpid());
+ retlen = snprintf(dump_path, PATH_MAX, "%s/jit-%i.dump",
+ jit_path, getpid());
+ if (retlen <= 0 || ((int) sizeof(dump_path)) <= retlen)
+ return NULL;

fd = open(dump_path, O_CREAT|O_TRUNC|O_RDWR, 0666);
if (fd == -1)
--
2.17.1

2018-07-10 22:59:16

by Jiri Olsa

Subject: Re: [PATCH] Check jvmti_agent snprintf return value to avoid build failures with GCC-8.1.1

On Tue, Jul 10, 2018 at 02:27:16PM -0400, William Cohen wrote:
> Newer versions of GCC perform static analysis to determine whether
> string truncation is possible with functions such as snprintf and
> provide a warning if truncation could occur. The make for
> jvmti_agent.c uses the compiler option that treats any compiler
> warnings as compiler errors. For GCC-8.1.1 in Fedora 28 this causes
> the build to fail. The return value of the snprint is now checked to
> ensure snprintf produced a NULL-terminated string. If the string for
> the path is invalid, the code does attempt to use the string.

hi,
I posted fix for this recently:
https://lore.kernel.org/lkml/[email protected]/

it also covers the perf_regs.c, which was failing with gcc8 for me

should be pulled in soon

thanks,
jirka

>
> Signed-off-by: William Cohen <[email protected]>
> ---
> tools/perf/jvmti/jvmti_agent.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/jvmti/jvmti_agent.c b/tools/perf/jvmti/jvmti_agent.c
> index 0c6d1002b524..30f14eafe4b3 100644
> --- a/tools/perf/jvmti/jvmti_agent.c
> +++ b/tools/perf/jvmti/jvmti_agent.c
> @@ -227,7 +227,7 @@ void *jvmti_open(void)
> {
> char dump_path[PATH_MAX];
> struct jitheader header;
> - int fd;
> + int retlen, fd;
> FILE *fp;
>
> init_arch_timestamp();
> @@ -249,7 +249,10 @@ void *jvmti_open(void)
> /*
> * jitdump file name
> */
> - snprintf(dump_path, PATH_MAX, "%s/jit-%i.dump", jit_path, getpid());
> + retlen = snprintf(dump_path, PATH_MAX, "%s/jit-%i.dump",
> + jit_path, getpid());
> + if (retlen <= 0 || ((int) sizeof(dump_path)) <= retlen)
> + return NULL;
>
> fd = open(dump_path, O_CREAT|O_TRUNC|O_RDWR, 0666);
> if (fd == -1)
> --
> 2.17.1
>

2018-07-11 00:11:43

by William Cohen

Subject: Re: [PATCH] Check jvmti_agent snprintf return value to avoid build failures with GCC-8.1.1

On 07/10/2018 06:58 PM, Jiri Olsa wrote:
> On Tue, Jul 10, 2018 at 02:27:16PM -0400, William Cohen wrote:
>> Newer versions of GCC perform static analysis to determine whether
>> string truncation is possible with functions such as snprintf and
>> provide a warning if truncation could occur. The make for
>> jvmti_agent.c uses the compiler option that treats any compiler
>> warnings as compiler errors. For GCC-8.1.1 in Fedora 28 this causes
>> the build to fail. The return value of the snprint is now checked to
>> ensure snprintf produced a NULL-terminated string. If the string for
>> the path is invalid, the code does attempt to use the string.
>
> hi,
> I posted fix for this recently:
> https://lore.kernel.org/lkml/[email protected]/
>
> it also covers the perf_regs.c, which was failing with gcc8 for me
>
> should be pulled in soon
>
> thanks,
> jirka

Hi Jirka,

Thanks. I hadn't seen the patch, and this failing to build for the past week has been bugging me. Glad to hear there is already a fix queued up.

-Will
>
>>
>> Signed-off-by: William Cohen <[email protected]>
>> ---
>> tools/perf/jvmti/jvmti_agent.c | 7 +++++--
>> 1 file changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/perf/jvmti/jvmti_agent.c b/tools/perf/jvmti/jvmti_agent.c
>> index 0c6d1002b524..30f14eafe4b3 100644
>> --- a/tools/perf/jvmti/jvmti_agent.c
>> +++ b/tools/perf/jvmti/jvmti_agent.c
>> @@ -227,7 +227,7 @@ void *jvmti_open(void)
>> {
>> char dump_path[PATH_MAX];
>> struct jitheader header;
>> - int fd;
>> + int retlen, fd;
>> FILE *fp;
>>
>> init_arch_timestamp();
>> @@ -249,7 +249,10 @@ void *jvmti_open(void)
>> /*
>> * jitdump file name
>> */
>> - snprintf(dump_path, PATH_MAX, "%s/jit-%i.dump", jit_path, getpid());
>> + retlen = snprintf(dump_path, PATH_MAX, "%s/jit-%i.dump",
>> + jit_path, getpid());
>> + if (retlen <= 0 || ((int) sizeof(dump_path)) <= retlen)
>> + return NULL;
>>
>> fd = open(dump_path, O_CREAT|O_TRUNC|O_RDWR, 0666);
>> if (fd == -1)
>> --
>> 2.17.1
>>