2016-11-01 20:06:22

by Thierry Moreau

Subject: Bus error on shm_open/ftruncate/mmap after changing kernel boot partition

Strange bug after changing the boot partition (the previous was on a
hard disk with a few media errors).

I must apologize that the old partition is re-formatted. (As an excuse
the other services on this server system work fine with the new partition.)

Anyway, the bogus code sequence is below. Should work fine everywhere,
modulo write permission on /dev/shm.

Same observation with privileged execution.

uname -a reports:

Linux hostname 4.1.3 #15 SMP Sun Aug 30 03:01:12 UTC 2015 x86_64
Intel(R) Pentium(R) CPU G3250 @ 3.20GHz GenuineIntel GNU/Linux

Obviously, any help appreciated.

Regards

- Thierry Moreau


============
/* mmap_test.cpp --
*/

#include <cstdio>
#include <cstring>
#include <cerrno>
#include <sys/mman.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>

int main(int argc, char *argv[])
{
int shm_fd;
shm_fd=shm_open(/*"/dev/shm/"*/"test",
O_RDWR | O_CREAT | O_EXCL,
S_IRUSR|S_IWUSR | S_IRGRP|S_IWGRP);
if (shm_fd>=0) {
if (ftruncate(shm_fd,0x20000)==0) {
void *region=mmap(0,0x20000,
PROT_READ|PROT_WRITE,MAP_SHARED,
shm_fd,0);
if (MAP_FAILED!=region) {
fprintf(stderr,"%d %p\n",__LINE__,region);
fprintf(stderr,"%d %d\n",__LINE__,*((int *)region));
/* ===== this triggers a bus error on the kernel instance */
}
else {
fprintf(stderr,"%d %d %s\n",__LINE__,errno,strerror(errno));
}
}
else {
fprintf(stderr,"%d %d %s\n",__LINE__,errno,strerror(errno));
}
}
else {
fprintf(stderr,"%d %d %s\n",__LINE__,errno,strerror(errno));
}
return errno;
}
============

2016-11-01 23:36:53

by Thierry Moreau

Subject: Re: Bus error on shm_open/ftruncate/mmap after changing kernel boot partition: FIXED

Fixed (but still puzzled) -- see below.

On 01/11/16 07:57 PM, Thierry Moreau wrote:
> Strange bug after changing the boot partition (the previous was on a
> hard disk with a few media errors).
>

Ah!

mounting tmpfs on /dev/shm (manually or through fstab) fixes this.

I am somehow puzzled as I think I observe the test program working
without this mount on some kernel instances.

> I must apologize that the old partition is re-formatted. (As an excuse
> the other services on this server system work fine with the new partition.)
>
> Anyway, the bogus code sequence is below. Should work fine everywhere,
> modulo write permission on /dev/shm.
>
> Same observation with privileged execution.
>
> uname -a reports:
>
> Linux hostname 4.1.3 #15 SMP Sun Aug 30 03:01:12 UTC 2015 x86_64
> Intel(R) Pentium(R) CPU G3250 @ 3.20GHz GenuineIntel GNU/Linux
>
> Obviously, any help appreciated.
>
> Regards
>
> - Thierry Moreau
>
>
> ============
> /* mmap_test.cpp --
> */
>
> #include <cstdio>
> #include <cstring>
> #include <cerrno>
> #include <sys/mman.h>
> #include <sys/stat.h>
> #include <fcntl.h>
> #include <unistd.h>
>
> int main(int argc, char *argv[])
> {
> int shm_fd;
> shm_fd=shm_open(/*"/dev/shm/"*/"test",
> O_RDWR | O_CREAT | O_EXCL,
> S_IRUSR|S_IWUSR | S_IRGRP|S_IWGRP);
> if (shm_fd>=0) {
> if (ftruncate(shm_fd,0x20000)==0) {
> void *region=mmap(0,0x20000,
> PROT_READ|PROT_WRITE,MAP_SHARED,
> shm_fd,0);
> if (MAP_FAILED!=region) {
> fprintf(stderr,"%d %p\n",__LINE__,region);
> fprintf(stderr,"%d %d\n",__LINE__,*((int *)region));
> /* ===== this triggers a bus error on the kernel instance */
> }
> else {
> fprintf(stderr,"%d %d %s\n",__LINE__,errno,strerror(errno));
> }
> }
> else {
> fprintf(stderr,"%d %d %s\n",__LINE__,errno,strerror(errno));
> }
> }
> else {
> fprintf(stderr,"%d %d %s\n",__LINE__,errno,strerror(errno));
> }
> return errno;
> }
> ============
>