core man page %e

core man page %e

[Jonny Grant]

Hello mtk

I've noticed "%e-%s.core" often gives something other than the filename and the name is truncated to 16 bytes:-

ThreadPoolServi-5.core
Chrome_InProcUt-5.core

This was set:
sysctl -w kernel.core_pattern="%e-%s.core"



In multithreaded applications it looks like %e is giving something other than the filename, perhaps a thread entry-point 
symbol name.

https://www.man7.org/linux/man-pages/man5/core.5.html

            %e  executable filename (without path prefix)

Could this be updated to:

            %e  executable filename or thread name truncated to 16 bytes

Cheers, Jonny

Re: core man page %e

[Jonny Grant]

Hi Mtk,

BTW, if I add "%P-%e-%s.core" every core starts with the same
"P1000-"

The man page shows:
  %P  PID of dumped process, as seen in the initial PID namespace
                (since Linux 3.12)

Do you see similar? I was only needing something to make the filename unique, so %t is enough.

Cheers, Jonny


On 30/05/2020 11:39, Jonny Grant wrote:
> Hello mtk
> 
> I've noticed "%e-%s.core" often gives something other than the filename and the name is truncated to 16 bytes:-
> 
> ThreadPoolServi-5.core
> Chrome_InProcUt-5.core
> 
> This was set:
> sysctl -w kernel.core_pattern="%e-%s.core"
> 
> 
> 
> In multithreaded applications it looks like %e is giving something other than the filename, perhaps a thread entry-point 
> symbol name.
> 
> https://www.man7.org/linux/man-pages/man5/core.5.html
> 
>             %e  executable filename (without path prefix)
> 
> Could this be updated to:
> 
>             %e  executable filename or thread name truncated to 16 bytes
> 
> Cheers, Jonny

Re: core man page %e

[Michael Kerrisk (man-pages)]

On 6/1/20 12:21 PM, Jonny Grant wrote:
> Hi Mtk,
> 
> BTW, if I add "%P-%e-%s.core" every core starts with the same
> "P1000-"
> 
> The man page shows:
>   %P  PID of dumped process, as seen in the initial PID namespace
>                 (since Linux 3.12)
> 
> Do you see similar? I was only needing something to make the filename
> unique, so %t is enough.
I don't see the same. What do you see when you verify the change with:

sysctl kernel.core_pattern

Thanks,

Michael


-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

Re: core man page %e

[Jonny Grant]

On 09/06/2020 09:25, Michael Kerrisk (man-pages) wrote:
> On 6/1/20 12:21 PM, Jonny Grant wrote:
>> Hi Mtk,
>>
>> BTW, if I add "%P-%e-%s.core" every core starts with the same
>> "P1000-"
>>
>> The man page shows:
>>    %P  PID of dumped process, as seen in the initial PID namespace
>>                  (since Linux 3.12)
>>
>> Do you see similar? I was only needing something to make the filename
>> unique, so %t is enough.
> I don't see the same. What do you see when you verify the change with:
> 
> sysctl kernel.core_pattern
> 
> Thanks,
> 
> Michael
> 

It must be some mistake I made, can't reproduce it today

Jonny

Re: core man page %e

[Michael Kerrisk (man-pages)]

Hello Jonny,

On 5/30/20 12:39 PM, Jonny Grant wrote:
> Hello mtk
> 
> I've noticed "%e-%s.core" often gives something other than the
> filename and the name is truncated to 16 bytes:-
> 
> ThreadPoolServi-5.core
> Chrome_InProcUt-5.core
> 
> This was set:
> sysctl -w kernel.core_pattern="%e-%s.core"
> 
> 
> 
> In multithreaded applications it looks like %e is giving something
> other than the filename, perhaps a thread entry-point symbol name.> 
> https://www.man7.org/linux/man-pages/man5/core.5.html
> 
>             %e  executable filename (without path prefix)
> 
> Could this be updated to:
> 
>             %e  executable filename or thread name truncated to 16 bytes

I changed it to this:

           %e  The process or thread's comm value, which typically is the
               same  as the executable filename (without path prefix, and
               truncated to a maximum of 15  characters),  but  may  have
               been  modified  to be something different; see the discus‐
               sion of /proc/[pid]/comm  and  /proc/[pid]/task/[tid]/comm
               in proc(5).

Thanks,

Michael

-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

Re: core man page %e

[Jonny Grant]

On 09/06/2020 11:18, Michael Kerrisk (man-pages) wrote:
> Hello Jonny,
> 
> On 5/30/20 12:39 PM, Jonny Grant wrote:
>> Hello mtk
>>
>> I've noticed "%e-%s.core" often gives something other than the
>> filename and the name is truncated to 16 bytes:-
>>
>> ThreadPoolServi-5.core
>> Chrome_InProcUt-5.core
>>
>> This was set:
>> sysctl -w kernel.core_pattern="%e-%s.core"
>>
>>
>>
>> In multithreaded applications it looks like %e is giving something
>> other than the filename, perhaps a thread entry-point symbol name.>
>> https://www.man7.org/linux/man-pages/man5/core.5.html
>>
>>              %e  executable filename (without path prefix)
>>
>> Could this be updated to:
>>
>>              %e  executable filename or thread name truncated to 16 bytes
> 
> I changed it to this:
> 
>             %e  The process or thread's comm value, which typically is the
>                 same  as the executable filename (without path prefix, and
>                 truncated to a maximum of 15  characters),  but  may  have
>                 been  modified  to be something different; see the discus‐
>                 sion of /proc/[pid]/comm  and  /proc/[pid]/task/[tid]/comm
>                 in proc(5).

Hi Michael

That looks good.

I saw $ top -H shows these similar 15 byte truncated names. It's a shame it is such a low limit.

19697 jonny     20   0  657252  46120  29096 S  2.6  0.3   0:18.84 gnome-terminal- 
 
 

19431 jonny     20   0 4823508 114384  75668 S  1.6  0.7   0:00.08 ThreadPoolForeg


Is it better to say "15 bytes" because utf8 binary filenames or thread names would be less characters?

For example:

$ ./おはようございます
$ cat /proc/892/comm
おはようご

That is 15 bytes of utf8
E3 81 8A E3 81 AF E3 82 88 E3 81 86 E3 81 94 0A

Cheers, Jonny