Date:   Mon, 15 Apr 2019 14:08:04 +0200 (CEST)
From:   John Kacur <jkacur@...hat.com>
To:     Stephen Rostedt <rostedt@...dmis.org>
cc:     linux-kernel@...r.kernel.org, Josef Bacik <jbacik@...com>
Subject: Re: BUG: trace-cmd: trace-cmd record -e failure



On Mon, 15 Apr 2019, John Kacur wrote:

> This is a long standing but in trace-cmd.
> 
> [root@...-0 tmp]# trace-cmd record -p function -e sched:* sleep 1
>   plugin 'function'
> none
> trace-cmd: Invalid argument
>   Failed filter of 
> /sys/kernel/debug/tracing/events/sched/sched_kthread_stop/filter
> 
> trace-cmd: No such file or directory
>   can not stat 'trace.dat.cpu0'
> 
> -------------------------------------------
> 
> This only happens on large machines, I can only get it to trigger at 
> around 200 cpus. However, it's not complicated, it's simply we're writing
> to the filter file, one per recorder thread until the size of the string
> exceeds the limit.
> 
> We're writing something crazy like this.
> 
> str = 
> (common_pid!=50883)||(common_pid!=50882)||(common_pid!=50881)||(common_pid!=50880)||(common_pid!=50879)||(common_pid!=50878)||(common_pid!=50877)||(common_pid!=50876)||(common_pid!=50875)||(common_pid!=50874)||(common_pid!=50873)||(common_pid!=50872)||(common_pid!=50871)||(common_pid!=50870)||(common_pid!=50869)||(common_pid!=50868)||(common_pid!=50867)||(common_pid!=50866)||(common_pid!=50865)||(common_pid!=50864)||(common_pid!=50863)||(common_pid!=50862)||(common_pid!=50861)||(common_pid!=50860)||(common_pid!=50859)||(common_pid!=50858)||(common_pid!=50857)||(common_pid!=50856)||(common_pid!=50855)||(common_pid!=50854)||(common_pid!=50853)||(common_pid!=50852)||(common_pid!=50851)||(common_pid!=50850)||(common_pid!=50849)||(common_pid!=50848)||(common_pid!=50847)||(common_pid!=50846)||(common_pid!=50845)||(common_pid!=50844)||(common_pid!=50843)||(common_pid!=50842)||(common_pid!=50841)||(common_pid!=50840)||(common_pid!=50839)||(common_pid!=50838)||(common_pid!=50837)||(common_pid!=50836)||(common_pid!=50835)||(common_pid!=50834)||(common_pid!=50833)||(common_pid!=50832)||(common_pid!=50831)||(common_pid!=50830)||(common_pid!=50829)||(common_pid!=50828)||(common_pid!=50827)||(common_pid!=50826)||(common_pid!=50825)||(common_pid!=50824)||(common_pid!=50823)||(common_pid!=50822)||(common_pid!=50821)||(common_pid!=50820)||(common_pid!=50819)||(common_pid!=50818)||(common_pid!=50817)||(common_pid!=50816)||(common_pid!=50815)||(common_pid!=50814)||(common_pid!=50813)||(common_pid!=50812)||(common_pid!=50811)||(common_pid!=50810)||(common_pid!=50809)||(common_pid!=50808)||(common_pid!=50807)||(common_pid!=50806)||(common_pid!=50805)||(common_pid!=50804)||(common_pid!=50803)||(common_pid!=50802)||(common_pid!=50801)||(common_pid!=50800)||(common_pid!=50799)||(common_pid!=50798)||(common_pid!=50797)||(common_pid!=50796)||(common_pid!=50795)||(common_pid!=50794)||(common_pid!=50793)||(common_pid!=50792)||(common_pid!=50791)||(common_pid!=50790)||(common_pid!=50789)||(common_pid!=50788)||(common_pid!=50787)||(common_pid!=50786)||(common_pid!=50785)||(common_pid!=50784)||(common_pid!=50783)||(common_pid!=50782)||(common_pid!=50781)||(common_pid!=50780)||(common_pid!=50779)||(common_pid!=50778)||(common_pid!=50777)||(common_pid!=50776)||(common_pid!=50775)||(common_pid!=50774)||(common_pid!=50773)||(common_pid!=50772)||(common_pid!=50771)||(common_pid!=50770)||(common_pid!=50769)||(common_pid!=50768)||(common_pid!=50767)||(common_pid!=50766)||(common_pid!=50765)||(common_pid!=50764)||(common_pid!=50763)||(common_pid!=50762)||(common_pid!=50761)||(common_pid!=50760)||(common_pid!=50759)||(common_pid!=50758)||(common_pid!=50757)||(common_pid!=50756)||(common_pid!=50755)||(common_pid!=50754)||(common_pid!=50753)||(common_pid!=50752)||(common_pid!=50751)||(common_pid!=50750)||(common_pid!=50749)||(common_pid!=50748)||(common_pid!=50747)||(common_pid!=50746)||(common_pid!=50745)||(common_pid!=50744)||(common_pid!=50743)||(common_pid!=50742)||(common_pid!=50741)||(common_pid!=50740)||(common_pid!=50739)||(common_pid!=50738)||(common_pid!=50737)||(common_pid!=50736)||(common_pid!=50735)||(common_pid!=50734)||(common_pid!=50733)||(common_pid!=50732)||(common_pid!=50731)||(common_pid!=50730)||(common_pid!=50729)||(common_pid!=50728)||(common_pid!=50727)||(common_pid!=50726)||(common_pid!=50725)||(common_pid!=50724)||(common_pid!=50723)||(common_pid!=50722)||(common_pid!=50721)||(common_pid!=50720)||(common_pid!=50719)||(common_pid!=50718)||(common_pid!=50717)||(common_pid!=50716)||(common_pid!=50715)||(common_pid!=50714)||(common_pid!=50713)||(common_pid!=50712)||(common_pid!=50711)||(common_pid!=50710)||(common_pid!=50709)||(common_pid!=50708)||(common_pid!=50707)||(common_pid!=50706)||(common_pid!=50705)||(common_pid!=50704)||(common_pid!=50703)||(common_pid!=50702)||(common_pid!=50701)||(common_pid!=50700)||(common_pid!=50699)||(common_pid!=50698)||(common_pid!=50697)||(common_pid!=50696)||(common_pid!=50695)||(common_pid!=50694)||(common_pid!=50693)||(common_pid!=50692)||(common_pid!=50691)||(common_pid!=50690)||(common_pid!=50689)||(common_pid!=50688)||(common_pid!=50687)||(common_pid!=50686)||(common_pid!=50685)||(common_pid!=50684)||(common_pid!=50683)||(common_pid!=50682)||(common_pid!=50681)||(common_pid!=50680)||(common_pid!=50679)||(common_pid!=50678)||(common_pid!=50677)||(common_pid!=50676)||(common_pid!=50675)||(common_pid!=50674)||(common_pid!=50673)||(common_pid!=50672)||(common_pid!=50671)||(common_pid!=50670)||(common_pid!=50669)||(common_pid!=50668)||(common_pid!=50667)||(common_pid!=50666)||(common_pid!=50665)||(common_pid!=50664)||(common_pid!=50663)||(common_pid!=50662)||(common_pid!=50661)||(common_pid!=50660)||(common_pid!=50659)||(common_pid!=50658)||(common_pid!=50657)||(common_pid!=50656)||(common_pid!=50655)||(common_pid!=50654)||(common_pid!=50653)||(common_pid!=50652)||(common_pid!=50651)||(common_pid!=50650)||(common_pid!=50649)||(common_pid!=50648)||(common_pid!=50647)||(common_pid!=50646)||(common_pid!=50645)||(common_pid!=50644)||(common_pid!=50643)||(common_pid!=50642)||(common_pid!=50641)||(common_pid!=50640)||(common_pid!=50639)||(common_pid!=50638)||(common_pid!=50637)||(common_pid!=50636)||(common_pid!=50635)||(common_pid!=50634)||(common_pid!=50633)||(common_pid!=50632)||(common_pid!=50631)||(common_pid!=50630)||(common_pid!=50629)||(common_pid!=50628)||(common_pid!=50627)||(common_pid!=50626)||(common_pid!=50625)||(common_pid!=50624)||(common_pid!=50623)||(common_pid!=50622)||(common_pid!=50621)||(common_pid!=50620)||(common_pid!=50619)||(common_pid!=50618)||(common_pid!=50617)||(common_pid!=50616)||(common_pid!=50615)||(common_pid!=50614)||(common_pid!=50613)||(common_pid!=50612)||(common_pid!=50611)||(common_pid!=50610)||(common_pid!=50609)||(common_pid!=50608)||(common_pid!=50607)||(common_pid!=50606)||(common_pid!=50605)||(common_pid!=50604)||(common_pid!=50603)||(common_pid!=50602)||(common_pid!=50601)||(common_pid!=50600)||(common_pid!=50599)||(common_pid!=50598)||(common_pid!=50597)||(common_pid!=50596), 
> strlen = 6046b
> 
> on an older version of traceback (but this bug still exists in the 
> latest), I did a gdb session like this
> 
> Breakpoint 2, write_file (
>     file=0x670e40 
> "/sys/kernel/debug/tracing/events/sched/sched_kthread_stop/filter", 
>     str=0x67f8c0 
> "(common_pid!=21759)||(common_pid!=21758)||(common_pid!=21757)||(common_pid!=21756)||(common_pid!=21755)||(common_pid!=21754)||(common_pid!=21753)||(common_pid!=21752)||(common_pid!=21751)||(common_pid"..., 
>     type=0x447dc4 "filter") at /root/trace-cmd/trace-record.c:1586
> 1586			die("Failed %s of %s\n", type, file);
> (gdb) print file
> $7 = 0x670e40 
> "/sys/kernel/debug/tracing/events/sched/sched_kthread_stop/filter"
> (gdb) bt
> #0  write_file (
>     file=0x670e40 
> "/sys/kernel/debug/tracing/events/sched/sched_kthread_stop/filter", 
>     str=0x67f8c0 
> "(common_pid!=21759)||(common_pid!=21758)||(common_pid!=21757)||(common_pid!=21756)||(common_pid!=21755)||(common_pid!=21754)||(common_pid!=21753)||(common_pid!=21752)||(common_pid!=21751)||(common_pid"..., 
>     type=0x447dc4 "filter") at /root/trace-cmd/trace-record.c:1586
> #1  0x000000000040abfd in write_filter (
>     file=0x670e40 
> "/sys/kernel/debug/tracing/events/sched/sched_kthread_stop/filter", 
>     filter=0x67f8c0 
> "(common_pid!=21759)||(common_pid!=21758)||(common_pid!=21757)||(common_pid!=21756)||(common_pid!=21755)||(common_pid!=21754)||(common_pid!=21753)||(common_pid!=21752)||(common_pid!=21751)||(common_pid"...)
>     at /root/trace-cmd/trace-record.c:1652
> #2  0x000000000040af81 in update_event (event=0x670d30, 
>     filter=0x67f8c0 
> "(common_pid!=21759)||(common_pid!=21758)||(common_pid!=21757)||(common_pid!=21756)||(common_pid!=21755)||(common_pid!=21754)||(common_pid!=21753)||(common_pid!=21752)||(common_pid!=21751)||(common_pid"..., 
>     filter_only=1, update=49 '1') at /root/trace-cmd/trace-record.c:1785
> #3  0x000000000040b653 in update_event_filters (
>     instance=0x65e5e0 <top_instance>) at 
> /root/trace-cmd/trace-record.c:2016
> #4  0x000000000040b83b in update_pid_event_filters (
>     instance=0x65e5e0 <top_instance>) at 
> /root/trace-cmd/trace-record.c:2081
> #5  0x0000000000409607 in update_task_filter ()
>     at /root/trace-cmd/trace-record.c:971
> #6  0x00000000004118be in trace_record (argc=6, argv=0x7fffffffe0f8)
>     at /root/trace-cmd/trace-record.c:4818
> #7  0x0000000000407cb7 in main (argc=6, argv=0x7fffffffe0f8)
>     at /root/trace-cmd/trace-cmd.c:133
> 
> Note even if you don't invoke the filter, you still write the sting to the 
> file.
> 
> Not sure how to fix this, but conceptually it's not hard, we need to add 
> some error checking and stop writing to the filter if we exceed the limit.
> The functions that do the writes however are shared in many paths.
> 
> In addition to that, perhaps we could write something saner to the file
> so that it takes longer to hit the limit.
> 
> Rewriting "common_pid" over and over again for example, is probably 
> unnecessary with some reworking.
> 
> John Kacur

I forgot to mention that the cause of the problem is the addition of the 
filting feature.

commit a5923c5c9f812d84472f60f8c9c3e13a6702dd97
Author: Josef Bacik <jbacik@...com>
Date:   Fri Dec 4 12:03:38 2015 -0500

    trace-cmd: Filter out specific pids
    
    Sometimes it is a good idea to filter out events from the recording 
processes
    and even other pids, for example an external app that links against 
libtracecmd
    that watches syscalls probably doesn't want its own syscalls showing 
up in the
    events.  So add the ability to filter _out_ certain pids for the 
events that are
    enabled.  The recorder threads will add their pids to this list, but 
they are
    only filtered if the library consumer filters out a pid or if the 
trace-cmd
    record user tries to filter anything out.
    
    Link: 
http://lkml.kernel.org/r/1449248618-30556-1-git-send-email-jbacik@fb.com
    
    Signed-off-by: Josef Bacik <jbacik@...com>
    Signed-off-by: Steven Rostedt <rostedt@...dmis.org>

Thanks

John

Powered by blists - more mailing lists