#1024811 linux: /proc/[pid]/stat unparsable

Package:
src:linux
Source:
linux
Submitter:
Thorsten Glaser
Date:
2022-11-26 08:03:05 UTC
Severity:
normal
Tags:
#1024811#5
Date:
2022-11-25 17:14:09 UTC
From:
To:
tglase@x61w:~ $ cp /bin/mksh mk\ sh
tglase@x61w:~ $ ./mk\ sh -c 'echo $$; sleep 10' &
[1] 12862
tglase@x61w:~ $ 12862
cat /proc/12862/stat
12862 (mk sh) S 12649 12862 12649 34838 12864 4194304 124 0 0 0 0 0 0 0 20 0 1 0 58970609 2977792 211 18446744073709551615 93898845827072 93898846076365 140724844598496 0 0 0 0 0 134307847 0 0 0 17 0 0 0 0 0 0 93898846102320 93898846110384 93898853036032 140724844603277 140724844603306 140724844603306 140724844605424 0
tglase@x61w:~ $ cat /proc/12862/stat | hd
00000000  31 32 38 36 32 20 28 6d  6b 20 73 68 29 20 53 20  |12862 (mk sh) S |
[…]

This directly violates the documentation of proc(5) which says…

              The  fields,  in order, with their proper scanf(3) format speci‐
              fiers, are listed below.  Whether or not certain of these fields

… while scanf(3) indicates:

       s      Matches  a  sequence  of  non-white-space  characters;  the next

The effect is that /proc/[pid]/stat cannot be parsed the way it is
documented, as it does not escape embedded whitespace characters;
this, as a result, makes this file realistically useless and anything
relying on it broken.
#1024811#10
Date:
2022-11-25 17:24:04 UTC
From:
To:
Dixi quod…

… nor parenthesēs:

tglase@x61w:~ $ ./mk\)sh -c 'echo $$; sleep 10' &
[1] 13375
tglase@x61w:~ $ 13375
tglase@x61w:~ $ cat /proc/13375/stat
13375 (mk)sh) S 13330 13375 13330 34837 13377 4194304 124 0 0 0 0 0 0 0 20 0 1 0 59029474 2977792 180 18446744073709551615 94911056490496 94911056739789 140721459110048 0 0 0 0 0 134307847 0 0 0 17 1 0 0 0 0 0 94911056765744 94911056773808 94911059955712 140721459115917 140721459115946 140721459115946 140721459118064 0

This is… sad — extremely so. It does not escape anything.
I found proc_task_name(), which has an escape parameter,
which is set to false here, but it’s only for /status
which must escape newlines.

It’s used with false in /stat and /comm… the latter indeed
needing no escapes.

I’d argue that it needs a tristate argument, 0 for /comm
to not escape anything, 1 for /status to escape newlines,
and 2 for /stat to escape whitespace (and perhaps also a
closing parenthesis, using \x29, so splitting both using
scanf as indicated in the manpage and using parenthesēs
as people seem to do on the ’net is fixed).

bye,
//mirabilos