ps: full command is too long

Question

Good day!

I use 'ps' to see command that starts process. The issue is that command is too long and 'ps' does not show it entirely.

Example: I use command 'ps -p 2755 | less' and have following output

  PID TTY      STAT   TIME COMMAND
2755 ?        Sl   305:05 /usr/java/jdk1.6.0_37/bin/java -Xms64m -Xmx512m -Dflume.monitoring.type=GANGLIA -Dflume.monitoring.hosts=prod.hostname.ru:8649 -cp /etc/flume-ng/conf/acrs-event:/usr/lib/flume-ng/lib/*:/etc/hadoop/conf:/usr/lib/hadoop/lib/activation-1.1.jar:/usr/lib/hadoop/lib/asm-3.2.jar:/usr/lib/hadoop/lib/avro-1.7.4.jar:/usr/lib/hadoop/lib/commons-beanutils-1.7.0.jar:/usr/lib/hadoop/lib/commons-beanutils-core-1.8.0.jar:/usr/lib/hadoop/lib/commons-cli-1.2.jar:/usr/lib/hadoop/lib/commons-codec-1.4.jar:/usr/lib/hadoop/lib/commons-collections-3.2.1.jar:/usr/lib/hadoop/lib/commons-compress-1.4.1.jar:/usr/lib/hadoop/lib/commons-configuration-1.6.jar:/usr/lib/hadoop/lib/commons-digester-1.8.jar:/usr/lib/hadoop/lib/commons-el-1.0.jar:/usr/lib/hadoop/lib/commons-httpclient-3.1.jar:/usr/lib/hadoop/lib/commons-io-2.1.jar:/usr/lib/hadoop/lib/commons-lang-2.5.jar:/usr/lib/hadoop/lib/commons-logging-1.1.1.jar:/usr/lib/hadoop/lib/commons-math-2.1.jar:/usr/lib/hadoop/lib/commons-net-3.1.jar:/usr/lib/hadoop/lib/guava-11.0.2.jar:/usr/lib/hadoop/lib/jackson-core-asl-1.8.8.jar:/usr/lib/hadoop/lib/jackson-jaxrs-1.8.8.jar:/usr/lib/hadoop/lib/jackson-mapper-asl-1.8.8.jar:/usr/lib/hadoop/lib/jackson-xc-1.8.8.jar:/usr/lib/hadoop/lib/jasper-compiler-5.5.23.jar:/usr/lib/hadoop/lib/jasper-runtime-5.5.23.jar:/usr/lib/hadoop/lib/jaxb-api-2.2.2.jar:/usr/lib/hadoop/lib/jaxb-impl-2.2.3-1.jar:/usr/lib/hadoop/lib/jersey-core-1.8.jar:/usr/lib/hadoop/lib/jersey-json-1.8.jar:/usr/lib/hadoop/lib/jersey-server-1.8.jar:/usr/lib/hadoop/lib/jets3t-0.6.1.jar:/usr/lib/hadoop/lib/jettison-1.1.jar:/usr/lib/hadoop/lib/jetty-6.1.26.cloudera.2.jar:/usr/lib/hadoop/lib/jetty-util-6.1.26.cloudera.2.jar:/usr/lib/hadoop/lib/jline-0.9.94.jar:/usr/lib/hadoop/lib/jsch-0.1.42.jar:/usr/lib/hadoop/lib/jsp-api-2.1.jar:/usr/lib/hadoop/lib/jsr305-1.3.9.jar:/usr/lib/hadoop/lib/junit-4.8.2.jar:/usr/lib/hadoop/lib/kfs-0.3.jar:/usr/lib/hadoop/lib/log4j-1.2.17.jar:/usr/lib/hadoop/lib/mockito-all-1.8.5.jar:/usr/lib/hadoop/lib/native:/usr/lib/hadoop/lib/paranamer-2.3.jar:/usr/lib/hadoop/lib/protobuf-java-2.4.0a.jar:/usr/lib/hadoop/lib/servlet-api-2.5.jar:/usr/lib/hadoop/lib/snappy-java-1.0.4.1.jar:/usr/lib/hadoop/lib/stax-api-1.0.1.jar:/usr/lib/hadoop/lib/xmlenc-0.52.jar:/usr/lib/hadoop/lib/xz-1.0.jar:/usr/lib/hadoop/lib/zookeeper-3.4.5-cdh4.3.0.jar:/usr/lib/hadoop/.//bin:/usr/lib/hadoop/.//cloudera:/usr/lib/hadoop/.//etc:/usr/lib/hadoop/.//hadoop-annotations-2.0.0-cdh4.3.0.jar:/usr/lib/hadoop/.//hadoop-annotations.jar:/usr/lib/hadoop/.//hadoop-auth-2.0.0-cdh4.3.0.jar:/usr/lib/hadoop/.//hadoop-auth.jar:/usr/lib/hadoop/.//hadoop-common-2.0.0-cdh4.3.0.jar:/usr/lib/hadoop/.//hadoop-common-2.0.0-cdh4.3.0-tests.jar:/usr/lib/hadoop/.//hadoop-common.jar:/usr/lib/hadoop/.//lib:/usr/lib/hadoop/.//libexec:/usr/lib/hadoop/.//sbin:/usr/lib/hadoop-hdfs/./:/usr/lib/hadoop-hdfs/lib/asm-3.2.jar:/usr/lib/hadoop-hdfs/lib/commons-cli-1.2.jar:/usr/lib/hadoop-hdfs/lib/commons-codec-1.4.jar:/usr/lib/hadoop-hdfs/lib/commons-daemon-1.0.3.jar:/usr/lib/hadoop-hdfs/lib/commons-el-1.0.jar:/usr/lib/hadoop-hdfs/lib/commons-io-2.1.jar:/usr/lib/hadoop-hdfs/lib/commons-lang-2.5.jar:/usr/lib/hadoop-hdfs/lib/commons-logging-1.1.1.jar:/usr/lib/hadoop-hdfs/lib/guava-11.0.2.jar:/usr/lib/hadoop-hdfs/lib/jackson-core-asl-1.8.8.jar:/usr/lib/hadoop-hdfs/lib/jackson-mapper-asl-1.8.8.jar:/usr/lib/hadoop-hdfs/lib/jasper-runtime-5.5.23.jar:/usr/lib/hadoop-hdfs/lib/jersey-core-1.8.jar:/usr/lib/hadoop-hdfs/lib/jersey-server-1.8.jar:/usr/lib/hadoop-hdfs/lib/jetty-6.1.26.cloudera.2.jar:/usr/lib/hadoop-hdfs/lib/jetty-util-6.1.26.cloudera.2.jar:/usr/lib/hadoop-hdfs/lib/jline-0.9.94.jar:/usr/lib/hadoop-hdfs/lib/jsp-api-2.1.jar:/usr/lib/hadoop-hdfs/lib/jsr305-1.3.9.jar:/usr/lib/hadoop-hdfs/lib/log4j-1.2.17.jar:/usr/lib/hadoop-hdfs/lib/protobuf-java-2.4.0a.jar:/usr/lib/hadoop-hdfs/lib/servlet-api-2.5.jar:/usr/lib/hadoop-hdfs/lib/xmlenc-0.52.jar:/usr/lib/hadoop-hdfs/lib/zookeeper-3.4.5-cdh4.3.0.jar:/usr/lib/hadoop-hdfs/.//bin:/usr/lib/hadoop-hdfs/.//cloudera:/usr/lib/hadoop-hdfs/.//hadoop-hdfs-2.0.0-cdh4.3.0.jar:/usr/lib/hadoop-hdfs/.//hadoop-hdfs-2.0.

So, the command line is too long and the command stops mid-phrase. How can I see it whole?

Answer 1

On Linux, with the ps from procps(-ng):

ps -fwwp 2755

In Linux versions prior to 4.2, it's still limited though (by the kernel (/proc/2755/cmdline) to 4k) and you can't get more except by asking the process to tell it to you or use a debugger.

$ sh -c 'sleep 1000' $(seq 4000) &
[1] 31149
$ gdb -p $! /bin/sh
[...]
Attaching to program: /bin/dash, process 31149
[...]
(gdb) bt
#0  0x00007f40d11f40aa in wait4 () at ../sysdeps/unix/syscall-template.S:81
[...]
#7  0x00007f40d115c995 in __libc_start_main (main=0x4022c0, argc=4003, ubp_av=0x7fff5b9f5a88, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fff5b9f5a78)
at libc-start.c:260
#8  0x00000000004024a5 in ?? ()
#9  0x00007fff5b9f5a78 in ?? ()
#10 0x0000000000000000 in ?? ()
(gdb) frame 7
#7  0x00007f40d115c995 in __libc_start_main (main=0x4022c0, argc=4003, ubp_av=0x7fff5b9f5a88, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fff5b9f5a78)
at libc-start.c:260
(gdb) x/4003s *ubp_av
0x7fff5b9ff83e: "sh"
0x7fff5b9ff841: "-c"
0x7fff5b9ff844: "sleep 1000"
0x7fff5b9ff84f: "1"
0x7fff5b9ff851: "2"
[...]
0x7fff5ba04212: "3999"
0x7fff5ba04217: "4000"

To print the 4th arg with up to 5000 characters:

(gdb) set print elements 5000
(gdb) p ubp_av[3]

If you want something non-intrusive, you could try and get the information from /proc/2755/mem (note that if the kernel.yama.ptrace_scope is not set to 0, you'll need superuser permissions for that). This below works for me (prints all the arguments and environment variables), but there's not much guarantee I would think (the error and unexpected input handling is left as an exercise to the reader):

$ perl -e '$p=shift;open MAPS, "/proc/$p/maps";
          ($m)=grep /\[stack\]/, <MAPS>;
          ($a,$b)=map hex, $m =~ /[\da-f]+/g;
          open MEM, "/proc/$p/mem" or die "open mem: $!";
          seek MEM,$a,0; read MEM, $c,$b-$a;
          print((split /\0{2,}/,$c)[-1])' "$!" | tr \\0 \\n | head
sh
-c
sleep 1000
1
2
3
4
5
6
7

(replace "$!" with the process id). The above uses the fact that Linux puts the strings pointed to by argv[], envp[] and the executed filename at the bottom of the stack of the process.

The above looks in that stack for the bottom-most string in between two sets of two or more consecutive NUL bytes. It doesn't work if any of the arguments or env strings is empty, because then you'll have a sequence of 2 NUL bytes in the middle of those argv or envp. Also, we don't know where the argv strings stop and where the envp ones start.

A work around for that would be to refine that heuristic by looking backwards for the actual content of argv[] (the pointers). This below works on i386 and amd64 architecture for ELF executables at least:

perl -le '$p=shift;open MAPS, "/proc/$p/maps";
      ($m)=grep /\[stack\]/, <MAPS>;
      ($a,$b)=map hex, $m =~ /[\da-f]+/g;
      open MEM, "/proc/$p/mem" or die "open mem: $!";
      seek MEM,$a,0; read MEM, $c,$b-$a;
      $c =~ /.*\0\0\K[^\0].*\0[^\0]*$/s;
      @a=unpack"L!*",substr$c,0,$-[0];
      for ($i = $#a; $i >=0 && $a[$i] != $a+$-[0];$i--) {}
      for ($i--; $i >= 0 && ($a[$i]>$a || $a[$i]==0); $i--) {}
      $argc=$a[$i++];
      print for unpack"(Z*)$argc",substr$c,$a[$i]-$a;' "$!"

Basically, it does the same as above, but once it has found the first string of argv[] (or at least one of the argv[] or envp[] strings if there are empties), it knows its address, so it looks backward in the top rest of the stack for a pointer with that same value. Then keeps looking backwards until it finds a number that can't be a pointer to those, and that is argc. Then the next integer is argv[0]. And knowing argv[0] and argc, it can display the list of arguments.

That doesn't work if the process has written to its argv[] possibly overriding some NUL delimiters or if argc is 0 (argc is generally at least 1 to include argv[0]) but should work in the general case at least for ELF executables.

In 4.2 and newer, /proc/<pid>/cmdline is no longer truncated, but ps itself has a maximum display width of 128K.

Answer 2

Add one or two -w flags. It makes the output wider. e.g. ps auxww.

Answer 3

In Linux kernel 4.2 and newer, /proc/<pid>/cmdline is no longer truncated and the following works nicely:

xargs -0 printf '%s\n' < /proc/2755/cmdline