119

Is there a limit to the amount of data that can be stored in an environment variable on Linux, and if so: what is it?

For Windows, I've found following KB article which summarizes to: Windows XP or later: 8191 characters Windows 2000/NT 4.0: 2047 characters

4
  • 3
    The actual limit on windows is 31,767. You read about the set command, which has it's commandline limit to 8191 characters. See this msdn article Still a random limitation, though. Commented Jan 29, 2012 at 23:50
  • 7
    8191 isn't that random. It's 2^13 - 1. Commented Mar 9, 2017 at 11:21
  • 1
    AWS Lambda: the total size of the set [of environment variables] does not exceed 4 KB (source) Commented May 27, 2019 at 9:17
  • 2
    @BarnabyDawson perhaps Christopher meant it in the sense that it was arbitrary? Commented Jun 4, 2019 at 13:46

8 Answers 8

100

I don't think there is a per-environment variable limit on Linux. The total size of all the environment variables put together is limited at execve() time. See "Limits on size of arguments and environment" here for more information.

A process may use setenv() or putenv() to grow the environment beyond the initial space allocated by exec.

Here's a quick and dirty program that creates a 256 MB environment variable.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

int main(void)
{
  size_t size = 1 << 28; /* 256 MB */
  char *var;

  var = malloc(size);
  if (var == NULL) {
  perror("malloc");
  return 1;
}

  memset(var, 'X', size);
  var[size - 1] = '\0';
  var[0] = 'A';
  var[1] = '=';

  if (putenv(var) != 0) {
  perror("putenv");
  return 1;
}

  /*  Demonstrate E2BIG failure explained by paxdiablo */
  execl("/bin/true", "true", (char *)NULL);
  perror("execl");


  printf("A=%s\n", getenv("A"));

  return 0;
}
4
  • 9
    +1 for the execve() info. You can actually add environment variables up to (at least) 8M but exec() calls will not work. This manifests itself in bash as "argument list is too long" whenever you then try to run a command.
    – paxdiablo
    Commented Jul 6, 2009 at 2:07
  • 1
    Also note that POSIX requires ARG_MAX to be at least _POSIX_ARG_MAX, which should be 4096.
    – ninjalj
    Commented Aug 5, 2010 at 21:09
  • 17
    The top answer (by Chipaca) on askUbuntu discusses using xargs --show-limits to get more info. Commented Jul 21, 2013 at 6:46
  • 1
    I see limit per arg string too - "Additionally, the limit per string is 32 pages (the kernel constant MAX_ARG_STRLEN)"
    – kiran01bm
    Commented Dec 16, 2019 at 5:36
39

Well, it's at least 4M on my box. At that point, I got bored and wandered off. Hopefully the terminal output will be finished before I'm back at work on Monday :-)

export b1=A
export b2=$b1$b1
export b4=$b2$b2
export b8=$b4$b4
export b16=$b8$b8
export b32=$b16$b16
export b64=$b32$b32
export b128=$b64$b64
export b256=$b128$b128
export b512=$b256$b256
export b1k=$b512$b512
export b2k=$b1k$b1k
export b4k=$b2k$b2k
export b8k=$b4k$b4k
export b16k=$b8k$b8k
export b32k=$b16k$b16k
export b64k=$b32k$b32k
export b128k=$b64k$b64k
export b256k=$b128k$b128k
export b512k=$b256k$b256k
export b1m=$b512k$b512k
export b2m=$b1m$b1m
export b4m=$b2m$b2m
echo $b4m
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
:    :    :    :    :    :    :    :    :    :    :    :
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

If you're worried that 4M may not be enough for your environment variable, you may want to rethink how you're doing things.

Perhaps it would be a better idea to put the information into a file and then use an environment variable to reference that file. I've seen cases where, if the variable is of the form @/path/to/any/fspec, it gets the actual information from the file path/to/any/fspec. If it doesn't begin with @, it uses the value of the environment variable itself.


Interestingly enough, with all those variables set, every single command starts complaining that the argument list is too long so, even though it lets you set them, it may not be able to start programs after you've done it (since it has to pass the environment to those programs).

4
  • 1
    I hope you used a script, and didn't type this all out by hand! Commented Jul 3, 2009 at 8:26
  • 2
    No, hence the "getting bored" bit. Vi-mode command line editing made it a little easier but I thought it might fail before I hit the 4M mark. C'est la vie.
    – paxdiablo
    Commented Jul 3, 2009 at 9:13
  • 1
    but why do ksh variables stop at 256K? Possibly CentOS too
    – phuclv
    Commented Nov 24, 2016 at 10:35
  • @LưuVĩnhPhúc: because programs have limits. They may be arbitrary limits or they may be based on available resources, but the limits exist. Quite possibly DavidK assumed his shell wouldn't be (mis-)used by lunatics who should know better :-)
    – paxdiablo
    Commented Nov 24, 2016 at 12:53
24

Here are two helpful commands:

  • getconf -a | grep ARG_MAX

  • true | xargs --show-limits

2
  • 1
    This should be the accepted answer. Both commands also work on macOS Ventura.
    – Teddy C
    Commented Mar 18, 2023 at 3:10
  • 1
    The first one works fine on my macOS Ventura 13.6.6, the other gives me the following error: xargs: illegal option -- -
    – Virus_7
    Commented Apr 18 at 21:35
10

I did a quick test on my Linux box with the following snippet:

a="1"
while true
do
    a=$a$a
    echo "$(date) $(numfmt --to=iec-i --suffix=B --padding=7 ${#a})" 
done

On my box (Gentoo 3.17.8-gentoo-r1) this results in (last lines of output):

Wed Jan  3 12:16:10 CET 2018   16MiB
Wed Jan  3 12:16:11 CET 2018   32MiB
Wed Jan  3 12:16:12 CET 2018   64MiB
Wed Jan  3 12:16:15 CET 2018  128MiB
Wed Jan  3 12:16:21 CET 2018  256MiB
Wed Jan  3 12:16:33 CET 2018  512MiB
xrealloc: cannot allocate 18446744071562068096 bytes

So: the limit is quite high!

3
  • 2
    18446744071562068096 bytes is about 15 exabytes. :) That seems like a big jump from 512MiB.
    – Marcus
    Commented May 16, 2019 at 21:47
  • 2
    @Marcus: it is... I guess that somehow 1Gb i.e. 1073741824 bytes i.e. 2^30 bytes overflows somewhere. The number of bytes in the error 18446744071562068096 is pretty close to 2^64 i.e. 18446744073709551616
    – cyberbird
    Commented Feb 7, 2020 at 16:29
  • 2
    I do have limited C/gdb skills but I think the issue is in the bash source file subst.c The function sub_append_string (line 726) uses a signed int variable called n to calculate the new size. The new size (2^31) would fit but there is an alignment calculation: n = (n + DEFAULT_ARRAY_SIZE) - (n % DEFAULT_ARRAY_SIZE); and I suspect the part: (n + DEFAULT_ARRAY_SIZE) overflows. All very nice but we are ofcourse way beyond any sane limit here..
    – cyberbird
    Commented Feb 9, 2020 at 12:47
1

Don't know exactly but a quick experiment shows that no error occurs e.g. with 64kB of value:

% perl -e 'print "#include <stdlib.h>\nint main() { return setenv(\"FOO\", \"", "x"x65536, "\", 1); }\n";'\
| gcc -x c -o envtest - && ./envtest && echo $?
0
1

I used this very quick and dirty php code (below), modifying it for different values, and found that it works for variable lengths up to 128k. After that, for whatever reason, it doesn't work; no exception is raised, no error is reported, but the value does not show up in the subshell.

Maybe this is a php-specific limit? Maybe there are php.ini settings that might affect it? Or maybe there's a limit on the size of vars that a subshell will inherit? Maybe there are relevant kernel or shell config settings..

Anyway, by default, in CentOS, the limit for setting a var in the environment via putenv in php seems to be about 128k.

<?php

  $s = 'abcdefghijklmnop';
  $s2 = "";
  for ($i = 0; $i < 8100; $i++) $s2 .= $s;
  $result = putenv('FOO='.$s2);
  print shell_exec('echo \'FOO: \'${FOO}');
  print "length of s2: ".strlen($s2)."\n";
  print "result = $result\n";
?>

Version info -

[root@localhost scratch]# php --version
PHP 5.2.6 (cli) (built: Dec  2 2008 16:32:08) 
<..snip..>

[root@localhost scratch]# uname -a
Linux localhost.localdomain 2.6.18-128.2.1.el5 #1 SMP Tue Jul 14 06:36:37 EDT 2009 x86_64 x86_64 x86_64 GNU/Linux

[root@localhost scratch]# cat /etc/redhat-release 
CentOS release 5.3 (Final)
0

The command line (with all argument) plus the environment variable should be less then 128k.

0
0

In my case it was due to buffer was limited when accepting a variable input value with read command. Solution was to add -e

Before read accessToken

After read -e accessToken

Docs: http://linuxcommand.org/lc3_man_pages/readh.html

Not the answer you're looking for? Browse other questions tagged or ask your own question.