3 ms·

Linux-insides: Limits on resources in Linux

73 points by 0xAX 9y ago

jwilk 9y ago
> $ strace ulimit -s 2>&1 | grep rl Does it actually work? On my system ulimit is a shell builtin, not a separate command: $ strace ulimit -s strace: Can't stat 'ulimit': No such file or directory
- Mic92 9y ago
  strace -e getrlimit,setrlimit bash -c 'ulimit -s 2 && echo foo'
- devnonymous 9y ago
  Bash has a few builtin that are clones of commands (mostly from coreutils). To execute the command instead of the builtin, just give the full path to the command (not sure and cannot verify, right now but this might mean either /bin/ulimit or /use/bin/ulimit ). There are other notable ones like echo, and time where the builtins have slightly different behaviour than the commands.
  - LukeShu 9y ago
    While what you said is true in general; ulimit is usually not a separate program. It modifies the state of the current process (similarly to `cd`), so it has to be built-in. Now, there's no solid reason why it couldn't work like `unshare` or `chroot` (which also wrap operations that modify the current process), and spawn a second child shell with the modified properties. But, that isn't how it works on any system I have ever used, isn't what users expect, and isn't in conformance with POSIX. To quote POSIX: Since `ulimit` affects the current shell execution environment, it is always provided as a shell regular built-in.
    - akira2501 9y ago
      > so it has to be built-in. We have fork() and exec(), and resource limits carry across exec(), so it does not /have/ to be a built-in.
rwmj 9y ago
Seems odd that it doesn't mention cgroups or the CLONE_* flags used to create namespaces for containers.
- LukeShu 9y ago
  Yes, cgroups can be used to impose limits on resources; and from the outside looking in, can look very similar to setrlimit(). But from the inside of the kernel looking out (generally, the perspective taken by the linux-insides book), they are very different subsystems; I expect that cgroups will come in quite a bit later. If the book were discussing how to limit resources, sure it should mention cgroups; but it is discussing how things work in the kernel, not how to get the kernel to do certain things.

73 points by 0xAX 9y ago

> $ strace ulimit -s 2>&1 | grep rl Does it actually work? On my system ulimit is a shell builtin, not a separate command: $ strace ulimit -s strace: Can't stat 'ulimit': No such file or directory

Mic92 9y ago
strace -e getrlimit,setrlimit bash -c 'ulimit -s 2 && echo foo'
devnonymous 9y ago
Bash has a few builtin that are clones of commands (mostly from coreutils). To execute the command instead of the builtin, just give the full path to the command (not sure and cannot verify, right now but this might mean either /bin/ulimit or /use/bin/ulimit ). There are other notable ones like echo, and time where the builtins have slightly different behaviour than the commands.
- LukeShu 9y ago
  While what you said is true in general; ulimit is usually not a separate program. It modifies the state of the current process (similarly to `cd`), so it has to be built-in. Now, there's no solid reason why it couldn't work like `unshare` or `chroot` (which also wrap operations that modify the current process), and spawn a second child shell with the modified properties. But, that isn't how it works on any system I have ever used, isn't what users expect, and isn't in conformance with POSIX. To quote POSIX: Since `ulimit` affects the current shell execution environment, it is always provided as a shell regular built-in.
  - akira2501 9y ago
    > so it has to be built-in. We have fork() and exec(), and resource limits carry across exec(), so it does not /have/ to be a built-in.

rwmj 9y ago

Seems odd that it doesn't mention cgroups or the CLONE_* flags used to create namespaces for containers.

LukeShu 9y ago
Yes, cgroups can be used to impose limits on resources; and from the outside looking in, can look very similar to setrlimit(). But from the inside of the kernel looking out (generally, the perspective taken by the linux-insides book), they are very different subsystems; I expect that cgroups will come in quite a bit later. If the book were discussing how to limit resources, sure it should mention cgroups; but it is discussing how things work in the kernel, not how to get the kernel to do certain things.