>
> i've seen such things: mmap fails with EPERM
> instead of ENOMEM when the system runs out of
> virtual address space, i don't remember the
> details though, you can debug it with strace.


Interesting.

if you figure it out i'd like to know when
> this happens (probably a kernel bug).


I'll see what I can find. If it's of interest/help I'm running on the
3.19.0-56-generic Linux Kernel.

putenv-doublefree was recently fixed, so
> your musl is probably not up to date.


That's very likely.

Thank you!

On Wed, Mar 30, 2016 at 2:12 PM, Rich Felker <dalias@libc.org> wrote:

> On Wed, Mar 30, 2016 at 08:04:43PM +0200, Szabolcs Nagy wrote:
> > * Max Ruttenberg <mruttenberg@emutechnology.com> [2016-03-30 13:22:19
> -0400]:
> > > this is what my src/regression/REPORT file looks like:
> > >
> > >
> ******************************************************************************************************************************
> > > src/regression/malloc-brk-fail.c:33: malloc did not fail with ENOMEM,
> got
> > > Operation not permitted
> > > FAIL ./src/regression/malloc-brk-fail-static.exe [status 1]
> > > src/regression/malloc-brk-fail.c:33: malloc did not fail with ENOMEM,
> got
> > > Operation not permitted
> > > FAIL ./src/regression/malloc-brk-fail.exe [status 1]
> > > src/regression/malloc-oom.c:16: expected ENOMEM, got Operation not
> permitted
> > > FAIL ./src/regression/malloc-oom-static.exe [status 1]
> > > src/regression/malloc-oom.c:16: expected ENOMEM, got Operation not
> permitted
> > > FAIL ./src/regression/malloc-oom.exe [status 1]
> > > FAIL ./src/regression/putenv-doublefree-static.exe [signal Segmentation
> > > fault]
> > > FAIL ./src/regression/putenv-doublefree.exe [signal Segmentation fault]
> > > src/regression/setenv-oom.c:23: expected ENOMEM, got Operation not
> permitted
> > > FAIL ./src/regression/setenv-oom-static.exe [status 1]
> > > src/regression/setenv-oom.c:23: expected ENOMEM, got Operation not
> permitted
> > > FAIL ./src/regression/setenv-oom.exe [status 1]
> > >
> ******************************************************************************************************************************
> > >
> > > Is the brk system call still kosher? I thought malloc was supposed to
> use
> > > something mmap.
> > >
> >
> > i've seen such things: mmap fails with EPERM
> > instead of ENOMEM when the system runs out of
> > virtual address space, i don't remember the
> > details though, you can debug it with strace.
> >
> > if you figure it out i'd like to know when
> > this happens (probably a kernel bug).
>
> We should probably work around this kernel bug by not trusting mmap to
> set errno and instead just setting it to ENOMEM on failure (in the
> malloc code at least; working around it in the mmap function may be
> more work becaus EPERM is a valid error for some usage cases).
>
> Rich
>



- Max