On Mon, Nov 12, 2018 at 5:52 PM, Andrew Morton <a...@linux-foundation.org> wrote: > On Mon, 12 Nov 2018 17:09:56 +0100 Oleg Nesterov <o...@redhat.com> wrote: > >> Large enterprise clients often times run applications out of networked >> file systems where the IT mandated layout of project volumes can end up >> leading to paths that are longer than 128 characters. Bumping this up to >> the next order of two solves this problem in all but the most egregious >> case while still fitting into a 512b slab. >> >> ... >> >> --- a/include/uapi/linux/binfmts.h >> +++ b/include/uapi/linux/binfmts.h >> @@ -16,6 +16,6 @@ struct pt_regs; >> #define MAX_ARG_STRINGS 0x7FFFFFFF >> >> /* sizeof(linux_binprm->buf) */ >> -#define BINPRM_BUF_SIZE 128 >> +#define BINPRM_BUF_SIZE 256
This comment needs updating too: fs/exec.c: * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes >> #endif /* _UAPI_LINUX_BINFMTS_H */ > > It does seem a rather silly restriction, and it's tempting to suggest > reworking the code so that linux_binprm.buf is dynamically sized to > accommodate even ludicrously large strings. > > But obviously 128 bytes has been enough for all this time, so that's > going too far. However it would be basically cost-free to increase > BINPRM_BUF_SIZE up to the point where sizeof(struct linux_binprm) == > PAGE_SIZE? Yeah, and this might be a useful detail included in a comment above the #define... Regardless: Acked-by: Kees Cook <keesc...@chromium.org> -- Kees Cook
