Re: [PATCH] execve.2: execve also returns E2BIG if a string is too long

[Matthew House <mattlloydhouse@gmail.com>]

On Wed, Oct 11, 2023 at 9:21 AM Rik van Riel <riel@surriel.com> wrote:
> On Wed, 2023-10-11 at 12:41 +0200, Alejandro Colomar wrote:
> > Hi Rik,
> >
> > On Tue, Oct 10, 2023 at 11:41:53PM -0400, Rik van Riel wrote:
> > > Document that if a command line or environment string is too long
> > > (> MAX_ARG_STRLEN), execve will also return E2BIG.
> >
> > That's already implied by the current text:
> >
> >        E2BIG  The total number of bytes in the environment (envp) and
> > argument
> >               list (argv) is too large.
> >
> > That means that
> >
> > size_t  bytes;
> >
> > bytes = 0;
> > for (char *e = envp; e != NULL; e++)
> >         bytes += strlen(e) + 1;  // I have doubts about the +1
> > for (char *a = argv; a != NULL; a++)
> >         bytes += strlen(a) + 1;  // Same doubts
> >
> > if (bytes > MAX_ARG_STRLEN)  // Maybe >= ?
> >         return -E2BIG;
>
> The code in fs/exec.c enforces MAX_ARG_STRLEN against
> each individual string, not against the total.
>
> If any string, either argument or environment, is larger
> than 32 * PAGE_SIZE, the kernel will return -E2BIG.
>
> do_execveat_common() has this code, which uses copy_strings
> to copy both the strings from the environment, and from
> the command line arguments:
>
>         retval = copy_strings(bprm->envc, envp, bprm);
>         if (retval < 0)
>                 goto out_free;
>
>         retval = copy_strings(bprm->argc, argv, bprm);
>         if (retval < 0)
>                 goto out_free;
>
> Inside copy_strings() we have this code:
>
>
>         while (argc-- > 0) {
> ...
>                 len = strnlen_user(str, MAX_ARG_STRLEN);
>                 if (!len)
>                         goto out;
>
>                 ret = -E2BIG;
>                 if (!valid_arg_len(bprm, len))
>                         goto out;
>
> The valid_arg_len() function does not need explanation:
>
> static bool valid_arg_len(struct linux_binprm *bprm, long len)
> {
>         return len <= MAX_ARG_STRLEN;
> }
>
>
> The current man page wording is very clear about the total
> length being enforced, but IMHO not as clear about the limit
> that gets enforced on each individual string.
>
> The total length limit of environment & commandline arguments
> is enforced by bprm_stack_limits(), and is checked against
> either 1/4 of the maximum stack size, or 3/4 of _STK_LIM, whichever
> is smaller. The MAX_ARG_STRLEN value does not come into play when
> enforcing the total.

To expand on this, there are basically two separate byte limits in
fs/exec.c, one for each individual argv/envp string, and another for all
strings and all pointers to them as a whole. To put the whole thing in
pseudocode, the checks work effectively like this, assuming I haven't made
any errors:

int argc, envc;
unsigned long bytes, limit;

/* assume that argv has already been adjusted to add an empty argv[0] */
argc = 0, envc = 0, bytes = 0;
for (char **a = argv; *a != NULL; a++, argc++) {
    if (strlen(*a) >= MAX_ARG_STRLEN)
        return -E2BIG;
    bytes += strlen(*a) + 1;
}
for (char **e = envp; *e != NULL; e++, envc++) {
    if (strlen(*e) >= MAX_ARG_STRLEN)
        return -E2BIG;
    bytes += strlen(*e) + 1;
}

if (argc > MAX_ARG_STRINGS || envc > MAX_ARG_STRINGS)
    return -E2BIG;
bytes += (argc + envc) * sizeof(void *);

limit = max(min(_STK_LIM / 4 * 3, rlim_stack.rlim_cur / 4), ARG_MAX);
if (bytes > limit)
    return -E2BIG;

Thank you,
Matthew House