From: Kees Cook <keescook@xxxxxxxxxxxx>
commit dcd46d897adb70d63e025f175a00a89797d31a43 upstream.
Quoting[1] Ariadne Conill:
"In several other operating systems, it is a hard requirement that the
second argument to execve(2) be the name of a program, thus prohibiting
a scenario where argc < 1. POSIX 2017 also recommends this behaviour,
but it is not an explicit requirement[2]:
The argument arg0 should point to a filename string that is
associated with the process being started by one of the exec
functions.
...
Interestingly, Michael Kerrisk opened an issue about this in 2008[3],
but there was no consensus to support fixing this issue then.
Hopefully now that CVE-2021-4034 shows practical exploitative use[4]
of this bug in a shellcode, we can reconsider.
This issue is being tracked in the KSPP issue tracker[5]."
While the initial code searches[6][7] turned up what appeared to be
mostly corner case tests, trying to that just reject argv == NULL
(or an immediately terminated pointer list) quickly started tripping[8]
existing userspace programs.
The next best approach is forcing a single empty string into argv and
adjusting argc to match. The number of programs depending on argc == 0
seems a smaller set than those calling execve with a NULL argv.
Account for the additional stack space in bprm_stack_limits(). Inject an
empty string when argc == 0 (and set argc = 1). Warn about the case so
userspace has some notice about the change:
process './argc0' launched './argc0' with NULL argv: empty string added
Additionally WARN() and reject NULL argv usage for kernel threads.
[1] https://lore.kernel.org/lkml/20220127000724.15106-1-ariadne@xxxxxxxxxxxxxxxx/
[2] https://pubs.opengroup.org/onlinepubs/9699919799/functions/exec.html
[3] https://bugzilla.kernel.org/show_bug.cgi?id=8408
[4] https://www.qualys.com/2022/01/25/cve-2021-4034/pwnkit.txt
[5] https://github.com/KSPP/linux/issues/176
[6] https://codesearch.debian.net/search?q=execve%5C+*%5C%28%5B%5E%2C%5D%2B%2C+*NULL&literal=0
[7] https://codesearch.debian.net/search?q=execlp%3F%5Cs*%5C%28%5B%5E%2C%5D%2B%2C%5Cs*NULL&literal=0
[8] https://lore.kernel.org/lkml/20220131144352.GE16385@xsang-OptiPlex-9020/
Reported-by: Ariadne Conill <ariadne@xxxxxxxxxxxxxxxx>
Reported-by: Michael Kerrisk <mtk.manpages@xxxxxxxxx>
Cc: Matthew Wilcox <willy@xxxxxxxxxxxxx>
Cc: Christian Brauner <brauner@xxxxxxxxxx>
Cc: Rich Felker <dalias@xxxxxxxx>
Cc: Eric Biederman <ebiederm@xxxxxxxxxxxx>
Cc: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>
Cc: linux-fsdevel@xxxxxxxxxxxxxxx
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
Acked-by: Christian Brauner <brauner@xxxxxxxxxx>
Acked-by: Ariadne Conill <ariadne@xxxxxxxxxxxxxxxx>
Acked-by: Andy Lutomirski <luto@xxxxxxxxxx>
Link: https://lore.kernel.org/r/20220201000947.2453721-1-keescook@xxxxxxxxxxxx
[vegard: fixed conflicts due to missing
886d7de631da71e30909980fdbf318f7caade262^- and
3950e975431bc914f7e81b8f2a2dbdf2064acb0f^-]
Signed-off-by: Vegard Nossum <vegard.nossum@xxxxxxxxxx>
---
fs/exec.c | 25 ++++++++++++++++++++++++-
1 file changed, 24 insertions(+), 1 deletion(-)
This has been tested in both argc == 0 and argc >= 1 cases, but I would
still appreciate a review given the differences with mainline. If it's
considered too risky I'm also fine with dropping it -- just wanted to
make sure this didn't fall through the cracks, as it does block a real
(albeit old by now) exploit.
diff --git a/fs/exec.c b/fs/exec.c
index 098de820abcc9..a7d78241082a2 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -454,6 +454,9 @@ static int prepare_arg_pages(struct linux_binprm *bprm,
unsigned long limit, ptr_size;
bprm->argc = count(argv, MAX_ARG_STRINGS);
+ if (bprm->argc == 0)
+ pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
+ current->comm, bprm->filename);
if (bprm->argc < 0)
return bprm->argc;
@@ -482,8 +485,14 @@ static int prepare_arg_pages(struct linux_binprm *bprm,
* the stack. They aren't stored until much later when we can't
* signal to the parent that the child has run out of stack space.
* Instead, calculate it here so it's possible to fail gracefully.
+ *
+ * In the case of argc = 0, make sure there is space for adding a
+ * empty string (which will bump argc to 1), to ensure confused
+ * userspace programs don't start processing from argv[1], thinking
+ * argc can never be 0, to keep them from walking envp by accident.
+ * See do_execveat_common().
*/
- ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
+ ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
if (limit <= ptr_size)
return -E2BIG;
limit -= ptr_size;
@@ -1848,6 +1857,20 @@ static int __do_execve_file(int fd, struct filename *filename,
if (retval < 0)
goto out;
+ /*
+ * When argv is empty, add an empty string ("") as argv[0] to
+ * ensure confused userspace programs that start processing
+ * from argv[1] won't end up walking envp. See also
+ * bprm_stack_limits().
+ */
+ if (bprm->argc == 0) {
+ const char *argv[] = { "", NULL };
+ retval = copy_strings_kernel(1, argv, bprm);
+ if (retval < 0)
+ goto out;
+ bprm->argc = 1;
+ }
+
retval = exec_binprm(bprm);
if (retval < 0)
goto out;
--
2.35.1.46.g38062e73e0
