Re: [PATCH] scsi: lpfc: replace deprecated strncpy with strscpy

[Kees Cook <keescook@chromium.org>]

On Fri, Feb 23, 2024 at 12:02:22PM -0800, Justin Stitt wrote:
> Hi,
> 
> On Wed, Feb 21, 2024 at 6:38 PM Kees Cook <kees@kernel.org> wrote:
> >
> >
> >
> > On February 21, 2024 4:41:52 PM PST, Justin Stitt <justinstitt@google.com> wrote:
> > >strncpy() is deprecated for use on NUL-terminated destination strings
> > >[1] and as such we should prefer more robust and less ambiguous string
> > >interfaces.
> > >
> > >We expect ae->value_string to be NUL-terminated because there's a
> > >comment that says as much; these attr strings are also used with other
> > >string APIs, further cementing the fact.
> > >
> > >Now, the question of whether or not to NUL-pad the destination buffer:
> > >lpfc_fdmi_rprt_defer() initializes vports (all zero-initialized), then
> > >we call lpfc_fdmi_cmd() with each vport and a mask. Then, inside of
> > >lpfc_fdmi_cmd() we check each bit in the mask to invoke the proper
> > >callback. Importantly, the zero-initialized vport is passed in as the
> > >"attr" parameter. Seeing this:
> > >|      struct lpfc_fdmi_attr_string *ae = attr;
> > >... we can tell that ae->value_string is entirely zero-initialized. Due
> > >to this, NUL-padding is _not_ required as it would be redundant.
> > >
> > >Conveniently, strscpy also returns the number of bytes copied into the
> > >destination buffer, eliminating the need for strnlen!
> > >
> > >Considering the above, a suitable replacement is `strscpy` [2].
> > >
> > >Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
> > >Link: https://manpages.debian.org/testing/linux-manual-4.8/strscpy.9.en.html [2]
> > >Link: https://github.com/KSPP/linux/issues/90
> > >Cc: linux-hardening@vger.kernel.org
> > >Signed-off-by: Justin Stitt <justinstitt@google.com>
> > >---
> > > drivers/scsi/lpfc/lpfc_ct.c | 5 ++---
> > > 1 file changed, 2 insertions(+), 3 deletions(-)
> > >
> > >diff --git a/drivers/scsi/lpfc/lpfc_ct.c b/drivers/scsi/lpfc/lpfc_ct.c
> > >index baae1f8279e0..42594ec87290 100644
> > >--- a/drivers/scsi/lpfc/lpfc_ct.c
> > >+++ b/drivers/scsi/lpfc/lpfc_ct.c
> > >@@ -2569,9 +2569,8 @@ lpfc_fdmi_set_attr_string(void *attr, uint16_t attrtype, char *attrstring)
> > >        * 64 bytes or less.
> > >        */
> > >
> > >-      strncpy(ae->value_string, attrstring, sizeof(ae->value_string));
> > >-      len = strnlen(ae->value_string, sizeof(ae->value_string));
> > >-      /* round string length to a 32bit boundary. Ensure there's a NULL */
> > >+      len = strscpy(ae->value_string, attrstring, sizeof(ae->value_string));
> >
> > This could be < 0 on error, and at least lpfc_fdmi_hba_attr_os_ver() may present more than 64 bytes...
> 
> Am I putting too much faith in this comment?
> 
> static inline int lpfc_fdmi_set_attr_string(void *attr, uint16_t
> attrtype, char *attrstring)
> ...
> /*
> * We are trusting the caller that if a fdmi string field
> * is capped at 64 bytes, the caller passes in a string of
> * 64 bytes or less.
> */

This comment is clearly wrong, given lpfc_fdmi_hba_attr_os_ver(). :)
But I feel like I'm misunderstanding it since it was added by the same
commit that added the 256-byte callers, commit 045c58c87560 ("scsi:
lpfc: Rework FDMI attribute registration for unintential padding")

> 
> I see lpfc_fdmi_hba_attr_os_ver() calls lpfc_fdmi_set_attr_string()
> with an attrstring sized at 256 bytes:
> char buf[256] = { 0 };
> 
> Can we really return -E2BIG from strscpy() if the dest buffer is the
> same size as the source buffer?

I see my confusion: I didn't check the size of ae->value_string, which I
assumed was 64 bytes. But it's 256, so in theory we can't overflow.

> I'm happy to just make the standard strncpy -> strscpy replacement and
> drop the len assignment. Let me know what you think, Kees.

For robustness, let's leave the strlen() in place...

-- 
Kees Cook