144448 - CKFW incorrectly uses ## operator to create cryptoki entry points

Status: RESOLVED FIXED

(NSS :: Libraries, defect, P2)

Description

[Ian McGreer]

The file nssck.api defines the macro __ADJOIN:

#define __ADJOIN(x,y) x##y

then uses it to define cryptoki entry points:

CK_RV CK_ENTRY
__ADJOIN(MODULE_NAME,C_Finalize)
(
  CK_VOID_PTR pReserved
)

As noted in K&R, this use of the ## operator creates an ambiguity.  The result
will be MODULE_NAMEC_Finalize, which is not the intention.

The fix is to define a second level macro that allows expansion of the
parameters before the ## operator is applied.  Here is the patch:

-#define __ADJOIN(x,y) x##y
+#define nssadjoin(x,y) x##y
+
+#define __ADJOIN(x,y) nssadjoin(x,y)

This fix is purely cosmetic.  It does not change functionality at all, and the
bug would only appear to someone who ran the code through a preprocessor (as I did).

Comment 1

[Wan-Teh Chang]

I had to deal with a similar problem with the stringize
operator # in nss.rc.

Just a stylistic nit: use all caps in macro names.

I also want to confirm that this change won't break
binary compatibility, right?  Do we ever call or
look up the function MODULE_NAMEC_Finalize by name
directly?

Comment 2

[Ian McGreer]

Will change it to NSSADJOIN.

This won't break compatibility -- the functions are never called directly, but
returned in a table of function pointers.

If a CKFW implementor has code which made calls to the functions, this would
break his or her builds.  But that is an unlikely situation, and not the
intended use of CKFW.  In addition, that implementor would have been making
calls to MODULE_NAMEC_XXX, so should have filed this bug already ;-)

Comment 3

[Wan-Teh Chang]

If the functions are never called directly, we should declare them
as static.  Ian, do you know how to make that change?

Comment 4

[Ian McGreer]

Attachment: implement cryptoki entry points as static functions (except for C_GetFunctionList)

Note, there is a seperate method for defining all of the entry points as API
functions (for example, C_GenerateKeyPair which calls
<module>C_GenerateKeyPair).  This patch implements the <module>C_XXX functions
as static functions, as I see no reason not to do that.

Comment 5

[Wan-Teh Chang]

Comment on attachment 83554 [details] [diff] [review]
implement cryptoki entry points as static functions (except for C_GetFunctionList)

r=wtc.	Please wait until the NSS_3_5_BRANCH has
been created to check in this patch.

Comment 6

[Ian McGreer]

Checked in to 3.6.