The previous change to assume() did address the coverity warning about
one direction of the shift in HASH_KEY, let's constrain the other in
HASH_SIZE as well.
To be fair, the hash table *will* break at 1G entries, but at that point
we have other problems RAM-wise. (Could bump the thing to 64-bit, but
then we need better item hash functions too on every single user.)
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
uint8_t minshift, maxshift;
};
-#define _HASH_SIZE(tabshift) \
- ((1U << (tabshift)) >> 1)
+#define _HASH_SIZE(tabshift) \
+ ({ \
+ assume((tabshift) <= 31); \
+ (1U << (tabshift)) >> 1; \
+ })
#define HASH_SIZE(head) \
_HASH_SIZE((head).tabshift)
#define _HASH_KEY(tabshift, val) \
({ \
- assume((tabshift) >= 2 && (tabshift) <= 33); \
+ assume((tabshift) >= 2 && (tabshift) <= 31); \
(val) >> (33 - (tabshift)); \
})
#define HASH_KEY(head, val) \