From: Antonio Quartulli <[email protected]>
The NLMSG_TAIL macro is confusing gcc when compiling with -O3, leading
to warnings like:
networking_sitnl.c:143:9: warning: writing 4 bytes into a region of size 0
[-Wstringop-overflow=]
143 | memcpy(RTA_DATA(rta), data, alen);
| ^
networking_sitnl.c:101:21: note: at offset [72, 88] into destination object ‘n’
of size 16
101 | struct nlmsghdr n;
| ^
Replacing the macro with a function is also not effective because gcc
will inline it and get confused again.
The only way out is to write a function that never gets inline'd and
replace the macro with it.
Tested on linux with gcc and clang.
Change-Id: I9306a590a10a7d5cba32abe06d269494fec41ba6
Signed-off-by: Antonio Quartulli <[email protected]>
Acked-by: Frank Lichtenheld <[email protected]>
---
This change was reviewed on Gerrit and approved by at least one
developer. I request to merge it to master.
Gerrit URL: https://gerrit.openvpn.net/c/openvpn/+/788
This mail reflects revision 5 of this Change.
Acked-by according to Gerrit (reflected above):
Frank Lichtenheld <[email protected]>
diff --git a/src/openvpn/networking_sitnl.c b/src/openvpn/networking_sitnl.c
index f53f5ee..6b750e8 100644
--- a/src/openvpn/networking_sitnl.c
+++ b/src/openvpn/networking_sitnl.c
@@ -52,21 +52,34 @@
} \
}
-#define NLMSG_TAIL(nmsg) \
- ((struct rtattr *)(((uint8_t *)(nmsg)) + NLMSG_ALIGN((nmsg)->nlmsg_len)))
-
#define SITNL_NEST(_msg, _max_size, _attr) \
({ \
- struct rtattr *_nest = NLMSG_TAIL(_msg); \
+ struct rtattr *_nest = sitnl_nlmsg_tail(_msg); \
SITNL_ADDATTR(_msg, _max_size, _attr, NULL, 0); \
_nest; \
})
-#define SITNL_NEST_END(_msg, _nest) \
- { \
- _nest->rta_len = (void *)NLMSG_TAIL(_msg) - (void *)_nest; \
+#define SITNL_NEST_END(_msg, _nest) \
+ { \
+ _nest->rta_len = (void *)sitnl_nlmsg_tail(_msg) - (void *)_nest; \
}
+/* This function was originally implemented as a macro, but compiling with
+ * gcc and -O3 was getting confused about the math and thus raising
+ * security warnings on subsequent memcpy() calls.
+ *
+ * Converting the macro to a function was not enough, because gcc was still
+ * inlining it and falling in the same math trap.
+ *
+ * The only way out to avoid any warning/error is to force the function to
+ * not be inline'd.
+ */
+static __attribute__ ((noinline)) void *
+sitnl_nlmsg_tail(const struct nlmsghdr *nlh)
+{
+ return (unsigned char *)nlh + NLMSG_ALIGN(nlh->nlmsg_len);
+}
+
/**
* Generic address data structure used to pass addresses and prefixes as
* argument to AF family agnostic functions
@@ -130,7 +143,7 @@
return -EMSGSIZE;
}
- rta = NLMSG_TAIL(n);
+ rta = sitnl_nlmsg_tail(n);
rta->rta_type = type;
rta->rta_len = len;
_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel
