-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA512
In the handling of netlink sockets is are two small and (probably always) non fatal glitches. Only together they could lead to page faults. It is quite unlikely that it happens, but you cannot say that it is impossible.
Best regards, Sven
We are currently telling the kernel that our netlink request has a special aligned size due to the usage of NLMSG_LENGTH. We don't provide enough space to send this data to the kernel and thus create invalid reads by the kernel when calling the sendto. The alignment is currently small enough that we cannot create page faults which makes these invalid reads alone a nonfatal problem. We should provide a buffer with the exact amount of data we write in nlh.nlmsg_len or even more.
Signed-off-by: Sven Eckelmann sven.eckelmann@gmx.de --- batman/linux/route.c | 138 ++++++++++++++++++++++++++------------------------ 1 files changed, 72 insertions(+), 66 deletions(-)
diff --git a/batman/linux/route.c b/batman/linux/route.c index 8fe6015..3ddce2e 100644 --- a/batman/linux/route.c +++ b/batman/linux/route.c @@ -182,11 +182,12 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src struct iovec iov; struct msghdr msg; struct nlmsghdr *nh; - struct { + struct req_s { struct nlmsghdr nlh; struct rtmsg rtm; char buff[4 * (sizeof(struct rtattr) + 4)]; - } req; + } *req; + char req_buf[NLMSG_LENGTH(sizeof(struct req_s))];
iov.iov_base = buf; iov.iov_len = sizeof(buf); @@ -227,8 +228,9 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src }
+ req = (struct req_s*)req_buf; memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(&req, 0, sizeof(req)); + memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK; @@ -242,39 +244,39 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src if (src_ip != 0) len += sizeof(struct rtattr) + 4;
- req.nlh.nlmsg_len = NLMSG_LENGTH(len); - req.nlh.nlmsg_pid = getpid(); - req.rtm.rtm_family = AF_INET; - req.rtm.rtm_table = rt_table; - req.rtm.rtm_dst_len = netmask; + req->nlh.nlmsg_len = NLMSG_LENGTH(len); + req->nlh.nlmsg_pid = getpid(); + req->rtm.rtm_family = AF_INET; + req->rtm.rtm_table = rt_table; + req->rtm.rtm_dst_len = netmask;
if (route_action == ROUTE_DEL) {
- req.nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; - req.nlh.nlmsg_type = RTM_DELROUTE; - req.rtm.rtm_scope = RT_SCOPE_NOWHERE; + req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + req->nlh.nlmsg_type = RTM_DELROUTE; + req->rtm.rtm_scope = RT_SCOPE_NOWHERE;
} else {
- req.nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_APPEND; - req.nlh.nlmsg_type = RTM_NEWROUTE; + req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_APPEND; + req->nlh.nlmsg_type = RTM_NEWROUTE;
if (route_type == ROUTE_TYPE_UNICAST && my_router == 0 && src_ip != 0) - req.rtm.rtm_scope = RT_SCOPE_LINK; + req->rtm.rtm_scope = RT_SCOPE_LINK; else - req.rtm.rtm_scope = RT_SCOPE_UNIVERSE; + req->rtm.rtm_scope = RT_SCOPE_UNIVERSE;
- req.rtm.rtm_protocol = RTPROT_STATIC; /* may be changed to some batman specific value - see <linux/rtnetlink.h> */ + req->rtm.rtm_protocol = RTPROT_STATIC; /* may be changed to some batman specific value - see <linux/rtnetlink.h> */
switch(route_type) { case ROUTE_TYPE_UNICAST: - req.rtm.rtm_type = RTN_UNICAST; + req->rtm.rtm_type = RTN_UNICAST; break; case ROUTE_TYPE_THROW: - req.rtm.rtm_type = RTN_THROW; + req->rtm.rtm_type = RTN_THROW; break; case ROUTE_TYPE_UNREACHABLE: - req.rtm.rtm_type = RTN_UNREACHABLE; + req->rtm.rtm_type = RTN_UNREACHABLE; break; default: debug_output(0, "Error - unknown route type (add_del_route): %i\n", route_type); @@ -282,28 +284,28 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src } }
- rta = (struct rtattr *)req.buff; + rta = (struct rtattr *)req->buff; rta->rta_type = RTA_DST; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + sizeof(struct rtattr), (char *)&dest, 4); + memcpy(((char *)req->buff) + sizeof(struct rtattr), (char *)&dest, 4);
if (route_type == ROUTE_TYPE_UNICAST) {
- rta = (struct rtattr *)(req.buff + sizeof(struct rtattr) + 4); + rta = (struct rtattr *)(req->buff + sizeof(struct rtattr) + 4); rta->rta_type = RTA_GATEWAY; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + 2 * sizeof(struct rtattr) + 4, (char *)&my_router, 4); + memcpy(((char *)req->buff) + 2 * sizeof(struct rtattr) + 4, (char *)&my_router, 4);
- rta = (struct rtattr *)(req.buff + 2 * sizeof(struct rtattr) + 8); + rta = (struct rtattr *)(req->buff + 2 * sizeof(struct rtattr) + 8); rta->rta_type = RTA_OIF; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + 3 * sizeof(struct rtattr) + 8, (char *)&ifi, 4); + memcpy(((char *)req->buff) + 3 * sizeof(struct rtattr) + 8, (char *)&ifi, 4);
if (src_ip != 0) { - rta = (struct rtattr *)(req.buff + 3 * sizeof(struct rtattr) + 12); + rta = (struct rtattr *)(req->buff + 3 * sizeof(struct rtattr) + 12); rta->rta_type = RTA_PREFSRC; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + 4 * sizeof(struct rtattr) + 12, (char *)&src_ip, 4); + memcpy(((char *)req->buff) + 4 * sizeof(struct rtattr) + 12, (char *)&src_ip, 4); }
} @@ -315,7 +317,7 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src
}
- if (sendto(netlink_sock, &req, req.nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output(0, "Error - can't send message to kernel via netlink socket for routing table manipulation: %s\n", strerror(errno)); close(netlink_sock); @@ -364,11 +366,12 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p struct iovec iov; struct msghdr msg; struct nlmsghdr *nh; - struct { + struct req_s { struct nlmsghdr nlh; struct rtmsg rtm; char buff[2 * (sizeof(struct rtattr) + 4)]; - } req; + } *req; + char req_buf[NLMSG_LENGTH(sizeof(struct req_s))];
iov.iov_base = buf; iov.iov_len = sizeof(buf); @@ -382,8 +385,9 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
+ req = (struct req_s*)req_buf; memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(&req, 0, sizeof(req)); + memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK; @@ -393,53 +397,53 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p if (prio != 0) len += sizeof(struct rtattr) + 4;
- req.nlh.nlmsg_len = NLMSG_LENGTH(len); - req.nlh.nlmsg_pid = getpid(); - req.rtm.rtm_family = AF_INET; - req.rtm.rtm_table = rt_table; + req->nlh.nlmsg_len = NLMSG_LENGTH(len); + req->nlh.nlmsg_pid = getpid(); + req->rtm.rtm_family = AF_INET; + req->rtm.rtm_table = rt_table;
if (rule_action == RULE_DEL) {
- req.nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; - req.nlh.nlmsg_type = RTM_DELRULE; - req.rtm.rtm_scope = RT_SCOPE_NOWHERE; + req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + req->nlh.nlmsg_type = RTM_DELRULE; + req->rtm.rtm_scope = RT_SCOPE_NOWHERE;
} else {
- req.nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL; - req.nlh.nlmsg_type = RTM_NEWRULE; - req.rtm.rtm_scope = RT_SCOPE_UNIVERSE; - req.rtm.rtm_protocol = RTPROT_STATIC; - req.rtm.rtm_type = RTN_UNICAST; + req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL; + req->nlh.nlmsg_type = RTM_NEWRULE; + req->rtm.rtm_scope = RT_SCOPE_UNIVERSE; + req->rtm.rtm_protocol = RTPROT_STATIC; + req->rtm.rtm_type = RTN_UNICAST;
}
switch(rule_type) { case RULE_TYPE_SRC: - rta = (struct rtattr *)req.buff; + rta = (struct rtattr *)req->buff; rta->rta_type = RTA_SRC; - req.rtm.rtm_src_len = netmask; + req->rtm.rtm_src_len = netmask; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + sizeof(struct rtattr), (char *)&network, 4); + memcpy(((char *)req->buff) + sizeof(struct rtattr), (char *)&network, 4); break; case RULE_TYPE_DST: - rta = (struct rtattr *)req.buff; + rta = (struct rtattr *)req->buff; rta->rta_type = RTA_DST; - req.rtm.rtm_dst_len = netmask; + req->rtm.rtm_dst_len = netmask; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + sizeof(struct rtattr), (char *)&network, 4); + memcpy(((char *)req->buff) + sizeof(struct rtattr), (char *)&network, 4); break; case RULE_TYPE_IIF: - rta = (struct rtattr *)req.buff; + rta = (struct rtattr *)req->buff; rta->rta_len = sizeof(struct rtattr) + 4;
if (rule_action == RULE_DEL) { rta->rta_type = RTA_SRC; - memcpy(((char *)&req.buff) + sizeof(struct rtattr), (char *)&network, 4); + memcpy(((char *)req->buff) + sizeof(struct rtattr), (char *)&network, 4); } else { rta->rta_type = RTA_IIF; - memcpy(((char *)&req.buff) + sizeof(struct rtattr), iif, 4); + memcpy(((char *)req->buff) + sizeof(struct rtattr), iif, 4); } break; default: @@ -448,10 +452,10 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
if (prio != 0) { - rta = (struct rtattr *)(req.buff + sizeof(struct rtattr) + 4); + rta = (struct rtattr *)(req->buff + sizeof(struct rtattr) + 4); rta->rta_type = RTA_PRIORITY; rta->rta_len = sizeof(struct rtattr) + 4; - memcpy(((char *)&req.buff) + 2 * sizeof(struct rtattr) + 4, (char *)&prio, 4); + memcpy(((char *)req->buff) + 2 * sizeof(struct rtattr) + 4, (char *)&prio, 4); }
@@ -463,7 +467,7 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
- if (sendto(netlink_sock, &req, req.nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output( 0, "Error - can't send message to kernel via netlink socket for routing rule manipulation: %s\n", strerror(errno)); close(netlink_sock); @@ -628,29 +632,31 @@ int flush_routes_rules(int8_t is_rule) struct msghdr msg; struct nlmsghdr *nh; struct rtmsg *rtm; - struct { + struct req_s { struct nlmsghdr nlh; struct rtmsg rtm; - } req; + } *req; + char req_buf[NLMSG_LENGTH(sizeof(struct req_s))]; + struct rtattr *rtap;
iov.iov_base = buf; iov.iov_len = sizeof(buf);
- + req = (struct req_s*)req_buf; memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(&req, 0, sizeof(req)); + memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK;
- req.nlh.nlmsg_len = NLMSG_LENGTH(sizeof(req)); - req.nlh.nlmsg_pid = getpid(); - req.rtm.rtm_family = AF_INET; + req->nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct req_s)); + req->nlh.nlmsg_pid = getpid(); + req->rtm.rtm_family = AF_INET;
- req.nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; - req.nlh.nlmsg_type = (is_rule ? RTM_GETRULE : RTM_GETROUTE); - req.rtm.rtm_scope = RTN_UNICAST; + req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req->nlh.nlmsg_type = (is_rule ? RTM_GETRULE : RTM_GETROUTE); + req->rtm.rtm_scope = RTN_UNICAST;
if ((netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE)) < 0) {
@@ -660,7 +666,7 @@ int flush_routes_rules(int8_t is_rule) }
- if (sendto(netlink_sock, &req, req.nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output(0, "Error - can't send message to kernel via netlink socket for flushing the routing table: %s\n", strerror(errno)); close(netlink_sock);
The parameter len of NLMSG_LENGTH is only the size of the payload of the request. If we use this function to calculate the size of the buffer we send to the kernel we would add (sizeof(nlmsghdr)+padding) and the size of our own structure which also has a nlmsghdr included. We must split the req structure into a header part and the payload to calculate the correct size of our buffer and having the payload always start at the correct address.
Signed-off-by: Sven Eckelmann sven.eckelmann@gmx.de --- batman/linux/route.c | 59 ++++++++++++++++++++++++++----------------------- 1 files changed, 31 insertions(+), 28 deletions(-)
diff --git a/batman/linux/route.c b/batman/linux/route.c index 3ddce2e..4be0179 100644 --- a/batman/linux/route.c +++ b/batman/linux/route.c @@ -182,8 +182,8 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src struct iovec iov; struct msghdr msg; struct nlmsghdr *nh; + struct nlmsghdr *nlh; struct req_s { - struct nlmsghdr nlh; struct rtmsg rtm; char buff[4 * (sizeof(struct rtattr) + 4)]; } *req; @@ -228,9 +228,10 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src }
- req = (struct req_s*)req_buf; + nlh = (struct nlmsghdr *)req_buf; + req = (struct req_s*)NLMSG_DATA(req_buf); memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); + memset(req_buf, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK; @@ -244,22 +245,22 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src if (src_ip != 0) len += sizeof(struct rtattr) + 4;
- req->nlh.nlmsg_len = NLMSG_LENGTH(len); - req->nlh.nlmsg_pid = getpid(); + nlh->nlmsg_len = NLMSG_LENGTH(len); + nlh->nlmsg_pid = getpid(); req->rtm.rtm_family = AF_INET; req->rtm.rtm_table = rt_table; req->rtm.rtm_dst_len = netmask;
if (route_action == ROUTE_DEL) {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; - req->nlh.nlmsg_type = RTM_DELROUTE; + nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + nlh->nlmsg_type = RTM_DELROUTE; req->rtm.rtm_scope = RT_SCOPE_NOWHERE;
} else {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_APPEND; - req->nlh.nlmsg_type = RTM_NEWROUTE; + nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_APPEND; + nlh->nlmsg_type = RTM_NEWROUTE;
if (route_type == ROUTE_TYPE_UNICAST && my_router == 0 && src_ip != 0) req->rtm.rtm_scope = RT_SCOPE_LINK; @@ -317,7 +318,7 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src
}
- if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req_buf, nlh->nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output(0, "Error - can't send message to kernel via netlink socket for routing table manipulation: %s\n", strerror(errno)); close(netlink_sock); @@ -366,8 +367,8 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p struct iovec iov; struct msghdr msg; struct nlmsghdr *nh; + struct nlmsghdr *nlh; struct req_s { - struct nlmsghdr nlh; struct rtmsg rtm; char buff[2 * (sizeof(struct rtattr) + 4)]; } *req; @@ -385,9 +386,10 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
- req = (struct req_s*)req_buf; + nlh = (struct nlmsghdr *)req_buf; + req = (struct req_s*)NLMSG_DATA(req_buf); memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); + memset(req_buf, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK; @@ -397,22 +399,22 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p if (prio != 0) len += sizeof(struct rtattr) + 4;
- req->nlh.nlmsg_len = NLMSG_LENGTH(len); - req->nlh.nlmsg_pid = getpid(); + nlh->nlmsg_len = NLMSG_LENGTH(len); + nlh->nlmsg_pid = getpid(); req->rtm.rtm_family = AF_INET; req->rtm.rtm_table = rt_table;
if (rule_action == RULE_DEL) {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; - req->nlh.nlmsg_type = RTM_DELRULE; + nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + nlh->nlmsg_type = RTM_DELRULE; req->rtm.rtm_scope = RT_SCOPE_NOWHERE;
} else {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL; - req->nlh.nlmsg_type = RTM_NEWRULE; + nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL; + nlh->nlmsg_type = RTM_NEWRULE; req->rtm.rtm_scope = RT_SCOPE_UNIVERSE; req->rtm.rtm_protocol = RTPROT_STATIC; req->rtm.rtm_type = RTN_UNICAST; @@ -467,7 +469,7 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
- if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req_buf, nlh->nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output( 0, "Error - can't send message to kernel via netlink socket for routing rule manipulation: %s\n", strerror(errno)); close(netlink_sock); @@ -632,8 +634,8 @@ int flush_routes_rules(int8_t is_rule) struct msghdr msg; struct nlmsghdr *nh; struct rtmsg *rtm; + struct nlmsghdr *nlh; struct req_s { - struct nlmsghdr nlh; struct rtmsg rtm; } *req; char req_buf[NLMSG_LENGTH(sizeof(struct req_s))]; @@ -643,19 +645,20 @@ int flush_routes_rules(int8_t is_rule) iov.iov_base = buf; iov.iov_len = sizeof(buf);
- req = (struct req_s*)req_buf; + nlh = (struct nlmsghdr *)req_buf; + req = (struct req_s*)NLMSG_DATA(req_buf); memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); + memset(req_buf, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK;
- req->nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct req_s)); - req->nlh.nlmsg_pid = getpid(); + nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct req_s)); + nlh->nlmsg_pid = getpid(); req->rtm.rtm_family = AF_INET;
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; - req->nlh.nlmsg_type = (is_rule ? RTM_GETRULE : RTM_GETROUTE); + nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + nlh->nlmsg_type = (is_rule ? RTM_GETRULE : RTM_GETROUTE); req->rtm.rtm_scope = RTN_UNICAST;
if ((netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE)) < 0) { @@ -666,7 +669,7 @@ int flush_routes_rules(int8_t is_rule) }
- if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req_buf, nlh->nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output(0, "Error - can't send message to kernel via netlink socket for flushing the routing table: %s\n", strerror(errno)); close(netlink_sock);
The parameter len of NLMSG_LENGTH is only the size of the payload of the request. If we use this function to calculate the size of the buffer we send to the kernel we would add (sizeof(nlmsghdr)+padding) and the size of our own structure which also has a nlmsghdr included. We must split the req structure into a header part and the payload to calculate the correct size of our buffer and having the payload always start at the correct address.
Signed-off-by: Sven Eckelmann sven.eckelmann@gmx.de --- batman/linux/route.c | 56 +++++++++++++++++++++++++------------------------- 1 files changed, 28 insertions(+), 28 deletions(-)
diff --git a/batman/linux/route.c b/batman/linux/route.c index 3ddce2e..99c295f 100644 --- a/batman/linux/route.c +++ b/batman/linux/route.c @@ -183,7 +183,6 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src struct msghdr msg; struct nlmsghdr *nh; struct req_s { - struct nlmsghdr nlh; struct rtmsg rtm; char buff[4 * (sizeof(struct rtattr) + 4)]; } *req; @@ -228,9 +227,10 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src }
- req = (struct req_s*)req_buf; + nh = (struct nlmsghdr *)req_buf; + req = (struct req_s*)NLMSG_DATA(req_buf); memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); + memset(req_buf, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK; @@ -244,22 +244,22 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src if (src_ip != 0) len += sizeof(struct rtattr) + 4;
- req->nlh.nlmsg_len = NLMSG_LENGTH(len); - req->nlh.nlmsg_pid = getpid(); + nh->nlmsg_len = NLMSG_LENGTH(len); + nh->nlmsg_pid = getpid(); req->rtm.rtm_family = AF_INET; req->rtm.rtm_table = rt_table; req->rtm.rtm_dst_len = netmask;
if (route_action == ROUTE_DEL) {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; - req->nlh.nlmsg_type = RTM_DELROUTE; + nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + nh->nlmsg_type = RTM_DELROUTE; req->rtm.rtm_scope = RT_SCOPE_NOWHERE;
} else {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_APPEND; - req->nlh.nlmsg_type = RTM_NEWROUTE; + nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_APPEND; + nh->nlmsg_type = RTM_NEWROUTE;
if (route_type == ROUTE_TYPE_UNICAST && my_router == 0 && src_ip != 0) req->rtm.rtm_scope = RT_SCOPE_LINK; @@ -317,7 +317,7 @@ void add_del_route(uint32_t dest, uint8_t netmask, uint32_t router, uint32_t src
}
- if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req_buf, nh->nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output(0, "Error - can't send message to kernel via netlink socket for routing table manipulation: %s\n", strerror(errno)); close(netlink_sock); @@ -367,7 +367,6 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p struct msghdr msg; struct nlmsghdr *nh; struct req_s { - struct nlmsghdr nlh; struct rtmsg rtm; char buff[2 * (sizeof(struct rtattr) + 4)]; } *req; @@ -385,9 +384,10 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
- req = (struct req_s*)req_buf; + nh = (struct nlmsghdr *)req_buf; + req = (struct req_s*)NLMSG_DATA(req_buf); memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); + memset(req_buf, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK; @@ -397,22 +397,22 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p if (prio != 0) len += sizeof(struct rtattr) + 4;
- req->nlh.nlmsg_len = NLMSG_LENGTH(len); - req->nlh.nlmsg_pid = getpid(); + nh->nlmsg_len = NLMSG_LENGTH(len); + nh->nlmsg_pid = getpid(); req->rtm.rtm_family = AF_INET; req->rtm.rtm_table = rt_table;
if (rule_action == RULE_DEL) {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; - req->nlh.nlmsg_type = RTM_DELRULE; + nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + nh->nlmsg_type = RTM_DELRULE; req->rtm.rtm_scope = RT_SCOPE_NOWHERE;
} else {
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL; - req->nlh.nlmsg_type = RTM_NEWRULE; + nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL; + nh->nlmsg_type = RTM_NEWRULE; req->rtm.rtm_scope = RT_SCOPE_UNIVERSE; req->rtm.rtm_protocol = RTPROT_STATIC; req->rtm.rtm_type = RTN_UNICAST; @@ -467,7 +467,7 @@ void add_del_rule(uint32_t network, uint8_t netmask, int8_t rt_table, uint32_t p }
- if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req_buf, nh->nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output( 0, "Error - can't send message to kernel via netlink socket for routing rule manipulation: %s\n", strerror(errno)); close(netlink_sock); @@ -633,7 +633,6 @@ int flush_routes_rules(int8_t is_rule) struct nlmsghdr *nh; struct rtmsg *rtm; struct req_s { - struct nlmsghdr nlh; struct rtmsg rtm; } *req; char req_buf[NLMSG_LENGTH(sizeof(struct req_s))]; @@ -643,19 +642,20 @@ int flush_routes_rules(int8_t is_rule) iov.iov_base = buf; iov.iov_len = sizeof(buf);
- req = (struct req_s*)req_buf; + nh = (struct nlmsghdr *)req_buf; + req = (struct req_s*)NLMSG_DATA(req_buf); memset(&nladdr, 0, sizeof(struct sockaddr_nl)); - memset(req, 0, NLMSG_LENGTH(sizeof(struct req_s))); + memset(req_buf, 0, NLMSG_LENGTH(sizeof(struct req_s))); memset(&msg, 0, sizeof(struct msghdr));
nladdr.nl_family = AF_NETLINK;
- req->nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct req_s)); - req->nlh.nlmsg_pid = getpid(); + nh->nlmsg_len = NLMSG_LENGTH(sizeof(struct req_s)); + nh->nlmsg_pid = getpid(); req->rtm.rtm_family = AF_INET;
- req->nlh.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; - req->nlh.nlmsg_type = (is_rule ? RTM_GETRULE : RTM_GETROUTE); + nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + nh->nlmsg_type = (is_rule ? RTM_GETRULE : RTM_GETROUTE); req->rtm.rtm_scope = RTN_UNICAST;
if ((netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE)) < 0) { @@ -666,7 +666,7 @@ int flush_routes_rules(int8_t is_rule) }
- if (sendto(netlink_sock, req, req->nlh.nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) { + if (sendto(netlink_sock, req_buf, nh->nlmsg_len, 0, (struct sockaddr *)&nladdr, sizeof(struct sockaddr_nl)) < 0) {
debug_output(0, "Error - can't send message to kernel via netlink socket for flushing the routing table: %s\n", strerror(errno)); close(netlink_sock);
On Thursday 21 May 2009 00:25:56 Sven Eckelmann wrote:
The parameter len of NLMSG_LENGTH is only the size of the payload of the request. If we use this function to calculate the size of the buffer we send to the kernel we would add (sizeof(nlmsghdr)+padding) and the size of our own structure which also has a nlmsghdr included. We must split the req structure into a header part and the payload to calculate the correct size of our buffer and having the payload always start at the correct address.
Signed-off-by: Sven Eckelmann sven.eckelmann@gmx.de
Great findings - the patches are in now. :-)
Regards, Marek
b.a.t.m.a.n@lists.open-mesh.org
-
Marek Lindner -
Sven Eckelmann