procd: Don't use syslog before its initialization
[project/procd.git] / jail / jail.c
index 56dc9ca..9d7483c 100644 (file)
 
 #include <stdlib.h>
 #include <unistd.h>
-#include <values.h>
 #include <errno.h>
-#include <stdio.h>
 #include <string.h>
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <libgen.h>
 #include <sched.h>
+#include <linux/limits.h>
+#include <signal.h>
 
-#include "elf.h"
 #include "capabilities.h"
+#include "elf.h"
+#include "fs.h"
+#include "jail.h"
+#include "log.h"
 
-#include <libubox/list.h>
 #include <libubox/uloop.h>
 
 #define STACK_SIZE     (1024 * 1024)
-#define OPT_ARGS       "P:S:C:n:r:w:d:psulo"
+#define OPT_ARGS       "S:C:n:h:r:w:d:psuloc"
 
 static struct {
-       char *path;
        char *name;
+       char *hostname;
        char **jail_argv;
        char *seccomp;
        char *capabilities;
+       int no_new_privs;
        int namespace;
        int procfs;
        int ronly;
        int sysfs;
 } opts;
 
-struct extra {
-       struct list_head list;
-
-       const char *path;
-       const char *name;
-       int readonly;
-};
-
-static LIST_HEAD(extras);
-
 extern int pivot_root(const char *new_root, const char *put_old);
 
 int debug = 0;
@@ -84,47 +77,38 @@ static int mkdir_p(char *dir, mode_t mask)
                return 0;
 
        if (ret)
-               ERROR("mkdir failed on %s: %s\n", dir, strerror(errno));
+               ERROR("mkdir(%s, %d) failed: %s\n", dir, mask, strerror(errno));
 
        return ret;
 }
 
-static int mount_bind(const char *root, const char *path, const char *name, int readonly, int error)
+int mount_bind(const char *root, const char *path, int readonly, int error)
 {
-       const char *p = path;
        struct stat s;
-       char old[256];
-       char new[256];
+       char new[PATH_MAX];
        int fd;
 
-       if (strstr(p, "local"))
-               p = "/lib";
-
-       snprintf(old, sizeof(old), "%s/%s", path, name);
-       snprintf(new, sizeof(new), "%s%s", root, p);
-
-       mkdir_p(new, 0755);
-
-       snprintf(new, sizeof(new), "%s%s/%s", root, p, name);
-
-       if (stat(old, &s)) {
-               ERROR("%s does not exist\n", old);
+       if (stat(path, &s)) {
+               ERROR("stat(%s) failed: %s\n", path, strerror(errno));
                return error;
        }
 
+       snprintf(new, sizeof(new), "%s%s", root, path);
        if (S_ISDIR(s.st_mode)) {
                mkdir_p(new, 0755);
        } else {
+               mkdir_p(dirname(new), 0755);
+               snprintf(new, sizeof(new), "%s%s", root, path);
                fd = creat(new, 0644);
                if (fd == -1) {
-                       ERROR("failed to create %s: %s\n", new, strerror(errno));
+                       ERROR("creat(%s) failed: %s\n", new, strerror(errno));
                        return -1;
                }
                close(fd);
        }
 
-       if (mount(old, new, NULL, MS_BIND, NULL)) {
-               ERROR("failed to mount -B %s %s: %s\n", old, new, strerror(errno));
+       if (mount(path, new, NULL, MS_BIND, NULL)) {
+               ERROR("failed to mount -B %s %s: %s\n", path, new, strerror(errno));
                return -1;
        }
 
@@ -133,64 +117,58 @@ static int mount_bind(const char *root, const char *path, const char *name, int
                return -1;
        }
 
-       DEBUG("mount -B %s %s\n", old, new);
+       DEBUG("mount -B %s %s (%s)\n", path, new, readonly?"ro":"rw");
 
        return 0;
 }
 
-static int build_jail_fs()
+static int build_jail_fs(void)
 {
-       struct library *l;
-       struct extra *m;
-
-       if (mount("tmpfs", opts.path, "tmpfs", MS_NOATIME, "mode=0755")) {
-               ERROR("tmpfs mount failed %s\n", strerror(errno));
+       char jail_root[] = "/tmp/ujail-XXXXXX";
+       if (mkdtemp(jail_root) == NULL) {
+               ERROR("mkdtemp(%s) failed: %s\n", jail_root, strerror(errno));
                return -1;
        }
 
-       if (chdir(opts.path)) {
-               ERROR("failed to chdir() in the jail root\n");
+       /* oldroot can't be MS_SHARED else pivot_root() fails */
+       if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
+               ERROR("private mount failed %s\n", strerror(errno));
                return -1;
        }
 
-       avl_init(&libraries, avl_strcmp, false, NULL);
-       alloc_library_path("/lib64");
-       alloc_library_path("/lib");
-       alloc_library_path("/usr/lib");
-       load_ldso_conf("/etc/ld.so.conf");
-
-       if (elf_load_deps(*opts.jail_argv)) {
-               ERROR("failed to load dependencies\n");
+       if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
+               ERROR("tmpfs mount failed %s\n", strerror(errno));
                return -1;
        }
 
-       if (opts.seccomp && elf_load_deps("libpreload-seccomp.so")) {
-               ERROR("failed to load libpreload-seccomp.so\n");
+       if (chdir(jail_root)) {
+               ERROR("chdir(%s) (jail_root) failed: %s\n", jail_root, strerror(errno));
                return -1;
        }
 
-       avl_for_each_element(&libraries, l, avl)
-               if (mount_bind(opts.path, l->path, l->name, 1, -1))
-                       return -1;
+       if (mount_all(jail_root)) {
+               ERROR("mount_all() failed\n");
+               return -1;
+       }
 
-       list_for_each_entry(m, &extras, list)
-               if (mount_bind(opts.path, m->path, m->name, m->readonly, 0))
-                       return -1;
+       char dirbuf[sizeof(jail_root) + 4];
+       snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
+       mkdir(dirbuf, 0755);
 
-       char *mpoint;
-       if (asprintf(&mpoint, "%s/old", opts.path) < 0) {
-               ERROR("failed to alloc pivot path: %s\n", strerror(errno));
+       if (pivot_root(jail_root, dirbuf) == -1) {
+               ERROR("pivot_root(%s, %s) failed: %s\n", jail_root, dirbuf, strerror(errno));
                return -1;
        }
-       mkdir_p(mpoint, 0755);
-       if (pivot_root(opts.path, mpoint) == -1) {
-               ERROR("pivot_root failed:%s\n", strerror(errno));
-               free(mpoint);
+       if (chdir("/")) {
+               ERROR("chdir(/) (after pivot_root) failed: %s\n", strerror(errno));
                return -1;
        }
-       free(mpoint);
+
+       snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
+       rmdir(dirbuf);
        umount2("/old", MNT_DETACH);
        rmdir("/old");
+
        if (opts.procfs) {
                mkdir("/proc", 0755);
                mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
@@ -209,10 +187,10 @@ static int build_jail_fs()
 static char** build_envp(const char *seccomp)
 {
        static char *envp[MAX_ENVP];
-       static char preload_var[64];
-       static char seccomp_var[64];
+       static char preload_var[PATH_MAX];
+       static char seccomp_var[PATH_MAX];
        static char debug_var[] = "LD_DEBUG=all";
-       char *preload_lib = find_lib("libpreload-seccomp.so");
+       const char *preload_lib = find_lib("libpreload-seccomp.so");
        int count = 0;
 
        if (seccomp && !preload_lib) {
@@ -237,9 +215,10 @@ static void usage(void)
        fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
        fprintf(stderr, "  -S <file>\tseccomp filter config\n");
        fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
+       fprintf(stderr, "  -c\t\tset PR_SET_NO_NEW_PRIVS\n");
        fprintf(stderr, "  -n <name>\tthe name of the jail\n");
        fprintf(stderr, "namespace jail options:\n");
-       fprintf(stderr, "  -P <path>\tpath where the jail will be staged\n");
+       fprintf(stderr, "  -h <hostname>\tchange the hostname of the jail\n");
        fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
        fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
        fprintf(stderr, "  -p\t\tjail has /proc\n");
@@ -256,41 +235,49 @@ ujail will not use namespace/build a jail,\n\
 and will only drop capabilities/apply seccomp filter.\n\n");
 }
 
-static int exec_jail()
+static int exec_jail(void *_notused)
 {
-       char **envp = build_envp(opts.seccomp);
-       if (!envp)
-               exit(EXIT_FAILURE);
-
        if (opts.capabilities && drop_capabilities(opts.capabilities))
                exit(EXIT_FAILURE);
 
-       INFO("exec-ing %s\n", *opts.jail_argv);
-       execve(*opts.jail_argv, opts.jail_argv, envp);
-       //we get there only if execve fails
-       ERROR("failed to execve %s: %s\n", *opts.jail_argv, strerror(errno));
-       exit(EXIT_FAILURE);
-}
+       if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
+                ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %s\n", strerror(errno));
+               exit(EXIT_FAILURE);
+       }
 
-static int spawn_jail(void *arg)
-{
-       if (opts.name && sethostname(opts.name, strlen(opts.name))) {
-               ERROR("failed to sethostname: %s\n", strerror(errno));
+       if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
+                       && sethostname(opts.hostname, strlen(opts.hostname))) {
+               ERROR("sethostname(%s) failed: %s\n", opts.hostname, strerror(errno));
+               exit(EXIT_FAILURE);
        }
 
-       if (build_jail_fs()) {
-               ERROR("failed to build jail fs");
+       if (opts.namespace && build_jail_fs()) {
+               ERROR("failed to build jail fs\n");
                exit(EXIT_FAILURE);
        }
 
-       return exec_jail();
+       char **envp = build_envp(opts.seccomp);
+       if (!envp)
+               exit(EXIT_FAILURE);
+
+       INFO("exec-ing %s\n", *opts.jail_argv);
+       execve(*opts.jail_argv, opts.jail_argv, envp);
+       /* we get there only if execve fails */
+       ERROR("failed to execve %s: %s\n", *opts.jail_argv, strerror(errno));
+       exit(EXIT_FAILURE);
 }
 
 static int jail_running = 1;
 static int jail_return_code = 0;
 
+static void jail_process_timeout_cb(struct uloop_timeout *t);
+static struct uloop_timeout jail_process_timeout = {
+       .cb = jail_process_timeout_cb,
+};
+
 static void jail_process_handler(struct uloop_process *c, int ret)
 {
+       uloop_timeout_cancel(&jail_process_timeout);
        if (WIFEXITED(ret)) {
                jail_return_code = WEXITSTATUS(ret);
                INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
@@ -306,31 +293,25 @@ static struct uloop_process jail_process = {
        .cb = jail_process_handler,
 };
 
-static void add_extra(char *name, int readonly)
+static void jail_process_timeout_cb(struct uloop_timeout *t)
 {
-       struct extra *f;
-
-       if (*name != '/') {
-               ERROR("%s is not an absolute path\n", name);
-               return;
-       }
-
-       f = calloc(1, sizeof(struct extra));
-
-       f->name = basename(name);
-       f->path = dirname(strdup(name));
-       f->readonly = readonly;
+       DEBUG("jail process failed to stop, sending SIGKILL\n");
+       kill(jail_process.pid, SIGKILL);
+}
 
-       list_add_tail(&f->list, &extras);
+static void jail_handle_signal(int signo)
+{
+       DEBUG("forwarding signal %d to the jailed process\n", signo);
+       kill(jail_process.pid, signo);
 }
 
 int main(int argc, char **argv)
 {
+       sigset_t sigmask;
        uid_t uid = getuid();
        char log[] = "/dev/log";
        char ubus[] = "/var/run/ubus.sock";
-       int ret = EXIT_SUCCESS;
-       int ch;
+       int ch, i;
 
        if (uid) {
                ERROR("not root, aborting: %s\n", strerror(errno));
@@ -338,6 +319,8 @@ int main(int argc, char **argv)
        }
 
        umask(022);
+       mount_list_init();
+       init_library_search();
 
        while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
                switch (ch) {
@@ -358,39 +341,40 @@ int main(int argc, char **argv)
                        break;
                case 'S':
                        opts.seccomp = optarg;
-                       add_extra(optarg, 1);
+                       add_mount(optarg, 1, -1);
                        break;
                case 'C':
                        opts.capabilities = optarg;
-                       add_extra(optarg, 1);
                        break;
-               case 'P':
-                       opts.namespace = 1;
-                       opts.path = optarg;
+               case 'c':
+                       opts.no_new_privs = 1;
                        break;
                case 'n':
                        opts.name = optarg;
                        break;
+               case 'h':
+                       opts.hostname = optarg;
+                       break;
                case 'r':
                        opts.namespace = 1;
-                       add_extra(optarg, 1);
+                       add_path_and_deps(optarg, 1, 0, 0);
                        break;
                case 'w':
                        opts.namespace = 1;
-                       add_extra(optarg, 0);
+                       add_path_and_deps(optarg, 0, 0, 0);
                        break;
                case 'u':
                        opts.namespace = 1;
-                       add_extra(ubus, 0);
+                       add_mount(ubus, 0, -1);
                        break;
                case 'l':
                        opts.namespace = 1;
-                       add_extra(log, 0);
+                       add_mount(log, 0, -1);
                        break;
                }
        }
 
-       //no <binary> param found
+       /* no <binary> param found */
        if (argc - optind < 1) {
                usage();
                return EXIT_FAILURE;
@@ -407,53 +391,65 @@ int main(int argc, char **argv)
 
        opts.jail_argv = &argv[optind];
 
+       if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
+               ERROR("failed to load dependencies\n");
+               return -1;
+       }
+
+       if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
+               ERROR("failed to load libpreload-seccomp.so\n");
+               return -1;
+       }
+
        if (opts.name)
                prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 
-       if (opts.namespace && !opts.path && asprintf(&opts.path, "/tmp/%s", basename(*opts.jail_argv)) == -1) {
-               ERROR("failed to asprintf root path: %s\n", strerror(errno));
-               return EXIT_FAILURE;
-       }
+       uloop_init();
 
-       if (opts.namespace && mkdir(opts.path, 0755)) {
-               ERROR("unable to create root path: %s (%s)\n", opts.path, strerror(errno));
-               return EXIT_FAILURE;
+       sigfillset(&sigmask);
+       for (i = 0; i < _NSIG; i++) {
+               struct sigaction s = { 0 };
+
+               if (!sigismember(&sigmask, i))
+                       continue;
+               if ((i == SIGCHLD) || (i == SIGPIPE))
+                       continue;
+
+               s.sa_handler = jail_handle_signal;
+               sigaction(i, &s, NULL);
        }
 
-       uloop_init();
        if (opts.namespace) {
-               jail_process.pid = clone(spawn_jail,
-                       child_stack + STACK_SIZE,
-                       CLONE_NEWUTS | CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD, argv);
+               add_mount("/dev/full", 0, -1);
+               add_mount("/dev/null", 0, -1);
+               add_mount("/dev/urandom", 0, -1);
+               add_mount("/dev/zero", 0, -1);
+
+               int flags = CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD;
+               if (opts.hostname)
+                       flags |= CLONE_NEWUTS;
+               jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, flags, NULL);
        } else {
                jail_process.pid = fork();
        }
 
        if (jail_process.pid > 0) {
-               //parent process
+               /* parent process */
                uloop_process_add(&jail_process);
                uloop_run();
-               uloop_done();
                if (jail_running) {
                        DEBUG("uloop interrupted, killing jail process\n");
                        kill(jail_process.pid, SIGTERM);
-                       waitpid(jail_process.pid, NULL, 0);
+                       uloop_timeout_set(&jail_process_timeout, 1000);
+                       uloop_run();
                }
+               uloop_done();
+               return jail_return_code;
        } else if (jail_process.pid == 0) {
-               //fork child process
-               return exec_jail();
+               /* fork child process */
+               return exec_jail(NULL);
        } else {
                ERROR("failed to clone/fork: %s\n", strerror(errno));
-               ret = EXIT_FAILURE;
-       }
-
-       if (opts.namespace && rmdir(opts.path)) {
-               ERROR("Unable to remove root path: %s (%s)\n", opts.path, strerror(errno));
-               ret = EXIT_FAILURE;
+               return EXIT_FAILURE;
        }
-
-       if (ret)
-               return ret;
-
-       return jail_return_code;
 }