static int __nonuserns_sysctl_op(struct sysctl_req *req, size_t nr_req, int op)
{
int dir, ret, exit_code = -1;;
dir = open("/proc/sys", O_RDONLY, O_DIRECTORY);
if (dir < 0) {
pr_perror("Can't open sysctl dir");
return -1;
}
while (nr_req--) {
int fd, flags;
if (op == CTL_READ)
flags = O_RDONLY;
else
flags = O_WRONLY;
fd = openat(dir, req->name, flags);
if (fd < 0) {
if (errno == ENOENT && (req->flags & CTL_FLAGS_OPTIONAL)) {
req++;
continue;
}
pr_perror("Can't open sysctl %s", req->name);
goto out;
}
ret = do_sysctl_op(fd, req, op);
if (ret)
goto out;
close(fd);
req++;
}
exit_code = 0;
out:
close(dir);
return exit_code;
}
static int __userns_sysctl_op(void *arg, int proc_fd, pid_t pid)
{
int fd, ret = -1, dir, i, status, *fds = NULL;
struct sysctl_userns_req *userns_req = arg;
int op = userns_req->op;
struct sysctl_req *req, **reqs = NULL;
sigset_t blockmask, oldmask;
pid_t worker;
// fix up the pointer
req = userns_req->reqs = (struct sysctl_req *) &userns_req[1];
/* For files in the IPC/UTS namespaces, restoring is more complicated
* than for net. Unprivileged users cannot even open these files, so
* they must be opened by usernsd. However, the value in the kernel is
* changed for the IPC/UTS namespace that write()s to the open sysctl
* file (not who opened it). So, we must set the value from inside the
* usernsd caller's namespace. We:
*
* 1. unsd opens the sysctl files
* 2. forks a task
* 3. setns()es to the UTS/IPC namespace of the caller
* 4. write()s to the files and exits
*/
dir = open("/proc/sys", O_RDONLY, O_DIRECTORY);
if (dir < 0) {
pr_perror("Can't open sysctl dir");
return -1;
}
fds = xmalloc(sizeof(int) * userns_req->nr_req);
if (!fds)
goto out;
reqs = xmalloc(sizeof(struct sysctl_req) * userns_req->nr_req);
if (!reqs)
goto out;
memset(fds, -1, sizeof(int) * userns_req->nr_req);
for (i = 0; i < userns_req->nr_req; i++) {
int arg_len = sysctl_userns_arg_size(req->type);
int name_len = strlen((char *) &req[1]) + 1;
int total_len = sizeof(*req) + arg_len + name_len;
int flags;
/* fix up the pointers */
req->name = (char *) &req[1];
req->arg = req->name + name_len;
if (((char *) req) + total_len >= ((char *) userns_req) + MAX_UNSFD_MSG_SIZE) {
pr_err("bad sysctl req %s, too big: %d\n", req->name, total_len);
goto out;
}
if (op == CTL_READ)
flags = O_RDONLY;
else
flags = O_WRONLY;
fd = openat(dir, req->name, flags);
if (fd < 0) {
if (errno == ENOENT && (req->flags & CTL_FLAGS_OPTIONAL))
continue;
pr_perror("Can't open sysctl %s", req->name);
goto out;
}
/* save a pointer to the req, so we don't need to recompute its
* location
*/
reqs[i] = req;
fds[i] = fd;
req = (struct sysctl_req *) (((char *) req) + total_len);
}
/*
* Don't let the sigchld_handler() mess with us
* calling waitpid() on the exited worker. The
* same is done in cr_system().
*/
sigemptyset(&blockmask);
sigaddset(&blockmask, SIGCHLD);
sigprocmask(SIG_BLOCK, &blockmask, &oldmask);
worker = fork();
if (worker < 0)
goto out;
if (!worker) {
int nsfd;
const char *nsname = ns_to_string(userns_req->ns);
BUG_ON(!nsname);
nsfd = openat(proc_fd, nsname, O_RDONLY);
if (nsfd < 0) {
pr_perror("failed to open pid %d's ns %s", pid, nsname);
exit(1);
}
if (setns(nsfd, 0) < 0) {
pr_perror("failed to setns to %d's ns %s", pid, nsname);
exit(1);
}
close(nsfd);
for (i = 0; i < userns_req->nr_req; i++) {
if (do_sysctl_op(fds[i], reqs[i], op) < 0)
exit(1);
}
exit(0);
}
if (waitpid(worker, &status, 0) != worker) {
pr_perror("worker didn't die?");
kill(worker, SIGKILL);
goto out;
}
sigprocmask(SIG_SETMASK, &oldmask, NULL);
if (!WIFEXITED(status) || WEXITSTATUS(status)) {
pr_err("worker failed: %d\n", status);
goto out;
}
ret = 0;
out:
if (fds) {
for (i = 0; i < userns_req->nr_req; i++) {
if (fds[i] < 0)
break;
close_safe(&fds[i]);
}
xfree(fds);
}
if (reqs)
xfree(reqs);
close_safe(&dir);
return ret;
}
