Files
sudo/src/exec_pty.c

1925 lines
57 KiB
C

/*
* SPDX-License-Identifier: ISC
*
* Copyright (c) 2009-2021 Todd C. Miller <Todd.Miller@sudo.ws>
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*/
/*
* This is an open source non-commercial project. Dear PVS-Studio, please check it.
* PVS-Studio Static Code Analyzer for C, C++ and C#: http://www.viva64.com
*/
#include <config.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <sys/wait.h>
#include <sys/ioctl.h>
#if defined(HAVE_STDINT_H)
# include <stdint.h>
#elif defined(HAVE_INTTYPES_H)
# include <inttypes.h>
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <termios.h> /* for struct winsize on HP-UX */
#include "sudo.h"
#include "sudo_exec.h"
#include "sudo_plugin.h"
#include "sudo_plugin_int.h"
/* Evaluates to true if the event has /dev/tty as its fd. */
#define USERTTY_EVENT(_ev) (sudo_ev_get_fd((_ev)) == io_fds[SFD_USERTTY])
#define TERM_COOKED 0
#define TERM_RAW 1
/* Tail queue of messages to send to the monitor. */
struct monitor_message {
TAILQ_ENTRY(monitor_message) entries;
struct command_status cstat;
};
TAILQ_HEAD(monitor_message_list, monitor_message);
struct exec_closure_pty {
struct command_details *details;
struct sudo_event_base *evbase;
struct sudo_event *backchannel_event;
struct sudo_event *fwdchannel_event;
struct sudo_event *sigint_event;
struct sudo_event *sigquit_event;
struct sudo_event *sigtstp_event;
struct sudo_event *sigterm_event;
struct sudo_event *sighup_event;
struct sudo_event *sigalrm_event;
struct sudo_event *sigusr1_event;
struct sudo_event *sigusr2_event;
struct sudo_event *sigchld_event;
struct sudo_event *sigwinch_event;
struct command_status *cstat;
struct monitor_message_list monitor_messages;
pid_t monitor_pid;
pid_t cmnd_pid;
pid_t ppgrp;
short rows;
short cols;
};
/*
* I/O buffer with associated read/write events and a logging action.
* Used to, e.g. pass data from the pty to the user's terminal
* and any I/O logging plugins.
*/
struct io_buffer;
typedef bool (*sudo_io_action_t)(const char *, unsigned int, struct io_buffer *);
struct io_buffer {
SLIST_ENTRY(io_buffer) entries;
struct exec_closure_pty *ec;
struct sudo_event *revent;
struct sudo_event *wevent;
sudo_io_action_t action;
int len; /* buffer length (how much produced) */
int off; /* write position (how much already consumed) */
char buf[64 * 1024];
};
SLIST_HEAD(io_buffer_list, io_buffer);
static char ptyname[PATH_MAX];
int io_fds[6] = { -1, -1, -1, -1, -1, -1};
static bool foreground, pipeline;
static int ttymode = TERM_COOKED;
static sigset_t ttyblock;
static struct io_buffer_list iobufs;
static const char *utmp_user;
static void del_io_events(bool nonblocking);
static void sync_ttysize(struct exec_closure_pty *ec);
static int safe_close(int fd);
static void ev_free_by_fd(struct sudo_event_base *evbase, int fd);
static pid_t check_foreground(struct exec_closure_pty *ec);
static void add_io_events(struct sudo_event_base *evbase);
static void schedule_signal(struct exec_closure_pty *ec, int signo);
/*
* Cleanup hook for sudo_fatal()/sudo_fatalx()
*/
void
pty_cleanup(void)
{
debug_decl(cleanup, SUDO_DEBUG_EXEC);
if (io_fds[SFD_USERTTY] != -1)
sudo_term_restore(io_fds[SFD_USERTTY], false);
if (utmp_user != NULL)
utmp_logout(ptyname, 0);
debug_return;
}
/*
* Allocate a pty if /dev/tty is a tty.
* Fills in io_fds[SFD_USERTTY], io_fds[SFD_LEADER], io_fds[SFD_FOLLOWER]
* and ptyname globals.
*/
static bool
pty_setup(struct command_details *details, const char *tty)
{
debug_decl(pty_setup, SUDO_DEBUG_EXEC);
io_fds[SFD_USERTTY] = open(_PATH_TTY, O_RDWR);
if (io_fds[SFD_USERTTY] == -1) {
sudo_debug_printf(SUDO_DEBUG_INFO, "%s: no %s, not allocating a pty",
__func__, _PATH_TTY);
debug_return_bool(false);
}
if (!get_pty(&io_fds[SFD_LEADER], &io_fds[SFD_FOLLOWER],
ptyname, sizeof(ptyname), details->cred.euid))
sudo_fatal("%s", U_("unable to allocate pty"));
/* Update tty name in command details (used by SELinux and AIX). */
details->tty = ptyname;
/* Add entry to utmp/utmpx? */
if (ISSET(details->flags, CD_SET_UTMP)) {
utmp_user =
details->utmp_user ? details->utmp_user : user_details.username;
utmp_login(tty, ptyname, io_fds[SFD_FOLLOWER], utmp_user);
}
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: %s fd %d, pty leader fd %d, pty follower fd %d",
__func__, _PATH_TTY, io_fds[SFD_USERTTY], io_fds[SFD_LEADER],
io_fds[SFD_FOLLOWER]);
debug_return_bool(true);
}
/*
* Make the tty follower the controlling tty.
* This is only used by the monitor but ptyname[] is static.
*/
int
pty_make_controlling(void)
{
if (io_fds[SFD_FOLLOWER] != -1) {
#ifdef TIOCSCTTY
if (ioctl(io_fds[SFD_FOLLOWER], TIOCSCTTY, NULL) != 0)
return -1;
#else
/* Set controlling tty by reopening pty follower. */
int fd = open(ptyname, O_RDWR);
if (fd == -1)
return -1;
close(fd);
#endif
}
return 0;
}
/* Call I/O plugin tty input log method. */
static bool
log_ttyin(const char *buf, unsigned int n, struct io_buffer *iob)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
bool ret = true;
debug_decl(log_ttyin, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->log_ttyin) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->log_ttyin(buf, n, &errstr);
if (rc <= 0) {
if (rc < 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->log_ttyin = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("I/O plugin error"),
iob->ec->details->info);
} else {
audit_reject(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("command rejected by I/O plugin"),
iob->ec->details->info);
}
ret = false;
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return_bool(ret);
}
/* Call I/O plugin stdin log method. */
static bool
log_stdin(const char *buf, unsigned int n, struct io_buffer *iob)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
bool ret = true;
debug_decl(log_stdin, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->log_stdin) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->log_stdin(buf, n, &errstr);
if (rc <= 0) {
if (rc < 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->log_stdin = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("I/O plugin error"),
iob->ec->details->info);
} else {
audit_reject(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("command rejected by I/O plugin"),
iob->ec->details->info);
}
ret = false;
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return_bool(ret);
}
/* Call I/O plugin tty output log method. */
static bool
log_ttyout(const char *buf, unsigned int n, struct io_buffer *iob)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
bool ret = true;
debug_decl(log_ttyout, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->log_ttyout) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->log_ttyout(buf, n, &errstr);
if (rc <= 0) {
if (rc < 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->log_ttyout = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("I/O plugin error"),
iob->ec->details->info);
} else {
audit_reject(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("command rejected by I/O plugin"),
iob->ec->details->info);
}
ret = false;
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
if (!ret) {
/*
* I/O plugin rejected the output, delete the write event
* (user's tty) so we do not display the rejected output.
*/
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: deleting and freeing devtty wevent %p", __func__, iob->wevent);
sudo_ev_free(iob->wevent);
iob->wevent = NULL;
iob->off = iob->len = 0;
}
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return_bool(ret);
}
/* Call I/O plugin stdout log method. */
static bool
log_stdout(const char *buf, unsigned int n, struct io_buffer *iob)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
bool ret = true;
debug_decl(log_stdout, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->log_stdout) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->log_stdout(buf, n, &errstr);
if (rc <= 0) {
if (rc < 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->log_stdout = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("I/O plugin error"),
iob->ec->details->info);
} else {
audit_reject(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("command rejected by I/O plugin"),
iob->ec->details->info);
}
ret = false;
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
if (!ret) {
/*
* I/O plugin rejected the output, delete the write event
* (user's stdout) so we do not display the rejected output.
*/
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: deleting and freeing stdout wevent %p", __func__, iob->wevent);
sudo_ev_free(iob->wevent);
iob->wevent = NULL;
iob->off = iob->len = 0;
}
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return_bool(ret);
}
/* Call I/O plugin stderr log method. */
static bool
log_stderr(const char *buf, unsigned int n, struct io_buffer *iob)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
bool ret = true;
debug_decl(log_stderr, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->log_stderr) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->log_stderr(buf, n, &errstr);
if (rc <= 0) {
if (rc < 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->log_stderr = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("I/O plugin error"),
iob->ec->details->info);
} else {
audit_reject(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("command rejected by I/O plugin"),
iob->ec->details->info);
}
ret = false;
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
if (!ret) {
/*
* I/O plugin rejected the output, delete the write event
* (user's stderr) so we do not display the rejected output.
*/
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: deleting and freeing stderr wevent %p", __func__, iob->wevent);
sudo_ev_free(iob->wevent);
iob->wevent = NULL;
iob->off = iob->len = 0;
}
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return_bool(ret);
}
/* Call I/O plugin suspend log method. */
static void
log_suspend(struct exec_closure_pty *ec, int signo)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
debug_decl(log_suspend, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->version < SUDO_API_MKVERSION(1, 13))
continue;
if (plugin->u.io->log_suspend) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->log_suspend(signo, &errstr);
if (rc <= 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->log_suspend = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("error logging suspend"),
ec->details->info);
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return;
}
/* Call I/O plugin window change log method. */
static void
log_winchange(struct exec_closure_pty *ec, unsigned int rows, unsigned int cols)
{
struct plugin_container *plugin;
const char *errstr = NULL;
sigset_t omask;
debug_decl(log_winchange, SUDO_DEBUG_EXEC);
sigprocmask(SIG_BLOCK, &ttyblock, &omask);
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->version < SUDO_API_MKVERSION(1, 12))
continue;
if (plugin->u.io->change_winsize) {
int rc;
sudo_debug_set_active_instance(plugin->debug_instance);
rc = plugin->u.io->change_winsize(rows, cols, &errstr);
if (rc <= 0) {
/* Error: disable plugin's I/O function. */
plugin->u.io->change_winsize = NULL;
audit_error(plugin->name, SUDO_IO_PLUGIN,
errstr ? errstr : _("error changing window size"),
ec->details->info);
break;
}
}
}
sudo_debug_set_active_instance(sudo_debug_instance);
sigprocmask(SIG_SETMASK, &omask, NULL);
debug_return;
}
/*
* Check whether we are running in the foregroup.
* Updates the foreground global and updates the window size.
* Returns 0 if there is no tty, the foreground process group ID
* on success, or -1 on failure (tty revoked).
*/
static pid_t
check_foreground(struct exec_closure_pty *ec)
{
int ret = 0;
debug_decl(check_foreground, SUDO_DEBUG_EXEC);
if (io_fds[SFD_USERTTY] != -1) {
if ((ret = tcgetpgrp(io_fds[SFD_USERTTY])) != -1) {
foreground = ret == ec->ppgrp;
/* Also check for window size changes. */
sync_ttysize(ec);
}
}
debug_return_int(ret);
}
/*
* Suspend sudo if the underlying command is suspended.
* Returns SIGCONT_FG if the command should be resumed in the
* foreground or SIGCONT_BG if it is a background process.
*/
static int
suspend_sudo(struct exec_closure_pty *ec, int signo)
{
char signame[SIG2STR_MAX];
struct sigaction sa, osa;
int ret = 0;
debug_decl(suspend_sudo, SUDO_DEBUG_EXEC);
switch (signo) {
case SIGTTOU:
case SIGTTIN:
/*
* If sudo is already the foreground process, just resume the command
* in the foreground. If not, we'll suspend sudo and resume later.
*/
if (!foreground) {
if (check_foreground(ec) == -1) {
/* User's tty was revoked. */
break;
}
}
if (foreground) {
if (ttymode != TERM_RAW) {
if (sudo_term_raw(io_fds[SFD_USERTTY], 0))
ttymode = TERM_RAW;
}
ret = SIGCONT_FG; /* resume command in foreground */
break;
}
FALLTHROUGH;
case SIGSTOP:
case SIGTSTP:
/* Flush any remaining output and deschedule I/O events. */
del_io_events(true);
/* Restore original tty mode before suspending. */
if (ttymode != TERM_COOKED)
sudo_term_restore(io_fds[SFD_USERTTY], false);
/* Log the suspend event. */
log_suspend(ec, signo);
if (sig2str(signo, signame) == -1)
(void)snprintf(signame, sizeof(signame), "%d", signo);
/* Suspend self and continue command when we resume. */
if (signo != SIGSTOP) {
memset(&sa, 0, sizeof(sa));
sigemptyset(&sa.sa_mask);
sa.sa_flags = SA_RESTART;
sa.sa_handler = SIG_DFL;
if (sudo_sigaction(signo, &sa, &osa) != 0)
sudo_warn(U_("unable to set handler for signal %d"), signo);
}
sudo_debug_printf(SUDO_DEBUG_INFO, "kill parent SIG%s", signame);
if (killpg(ec->ppgrp, signo) != 0)
sudo_warn("killpg(%d, SIG%s)", (int)ec->ppgrp, signame);
/* Log the resume event. */
log_suspend(ec, SIGCONT);
/* Check foreground/background status on resume. */
if (check_foreground(ec) == -1) {
/* User's tty was revoked. */
break;
}
/*
* We always resume the command in the foreground if sudo itself
* is the foreground process. This helps work around poorly behaved
* programs that catch SIGTTOU/SIGTTIN but suspend themselves with
* SIGSTOP. At worst, sudo will go into the background but upon
* resume the command will be runnable. Otherwise, we can get into
* a situation where the command will immediately suspend itself.
*/
sudo_debug_printf(SUDO_DEBUG_INFO, "parent is in %s, ttymode %d -> %d",
foreground ? "foreground" : "background", ttymode,
foreground ? TERM_RAW : TERM_COOKED);
if (foreground) {
/* Foreground process, set tty to raw mode. */
if (sudo_term_raw(io_fds[SFD_USERTTY], 0))
ttymode = TERM_RAW;
} else {
/* Background process, no access to tty. */
ttymode = TERM_COOKED;
}
if (signo != SIGSTOP) {
if (sudo_sigaction(signo, &osa, NULL) != 0)
sudo_warn(U_("unable to restore handler for signal %d"), signo);
}
ret = ttymode == TERM_RAW ? SIGCONT_FG : SIGCONT_BG;
break;
}
debug_return_int(ret);
}
/*
* SIGTTIN signal handler for read_callback that just sets a flag.
*/
static volatile sig_atomic_t got_sigttin;
static void
sigttin(int signo)
{
got_sigttin = 1;
}
/*
* Read an iobuf that is ready.
*/
static void
read_callback(int fd, int what, void *v)
{
struct io_buffer *iob = v;
struct sudo_event_base *evbase = sudo_ev_get_base(iob->revent);
struct sigaction sa, osa;
int saved_errno;
ssize_t n;
debug_decl(read_callback, SUDO_DEBUG_EXEC);
/*
* We ignore SIGTTIN by default but we need to handle it when reading
* from the terminal. A signal event won't work here because the
* read() would be restarted, preventing the callback from running.
*/
memset(&sa, 0, sizeof(sa));
sigemptyset(&sa.sa_mask);
sa.sa_handler = sigttin;
got_sigttin = 0;
sigaction(SIGTTIN, &sa, &osa);
n = read(fd, iob->buf + iob->len, sizeof(iob->buf) - iob->len);
saved_errno = errno;
sigaction(SIGTTIN, &osa, NULL);
errno = saved_errno;
switch (n) {
case -1:
if (got_sigttin) {
/* Schedule SIGTTIN to be forwarded to the command. */
schedule_signal(iob->ec, SIGTTIN);
}
if (errno == EAGAIN || errno == EINTR)
break;
/* treat read error as fatal and close the fd */
sudo_debug_printf(SUDO_DEBUG_ERROR,
"error reading fd %d: %s", fd, strerror(errno));
FALLTHROUGH;
case 0:
/* got EOF or pty has gone away */
if (n == 0) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"read EOF from fd %d", fd);
}
safe_close(fd);
ev_free_by_fd(evbase, fd);
/* If writer already consumed the buffer, close it too. */
if (iob->wevent != NULL && iob->off == iob->len) {
safe_close(sudo_ev_get_fd(iob->wevent));
ev_free_by_fd(evbase, sudo_ev_get_fd(iob->wevent));
iob->off = iob->len = 0;
}
break;
default:
sudo_debug_printf(SUDO_DEBUG_INFO,
"read %zd bytes from fd %d", n, fd);
if (!iob->action(iob->buf + iob->len, n, iob)) {
terminate_command(iob->ec->cmnd_pid, true);
iob->ec->cmnd_pid = -1;
}
iob->len += n;
/* Enable writer now that there is data in the buffer. */
if (iob->wevent != NULL) {
if (sudo_ev_add(evbase, iob->wevent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
/* Re-enable reader if buffer is not full. */
if (iob->len != sizeof(iob->buf)) {
if (sudo_ev_add(evbase, iob->revent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
break;
}
}
/*
* SIGTTOU signal handler for write_callback that just sets a flag.
*/
static volatile sig_atomic_t got_sigttou;
static void
sigttou(int signo)
{
got_sigttou = 1;
}
/*
* Write an iobuf that is ready.
*/
static void
write_callback(int fd, int what, void *v)
{
struct io_buffer *iob = v;
struct sudo_event_base *evbase = sudo_ev_get_base(iob->wevent);
struct sigaction sa, osa;
int saved_errno;
ssize_t n;
debug_decl(write_callback, SUDO_DEBUG_EXEC);
/*
* We ignore SIGTTOU by default but we need to handle it when writing
* to the terminal. A signal event won't work here because the
* write() would be restarted, preventing the callback from running.
*/
memset(&sa, 0, sizeof(sa));
sigemptyset(&sa.sa_mask);
sa.sa_handler = sigttou;
got_sigttou = 0;
sigaction(SIGTTOU, &sa, &osa);
n = write(fd, iob->buf + iob->off, iob->len - iob->off);
saved_errno = errno;
sigaction(SIGTTOU, &osa, NULL);
errno = saved_errno;
if (n == -1) {
switch (errno) {
case EPIPE:
case ENXIO:
case EIO:
case EBADF:
/* other end of pipe closed or pty revoked */
sudo_debug_printf(SUDO_DEBUG_INFO,
"unable to write %d bytes to fd %d",
iob->len - iob->off, fd);
/* Close reader if there is one. */
if (iob->revent != NULL) {
safe_close(sudo_ev_get_fd(iob->revent));
ev_free_by_fd(evbase, sudo_ev_get_fd(iob->revent));
}
safe_close(fd);
ev_free_by_fd(evbase, fd);
break;
case EINTR:
if (got_sigttou) {
/* Schedule SIGTTOU to be forwarded to the command. */
schedule_signal(iob->ec, SIGTTOU);
}
FALLTHROUGH;
case EAGAIN:
/* not an error */
break;
default:
/* XXX - need a way to distinguish non-exec error. */
iob->ec->cstat->type = CMD_ERRNO;
iob->ec->cstat->val = errno;
sudo_debug_printf(SUDO_DEBUG_ERROR,
"error writing fd %d: %s", fd, strerror(errno));
sudo_ev_loopbreak(evbase);
break;
}
} else {
sudo_debug_printf(SUDO_DEBUG_INFO,
"wrote %zd bytes to fd %d", n, fd);
iob->off += n;
/* Reset buffer if fully consumed. */
if (iob->off == iob->len) {
iob->off = iob->len = 0;
/* Forward the EOF from reader to writer. */
if (iob->revent == NULL) {
safe_close(fd);
ev_free_by_fd(evbase, fd);
}
}
/* Re-enable writer if buffer is not empty. */
if (iob->len > iob->off) {
if (sudo_ev_add(evbase, iob->wevent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
/* Enable reader if buffer is not full. */
if (iob->revent != NULL &&
(ttymode == TERM_RAW || !USERTTY_EVENT(iob->revent))) {
if (iob->len != sizeof(iob->buf)) {
if (sudo_ev_add(evbase, iob->revent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
}
}
}
static void
io_buf_new(int rfd, int wfd,
bool (*action)(const char *, unsigned int, struct io_buffer *),
struct exec_closure_pty *ec, struct io_buffer_list *head)
{
int n;
struct io_buffer *iob;
debug_decl(io_buf_new, SUDO_DEBUG_EXEC);
/* Set non-blocking mode. */
n = fcntl(rfd, F_GETFL, 0);
if (n != -1 && !ISSET(n, O_NONBLOCK))
(void) fcntl(rfd, F_SETFL, n | O_NONBLOCK);
n = fcntl(wfd, F_GETFL, 0);
if (n != -1 && !ISSET(n, O_NONBLOCK))
(void) fcntl(wfd, F_SETFL, n | O_NONBLOCK);
/* Allocate and add to head of list. */
if ((iob = malloc(sizeof(*iob))) == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
iob->ec = ec;
iob->revent = sudo_ev_alloc(rfd, SUDO_EV_READ, read_callback, iob);
iob->wevent = sudo_ev_alloc(wfd, SUDO_EV_WRITE, write_callback, iob);
iob->len = 0;
iob->off = 0;
iob->action = action;
iob->buf[0] = '\0';
if (iob->revent == NULL || iob->wevent == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
SLIST_INSERT_HEAD(head, iob, entries);
debug_return;
}
/*
* We already closed the follower so reads from the leader will not block.
*/
static void
pty_finish(struct command_status *cstat)
{
struct io_buffer *iob;
int n;
debug_decl(pty_finish, SUDO_DEBUG_EXEC);
/* Flush any remaining output (the plugin already got it). */
if (io_fds[SFD_USERTTY] != -1) {
n = fcntl(io_fds[SFD_USERTTY], F_GETFL, 0);
if (n != -1 && ISSET(n, O_NONBLOCK)) {
CLR(n, O_NONBLOCK);
(void) fcntl(io_fds[SFD_USERTTY], F_SETFL, n);
}
}
del_io_events(false);
/* Free I/O buffers. */
while ((iob = SLIST_FIRST(&iobufs)) != NULL) {
SLIST_REMOVE_HEAD(&iobufs, entries);
if (iob->revent != NULL)
sudo_ev_free(iob->revent);
if (iob->wevent != NULL)
sudo_ev_free(iob->wevent);
free(iob);
}
/* Restore terminal settings. */
if (io_fds[SFD_USERTTY] != -1)
sudo_term_restore(io_fds[SFD_USERTTY], false);
/* Update utmp */
if (utmp_user != NULL)
utmp_logout(ptyname, cstat->type == CMD_WSTATUS ? cstat->val : 0);
debug_return;
}
/*
* Send command status to the monitor (signal or window size change).
*/
static void
send_command_status(struct exec_closure_pty *ec, int type, int val)
{
struct monitor_message *msg;
debug_decl(send_command, SUDO_DEBUG_EXEC);
if ((msg = calloc(1, sizeof(*msg))) == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
msg->cstat.type = type;
msg->cstat.val = val;
TAILQ_INSERT_TAIL(&ec->monitor_messages, msg, entries);
if (sudo_ev_add(ec->evbase, ec->fwdchannel_event, NULL, true) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
/* Restart event loop to send the command immediately. */
sudo_ev_loopcontinue(ec->evbase);
debug_return;
}
/*
* Schedule a signal to be forwarded.
*/
static void
schedule_signal(struct exec_closure_pty *ec, int signo)
{
char signame[SIG2STR_MAX];
debug_decl(schedule_signal, SUDO_DEBUG_EXEC);
if (signo == 0)
debug_return;
if (signo == SIGCONT_FG)
strlcpy(signame, "CONT_FG", sizeof(signame));
else if (signo == SIGCONT_BG)
strlcpy(signame, "CONT_BG", sizeof(signame));
else if (sig2str(signo, signame) == -1)
(void)snprintf(signame, sizeof(signame), "%d", signo);
sudo_debug_printf(SUDO_DEBUG_DIAG, "scheduled SIG%s for command", signame);
send_command_status(ec, CMD_SIGNO, signo);
debug_return;
}
static void
backchannel_cb(int fd, int what, void *v)
{
struct exec_closure_pty *ec = v;
struct command_status cstat;
ssize_t nread;
debug_decl(backchannel_cb, SUDO_DEBUG_EXEC);
/*
* Read command status from the monitor.
* Note that the backchannel is a *blocking* socket.
*/
nread = recv(fd, &cstat, sizeof(cstat), MSG_WAITALL);
switch (nread) {
case -1:
switch (errno) {
case EINTR:
case EAGAIN:
/* Nothing ready. */
break;
default:
if (ec->cstat->val == CMD_INVALID) {
ec->cstat->type = CMD_ERRNO;
ec->cstat->val = errno;
sudo_debug_printf(SUDO_DEBUG_ERROR,
"%s: failed to read command status: %s",
__func__, strerror(errno));
sudo_ev_loopbreak(ec->evbase);
}
break;
}
break;
case 0:
/* EOF, monitor exited or was killed. */
sudo_debug_printf(SUDO_DEBUG_INFO,
"EOF on backchannel, monitor dead?");
if (ec->cstat->type == CMD_INVALID) {
/* XXX - need new CMD_ type for monitor errors. */
ec->cstat->type = CMD_ERRNO;
ec->cstat->val = ECONNRESET;
}
sudo_ev_loopexit(ec->evbase);
break;
case sizeof(cstat):
/* Check command status. */
switch (cstat.type) {
case CMD_PID:
ec->cmnd_pid = cstat.val;
sudo_debug_printf(SUDO_DEBUG_INFO, "executed %s, pid %d",
ec->details->command, (int)ec->cmnd_pid);
break;
case CMD_WSTATUS:
if (WIFSTOPPED(cstat.val)) {
int signo;
/* Suspend parent and tell monitor how to resume on return. */
sudo_debug_printf(SUDO_DEBUG_INFO,
"command stopped, suspending parent");
signo = suspend_sudo(ec, WSTOPSIG(cstat.val));
schedule_signal(ec, signo);
/* Re-enable I/O events */
add_io_events(ec->evbase);
} else {
/* Command exited or was killed, either way we are done. */
sudo_debug_printf(SUDO_DEBUG_INFO, "command exited or was killed");
sudo_ev_loopexit(ec->evbase);
*ec->cstat = cstat;
}
break;
case CMD_ERRNO:
/* Monitor was unable to execute command or broken pipe. */
sudo_debug_printf(SUDO_DEBUG_INFO, "errno from monitor: %s",
strerror(cstat.val));
sudo_ev_loopbreak(ec->evbase);
*ec->cstat = cstat;
break;
}
/* Keep reading command status messages until EAGAIN or EOF. */
break;
default:
/* Short read, should not happen. */
if (ec->cstat->val == CMD_INVALID) {
ec->cstat->type = CMD_ERRNO;
ec->cstat->val = EIO;
sudo_debug_printf(SUDO_DEBUG_ERROR,
"%s: failed to read command status: short read", __func__);
sudo_ev_loopbreak(ec->evbase);
}
break;
}
debug_return;
}
/*
* Handle changes to the monitors's status (SIGCHLD).
*/
static void
handle_sigchld_pty(struct exec_closure_pty *ec)
{
int n, status;
pid_t pid;
debug_decl(handle_sigchld_pty, SUDO_DEBUG_EXEC);
/*
* Monitor process was signaled; wait for it as needed.
*/
do {
pid = waitpid(ec->monitor_pid, &status, WUNTRACED|WNOHANG);
} while (pid == -1 && errno == EINTR);
switch (pid) {
case 0:
errno = ECHILD;
FALLTHROUGH;
case -1:
sudo_warn(U_("%s: %s"), __func__, "waitpid");
debug_return;
}
/*
* If the monitor dies we get notified via backchannel_cb().
* If it was stopped, we should stop too (the command keeps
* running in its pty) and continue it when we come back.
*/
if (WIFSTOPPED(status)) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"monitor stopped, suspending sudo");
n = suspend_sudo(ec, WSTOPSIG(status));
kill(pid, SIGCONT);
schedule_signal(ec, n);
/* Re-enable I/O events */
add_io_events(ec->evbase);
} else if (WIFSIGNALED(status)) {
char signame[SIG2STR_MAX];
if (sig2str(WTERMSIG(status), signame) == -1)
(void)snprintf(signame, sizeof(signame), "%d", WTERMSIG(status));
sudo_debug_printf(SUDO_DEBUG_INFO, "%s: monitor (%d) killed, SIG%s",
__func__, (int)ec->monitor_pid, signame);
ec->monitor_pid = -1;
} else {
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: monitor exited, status %d", __func__, WEXITSTATUS(status));
ec->monitor_pid = -1;
}
debug_return;
}
/* Signal callback */
static void
signal_cb_pty(int signo, int what, void *v)
{
struct sudo_ev_siginfo_container *sc = v;
struct exec_closure_pty *ec = sc->closure;
char signame[SIG2STR_MAX];
debug_decl(signal_cb_pty, SUDO_DEBUG_EXEC);
if (ec->monitor_pid == -1)
debug_return;
if (sig2str(signo, signame) == -1)
(void)snprintf(signame, sizeof(signame), "%d", signo);
sudo_debug_printf(SUDO_DEBUG_DIAG,
"%s: evbase %p, monitor: %d, signo %s(%d), cstat %p", __func__,
ec->evbase, (int)ec->monitor_pid, signame, signo, ec->cstat);
switch (signo) {
case SIGCHLD:
handle_sigchld_pty(ec);
break;
case SIGWINCH:
sync_ttysize(ec);
break;
default:
/*
* Do not forward signals sent by a process in the command's process
* group, as we don't want the command to indirectly kill itself.
* For example, this can happen with some versions of reboot that
* call kill(-1, SIGTERM) to kill all other processes.
*/
if (USER_SIGNALED(sc->siginfo) && sc->siginfo->si_pid != 0) {
pid_t si_pgrp = getpgid(sc->siginfo->si_pid);
if (si_pgrp != -1) {
if (si_pgrp == ec->ppgrp || si_pgrp == ec->cmnd_pid)
debug_return;
} else if (sc->siginfo->si_pid == ec->cmnd_pid) {
debug_return;
}
}
/* Schedule signal to be forwarded to the command. */
schedule_signal(ec, signo);
break;
}
debug_return;
}
/*
* Forward signals in monitor_messages to the monitor so it can
* deliver them to the command.
*/
static void
fwdchannel_cb(int sock, int what, void *v)
{
struct exec_closure_pty *ec = v;
char signame[SIG2STR_MAX];
struct monitor_message *msg;
ssize_t nsent;
debug_decl(fwdchannel_cb, SUDO_DEBUG_EXEC);
while ((msg = TAILQ_FIRST(&ec->monitor_messages)) != NULL) {
switch (msg->cstat.type) {
case CMD_SIGNO:
if (msg->cstat.val == SIGCONT_FG)
strlcpy(signame, "CONT_FG", sizeof(signame));
else if (msg->cstat.val == SIGCONT_BG)
strlcpy(signame, "CONT_BG", sizeof(signame));
else if (sig2str(msg->cstat.val, signame) == -1)
(void)snprintf(signame, sizeof(signame), "%d", msg->cstat.val);
sudo_debug_printf(SUDO_DEBUG_INFO,
"sending SIG%s to monitor over backchannel", signame);
break;
case CMD_TTYWINCH:
sudo_debug_printf(SUDO_DEBUG_INFO, "sending window size change "
"to monitor over backchannelL %d x %d",
msg->cstat.val & 0xffff, (msg->cstat.val >> 16) & 0xffff);
break;
default:
sudo_debug_printf(SUDO_DEBUG_INFO,
"sending cstat type %d, value %d to monitor over backchannel",
msg->cstat.type, msg->cstat.val);
break;
}
TAILQ_REMOVE(&ec->monitor_messages, msg, entries);
nsent = send(sock, &msg->cstat, sizeof(msg->cstat), 0);
if (nsent != sizeof(msg->cstat)) {
if (errno == EPIPE) {
sudo_debug_printf(SUDO_DEBUG_ERROR,
"broken pipe writing to monitor over backchannel");
/* Other end of socket gone, empty out monitor_messages. */
free(msg);
while ((msg = TAILQ_FIRST(&ec->monitor_messages)) != NULL) {
TAILQ_REMOVE(&ec->monitor_messages, msg, entries);
free(msg);
}
/* XXX - need new CMD_ type for monitor errors. */
ec->cstat->type = CMD_ERRNO;
ec->cstat->val = errno;
sudo_ev_loopbreak(ec->evbase);
}
break;
}
free(msg);
}
}
/*
* Fill in the exec closure and setup initial exec events.
* Allocates events for the signal pipe and backchannel.
* Forwarded signals on the backchannel are enabled on demand.
*/
static void
fill_exec_closure_pty(struct exec_closure_pty *ec, struct command_status *cstat,
struct command_details *details, pid_t ppgrp, int backchannel)
{
debug_decl(fill_exec_closure_pty, SUDO_DEBUG_EXEC);
/* Fill in the non-event part of the closure. */
ec->cmnd_pid = -1;
ec->ppgrp = ppgrp;
ec->cstat = cstat;
ec->details = details;
ec->rows = user_details.ts_rows;
ec->cols = user_details.ts_cols;
TAILQ_INIT(&ec->monitor_messages);
/* Reset cstat for running the command. */
cstat->type = CMD_INVALID;
cstat->val = 0;
/* Setup event base and events. */
ec->evbase = details->evbase;
details->evbase = NULL;
/* Event for command status via backchannel. */
ec->backchannel_event = sudo_ev_alloc(backchannel,
SUDO_EV_READ|SUDO_EV_PERSIST, backchannel_cb, ec);
if (ec->backchannel_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->backchannel_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
sudo_debug_printf(SUDO_DEBUG_INFO, "backchannel fd %d\n", backchannel);
/* Events for local signals. */
ec->sigint_event = sudo_ev_alloc(SIGINT,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigint_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigint_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigquit_event = sudo_ev_alloc(SIGQUIT,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigquit_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigquit_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigtstp_event = sudo_ev_alloc(SIGTSTP,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigtstp_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigtstp_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigterm_event = sudo_ev_alloc(SIGTERM,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigterm_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigterm_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sighup_event = sudo_ev_alloc(SIGHUP,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sighup_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sighup_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigalrm_event = sudo_ev_alloc(SIGALRM,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigalrm_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigalrm_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigusr1_event = sudo_ev_alloc(SIGUSR1,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigusr1_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigusr1_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigusr2_event = sudo_ev_alloc(SIGUSR2,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigusr2_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigusr2_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigchld_event = sudo_ev_alloc(SIGCHLD,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigchld_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigchld_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
ec->sigwinch_event = sudo_ev_alloc(SIGWINCH,
SUDO_EV_SIGINFO, signal_cb_pty, ec);
if (ec->sigwinch_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
if (sudo_ev_add(ec->evbase, ec->sigwinch_event, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
/* The signal forwarding event gets added on demand. */
ec->fwdchannel_event = sudo_ev_alloc(backchannel,
SUDO_EV_WRITE, fwdchannel_cb, ec);
if (ec->fwdchannel_event == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
/* Set the default event base. */
sudo_ev_base_setdef(ec->evbase);
debug_return;
}
/*
* Free the dynamically-allocated contents of the exec closure.
*/
static void
free_exec_closure_pty(struct exec_closure_pty *ec)
{
struct monitor_message *msg;
debug_decl(free_exec_closure_pty, SUDO_DEBUG_EXEC);
/* Free any remaining intercept resources. */
intercept_cleanup();
sudo_ev_base_free(ec->evbase);
sudo_ev_free(ec->backchannel_event);
sudo_ev_free(ec->fwdchannel_event);
sudo_ev_free(ec->sigint_event);
sudo_ev_free(ec->sigquit_event);
sudo_ev_free(ec->sigtstp_event);
sudo_ev_free(ec->sigterm_event);
sudo_ev_free(ec->sighup_event);
sudo_ev_free(ec->sigalrm_event);
sudo_ev_free(ec->sigusr1_event);
sudo_ev_free(ec->sigusr2_event);
sudo_ev_free(ec->sigchld_event);
sudo_ev_free(ec->sigwinch_event);
while ((msg = TAILQ_FIRST(&ec->monitor_messages)) != NULL) {
TAILQ_REMOVE(&ec->monitor_messages, msg, entries);
free(msg);
}
debug_return;
}
/*
* Execute a command in a pty, potentially with I/O logging, and
* wait for it to finish.
* This is a little bit tricky due to how POSIX job control works and
* we fact that we have two different controlling terminals to deal with.
*/
bool
exec_pty(struct command_details *details, struct command_status *cstat)
{
int io_pipe[3][2] = { { -1, -1 }, { -1, -1 }, { -1, -1 } };
bool interpose[3] = { false, false, false };
int sv[2], intercept_sv[2] = { -1, -1 };
struct exec_closure_pty ec = { 0 };
struct plugin_container *plugin;
int evloop_retries = -1;
sigset_t set, oset;
struct sigaction sa;
struct stat sb;
pid_t ppgrp;
debug_decl(exec_pty, SUDO_DEBUG_EXEC);
/*
* Allocate a pty.
*/
if (!pty_setup(details, user_details.tty)) {
if (TAILQ_EMPTY(&io_plugins)) {
/* Not logging I/O and didn't allocate a pty. */
debug_return_bool(false);
}
}
/*
* We communicate with the monitor over a bi-directional pair of sockets.
* Parent sends signal info to monitor and monitor sends back wait status.
*/
if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1 ||
fcntl(sv[0], F_SETFD, FD_CLOEXEC) == -1 ||
fcntl(sv[1], F_SETFD, FD_CLOEXEC) == -1)
sudo_fatal("%s", U_("unable to create sockets"));
/*
* Allocate a socketpair for communicating with sudo_intercept.so.
* This must be inherited across exec, hence no FD_CLOEXEC.
*/
if (ISSET(details->flags, CD_INTERCEPT|CD_LOG_SUBCMDS)) {
if (socketpair(PF_UNIX, SOCK_STREAM, 0, intercept_sv) == -1)
sudo_fatal("%s", U_("unable to create sockets"));
}
/*
* We don't want to receive SIGTTIN/SIGTTOU.
* XXX - this affects tcsetattr() and tcsetpgrp() too.
*/
memset(&sa, 0, sizeof(sa));
sigemptyset(&sa.sa_mask);
sa.sa_flags = SA_RESTART;
sa.sa_handler = SIG_IGN;
if (sudo_sigaction(SIGTTIN, &sa, NULL) != 0)
sudo_warn(U_("unable to set handler for signal %d"), SIGTTIN);
if (sudo_sigaction(SIGTTOU, &sa, NULL) != 0)
sudo_warn(U_("unable to set handler for signal %d"), SIGTTOU);
/*
* The policy plugin's session init must be run before we fork
* or certain pam modules won't be able to track their state.
*/
if (policy_init_session(details) != true)
sudo_fatalx("%s", U_("policy plugin failed session initialization"));
/*
* Child will run the command in the pty, parent will pass data
* to and from pty.
*/
/* So we can block tty-generated signals */
sigemptyset(&ttyblock);
sigaddset(&ttyblock, SIGINT);
sigaddset(&ttyblock, SIGQUIT);
sigaddset(&ttyblock, SIGTSTP);
sigaddset(&ttyblock, SIGTTIN);
sigaddset(&ttyblock, SIGTTOU);
ppgrp = getpgrp(); /* parent's pgrp, so child can signal us */
/* Determine whether any of std{in,out,err} should be logged. */
TAILQ_FOREACH(plugin, &io_plugins, entries) {
if (plugin->u.io->log_stdin)
interpose[STDIN_FILENO] = true;
if (plugin->u.io->log_stdout)
interpose[STDOUT_FILENO] = true;
if (plugin->u.io->log_stderr)
interpose[STDERR_FILENO] = true;
}
/*
* Setup stdin/stdout/stderr for command, to be duped after forking.
* In background mode there is no stdin.
*/
if (!ISSET(details->flags, CD_BACKGROUND))
io_fds[SFD_STDIN] = io_fds[SFD_FOLLOWER];
io_fds[SFD_STDOUT] = io_fds[SFD_FOLLOWER];
io_fds[SFD_STDERR] = io_fds[SFD_FOLLOWER];
if (io_fds[SFD_USERTTY] != -1) {
/* Read from /dev/tty, write to pty leader */
if (!ISSET(details->flags, CD_BACKGROUND)) {
io_buf_new(io_fds[SFD_USERTTY], io_fds[SFD_LEADER],
log_ttyin, &ec, &iobufs);
}
/* Read from pty leader, write to /dev/tty */
io_buf_new(io_fds[SFD_LEADER], io_fds[SFD_USERTTY],
log_ttyout, &ec, &iobufs);
/* Are we the foreground process? */
foreground = tcgetpgrp(io_fds[SFD_USERTTY]) == ppgrp;
}
/*
* If stdin, stdout or stderr is not a tty and logging is enabled,
* use a pipe to interpose ourselves instead of using the pty fd.
*/
if (io_fds[SFD_STDIN] == -1 || !isatty(STDIN_FILENO)) {
if (!interpose[STDIN_FILENO]) {
/* Not logging stdin, do not interpose. */
sudo_debug_printf(SUDO_DEBUG_INFO,
"stdin not a tty, not logging");
if (fstat(STDIN_FILENO, &sb) == 0 && S_ISFIFO(sb.st_mode))
pipeline = true;
io_fds[SFD_STDIN] = dup(STDIN_FILENO);
if (io_fds[SFD_STDIN] == -1)
sudo_fatal("dup");
} else {
sudo_debug_printf(SUDO_DEBUG_INFO,
"stdin not a tty, creating a pipe");
pipeline = true;
if (pipe2(io_pipe[STDIN_FILENO], O_CLOEXEC) != 0)
sudo_fatal("%s", U_("unable to create pipe"));
io_buf_new(STDIN_FILENO, io_pipe[STDIN_FILENO][1],
log_stdin, &ec, &iobufs);
io_fds[SFD_STDIN] = io_pipe[STDIN_FILENO][0];
}
}
if (io_fds[SFD_STDOUT] == -1 || !isatty(STDOUT_FILENO)) {
if (!interpose[STDOUT_FILENO]) {
/* Not logging stdout, do not interpose. */
sudo_debug_printf(SUDO_DEBUG_INFO,
"stdout not a tty, not logging");
if (fstat(STDOUT_FILENO, &sb) == 0 && S_ISFIFO(sb.st_mode))
pipeline = true;
io_fds[SFD_STDOUT] = dup(STDOUT_FILENO);
if (io_fds[SFD_STDOUT] == -1)
sudo_fatal("dup");
} else {
sudo_debug_printf(SUDO_DEBUG_INFO,
"stdout not a tty, creating a pipe");
pipeline = true;
if (pipe2(io_pipe[STDOUT_FILENO], O_CLOEXEC) != 0)
sudo_fatal("%s", U_("unable to create pipe"));
io_buf_new(io_pipe[STDOUT_FILENO][0], STDOUT_FILENO,
log_stdout, &ec, &iobufs);
io_fds[SFD_STDOUT] = io_pipe[STDOUT_FILENO][1];
}
}
if (io_fds[SFD_STDERR] == -1 || !isatty(STDERR_FILENO)) {
if (!interpose[STDERR_FILENO]) {
/* Not logging stderr, do not interpose. */
sudo_debug_printf(SUDO_DEBUG_INFO,
"stderr not a tty, not logging");
if (fstat(STDERR_FILENO, &sb) == 0 && S_ISFIFO(sb.st_mode))
pipeline = true;
io_fds[SFD_STDERR] = dup(STDERR_FILENO);
if (io_fds[SFD_STDERR] == -1)
sudo_fatal("dup");
} else {
sudo_debug_printf(SUDO_DEBUG_INFO,
"stderr not a tty, creating a pipe");
if (pipe2(io_pipe[STDERR_FILENO], O_CLOEXEC) != 0)
sudo_fatal("%s", U_("unable to create pipe"));
io_buf_new(io_pipe[STDERR_FILENO][0], STDERR_FILENO,
log_stderr, &ec, &iobufs);
io_fds[SFD_STDERR] = io_pipe[STDERR_FILENO][1];
}
}
if (foreground) {
/* Copy terminal attrs from user tty -> pty follower. */
if (!sudo_term_copy(io_fds[SFD_USERTTY], io_fds[SFD_FOLLOWER])) {
sudo_debug_printf(SUDO_DEBUG_ERROR|SUDO_DEBUG_ERRNO,
"%s: unable to copy terminal settings to pty", __func__);
foreground = false;
} else {
/* Start in raw mode unless part of a pipeline or backgrounded. */
if (!pipeline && !ISSET(details->flags, CD_EXEC_BG)) {
if (sudo_term_raw(io_fds[SFD_USERTTY], 0))
ttymode = TERM_RAW;
}
}
}
/*
* Block signals until we have our handlers setup in the parent so
* we don't miss SIGCHLD if the command exits immediately.
*/
sigfillset(&set);
sigprocmask(SIG_BLOCK, &set, &oset);
/* Check for early termination or suspend signals before we fork. */
if (sudo_terminated(cstat)) {
sigprocmask(SIG_SETMASK, &oset, NULL);
debug_return_int(true);
}
ec.monitor_pid = sudo_debug_fork();
switch (ec.monitor_pid) {
case -1:
sudo_fatal("%s", U_("unable to fork"));
break;
case 0:
/* child */
close(sv[0]);
if (intercept_sv[0] != -1)
close(intercept_sv[0]);
/* Close the other end of the stdin/stdout/stderr pipes and exec. */
if (io_pipe[STDIN_FILENO][1] != -1)
close(io_pipe[STDIN_FILENO][1]);
if (io_pipe[STDOUT_FILENO][0] != -1)
close(io_pipe[STDOUT_FILENO][0]);
if (io_pipe[STDERR_FILENO][0] != -1)
close(io_pipe[STDERR_FILENO][0]);
/*
* If stdin/stdout is not a tty, start command in the background
* since it might be part of a pipeline that reads from /dev/tty.
* In this case, we rely on the command receiving SIGTTOU or SIGTTIN
* when it needs access to the controlling tty.
*/
exec_monitor(details, &oset, foreground && !pipeline, sv[1],
intercept_sv[1]);
cstat->type = CMD_ERRNO;
cstat->val = errno;
if (send(sv[1], cstat, sizeof(*cstat), 0) == -1) {
sudo_debug_printf(SUDO_DEBUG_ERROR|SUDO_DEBUG_ERRNO,
"%s: unable to send status to parent", __func__);
}
_exit(EXIT_FAILURE);
}
/*
* We close the pty follower so only the monitor and command have a
* reference to it. This ensures that we can don't block reading
* from the leader when the command and monitor have exited.
*/
if (io_fds[SFD_FOLLOWER] != -1) {
close(io_fds[SFD_FOLLOWER]);
io_fds[SFD_FOLLOWER] = -1;
}
/* Tell the monitor to continue now that the follower is closed. */
cstat->type = CMD_SIGNO;
cstat->val = 0;
while (send(sv[0], cstat, sizeof(*cstat), 0) == -1) {
if (errno != EINTR && errno != EAGAIN)
sudo_fatal("%s", U_("unable to send message to monitor process"));
}
/* Close the other end of the stdin/stdout/stderr pipes and socketpair. */
if (io_pipe[STDIN_FILENO][0] != -1)
close(io_pipe[STDIN_FILENO][0]);
if (io_pipe[STDOUT_FILENO][1] != -1)
close(io_pipe[STDOUT_FILENO][1]);
if (io_pipe[STDERR_FILENO][1] != -1)
close(io_pipe[STDERR_FILENO][1]);
close(sv[1]);
/* No longer need execfd. */
if (details->execfd != -1) {
close(details->execfd);
details->execfd = -1;
}
/* Set command timeout if specified. */
if (ISSET(details->flags, CD_SET_TIMEOUT))
alarm(details->timeout);
/*
* Fill in exec closure, allocate event base, signal events and
* the backchannel event.
*/
fill_exec_closure_pty(&ec, cstat, details, ppgrp, sv[0]);
/* Create event and closure for intercept mode. */
if (intercept_sv[0] != -1) {
if (!intercept_setup(intercept_sv[0], ec.evbase, details))
exit(EXIT_FAILURE);
}
/* Restore signal mask now that signal handlers are setup. */
sigprocmask(SIG_SETMASK, &oset, NULL);
/*
* I/O logging must be in the C locale for floating point numbers
* to be logged consistently.
*/
setlocale(LC_ALL, "C");
/*
* In the event loop we pass input from user tty to leader
* and pass output from leader to stdout and IO plugin.
* Try to recover on ENXIO, it means the tty was revoked.
*/
add_io_events(ec.evbase);
do {
if (sudo_ev_dispatch(ec.evbase) == -1)
sudo_warn("%s", U_("error in event loop"));
if (sudo_ev_got_break(ec.evbase)) {
/* error from callback or monitor died */
sudo_debug_printf(SUDO_DEBUG_ERROR, "event loop exited prematurely");
/* XXX: no good way to know if we should terminate the command. */
if (cstat->val == CMD_INVALID && ec.cmnd_pid != -1) {
/* no status message, kill command */
terminate_command(ec.cmnd_pid, true);
ec.cmnd_pid = -1;
/* TODO: need way to pass an error to the sudo front end */
cstat->type = CMD_WSTATUS;
cstat->val = W_EXITCODE(1, SIGKILL);
}
} else if (!sudo_ev_got_exit(ec.evbase)) {
switch (errno) {
case ENXIO:
case EIO:
case EBADF:
/* /dev/tty was revoked, remove tty events and retry (once) */
if (evloop_retries == -1 && io_fds[SFD_USERTTY] != -1) {
ev_free_by_fd(ec.evbase, io_fds[SFD_USERTTY]);
evloop_retries = 1;
}
break;
}
}
} while (evloop_retries-- > 0);
/* Flush any remaining output, free I/O bufs and events, do logout. */
pty_finish(cstat);
/* Free things up. */
free_exec_closure_pty(&ec);
debug_return_bool(true);
}
/*
* Schedule I/O events before starting the main event loop or
* resuming from suspend.
*/
static void
add_io_events(struct sudo_event_base *evbase)
{
struct io_buffer *iob;
debug_decl(add_io_events, SUDO_DEBUG_EXEC);
/*
* Schedule all readers as long as the buffer is not full.
* Schedule writers that contain buffered data.
* Normally, write buffers are added on demand when data is read.
*/
SLIST_FOREACH(iob, &iobufs, entries) {
/* Don't read from /dev/tty if we are not in the foreground. */
if (iob->revent != NULL &&
(ttymode == TERM_RAW || !USERTTY_EVENT(iob->revent))) {
if (iob->len != sizeof(iob->buf)) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"added I/O revent %p, fd %d, events %d",
iob->revent, iob->revent->fd, iob->revent->events);
if (sudo_ev_add(evbase, iob->revent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
}
if (iob->wevent != NULL) {
/* Enable writer if buffer is not empty. */
if (iob->len > iob->off) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"added I/O wevent %p, fd %d, events %d",
iob->wevent, iob->wevent->fd, iob->wevent->events);
if (sudo_ev_add(evbase, iob->wevent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
}
}
debug_return;
}
/*
* Flush any output buffered in iobufs or readable from fds other
* than /dev/tty. Removes I/O events from the event base when done.
*/
static void
del_io_events(bool nonblocking)
{
struct io_buffer *iob;
struct sudo_event_base *evbase;
debug_decl(del_io_events, SUDO_DEBUG_EXEC);
/* Remove iobufs from existing event base. */
SLIST_FOREACH(iob, &iobufs, entries) {
if (iob->revent != NULL) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"deleted I/O revent %p, fd %d, events %d",
iob->revent, iob->revent->fd, iob->revent->events);
sudo_ev_del(NULL, iob->revent);
}
if (iob->wevent != NULL) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"deleted I/O wevent %p, fd %d, events %d",
iob->wevent, iob->wevent->fd, iob->wevent->events);
sudo_ev_del(NULL, iob->wevent);
}
}
/* Create temporary event base for flushing. */
evbase = sudo_ev_base_alloc();
if (evbase == NULL)
sudo_fatalx(U_("%s: %s"), __func__, U_("unable to allocate memory"));
/* Avoid reading from /dev/tty, just flush existing data. */
SLIST_FOREACH(iob, &iobufs, entries) {
/* Don't read from /dev/tty while flushing. */
if (iob->revent != NULL && !USERTTY_EVENT(iob->revent)) {
if (iob->len != sizeof(iob->buf)) {
if (sudo_ev_add(evbase, iob->revent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
}
/* Flush any write buffers with data in them. */
if (iob->wevent != NULL) {
if (iob->len > iob->off) {
if (sudo_ev_add(evbase, iob->wevent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
}
}
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: flushing remaining I/O buffers (nonblocking)", __func__);
(void) sudo_ev_loop(evbase, SUDO_EVLOOP_NONBLOCK);
/*
* If not in non-blocking mode, make sure we flush write buffers.
* We don't want to read from the pty or stdin since that might block
* and the command is no longer running anyway.
*/
if (!nonblocking) {
/* Clear out iobufs from event base. */
SLIST_FOREACH(iob, &iobufs, entries) {
if (iob->revent != NULL && !USERTTY_EVENT(iob->revent))
sudo_ev_del(evbase, iob->revent);
if (iob->wevent != NULL)
sudo_ev_del(evbase, iob->wevent);
}
SLIST_FOREACH(iob, &iobufs, entries) {
/* Flush any write buffers with data in them. */
if (iob->wevent != NULL) {
if (iob->len > iob->off) {
if (sudo_ev_add(evbase, iob->wevent, NULL, false) == -1)
sudo_fatal("%s", U_("unable to add event to queue"));
}
}
}
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: flushing remaining write buffers (blocking)", __func__);
(void) sudo_ev_dispatch(evbase);
/* We should now have flushed all write buffers. */
SLIST_FOREACH(iob, &iobufs, entries) {
if (iob->wevent != NULL) {
if (iob->len > iob->off) {
sudo_debug_printf(SUDO_DEBUG_ERROR,
"unflushed data: wevent %p, fd %d, events %d",
iob->wevent, iob->wevent->fd, iob->wevent->events);
}
}
}
}
/* Free temporary event base, removing its events. */
sudo_ev_base_free(evbase);
debug_return;
}
/*
* Check for tty size changes.
* Passes the new window size to the I/O plugin and to the monitor.
*/
static void
sync_ttysize(struct exec_closure_pty *ec)
{
struct winsize wsize;
debug_decl(sync_ttysize, SUDO_DEBUG_EXEC);
if (ioctl(io_fds[SFD_USERTTY], TIOCGWINSZ, &wsize) == 0) {
if (wsize.ws_row != ec->rows || wsize.ws_col != ec->cols) {
const unsigned int wsize_packed = (wsize.ws_row & 0xffff) |
((wsize.ws_col & 0xffff) << 16);
/* Log window change event. */
log_winchange(ec, wsize.ws_row, wsize.ws_col);
/* Send window change event to monitor process. */
send_command_status(ec, CMD_TTYWINCH, wsize_packed);
/* Update rows/cols. */
ec->rows = wsize.ws_row;
ec->cols = wsize.ws_col;
}
}
debug_return;
}
/*
* Remove and free any events associated with the specified
* file descriptor present in the I/O buffers list.
*/
static void
ev_free_by_fd(struct sudo_event_base *evbase, int fd)
{
struct io_buffer *iob;
debug_decl(ev_free_by_fd, SUDO_DEBUG_EXEC);
/* Deschedule any users of the fd and free up the events. */
SLIST_FOREACH(iob, &iobufs, entries) {
if (iob->revent != NULL) {
if (sudo_ev_get_fd(iob->revent) == fd) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: deleting and freeing revent %p with fd %d",
__func__, iob->revent, fd);
sudo_ev_free(iob->revent);
iob->revent = NULL;
}
}
if (iob->wevent != NULL) {
if (sudo_ev_get_fd(iob->wevent) == fd) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: deleting and freeing wevent %p with fd %d",
__func__, iob->wevent, fd);
sudo_ev_free(iob->wevent);
iob->wevent = NULL;
}
}
}
debug_return;
}
/*
* Only close the fd if it is not /dev/tty or std{in,out,err}.
* Return value is the same as close(2).
*/
static int
safe_close(int fd)
{
debug_decl(safe_close, SUDO_DEBUG_EXEC);
/* Avoid closing /dev/tty or std{in,out,err}. */
if (fd < 3 || fd == io_fds[SFD_USERTTY]) {
sudo_debug_printf(SUDO_DEBUG_INFO,
"%s: not closing fd %d (%s)", __func__, fd, _PATH_TTY);
errno = EINVAL;
debug_return_int(-1);
}
sudo_debug_printf(SUDO_DEBUG_INFO, "%s: closing fd %d", __func__, fd);
debug_return_int(close(fd));
}