1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2013 DEY Storage Systems, Inc.
24 * Copyright (c) 2014 Gary Mills
25 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
26 * Copyright 2019 Joyent, Inc.
27 * Copyright 2019 OmniOS Community Edition (OmniOSce) Association.
28 */
29
30 /*
31 * zlogin provides three types of login which allow users in the global
32 * zone to access non-global zones.
33 *
34 * - "interactive login" is similar to rlogin(1); for example, the user could
35 * issue 'zlogin my-zone' or 'zlogin -e ^ -l me my-zone'. The user is
36 * granted a new pty (which is then shoved into the zone), and an I/O
37 * loop between parent and child processes takes care of the interactive
38 * session. In this mode, login(1) (and its -c option, which means
39 * "already authenticated") is employed to take care of the initialization
40 * of the user's session.
41 *
42 * - "non-interactive login" is similar to su(1M); the user could issue
43 * 'zlogin my-zone ls -l' and the command would be run as specified.
44 * In this mode, zlogin sets up pipes as the communication channel, and
45 * 'su' is used to do the login setup work.
46 *
47 * - "console login" is the equivalent to accessing the tip line for a
48 * zone. For example, the user can issue 'zlogin -C my-zone'.
49 * In this mode, zlogin contacts the zoneadmd process via unix domain
50 * socket. If zoneadmd is not running, it starts it. This allows the
51 * console to be available anytime the zone is installed, regardless of
52 * whether it is running.
53 */
54
55 #include <sys/socket.h>
56 #include <sys/termios.h>
57 #include <sys/utsname.h>
58 #include <sys/stat.h>
59 #include <sys/types.h>
60 #include <sys/contract/process.h>
61 #include <sys/ctfs.h>
62 #include <sys/brand.h>
63 #include <sys/wait.h>
64 #include <alloca.h>
65 #include <assert.h>
66 #include <ctype.h>
67 #include <paths.h>
68 #include <door.h>
69 #include <errno.h>
70 #include <nss_dbdefs.h>
71 #include <poll.h>
72 #include <priv.h>
73 #include <pwd.h>
74 #include <unistd.h>
75 #include <utmpx.h>
76 #include <sac.h>
77 #include <signal.h>
78 #include <stdarg.h>
79 #include <stdio.h>
80 #include <stdlib.h>
81 #include <string.h>
82 #include <strings.h>
83 #include <stropts.h>
84 #include <wait.h>
85 #include <zone.h>
86 #include <fcntl.h>
87 #include <libdevinfo.h>
88 #include <libintl.h>
89 #include <locale.h>
90 #include <libzonecfg.h>
91 #include <libcontract.h>
92 #include <libbrand.h>
93 #include <auth_list.h>
94 #include <auth_attr.h>
95 #include <secdb.h>
96
97 static int masterfd;
98 static struct termios save_termios;
99 static struct termios effective_termios;
100 static int save_fd;
101 static struct winsize winsize;
102 static volatile int dead;
103 static volatile pid_t child_pid = -1;
104 static int interactive = 0;
105 static priv_set_t *dropprivs;
106
107 static int nocmdchar = 0;
108 static int failsafe = 0;
109 static int disconnect = 0;
110 static char cmdchar = '~';
111 static int quiet = 0;
112
113 static int pollerr = 0;
114
115 static const char *pname;
116 static char *username;
117
118 extern int __xpg4; /* 0 if not an xpg4/6-compiled program */
119
120 /*
121 * When forced_login is true, the user is not prompted
122 * for an authentication password in the target zone.
123 */
124 static boolean_t forced_login = B_FALSE;
125
126 #if !defined(TEXT_DOMAIN) /* should be defined by cc -D */
127 #define TEXT_DOMAIN "SYS_TEST" /* Use this only if it wasn't */
128 #endif
129
130 #define SUPATH "/usr/bin/su"
131 #define FAILSAFESHELL "/sbin/sh"
132 #define DEFAULTSHELL "/sbin/sh"
133 #define DEF_PATH "/usr/sbin:/usr/bin"
134
135 #define CLUSTER_BRAND_NAME "cluster"
136
137 /*
138 * The ZLOGIN_BUFSIZ is larger than PIPE_BUF so we can be sure we're clearing
139 * out the pipe when the child is exiting. The ZLOGIN_RDBUFSIZ must be less
140 * than ZLOGIN_BUFSIZ (because we share the buffer in doio). This value is
141 * also chosen in conjunction with the HI_WATER setting to make sure we
142 * don't fill up the pipe. We can write FIFOHIWAT (16k) into the pipe before
143 * blocking. By having ZLOGIN_RDBUFSIZ set to 1k and HI_WATER set to 8k, we
144 * know we can always write a ZLOGIN_RDBUFSIZ chunk into the pipe when there
145 * is less than HI_WATER data already in the pipe.
146 */
147 #define ZLOGIN_BUFSIZ 8192
148 #define ZLOGIN_RDBUFSIZ 1024
149 #define HI_WATER 8192
150
151 /*
152 * See canonify() below. CANONIFY_LEN is the maximum length that a
153 * "canonical" sequence will expand to (backslash, three octal digits, NUL).
154 */
155 #define CANONIFY_LEN 5
156
157 static void
158 usage(void)
159 {
160 (void) fprintf(stderr, gettext("usage: %s [ -dnQCES ] [ -e cmdchar ] "
161 "[-l user] zonename [command [args ...] ]\n"), pname);
162 exit(2);
163 }
164
165 static const char *
166 getpname(const char *arg0)
167 {
168 const char *p = strrchr(arg0, '/');
169
170 if (p == NULL)
171 p = arg0;
172 else
173 p++;
174
175 pname = p;
176 return (p);
177 }
178
179 static void
180 zerror(const char *fmt, ...)
181 {
182 va_list alist;
183
184 (void) fprintf(stderr, "%s: ", pname);
185 va_start(alist, fmt);
186 (void) vfprintf(stderr, fmt, alist);
187 va_end(alist);
188 (void) fprintf(stderr, "\n");
189 }
190
191 static void
192 zperror(const char *str)
193 {
194 const char *estr;
195
196 if ((estr = strerror(errno)) != NULL)
197 (void) fprintf(stderr, "%s: %s: %s\n", pname, str, estr);
198 else
199 (void) fprintf(stderr, "%s: %s: errno %d\n", pname, str, errno);
200 }
201
202 /*
203 * The first part of our privilege dropping scheme needs to be called before
204 * fork(), since we must have it for security; we don't want to be surprised
205 * later that we couldn't allocate the privset.
206 */
207 static int
208 prefork_dropprivs()
209 {
210 if ((dropprivs = priv_allocset()) == NULL)
211 return (1);
212
213 priv_basicset(dropprivs);
214 (void) priv_delset(dropprivs, PRIV_PROC_INFO);
215 (void) priv_delset(dropprivs, PRIV_PROC_FORK);
216 (void) priv_delset(dropprivs, PRIV_PROC_EXEC);
217 (void) priv_delset(dropprivs, PRIV_FILE_LINK_ANY);
218
219 /*
220 * We need to keep the basic privilege PROC_SESSION and all unknown
221 * basic privileges as well as the privileges PROC_ZONE and
222 * PROC_OWNER in order to query session information and
223 * send signals.
224 */
225 if (interactive == 0) {
226 (void) priv_addset(dropprivs, PRIV_PROC_ZONE);
227 (void) priv_addset(dropprivs, PRIV_PROC_OWNER);
228 } else {
229 (void) priv_delset(dropprivs, PRIV_PROC_SESSION);
230 }
231
232 return (0);
233 }
234
235 /*
236 * The second part of the privilege drop. We are paranoid about being attacked
237 * by the zone, so we drop all privileges. This should prevent a compromise
238 * which gets us to fork(), exec(), symlink(), etc.
239 */
240 static void
241 postfork_dropprivs()
242 {
243 if ((setppriv(PRIV_SET, PRIV_PERMITTED, dropprivs)) == -1) {
244 zperror(gettext("Warning: could not set permitted privileges"));
245 }
246 if ((setppriv(PRIV_SET, PRIV_LIMIT, dropprivs)) == -1) {
247 zperror(gettext("Warning: could not set limit privileges"));
248 }
249 if ((setppriv(PRIV_SET, PRIV_INHERITABLE, dropprivs)) == -1) {
250 zperror(gettext("Warning: could not set inheritable "
251 "privileges"));
252 }
253 }
254
255 /*
256 * Create the unix domain socket and call the zoneadmd server; handshake
257 * with it to determine whether it will allow us to connect.
258 */
259 static int
260 get_console_master(const char *zname)
261 {
262 int sockfd = -1;
263 struct sockaddr_un servaddr;
264 char clientid[MAXPATHLEN];
265 char handshake[MAXPATHLEN], c;
266 int msglen;
267 int i = 0, err = 0;
268
269 if ((sockfd = socket(AF_UNIX, SOCK_STREAM, 0)) == -1) {
270 zperror(gettext("could not create socket"));
271 return (-1);
272 }
273
274 bzero(&servaddr, sizeof (servaddr));
275 servaddr.sun_family = AF_UNIX;
276 (void) snprintf(servaddr.sun_path, sizeof (servaddr.sun_path),
277 "%s/%s.console_sock", ZONES_TMPDIR, zname);
278
279 if (connect(sockfd, (struct sockaddr *)&servaddr,
280 sizeof (servaddr)) == -1) {
281 zperror(gettext("Could not connect to zone console"));
282 goto bad;
283 }
284 masterfd = sockfd;
285
286 msglen = snprintf(clientid, sizeof (clientid), "IDENT %lu %s %d\n",
287 getpid(), setlocale(LC_MESSAGES, NULL), disconnect);
288
289 if (msglen >= sizeof (clientid) || msglen < 0) {
290 zerror("protocol error");
291 goto bad;
292 }
293
294 if (write(masterfd, clientid, msglen) != msglen) {
295 zerror("protocol error");
296 goto bad;
297 }
298
299 bzero(handshake, sizeof (handshake));
300
301 /*
302 * Take care not to accumulate more than our fill, and leave room for
303 * the NUL at the end.
304 */
305 while ((err = read(masterfd, &c, 1)) == 1) {
306 if (i >= (sizeof (handshake) - 1))
307 break;
308 if (c == '\n')
309 break;
310 handshake[i] = c;
311 i++;
312 }
313
314 /*
315 * If something went wrong during the handshake we bail; perhaps
316 * the server died off.
317 */
318 if (err == -1) {
319 zperror(gettext("Could not connect to zone console"));
320 goto bad;
321 }
322
323 if (strncmp(handshake, "OK", sizeof (handshake)) == 0)
324 return (0);
325
326 zerror(gettext("Console is already in use by process ID %s."),
327 handshake);
328 bad:
329 (void) close(sockfd);
330 masterfd = -1;
331 return (-1);
332 }
333
334
335 /*
336 * Routines to handle pty creation upon zone entry and to shuttle I/O back
337 * and forth between the two terminals. We also compute and store the
338 * name of the slave terminal associated with the master side.
339 */
340 static int
341 get_master_pty()
342 {
343 if ((masterfd = open("/dev/ptmx", O_RDWR|O_NONBLOCK)) < 0) {
344 zperror(gettext("failed to obtain a pseudo-tty"));
345 return (-1);
346 }
347 if (tcgetattr(STDIN_FILENO, &save_termios) == -1) {
348 zperror(gettext("failed to get terminal settings from stdin"));
349 return (-1);
350 }
351 (void) ioctl(STDIN_FILENO, TIOCGWINSZ, (char *)&winsize);
352
353 return (0);
354 }
355
356 /*
357 * This is a bit tricky; normally a pts device will belong to the zone it
358 * is granted to. But in the case of "entering" a zone, we need to establish
359 * the pty before entering the zone so that we can vector I/O to and from it
360 * from the global zone.
361 *
362 * We use the zonept() call to let the ptm driver know what we are up to;
363 * the only other hairy bit is the setting of zoneslavename (which happens
364 * above, in get_master_pty()).
365 */
366 static int
367 init_slave_pty(zoneid_t zoneid, char *devroot)
368 {
369 int slavefd = -1;
370 char *slavename, zoneslavename[MAXPATHLEN];
371
372 /*
373 * Set slave permissions, zone the pts, then unlock it.
374 */
375 if (grantpt(masterfd) != 0) {
376 zperror(gettext("grantpt failed"));
377 return (-1);
378 }
379
380 if (unlockpt(masterfd) != 0) {
381 zperror(gettext("unlockpt failed"));
382 return (-1);
383 }
384
385 /*
386 * We must open the slave side before zoning this pty; otherwise
387 * the kernel would refuse us the open-- zoning a pty makes it
388 * inaccessible to the global zone. Note we are trying to open
389 * the device node via the $ZONEROOT/dev path for this pty.
390 *
391 * Later we'll close the slave out when once we've opened it again
392 * from within the target zone. Blarg.
393 */
394 if ((slavename = ptsname(masterfd)) == NULL) {
395 zperror(gettext("failed to get name for pseudo-tty"));
396 return (-1);
397 }
398
399 (void) snprintf(zoneslavename, sizeof (zoneslavename), "%s%s",
400 devroot, slavename);
401
402 if ((slavefd = open(zoneslavename, O_RDWR)) < 0) {
403 zerror(gettext("failed to open %s: %s"), zoneslavename,
404 strerror(errno));
405 return (-1);
406 }
407
408 /*
409 * Push hardware emulation (ptem), line discipline (ldterm),
410 * and V7/4BSD/Xenix compatibility (ttcompat) modules.
411 */
412 if (ioctl(slavefd, I_PUSH, "ptem") == -1) {
413 zperror(gettext("failed to push ptem module"));
414 if (!failsafe)
415 goto bad;
416 }
417
418 /*
419 * Anchor the stream to prevent malicious I_POPs; we prefer to do
420 * this prior to entering the zone so that we can detect any errors
421 * early, and so that we can set the anchor from the global zone.
422 */
423 if (ioctl(slavefd, I_ANCHOR) == -1) {
424 zperror(gettext("failed to set stream anchor"));
425 if (!failsafe)
426 goto bad;
427 }
428
429 if (ioctl(slavefd, I_PUSH, "ldterm") == -1) {
430 zperror(gettext("failed to push ldterm module"));
431 if (!failsafe)
432 goto bad;
433 }
434 if (ioctl(slavefd, I_PUSH, "ttcompat") == -1) {
435 zperror(gettext("failed to push ttcompat module"));
436 if (!failsafe)
437 goto bad;
438 }
439
440 /*
441 * Propagate terminal settings from the external term to the new one.
442 */
443 if (tcsetattr(slavefd, TCSAFLUSH, &save_termios) == -1) {
444 zperror(gettext("failed to set terminal settings"));
445 if (!failsafe)
446 goto bad;
447 }
448 (void) ioctl(slavefd, TIOCSWINSZ, (char *)&winsize);
449
450 if (zonept(masterfd, zoneid) != 0) {
451 zperror(gettext("could not set zoneid of pty"));
452 goto bad;
453 }
454
455 return (slavefd);
456
457 bad:
458 (void) close(slavefd);
459 return (-1);
460 }
461
462 /*
463 * Place terminal into raw mode.
464 */
465 static int
466 set_tty_rawmode(int fd)
467 {
468 struct termios term;
469 if (tcgetattr(fd, &term) < 0) {
470 zperror(gettext("failed to get user terminal settings"));
471 return (-1);
472 }
473
474 /* Stash for later, so we can revert back to previous mode */
475 save_termios = term;
476 save_fd = fd;
477
478 /* disable 8->7 bit strip, start/stop, enable any char to restart */
479 term.c_iflag &= ~(ISTRIP|IXON|IXANY);
480 /* disable NL->CR, CR->NL, ignore CR, UPPER->lower */
481 term.c_iflag &= ~(INLCR|ICRNL|IGNCR|IUCLC);
482 /* disable output post-processing */
483 term.c_oflag &= ~OPOST;
484 /* disable canonical mode, signal chars, echo & extended functions */
485 term.c_lflag &= ~(ICANON|ISIG|ECHO|IEXTEN);
486
487 term.c_cc[VMIN] = 1; /* byte-at-a-time */
488 term.c_cc[VTIME] = 0;
489
490 if (tcsetattr(STDIN_FILENO, TCSAFLUSH, &term)) {
491 zperror(gettext("failed to set user terminal to raw mode"));
492 return (-1);
493 }
494
495 /*
496 * We need to know the value of VEOF so that we can properly process for
497 * client-side ~<EOF>. But we have obliterated VEOF in term,
498 * because VMIN overloads the same array slot in non-canonical mode.
499 * Stupid @&^%!
500 *
501 * So here we construct the "effective" termios from the current
502 * terminal settings, and the corrected VEOF and VEOL settings.
503 */
504 if (tcgetattr(STDIN_FILENO, &effective_termios) < 0) {
505 zperror(gettext("failed to get user terminal settings"));
506 return (-1);
507 }
508 effective_termios.c_cc[VEOF] = save_termios.c_cc[VEOF];
509 effective_termios.c_cc[VEOL] = save_termios.c_cc[VEOL];
510
511 return (0);
512 }
513
514 /*
515 * Copy terminal window size from our terminal to the pts.
516 */
517 /*ARGSUSED*/
518 static void
519 sigwinch(int s)
520 {
521 struct winsize ws;
522
523 if (ioctl(0, TIOCGWINSZ, &ws) == 0)
524 (void) ioctl(masterfd, TIOCSWINSZ, &ws);
525 }
526
527 static volatile int close_on_sig = -1;
528
529 static void
530 /*ARGSUSED*/
531 sigcld(int s)
532 {
533 int status;
534 pid_t pid;
535
536 /*
537 * Peek at the exit status. If this isn't the process we cared
538 * about, then just reap it.
539 */
540 if ((pid = waitpid(child_pid, &status, WNOHANG|WNOWAIT)) != -1) {
541 if (pid == child_pid &&
542 (WIFEXITED(status) || WIFSIGNALED(status))) {
543 dead = 1;
544 if (close_on_sig != -1) {
545 (void) write(close_on_sig, "a", 1);
546 (void) close(close_on_sig);
547 close_on_sig = -1;
548 }
549 } else {
550 (void) waitpid(pid, &status, WNOHANG);
551 }
552 }
553 }
554
555 /*
556 * Some signals (currently, SIGINT) must be forwarded on to the process
557 * group of the child process.
558 */
559 static void
560 sig_forward(int s)
561 {
562 if (child_pid != -1) {
563 (void) sigsend(P_PGID, child_pid, s);
564 }
565 }
566
567 /*
568 * reset terminal settings for global environment
569 */
570 static void
571 reset_tty()
572 {
573 (void) tcsetattr(save_fd, TCSADRAIN, &save_termios);
574 }
575
576 /*
577 * Convert character to printable representation, for display with locally
578 * echoed command characters (like when we need to display ~^D)
579 */
580 static void
581 canonify(char c, char *cc)
582 {
583 if (isprint(c)) {
584 cc[0] = c;
585 cc[1] = '\0';
586 } else if (c >= 0 && c <= 31) { /* ^@ through ^_ */
587 cc[0] = '^';
588 cc[1] = c + '@';
589 cc[2] = '\0';
590 } else {
591 cc[0] = '\\';
592 cc[1] = ((c >> 6) & 7) + '0';
593 cc[2] = ((c >> 3) & 7) + '0';
594 cc[3] = (c & 7) + '0';
595 cc[4] = '\0';
596 }
597 }
598
599 /*
600 * process_user_input watches the input stream for the escape sequence for
601 * 'quit' (by default, tilde-period). Because we might be fed just one
602 * keystroke at a time, state associated with the user input (are we at the
603 * beginning of the line? are we locally echoing the next character?) is
604 * maintained by beginning_of_line and local_echo across calls to the routine.
605 * If the write to outfd fails, we'll try to read from infd in an attempt
606 * to prevent deadlock between the two processes.
607 *
608 * This routine returns -1 when the 'quit' escape sequence has been issued,
609 * or an error is encountered, 1 if stdin is EOF, and 0 otherwise.
610 */
611 static int
612 process_user_input(int outfd, int infd)
613 {
614 static boolean_t beginning_of_line = B_TRUE;
615 static boolean_t local_echo = B_FALSE;
616 char ibuf[ZLOGIN_BUFSIZ];
617 int nbytes;
618 char *buf = ibuf;
619
620 nbytes = read(STDIN_FILENO, ibuf, ZLOGIN_RDBUFSIZ);
621 if (nbytes == -1 && (errno != EINTR || dead))
622 return (-1);
623
624 if (nbytes == -1) /* The read was interrupted. */
625 return (0);
626
627 /* 0 read means EOF, close the pipe to the child */
628 if (nbytes == 0)
629 return (1);
630
631 for (char c = *buf; nbytes > 0; c = *buf, --nbytes) {
632 buf++;
633 if (beginning_of_line && !nocmdchar) {
634 beginning_of_line = B_FALSE;
635 if (c == cmdchar) {
636 local_echo = B_TRUE;
637 continue;
638 }
639 } else if (local_echo) {
640 local_echo = B_FALSE;
641 if (c == '.' || c == effective_termios.c_cc[VEOF]) {
642 char cc[CANONIFY_LEN];
643
644 canonify(c, cc);
645 (void) write(STDOUT_FILENO, &cmdchar, 1);
646 (void) write(STDOUT_FILENO, cc, strlen(cc));
647 return (-1);
648 }
649 }
650 retry:
651 if (write(outfd, &c, 1) <= 0) {
652 /*
653 * Since the fd we are writing to is opened with
654 * O_NONBLOCK it is possible to get EAGAIN if the
655 * pipe is full. One way this could happen is if we
656 * are writing a lot of data into the pipe in this loop
657 * and the application on the other end is echoing that
658 * data back out to its stdout. The output pipe can
659 * fill up since we are stuck here in this loop and not
660 * draining the other pipe. We can try to read some of
661 * the data to see if we can drain the pipe so that the
662 * application can continue to make progress. The read
663 * is non-blocking so we won't hang here. We also wait
664 * a bit before retrying since there could be other
665 * reasons why the pipe is full and we don't want to
666 * continuously retry.
667 */
668 if (errno == EAGAIN) {
669 struct timespec rqtp;
670 int ln;
671 char obuf[ZLOGIN_BUFSIZ];
672
673 if ((ln = read(infd, obuf, ZLOGIN_BUFSIZ)) > 0)
674 (void) write(STDOUT_FILENO, obuf, ln);
675
676 /* sleep for 10 milliseconds */
677 rqtp.tv_sec = 0;
678 rqtp.tv_nsec = MSEC2NSEC(10);
679 (void) nanosleep(&rqtp, NULL);
680 if (!dead)
681 goto retry;
682 }
683
684 return (-1);
685 }
686 beginning_of_line = (c == '\r' || c == '\n' ||
687 c == effective_termios.c_cc[VKILL] ||
688 c == effective_termios.c_cc[VEOL] ||
689 c == effective_termios.c_cc[VSUSP] ||
690 c == effective_termios.c_cc[VINTR]);
691 }
692 return (0);
693 }
694
695 /*
696 * This function prevents deadlock between zlogin and the application in the
697 * zone that it is talking to. This can happen when we read from zlogin's
698 * stdin and write the data down the pipe to the application. If the pipe
699 * is full, we'll block in the write. Because zlogin could be blocked in
700 * the write, it would never read the application's stdout/stderr so the
701 * application can then block on those writes (when the pipe fills up). If the
702 * the application gets blocked this way, it can never get around to reading
703 * its stdin so that zlogin can unblock from its write. Once in this state,
704 * the two processes are deadlocked.
705 *
706 * To prevent this, we want to verify that we can write into the pipe before we
707 * read from our stdin. If the pipe already is pretty full, we bypass the read
708 * for now. We'll circle back here again after the poll() so that we can
709 * try again. When this function is called, we already know there is data
710 * ready to read on STDIN_FILENO. We return -1 if there is a problem, 1 if
711 * stdin is EOF, and 0 if everything is ok (even though we might not have
712 * read/written any data into the pipe on this iteration).
713 */
714 static int
715 process_raw_input(int stdin_fd, int appin_fd)
716 {
717 int cc;
718 struct stat64 sb;
719 char ibuf[ZLOGIN_RDBUFSIZ];
720
721 /* Check how much data is already in the pipe */
722 if (fstat64(appin_fd, &sb) == -1) {
723 perror("stat failed");
724 return (-1);
725 }
726
727 if (dead)
728 return (-1);
729
730 /*
731 * The pipe already has a lot of data in it, don't write any more
732 * right now.
733 */
734 if (sb.st_size >= HI_WATER)
735 return (0);
736
737 cc = read(STDIN_FILENO, ibuf, ZLOGIN_RDBUFSIZ);
738 if (cc == -1 && (errno != EINTR || dead))
739 return (-1);
740
741 if (cc == -1) /* The read was interrupted. */
742 return (0);
743
744 /* 0 read means EOF, close the pipe to the child */
745 if (cc == 0)
746 return (1);
747
748 /*
749 * stdin_fd is stdin of the target; so, the thing we'll write the user
750 * data *to*.
751 */
752 if (write(stdin_fd, ibuf, cc) == -1)
753 return (-1);
754
755 return (0);
756 }
757
758 /*
759 * Write the output from the application running in the zone. We can get
760 * a signal during the write (usually it would be SIGCHLD when the application
761 * has exited) so we loop to make sure we have written all of the data we read.
762 */
763 static int
764 process_output(int in_fd, int out_fd)
765 {
766 int wrote = 0;
767 int cc;
768 char ibuf[ZLOGIN_BUFSIZ];
769
770 cc = read(in_fd, ibuf, ZLOGIN_BUFSIZ);
771 if (cc == -1 && (errno != EINTR || dead))
772 return (-1);
773 if (cc == 0) {
774 /*
775 * A return value of 0 when calling read() on a terminal
776 * indicates end-of-file pre-XPG4 and no data available
777 * for XPG4 and above.
778 */
779 if (__xpg4 == 0)
780 return (-1);
781 return (0);
782 }
783 if (cc == -1) /* The read was interrupted. */
784 return (0);
785
786 do {
787 int len;
788
789 len = write(out_fd, ibuf + wrote, cc - wrote);
790 if (len == -1 && errno != EINTR)
791 return (-1);
792 if (len != -1)
793 wrote += len;
794 } while (wrote < cc);
795
796 return (0);
797 }
798
799 /*
800 * This is the main I/O loop, and is shared across all zlogin modes.
801 * Parameters:
802 * stdin_fd: The fd representing 'stdin' for the slave side; input to
803 * the zone will be written here.
804 *
805 * appin_fd: The fd representing the other end of the 'stdin' pipe (when
806 * we're running non-interactive); used in process_raw_input
807 * to ensure we don't fill up the application's stdin pipe.
808 *
809 * stdout_fd: The fd representing 'stdout' for the slave side; output
810 * from the zone will arrive here.
811 *
812 * stderr_fd: The fd representing 'stderr' for the slave side; output
813 * from the zone will arrive here.
814 *
815 * raw_mode: If TRUE, then no processing (for example, for '~.') will
816 * be performed on the input coming from STDIN.
817 *
818 * stderr_fd may be specified as -1 if there is no stderr (only non-interactive
819 * mode supplies a stderr).
820 *
821 */
822 static void
823 doio(int stdin_fd, int appin_fd, int stdout_fd, int stderr_fd, int sig_fd,
824 boolean_t raw_mode)
825 {
826 struct pollfd pollfds[4];
827 char ibuf[ZLOGIN_BUFSIZ];
828 int cc, ret;
829
830 /* read from stdout of zone and write to stdout of global zone */
831 pollfds[0].fd = stdout_fd;
832 pollfds[0].events = POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI;
833
834 /* read from stderr of zone and write to stderr of global zone */
835 pollfds[1].fd = stderr_fd;
836 pollfds[1].events = pollfds[0].events;
837
838 /* read from stdin of global zone and write to stdin of zone */
839 pollfds[2].fd = STDIN_FILENO;
840 pollfds[2].events = pollfds[0].events;
841
842 /* read from signalling pipe so we know when child dies */
843 pollfds[3].fd = sig_fd;
844 pollfds[3].events = pollfds[0].events;
845
846 for (;;) {
847 pollfds[0].revents = pollfds[1].revents =
848 pollfds[2].revents = pollfds[3].revents = 0;
849
850 if (dead)
851 break;
852
853 /*
854 * There is a race condition here where we can receive the
855 * child death signal, set the dead flag, but since we have
856 * passed the test above, we would go into poll and hang.
857 * To avoid this we use the sig_fd as an additional poll fd.
858 * The signal handler writes into the other end of this pipe
859 * when the child dies so that the poll will always see that
860 * input and proceed. We just loop around at that point and
861 * then notice the dead flag.
862 */
863
864 ret = poll(pollfds,
865 sizeof (pollfds) / sizeof (struct pollfd), -1);
866
867 if (ret == -1 && errno != EINTR) {
868 perror("poll failed");
869 break;
870 }
871
872 if (errno == EINTR && dead) {
873 break;
874 }
875
876 /* event from master side stdout */
877 if (pollfds[0].revents) {
878 if (pollfds[0].revents &
879 (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
880 if (process_output(stdout_fd, STDOUT_FILENO)
881 != 0)
882 break;
883 } else {
884 pollerr = pollfds[0].revents;
885 break;
886 }
887 }
888
889 /* event from master side stderr */
890 if (pollfds[1].revents) {
891 if (pollfds[1].revents &
892 (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
893 if (process_output(stderr_fd, STDERR_FILENO)
894 != 0)
895 break;
896 } else {
897 pollerr = pollfds[1].revents;
898 break;
899 }
900 }
901
902 /* event from user STDIN side */
903 if (pollfds[2].revents) {
904 if (pollfds[2].revents &
905 (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
906 /*
907 * stdin fd is stdin of the target; so,
908 * the thing we'll write the user data *to*.
909 *
910 * Also, unlike on the output side, we
911 * close the pipe on a zero-length message.
912 */
913 int res;
914
915 if (raw_mode)
916 res = process_raw_input(stdin_fd,
917 appin_fd);
918 else
919 res = process_user_input(stdin_fd,
920 stdout_fd);
921
922 if (res < 0)
923 break;
924 if (res > 0) {
925 /* EOF (close) child's stdin_fd */
926 pollfds[2].fd = -1;
927 while ((res = close(stdin_fd)) != 0 &&
928 errno == EINTR)
929 ;
930 if (res != 0)
931 break;
932 }
933
934 } else if (raw_mode && pollfds[2].revents & POLLHUP) {
935 /*
936 * It's OK to get a POLLHUP on STDIN-- it
937 * always happens if you do:
938 *
939 * echo foo | zlogin <zone> <command>
940 *
941 * We reset fd to -1 in this case to clear
942 * the condition and close the pipe (EOF) to
943 * the other side in order to wrap things up.
944 */
945 int res;
946
947 pollfds[2].fd = -1;
948 while ((res = close(stdin_fd)) != 0 &&
949 errno == EINTR)
950 ;
951 if (res != 0)
952 break;
953 } else {
954 pollerr = pollfds[2].revents;
955 break;
956 }
957 }
958 }
959
960 /*
961 * We are in the midst of dying, but try to poll with a short
962 * timeout to see if we can catch the last bit of I/O from the
963 * children.
964 */
965 retry:
966 pollfds[0].revents = pollfds[1].revents = 0;
967 (void) poll(pollfds, 2, 100);
968 if (pollfds[0].revents &
969 (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
970 if ((cc = read(stdout_fd, ibuf, ZLOGIN_BUFSIZ)) > 0) {
971 (void) write(STDOUT_FILENO, ibuf, cc);
972 goto retry;
973 }
974 }
975 if (pollfds[1].revents &
976 (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
977 if ((cc = read(stderr_fd, ibuf, ZLOGIN_BUFSIZ)) > 0) {
978 (void) write(STDERR_FILENO, ibuf, cc);
979 goto retry;
980 }
981 }
982 }
983
984 /*
985 * Fetch the user_cmd brand hook for getting a user's passwd(4) entry.
986 */
987 static const char *
988 zone_get_user_cmd(brand_handle_t bh, const char *login, char *user_cmd,
989 size_t len)
990 {
991 bzero(user_cmd, sizeof (user_cmd));
992 if (brand_get_user_cmd(bh, login, user_cmd, len) != 0)
993 return (NULL);
994
995 return (user_cmd);
996 }
997
998 /* From libc */
999 extern int str2passwd(const char *, int, void *, char *, int);
1000
1001 /*
1002 * exec() the user_cmd brand hook, and convert the output string to a
1003 * struct passwd. This is to be called after zone_enter().
1004 *
1005 */
1006 static struct passwd *
1007 zone_get_user_pw(const char *user_cmd, struct passwd *pwent, char *pwbuf,
1008 int pwbuflen)
1009 {
1010 char pwline[NSS_BUFLEN_PASSWD];
1011 char *cin = NULL;
1012 FILE *fin;
1013 int status;
1014
1015 assert(getzoneid() != GLOBAL_ZONEID);
1016
1017 if ((fin = popen(user_cmd, "r")) == NULL)
1018 return (NULL);
1019
1020 while (cin == NULL && !feof(fin))
1021 cin = fgets(pwline, sizeof (pwline), fin);
1022
1023 if (cin == NULL) {
1024 (void) pclose(fin);
1025 return (NULL);
1026 }
1027
1028 status = pclose(fin);
1029 if (!WIFEXITED(status))
1030 return (NULL);
1031 if (WEXITSTATUS(status) != 0)
1032 return (NULL);
1033
1034 if (str2passwd(pwline, sizeof (pwline), pwent, pwbuf, pwbuflen) == 0)
1035 return (pwent);
1036 else
1037 return (NULL);
1038 }
1039
1040 static char **
1041 zone_login_cmd(brand_handle_t bh, const char *login)
1042 {
1043 static char result_buf[ARG_MAX];
1044 char **new_argv, *ptr, *lasts;
1045 int n, a;
1046
1047 /* Get the login command for the target zone. */
1048 bzero(result_buf, sizeof (result_buf));
1049
1050 if (forced_login) {
1051 if (brand_get_forcedlogin_cmd(bh, login,
1052 result_buf, sizeof (result_buf)) != 0)
1053 return (NULL);
1054 } else {
1055 if (brand_get_login_cmd(bh, login,
1056 result_buf, sizeof (result_buf)) != 0)
1057 return (NULL);
1058 }
1059
1060 /*
1061 * We got back a string that we'd like to execute. But since
1062 * we're not doing the execution via a shell we'll need to convert
1063 * the exec string to an array of strings. We'll do that here
1064 * but we're going to be very simplistic about it and break stuff
1065 * up based on spaces. We're not even going to support any kind
1066 * of quoting or escape characters. It's truly amazing that
1067 * there is no library function in OpenSolaris to do this for us.
1068 */
1069
1070 /*
1071 * Be paranoid. Since we're deliniating based on spaces make
1072 * sure there are no adjacent spaces.
1073 */
1074 if (strstr(result_buf, " ") != NULL)
1075 return (NULL);
1076
1077 /* Remove any trailing whitespace. */
1078 n = strlen(result_buf);
1079 if (result_buf[n - 1] == ' ')
1080 result_buf[n - 1] = '\0';
1081
1082 /* Count how many elements there are in the exec string. */
1083 ptr = result_buf;
1084 for (n = 2; ((ptr = strchr(ptr + 1, (int)' ')) != NULL); n++)
1085 ;
1086
1087 /* Allocate the argv array that we're going to return. */
1088 if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1089 return (NULL);
1090
1091 /* Tokenize the exec string and return. */
1092 a = 0;
1093 new_argv[a++] = result_buf;
1094 if (n > 2) {
1095 (void) strtok_r(result_buf, " ", &lasts);
1096 while ((new_argv[a++] = strtok_r(NULL, " ", &lasts)) != NULL)
1097 ;
1098 } else {
1099 new_argv[a++] = NULL;
1100 }
1101 assert(n == a);
1102 return (new_argv);
1103 }
1104
1105 /*
1106 * Prepare argv array for exec'd process; if we're passing commands to the
1107 * new process, then use su(1M) to do the invocation. Otherwise, use
1108 * 'login -z <from_zonename> -f' (-z is an undocumented option which tells
1109 * login that we're coming from another zone, and to disregard its CONSOLE
1110 * checks).
1111 */
1112 static char **
1113 prep_args(brand_handle_t bh, const char *login, char **argv)
1114 {
1115 int argc = 0, a = 0, i, n = -1;
1116 char **new_argv;
1117
1118 if (argv != NULL) {
1119 size_t subshell_len = 1;
1120 char *subshell;
1121
1122 while (argv[argc] != NULL)
1123 argc++;
1124
1125 for (i = 0; i < argc; i++) {
1126 subshell_len += strlen(argv[i]) + 1;
1127 }
1128 if ((subshell = calloc(1, subshell_len)) == NULL)
1129 return (NULL);
1130
1131 for (i = 0; i < argc; i++) {
1132 (void) strcat(subshell, argv[i]);
1133 (void) strcat(subshell, " ");
1134 }
1135
1136 if (failsafe) {
1137 n = 4;
1138 if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1139 return (NULL);
1140
1141 new_argv[a++] = FAILSAFESHELL;
1142 } else {
1143 n = 5;
1144 if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1145 return (NULL);
1146
1147 new_argv[a++] = SUPATH;
1148 if (strcmp(login, "root") != 0) {
1149 new_argv[a++] = "-";
1150 n++;
1151 }
1152 new_argv[a++] = (char *)login;
1153 }
1154 new_argv[a++] = "-c";
1155 new_argv[a++] = subshell;
1156 new_argv[a++] = NULL;
1157 assert(a == n);
1158 } else {
1159 if (failsafe) {
1160 n = 2;
1161 if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1162 return (NULL);
1163 new_argv[a++] = FAILSAFESHELL;
1164 new_argv[a++] = NULL;
1165 assert(n == a);
1166 } else {
1167 new_argv = zone_login_cmd(bh, login);
1168 }
1169 }
1170
1171 return (new_argv);
1172 }
1173
1174 /*
1175 * Helper routine for prep_env below.
1176 */
1177 static char *
1178 add_env(char *name, char *value)
1179 {
1180 size_t sz = strlen(name) + strlen(value) + 2; /* name, =, value, NUL */
1181 char *str;
1182
1183 if ((str = malloc(sz)) == NULL)
1184 return (NULL);
1185
1186 (void) snprintf(str, sz, "%s=%s", name, value);
1187 return (str);
1188 }
1189
1190 /*
1191 * Prepare envp array for exec'd process.
1192 */
1193 static char **
1194 prep_env()
1195 {
1196 int e = 0, size = 1;
1197 char **new_env, *estr;
1198 char *term = getenv("TERM");
1199
1200 size++; /* for $PATH */
1201 if (term != NULL)
1202 size++;
1203
1204 /*
1205 * In failsafe mode we set $HOME, since '-l' isn't valid in this mode.
1206 * We also set $SHELL, since neither login nor su will be around to do
1207 * it.
1208 */
1209 if (failsafe)
1210 size += 2;
1211
1212 if ((new_env = malloc(sizeof (char *) * size)) == NULL)
1213 return (NULL);
1214
1215 if ((estr = add_env("PATH", DEF_PATH)) == NULL)
1216 return (NULL);
1217 new_env[e++] = estr;
1218
1219 if (term != NULL) {
1220 if ((estr = add_env("TERM", term)) == NULL)
1221 return (NULL);
1222 new_env[e++] = estr;
1223 }
1224
1225 if (failsafe) {
1226 if ((estr = add_env("HOME", "/")) == NULL)
1227 return (NULL);
1228 new_env[e++] = estr;
1229
1230 if ((estr = add_env("SHELL", FAILSAFESHELL)) == NULL)
1231 return (NULL);
1232 new_env[e++] = estr;
1233 }
1234
1235 new_env[e++] = NULL;
1236
1237 assert(e == size);
1238
1239 return (new_env);
1240 }
1241
1242 /*
1243 * Finish the preparation of the envp array for exec'd non-interactive
1244 * zlogins. This is called in the child process *after* we zone_enter(), since
1245 * it derives things we can only know within the zone, such as $HOME, $SHELL,
1246 * etc. We need only do this in the non-interactive, mode, since otherwise
1247 * login(1) will do it. We don't do this in failsafe mode, since it presents
1248 * additional ways in which the command could fail, and we'd prefer to avoid
1249 * that.
1250 */
1251 static char **
1252 prep_env_noninteractive(const char *user_cmd, char **env)
1253 {
1254 size_t size;
1255 char **new_env;
1256 int e, i;
1257 char *estr;
1258 char varmail[LOGNAME_MAX + 11]; /* strlen(/var/mail/) = 10, NUL */
1259 char pwbuf[NSS_BUFLEN_PASSWD + 1];
1260 struct passwd pwent;
1261 struct passwd *pw = NULL;
1262
1263 assert(env != NULL);
1264 assert(failsafe == 0);
1265
1266 /*
1267 * Exec the "user_cmd" brand hook to get a pwent for the
1268 * login user. If this fails, HOME will be set to "/", SHELL
1269 * will be set to $DEFAULTSHELL, and we will continue to exec
1270 * SUPATH <login> -c <cmd>.
1271 */
1272 pw = zone_get_user_pw(user_cmd, &pwent, pwbuf, sizeof (pwbuf));
1273
1274 /*
1275 * Get existing envp size.
1276 */
1277 for (size = 0; env[size] != NULL; size++)
1278 ;
1279
1280 e = size;
1281
1282 /*
1283 * Finish filling out the environment; we duplicate the environment
1284 * setup described in login(1), for lack of a better precedent.
1285 */
1286 if (pw != NULL)
1287 size += 3; /* LOGNAME, HOME, MAIL */
1288 else
1289 size += 1; /* HOME */
1290
1291 size++; /* always fill in SHELL */
1292 size++; /* terminating NULL */
1293
1294 if ((new_env = malloc(sizeof (char *) * size)) == NULL)
1295 goto malloc_fail;
1296
1297 /*
1298 * Copy existing elements of env into new_env.
1299 */
1300 for (i = 0; env[i] != NULL; i++) {
1301 if ((new_env[i] = strdup(env[i])) == NULL)
1302 goto malloc_fail;
1303 }
1304 assert(e == i);
1305
1306 if (pw != NULL) {
1307 if ((estr = add_env("LOGNAME", pw->pw_name)) == NULL)
1308 goto malloc_fail;
1309 new_env[e++] = estr;
1310
1311 if ((estr = add_env("HOME", pw->pw_dir)) == NULL)
1312 goto malloc_fail;
1313 new_env[e++] = estr;
1314
1315 if (chdir(pw->pw_dir) != 0)
1316 zerror(gettext("Could not chdir to home directory "
1317 "%s: %s"), pw->pw_dir, strerror(errno));
1318
1319 (void) snprintf(varmail, sizeof (varmail), "/var/mail/%s",
1320 pw->pw_name);
1321 if ((estr = add_env("MAIL", varmail)) == NULL)
1322 goto malloc_fail;
1323 new_env[e++] = estr;
1324 } else {
1325 if ((estr = add_env("HOME", "/")) == NULL)
1326 goto malloc_fail;
1327 new_env[e++] = estr;
1328 }
1329
1330 if (pw != NULL && strlen(pw->pw_shell) > 0) {
1331 if ((estr = add_env("SHELL", pw->pw_shell)) == NULL)
1332 goto malloc_fail;
1333 new_env[e++] = estr;
1334 } else {
1335 if ((estr = add_env("SHELL", DEFAULTSHELL)) == NULL)
1336 goto malloc_fail;
1337 new_env[e++] = estr;
1338 }
1339
1340 new_env[e++] = NULL; /* add terminating NULL */
1341
1342 assert(e == size);
1343 return (new_env);
1344
1345 malloc_fail:
1346 zperror(gettext("failed to allocate memory for process environment"));
1347 return (NULL);
1348 }
1349
1350 static int
1351 close_func(void *slavefd, int fd)
1352 {
1353 if (fd != *(int *)slavefd)
1354 (void) close(fd);
1355 return (0);
1356 }
1357
1358 static void
1359 set_cmdchar(char *cmdcharstr)
1360 {
1361 char c;
1362 long lc;
1363
1364 if ((c = *cmdcharstr) != '\\') {
1365 cmdchar = c;
1366 return;
1367 }
1368
1369 c = cmdcharstr[1];
1370 if (c == '\0' || c == '\\') {
1371 cmdchar = '\\';
1372 return;
1373 }
1374
1375 if (c < '0' || c > '7') {
1376 zerror(gettext("Unrecognized escape character option %s"),
1377 cmdcharstr);
1378 usage();
1379 }
1380
1381 lc = strtol(cmdcharstr + 1, NULL, 8);
1382 if (lc < 0 || lc > 255) {
1383 zerror(gettext("Octal escape character '%s' too large"),
1384 cmdcharstr);
1385 usage();
1386 }
1387 cmdchar = (char)lc;
1388 }
1389
1390 static int
1391 setup_utmpx(char *slavename)
1392 {
1393 struct utmpx ut;
1394
1395 bzero(&ut, sizeof (ut));
1396 (void) strncpy(ut.ut_user, ".zlogin", sizeof (ut.ut_user));
1397 (void) strncpy(ut.ut_line, slavename, sizeof (ut.ut_line));
1398 ut.ut_pid = getpid();
1399 ut.ut_id[0] = 'z';
1400 ut.ut_id[1] = ut.ut_id[2] = ut.ut_id[3] = (char)SC_WILDC;
1401 ut.ut_type = LOGIN_PROCESS;
1402 (void) time(&ut.ut_tv.tv_sec);
1403
1404 if (makeutx(&ut) == NULL) {
1405 zerror(gettext("makeutx failed"));
1406 return (-1);
1407 }
1408 return (0);
1409 }
1410
1411 static void
1412 release_lock_file(int lockfd)
1413 {
1414 (void) close(lockfd);
1415 }
1416
1417 static int
1418 grab_lock_file(const char *zone_name, int *lockfd)
1419 {
1420 char pathbuf[PATH_MAX];
1421 struct flock flock;
1422
1423 if (mkdir(ZONES_TMPDIR, S_IRWXU) < 0 && errno != EEXIST) {
1424 zerror(gettext("could not mkdir %s: %s"), ZONES_TMPDIR,
1425 strerror(errno));
1426 return (-1);
1427 }
1428 (void) chmod(ZONES_TMPDIR, S_IRWXU);
1429 (void) snprintf(pathbuf, sizeof (pathbuf), "%s/%s.zoneadm.lock",
1430 ZONES_TMPDIR, zone_name);
1431
1432 if ((*lockfd = open(pathbuf, O_RDWR|O_CREAT, S_IRUSR|S_IWUSR)) < 0) {
1433 zerror(gettext("could not open %s: %s"), pathbuf,
1434 strerror(errno));
1435 return (-1);
1436 }
1437 /*
1438 * Lock the file to synchronize with other zoneadmds
1439 */
1440 flock.l_type = F_WRLCK;
1441 flock.l_whence = SEEK_SET;
1442 flock.l_start = (off_t)0;
1443 flock.l_len = (off_t)0;
1444 if (fcntl(*lockfd, F_SETLKW, &flock) < 0) {
1445 zerror(gettext("unable to lock %s: %s"), pathbuf,
1446 strerror(errno));
1447 release_lock_file(*lockfd);
1448 return (-1);
1449 }
1450 return (Z_OK);
1451 }
1452
1453 static int
1454 start_zoneadmd(const char *zone_name)
1455 {
1456 pid_t retval;
1457 int pstatus = 0, error = -1, lockfd, doorfd;
1458 struct door_info info;
1459 char doorpath[MAXPATHLEN];
1460
1461 (void) snprintf(doorpath, sizeof (doorpath), ZONE_DOOR_PATH, zone_name);
1462
1463 if (grab_lock_file(zone_name, &lockfd) != Z_OK)
1464 return (-1);
1465 /*
1466 * We must do the door check with the lock held. Otherwise, we
1467 * might race against another zoneadm/zlogin process and wind
1468 * up with two processes trying to start zoneadmd at the same
1469 * time. zoneadmd will detect this, and fail, but we prefer this
1470 * to be as seamless as is practical, from a user perspective.
1471 */
1472 if ((doorfd = open(doorpath, O_RDONLY)) < 0) {
1473 if (errno != ENOENT) {
1474 zerror("failed to open %s: %s", doorpath,
1475 strerror(errno));
1476 goto out;
1477 }
1478 } else {
1479 /*
1480 * Seems to be working ok.
1481 */
1482 if (door_info(doorfd, &info) == 0 &&
1483 ((info.di_attributes & DOOR_REVOKED) == 0)) {
1484 error = 0;
1485 goto out;
1486 }
1487 }
1488
1489 if ((child_pid = fork()) == -1) {
1490 zperror(gettext("could not fork"));
1491 goto out;
1492 } else if (child_pid == 0) {
1493 /* child process */
1494 (void) execl("/usr/lib/zones/zoneadmd", "zoneadmd", "-z",
1495 zone_name, NULL);
1496 zperror(gettext("could not exec zoneadmd"));
1497 _exit(1);
1498 }
1499
1500 /* parent process */
1501 do {
1502 retval = waitpid(child_pid, &pstatus, 0);
1503 } while (retval != child_pid);
1504 if (WIFSIGNALED(pstatus) ||
1505 (WIFEXITED(pstatus) && WEXITSTATUS(pstatus) != 0)) {
1506 zerror(gettext("could not start %s"), "zoneadmd");
1507 goto out;
1508 }
1509 error = 0;
1510 out:
1511 release_lock_file(lockfd);
1512 (void) close(doorfd);
1513 return (error);
1514 }
1515
1516 static int
1517 init_template(void)
1518 {
1519 int fd;
1520 int err = 0;
1521
1522 fd = open64(CTFS_ROOT "/process/template", O_RDWR);
1523 if (fd == -1)
1524 return (-1);
1525
1526 /*
1527 * zlogin doesn't do anything with the contract.
1528 * Deliver no events, don't inherit, and allow it to be orphaned.
1529 */
1530 err |= ct_tmpl_set_critical(fd, 0);
1531 err |= ct_tmpl_set_informative(fd, 0);
1532 err |= ct_pr_tmpl_set_fatal(fd, CT_PR_EV_HWERR);
1533 err |= ct_pr_tmpl_set_param(fd, CT_PR_PGRPONLY | CT_PR_REGENT);
1534 if (err || ct_tmpl_activate(fd)) {
1535 (void) close(fd);
1536 return (-1);
1537 }
1538
1539 return (fd);
1540 }
1541
1542 static int
1543 noninteractive_login(char *zonename, const char *user_cmd, zoneid_t zoneid,
1544 char **new_args, char **new_env)
1545 {
1546 pid_t retval;
1547 int stdin_pipe[2], stdout_pipe[2], stderr_pipe[2], dead_child_pipe[2];
1548 int child_status;
1549 int tmpl_fd;
1550 sigset_t block_cld;
1551
1552 if ((tmpl_fd = init_template()) == -1) {
1553 reset_tty();
1554 zperror(gettext("could not create contract"));
1555 return (1);
1556 }
1557
1558 if (pipe(stdin_pipe) != 0) {
1559 zperror(gettext("could not create STDIN pipe"));
1560 return (1);
1561 }
1562 /*
1563 * When the user types ^D, we get a zero length message on STDIN.
1564 * We need to echo that down the pipe to send it to the other side;
1565 * but by default, pipes don't propagate zero-length messages. We
1566 * toggle that behavior off using I_SWROPT. See streamio(7i).
1567 */
1568 if (ioctl(stdin_pipe[0], I_SWROPT, SNDZERO) != 0) {
1569 zperror(gettext("could not configure STDIN pipe"));
1570 return (1);
1571
1572 }
1573 if (pipe(stdout_pipe) != 0) {
1574 zperror(gettext("could not create STDOUT pipe"));
1575 return (1);
1576 }
1577 if (pipe(stderr_pipe) != 0) {
1578 zperror(gettext("could not create STDERR pipe"));
1579 return (1);
1580 }
1581
1582 if (pipe(dead_child_pipe) != 0) {
1583 zperror(gettext("could not create signalling pipe"));
1584 return (1);
1585 }
1586 close_on_sig = dead_child_pipe[0];
1587
1588 /*
1589 * If any of the pipe FD's winds up being less than STDERR, then we
1590 * have a mess on our hands-- and we are lacking some of the I/O
1591 * streams we would expect anyway. So we bail.
1592 */
1593 if (stdin_pipe[0] <= STDERR_FILENO ||
1594 stdin_pipe[1] <= STDERR_FILENO ||
1595 stdout_pipe[0] <= STDERR_FILENO ||
1596 stdout_pipe[1] <= STDERR_FILENO ||
1597 stderr_pipe[0] <= STDERR_FILENO ||
1598 stderr_pipe[1] <= STDERR_FILENO ||
1599 dead_child_pipe[0] <= STDERR_FILENO ||
1600 dead_child_pipe[1] <= STDERR_FILENO) {
1601 zperror(gettext("process lacks valid STDIN, STDOUT, STDERR"));
1602 return (1);
1603 }
1604
1605 if (prefork_dropprivs() != 0) {
1606 zperror(gettext("could not allocate privilege set"));
1607 return (1);
1608 }
1609
1610 (void) sigset(SIGCLD, sigcld);
1611 (void) sigemptyset(&block_cld);
1612 (void) sigaddset(&block_cld, SIGCLD);
1613 (void) sigprocmask(SIG_BLOCK, &block_cld, NULL);
1614
1615 if ((child_pid = fork()) == -1) {
1616 (void) ct_tmpl_clear(tmpl_fd);
1617 (void) close(tmpl_fd);
1618 zperror(gettext("could not fork"));
1619 return (1);
1620 } else if (child_pid == 0) { /* child process */
1621 (void) ct_tmpl_clear(tmpl_fd);
1622
1623 /*
1624 * Do a dance to get the pipes hooked up as FD's 0, 1 and 2.
1625 */
1626 (void) close(STDIN_FILENO);
1627 (void) close(STDOUT_FILENO);
1628 (void) close(STDERR_FILENO);
1629 (void) dup2(stdin_pipe[1], STDIN_FILENO);
1630 (void) dup2(stdout_pipe[1], STDOUT_FILENO);
1631 (void) dup2(stderr_pipe[1], STDERR_FILENO);
1632 (void) closefrom(STDERR_FILENO + 1);
1633
1634 (void) sigset(SIGCLD, SIG_DFL);
1635 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
1636 /*
1637 * In case any of stdin, stdout or stderr are streams,
1638 * anchor them to prevent malicious I_POPs.
1639 */
1640 (void) ioctl(STDIN_FILENO, I_ANCHOR);
1641 (void) ioctl(STDOUT_FILENO, I_ANCHOR);
1642 (void) ioctl(STDERR_FILENO, I_ANCHOR);
1643
1644 if (zone_enter(zoneid) == -1) {
1645 zerror(gettext("could not enter zone %s: %s"),
1646 zonename, strerror(errno));
1647 _exit(1);
1648 }
1649
1650 /*
1651 * For non-native zones, tell libc where it can find locale
1652 * specific getttext() messages.
1653 */
1654 if (access("/.SUNWnative/usr/lib/locale", R_OK) == 0)
1655 (void) bindtextdomain(TEXT_DOMAIN,
1656 "/.SUNWnative/usr/lib/locale");
1657 else if (access("/native/usr/lib/locale", R_OK) == 0)
1658 (void) bindtextdomain(TEXT_DOMAIN,
1659 "/native/usr/lib/locale");
1660
1661 if (!failsafe)
1662 new_env = prep_env_noninteractive(user_cmd, new_env);
1663
1664 if (new_env == NULL) {
1665 _exit(1);
1666 }
1667
1668 /*
1669 * Move into a new process group; the zone_enter will have
1670 * placed us into zsched's session, and we want to be in
1671 * a unique process group.
1672 */
1673 (void) setpgid(getpid(), getpid());
1674
1675 /*
1676 * The child needs to run as root to
1677 * execute the su program.
1678 */
1679 if (setuid(0) == -1) {
1680 zperror(gettext("insufficient privilege"));
1681 return (1);
1682 }
1683
1684 (void) execve(new_args[0], new_args, new_env);
1685 zperror(gettext("exec failure"));
1686 _exit(1);
1687 }
1688 /* parent */
1689
1690 /* close pipe sides written by child */
1691 (void) close(stdout_pipe[1]);
1692 (void) close(stderr_pipe[1]);
1693
1694 (void) sigset(SIGINT, sig_forward);
1695
1696 postfork_dropprivs();
1697
1698 (void) ct_tmpl_clear(tmpl_fd);
1699 (void) close(tmpl_fd);
1700
1701 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
1702 doio(stdin_pipe[0], stdin_pipe[1], stdout_pipe[0], stderr_pipe[0],
1703 dead_child_pipe[1], B_TRUE);
1704 do {
1705 retval = waitpid(child_pid, &child_status, 0);
1706 if (retval == -1) {
1707 child_status = 0;
1708 }
1709 } while (retval != child_pid && errno != ECHILD);
1710
1711 return (WEXITSTATUS(child_status));
1712 }
1713
1714 static char *
1715 get_username()
1716 {
1717 uid_t uid;
1718 struct passwd *nptr;
1719
1720 /*
1721 * Authorizations are checked to restrict access based on the
1722 * requested operation and zone name, It is assumed that the
1723 * program is running with all privileges, but that the real
1724 * user ID is that of the user or role on whose behalf we are
1725 * operating. So we start by getting the username that will be
1726 * used for subsequent authorization checks.
1727 */
1728
1729 uid = getuid();
1730 if ((nptr = getpwuid(uid)) == NULL) {
1731 zerror(gettext("could not get user name."));
1732 _exit(1);
1733 }
1734 return (nptr->pw_name);
1735 }
1736
1737 int
1738 main(int argc, char **argv)
1739 {
1740 int arg, console = 0;
1741 zoneid_t zoneid;
1742 zone_state_t st;
1743 char *login = "root";
1744 int lflag = 0;
1745 int nflag = 0;
1746 char *zonename = NULL;
1747 char **proc_args = NULL;
1748 char **new_args, **new_env;
1749 sigset_t block_cld;
1750 char devroot[MAXPATHLEN];
1751 char *slavename, slaveshortname[MAXPATHLEN];
1752 priv_set_t *privset;
1753 int tmpl_fd;
1754 char zonebrand[MAXNAMELEN];
1755 char default_brand[MAXNAMELEN];
1756 struct stat sb;
1757 char kernzone[ZONENAME_MAX];
1758 brand_handle_t bh;
1759 char user_cmd[MAXPATHLEN];
1760 char authname[MAXAUTHS];
1761
1762 (void) setlocale(LC_ALL, "");
1763 (void) textdomain(TEXT_DOMAIN);
1764
1765 (void) getpname(argv[0]);
1766 username = get_username();
1767
1768 while ((arg = getopt(argc, argv, "dnECR:Se:l:Q")) != EOF) {
1769 switch (arg) {
1770 case 'C':
1771 console = 1;
1772 break;
1773 case 'E':
1774 nocmdchar = 1;
1775 break;
1776 case 'R': /* undocumented */
1777 if (*optarg != '/') {
1778 zerror(gettext("root path must be absolute."));
1779 exit(2);
1780 }
1781 if (stat(optarg, &sb) == -1 || !S_ISDIR(sb.st_mode)) {
1782 zerror(
1783 gettext("root path must be a directory."));
1784 exit(2);
1785 }
1786 zonecfg_set_root(optarg);
1787 break;
1788 case 'Q':
1789 quiet = 1;
1790 break;
1791 case 'S':
1792 failsafe = 1;
1793 break;
1794 case 'd':
1795 disconnect = 1;
1796 break;
1797 case 'e':
1798 set_cmdchar(optarg);
1799 break;
1800 case 'l':
1801 login = optarg;
1802 lflag = 1;
1803 break;
1804 case 'n':
1805 nflag = 1;
1806 break;
1807 default:
1808 usage();
1809 }
1810 }
1811
1812 if (console != 0) {
1813
1814 if (lflag != 0) {
1815 zerror(gettext(
1816 "-l may not be specified for console login"));
1817 usage();
1818 }
1819
1820 if (nflag != 0) {
1821 zerror(gettext(
1822 "-n may not be specified for console login"));
1823 usage();
1824 }
1825
1826 if (failsafe != 0) {
1827 zerror(gettext(
1828 "-S may not be specified for console login"));
1829 usage();
1830 }
1831
1832 if (zonecfg_in_alt_root()) {
1833 zerror(gettext(
1834 "-R may not be specified for console login"));
1835 exit(2);
1836 }
1837
1838 }
1839
1840 if (failsafe != 0 && lflag != 0) {
1841 zerror(gettext("-l may not be specified for failsafe login"));
1842 usage();
1843 }
1844
1845 if (!console && disconnect != 0) {
1846 zerror(gettext(
1847 "-d may only be specified with console login"));
1848 usage();
1849 }
1850
1851 if (optind == (argc - 1)) {
1852 /*
1853 * zone name, no process name; this should be an interactive
1854 * as long as STDIN is really a tty.
1855 */
1856 if (nflag != 0) {
1857 zerror(gettext(
1858 "-n may not be specified for interactive login"));
1859 usage();
1860 }
1861 if (isatty(STDIN_FILENO))
1862 interactive = 1;
1863 zonename = argv[optind];
1864 } else if (optind < (argc - 1)) {
1865 if (console) {
1866 zerror(gettext("Commands may not be specified for "
1867 "console login."));
1868 usage();
1869 }
1870 /* zone name and process name, and possibly some args */
1871 zonename = argv[optind];
1872 proc_args = &argv[optind + 1];
1873 interactive = 0;
1874 } else {
1875 usage();
1876 }
1877
1878 if (getzoneid() != GLOBAL_ZONEID) {
1879 zerror(gettext("'%s' may only be used from the global zone"),
1880 pname);
1881 return (1);
1882 }
1883
1884 if (strcmp(zonename, GLOBAL_ZONENAME) == 0) {
1885 zerror(gettext("'%s' not applicable to the global zone"),
1886 pname);
1887 return (1);
1888 }
1889
1890 if (zone_get_state(zonename, &st) != Z_OK) {
1891 zerror(gettext("zone '%s' unknown"), zonename);
1892 return (1);
1893 }
1894
1895 if (st < ZONE_STATE_INSTALLED) {
1896 zerror(gettext("cannot login to a zone which is '%s'"),
1897 zone_state_str(st));
1898 return (1);
1899 }
1900
1901 /*
1902 * In both console and non-console cases, we require all privs.
1903 * In the console case, because we may need to startup zoneadmd.
1904 * In the non-console case in order to do zone_enter(2), zonept()
1905 * and other tasks.
1906 */
1907
1908 if ((privset = priv_allocset()) == NULL) {
1909 zperror(gettext("priv_allocset failed"));
1910 return (1);
1911 }
1912
1913 if (getppriv(PRIV_EFFECTIVE, privset) != 0) {
1914 zperror(gettext("getppriv failed"));
1915 priv_freeset(privset);
1916 return (1);
1917 }
1918
1919 if (priv_isfullset(privset) == B_FALSE) {
1920 zerror(gettext("You lack sufficient privilege to run "
1921 "this command (all privs required)"));
1922 priv_freeset(privset);
1923 return (1);
1924 }
1925 priv_freeset(privset);
1926
1927 /*
1928 * Check if user is authorized for requested usage of the zone
1929 */
1930
1931 (void) snprintf(authname, MAXAUTHS, "%s%s%s",
1932 ZONE_MANAGE_AUTH, KV_OBJECT, zonename);
1933 if (chkauthattr(authname, username) == 0) {
1934 if (console) {
1935 zerror(gettext("%s is not authorized for console "
1936 "access to %s zone."),
1937 username, zonename);
1938 return (1);
1939 } else {
1940 (void) snprintf(authname, MAXAUTHS, "%s%s%s",
1941 ZONE_LOGIN_AUTH, KV_OBJECT, zonename);
1942 if (failsafe || !interactive) {
1943 zerror(gettext("%s is not authorized for "
1944 "failsafe or non-interactive login "
1945 "to %s zone."), username, zonename);
1946 return (1);
1947 } else if (chkauthattr(authname, username) == 0) {
1948 zerror(gettext("%s is not authorized "
1949 " to login to %s zone."),
1950 username, zonename);
1951 return (1);
1952 }
1953 }
1954 } else {
1955 forced_login = B_TRUE;
1956 }
1957
1958 /*
1959 * The console is a separate case from the rest of the code; handle
1960 * it first.
1961 */
1962 if (console) {
1963 /*
1964 * Ensure that zoneadmd for this zone is running.
1965 */
1966 if (start_zoneadmd(zonename) == -1)
1967 return (1);
1968
1969 /*
1970 * Make contact with zoneadmd.
1971 */
1972 if (get_console_master(zonename) == -1)
1973 return (1);
1974
1975 if (!quiet)
1976 (void) printf(
1977 gettext("[Connected to zone '%s' console]\n"),
1978 zonename);
1979
1980 if (set_tty_rawmode(STDIN_FILENO) == -1) {
1981 reset_tty();
1982 zperror(gettext("failed to set stdin pty to raw mode"));
1983 return (1);
1984 }
1985
1986 (void) sigset(SIGWINCH, sigwinch);
1987 (void) sigwinch(0);
1988
1989 /*
1990 * Run the I/O loop until we get disconnected.
1991 */
1992 doio(masterfd, -1, masterfd, -1, -1, B_FALSE);
1993 reset_tty();
1994 if (!quiet)
1995 (void) printf(
1996 gettext("\n[Connection to zone '%s' console "
1997 "closed]\n"), zonename);
1998
1999 return (0);
2000 }
2001
2002 if (st != ZONE_STATE_RUNNING && st != ZONE_STATE_MOUNTED) {
2003 zerror(gettext("login allowed only to running zones "
2004 "(%s is '%s')."), zonename, zone_state_str(st));
2005 return (1);
2006 }
2007
2008 (void) strlcpy(kernzone, zonename, sizeof (kernzone));
2009 if (zonecfg_in_alt_root()) {
2010 FILE *fp = zonecfg_open_scratch("", B_FALSE);
2011
2012 if (fp == NULL || zonecfg_find_scratch(fp, zonename,
2013 zonecfg_get_root(), kernzone, sizeof (kernzone)) == -1) {
2014 zerror(gettext("cannot find scratch zone %s"),
2015 zonename);
2016 if (fp != NULL)
2017 zonecfg_close_scratch(fp);
2018 return (1);
2019 }
2020 zonecfg_close_scratch(fp);
2021 }
2022
2023 if ((zoneid = getzoneidbyname(kernzone)) == -1) {
2024 zerror(gettext("failed to get zoneid for zone '%s'"),
2025 zonename);
2026 return (1);
2027 }
2028
2029 /*
2030 * We need the zone root path only if we are setting up a pty.
2031 */
2032 if (zone_get_devroot(zonename, devroot, sizeof (devroot)) == -1) {
2033 zerror(gettext("could not get dev path for zone %s"),
2034 zonename);
2035 return (1);
2036 }
2037
2038 if (zone_get_brand(zonename, zonebrand, sizeof (zonebrand)) != Z_OK) {
2039 zerror(gettext("could not get brand for zone %s"), zonename);
2040 return (1);
2041 }
2042 /*
2043 * In the alternate root environment, the only supported
2044 * operations are mount and unmount. In this case, just treat
2045 * the zone as native if it is cluster. Cluster zones can be
2046 * native for the purpose of LU or upgrade, and the cluster
2047 * brand may not exist in the miniroot (such as in net install
2048 * upgrade).
2049 */
2050 if (zonecfg_default_brand(default_brand,
2051 sizeof (default_brand)) != Z_OK) {
2052 zerror(gettext("unable to determine default brand"));
2053 return (1);
2054 }
2055 if (zonecfg_in_alt_root() &&
2056 strcmp(zonebrand, CLUSTER_BRAND_NAME) == 0) {
2057 (void) strlcpy(zonebrand, default_brand, sizeof (zonebrand));
2058 }
2059
2060 if ((bh = brand_open(zonebrand)) == NULL) {
2061 zerror(gettext("could not open brand for zone %s"), zonename);
2062 return (1);
2063 }
2064
2065 if ((new_args = prep_args(bh, login, proc_args)) == NULL) {
2066 zperror(gettext("could not assemble new arguments"));
2067 brand_close(bh);
2068 return (1);
2069 }
2070 /*
2071 * Get the brand specific user_cmd. This command is used to get
2072 * a passwd(4) entry for login.
2073 */
2074 if (!interactive && !failsafe) {
2075 if (zone_get_user_cmd(bh, login, user_cmd,
2076 sizeof (user_cmd)) == NULL) {
2077 zerror(gettext("could not get user_cmd for zone %s"),
2078 zonename);
2079 brand_close(bh);
2080 return (1);
2081 }
2082 }
2083 brand_close(bh);
2084
2085 if ((new_env = prep_env()) == NULL) {
2086 zperror(gettext("could not assemble new environment"));
2087 return (1);
2088 }
2089
2090 if (!interactive) {
2091 if (nflag) {
2092 int nfd;
2093
2094 if ((nfd = open(_PATH_DEVNULL, O_RDONLY)) < 0) {
2095 zperror(gettext("failed to open null device"));
2096 return (1);
2097 }
2098 if (nfd != STDIN_FILENO) {
2099 if (dup2(nfd, STDIN_FILENO) < 0) {
2100 zperror(gettext(
2101 "failed to dup2 null device"));
2102 return (1);
2103 }
2104 (void) close(nfd);
2105 }
2106 /* /dev/null is now standard input */
2107 }
2108 return (noninteractive_login(zonename, user_cmd, zoneid,
2109 new_args, new_env));
2110 }
2111
2112 if (zonecfg_in_alt_root()) {
2113 zerror(gettext("cannot use interactive login with scratch "
2114 "zone"));
2115 return (1);
2116 }
2117
2118 /*
2119 * Things are more complex in interactive mode; we get the
2120 * master side of the pty, then place the user's terminal into
2121 * raw mode.
2122 */
2123 if (get_master_pty() == -1) {
2124 zerror(gettext("could not setup master pty device"));
2125 return (1);
2126 }
2127
2128 /*
2129 * Compute the "short name" of the pts. /dev/pts/2 --> pts/2
2130 */
2131 if ((slavename = ptsname(masterfd)) == NULL) {
2132 zperror(gettext("failed to get name for pseudo-tty"));
2133 return (1);
2134 }
2135 if (strncmp(slavename, "/dev/", strlen("/dev/")) == 0)
2136 (void) strlcpy(slaveshortname, slavename + strlen("/dev/"),
2137 sizeof (slaveshortname));
2138 else
2139 (void) strlcpy(slaveshortname, slavename,
2140 sizeof (slaveshortname));
2141
2142 if (!quiet)
2143 (void) printf(gettext("[Connected to zone '%s' %s]\n"),
2144 zonename, slaveshortname);
2145
2146 if (set_tty_rawmode(STDIN_FILENO) == -1) {
2147 reset_tty();
2148 zperror(gettext("failed to set stdin pty to raw mode"));
2149 return (1);
2150 }
2151
2152 if (prefork_dropprivs() != 0) {
2153 reset_tty();
2154 zperror(gettext("could not allocate privilege set"));
2155 return (1);
2156 }
2157
2158 /*
2159 * We must mask SIGCLD until after we have coped with the fork
2160 * sufficiently to deal with it; otherwise we can race and receive the
2161 * signal before child_pid has been initialized (yes, this really
2162 * happens).
2163 */
2164 (void) sigset(SIGCLD, sigcld);
2165 (void) sigemptyset(&block_cld);
2166 (void) sigaddset(&block_cld, SIGCLD);
2167 (void) sigprocmask(SIG_BLOCK, &block_cld, NULL);
2168
2169 /*
2170 * We activate the contract template at the last minute to
2171 * avoid intermediate functions that could be using fork(2)
2172 * internally.
2173 */
2174 if ((tmpl_fd = init_template()) == -1) {
2175 reset_tty();
2176 zperror(gettext("could not create contract"));
2177 return (1);
2178 }
2179
2180 if ((child_pid = fork()) == -1) {
2181 (void) ct_tmpl_clear(tmpl_fd);
2182 reset_tty();
2183 zperror(gettext("could not fork"));
2184 return (1);
2185 } else if (child_pid == 0) { /* child process */
2186 int slavefd, newslave;
2187
2188 (void) ct_tmpl_clear(tmpl_fd);
2189 (void) close(tmpl_fd);
2190
2191 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
2192
2193 if ((slavefd = init_slave_pty(zoneid, devroot)) == -1)
2194 return (1);
2195
2196 /*
2197 * Close all fds except for the slave pty.
2198 */
2199 (void) fdwalk(close_func, &slavefd);
2200
2201 /*
2202 * Temporarily dup slavefd to stderr; that way if we have
2203 * to print out that zone_enter failed, the output will
2204 * have somewhere to go.
2205 */
2206 if (slavefd != STDERR_FILENO)
2207 (void) dup2(slavefd, STDERR_FILENO);
2208
2209 if (zone_enter(zoneid) == -1) {
2210 zerror(gettext("could not enter zone %s: %s"),
2211 zonename, strerror(errno));
2212 return (1);
2213 }
2214
2215 if (slavefd != STDERR_FILENO)
2216 (void) close(STDERR_FILENO);
2217
2218 /*
2219 * We take pains to get this process into a new process
2220 * group, and subsequently a new session. In this way,
2221 * we'll have a session which doesn't yet have a controlling
2222 * terminal. When we open the slave, it will become the
2223 * controlling terminal; no PIDs concerning pgrps or sids
2224 * will leak inappropriately into the zone.
2225 */
2226 (void) setpgrp();
2227
2228 /*
2229 * We need the slave pty to be referenced from the zone's
2230 * /dev in order to ensure that the devt's, etc are all
2231 * correct. Otherwise we break ttyname and the like.
2232 */
2233 if ((newslave = open(slavename, O_RDWR)) == -1) {
2234 (void) close(slavefd);
2235 return (1);
2236 }
2237 (void) close(slavefd);
2238 slavefd = newslave;
2239
2240 /*
2241 * dup the slave to the various FDs, so that when the
2242 * spawned process does a write/read it maps to the slave
2243 * pty.
2244 */
2245 (void) dup2(slavefd, STDIN_FILENO);
2246 (void) dup2(slavefd, STDOUT_FILENO);
2247 (void) dup2(slavefd, STDERR_FILENO);
2248 if (slavefd != STDIN_FILENO && slavefd != STDOUT_FILENO &&
2249 slavefd != STDERR_FILENO) {
2250 (void) close(slavefd);
2251 }
2252
2253 /*
2254 * In failsafe mode, we don't use login(1), so don't try
2255 * setting up a utmpx entry.
2256 */
2257 if (!failsafe)
2258 if (setup_utmpx(slaveshortname) == -1)
2259 return (1);
2260
2261 /*
2262 * The child needs to run as root to
2263 * execute the brand's login program.
2264 */
2265 if (setuid(0) == -1) {
2266 zperror(gettext("insufficient privilege"));
2267 return (1);
2268 }
2269
2270 (void) execve(new_args[0], new_args, new_env);
2271 zperror(gettext("exec failure"));
2272 return (1);
2273 }
2274
2275 (void) ct_tmpl_clear(tmpl_fd);
2276 (void) close(tmpl_fd);
2277
2278 /*
2279 * The rest is only for the parent process.
2280 */
2281 (void) sigset(SIGWINCH, sigwinch);
2282
2283 postfork_dropprivs();
2284
2285 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
2286 doio(masterfd, -1, masterfd, -1, -1, B_FALSE);
2287
2288 reset_tty();
2289 if (!quiet)
2290 (void) fprintf(stderr,
2291 gettext("\n[Connection to zone '%s' %s closed]\n"),
2292 zonename, slaveshortname);
2293
2294 if (pollerr != 0) {
2295 (void) fprintf(stderr, gettext("Error: connection closed due "
2296 "to unexpected pollevents=0x%x.\n"), pollerr);
2297 return (1);
2298 }
2299
2300 return (0);
2301 }