2 * Copyright (C) 2009-2010 Red Hat Inc.
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #define _BSD_SOURCE /* for mkdtemp, usleep */
34 #include <sys/select.h>
38 #include <rpc/types.h>
45 #ifdef HAVE_SYS_TYPES_H
46 #include <sys/types.h>
49 #ifdef HAVE_SYS_WAIT_H
53 #ifdef HAVE_SYS_SOCKET_H
54 #include <sys/socket.h>
61 #include <arpa/inet.h>
62 #include <netinet/in.h>
64 #include "glthread/lock.h"
67 #include "guestfs-internal.h"
68 #include "guestfs-internal-actions.h"
69 #include "guestfs_protocol.h"
71 static int qemu_supports (guestfs_h *g, const char *option);
73 /* Add a string to the current command line. */
75 incr_cmdline_size (guestfs_h *g)
77 if (g->cmdline == NULL) {
78 /* g->cmdline[0] is reserved for argv[0], set in guestfs_launch. */
80 g->cmdline = safe_malloc (g, sizeof (char *));
85 g->cmdline = safe_realloc (g, g->cmdline, sizeof (char *) * g->cmdline_size);
89 add_cmdline (guestfs_h *g, const char *str)
91 if (g->state != CONFIG) {
93 _("command line cannot be altered after qemu subprocess launched"));
97 incr_cmdline_size (g);
98 g->cmdline[g->cmdline_size-1] = safe_strdup (g, str);
103 guestfs__config (guestfs_h *g,
104 const char *qemu_param, const char *qemu_value)
106 if (qemu_param[0] != '-') {
107 error (g, _("guestfs_config: parameter must begin with '-' character"));
111 /* A bit fascist, but the user will probably break the extra
112 * parameters that we add if they try to set any of these.
114 if (STREQ (qemu_param, "-kernel") ||
115 STREQ (qemu_param, "-initrd") ||
116 STREQ (qemu_param, "-nographic") ||
117 STREQ (qemu_param, "-serial") ||
118 STREQ (qemu_param, "-full-screen") ||
119 STREQ (qemu_param, "-std-vga") ||
120 STREQ (qemu_param, "-vnc")) {
121 error (g, _("guestfs_config: parameter '%s' isn't allowed"), qemu_param);
125 if (add_cmdline (g, qemu_param) != 0) return -1;
127 if (qemu_value != NULL) {
128 if (add_cmdline (g, qemu_value) != 0) return -1;
135 guestfs__add_drive_with_if (guestfs_h *g, const char *filename,
136 const char *drive_if)
138 size_t len = strlen (filename) + 64;
141 if (strchr (filename, ',') != NULL) {
142 error (g, _("filename cannot contain ',' (comma) character"));
146 /* cache=off improves reliability in the event of a host crash.
148 * However this option causes qemu to try to open the file with
149 * O_DIRECT. This fails on some filesystem types (notably tmpfs).
150 * So we check if we can open the file with or without O_DIRECT,
151 * and use cache=off (or not) accordingly.
153 * This test also checks for the presence of the file, which
154 * is a documented semantic of this interface.
156 int fd = open (filename, O_RDONLY|O_DIRECT);
159 snprintf (buf, len, "file=%s,cache=off,if=%s", filename, drive_if);
161 fd = open (filename, O_RDONLY);
164 snprintf (buf, len, "file=%s,if=%s", filename, drive_if);
166 perrorf (g, "%s", filename);
171 return guestfs__config (g, "-drive", buf);
175 guestfs__add_drive_ro_with_if (guestfs_h *g, const char *filename,
176 const char *drive_if)
178 if (strchr (filename, ',') != NULL) {
179 error (g, _("filename cannot contain ',' (comma) character"));
183 if (access (filename, F_OK) == -1) {
184 perrorf (g, "%s", filename);
188 size_t len = strlen (filename) + 64;
191 snprintf (buf, len, "file=%s,snapshot=on,if=%s", filename, drive_if);
193 return guestfs__config (g, "-drive", buf);
197 guestfs__add_drive (guestfs_h *g, const char *filename)
199 return guestfs__add_drive_with_if (g, filename, DRIVE_IF);
203 guestfs__add_drive_ro (guestfs_h *g, const char *filename)
205 return guestfs__add_drive_ro_with_if (g, filename, DRIVE_IF);
209 guestfs__add_cdrom (guestfs_h *g, const char *filename)
211 if (strchr (filename, ',') != NULL) {
212 error (g, _("filename cannot contain ',' (comma) character"));
216 if (access (filename, F_OK) == -1) {
217 perrorf (g, "%s", filename);
221 return guestfs__config (g, "-cdrom", filename);
224 static int is_openable (guestfs_h *g, const char *path, int flags);
225 static void print_cmdline (guestfs_h *g);
228 guestfs__launch (guestfs_h *g)
234 struct sockaddr_un addr;
238 error (g, _("you must call guestfs_add_drive before guestfs_launch"));
242 if (g->state != CONFIG) {
243 error (g, _("the libguestfs handle has already been launched"));
247 /* Start the clock ... */
248 gettimeofday (&g->launch_t, NULL);
250 /* Make the temporary directory. */
252 const char *tmpdir = guestfs___tmpdir ();
253 char dir_template[strlen (tmpdir) + 32];
254 sprintf (dir_template, "%s/libguestfsXXXXXX", tmpdir);
256 g->tmpdir = safe_strdup (g, dir_template);
257 if (mkdtemp (g->tmpdir) == NULL) {
258 perrorf (g, _("%s: cannot create temporary directory"), dir_template);
263 /* Allow anyone to read the temporary directory. The socket in this
264 * directory won't be readable but anyone can see it exists if they
265 * want. (RHBZ#610880).
267 if (chmod (g->tmpdir, 0755) == -1)
268 fprintf (stderr, "chmod: %s: %m (ignored)\n", g->tmpdir);
270 /* Locate and/or build the appliance. */
271 char *kernel = NULL, *initrd = NULL, *appliance = NULL;
272 if (guestfs___build_appliance (g, &kernel, &initrd, &appliance) == -1)
276 guestfs___print_timestamped_message (g, "begin testing qemu features");
278 /* Get qemu help text and version. */
279 if (qemu_supports (g, NULL) == -1)
282 /* Using virtio-serial, we need to create a local Unix domain socket
283 * for qemu to connect to.
285 snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
288 g->sock = socket (AF_UNIX, SOCK_STREAM, 0);
290 perrorf (g, "socket");
294 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
295 perrorf (g, "fcntl");
299 addr.sun_family = AF_UNIX;
300 strncpy (addr.sun_path, unixsock, UNIX_PATH_MAX);
301 addr.sun_path[UNIX_PATH_MAX-1] = '\0';
303 if (bind (g->sock, &addr, sizeof addr) == -1) {
308 if (listen (g->sock, 1) == -1) {
309 perrorf (g, "listen");
314 if (pipe (wfd) == -1 || pipe (rfd) == -1) {
321 guestfs___print_timestamped_message (g, "finished testing qemu features");
335 if (r == 0) { /* Child (qemu). */
338 /* Set up the full command line. Do this in the subprocess so we
339 * don't need to worry about cleaning up.
341 g->cmdline[0] = g->qemu;
343 if (qemu_supports (g, "-nodefconfig"))
344 add_cmdline (g, "-nodefconfig");
346 /* qemu sometimes needs this option to enable hardware
347 * virtualization, but some versions of 'qemu-kvm' will use KVM
348 * regardless (even where this option appears in the help text).
349 * It is rumoured that there are versions of qemu where supplying
350 * this option when hardware virtualization is not available will
351 * cause qemu to fail, so we we have to check at least that
352 * /dev/kvm is openable. That's not reliable, since /dev/kvm
353 * might be openable by qemu but not by us (think: SELinux) in
354 * which case the user would not get hardware virtualization,
355 * although at least shouldn't fail. A giant clusterfuck with the
356 * qemu command line, again.
358 if (qemu_supports (g, "-enable-kvm") &&
359 is_openable (g, "/dev/kvm", O_RDWR))
360 add_cmdline (g, "-enable-kvm");
362 /* Newer versions of qemu (from around 2009/12) changed the
363 * behaviour of monitors so that an implicit '-monitor stdio' is
364 * assumed if we are in -nographic mode and there is no other
365 * -monitor option. Only a single stdio device is allowed, so
366 * this broke the '-serial stdio' option. There is a new flag
367 * called -nodefaults which gets rid of all this default crud, so
368 * let's use that to avoid this and any future surprises.
370 if (qemu_supports (g, "-nodefaults"))
371 add_cmdline (g, "-nodefaults");
373 add_cmdline (g, "-nographic");
375 snprintf (buf, sizeof buf, "%d", g->memsize);
376 add_cmdline (g, "-m");
377 add_cmdline (g, buf);
379 /* Force exit instead of reboot on panic */
380 add_cmdline (g, "-no-reboot");
382 /* These options recommended by KVM developers to improve reliability. */
383 if (qemu_supports (g, "-no-hpet"))
384 add_cmdline (g, "-no-hpet");
386 if (qemu_supports (g, "-rtc-td-hack"))
387 add_cmdline (g, "-rtc-td-hack");
389 /* Create the virtio serial bus. */
390 add_cmdline (g, "-device");
391 add_cmdline (g, "virtio-serial");
394 /* Use virtio-console (a variant form of virtio-serial) for the
395 * guest's serial console.
397 add_cmdline (g, "-chardev");
398 add_cmdline (g, "stdio,id=console");
399 add_cmdline (g, "-device");
400 add_cmdline (g, "virtconsole,chardev=console,name=org.libguestfs.console.0");
402 /* When the above works ... until then: */
403 add_cmdline (g, "-serial");
404 add_cmdline (g, "stdio");
407 /* Set up virtio-serial for the communications channel. */
408 add_cmdline (g, "-chardev");
409 snprintf (buf, sizeof buf, "socket,path=%s,id=channel0", unixsock);
410 add_cmdline (g, buf);
411 add_cmdline (g, "-device");
412 add_cmdline (g, "virtserialport,chardev=channel0,name=org.libguestfs.channel.0");
414 /* Enable user networking. */
415 if (g->enable_network) {
416 add_cmdline (g, "-netdev");
417 add_cmdline (g, "user,id=usernet");
418 add_cmdline (g, "-device");
419 add_cmdline (g, NET_IF ",netdev=usernet");
422 #define LINUX_CMDLINE \
423 "panic=1 " /* force kernel to panic if daemon exits */ \
424 "console=ttyS0 " /* serial console */ \
425 "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \
426 "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \
427 "acpi=off " /* we don't need ACPI, turn it off */ \
428 "printk.time=1 " /* display timestamp before kernel messages */ \
429 "cgroup_disable=memory " /* saves us about 5 MB of RAM */
431 /* Linux kernel command line. */
432 snprintf (buf, sizeof buf,
434 "%s " /* (selinux) */
435 "%s " /* (verbose) */
436 "TERM=%s " /* (TERM environment variable) */
438 g->selinux ? "selinux=1 enforcing=0" : "selinux=0",
439 g->verbose ? "guestfs_verbose=1" : "",
440 getenv ("TERM") ? : "linux",
441 g->append ? g->append : "");
443 add_cmdline (g, "-kernel");
444 add_cmdline (g, kernel);
445 add_cmdline (g, "-initrd");
446 add_cmdline (g, initrd);
447 add_cmdline (g, "-append");
448 add_cmdline (g, buf);
450 /* Add the ext2 appliance drive (last of all). */
452 const char *cachemode = "";
453 if (qemu_supports (g, "cache=")) {
454 if (qemu_supports (g, "unsafe"))
455 cachemode = ",cache=unsafe";
456 else if (qemu_supports (g, "writeback"))
457 cachemode = ",cache=writeback";
460 char buf2[PATH_MAX + 64];
461 add_cmdline (g, "-drive");
462 snprintf (buf2, sizeof buf2, "file=%s,snapshot=on,if=" DRIVE_IF "%s",
463 appliance, cachemode);
464 add_cmdline (g, buf2);
467 /* Finish off the command line. */
468 incr_cmdline_size (g);
469 g->cmdline[g->cmdline_size-1] = NULL;
475 /* Set up stdin, stdout. */
481 if (dup (wfd[0]) == -1) {
483 perror ("dup failed");
484 _exit (EXIT_FAILURE);
486 if (dup (rfd[1]) == -1)
494 /* Set up a new process group, so we can signal this process
495 * and all subprocesses (eg. if qemu is really a shell script).
500 setenv ("LC_ALL", "C", 1);
502 execv (g->qemu, g->cmdline); /* Run qemu. */
504 _exit (EXIT_FAILURE);
507 /* Parent (library). */
515 /* Fork the recovery process off which will kill qemu if the parent
516 * process fails to do so (eg. if the parent segfaults).
519 if (g->recovery_proc) {
522 pid_t qemu_pid = g->pid;
523 pid_t parent_pid = getppid ();
525 /* Writing to argv is hideously complicated and error prone. See:
526 * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain
529 /* Loop around waiting for one or both of the other processes to
530 * disappear. It's fair to say this is very hairy. The PIDs that
531 * we are looking at might be reused by another process. We are
532 * effectively polling. Is the cure worse than the disease?
535 if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */
536 _exit (EXIT_SUCCESS);
537 if (kill (parent_pid, 0) == -1) {
538 /* Parent's gone away, qemu still around, so kill qemu. */
540 _exit (EXIT_SUCCESS);
546 /* Don't worry, if the fork failed, this will be -1. The recovery
547 * process isn't essential.
553 /* Close the other ends of the pipe. */
557 if (fcntl (wfd[1], F_SETFL, O_NONBLOCK) == -1 ||
558 fcntl (rfd[0], F_SETFL, O_NONBLOCK) == -1) {
559 perrorf (g, "fcntl");
563 g->fd[0] = wfd[1]; /* stdin of child */
564 g->fd[1] = rfd[0]; /* stdout of child */
566 g->fd[0] = open ("/dev/null", O_RDWR);
567 if (g->fd[0] == -1) {
568 perrorf (g, "open /dev/null");
571 g->fd[1] = dup (g->fd[0]);
572 if (g->fd[1] == -1) {
579 g->state = LAUNCHING;
581 /* Wait for qemu to start and to connect back to us via
582 * virtio-serial and send the GUESTFS_LAUNCH_FLAG message.
584 r = guestfs___accept_from_daemon (g);
588 close (g->sock); /* Close the listening socket. */
589 g->sock = r; /* This is the accepted data socket. */
591 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
592 perrorf (g, "fcntl");
598 r = guestfs___recv_from_daemon (g, &size, &buf);
601 if (r == -1) return -1;
603 if (size != GUESTFS_LAUNCH_FLAG) {
604 error (g, _("guestfs_launch failed, see earlier error messages"));
609 guestfs___print_timestamped_message (g, "appliance is up");
611 /* This is possible in some really strange situations, such as
612 * guestfsd starts up OK but then qemu immediately exits. Check for
613 * it because the caller is probably expecting to be able to send
614 * commands after this function returns.
616 if (g->state != READY) {
617 error (g, _("qemu launched and contacted daemon, but state != READY"));
628 if (g->pid > 0) kill (g->pid, 9);
629 if (g->recoverypid > 0) kill (g->recoverypid, 9);
630 waitpid (g->pid, NULL, 0);
631 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
636 memset (&g->launch_t, 0, sizeof g->launch_t);
651 guestfs___tmpdir (void)
661 const char *t = getenv ("TMPDIR");
667 /* This function is used to print the qemu command line before it gets
668 * executed, when in verbose mode.
671 print_cmdline (guestfs_h *g)
676 while (g->cmdline[i]) {
677 if (g->cmdline[i][0] == '-') /* -option starts a new line */
678 fprintf (stderr, " \\\n ");
680 if (i > 0) fputc (' ', stderr);
682 /* Does it need shell quoting? This only deals with simple cases. */
683 needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]);
685 if (needs_quote) fputc ('\'', stderr);
686 fprintf (stderr, "%s", g->cmdline[i]);
687 if (needs_quote) fputc ('\'', stderr);
691 fputc ('\n', stderr);
694 /* Compute Y - X and return the result in milliseconds.
695 * Approximately the same as this code:
696 * http://www.mpp.mpg.de/~huber/util/timevaldiff.c
699 timeval_diff (const struct timeval *x, const struct timeval *y)
703 msec = (y->tv_sec - x->tv_sec) * 1000;
704 msec += (y->tv_usec - x->tv_usec) / 1000;
709 guestfs___print_timestamped_message (guestfs_h *g, const char *fs, ...)
717 err = vasprintf (&msg, fs, args);
722 gettimeofday (&tv, NULL);
724 fprintf (stderr, "[%05" PRIi64 "ms] %s\n",
725 timeval_diff (&g->launch_t, &tv), msg);
730 static int read_all (guestfs_h *g, FILE *fp, char **ret);
732 /* Test qemu binary (or wrapper) runs, and do 'qemu -help' and
733 * 'qemu -version' so we know what options this qemu supports and
737 test_qemu (guestfs_h *g)
742 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -nographic -help", g->qemu);
744 fp = popen (cmd, "r");
745 /* qemu -help should always work (qemu -version OTOH wasn't
746 * supported by qemu 0.9). If this command doesn't work then it
747 * probably indicates that the qemu binary is missing.
750 /* XXX This error is never printed, even if the qemu binary
751 * doesn't exist. Why?
754 perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd);
758 if (read_all (g, fp, &g->qemu_help) == -1)
761 if (pclose (fp) == -1)
764 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -nographic -version 2>/dev/null",
767 fp = popen (cmd, "r");
769 /* Intentionally ignore errors. */
770 read_all (g, fp, &g->qemu_version);
778 read_all (guestfs_h *g, FILE *fp, char **ret)
785 *ret = safe_realloc (g, *ret, n + 1);
790 *ret = safe_realloc (g, *ret, n + BUFSIZ);
792 r = fread (p, 1, BUFSIZ, fp);
801 /* Test if option is supported by qemu command line (just by grepping
804 * The first time this is used, it has to run the external qemu
805 * binary. If that fails, it returns -1.
807 * To just do the first-time run of the qemu binary, call this with
808 * option == NULL, in which case it will return -1 if there was an
812 qemu_supports (guestfs_h *g, const char *option)
815 if (test_qemu (g) == -1)
822 return strstr (g->qemu_help, option) != NULL;
825 /* Check if a file can be opened. */
827 is_openable (guestfs_h *g, const char *path, int flags)
829 int fd = open (path, flags);
839 /* You had to call this function after launch in versions <= 1.0.70,
840 * but it is now a no-op.
843 guestfs__wait_ready (guestfs_h *g)
845 if (g->state != READY) {
846 error (g, _("qemu has not been launched yet"));
854 guestfs__kill_subprocess (guestfs_h *g)
856 if (g->state == CONFIG) {
857 error (g, _("no subprocess to kill"));
862 fprintf (stderr, "sending SIGTERM to process %d\n", g->pid);
864 if (g->pid > 0) kill (g->pid, SIGTERM);
865 if (g->recoverypid > 0) kill (g->recoverypid, 9);
870 /* Access current state. */
872 guestfs__is_config (guestfs_h *g)
874 return g->state == CONFIG;
878 guestfs__is_launching (guestfs_h *g)
880 return g->state == LAUNCHING;
884 guestfs__is_ready (guestfs_h *g)
886 return g->state == READY;
890 guestfs__is_busy (guestfs_h *g)
892 return g->state == BUSY;
896 guestfs__get_state (guestfs_h *g)