2 * Copyright (C) 2009 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>
37 #include <rpc/types.h>
44 #ifdef HAVE_SYS_TYPES_H
45 #include <sys/types.h>
48 #ifdef HAVE_SYS_WAIT_H
52 #ifdef HAVE_SYS_SOCKET_H
53 #include <sys/socket.h>
60 #include <arpa/inet.h>
61 #include <netinet/in.h>
64 #include "glthread/lock.h"
65 #include "ignore-value.h"
68 #include "guestfs-internal.h"
69 #include "guestfs-internal-actions.h"
70 #include "guestfs_protocol.h"
74 #define _(str) dgettext(PACKAGE, (str))
75 //#define N_(str) dgettext(PACKAGE, (str))
81 #define error guestfs_error
82 #define perrorf guestfs_perrorf
83 #define safe_malloc guestfs_safe_malloc
84 #define safe_realloc guestfs_safe_realloc
85 #define safe_strdup guestfs_safe_strdup
86 //#define safe_memdup guestfs_safe_memdup
88 static void default_error_cb (guestfs_h *g, void *data, const char *msg);
89 static int send_to_daemon (guestfs_h *g, const void *v_buf, size_t n);
90 static int recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn);
91 static int accept_from_daemon (guestfs_h *g);
92 static int check_peer_euid (guestfs_h *g, int sock, uid_t *rtn);
93 static void close_handles (void);
95 #define UNIX_PATH_MAX 108
97 /* Also in guestfsd.c */
98 #define GUESTFWD_ADDR "10.0.2.4"
99 #define GUESTFWD_PORT "6666"
101 /* GuestFS handle and connection. */
102 enum state { CONFIG, LAUNCHING, READY, BUSY, NO_HANDLE };
106 struct guestfs_h *next; /* Linked list of open handles. */
108 /* State: see the state machine diagram in the man page guestfs(3). */
111 int fd[2]; /* Stdin/stdout of qemu. */
112 int sock; /* Daemon communications socket. */
113 pid_t pid; /* Qemu PID. */
114 pid_t recoverypid; /* Recovery process PID. */
116 struct timeval launch_t; /* The time that we called guestfs_launch. */
118 char *tmpdir; /* Temporary directory containing socket. */
120 char *qemu_help, *qemu_version; /* Output of qemu -help, qemu -version. */
122 char **cmdline; /* Qemu command line. */
131 char *path; /* Path to kernel, initrd. */
132 char *qemu; /* Qemu binary. */
133 char *append; /* Append to kernel command line. */
135 int memsize; /* Size of RAM (megabytes). */
137 int selinux; /* selinux enabled? */
142 guestfs_abort_cb abort_cb;
143 guestfs_error_handler_cb error_cb;
144 void * error_cb_data;
145 guestfs_log_message_cb log_message_cb;
146 void * log_message_cb_data;
147 guestfs_subprocess_quit_cb subprocess_quit_cb;
148 void * subprocess_quit_cb_data;
149 guestfs_launch_done_cb launch_done_cb;
150 void * launch_done_cb_data;
155 gl_lock_define_initialized (static, handles_lock);
156 static guestfs_h *handles = NULL;
157 static int atexit_handler_set = 0;
160 guestfs_create (void)
165 g = malloc (sizeof (*g));
168 memset (g, 0, sizeof (*g));
177 g->error_cb = default_error_cb;
178 g->error_cb_data = NULL;
180 g->recovery_proc = 1;
182 str = getenv ("LIBGUESTFS_DEBUG");
183 g->verbose = str != NULL && STREQ (str, "1");
185 str = getenv ("LIBGUESTFS_TRACE");
186 g->trace = str != NULL && STREQ (str, "1");
188 str = getenv ("LIBGUESTFS_PATH");
189 g->path = str != NULL ? strdup (str) : strdup (GUESTFS_DEFAULT_PATH);
190 if (!g->path) goto error;
192 str = getenv ("LIBGUESTFS_QEMU");
193 g->qemu = str != NULL ? strdup (str) : strdup (QEMU);
194 if (!g->qemu) goto error;
196 str = getenv ("LIBGUESTFS_APPEND");
198 g->append = strdup (str);
199 if (!g->append) goto error;
202 /* Choose a suitable memory size. Previously we tried to choose
203 * a minimal memory size, but this isn't really necessary since
204 * recent QEMU and KVM don't do anything nasty like locking
205 * memory into core any more. Thus we can safely choose a
206 * large, generous amount of memory, and it'll just get swapped
207 * on smaller systems.
209 str = getenv ("LIBGUESTFS_MEMSIZE");
211 if (sscanf (str, "%d", &g->memsize) != 1 || g->memsize <= 256) {
212 fprintf (stderr, "libguestfs: non-numeric or too small value for LIBGUESTFS_MEMSIZE\n");
218 /* Start with large serial numbers so they are easy to spot
219 * inside the protocol.
221 g->msg_next_serial = 0x00123400;
223 /* Link the handles onto a global list. */
224 gl_lock_lock (handles_lock);
227 if (!atexit_handler_set) {
228 atexit (close_handles);
229 atexit_handler_set = 1;
231 gl_lock_unlock (handles_lock);
234 fprintf (stderr, "new guestfs handle %p\n", g);
247 guestfs_close (guestfs_h *g)
253 if (g->state == NO_HANDLE) {
254 /* Not safe to call 'error' here, so ... */
255 fprintf (stderr, _("guestfs_close: called twice on the same handle\n"));
260 fprintf (stderr, "closing guestfs handle %p (state %d)\n", g, g->state);
262 /* Try to sync if autosync flag is set. */
263 if (g->autosync && g->state == READY) {
264 guestfs_umount_all (g);
268 /* Remove any handlers that might be called back before we kill the
271 g->log_message_cb = NULL;
273 if (g->state != CONFIG)
274 guestfs_kill_subprocess (g);
287 /* Wait for subprocess(es) to exit. */
288 waitpid (g->pid, NULL, 0);
289 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
291 /* Remove tmpfiles. */
293 snprintf (filename, sizeof filename, "%s/sock", g->tmpdir);
296 snprintf (filename, sizeof filename, "%s/initrd", g->tmpdir);
299 snprintf (filename, sizeof filename, "%s/kernel", g->tmpdir);
308 for (i = 0; i < g->cmdline_size; ++i)
309 free (g->cmdline[i]);
313 /* Mark the handle as dead before freeing it. */
314 g->state = NO_HANDLE;
316 gl_lock_lock (handles_lock);
320 for (gg = handles; gg->next != g; gg = gg->next)
324 gl_lock_unlock (handles_lock);
326 free (g->last_error);
331 free (g->qemu_version);
335 /* Close all open handles (called from atexit(3)). */
339 while (handles) guestfs_close (handles);
343 guestfs_last_error (guestfs_h *g)
345 return g->last_error;
349 set_last_error (guestfs_h *g, const char *msg)
351 free (g->last_error);
352 g->last_error = strdup (msg);
356 default_error_cb (guestfs_h *g, void *data, const char *msg)
358 fprintf (stderr, _("libguestfs: error: %s\n"), msg);
362 guestfs_error (guestfs_h *g, const char *fs, ...)
368 int err = vasprintf (&msg, fs, args);
373 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
374 set_last_error (g, msg);
380 guestfs_perrorf (guestfs_h *g, const char *fs, ...)
387 int err = vasprintf (&msg, fs, args);
394 strerror_r (errnum, buf, sizeof buf);
398 buf = strerror_r (errnum, _buf, sizeof _buf);
401 msg = safe_realloc (g, msg, strlen (msg) + 2 + strlen (buf) + 1);
405 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
406 set_last_error (g, msg);
412 guestfs_safe_malloc (guestfs_h *g, size_t nbytes)
414 void *ptr = malloc (nbytes);
415 if (nbytes > 0 && !ptr) g->abort_cb ();
419 /* Return 1 if an array of N objects, each of size S, cannot exist due
420 to size arithmetic overflow. S must be positive and N must be
421 nonnegative. This is a macro, not an inline function, so that it
422 works correctly even when SIZE_MAX < N.
424 By gnulib convention, SIZE_MAX represents overflow in size
425 calculations, so the conservative dividend to use here is
426 SIZE_MAX - 1, since SIZE_MAX might represent an overflowed value.
427 However, malloc (SIZE_MAX) fails on all known hosts where
428 sizeof (ptrdiff_t) <= sizeof (size_t), so do not bother to test for
429 exactly-SIZE_MAX allocations on such hosts; this avoids a test and
430 branch when S is known to be 1. */
431 # define xalloc_oversized(n, s) \
432 ((size_t) (sizeof (ptrdiff_t) <= sizeof (size_t) ? -1 : -2) / (s) < (n))
434 /* Technically we should add an autoconf test for this, testing for the desired
435 functionality, like what's done in gnulib, but for now, this is fine. */
436 #define HAVE_GNU_CALLOC (__GLIBC__ >= 2)
438 /* Allocate zeroed memory for N elements of S bytes, with error
439 checking. S must be nonzero. */
441 guestfs_safe_calloc (guestfs_h *g, size_t n, size_t s)
443 /* From gnulib's calloc function in xmalloc.c. */
445 /* Test for overflow, since some calloc implementations don't have
446 proper overflow checks. But omit overflow and size-zero tests if
447 HAVE_GNU_CALLOC, since GNU calloc catches overflow and never
448 returns NULL if successful. */
449 if ((! HAVE_GNU_CALLOC && xalloc_oversized (n, s))
450 || (! (p = calloc (n, s)) && (HAVE_GNU_CALLOC || n != 0)))
456 guestfs_safe_realloc (guestfs_h *g, void *ptr, int nbytes)
458 void *p = realloc (ptr, nbytes);
459 if (nbytes > 0 && !p) g->abort_cb ();
464 guestfs_safe_strdup (guestfs_h *g, const char *str)
466 char *s = strdup (str);
467 if (!s) g->abort_cb ();
472 guestfs_safe_memdup (guestfs_h *g, void *ptr, size_t size)
474 void *p = malloc (size);
475 if (!p) g->abort_cb ();
476 memcpy (p, ptr, size);
481 xwrite (int fd, const void *v_buf, size_t len)
483 const char *buf = v_buf;
487 r = write (fd, buf, len);
499 guestfs_set_out_of_memory_handler (guestfs_h *g, guestfs_abort_cb cb)
505 guestfs_get_out_of_memory_handler (guestfs_h *g)
511 guestfs_set_error_handler (guestfs_h *g, guestfs_error_handler_cb cb, void *data)
514 g->error_cb_data = data;
517 guestfs_error_handler_cb
518 guestfs_get_error_handler (guestfs_h *g, void **data_rtn)
520 if (data_rtn) *data_rtn = g->error_cb_data;
525 guestfs__set_verbose (guestfs_h *g, int v)
532 guestfs__get_verbose (guestfs_h *g)
538 guestfs__set_autosync (guestfs_h *g, int a)
545 guestfs__get_autosync (guestfs_h *g)
551 guestfs__set_path (guestfs_h *g, const char *path)
558 safe_strdup (g, GUESTFS_DEFAULT_PATH) : safe_strdup (g, path);
563 guestfs__get_path (guestfs_h *g)
569 guestfs__set_qemu (guestfs_h *g, const char *qemu)
574 g->qemu = qemu == NULL ? safe_strdup (g, QEMU) : safe_strdup (g, qemu);
579 guestfs__get_qemu (guestfs_h *g)
585 guestfs__set_append (guestfs_h *g, const char *append)
590 g->append = append ? safe_strdup (g, append) : NULL;
595 guestfs__get_append (guestfs_h *g)
601 guestfs__set_memsize (guestfs_h *g, int memsize)
603 g->memsize = memsize;
608 guestfs__get_memsize (guestfs_h *g)
614 guestfs__set_selinux (guestfs_h *g, int selinux)
616 g->selinux = selinux;
621 guestfs__get_selinux (guestfs_h *g)
627 guestfs__get_pid (guestfs_h *g)
632 error (g, "get_pid: no qemu subprocess");
637 struct guestfs_version *
638 guestfs__version (guestfs_h *g)
640 struct guestfs_version *r;
642 r = safe_malloc (g, sizeof *r);
643 r->major = PACKAGE_VERSION_MAJOR;
644 r->minor = PACKAGE_VERSION_MINOR;
645 r->release = PACKAGE_VERSION_RELEASE;
646 r->extra = safe_strdup (g, PACKAGE_VERSION_EXTRA);
651 guestfs__set_trace (guestfs_h *g, int t)
658 guestfs__get_trace (guestfs_h *g)
664 guestfs__set_direct (guestfs_h *g, int d)
671 guestfs__get_direct (guestfs_h *g)
677 guestfs__set_recovery_proc (guestfs_h *g, int f)
679 g->recovery_proc = !!f;
684 guestfs__get_recovery_proc (guestfs_h *g)
686 return g->recovery_proc;
689 /* Add a string to the current command line. */
691 incr_cmdline_size (guestfs_h *g)
693 if (g->cmdline == NULL) {
694 /* g->cmdline[0] is reserved for argv[0], set in guestfs_launch. */
696 g->cmdline = safe_malloc (g, sizeof (char *));
697 g->cmdline[0] = NULL;
701 g->cmdline = safe_realloc (g, g->cmdline, sizeof (char *) * g->cmdline_size);
705 add_cmdline (guestfs_h *g, const char *str)
707 if (g->state != CONFIG) {
709 _("command line cannot be altered after qemu subprocess launched"));
713 incr_cmdline_size (g);
714 g->cmdline[g->cmdline_size-1] = safe_strdup (g, str);
719 guestfs__config (guestfs_h *g,
720 const char *qemu_param, const char *qemu_value)
722 if (qemu_param[0] != '-') {
723 error (g, _("guestfs_config: parameter must begin with '-' character"));
727 /* A bit fascist, but the user will probably break the extra
728 * parameters that we add if they try to set any of these.
730 if (STREQ (qemu_param, "-kernel") ||
731 STREQ (qemu_param, "-initrd") ||
732 STREQ (qemu_param, "-nographic") ||
733 STREQ (qemu_param, "-serial") ||
734 STREQ (qemu_param, "-full-screen") ||
735 STREQ (qemu_param, "-std-vga") ||
736 STREQ (qemu_param, "-vnc")) {
737 error (g, _("guestfs_config: parameter '%s' isn't allowed"), qemu_param);
741 if (add_cmdline (g, qemu_param) != 0) return -1;
743 if (qemu_value != NULL) {
744 if (add_cmdline (g, qemu_value) != 0) return -1;
751 guestfs__add_drive (guestfs_h *g, const char *filename)
753 size_t len = strlen (filename) + 64;
756 if (strchr (filename, ',') != NULL) {
757 error (g, _("filename cannot contain ',' (comma) character"));
761 /* cache=off improves reliability in the event of a host crash.
763 * However this option causes qemu to try to open the file with
764 * O_DIRECT. This fails on some filesystem types (notably tmpfs).
765 * So we check if we can open the file with or without O_DIRECT,
766 * and use cache=off (or not) accordingly.
768 * This test also checks for the presence of the file, which
769 * is a documented semantic of this interface.
771 int fd = open (filename, O_RDONLY|O_DIRECT);
774 snprintf (buf, len, "file=%s,cache=off,if=" DRIVE_IF, filename);
776 fd = open (filename, O_RDONLY);
779 snprintf (buf, len, "file=%s,if=" DRIVE_IF, filename);
781 perrorf (g, "%s", filename);
786 return guestfs__config (g, "-drive", buf);
790 guestfs__add_drive_ro (guestfs_h *g, const char *filename)
792 size_t len = strlen (filename) + 64;
795 if (strchr (filename, ',') != NULL) {
796 error (g, _("filename cannot contain ',' (comma) character"));
800 if (access (filename, F_OK) == -1) {
801 perrorf (g, "%s", filename);
805 snprintf (buf, len, "file=%s,snapshot=on,if=%s", filename, DRIVE_IF);
807 return guestfs__config (g, "-drive", buf);
811 guestfs__add_cdrom (guestfs_h *g, const char *filename)
813 if (strchr (filename, ',') != NULL) {
814 error (g, _("filename cannot contain ',' (comma) character"));
818 if (access (filename, F_OK) == -1) {
819 perrorf (g, "%s", filename);
823 return guestfs__config (g, "-cdrom", filename);
826 /* Returns true iff file is contained in dir. */
828 dir_contains_file (const char *dir, const char *file)
830 int dirlen = strlen (dir);
831 int filelen = strlen (file);
832 int len = dirlen+filelen+2;
835 snprintf (path, len, "%s/%s", dir, file);
836 return access (path, F_OK) == 0;
839 /* Returns true iff every listed file is contained in 'dir'. */
841 dir_contains_files (const char *dir, ...)
846 va_start (args, dir);
847 while ((file = va_arg (args, const char *)) != NULL) {
848 if (!dir_contains_file (dir, file)) {
857 static void print_timestamped_message (guestfs_h *g, const char *fs, ...);
858 static int build_supermin_appliance (guestfs_h *g, const char *path, char **kernel, char **initrd);
859 static int test_qemu (guestfs_h *g);
860 static int qemu_supports (guestfs_h *g, const char *option);
861 static void print_cmdline (guestfs_h *g);
863 static const char *kernel_name = "vmlinuz." REPO "." host_cpu;
864 static const char *initrd_name = "initramfs." REPO "." host_cpu ".img";
865 static const char *supermin_name =
866 "initramfs." REPO "." host_cpu ".supermin.img";
867 static const char *supermin_hostfiles_name =
868 "initramfs." REPO "." host_cpu ".supermin.hostfiles";
871 guestfs__launch (guestfs_h *g)
874 char dir_template[PATH_MAX];
879 char *path, *pelem, *pend;
880 char *kernel = NULL, *initrd = NULL;
881 int null_vmchannel_sock;
883 struct sockaddr_un addr;
885 /* Start the clock ... */
886 gettimeofday (&g->launch_t, NULL);
894 tmpdir = getenv ("TMPDIR") ? : tmpdir;
895 snprintf (dir_template, sizeof dir_template, "%s/libguestfsXXXXXX", tmpdir);
899 error (g, _("you must call guestfs_add_drive before guestfs_launch"));
903 if (g->state != CONFIG) {
904 error (g, _("qemu has already been launched"));
908 /* Make the temporary directory. */
910 g->tmpdir = safe_strdup (g, dir_template);
911 if (mkdtemp (g->tmpdir) == NULL) {
912 perrorf (g, _("%s: cannot create temporary directory"), dir_template);
917 /* First search g->path for the supermin appliance, and try to
918 * synthesize a kernel and initrd from that. If it fails, we
919 * try the path search again looking for a backup ordinary
922 pelem = path = safe_strdup (g, g->path);
924 pend = strchrnul (pelem, ':');
925 pmore = *pend == ':';
929 /* Empty element of "." means cwd. */
930 if (len == 0 || (len == 1 && *pelem == '.')) {
933 "looking for supermin appliance in current directory\n");
934 if (dir_contains_files (".",
935 supermin_name, supermin_hostfiles_name,
936 "kmod.whitelist", NULL)) {
937 if (build_supermin_appliance (g, ".", &kernel, &initrd) == -1)
942 /* Look at <path>/supermin* etc. */
945 fprintf (stderr, "looking for supermin appliance in %s\n", pelem);
947 if (dir_contains_files (pelem,
948 supermin_name, supermin_hostfiles_name,
949 "kmod.whitelist", NULL)) {
950 if (build_supermin_appliance (g, pelem, &kernel, &initrd) == -1)
961 if (kernel == NULL || initrd == NULL) {
962 /* Search g->path for the kernel and initrd. */
963 pelem = path = safe_strdup (g, g->path);
965 pend = strchrnul (pelem, ':');
966 pmore = *pend == ':';
970 /* Empty element or "." means cwd. */
971 if (len == 0 || (len == 1 && *pelem == '.')) {
974 "looking for appliance in current directory\n");
975 if (dir_contains_files (".", kernel_name, initrd_name, NULL)) {
976 kernel = safe_strdup (g, kernel_name);
977 initrd = safe_strdup (g, initrd_name);
981 /* Look at <path>/kernel etc. */
984 fprintf (stderr, "looking for appliance in %s\n", pelem);
986 if (dir_contains_files (pelem, kernel_name, initrd_name, NULL)) {
987 kernel = safe_malloc (g, len + strlen (kernel_name) + 2);
988 initrd = safe_malloc (g, len + strlen (initrd_name) + 2);
989 sprintf (kernel, "%s/%s", pelem, kernel_name);
990 sprintf (initrd, "%s/%s", pelem, initrd_name);
1001 if (kernel == NULL || initrd == NULL) {
1002 error (g, _("cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)"),
1003 kernel_name, initrd_name, g->path);
1008 print_timestamped_message (g, "begin testing qemu features");
1010 /* Get qemu help text and version. */
1011 if (test_qemu (g) == -1)
1014 /* Choose which vmchannel implementation to use. */
1015 if (qemu_supports (g, "-net user")) {
1016 /* The "null vmchannel" implementation. Requires SLIRP (user mode
1017 * networking in qemu) but no other vmchannel support. The daemon
1018 * will connect back to a random port number on localhost.
1020 struct sockaddr_in addr;
1021 socklen_t addrlen = sizeof addr;
1023 g->sock = socket (AF_INET, SOCK_STREAM, IPPROTO_TCP);
1024 if (g->sock == -1) {
1025 perrorf (g, "socket");
1028 addr.sin_family = AF_INET;
1029 addr.sin_port = htons (0);
1030 addr.sin_addr.s_addr = htonl (INADDR_LOOPBACK);
1031 if (bind (g->sock, (struct sockaddr *) &addr, addrlen) == -1) {
1032 perrorf (g, "bind");
1036 if (listen (g->sock, 256) == -1) {
1037 perrorf (g, "listen");
1041 if (getsockname (g->sock, (struct sockaddr *) &addr, &addrlen) == -1) {
1042 perrorf (g, "getsockname");
1046 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1047 perrorf (g, "fcntl");
1051 null_vmchannel_sock = ntohs (addr.sin_port);
1053 fprintf (stderr, "null_vmchannel_sock = %d\n", null_vmchannel_sock);
1055 /* Using some vmchannel impl. We need to create a local Unix
1056 * domain socket for qemu to use.
1058 snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
1060 null_vmchannel_sock = 0;
1064 if (pipe (wfd) == -1 || pipe (rfd) == -1) {
1065 perrorf (g, "pipe");
1071 print_timestamped_message (g, "finished testing qemu features");
1075 perrorf (g, "fork");
1085 if (r == 0) { /* Child (qemu). */
1087 const char *vmchannel = NULL;
1089 /* Set up the full command line. Do this in the subprocess so we
1090 * don't need to worry about cleaning up.
1092 g->cmdline[0] = g->qemu;
1094 /* Newer versions of qemu (from around 2009/12) changed the
1095 * behaviour of monitors so that an implicit '-monitor stdio' is
1096 * assumed if we are in -nographic mode and there is no other
1097 * -monitor option. Only a single stdio device is allowed, so
1098 * this broke the '-serial stdio' option. There is a new flag
1099 * called -nodefaults which gets rid of all this default crud, so
1100 * let's use that to avoid this and any future surprises.
1102 if (qemu_supports (g, "-nodefaults"))
1103 add_cmdline (g, "-nodefaults");
1105 add_cmdline (g, "-nographic");
1106 add_cmdline (g, "-serial");
1107 add_cmdline (g, "stdio");
1109 snprintf (buf, sizeof buf, "%d", g->memsize);
1110 add_cmdline (g, "-m");
1111 add_cmdline (g, buf);
1113 /* Force exit instead of reboot on panic */
1114 add_cmdline (g, "-no-reboot");
1116 /* These options recommended by KVM developers to improve reliability. */
1117 if (qemu_supports (g, "-no-hpet"))
1118 add_cmdline (g, "-no-hpet");
1120 if (qemu_supports (g, "-rtc-td-hack"))
1121 add_cmdline (g, "-rtc-td-hack");
1123 /* If qemu has SLIRP (user mode network) enabled then we can get
1124 * away with "no vmchannel", where we just connect back to a random
1127 if (null_vmchannel_sock) {
1128 add_cmdline (g, "-net");
1129 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1131 snprintf (buf, sizeof buf,
1132 "guestfs_vmchannel=tcp:10.0.2.2:%d", null_vmchannel_sock);
1133 vmchannel = strdup (buf);
1136 /* New-style -net user,guestfwd=... syntax for guestfwd. See:
1138 * http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=c92ef6a22d3c71538fcc48fb61ad353f7ba03b62
1140 * The original suggested format doesn't work, see:
1142 * http://lists.gnu.org/archive/html/qemu-devel/2009-07/msg01654.html
1144 * However Gerd Hoffman privately suggested to me using -chardev
1145 * instead, which does work.
1147 else if (qemu_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) {
1148 snprintf (buf, sizeof buf,
1149 "socket,id=guestfsvmc,path=%s,server,nowait", unixsock);
1151 add_cmdline (g, "-chardev");
1152 add_cmdline (g, buf);
1154 snprintf (buf, sizeof buf,
1155 "user,vlan=0,net=10.0.2.0/8,"
1156 "guestfwd=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT
1157 "-chardev:guestfsvmc");
1159 add_cmdline (g, "-net");
1160 add_cmdline (g, buf);
1162 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1165 /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel
1166 * syntax, or if not then we'll get a quick failure.
1169 snprintf (buf, sizeof buf,
1170 "channel," GUESTFWD_PORT ":unix:%s,server,nowait", unixsock);
1172 add_cmdline (g, "-net");
1173 add_cmdline (g, buf);
1174 add_cmdline (g, "-net");
1175 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1177 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1179 add_cmdline (g, "-net");
1180 add_cmdline (g, "nic,model=" NET_IF ",vlan=0");
1182 #define LINUX_CMDLINE \
1183 "panic=1 " /* force kernel to panic if daemon exits */ \
1184 "console=ttyS0 " /* serial console */ \
1185 "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \
1186 "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \
1187 "acpi=off " /* we don't need ACPI, turn it off */ \
1188 "printk.time=1 " /* display timestamp before kernel messages */ \
1189 "cgroup_disable=memory " /* saves us about 5 MB of RAM */
1191 /* Linux kernel command line. */
1192 snprintf (buf, sizeof buf,
1194 "%s " /* (selinux) */
1195 "%s " /* (vmchannel) */
1196 "%s " /* (verbose) */
1197 "%s", /* (append) */
1198 g->selinux ? "selinux=1 enforcing=0" : "selinux=0",
1199 vmchannel ? vmchannel : "",
1200 g->verbose ? "guestfs_verbose=1" : "",
1201 g->append ? g->append : "");
1203 add_cmdline (g, "-kernel");
1204 add_cmdline (g, (char *) kernel);
1205 add_cmdline (g, "-initrd");
1206 add_cmdline (g, (char *) initrd);
1207 add_cmdline (g, "-append");
1208 add_cmdline (g, buf);
1210 /* Finish off the command line. */
1211 incr_cmdline_size (g);
1212 g->cmdline[g->cmdline_size-1] = NULL;
1218 /* Set up stdin, stdout. */
1224 if (dup (wfd[0]) == -1) {
1226 perror ("dup failed");
1229 if (dup (rfd[1]) == -1)
1237 /* Set up a new process group, so we can signal this process
1238 * and all subprocesses (eg. if qemu is really a shell script).
1243 setenv ("LC_ALL", "C", 1);
1245 execv (g->qemu, g->cmdline); /* Run qemu. */
1250 /* Parent (library). */
1258 /* Fork the recovery process off which will kill qemu if the parent
1259 * process fails to do so (eg. if the parent segfaults).
1261 g->recoverypid = -1;
1262 if (g->recovery_proc) {
1265 pid_t qemu_pid = g->pid;
1266 pid_t parent_pid = getppid ();
1268 /* Writing to argv is hideously complicated and error prone. See:
1269 * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain
1272 /* Loop around waiting for one or both of the other processes to
1273 * disappear. It's fair to say this is very hairy. The PIDs that
1274 * we are looking at might be reused by another process. We are
1275 * effectively polling. Is the cure worse than the disease?
1278 if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */
1280 if (kill (parent_pid, 0) == -1) {
1281 /* Parent's gone away, qemu still around, so kill qemu. */
1289 /* Don't worry, if the fork failed, this will be -1. The recovery
1290 * process isn't essential.
1296 /* Close the other ends of the pipe. */
1300 if (fcntl (wfd[1], F_SETFL, O_NONBLOCK) == -1 ||
1301 fcntl (rfd[0], F_SETFL, O_NONBLOCK) == -1) {
1302 perrorf (g, "fcntl");
1306 g->fd[0] = wfd[1]; /* stdin of child */
1307 g->fd[1] = rfd[0]; /* stdout of child */
1309 g->fd[0] = open ("/dev/null", O_RDWR);
1310 if (g->fd[0] == -1) {
1311 perrorf (g, "open /dev/null");
1314 g->fd[1] = dup (g->fd[0]);
1315 if (g->fd[1] == -1) {
1322 if (null_vmchannel_sock) {
1326 /* Null vmchannel implementation: We listen on g->sock for a
1327 * connection. The connection could come from any local process
1328 * so we must check it comes from the appliance (or at least
1329 * from our UID) for security reasons.
1331 while (sock == -1) {
1332 sock = accept_from_daemon (g);
1336 if (check_peer_euid (g, sock, &uid) == -1)
1338 if (uid != geteuid ()) {
1340 "libguestfs: warning: unexpected connection from UID %d to port %d\n",
1341 uid, null_vmchannel_sock);
1347 if (fcntl (sock, F_SETFL, O_NONBLOCK) == -1) {
1348 perrorf (g, "fcntl");
1355 /* Other vmchannel. Open the Unix socket.
1357 * The vmchannel implementation that got merged with qemu sucks in
1358 * a number of ways. Both ends do connect(2), which means that no
1359 * one knows what, if anything, is connected to the other end, or
1360 * if it becomes disconnected. Even worse, we have to wait some
1361 * indeterminate time for qemu to create the socket and connect to
1362 * it (which happens very early in qemu's start-up), so any code
1363 * that uses vmchannel is inherently racy. Hence this silly loop.
1365 g->sock = socket (AF_UNIX, SOCK_STREAM, 0);
1366 if (g->sock == -1) {
1367 perrorf (g, "socket");
1371 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1372 perrorf (g, "fcntl");
1376 addr.sun_family = AF_UNIX;
1377 strncpy (addr.sun_path, unixsock, UNIX_PATH_MAX);
1378 addr.sun_path[UNIX_PATH_MAX-1] = '\0';
1381 /* Always sleep at least once to give qemu a small chance to start up. */
1384 r = connect (g->sock, (struct sockaddr *) &addr, sizeof addr);
1385 if ((r == -1 && errno == EINPROGRESS) || r == 0)
1388 if (errno != ENOENT)
1389 perrorf (g, "connect");
1394 error (g, _("failed to connect to vmchannel socket"));
1400 g->state = LAUNCHING;
1402 /* Wait for qemu to start and to connect back to us via vmchannel and
1403 * send the GUESTFS_LAUNCH_FLAG message.
1407 r = recv_from_daemon (g, &size, &buf);
1410 if (r == -1) return -1;
1412 if (size != GUESTFS_LAUNCH_FLAG) {
1413 error (g, _("guestfs_launch failed, see earlier error messages"));
1418 print_timestamped_message (g, "appliance is up");
1420 /* This is possible in some really strange situations, such as
1421 * guestfsd starts up OK but then qemu immediately exits. Check for
1422 * it because the caller is probably expecting to be able to send
1423 * commands after this function returns.
1425 if (g->state != READY) {
1426 error (g, _("qemu launched and contacted daemon, but state != READY"));
1438 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1439 waitpid (g->pid, NULL, 0);
1440 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1445 memset (&g->launch_t, 0, sizeof g->launch_t);
1458 /* This function is used to print the qemu command line before it gets
1459 * executed, when in verbose mode.
1462 print_cmdline (guestfs_h *g)
1467 while (g->cmdline[i]) {
1468 if (g->cmdline[i][0] == '-') /* -option starts a new line */
1469 fprintf (stderr, " \\\n ");
1471 if (i > 0) fputc (' ', stderr);
1473 /* Does it need shell quoting? This only deals with simple cases. */
1474 needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]);
1476 if (needs_quote) fputc ('\'', stderr);
1477 fprintf (stderr, "%s", g->cmdline[i]);
1478 if (needs_quote) fputc ('\'', stderr);
1482 fputc ('\n', stderr);
1485 /* This function does the hard work of building the supermin appliance
1486 * on the fly. 'path' is the directory containing the control files.
1487 * 'kernel' and 'initrd' are where we will return the names of the
1488 * kernel and initrd (only initrd is built). The work is done by
1489 * an external script. We just tell it where to put the result.
1492 build_supermin_appliance (guestfs_h *g, const char *path,
1493 char **kernel, char **initrd)
1499 print_timestamped_message (g, "begin building supermin appliance");
1501 len = strlen (g->tmpdir);
1502 *kernel = safe_malloc (g, len + 8);
1503 snprintf (*kernel, len+8, "%s/kernel", g->tmpdir);
1504 *initrd = safe_malloc (g, len + 8);
1505 snprintf (*initrd, len+8, "%s/initrd", g->tmpdir);
1507 snprintf (cmd, sizeof cmd,
1509 "libguestfs-supermin-helper '%s' %s %s",
1511 path, *kernel, *initrd);
1514 if (r == -1 || WEXITSTATUS(r) != 0) {
1515 error (g, _("external command failed: %s"), cmd);
1518 *kernel = *initrd = NULL;
1523 print_timestamped_message (g, "finished building supermin appliance");
1528 /* Compute Y - X and return the result in milliseconds.
1529 * Approximately the same as this code:
1530 * http://www.mpp.mpg.de/~huber/util/timevaldiff.c
1533 timeval_diff (const struct timeval *x, const struct timeval *y)
1537 msec = (y->tv_sec - x->tv_sec) * 1000;
1538 msec += (y->tv_usec - x->tv_usec) / 1000;
1543 print_timestamped_message (guestfs_h *g, const char *fs, ...)
1548 struct timeval tv, diff;
1550 va_start (args, fs);
1551 err = vasprintf (&msg, fs, args);
1554 if (err < 0) return;
1556 gettimeofday (&tv, NULL);
1558 fprintf (stderr, "[%05" PRIi64 "ms] %s\n",
1559 timeval_diff (&g->launch_t, &tv), msg);
1564 static int read_all (guestfs_h *g, FILE *fp, char **ret);
1566 /* Test qemu binary (or wrapper) runs, and do 'qemu -help' and
1567 * 'qemu -version' so we know what options this qemu supports and
1571 test_qemu (guestfs_h *g)
1576 free (g->qemu_help);
1577 free (g->qemu_version);
1578 g->qemu_help = NULL;
1579 g->qemu_version = NULL;
1581 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -help", g->qemu);
1583 fp = popen (cmd, "r");
1584 /* qemu -help should always work (qemu -version OTOH wasn't
1585 * supported by qemu 0.9). If this command doesn't work then it
1586 * probably indicates that the qemu binary is missing.
1589 /* XXX This error is never printed, even if the qemu binary
1590 * doesn't exist. Why?
1593 perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd);
1597 if (read_all (g, fp, &g->qemu_help) == -1)
1600 if (pclose (fp) == -1)
1603 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -version 2>/dev/null", g->qemu);
1605 fp = popen (cmd, "r");
1607 /* Intentionally ignore errors. */
1608 read_all (g, fp, &g->qemu_version);
1616 read_all (guestfs_h *g, FILE *fp, char **ret)
1623 *ret = safe_realloc (g, *ret, n + 1);
1628 *ret = safe_realloc (g, *ret, n + BUFSIZ);
1630 r = fread (p, 1, BUFSIZ, fp);
1632 perrorf (g, "read");
1639 /* Test if option is supported by qemu command line (just by grepping
1643 qemu_supports (guestfs_h *g, const char *option)
1645 return g->qemu_help && strstr (g->qemu_help, option) != NULL;
1648 /* Check the peer effective UID for a TCP socket. Ideally we'd like
1649 * SO_PEERCRED for a loopback TCP socket. This isn't possible on
1650 * Linux (but it is on Solaris!) so we read /proc/net/tcp instead.
1653 check_peer_euid (guestfs_h *g, int sock, uid_t *rtn)
1655 struct sockaddr_in peer;
1656 socklen_t addrlen = sizeof peer;
1658 if (getpeername (sock, (struct sockaddr *) &peer, &addrlen) == -1) {
1659 perrorf (g, "getpeername");
1663 if (peer.sin_family != AF_INET ||
1664 ntohl (peer.sin_addr.s_addr) != INADDR_LOOPBACK) {
1665 error (g, "check_peer_euid: unexpected connection from non-IPv4, non-loopback peer (family = %d, addr = %s)",
1666 peer.sin_family, inet_ntoa (peer.sin_addr));
1670 struct sockaddr_in our;
1671 addrlen = sizeof our;
1672 if (getsockname (sock, (struct sockaddr *) &our, &addrlen) == -1) {
1673 perrorf (g, "getsockname");
1677 FILE *fp = fopen ("/proc/net/tcp", "r");
1679 perrorf (g, "/proc/net/tcp");
1684 if (fgets (line, sizeof line, fp) == NULL) { /* Drop first line. */
1685 error (g, "unexpected end of file in /proc/net/tcp");
1690 while (fgets (line, sizeof line, fp) != NULL) {
1691 unsigned line_our_addr, line_our_port, line_peer_addr, line_peer_port;
1692 int dummy0, dummy1, dummy2, dummy3, dummy4, dummy5, dummy6;
1695 if (sscanf (line, "%d:%08X:%04X %08X:%04X %02X %08X:%08X %02X:%08X %08X %d",
1697 &line_our_addr, &line_our_port,
1698 &line_peer_addr, &line_peer_port,
1699 &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, &dummy6,
1701 /* Note about /proc/net/tcp: local_address and rem_address are
1702 * always in network byte order. However the port part is
1703 * always in host byte order.
1705 * The sockname and peername that we got above are in network
1706 * byte order. So we have to byte swap the port but not the
1709 if (line_our_addr == our.sin_addr.s_addr &&
1710 line_our_port == ntohs (our.sin_port) &&
1711 line_peer_addr == peer.sin_addr.s_addr &&
1712 line_peer_port == ntohs (peer.sin_port)) {
1720 error (g, "check_peer_euid: no matching TCP connection found in /proc/net/tcp");
1725 /* You had to call this function after launch in versions <= 1.0.70,
1726 * but it is now a no-op.
1729 guestfs__wait_ready (guestfs_h *g)
1731 if (g->state != READY) {
1732 error (g, _("qemu has not been launched yet"));
1740 guestfs__kill_subprocess (guestfs_h *g)
1742 if (g->state == CONFIG) {
1743 error (g, _("no subprocess to kill"));
1748 fprintf (stderr, "sending SIGTERM to process %d\n", g->pid);
1750 kill (g->pid, SIGTERM);
1751 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1756 /* Access current state. */
1758 guestfs__is_config (guestfs_h *g)
1760 return g->state == CONFIG;
1764 guestfs__is_launching (guestfs_h *g)
1766 return g->state == LAUNCHING;
1770 guestfs__is_ready (guestfs_h *g)
1772 return g->state == READY;
1776 guestfs__is_busy (guestfs_h *g)
1778 return g->state == BUSY;
1782 guestfs__get_state (guestfs_h *g)
1788 guestfs_set_log_message_callback (guestfs_h *g,
1789 guestfs_log_message_cb cb, void *opaque)
1791 g->log_message_cb = cb;
1792 g->log_message_cb_data = opaque;
1796 guestfs_set_subprocess_quit_callback (guestfs_h *g,
1797 guestfs_subprocess_quit_cb cb, void *opaque)
1799 g->subprocess_quit_cb = cb;
1800 g->subprocess_quit_cb_data = opaque;
1804 guestfs_set_launch_done_callback (guestfs_h *g,
1805 guestfs_launch_done_cb cb, void *opaque)
1807 g->launch_done_cb = cb;
1808 g->launch_done_cb_data = opaque;
1811 /*----------------------------------------------------------------------*/
1813 /* This is the code used to send and receive RPC messages and (for
1814 * certain types of message) to perform file transfers. This code is
1815 * driven from the generated actions (src/guestfs-actions.c). There
1816 * are five different cases to consider:
1818 * (1) A non-daemon function. There is no RPC involved at all, it's
1819 * all handled inside the library.
1821 * (2) A simple RPC (eg. "mount"). We write the request, then read
1822 * the reply. The sequence of calls is:
1824 * guestfs___set_busy
1827 * guestfs___end_busy
1829 * (3) An RPC with FileOut parameters (eg. "upload"). We write the
1830 * request, then write the file(s), then read the reply. The sequence
1833 * guestfs___set_busy
1835 * guestfs___send_file (possibly multiple times)
1837 * guestfs___end_busy
1839 * (4) An RPC with FileIn parameters (eg. "download"). We write the
1840 * request, then read the reply, then read the file(s). The sequence
1843 * guestfs___set_busy
1846 * guestfs___recv_file (possibly multiple times)
1847 * guestfs___end_busy
1849 * (5) Both FileOut and FileIn parameters. There are no calls like
1850 * this in the current API, but they would be implemented as a
1851 * combination of cases (3) and (4).
1853 * During all writes and reads, we also select(2) on qemu stdout
1854 * looking for messages (guestfsd stderr and guest kernel dmesg), and
1855 * anything received is passed up through the log_message_cb. This is
1856 * also the reason why all the sockets are non-blocking. We also have
1857 * to check for EOF (qemu died). All of this is handled by the
1858 * functions send_to_daemon and recv_from_daemon.
1862 guestfs___set_busy (guestfs_h *g)
1864 if (g->state != READY) {
1865 error (g, _("guestfs_set_busy: called when in state %d != READY"),
1874 guestfs___end_busy (guestfs_h *g)
1888 error (g, _("guestfs_end_busy: called when in state %d"), g->state);
1894 /* This is called if we detect EOF, ie. qemu died. */
1896 child_cleanup (guestfs_h *g)
1899 fprintf (stderr, "child_cleanup: %p: child process died\n", g);
1901 /*kill (g->pid, SIGTERM);*/
1902 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1903 waitpid (g->pid, NULL, 0);
1904 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1913 memset (&g->launch_t, 0, sizeof g->launch_t);
1915 if (g->subprocess_quit_cb)
1916 g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
1920 read_log_message_or_eof (guestfs_h *g, int fd)
1928 "read_log_message_or_eof: %p g->state = %d, fd = %d\n",
1932 /* QEMU's console emulates a 16550A serial port. The real 16550A
1933 * device has a small FIFO buffer (16 bytes) which means here we see
1934 * lots of small reads of 1-16 bytes in length, usually single
1937 n = read (fd, buf, sizeof buf);
1939 /* Hopefully this indicates the qemu child process has died. */
1945 if (errno == EINTR || errno == EAGAIN)
1948 perrorf (g, "read");
1952 /* In verbose mode, copy all log messages to stderr. */
1954 ignore_value (write (STDERR_FILENO, buf, n));
1956 /* It's an actual log message, send it upwards if anyone is listening. */
1957 if (g->log_message_cb)
1958 g->log_message_cb (g, g->log_message_cb_data, buf, n);
1964 check_for_daemon_cancellation_or_eof (guestfs_h *g, int fd)
1973 "check_for_daemon_cancellation_or_eof: %p g->state = %d, fd = %d\n",
1976 n = read (fd, buf, 4);
1978 /* Hopefully this indicates the qemu child process has died. */
1984 if (errno == EINTR || errno == EAGAIN)
1987 perrorf (g, "read");
1991 xdrmem_create (&xdr, buf, 4, XDR_DECODE);
1992 xdr_uint32_t (&xdr, &flag);
1995 if (flag != GUESTFS_CANCEL_FLAG) {
1996 error (g, _("check_for_daemon_cancellation_or_eof: read 0x%x from daemon, expected 0x%x\n"),
1997 flag, GUESTFS_CANCEL_FLAG);
2004 /* This writes the whole N bytes of BUF to the daemon socket.
2006 * If the whole write is successful, it returns 0.
2007 * If there was an error, it returns -1.
2008 * If the daemon sent a cancellation message, it returns -2.
2010 * It also checks qemu stdout for log messages and passes those up
2011 * through log_message_cb.
2013 * It also checks for EOF (qemu died) and passes that up through the
2014 * child_cleanup function above.
2017 send_to_daemon (guestfs_h *g, const void *v_buf, size_t n)
2019 const char *buf = v_buf;
2025 "send_to_daemon: %p g->state = %d, n = %zu\n", g, g->state, n);
2030 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2031 FD_SET (g->sock, &rset); /* Read socket for cancellation & EOF. */
2032 FD_SET (g->sock, &wset); /* Write to socket to send the data. */
2034 int max_fd = MAX (g->sock, g->fd[1]);
2039 int r = select (max_fd+1, &rset2, &wset2, NULL, NULL);
2041 if (errno == EINTR || errno == EAGAIN)
2043 perrorf (g, "select");
2047 if (FD_ISSET (g->fd[1], &rset2)) {
2048 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2051 if (FD_ISSET (g->sock, &rset2)) {
2052 r = check_for_daemon_cancellation_or_eof (g, g->sock);
2056 if (FD_ISSET (g->sock, &wset2)) {
2057 r = write (g->sock, buf, n);
2059 if (errno == EINTR || errno == EAGAIN)
2061 perrorf (g, "write");
2062 if (errno == EPIPE) /* Disconnected from guest (RHBZ#508713). */
2074 /* This reads a single message, file chunk, launch flag or
2075 * cancellation flag from the daemon. If something was read, it
2076 * returns 0, otherwise -1.
2078 * Both size_rtn and buf_rtn must be passed by the caller as non-NULL.
2080 * *size_rtn returns the size of the returned message or it may be
2081 * GUESTFS_LAUNCH_FLAG or GUESTFS_CANCEL_FLAG.
2083 * *buf_rtn is returned containing the message (if any) or will be set
2084 * to NULL. *buf_rtn must be freed by the caller.
2086 * It also checks qemu stdout for log messages and passes those up
2087 * through log_message_cb.
2089 * It also checks for EOF (qemu died) and passes that up through the
2090 * child_cleanup function above.
2093 recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn)
2099 "recv_from_daemon: %p g->state = %d, size_rtn = %p, buf_rtn = %p\n",
2100 g, g->state, size_rtn, buf_rtn);
2104 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2105 FD_SET (g->sock, &rset); /* Read socket for data & EOF. */
2107 int max_fd = MAX (g->sock, g->fd[1]);
2113 /* nr is the size of the message, but we prime it as -4 because we
2114 * have to read the message length word first.
2118 while (nr < (ssize_t) *size_rtn) {
2120 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2122 if (errno == EINTR || errno == EAGAIN)
2124 perrorf (g, "select");
2130 if (FD_ISSET (g->fd[1], &rset2)) {
2131 if (read_log_message_or_eof (g, g->fd[1]) == -1) {
2137 if (FD_ISSET (g->sock, &rset2)) {
2138 if (nr < 0) { /* Have we read the message length word yet? */
2139 r = read (g->sock, lenbuf+nr+4, -nr);
2141 if (errno == EINTR || errno == EAGAIN)
2144 perrorf (g, "read");
2145 /* Under some circumstances we see "Connection reset by peer"
2146 * here when the child dies suddenly. Catch this and call
2147 * the cleanup function, same as for EOF.
2149 if (err == ECONNRESET)
2154 error (g, _("unexpected end of file when reading from daemon"));
2160 if (nr < 0) /* Still not got the whole length word. */
2164 xdrmem_create (&xdr, lenbuf, 4, XDR_DECODE);
2165 xdr_uint32_t (&xdr, size_rtn);
2168 if (*size_rtn == GUESTFS_LAUNCH_FLAG) {
2169 if (g->state != LAUNCHING)
2170 error (g, _("received magic signature from guestfsd, but in state %d"),
2174 if (g->launch_done_cb)
2175 g->launch_done_cb (g, g->launch_done_cb_data);
2179 else if (*size_rtn == GUESTFS_CANCEL_FLAG)
2181 /* If this happens, it's pretty bad and we've probably lost
2184 else if (*size_rtn > GUESTFS_MESSAGE_MAX) {
2185 error (g, _("message length (%u) > maximum possible size (%d)"),
2186 (unsigned) *size_rtn, GUESTFS_MESSAGE_MAX);
2190 /* Allocate the complete buffer, size now known. */
2191 *buf_rtn = safe_malloc (g, *size_rtn);
2195 size_t sizetoread = *size_rtn - nr;
2196 if (sizetoread > BUFSIZ) sizetoread = BUFSIZ;
2198 r = read (g->sock, (char *) (*buf_rtn) + nr, sizetoread);
2200 if (errno == EINTR || errno == EAGAIN)
2202 perrorf (g, "read");
2208 error (g, _("unexpected end of file when reading from daemon"));
2218 /* Got the full message, caller can start processing it. */
2219 #ifdef ENABLE_PACKET_DUMP
2223 for (i = 0; i < nr; i += 16) {
2224 printf ("%04zx: ", i);
2225 for (j = i; j < MIN (i+16, nr); ++j)
2226 printf ("%02x ", (*(unsigned char **)buf_rtn)[j]);
2227 for (; j < i+16; ++j)
2230 for (j = i; j < MIN (i+16, nr); ++j)
2231 if (c_isprint ((*(char **)buf_rtn)[j]))
2232 printf ("%c", (*(char **)buf_rtn)[j]);
2235 for (; j < i+16; ++j)
2245 /* This is very much like recv_from_daemon above, but g->sock is
2246 * a listening socket and we are accepting a new connection on
2247 * that socket instead of reading anything. Returns the newly
2251 accept_from_daemon (guestfs_h *g)
2257 "accept_from_daemon: %p g->state = %d\n", g, g->state);
2261 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2262 FD_SET (g->sock, &rset); /* Read socket for accept. */
2264 int max_fd = MAX (g->sock, g->fd[1]);
2267 while (sock == -1) {
2269 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2271 if (errno == EINTR || errno == EAGAIN)
2273 perrorf (g, "select");
2277 if (FD_ISSET (g->fd[1], &rset2)) {
2278 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2281 if (FD_ISSET (g->sock, &rset2)) {
2282 sock = accept (g->sock, NULL, NULL);
2284 if (errno == EINTR || errno == EAGAIN)
2286 perrorf (g, "accept");
2296 guestfs___send (guestfs_h *g, int proc_nr, xdrproc_t xdrp, char *args)
2298 struct guestfs_message_header hdr;
2301 int serial = g->msg_next_serial++;
2304 size_t msg_out_size;
2306 if (g->state != BUSY) {
2307 error (g, _("guestfs___send: state %d != BUSY"), g->state);
2311 /* We have to allocate this message buffer on the heap because
2312 * it is quite large (although will be mostly unused). We
2313 * can't allocate it on the stack because in some environments
2314 * we have quite limited stack space available, notably when
2315 * running in the JVM.
2317 msg_out = safe_malloc (g, GUESTFS_MESSAGE_MAX + 4);
2318 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MESSAGE_MAX, XDR_ENCODE);
2320 /* Serialize the header. */
2321 hdr.prog = GUESTFS_PROGRAM;
2322 hdr.vers = GUESTFS_PROTOCOL_VERSION;
2324 hdr.direction = GUESTFS_DIRECTION_CALL;
2325 hdr.serial = serial;
2326 hdr.status = GUESTFS_STATUS_OK;
2328 if (!xdr_guestfs_message_header (&xdr, &hdr)) {
2329 error (g, _("xdr_guestfs_message_header failed"));
2333 /* Serialize the args. If any, because some message types
2334 * have no parameters.
2337 if (!(*xdrp) (&xdr, args)) {
2338 error (g, _("dispatch failed to marshal args"));
2343 /* Get the actual length of the message, resize the buffer to match
2344 * the actual length, and write the length word at the beginning.
2346 len = xdr_getpos (&xdr);
2349 msg_out = safe_realloc (g, msg_out, len + 4);
2350 msg_out_size = len + 4;
2352 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2353 xdr_uint32_t (&xdr, &len);
2356 r = send_to_daemon (g, msg_out, msg_out_size);
2357 if (r == -2) /* Ignore stray daemon cancellations. */
2370 static int cancel = 0; /* XXX Implement file cancellation. */
2371 static int send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t len);
2372 static int send_file_data (guestfs_h *g, const char *buf, size_t len);
2373 static int send_file_cancellation (guestfs_h *g);
2374 static int send_file_complete (guestfs_h *g);
2380 * -2 daemon cancelled (we must read the error message)
2383 guestfs___send_file (guestfs_h *g, const char *filename)
2385 char buf[GUESTFS_MAX_CHUNK_SIZE];
2388 fd = open (filename, O_RDONLY);
2390 perrorf (g, "open: %s", filename);
2391 send_file_cancellation (g);
2392 /* Daemon sees cancellation and won't reply, so caller can
2398 /* Send file in chunked encoding. */
2400 r = read (fd, buf, sizeof buf);
2401 if (r == -1 && (errno == EINTR || errno == EAGAIN))
2404 err = send_file_data (g, buf, r);
2406 if (err == -2) /* daemon sent cancellation */
2407 send_file_cancellation (g);
2412 if (cancel) { /* cancel from either end */
2413 send_file_cancellation (g);
2418 perrorf (g, "read: %s", filename);
2419 send_file_cancellation (g);
2423 /* End of file, but before we send that, we need to close
2424 * the file and check for errors.
2426 if (close (fd) == -1) {
2427 perrorf (g, "close: %s", filename);
2428 send_file_cancellation (g);
2432 return send_file_complete (g);
2435 /* Send a chunk of file data. */
2437 send_file_data (guestfs_h *g, const char *buf, size_t len)
2439 return send_file_chunk (g, 0, buf, len);
2442 /* Send a cancellation message. */
2444 send_file_cancellation (guestfs_h *g)
2446 return send_file_chunk (g, 1, NULL, 0);
2449 /* Send a file complete chunk. */
2451 send_file_complete (guestfs_h *g)
2454 return send_file_chunk (g, 0, buf, 0);
2458 send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t buflen)
2462 guestfs_chunk chunk;
2465 size_t msg_out_size;
2467 if (g->state != BUSY) {
2468 error (g, _("send_file_chunk: state %d != READY"), g->state);
2472 /* Allocate the chunk buffer. Don't use the stack to avoid
2473 * excessive stack usage and unnecessary copies.
2475 msg_out = safe_malloc (g, GUESTFS_MAX_CHUNK_SIZE + 4 + 48);
2476 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MAX_CHUNK_SIZE + 48, XDR_ENCODE);
2478 /* Serialize the chunk. */
2479 chunk.cancel = cancel;
2480 chunk.data.data_len = buflen;
2481 chunk.data.data_val = (char *) buf;
2483 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2484 error (g, _("xdr_guestfs_chunk failed (buf = %p, buflen = %zu)"),
2490 len = xdr_getpos (&xdr);
2493 /* Reduce the size of the outgoing message buffer to the real length. */
2494 msg_out = safe_realloc (g, msg_out, len + 4);
2495 msg_out_size = len + 4;
2497 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2498 xdr_uint32_t (&xdr, &len);
2500 r = send_to_daemon (g, msg_out, msg_out_size);
2502 /* Did the daemon send a cancellation message? */
2505 fprintf (stderr, "got daemon cancellation\n");
2521 /* Receive a reply. */
2523 guestfs___recv (guestfs_h *g, const char *fn,
2524 guestfs_message_header *hdr,
2525 guestfs_message_error *err,
2526 xdrproc_t xdrp, char *ret)
2534 r = recv_from_daemon (g, &size, &buf);
2538 /* This can happen if a cancellation happens right at the end
2539 * of us sending a FileIn parameter to the daemon. Discard. The
2540 * daemon should send us an error message next.
2542 if (size == GUESTFS_CANCEL_FLAG)
2545 if (size == GUESTFS_LAUNCH_FLAG) {
2546 error (g, "%s: received unexpected launch flag from daemon when expecting reply", fn);
2550 xdrmem_create (&xdr, buf, size, XDR_DECODE);
2552 if (!xdr_guestfs_message_header (&xdr, hdr)) {
2553 error (g, "%s: failed to parse reply header", fn);
2558 if (hdr->status == GUESTFS_STATUS_ERROR) {
2559 if (!xdr_guestfs_message_error (&xdr, err)) {
2560 error (g, "%s: failed to parse reply error", fn);
2566 if (xdrp && ret && !xdrp (&xdr, ret)) {
2567 error (g, "%s: failed to parse reply", fn);
2579 /* Receive a file. */
2581 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2582 static ssize_t receive_file_data (guestfs_h *g, void **buf);
2585 guestfs___recv_file (guestfs_h *g, const char *filename)
2591 fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2593 perrorf (g, "open: %s", filename);
2597 /* Receive the file in chunked encoding. */
2598 while ((r = receive_file_data (g, &buf)) > 0) {
2599 if (xwrite (fd, buf, r) == -1) {
2600 perrorf (g, "%s: write", filename);
2608 error (g, _("%s: error in chunked encoding"), filename);
2612 if (close (fd) == -1) {
2613 perrorf (g, "close: %s", filename);
2620 /* Send cancellation message to daemon, then wait until it
2621 * cancels (just throwing away data).
2625 uint32_t flag = GUESTFS_CANCEL_FLAG;
2628 fprintf (stderr, "%s: waiting for daemon to acknowledge cancellation\n",
2631 xdrmem_create (&xdr, fbuf, sizeof fbuf, XDR_ENCODE);
2632 xdr_uint32_t (&xdr, &flag);
2635 if (xwrite (g->sock, fbuf, sizeof fbuf) == -1) {
2636 perrorf (g, _("write to daemon socket"));
2640 while (receive_file_data (g, NULL) > 0)
2641 ; /* just discard it */
2646 /* Receive a chunk of file data. */
2647 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2649 receive_file_data (guestfs_h *g, void **buf_r)
2655 guestfs_chunk chunk;
2657 r = recv_from_daemon (g, &len, &buf);
2659 error (g, _("receive_file_data: parse error in reply callback"));
2663 if (len == GUESTFS_LAUNCH_FLAG || len == GUESTFS_CANCEL_FLAG) {
2664 error (g, _("receive_file_data: unexpected flag received when reading file chunks"));
2668 memset (&chunk, 0, sizeof chunk);
2670 xdrmem_create (&xdr, buf, len, XDR_DECODE);
2671 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2672 error (g, _("failed to parse file chunk"));
2677 /* After decoding, the original buffer is no longer used. */
2681 error (g, _("file receive cancelled by daemon"));
2682 free (chunk.data.data_val);
2686 if (chunk.data.data_len == 0) { /* end of transfer */
2687 free (chunk.data.data_val);
2691 if (buf_r) *buf_r = chunk.data.data_val;
2692 else free (chunk.data.data_val); /* else caller frees */
2694 return chunk.data.data_len;