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 "guestfs-internal-actions.h"
65 #include "guestfs_protocol.h"
67 #include "ignore-value.h"
71 #define _(str) dgettext(PACKAGE, (str))
72 //#define N_(str) dgettext(PACKAGE, (str))
78 #define error guestfs_error
79 #define perrorf guestfs_perrorf
80 #define safe_malloc guestfs_safe_malloc
81 #define safe_realloc guestfs_safe_realloc
82 #define safe_strdup guestfs_safe_strdup
83 //#define safe_memdup guestfs_safe_memdup
85 static void default_error_cb (guestfs_h *g, void *data, const char *msg);
86 static int send_to_daemon (guestfs_h *g, const void *v_buf, size_t n);
87 static int recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn);
88 static int accept_from_daemon (guestfs_h *g);
89 static int check_peer_euid (guestfs_h *g, int sock, uid_t *rtn);
90 static void close_handles (void);
92 #define UNIX_PATH_MAX 108
94 /* Also in guestfsd.c */
95 #define GUESTFWD_ADDR "10.0.2.4"
96 #define GUESTFWD_PORT "6666"
98 /* GuestFS handle and connection. */
99 enum state { CONFIG, LAUNCHING, READY, BUSY, NO_HANDLE };
103 struct guestfs_h *next; /* Linked list of open handles. */
105 /* State: see the state machine diagram in the man page guestfs(3). */
108 int fd[2]; /* Stdin/stdout of qemu. */
109 int sock; /* Daemon communications socket. */
110 pid_t pid; /* Qemu PID. */
111 pid_t recoverypid; /* Recovery process PID. */
113 struct timeval launch_t; /* The time that we called guestfs_launch. */
115 char *tmpdir; /* Temporary directory containing socket. */
117 char *qemu_help, *qemu_version; /* Output of qemu -help, qemu -version. */
119 char **cmdline; /* Qemu command line. */
128 char *path; /* Path to kernel, initrd. */
129 char *qemu; /* Qemu binary. */
130 char *append; /* Append to kernel command line. */
132 int memsize; /* Size of RAM (megabytes). */
134 int selinux; /* selinux enabled? */
139 guestfs_abort_cb abort_cb;
140 guestfs_error_handler_cb error_cb;
141 void * error_cb_data;
142 guestfs_log_message_cb log_message_cb;
143 void * log_message_cb_data;
144 guestfs_subprocess_quit_cb subprocess_quit_cb;
145 void * subprocess_quit_cb_data;
146 guestfs_launch_done_cb launch_done_cb;
147 void * launch_done_cb_data;
152 static guestfs_h *handles = NULL;
153 static int atexit_handler_set = 0;
156 guestfs_create (void)
161 g = malloc (sizeof (*g));
164 memset (g, 0, sizeof (*g));
173 g->error_cb = default_error_cb;
174 g->error_cb_data = NULL;
176 g->recovery_proc = 1;
178 str = getenv ("LIBGUESTFS_DEBUG");
179 g->verbose = str != NULL && STREQ (str, "1");
181 str = getenv ("LIBGUESTFS_TRACE");
182 g->trace = str != NULL && STREQ (str, "1");
184 str = getenv ("LIBGUESTFS_PATH");
185 g->path = str != NULL ? strdup (str) : strdup (GUESTFS_DEFAULT_PATH);
186 if (!g->path) goto error;
188 str = getenv ("LIBGUESTFS_QEMU");
189 g->qemu = str != NULL ? strdup (str) : strdup (QEMU);
190 if (!g->qemu) goto error;
192 str = getenv ("LIBGUESTFS_APPEND");
194 g->append = strdup (str);
195 if (!g->append) goto error;
198 /* Choose a suitable memory size. Previously we tried to choose
199 * a minimal memory size, but this isn't really necessary since
200 * recent QEMU and KVM don't do anything nasty like locking
201 * memory into core any more. Thus we can safely choose a
202 * large, generous amount of memory, and it'll just get swapped
203 * on smaller systems.
205 str = getenv ("LIBGUESTFS_MEMSIZE");
207 if (sscanf (str, "%d", &g->memsize) != 1 || g->memsize <= 256) {
208 fprintf (stderr, "libguestfs: non-numeric or too small value for LIBGUESTFS_MEMSIZE\n");
214 /* Start with large serial numbers so they are easy to spot
215 * inside the protocol.
217 g->msg_next_serial = 0x00123400;
219 /* Link the handles onto a global list. This is the one area
220 * where the library needs to be made thread-safe. (XXX)
222 /* acquire mutex (XXX) */
225 if (!atexit_handler_set) {
226 atexit (close_handles);
227 atexit_handler_set = 1;
229 /* release mutex (XXX) */
232 fprintf (stderr, "new guestfs handle %p\n", g);
245 guestfs_close (guestfs_h *g)
251 if (g->state == NO_HANDLE) {
252 /* Not safe to call 'error' here, so ... */
253 fprintf (stderr, _("guestfs_close: called twice on the same handle\n"));
258 fprintf (stderr, "closing guestfs handle %p (state %d)\n", g, g->state);
260 /* Try to sync if autosync flag is set. */
261 if (g->autosync && g->state == READY) {
262 guestfs_umount_all (g);
266 /* Remove any handlers that might be called back before we kill the
269 g->log_message_cb = NULL;
271 if (g->state != CONFIG)
272 guestfs_kill_subprocess (g);
285 /* Wait for subprocess(es) to exit. */
286 waitpid (g->pid, NULL, 0);
287 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
289 /* Remove tmpfiles. */
291 snprintf (filename, sizeof filename, "%s/sock", g->tmpdir);
294 snprintf (filename, sizeof filename, "%s/initrd", g->tmpdir);
297 snprintf (filename, sizeof filename, "%s/kernel", g->tmpdir);
306 for (i = 0; i < g->cmdline_size; ++i)
307 free (g->cmdline[i]);
311 /* Mark the handle as dead before freeing it. */
312 g->state = NO_HANDLE;
314 /* acquire mutex (XXX) */
318 for (gg = handles; gg->next != g; gg = gg->next)
322 /* release mutex (XXX) */
324 free (g->last_error);
329 free (g->qemu_version);
333 /* Close all open handles (called from atexit(3)). */
337 while (handles) guestfs_close (handles);
341 guestfs_last_error (guestfs_h *g)
343 return g->last_error;
347 set_last_error (guestfs_h *g, const char *msg)
349 free (g->last_error);
350 g->last_error = strdup (msg);
354 default_error_cb (guestfs_h *g, void *data, const char *msg)
356 fprintf (stderr, _("libguestfs: error: %s\n"), msg);
360 guestfs_error (guestfs_h *g, const char *fs, ...)
366 int err = vasprintf (&msg, fs, args);
371 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
372 set_last_error (g, msg);
378 guestfs_perrorf (guestfs_h *g, const char *fs, ...)
385 int err = vasprintf (&msg, fs, args);
392 strerror_r (errnum, buf, sizeof buf);
396 buf = strerror_r (errnum, _buf, sizeof _buf);
399 msg = safe_realloc (g, msg, strlen (msg) + 2 + strlen (buf) + 1);
403 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
404 set_last_error (g, msg);
410 guestfs_safe_malloc (guestfs_h *g, size_t nbytes)
412 void *ptr = malloc (nbytes);
413 if (nbytes > 0 && !ptr) g->abort_cb ();
417 /* Return 1 if an array of N objects, each of size S, cannot exist due
418 to size arithmetic overflow. S must be positive and N must be
419 nonnegative. This is a macro, not an inline function, so that it
420 works correctly even when SIZE_MAX < N.
422 By gnulib convention, SIZE_MAX represents overflow in size
423 calculations, so the conservative dividend to use here is
424 SIZE_MAX - 1, since SIZE_MAX might represent an overflowed value.
425 However, malloc (SIZE_MAX) fails on all known hosts where
426 sizeof (ptrdiff_t) <= sizeof (size_t), so do not bother to test for
427 exactly-SIZE_MAX allocations on such hosts; this avoids a test and
428 branch when S is known to be 1. */
429 # define xalloc_oversized(n, s) \
430 ((size_t) (sizeof (ptrdiff_t) <= sizeof (size_t) ? -1 : -2) / (s) < (n))
432 /* Technically we should add an autoconf test for this, testing for the desired
433 functionality, like what's done in gnulib, but for now, this is fine. */
434 #define HAVE_GNU_CALLOC (__GLIBC__ >= 2)
436 /* Allocate zeroed memory for N elements of S bytes, with error
437 checking. S must be nonzero. */
439 guestfs_safe_calloc (guestfs_h *g, size_t n, size_t s)
441 /* From gnulib's calloc function in xmalloc.c. */
443 /* Test for overflow, since some calloc implementations don't have
444 proper overflow checks. But omit overflow and size-zero tests if
445 HAVE_GNU_CALLOC, since GNU calloc catches overflow and never
446 returns NULL if successful. */
447 if ((! HAVE_GNU_CALLOC && xalloc_oversized (n, s))
448 || (! (p = calloc (n, s)) && (HAVE_GNU_CALLOC || n != 0)))
454 guestfs_safe_realloc (guestfs_h *g, void *ptr, int nbytes)
456 void *p = realloc (ptr, nbytes);
457 if (nbytes > 0 && !p) g->abort_cb ();
462 guestfs_safe_strdup (guestfs_h *g, const char *str)
464 char *s = strdup (str);
465 if (!s) g->abort_cb ();
470 guestfs_safe_memdup (guestfs_h *g, void *ptr, size_t size)
472 void *p = malloc (size);
473 if (!p) g->abort_cb ();
474 memcpy (p, ptr, size);
479 xwrite (int fd, const void *v_buf, size_t len)
481 const char *buf = v_buf;
485 r = write (fd, buf, len);
497 guestfs_set_out_of_memory_handler (guestfs_h *g, guestfs_abort_cb cb)
503 guestfs_get_out_of_memory_handler (guestfs_h *g)
509 guestfs_set_error_handler (guestfs_h *g, guestfs_error_handler_cb cb, void *data)
512 g->error_cb_data = data;
515 guestfs_error_handler_cb
516 guestfs_get_error_handler (guestfs_h *g, void **data_rtn)
518 if (data_rtn) *data_rtn = g->error_cb_data;
523 guestfs__set_verbose (guestfs_h *g, int v)
530 guestfs__get_verbose (guestfs_h *g)
536 guestfs__set_autosync (guestfs_h *g, int a)
543 guestfs__get_autosync (guestfs_h *g)
549 guestfs__set_path (guestfs_h *g, const char *path)
556 safe_strdup (g, GUESTFS_DEFAULT_PATH) : safe_strdup (g, path);
561 guestfs__get_path (guestfs_h *g)
567 guestfs__set_qemu (guestfs_h *g, const char *qemu)
572 g->qemu = qemu == NULL ? safe_strdup (g, QEMU) : safe_strdup (g, qemu);
577 guestfs__get_qemu (guestfs_h *g)
583 guestfs__set_append (guestfs_h *g, const char *append)
588 g->append = append ? safe_strdup (g, append) : NULL;
593 guestfs__get_append (guestfs_h *g)
599 guestfs__set_memsize (guestfs_h *g, int memsize)
601 g->memsize = memsize;
606 guestfs__get_memsize (guestfs_h *g)
612 guestfs__set_selinux (guestfs_h *g, int selinux)
614 g->selinux = selinux;
619 guestfs__get_selinux (guestfs_h *g)
625 guestfs__get_pid (guestfs_h *g)
630 error (g, "get_pid: no qemu subprocess");
635 struct guestfs_version *
636 guestfs__version (guestfs_h *g)
638 struct guestfs_version *r;
640 r = safe_malloc (g, sizeof *r);
641 r->major = PACKAGE_VERSION_MAJOR;
642 r->minor = PACKAGE_VERSION_MINOR;
643 r->release = PACKAGE_VERSION_RELEASE;
644 r->extra = safe_strdup (g, PACKAGE_VERSION_EXTRA);
649 guestfs__set_trace (guestfs_h *g, int t)
656 guestfs__get_trace (guestfs_h *g)
662 guestfs__set_direct (guestfs_h *g, int d)
669 guestfs__get_direct (guestfs_h *g)
675 guestfs__set_recovery_proc (guestfs_h *g, int f)
677 g->recovery_proc = !!f;
682 guestfs__get_recovery_proc (guestfs_h *g)
684 return g->recovery_proc;
687 /* Add a string to the current command line. */
689 incr_cmdline_size (guestfs_h *g)
691 if (g->cmdline == NULL) {
692 /* g->cmdline[0] is reserved for argv[0], set in guestfs_launch. */
694 g->cmdline = safe_malloc (g, sizeof (char *));
695 g->cmdline[0] = NULL;
699 g->cmdline = safe_realloc (g, g->cmdline, sizeof (char *) * g->cmdline_size);
703 add_cmdline (guestfs_h *g, const char *str)
705 if (g->state != CONFIG) {
707 _("command line cannot be altered after qemu subprocess launched"));
711 incr_cmdline_size (g);
712 g->cmdline[g->cmdline_size-1] = safe_strdup (g, str);
717 guestfs__config (guestfs_h *g,
718 const char *qemu_param, const char *qemu_value)
720 if (qemu_param[0] != '-') {
721 error (g, _("guestfs_config: parameter must begin with '-' character"));
725 /* A bit fascist, but the user will probably break the extra
726 * parameters that we add if they try to set any of these.
728 if (STREQ (qemu_param, "-kernel") ||
729 STREQ (qemu_param, "-initrd") ||
730 STREQ (qemu_param, "-nographic") ||
731 STREQ (qemu_param, "-serial") ||
732 STREQ (qemu_param, "-full-screen") ||
733 STREQ (qemu_param, "-std-vga") ||
734 STREQ (qemu_param, "-vnc")) {
735 error (g, _("guestfs_config: parameter '%s' isn't allowed"), qemu_param);
739 if (add_cmdline (g, qemu_param) != 0) return -1;
741 if (qemu_value != NULL) {
742 if (add_cmdline (g, qemu_value) != 0) return -1;
749 guestfs__add_drive (guestfs_h *g, const char *filename)
751 size_t len = strlen (filename) + 64;
754 if (strchr (filename, ',') != NULL) {
755 error (g, _("filename cannot contain ',' (comma) character"));
759 /* cache=off improves reliability in the event of a host crash.
761 * However this option causes qemu to try to open the file with
762 * O_DIRECT. This fails on some filesystem types (notably tmpfs).
763 * So we check if we can open the file with or without O_DIRECT,
764 * and use cache=off (or not) accordingly.
766 * This test also checks for the presence of the file, which
767 * is a documented semantic of this interface.
769 int fd = open (filename, O_RDONLY|O_DIRECT);
772 snprintf (buf, len, "file=%s,cache=off,if=" DRIVE_IF, filename);
774 fd = open (filename, O_RDONLY);
777 snprintf (buf, len, "file=%s,if=" DRIVE_IF, filename);
779 perrorf (g, "%s", filename);
784 return guestfs__config (g, "-drive", buf);
788 guestfs__add_drive_ro (guestfs_h *g, const char *filename)
790 size_t len = strlen (filename) + 64;
793 if (strchr (filename, ',') != NULL) {
794 error (g, _("filename cannot contain ',' (comma) character"));
798 if (access (filename, F_OK) == -1) {
799 perrorf (g, "%s", filename);
803 snprintf (buf, len, "file=%s,snapshot=on,if=%s", filename, DRIVE_IF);
805 return guestfs__config (g, "-drive", buf);
809 guestfs__add_cdrom (guestfs_h *g, const char *filename)
811 if (strchr (filename, ',') != NULL) {
812 error (g, _("filename cannot contain ',' (comma) character"));
816 if (access (filename, F_OK) == -1) {
817 perrorf (g, "%s", filename);
821 return guestfs__config (g, "-cdrom", filename);
824 /* Returns true iff file is contained in dir. */
826 dir_contains_file (const char *dir, const char *file)
828 int dirlen = strlen (dir);
829 int filelen = strlen (file);
830 int len = dirlen+filelen+2;
833 snprintf (path, len, "%s/%s", dir, file);
834 return access (path, F_OK) == 0;
837 /* Returns true iff every listed file is contained in 'dir'. */
839 dir_contains_files (const char *dir, ...)
844 va_start (args, dir);
845 while ((file = va_arg (args, const char *)) != NULL) {
846 if (!dir_contains_file (dir, file)) {
855 static void print_timestamped_message (guestfs_h *g, const char *fs, ...);
856 static int build_supermin_appliance (guestfs_h *g, const char *path, char **kernel, char **initrd);
857 static int test_qemu (guestfs_h *g);
858 static int qemu_supports (guestfs_h *g, const char *option);
859 static void print_cmdline (guestfs_h *g);
861 static const char *kernel_name = "vmlinuz." REPO "." host_cpu;
862 static const char *initrd_name = "initramfs." REPO "." host_cpu ".img";
863 static const char *supermin_name =
864 "initramfs." REPO "." host_cpu ".supermin.img";
865 static const char *supermin_hostfiles_name =
866 "initramfs." REPO "." host_cpu ".supermin.hostfiles";
869 guestfs__launch (guestfs_h *g)
872 char dir_template[PATH_MAX];
877 char *path, *pelem, *pend;
878 char *kernel = NULL, *initrd = NULL;
879 int null_vmchannel_sock;
881 struct sockaddr_un addr;
883 /* Start the clock ... */
884 gettimeofday (&g->launch_t, NULL);
892 tmpdir = getenv ("TMPDIR") ? : tmpdir;
893 snprintf (dir_template, sizeof dir_template, "%s/libguestfsXXXXXX", tmpdir);
897 error (g, _("you must call guestfs_add_drive before guestfs_launch"));
901 if (g->state != CONFIG) {
902 error (g, _("qemu has already been launched"));
906 /* Make the temporary directory. */
908 g->tmpdir = safe_strdup (g, dir_template);
909 if (mkdtemp (g->tmpdir) == NULL) {
910 perrorf (g, _("%s: cannot create temporary directory"), dir_template);
915 /* First search g->path for the supermin appliance, and try to
916 * synthesize a kernel and initrd from that. If it fails, we
917 * try the path search again looking for a backup ordinary
920 pelem = path = safe_strdup (g, g->path);
922 pend = strchrnul (pelem, ':');
923 pmore = *pend == ':';
927 /* Empty element of "." means cwd. */
928 if (len == 0 || (len == 1 && *pelem == '.')) {
931 "looking for supermin appliance in current directory\n");
932 if (dir_contains_files (".",
933 supermin_name, supermin_hostfiles_name,
934 "kmod.whitelist", NULL)) {
935 if (build_supermin_appliance (g, ".", &kernel, &initrd) == -1)
940 /* Look at <path>/supermin* etc. */
943 fprintf (stderr, "looking for supermin appliance in %s\n", pelem);
945 if (dir_contains_files (pelem,
946 supermin_name, supermin_hostfiles_name,
947 "kmod.whitelist", NULL)) {
948 if (build_supermin_appliance (g, pelem, &kernel, &initrd) == -1)
959 if (kernel == NULL || initrd == NULL) {
960 /* Search g->path for the kernel and initrd. */
961 pelem = path = safe_strdup (g, g->path);
963 pend = strchrnul (pelem, ':');
964 pmore = *pend == ':';
968 /* Empty element or "." means cwd. */
969 if (len == 0 || (len == 1 && *pelem == '.')) {
972 "looking for appliance in current directory\n");
973 if (dir_contains_files (".", kernel_name, initrd_name, NULL)) {
974 kernel = safe_strdup (g, kernel_name);
975 initrd = safe_strdup (g, initrd_name);
979 /* Look at <path>/kernel etc. */
982 fprintf (stderr, "looking for appliance in %s\n", pelem);
984 if (dir_contains_files (pelem, kernel_name, initrd_name, NULL)) {
985 kernel = safe_malloc (g, len + strlen (kernel_name) + 2);
986 initrd = safe_malloc (g, len + strlen (initrd_name) + 2);
987 sprintf (kernel, "%s/%s", pelem, kernel_name);
988 sprintf (initrd, "%s/%s", pelem, initrd_name);
999 if (kernel == NULL || initrd == NULL) {
1000 error (g, _("cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)"),
1001 kernel_name, initrd_name, g->path);
1006 print_timestamped_message (g, "begin testing qemu features");
1008 /* Get qemu help text and version. */
1009 if (test_qemu (g) == -1)
1012 /* Choose which vmchannel implementation to use. */
1013 if (qemu_supports (g, "-net user")) {
1014 /* The "null vmchannel" implementation. Requires SLIRP (user mode
1015 * networking in qemu) but no other vmchannel support. The daemon
1016 * will connect back to a random port number on localhost.
1018 struct sockaddr_in addr;
1019 socklen_t addrlen = sizeof addr;
1021 g->sock = socket (AF_INET, SOCK_STREAM, IPPROTO_TCP);
1022 if (g->sock == -1) {
1023 perrorf (g, "socket");
1026 addr.sin_family = AF_INET;
1027 addr.sin_port = htons (0);
1028 addr.sin_addr.s_addr = htonl (INADDR_LOOPBACK);
1029 if (bind (g->sock, (struct sockaddr *) &addr, addrlen) == -1) {
1030 perrorf (g, "bind");
1034 if (listen (g->sock, 256) == -1) {
1035 perrorf (g, "listen");
1039 if (getsockname (g->sock, (struct sockaddr *) &addr, &addrlen) == -1) {
1040 perrorf (g, "getsockname");
1044 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1045 perrorf (g, "fcntl");
1049 null_vmchannel_sock = ntohs (addr.sin_port);
1051 fprintf (stderr, "null_vmchannel_sock = %d\n", null_vmchannel_sock);
1053 /* Using some vmchannel impl. We need to create a local Unix
1054 * domain socket for qemu to use.
1056 snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
1058 null_vmchannel_sock = 0;
1062 if (pipe (wfd) == -1 || pipe (rfd) == -1) {
1063 perrorf (g, "pipe");
1069 print_timestamped_message (g, "finished testing qemu features");
1073 perrorf (g, "fork");
1083 if (r == 0) { /* Child (qemu). */
1085 const char *vmchannel = NULL;
1087 /* Set up the full command line. Do this in the subprocess so we
1088 * don't need to worry about cleaning up.
1090 g->cmdline[0] = g->qemu;
1092 snprintf (buf, sizeof buf, "%d", g->memsize);
1093 add_cmdline (g, "-m");
1094 add_cmdline (g, buf);
1096 add_cmdline (g, "-no-reboot"); /* Force exit instead of reboot on panic */
1097 add_cmdline (g, "-nographic");
1098 add_cmdline (g, "-serial");
1099 add_cmdline (g, "stdio");
1101 /* These options recommended by KVM developers to improve reliability. */
1102 if (qemu_supports (g, "-no-hpet"))
1103 add_cmdline (g, "-no-hpet");
1105 if (qemu_supports (g, "-rtc-td-hack"))
1106 add_cmdline (g, "-rtc-td-hack");
1108 /* If qemu has SLIRP (user mode network) enabled then we can get
1109 * away with "no vmchannel", where we just connect back to a random
1112 if (null_vmchannel_sock) {
1113 add_cmdline (g, "-net");
1114 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1116 snprintf (buf, sizeof buf,
1117 "guestfs_vmchannel=tcp:10.0.2.2:%d", null_vmchannel_sock);
1118 vmchannel = strdup (buf);
1121 /* New-style -net user,guestfwd=... syntax for guestfwd. See:
1123 * http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=c92ef6a22d3c71538fcc48fb61ad353f7ba03b62
1125 * The original suggested format doesn't work, see:
1127 * http://lists.gnu.org/archive/html/qemu-devel/2009-07/msg01654.html
1129 * However Gerd Hoffman privately suggested to me using -chardev
1130 * instead, which does work.
1132 else if (qemu_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) {
1133 snprintf (buf, sizeof buf,
1134 "socket,id=guestfsvmc,path=%s,server,nowait", unixsock);
1136 add_cmdline (g, "-chardev");
1137 add_cmdline (g, buf);
1139 snprintf (buf, sizeof buf,
1140 "user,vlan=0,net=10.0.2.0/8,"
1141 "guestfwd=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT
1142 "-chardev:guestfsvmc");
1144 add_cmdline (g, "-net");
1145 add_cmdline (g, buf);
1147 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1150 /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel
1151 * syntax, or if not then we'll get a quick failure.
1154 snprintf (buf, sizeof buf,
1155 "channel," GUESTFWD_PORT ":unix:%s,server,nowait", unixsock);
1157 add_cmdline (g, "-net");
1158 add_cmdline (g, buf);
1159 add_cmdline (g, "-net");
1160 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1162 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1164 add_cmdline (g, "-net");
1165 add_cmdline (g, "nic,model=" NET_IF ",vlan=0");
1167 #define LINUX_CMDLINE \
1168 "panic=1 " /* force kernel to panic if daemon exits */ \
1169 "console=ttyS0 " /* serial console */ \
1170 "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \
1171 "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \
1172 "acpi=off " /* we don't need ACPI, turn it off */ \
1173 "printk.time=1 " /* display timestamp before kernel messages */ \
1174 "cgroup_disable=memory " /* saves us about 5 MB of RAM */
1176 /* Linux kernel command line. */
1177 snprintf (buf, sizeof buf,
1179 "%s " /* (selinux) */
1180 "%s " /* (vmchannel) */
1181 "%s " /* (verbose) */
1182 "%s", /* (append) */
1183 g->selinux ? "selinux=1 enforcing=0" : "selinux=0",
1184 vmchannel ? vmchannel : "",
1185 g->verbose ? "guestfs_verbose=1" : "",
1186 g->append ? g->append : "");
1188 add_cmdline (g, "-kernel");
1189 add_cmdline (g, (char *) kernel);
1190 add_cmdline (g, "-initrd");
1191 add_cmdline (g, (char *) initrd);
1192 add_cmdline (g, "-append");
1193 add_cmdline (g, buf);
1195 /* Finish off the command line. */
1196 incr_cmdline_size (g);
1197 g->cmdline[g->cmdline_size-1] = NULL;
1203 /* Set up stdin, stdout. */
1209 if (dup (wfd[0]) == -1) {
1211 perror ("dup failed");
1214 if (dup (rfd[1]) == -1)
1222 /* Set up a new process group, so we can signal this process
1223 * and all subprocesses (eg. if qemu is really a shell script).
1228 setenv ("LC_ALL", "C", 1);
1230 execv (g->qemu, g->cmdline); /* Run qemu. */
1235 /* Parent (library). */
1243 /* Fork the recovery process off which will kill qemu if the parent
1244 * process fails to do so (eg. if the parent segfaults).
1246 g->recoverypid = -1;
1247 if (g->recovery_proc) {
1250 pid_t qemu_pid = g->pid;
1251 pid_t parent_pid = getppid ();
1253 /* Writing to argv is hideously complicated and error prone. See:
1254 * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain
1257 /* Loop around waiting for one or both of the other processes to
1258 * disappear. It's fair to say this is very hairy. The PIDs that
1259 * we are looking at might be reused by another process. We are
1260 * effectively polling. Is the cure worse than the disease?
1263 if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */
1265 if (kill (parent_pid, 0) == -1) {
1266 /* Parent's gone away, qemu still around, so kill qemu. */
1274 /* Don't worry, if the fork failed, this will be -1. The recovery
1275 * process isn't essential.
1281 /* Close the other ends of the pipe. */
1285 if (fcntl (wfd[1], F_SETFL, O_NONBLOCK) == -1 ||
1286 fcntl (rfd[0], F_SETFL, O_NONBLOCK) == -1) {
1287 perrorf (g, "fcntl");
1291 g->fd[0] = wfd[1]; /* stdin of child */
1292 g->fd[1] = rfd[0]; /* stdout of child */
1294 g->fd[0] = open ("/dev/null", O_RDWR);
1295 if (g->fd[0] == -1) {
1296 perrorf (g, "open /dev/null");
1299 g->fd[1] = dup (g->fd[0]);
1300 if (g->fd[1] == -1) {
1307 if (null_vmchannel_sock) {
1311 /* Null vmchannel implementation: We listen on g->sock for a
1312 * connection. The connection could come from any local process
1313 * so we must check it comes from the appliance (or at least
1314 * from our UID) for security reasons.
1316 while (sock == -1) {
1317 sock = accept_from_daemon (g);
1321 if (check_peer_euid (g, sock, &uid) == -1)
1323 if (uid != geteuid ()) {
1325 "libguestfs: warning: unexpected connection from UID %d to port %d\n",
1326 uid, null_vmchannel_sock);
1332 if (fcntl (sock, F_SETFL, O_NONBLOCK) == -1) {
1333 perrorf (g, "fcntl");
1340 /* Other vmchannel. Open the Unix socket.
1342 * The vmchannel implementation that got merged with qemu sucks in
1343 * a number of ways. Both ends do connect(2), which means that no
1344 * one knows what, if anything, is connected to the other end, or
1345 * if it becomes disconnected. Even worse, we have to wait some
1346 * indeterminate time for qemu to create the socket and connect to
1347 * it (which happens very early in qemu's start-up), so any code
1348 * that uses vmchannel is inherently racy. Hence this silly loop.
1350 g->sock = socket (AF_UNIX, SOCK_STREAM, 0);
1351 if (g->sock == -1) {
1352 perrorf (g, "socket");
1356 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1357 perrorf (g, "fcntl");
1361 addr.sun_family = AF_UNIX;
1362 strncpy (addr.sun_path, unixsock, UNIX_PATH_MAX);
1363 addr.sun_path[UNIX_PATH_MAX-1] = '\0';
1366 /* Always sleep at least once to give qemu a small chance to start up. */
1369 r = connect (g->sock, (struct sockaddr *) &addr, sizeof addr);
1370 if ((r == -1 && errno == EINPROGRESS) || r == 0)
1373 if (errno != ENOENT)
1374 perrorf (g, "connect");
1379 error (g, _("failed to connect to vmchannel socket"));
1385 g->state = LAUNCHING;
1387 /* Wait for qemu to start and to connect back to us via vmchannel and
1388 * send the GUESTFS_LAUNCH_FLAG message.
1392 r = recv_from_daemon (g, &size, &buf);
1395 if (r == -1) return -1;
1397 if (size != GUESTFS_LAUNCH_FLAG) {
1398 error (g, _("guestfs_launch failed, see earlier error messages"));
1403 print_timestamped_message (g, "appliance is up");
1405 /* This is possible in some really strange situations, such as
1406 * guestfsd starts up OK but then qemu immediately exits. Check for
1407 * it because the caller is probably expecting to be able to send
1408 * commands after this function returns.
1410 if (g->state != READY) {
1411 error (g, _("qemu launched and contacted daemon, but state != READY"));
1423 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1424 waitpid (g->pid, NULL, 0);
1425 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1430 memset (&g->launch_t, 0, sizeof g->launch_t);
1443 /* This function is used to print the qemu command line before it gets
1444 * executed, when in verbose mode.
1447 print_cmdline (guestfs_h *g)
1452 while (g->cmdline[i]) {
1453 if (g->cmdline[i][0] == '-') /* -option starts a new line */
1454 fprintf (stderr, " \\\n ");
1456 if (i > 0) fputc (' ', stderr);
1458 /* Does it need shell quoting? This only deals with simple cases. */
1459 needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]);
1461 if (needs_quote) fputc ('\'', stderr);
1462 fprintf (stderr, "%s", g->cmdline[i]);
1463 if (needs_quote) fputc ('\'', stderr);
1467 fputc ('\n', stderr);
1470 /* This function does the hard work of building the supermin appliance
1471 * on the fly. 'path' is the directory containing the control files.
1472 * 'kernel' and 'initrd' are where we will return the names of the
1473 * kernel and initrd (only initrd is built). The work is done by
1474 * an external script. We just tell it where to put the result.
1477 build_supermin_appliance (guestfs_h *g, const char *path,
1478 char **kernel, char **initrd)
1484 print_timestamped_message (g, "begin building supermin appliance");
1486 len = strlen (g->tmpdir);
1487 *kernel = safe_malloc (g, len + 8);
1488 snprintf (*kernel, len+8, "%s/kernel", g->tmpdir);
1489 *initrd = safe_malloc (g, len + 8);
1490 snprintf (*initrd, len+8, "%s/initrd", g->tmpdir);
1492 snprintf (cmd, sizeof cmd,
1494 "libguestfs-supermin-helper '%s' %s %s",
1496 path, *kernel, *initrd);
1499 if (r == -1 || WEXITSTATUS(r) != 0) {
1500 error (g, _("external command failed: %s"), cmd);
1503 *kernel = *initrd = NULL;
1508 print_timestamped_message (g, "finished building supermin appliance");
1513 /* Compute Y - X and return the result in milliseconds.
1514 * Approximately the same as this code:
1515 * http://www.mpp.mpg.de/~huber/util/timevaldiff.c
1518 timeval_diff (const struct timeval *x, const struct timeval *y)
1522 msec = (y->tv_sec - x->tv_sec) * 1000;
1523 msec += (y->tv_usec - x->tv_usec) / 1000;
1528 print_timestamped_message (guestfs_h *g, const char *fs, ...)
1533 struct timeval tv, diff;
1535 va_start (args, fs);
1536 err = vasprintf (&msg, fs, args);
1539 if (err < 0) return;
1541 gettimeofday (&tv, NULL);
1543 fprintf (stderr, "[%05" PRIi64 "ms] %s\n",
1544 timeval_diff (&g->launch_t, &tv), msg);
1549 static int read_all (guestfs_h *g, FILE *fp, char **ret);
1551 /* Test qemu binary (or wrapper) runs, and do 'qemu -help' and
1552 * 'qemu -version' so we know what options this qemu supports and
1556 test_qemu (guestfs_h *g)
1561 free (g->qemu_help);
1562 free (g->qemu_version);
1563 g->qemu_help = NULL;
1564 g->qemu_version = NULL;
1566 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -help", g->qemu);
1568 fp = popen (cmd, "r");
1569 /* qemu -help should always work (qemu -version OTOH wasn't
1570 * supported by qemu 0.9). If this command doesn't work then it
1571 * probably indicates that the qemu binary is missing.
1574 /* XXX This error is never printed, even if the qemu binary
1575 * doesn't exist. Why?
1578 perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd);
1582 if (read_all (g, fp, &g->qemu_help) == -1)
1585 if (pclose (fp) == -1)
1588 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -version 2>/dev/null", g->qemu);
1590 fp = popen (cmd, "r");
1592 /* Intentionally ignore errors. */
1593 read_all (g, fp, &g->qemu_version);
1601 read_all (guestfs_h *g, FILE *fp, char **ret)
1608 *ret = safe_realloc (g, *ret, n + 1);
1613 *ret = safe_realloc (g, *ret, n + BUFSIZ);
1615 r = fread (p, 1, BUFSIZ, fp);
1617 perrorf (g, "read");
1624 /* Test if option is supported by qemu command line (just by grepping
1628 qemu_supports (guestfs_h *g, const char *option)
1630 return g->qemu_help && strstr (g->qemu_help, option) != NULL;
1633 /* Check the peer effective UID for a TCP socket. Ideally we'd like
1634 * SO_PEERCRED for a loopback TCP socket. This isn't possible on
1635 * Linux (but it is on Solaris!) so we read /proc/net/tcp instead.
1638 check_peer_euid (guestfs_h *g, int sock, uid_t *rtn)
1640 struct sockaddr_in peer;
1641 socklen_t addrlen = sizeof peer;
1643 if (getpeername (sock, (struct sockaddr *) &peer, &addrlen) == -1) {
1644 perrorf (g, "getpeername");
1648 if (peer.sin_family != AF_INET ||
1649 ntohl (peer.sin_addr.s_addr) != INADDR_LOOPBACK) {
1650 error (g, "check_peer_euid: unexpected connection from non-IPv4, non-loopback peer (family = %d, addr = %s)",
1651 peer.sin_family, inet_ntoa (peer.sin_addr));
1655 struct sockaddr_in our;
1656 addrlen = sizeof our;
1657 if (getsockname (sock, (struct sockaddr *) &our, &addrlen) == -1) {
1658 perrorf (g, "getsockname");
1662 FILE *fp = fopen ("/proc/net/tcp", "r");
1664 perrorf (g, "/proc/net/tcp");
1669 if (fgets (line, sizeof line, fp) == NULL) { /* Drop first line. */
1670 error (g, "unexpected end of file in /proc/net/tcp");
1675 while (fgets (line, sizeof line, fp) != NULL) {
1676 unsigned line_our_addr, line_our_port, line_peer_addr, line_peer_port;
1677 int dummy0, dummy1, dummy2, dummy3, dummy4, dummy5, dummy6;
1680 if (sscanf (line, "%d:%08X:%04X %08X:%04X %02X %08X:%08X %02X:%08X %08X %d",
1682 &line_our_addr, &line_our_port,
1683 &line_peer_addr, &line_peer_port,
1684 &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, &dummy6,
1686 /* Note about /proc/net/tcp: local_address and rem_address are
1687 * always in network byte order. However the port part is
1688 * always in host byte order.
1690 * The sockname and peername that we got above are in network
1691 * byte order. So we have to byte swap the port but not the
1694 if (line_our_addr == our.sin_addr.s_addr &&
1695 line_our_port == ntohs (our.sin_port) &&
1696 line_peer_addr == peer.sin_addr.s_addr &&
1697 line_peer_port == ntohs (peer.sin_port)) {
1705 error (g, "check_peer_euid: no matching TCP connection found in /proc/net/tcp");
1710 /* You had to call this function after launch in versions <= 1.0.70,
1711 * but it is now a no-op.
1714 guestfs__wait_ready (guestfs_h *g)
1716 if (g->state != READY) {
1717 error (g, _("qemu has not been launched yet"));
1725 guestfs__kill_subprocess (guestfs_h *g)
1727 if (g->state == CONFIG) {
1728 error (g, _("no subprocess to kill"));
1733 fprintf (stderr, "sending SIGTERM to process %d\n", g->pid);
1735 kill (g->pid, SIGTERM);
1736 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1741 /* Access current state. */
1743 guestfs__is_config (guestfs_h *g)
1745 return g->state == CONFIG;
1749 guestfs__is_launching (guestfs_h *g)
1751 return g->state == LAUNCHING;
1755 guestfs__is_ready (guestfs_h *g)
1757 return g->state == READY;
1761 guestfs__is_busy (guestfs_h *g)
1763 return g->state == BUSY;
1767 guestfs__get_state (guestfs_h *g)
1773 guestfs_set_log_message_callback (guestfs_h *g,
1774 guestfs_log_message_cb cb, void *opaque)
1776 g->log_message_cb = cb;
1777 g->log_message_cb_data = opaque;
1781 guestfs_set_subprocess_quit_callback (guestfs_h *g,
1782 guestfs_subprocess_quit_cb cb, void *opaque)
1784 g->subprocess_quit_cb = cb;
1785 g->subprocess_quit_cb_data = opaque;
1789 guestfs_set_launch_done_callback (guestfs_h *g,
1790 guestfs_launch_done_cb cb, void *opaque)
1792 g->launch_done_cb = cb;
1793 g->launch_done_cb_data = opaque;
1796 /*----------------------------------------------------------------------*/
1798 /* This is the code used to send and receive RPC messages and (for
1799 * certain types of message) to perform file transfers. This code is
1800 * driven from the generated actions (src/guestfs-actions.c). There
1801 * are five different cases to consider:
1803 * (1) A non-daemon function. There is no RPC involved at all, it's
1804 * all handled inside the library.
1806 * (2) A simple RPC (eg. "mount"). We write the request, then read
1807 * the reply. The sequence of calls is:
1809 * guestfs___set_busy
1812 * guestfs___end_busy
1814 * (3) An RPC with FileOut parameters (eg. "upload"). We write the
1815 * request, then write the file(s), then read the reply. The sequence
1818 * guestfs___set_busy
1820 * guestfs___send_file (possibly multiple times)
1822 * guestfs___end_busy
1824 * (4) An RPC with FileIn parameters (eg. "download"). We write the
1825 * request, then read the reply, then read the file(s). The sequence
1828 * guestfs___set_busy
1831 * guestfs___recv_file (possibly multiple times)
1832 * guestfs___end_busy
1834 * (5) Both FileOut and FileIn parameters. There are no calls like
1835 * this in the current API, but they would be implemented as a
1836 * combination of cases (3) and (4).
1838 * During all writes and reads, we also select(2) on qemu stdout
1839 * looking for messages (guestfsd stderr and guest kernel dmesg), and
1840 * anything received is passed up through the log_message_cb. This is
1841 * also the reason why all the sockets are non-blocking. We also have
1842 * to check for EOF (qemu died). All of this is handled by the
1843 * functions send_to_daemon and recv_from_daemon.
1847 guestfs___set_busy (guestfs_h *g)
1849 if (g->state != READY) {
1850 error (g, _("guestfs_set_busy: called when in state %d != READY"),
1859 guestfs___end_busy (guestfs_h *g)
1873 error (g, _("guestfs_end_busy: called when in state %d"), g->state);
1879 /* This is called if we detect EOF, ie. qemu died. */
1881 child_cleanup (guestfs_h *g)
1884 fprintf (stderr, "child_cleanup: %p: child process died\n", g);
1886 /*kill (g->pid, SIGTERM);*/
1887 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1888 waitpid (g->pid, NULL, 0);
1889 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1898 memset (&g->launch_t, 0, sizeof g->launch_t);
1900 if (g->subprocess_quit_cb)
1901 g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
1905 read_log_message_or_eof (guestfs_h *g, int fd)
1913 "read_log_message_or_eof: %p g->state = %d, fd = %d\n",
1917 /* QEMU's console emulates a 16550A serial port. The real 16550A
1918 * device has a small FIFO buffer (16 bytes) which means here we see
1919 * lots of small reads of 1-16 bytes in length, usually single
1922 n = read (fd, buf, sizeof buf);
1924 /* Hopefully this indicates the qemu child process has died. */
1930 if (errno == EINTR || errno == EAGAIN)
1933 perrorf (g, "read");
1937 /* In verbose mode, copy all log messages to stderr. */
1939 ignore_value (write (STDERR_FILENO, buf, n));
1941 /* It's an actual log message, send it upwards if anyone is listening. */
1942 if (g->log_message_cb)
1943 g->log_message_cb (g, g->log_message_cb_data, buf, n);
1949 check_for_daemon_cancellation_or_eof (guestfs_h *g, int fd)
1958 "check_for_daemon_cancellation_or_eof: %p g->state = %d, fd = %d\n",
1961 n = read (fd, buf, 4);
1963 /* Hopefully this indicates the qemu child process has died. */
1969 if (errno == EINTR || errno == EAGAIN)
1972 perrorf (g, "read");
1976 xdrmem_create (&xdr, buf, 4, XDR_DECODE);
1977 xdr_uint32_t (&xdr, &flag);
1980 if (flag != GUESTFS_CANCEL_FLAG) {
1981 error (g, _("check_for_daemon_cancellation_or_eof: read 0x%x from daemon, expected 0x%x\n"),
1982 flag, GUESTFS_CANCEL_FLAG);
1989 /* This writes the whole N bytes of BUF to the daemon socket.
1991 * If the whole write is successful, it returns 0.
1992 * If there was an error, it returns -1.
1993 * If the daemon sent a cancellation message, it returns -2.
1995 * It also checks qemu stdout for log messages and passes those up
1996 * through log_message_cb.
1998 * It also checks for EOF (qemu died) and passes that up through the
1999 * child_cleanup function above.
2002 send_to_daemon (guestfs_h *g, const void *v_buf, size_t n)
2004 const char *buf = v_buf;
2010 "send_to_daemon: %p g->state = %d, n = %zu\n", g, g->state, n);
2015 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2016 FD_SET (g->sock, &rset); /* Read socket for cancellation & EOF. */
2017 FD_SET (g->sock, &wset); /* Write to socket to send the data. */
2019 int max_fd = MAX (g->sock, g->fd[1]);
2024 int r = select (max_fd+1, &rset2, &wset2, NULL, NULL);
2026 if (errno == EINTR || errno == EAGAIN)
2028 perrorf (g, "select");
2032 if (FD_ISSET (g->fd[1], &rset2)) {
2033 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2036 if (FD_ISSET (g->sock, &rset2)) {
2037 r = check_for_daemon_cancellation_or_eof (g, g->sock);
2041 if (FD_ISSET (g->sock, &wset2)) {
2042 r = write (g->sock, buf, n);
2044 if (errno == EINTR || errno == EAGAIN)
2046 perrorf (g, "write");
2047 if (errno == EPIPE) /* Disconnected from guest (RHBZ#508713). */
2059 /* This reads a single message, file chunk, launch flag or
2060 * cancellation flag from the daemon. If something was read, it
2061 * returns 0, otherwise -1.
2063 * Both size_rtn and buf_rtn must be passed by the caller as non-NULL.
2065 * *size_rtn returns the size of the returned message or it may be
2066 * GUESTFS_LAUNCH_FLAG or GUESTFS_CANCEL_FLAG.
2068 * *buf_rtn is returned containing the message (if any) or will be set
2069 * to NULL. *buf_rtn must be freed by the caller.
2071 * It also checks qemu stdout for log messages and passes those up
2072 * through log_message_cb.
2074 * It also checks for EOF (qemu died) and passes that up through the
2075 * child_cleanup function above.
2078 recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn)
2084 "recv_from_daemon: %p g->state = %d, size_rtn = %p, buf_rtn = %p\n",
2085 g, g->state, size_rtn, buf_rtn);
2089 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2090 FD_SET (g->sock, &rset); /* Read socket for data & EOF. */
2092 int max_fd = MAX (g->sock, g->fd[1]);
2098 /* nr is the size of the message, but we prime it as -4 because we
2099 * have to read the message length word first.
2103 while (nr < (ssize_t) *size_rtn) {
2105 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2107 if (errno == EINTR || errno == EAGAIN)
2109 perrorf (g, "select");
2115 if (FD_ISSET (g->fd[1], &rset2)) {
2116 if (read_log_message_or_eof (g, g->fd[1]) == -1) {
2122 if (FD_ISSET (g->sock, &rset2)) {
2123 if (nr < 0) { /* Have we read the message length word yet? */
2124 r = read (g->sock, lenbuf+nr+4, -nr);
2126 if (errno == EINTR || errno == EAGAIN)
2129 perrorf (g, "read");
2130 /* Under some circumstances we see "Connection reset by peer"
2131 * here when the child dies suddenly. Catch this and call
2132 * the cleanup function, same as for EOF.
2134 if (err == ECONNRESET)
2139 error (g, _("unexpected end of file when reading from daemon"));
2145 if (nr < 0) /* Still not got the whole length word. */
2149 xdrmem_create (&xdr, lenbuf, 4, XDR_DECODE);
2150 xdr_uint32_t (&xdr, size_rtn);
2153 if (*size_rtn == GUESTFS_LAUNCH_FLAG) {
2154 if (g->state != LAUNCHING)
2155 error (g, _("received magic signature from guestfsd, but in state %d"),
2159 if (g->launch_done_cb)
2160 g->launch_done_cb (g, g->launch_done_cb_data);
2164 else if (*size_rtn == GUESTFS_CANCEL_FLAG)
2166 /* If this happens, it's pretty bad and we've probably lost
2169 else if (*size_rtn > GUESTFS_MESSAGE_MAX) {
2170 error (g, _("message length (%u) > maximum possible size (%d)"),
2171 (unsigned) *size_rtn, GUESTFS_MESSAGE_MAX);
2175 /* Allocate the complete buffer, size now known. */
2176 *buf_rtn = safe_malloc (g, *size_rtn);
2180 size_t sizetoread = *size_rtn - nr;
2181 if (sizetoread > BUFSIZ) sizetoread = BUFSIZ;
2183 r = read (g->sock, (char *) (*buf_rtn) + nr, sizetoread);
2185 if (errno == EINTR || errno == EAGAIN)
2187 perrorf (g, "read");
2193 error (g, _("unexpected end of file when reading from daemon"));
2203 /* Got the full message, caller can start processing it. */
2204 #ifdef ENABLE_PACKET_DUMP
2208 for (i = 0; i < nr; i += 16) {
2209 printf ("%04zx: ", i);
2210 for (j = i; j < MIN (i+16, nr); ++j)
2211 printf ("%02x ", (*(unsigned char **)buf_rtn)[j]);
2212 for (; j < i+16; ++j)
2215 for (j = i; j < MIN (i+16, nr); ++j)
2216 if (c_isprint ((*(char **)buf_rtn)[j]))
2217 printf ("%c", (*(char **)buf_rtn)[j]);
2220 for (; j < i+16; ++j)
2230 /* This is very much like recv_from_daemon above, but g->sock is
2231 * a listening socket and we are accepting a new connection on
2232 * that socket instead of reading anything. Returns the newly
2236 accept_from_daemon (guestfs_h *g)
2242 "accept_from_daemon: %p g->state = %d\n", g, g->state);
2246 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2247 FD_SET (g->sock, &rset); /* Read socket for accept. */
2249 int max_fd = MAX (g->sock, g->fd[1]);
2252 while (sock == -1) {
2254 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2256 if (errno == EINTR || errno == EAGAIN)
2258 perrorf (g, "select");
2262 if (FD_ISSET (g->fd[1], &rset2)) {
2263 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2266 if (FD_ISSET (g->sock, &rset2)) {
2267 sock = accept (g->sock, NULL, NULL);
2269 if (errno == EINTR || errno == EAGAIN)
2271 perrorf (g, "accept");
2281 guestfs___send (guestfs_h *g, int proc_nr, xdrproc_t xdrp, char *args)
2283 struct guestfs_message_header hdr;
2286 int serial = g->msg_next_serial++;
2289 size_t msg_out_size;
2291 if (g->state != BUSY) {
2292 error (g, _("guestfs___send: state %d != BUSY"), g->state);
2296 /* We have to allocate this message buffer on the heap because
2297 * it is quite large (although will be mostly unused). We
2298 * can't allocate it on the stack because in some environments
2299 * we have quite limited stack space available, notably when
2300 * running in the JVM.
2302 msg_out = safe_malloc (g, GUESTFS_MESSAGE_MAX + 4);
2303 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MESSAGE_MAX, XDR_ENCODE);
2305 /* Serialize the header. */
2306 hdr.prog = GUESTFS_PROGRAM;
2307 hdr.vers = GUESTFS_PROTOCOL_VERSION;
2309 hdr.direction = GUESTFS_DIRECTION_CALL;
2310 hdr.serial = serial;
2311 hdr.status = GUESTFS_STATUS_OK;
2313 if (!xdr_guestfs_message_header (&xdr, &hdr)) {
2314 error (g, _("xdr_guestfs_message_header failed"));
2318 /* Serialize the args. If any, because some message types
2319 * have no parameters.
2322 if (!(*xdrp) (&xdr, args)) {
2323 error (g, _("dispatch failed to marshal args"));
2328 /* Get the actual length of the message, resize the buffer to match
2329 * the actual length, and write the length word at the beginning.
2331 len = xdr_getpos (&xdr);
2334 msg_out = safe_realloc (g, msg_out, len + 4);
2335 msg_out_size = len + 4;
2337 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2338 xdr_uint32_t (&xdr, &len);
2341 r = send_to_daemon (g, msg_out, msg_out_size);
2342 if (r == -2) /* Ignore stray daemon cancellations. */
2355 static int cancel = 0; /* XXX Implement file cancellation. */
2356 static int send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t len);
2357 static int send_file_data (guestfs_h *g, const char *buf, size_t len);
2358 static int send_file_cancellation (guestfs_h *g);
2359 static int send_file_complete (guestfs_h *g);
2365 * -2 daemon cancelled (we must read the error message)
2368 guestfs___send_file (guestfs_h *g, const char *filename)
2370 char buf[GUESTFS_MAX_CHUNK_SIZE];
2373 fd = open (filename, O_RDONLY);
2375 perrorf (g, "open: %s", filename);
2376 send_file_cancellation (g);
2377 /* Daemon sees cancellation and won't reply, so caller can
2383 /* Send file in chunked encoding. */
2385 r = read (fd, buf, sizeof buf);
2386 if (r == -1 && (errno == EINTR || errno == EAGAIN))
2389 err = send_file_data (g, buf, r);
2391 if (err == -2) /* daemon sent cancellation */
2392 send_file_cancellation (g);
2397 if (cancel) { /* cancel from either end */
2398 send_file_cancellation (g);
2403 perrorf (g, "read: %s", filename);
2404 send_file_cancellation (g);
2408 /* End of file, but before we send that, we need to close
2409 * the file and check for errors.
2411 if (close (fd) == -1) {
2412 perrorf (g, "close: %s", filename);
2413 send_file_cancellation (g);
2417 return send_file_complete (g);
2420 /* Send a chunk of file data. */
2422 send_file_data (guestfs_h *g, const char *buf, size_t len)
2424 return send_file_chunk (g, 0, buf, len);
2427 /* Send a cancellation message. */
2429 send_file_cancellation (guestfs_h *g)
2431 return send_file_chunk (g, 1, NULL, 0);
2434 /* Send a file complete chunk. */
2436 send_file_complete (guestfs_h *g)
2439 return send_file_chunk (g, 0, buf, 0);
2443 send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t buflen)
2447 guestfs_chunk chunk;
2450 size_t msg_out_size;
2452 if (g->state != BUSY) {
2453 error (g, _("send_file_chunk: state %d != READY"), g->state);
2457 /* Allocate the chunk buffer. Don't use the stack to avoid
2458 * excessive stack usage and unnecessary copies.
2460 msg_out = safe_malloc (g, GUESTFS_MAX_CHUNK_SIZE + 4 + 48);
2461 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MAX_CHUNK_SIZE + 48, XDR_ENCODE);
2463 /* Serialize the chunk. */
2464 chunk.cancel = cancel;
2465 chunk.data.data_len = buflen;
2466 chunk.data.data_val = (char *) buf;
2468 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2469 error (g, _("xdr_guestfs_chunk failed (buf = %p, buflen = %zu)"),
2475 len = xdr_getpos (&xdr);
2478 /* Reduce the size of the outgoing message buffer to the real length. */
2479 msg_out = safe_realloc (g, msg_out, len + 4);
2480 msg_out_size = len + 4;
2482 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2483 xdr_uint32_t (&xdr, &len);
2485 r = send_to_daemon (g, msg_out, msg_out_size);
2487 /* Did the daemon send a cancellation message? */
2490 fprintf (stderr, "got daemon cancellation\n");
2506 /* Receive a reply. */
2508 guestfs___recv (guestfs_h *g, const char *fn,
2509 guestfs_message_header *hdr,
2510 guestfs_message_error *err,
2511 xdrproc_t xdrp, char *ret)
2519 r = recv_from_daemon (g, &size, &buf);
2523 /* This can happen if a cancellation happens right at the end
2524 * of us sending a FileIn parameter to the daemon. Discard. The
2525 * daemon should send us an error message next.
2527 if (size == GUESTFS_CANCEL_FLAG)
2530 if (size == GUESTFS_LAUNCH_FLAG) {
2531 error (g, "%s: received unexpected launch flag from daemon when expecting reply", fn);
2535 xdrmem_create (&xdr, buf, size, XDR_DECODE);
2537 if (!xdr_guestfs_message_header (&xdr, hdr)) {
2538 error (g, "%s: failed to parse reply header", fn);
2543 if (hdr->status == GUESTFS_STATUS_ERROR) {
2544 if (!xdr_guestfs_message_error (&xdr, err)) {
2545 error (g, "%s: failed to parse reply error", fn);
2551 if (xdrp && ret && !xdrp (&xdr, ret)) {
2552 error (g, "%s: failed to parse reply", fn);
2564 /* Receive a file. */
2566 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2567 static ssize_t receive_file_data (guestfs_h *g, void **buf);
2570 guestfs___recv_file (guestfs_h *g, const char *filename)
2576 fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2578 perrorf (g, "open: %s", filename);
2582 /* Receive the file in chunked encoding. */
2583 while ((r = receive_file_data (g, &buf)) > 0) {
2584 if (xwrite (fd, buf, r) == -1) {
2585 perrorf (g, "%s: write", filename);
2593 error (g, _("%s: error in chunked encoding"), filename);
2597 if (close (fd) == -1) {
2598 perrorf (g, "close: %s", filename);
2605 /* Send cancellation message to daemon, then wait until it
2606 * cancels (just throwing away data).
2610 uint32_t flag = GUESTFS_CANCEL_FLAG;
2613 fprintf (stderr, "%s: waiting for daemon to acknowledge cancellation\n",
2616 xdrmem_create (&xdr, fbuf, sizeof fbuf, XDR_ENCODE);
2617 xdr_uint32_t (&xdr, &flag);
2620 if (xwrite (g->sock, fbuf, sizeof fbuf) == -1) {
2621 perrorf (g, _("write to daemon socket"));
2625 while (receive_file_data (g, NULL) > 0)
2626 ; /* just discard it */
2631 /* Receive a chunk of file data. */
2632 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2634 receive_file_data (guestfs_h *g, void **buf_r)
2640 guestfs_chunk chunk;
2642 r = recv_from_daemon (g, &len, &buf);
2644 error (g, _("receive_file_data: parse error in reply callback"));
2648 if (len == GUESTFS_LAUNCH_FLAG || len == GUESTFS_CANCEL_FLAG) {
2649 error (g, _("receive_file_data: unexpected flag received when reading file chunks"));
2653 memset (&chunk, 0, sizeof chunk);
2655 xdrmem_create (&xdr, buf, len, XDR_DECODE);
2656 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2657 error (g, _("failed to parse file chunk"));
2662 /* After decoding, the original buffer is no longer used. */
2666 error (g, _("file receive cancelled by daemon"));
2667 free (chunk.data.data_val);
2671 if (chunk.data.data_len == 0) { /* end of transfer */
2672 free (chunk.data.data_val);
2676 if (buf_r) *buf_r = chunk.data.data_val;
2677 else free (chunk.data.data_val); /* else caller frees */
2679 return chunk.data.data_len;