2 * Copyright (C) 2009-2010 Red Hat Inc.
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #define _BSD_SOURCE /* for mkdtemp, usleep */
34 #include <sys/select.h>
38 #include <rpc/types.h>
45 #ifdef HAVE_SYS_TYPES_H
46 #include <sys/types.h>
49 #ifdef HAVE_SYS_WAIT_H
53 #ifdef HAVE_SYS_SOCKET_H
54 #include <sys/socket.h>
61 #include <arpa/inet.h>
62 #include <netinet/in.h>
65 #include "glthread/lock.h"
66 #include "ignore-value.h"
69 #include "guestfs-internal.h"
70 #include "guestfs-internal-actions.h"
71 #include "guestfs_protocol.h"
75 #define _(str) dgettext(PACKAGE, (str))
76 //#define N_(str) dgettext(PACKAGE, (str))
82 #define error guestfs_error
83 #define perrorf guestfs_perrorf
84 #define safe_malloc guestfs_safe_malloc
85 #define safe_realloc guestfs_safe_realloc
86 #define safe_strdup guestfs_safe_strdup
87 //#define safe_memdup guestfs_safe_memdup
89 static void default_error_cb (guestfs_h *g, void *data, const char *msg);
90 static int send_to_daemon (guestfs_h *g, const void *v_buf, size_t n);
91 static int recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn);
92 static int accept_from_daemon (guestfs_h *g);
93 static int check_peer_euid (guestfs_h *g, int sock, uid_t *rtn);
94 static void close_handles (void);
95 static int qemu_supports (guestfs_h *g, const char *option);
97 #define UNIX_PATH_MAX 108
100 #define MAX(a,b) ((a)>(b)?(a):(b))
103 /* Also in guestfsd.c */
104 #define GUESTFWD_ADDR "10.0.2.4"
105 #define GUESTFWD_PORT "6666"
107 /* GuestFS handle and connection. */
108 enum state { CONFIG, LAUNCHING, READY, BUSY, NO_HANDLE };
112 struct guestfs_h *next; /* Linked list of open handles. */
114 /* State: see the state machine diagram in the man page guestfs(3). */
117 int fd[2]; /* Stdin/stdout of qemu. */
118 int sock; /* Daemon communications socket. */
119 pid_t pid; /* Qemu PID. */
120 pid_t recoverypid; /* Recovery process PID. */
122 struct timeval launch_t; /* The time that we called guestfs_launch. */
124 char *tmpdir; /* Temporary directory containing socket. */
126 char *qemu_help, *qemu_version; /* Output of qemu -help, qemu -version. */
128 char **cmdline; /* Qemu command line. */
137 char *path; /* Path to kernel, initrd. */
138 char *qemu; /* Qemu binary. */
139 char *append; /* Append to kernel command line. */
141 int memsize; /* Size of RAM (megabytes). */
143 int selinux; /* selinux enabled? */
148 guestfs_abort_cb abort_cb;
149 guestfs_error_handler_cb error_cb;
150 void * error_cb_data;
151 guestfs_log_message_cb log_message_cb;
152 void * log_message_cb_data;
153 guestfs_subprocess_quit_cb subprocess_quit_cb;
154 void * subprocess_quit_cb_data;
155 guestfs_launch_done_cb launch_done_cb;
156 void * launch_done_cb_data;
161 gl_lock_define_initialized (static, handles_lock);
162 static guestfs_h *handles = NULL;
163 static int atexit_handler_set = 0;
166 guestfs_create (void)
171 g = malloc (sizeof (*g));
174 memset (g, 0, sizeof (*g));
183 g->error_cb = default_error_cb;
184 g->error_cb_data = NULL;
186 g->recovery_proc = 1;
188 str = getenv ("LIBGUESTFS_DEBUG");
189 g->verbose = str != NULL && STREQ (str, "1");
191 str = getenv ("LIBGUESTFS_TRACE");
192 g->trace = str != NULL && STREQ (str, "1");
194 str = getenv ("LIBGUESTFS_PATH");
195 g->path = str != NULL ? strdup (str) : strdup (GUESTFS_DEFAULT_PATH);
196 if (!g->path) goto error;
198 str = getenv ("LIBGUESTFS_QEMU");
199 g->qemu = str != NULL ? strdup (str) : strdup (QEMU);
200 if (!g->qemu) goto error;
202 str = getenv ("LIBGUESTFS_APPEND");
204 g->append = strdup (str);
205 if (!g->append) goto error;
208 /* Choose a suitable memory size. Previously we tried to choose
209 * a minimal memory size, but this isn't really necessary since
210 * recent QEMU and KVM don't do anything nasty like locking
211 * memory into core any more. Thus we can safely choose a
212 * large, generous amount of memory, and it'll just get swapped
213 * on smaller systems.
215 str = getenv ("LIBGUESTFS_MEMSIZE");
217 if (sscanf (str, "%d", &g->memsize) != 1 || g->memsize <= 256) {
218 fprintf (stderr, "libguestfs: non-numeric or too small value for LIBGUESTFS_MEMSIZE\n");
224 /* Start with large serial numbers so they are easy to spot
225 * inside the protocol.
227 g->msg_next_serial = 0x00123400;
229 /* Link the handles onto a global list. */
230 gl_lock_lock (handles_lock);
233 if (!atexit_handler_set) {
234 atexit (close_handles);
235 atexit_handler_set = 1;
237 gl_lock_unlock (handles_lock);
240 fprintf (stderr, "new guestfs handle %p\n", g);
253 guestfs_close (guestfs_h *g)
259 if (g->state == NO_HANDLE) {
260 /* Not safe to call 'error' here, so ... */
261 fprintf (stderr, _("guestfs_close: called twice on the same handle\n"));
266 fprintf (stderr, "closing guestfs handle %p (state %d)\n", g, g->state);
268 /* Try to sync if autosync flag is set. */
269 if (g->autosync && g->state == READY) {
270 guestfs_umount_all (g);
274 /* Remove any handlers that might be called back before we kill the
277 g->log_message_cb = NULL;
279 if (g->state != CONFIG)
280 guestfs_kill_subprocess (g);
293 /* Wait for subprocess(es) to exit. */
294 waitpid (g->pid, NULL, 0);
295 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
297 /* Remove tmpfiles. */
299 snprintf (filename, sizeof filename, "%s/sock", g->tmpdir);
302 snprintf (filename, sizeof filename, "%s/initrd", g->tmpdir);
305 snprintf (filename, sizeof filename, "%s/kernel", g->tmpdir);
314 for (i = 0; i < g->cmdline_size; ++i)
315 free (g->cmdline[i]);
319 /* Mark the handle as dead before freeing it. */
320 g->state = NO_HANDLE;
322 gl_lock_lock (handles_lock);
326 for (gg = handles; gg->next != g; gg = gg->next)
330 gl_lock_unlock (handles_lock);
332 free (g->last_error);
337 free (g->qemu_version);
341 /* Close all open handles (called from atexit(3)). */
345 while (handles) guestfs_close (handles);
349 guestfs_last_error (guestfs_h *g)
351 return g->last_error;
355 set_last_error (guestfs_h *g, const char *msg)
357 free (g->last_error);
358 g->last_error = strdup (msg);
362 default_error_cb (guestfs_h *g, void *data, const char *msg)
364 fprintf (stderr, _("libguestfs: error: %s\n"), msg);
368 guestfs_error (guestfs_h *g, const char *fs, ...)
374 int err = vasprintf (&msg, fs, args);
379 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
380 set_last_error (g, msg);
386 guestfs_perrorf (guestfs_h *g, const char *fs, ...)
393 int err = vasprintf (&msg, fs, args);
398 #if !defined(_GNU_SOURCE) || defined(__APPLE__)
400 strerror_r (errnum, buf, sizeof buf);
404 buf = strerror_r (errnum, _buf, sizeof _buf);
407 msg = safe_realloc (g, msg, strlen (msg) + 2 + strlen (buf) + 1);
411 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
412 set_last_error (g, msg);
418 guestfs_safe_malloc (guestfs_h *g, size_t nbytes)
420 void *ptr = malloc (nbytes);
421 if (nbytes > 0 && !ptr) g->abort_cb ();
425 /* Return 1 if an array of N objects, each of size S, cannot exist due
426 to size arithmetic overflow. S must be positive and N must be
427 nonnegative. This is a macro, not an inline function, so that it
428 works correctly even when SIZE_MAX < N.
430 By gnulib convention, SIZE_MAX represents overflow in size
431 calculations, so the conservative dividend to use here is
432 SIZE_MAX - 1, since SIZE_MAX might represent an overflowed value.
433 However, malloc (SIZE_MAX) fails on all known hosts where
434 sizeof (ptrdiff_t) <= sizeof (size_t), so do not bother to test for
435 exactly-SIZE_MAX allocations on such hosts; this avoids a test and
436 branch when S is known to be 1. */
437 # define xalloc_oversized(n, s) \
438 ((size_t) (sizeof (ptrdiff_t) <= sizeof (size_t) ? -1 : -2) / (s) < (n))
440 /* Technically we should add an autoconf test for this, testing for the desired
441 functionality, like what's done in gnulib, but for now, this is fine. */
442 #if defined(__GLIBC__)
443 #define HAVE_GNU_CALLOC (__GLIBC__ >= 2)
445 #define HAVE_GNU_CALLOC 0
448 /* Allocate zeroed memory for N elements of S bytes, with error
449 checking. S must be nonzero. */
451 guestfs_safe_calloc (guestfs_h *g, size_t n, size_t s)
453 /* From gnulib's calloc function in xmalloc.c. */
455 /* Test for overflow, since some calloc implementations don't have
456 proper overflow checks. But omit overflow and size-zero tests if
457 HAVE_GNU_CALLOC, since GNU calloc catches overflow and never
458 returns NULL if successful. */
459 if ((! HAVE_GNU_CALLOC && xalloc_oversized (n, s))
460 || (! (p = calloc (n, s)) && (HAVE_GNU_CALLOC || n != 0)))
466 guestfs_safe_realloc (guestfs_h *g, void *ptr, int nbytes)
468 void *p = realloc (ptr, nbytes);
469 if (nbytes > 0 && !p) g->abort_cb ();
474 guestfs_safe_strdup (guestfs_h *g, const char *str)
476 char *s = strdup (str);
477 if (!s) g->abort_cb ();
482 guestfs_safe_memdup (guestfs_h *g, void *ptr, size_t size)
484 void *p = malloc (size);
485 if (!p) g->abort_cb ();
486 memcpy (p, ptr, size);
491 xwrite (int fd, const void *v_buf, size_t len)
493 const char *buf = v_buf;
497 r = write (fd, buf, len);
509 guestfs_set_out_of_memory_handler (guestfs_h *g, guestfs_abort_cb cb)
515 guestfs_get_out_of_memory_handler (guestfs_h *g)
521 guestfs_set_error_handler (guestfs_h *g, guestfs_error_handler_cb cb, void *data)
524 g->error_cb_data = data;
527 guestfs_error_handler_cb
528 guestfs_get_error_handler (guestfs_h *g, void **data_rtn)
530 if (data_rtn) *data_rtn = g->error_cb_data;
535 guestfs__set_verbose (guestfs_h *g, int v)
542 guestfs__get_verbose (guestfs_h *g)
548 guestfs__set_autosync (guestfs_h *g, int a)
555 guestfs__get_autosync (guestfs_h *g)
561 guestfs__set_path (guestfs_h *g, const char *path)
568 safe_strdup (g, GUESTFS_DEFAULT_PATH) : safe_strdup (g, path);
573 guestfs__get_path (guestfs_h *g)
579 guestfs__set_qemu (guestfs_h *g, const char *qemu)
584 g->qemu = qemu == NULL ? safe_strdup (g, QEMU) : safe_strdup (g, qemu);
589 guestfs__get_qemu (guestfs_h *g)
595 guestfs__set_append (guestfs_h *g, const char *append)
600 g->append = append ? safe_strdup (g, append) : NULL;
605 guestfs__get_append (guestfs_h *g)
611 guestfs__set_memsize (guestfs_h *g, int memsize)
613 g->memsize = memsize;
618 guestfs__get_memsize (guestfs_h *g)
624 guestfs__set_selinux (guestfs_h *g, int selinux)
626 g->selinux = selinux;
631 guestfs__get_selinux (guestfs_h *g)
637 guestfs__get_pid (guestfs_h *g)
642 error (g, "get_pid: no qemu subprocess");
647 struct guestfs_version *
648 guestfs__version (guestfs_h *g)
650 struct guestfs_version *r;
652 r = safe_malloc (g, sizeof *r);
653 r->major = PACKAGE_VERSION_MAJOR;
654 r->minor = PACKAGE_VERSION_MINOR;
655 r->release = PACKAGE_VERSION_RELEASE;
656 r->extra = safe_strdup (g, PACKAGE_VERSION_EXTRA);
661 guestfs__set_trace (guestfs_h *g, int t)
668 guestfs__get_trace (guestfs_h *g)
674 guestfs__set_direct (guestfs_h *g, int d)
681 guestfs__get_direct (guestfs_h *g)
687 guestfs__set_recovery_proc (guestfs_h *g, int f)
689 g->recovery_proc = !!f;
694 guestfs__get_recovery_proc (guestfs_h *g)
696 return g->recovery_proc;
699 /* Add a string to the current command line. */
701 incr_cmdline_size (guestfs_h *g)
703 if (g->cmdline == NULL) {
704 /* g->cmdline[0] is reserved for argv[0], set in guestfs_launch. */
706 g->cmdline = safe_malloc (g, sizeof (char *));
707 g->cmdline[0] = NULL;
711 g->cmdline = safe_realloc (g, g->cmdline, sizeof (char *) * g->cmdline_size);
715 add_cmdline (guestfs_h *g, const char *str)
717 if (g->state != CONFIG) {
719 _("command line cannot be altered after qemu subprocess launched"));
723 incr_cmdline_size (g);
724 g->cmdline[g->cmdline_size-1] = safe_strdup (g, str);
729 guestfs__config (guestfs_h *g,
730 const char *qemu_param, const char *qemu_value)
732 if (qemu_param[0] != '-') {
733 error (g, _("guestfs_config: parameter must begin with '-' character"));
737 /* A bit fascist, but the user will probably break the extra
738 * parameters that we add if they try to set any of these.
740 if (STREQ (qemu_param, "-kernel") ||
741 STREQ (qemu_param, "-initrd") ||
742 STREQ (qemu_param, "-nographic") ||
743 STREQ (qemu_param, "-serial") ||
744 STREQ (qemu_param, "-full-screen") ||
745 STREQ (qemu_param, "-std-vga") ||
746 STREQ (qemu_param, "-vnc")) {
747 error (g, _("guestfs_config: parameter '%s' isn't allowed"), qemu_param);
751 if (add_cmdline (g, qemu_param) != 0) return -1;
753 if (qemu_value != NULL) {
754 if (add_cmdline (g, qemu_value) != 0) return -1;
761 guestfs__add_drive_with_if (guestfs_h *g, const char *filename,
762 const char *drive_if)
764 size_t len = strlen (filename) + 64;
767 if (strchr (filename, ',') != NULL) {
768 error (g, _("filename cannot contain ',' (comma) character"));
772 /* cache=off improves reliability in the event of a host crash.
774 * However this option causes qemu to try to open the file with
775 * O_DIRECT. This fails on some filesystem types (notably tmpfs).
776 * So we check if we can open the file with or without O_DIRECT,
777 * and use cache=off (or not) accordingly.
779 * This test also checks for the presence of the file, which
780 * is a documented semantic of this interface.
782 int fd = open (filename, O_RDONLY|O_DIRECT);
785 snprintf (buf, len, "file=%s,cache=off,if=%s", filename, drive_if);
787 fd = open (filename, O_RDONLY);
790 snprintf (buf, len, "file=%s,if=%s", filename, drive_if);
792 perrorf (g, "%s", filename);
797 return guestfs__config (g, "-drive", buf);
801 guestfs__add_drive_ro_with_if (guestfs_h *g, const char *filename,
802 const char *drive_if)
804 if (strchr (filename, ',') != NULL) {
805 error (g, _("filename cannot contain ',' (comma) character"));
809 if (access (filename, F_OK) == -1) {
810 perrorf (g, "%s", filename);
814 if (qemu_supports (g, NULL) == -1)
817 /* Only SCSI and virtio drivers support readonly mode.
818 * This is only supported as a QEMU feature since 2010/01.
821 if ((STREQ (drive_if, "scsi") || STREQ (drive_if, "virtio")) &&
822 qemu_supports (g, "readonly=on"))
825 size_t len = strlen (filename) + 100;
828 snprintf (buf, len, "file=%s,snapshot=on,%sif=%s",
830 supports_ro ? "readonly=on," : "",
833 return guestfs__config (g, "-drive", buf);
837 guestfs__add_drive (guestfs_h *g, const char *filename)
839 return guestfs__add_drive_with_if (g, filename, DRIVE_IF);
843 guestfs__add_drive_ro (guestfs_h *g, const char *filename)
845 return guestfs__add_drive_ro_with_if (g, filename, DRIVE_IF);
849 guestfs__add_cdrom (guestfs_h *g, const char *filename)
851 if (strchr (filename, ',') != NULL) {
852 error (g, _("filename cannot contain ',' (comma) character"));
856 if (access (filename, F_OK) == -1) {
857 perrorf (g, "%s", filename);
861 return guestfs__config (g, "-cdrom", filename);
864 /* Returns true iff file is contained in dir. */
866 dir_contains_file (const char *dir, const char *file)
868 int dirlen = strlen (dir);
869 int filelen = strlen (file);
870 int len = dirlen+filelen+2;
873 snprintf (path, len, "%s/%s", dir, file);
874 return access (path, F_OK) == 0;
877 /* Returns true iff every listed file is contained in 'dir'. */
879 dir_contains_files (const char *dir, ...)
884 va_start (args, dir);
885 while ((file = va_arg (args, const char *)) != NULL) {
886 if (!dir_contains_file (dir, file)) {
895 static void print_timestamped_message (guestfs_h *g, const char *fs, ...);
896 static int build_supermin_appliance (guestfs_h *g, const char *path, char **kernel, char **initrd);
897 static int is_openable (guestfs_h *g, const char *path, int flags);
898 static void print_cmdline (guestfs_h *g);
900 static const char *kernel_name = "vmlinuz." REPO "." host_cpu;
901 static const char *initrd_name = "initramfs." REPO "." host_cpu ".img";
902 static const char *supermin_name =
903 "initramfs." REPO "." host_cpu ".supermin.img";
904 static const char *supermin_hostfiles_name =
905 "initramfs." REPO "." host_cpu ".supermin.hostfiles";
908 guestfs__launch (guestfs_h *g)
911 char dir_template[PATH_MAX];
916 char *path, *pelem, *pend;
917 char *kernel = NULL, *initrd = NULL;
918 int null_vmchannel_sock;
920 struct sockaddr_un addr;
922 /* Start the clock ... */
923 gettimeofday (&g->launch_t, NULL);
931 tmpdir = getenv ("TMPDIR") ? : tmpdir;
932 snprintf (dir_template, sizeof dir_template, "%s/libguestfsXXXXXX", tmpdir);
936 error (g, _("you must call guestfs_add_drive before guestfs_launch"));
940 if (g->state != CONFIG) {
941 error (g, _("qemu has already been launched"));
945 /* Make the temporary directory. */
947 g->tmpdir = safe_strdup (g, dir_template);
948 if (mkdtemp (g->tmpdir) == NULL) {
949 perrorf (g, _("%s: cannot create temporary directory"), dir_template);
954 /* First search g->path for the supermin appliance, and try to
955 * synthesize a kernel and initrd from that. If it fails, we
956 * try the path search again looking for a backup ordinary
959 pelem = path = safe_strdup (g, g->path);
961 pend = strchrnul (pelem, ':');
962 pmore = *pend == ':';
966 /* Empty element of "." means cwd. */
967 if (len == 0 || (len == 1 && *pelem == '.')) {
970 "looking for supermin appliance in current directory\n");
971 if (dir_contains_files (".",
972 supermin_name, supermin_hostfiles_name,
973 "kmod.whitelist", NULL)) {
974 if (build_supermin_appliance (g, ".", &kernel, &initrd) == -1)
979 /* Look at <path>/supermin* etc. */
982 fprintf (stderr, "looking for supermin appliance in %s\n", pelem);
984 if (dir_contains_files (pelem,
985 supermin_name, supermin_hostfiles_name,
986 "kmod.whitelist", NULL)) {
987 if (build_supermin_appliance (g, pelem, &kernel, &initrd) == -1)
998 if (kernel == NULL || initrd == NULL) {
999 /* Search g->path for the kernel and initrd. */
1000 pelem = path = safe_strdup (g, g->path);
1002 pend = strchrnul (pelem, ':');
1003 pmore = *pend == ':';
1007 /* Empty element or "." means cwd. */
1008 if (len == 0 || (len == 1 && *pelem == '.')) {
1011 "looking for appliance in current directory\n");
1012 if (dir_contains_files (".", kernel_name, initrd_name, NULL)) {
1013 kernel = safe_strdup (g, kernel_name);
1014 initrd = safe_strdup (g, initrd_name);
1018 /* Look at <path>/kernel etc. */
1021 fprintf (stderr, "looking for appliance in %s\n", pelem);
1023 if (dir_contains_files (pelem, kernel_name, initrd_name, NULL)) {
1024 kernel = safe_malloc (g, len + strlen (kernel_name) + 2);
1025 initrd = safe_malloc (g, len + strlen (initrd_name) + 2);
1026 sprintf (kernel, "%s/%s", pelem, kernel_name);
1027 sprintf (initrd, "%s/%s", pelem, initrd_name);
1038 if (kernel == NULL || initrd == NULL) {
1039 error (g, _("cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)"),
1040 kernel_name, initrd_name, g->path);
1045 print_timestamped_message (g, "begin testing qemu features");
1047 /* Get qemu help text and version. */
1048 if (qemu_supports (g, NULL) == -1)
1051 /* Choose which vmchannel implementation to use. */
1052 if (qemu_supports (g, "-net user")) {
1053 /* The "null vmchannel" implementation. Requires SLIRP (user mode
1054 * networking in qemu) but no other vmchannel support. The daemon
1055 * will connect back to a random port number on localhost.
1057 struct sockaddr_in addr;
1058 socklen_t addrlen = sizeof addr;
1060 g->sock = socket (AF_INET, SOCK_STREAM, IPPROTO_TCP);
1061 if (g->sock == -1) {
1062 perrorf (g, "socket");
1065 addr.sin_family = AF_INET;
1066 addr.sin_port = htons (0);
1067 addr.sin_addr.s_addr = htonl (INADDR_LOOPBACK);
1068 if (bind (g->sock, (struct sockaddr *) &addr, addrlen) == -1) {
1069 perrorf (g, "bind");
1073 if (listen (g->sock, 256) == -1) {
1074 perrorf (g, "listen");
1078 if (getsockname (g->sock, (struct sockaddr *) &addr, &addrlen) == -1) {
1079 perrorf (g, "getsockname");
1083 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1084 perrorf (g, "fcntl");
1088 null_vmchannel_sock = ntohs (addr.sin_port);
1090 fprintf (stderr, "null_vmchannel_sock = %d\n", null_vmchannel_sock);
1092 /* Using some vmchannel impl. We need to create a local Unix
1093 * domain socket for qemu to use.
1095 snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
1097 null_vmchannel_sock = 0;
1101 if (pipe (wfd) == -1 || pipe (rfd) == -1) {
1102 perrorf (g, "pipe");
1108 print_timestamped_message (g, "finished testing qemu features");
1112 perrorf (g, "fork");
1122 if (r == 0) { /* Child (qemu). */
1124 const char *vmchannel = NULL;
1126 /* Set up the full command line. Do this in the subprocess so we
1127 * don't need to worry about cleaning up.
1129 g->cmdline[0] = g->qemu;
1131 /* qemu sometimes needs this option to enable hardware
1132 * virtualization, but some versions of 'qemu-kvm' will use KVM
1133 * regardless (even where this option appears in the help text).
1134 * It is rumoured that there are versions of qemu where supplying
1135 * this option when hardware virtualization is not available will
1136 * cause qemu to fail, so we we have to check at least that
1137 * /dev/kvm is openable. That's not reliable, since /dev/kvm
1138 * might be openable by qemu but not by us (think: SELinux) in
1139 * which case the user would not get hardware virtualization,
1140 * although at least shouldn't fail. A giant clusterfuck with the
1141 * qemu command line, again.
1143 if (qemu_supports (g, "-enable-kvm") &&
1144 is_openable (g, "/dev/kvm", O_RDWR))
1145 add_cmdline (g, "-enable-kvm");
1147 /* Newer versions of qemu (from around 2009/12) changed the
1148 * behaviour of monitors so that an implicit '-monitor stdio' is
1149 * assumed if we are in -nographic mode and there is no other
1150 * -monitor option. Only a single stdio device is allowed, so
1151 * this broke the '-serial stdio' option. There is a new flag
1152 * called -nodefaults which gets rid of all this default crud, so
1153 * let's use that to avoid this and any future surprises.
1155 if (qemu_supports (g, "-nodefaults"))
1156 add_cmdline (g, "-nodefaults");
1158 add_cmdline (g, "-nographic");
1159 add_cmdline (g, "-serial");
1160 add_cmdline (g, "stdio");
1162 snprintf (buf, sizeof buf, "%d", g->memsize);
1163 add_cmdline (g, "-m");
1164 add_cmdline (g, buf);
1166 /* Force exit instead of reboot on panic */
1167 add_cmdline (g, "-no-reboot");
1169 /* These options recommended by KVM developers to improve reliability. */
1170 if (qemu_supports (g, "-no-hpet"))
1171 add_cmdline (g, "-no-hpet");
1173 if (qemu_supports (g, "-rtc-td-hack"))
1174 add_cmdline (g, "-rtc-td-hack");
1176 /* If qemu has SLIRP (user mode network) enabled then we can get
1177 * away with "no vmchannel", where we just connect back to a random
1180 if (null_vmchannel_sock) {
1181 add_cmdline (g, "-net");
1182 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1184 snprintf (buf, sizeof buf,
1185 "guestfs_vmchannel=tcp:10.0.2.2:%d", null_vmchannel_sock);
1186 vmchannel = strdup (buf);
1189 /* New-style -net user,guestfwd=... syntax for guestfwd. See:
1191 * http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=c92ef6a22d3c71538fcc48fb61ad353f7ba03b62
1193 * The original suggested format doesn't work, see:
1195 * http://lists.gnu.org/archive/html/qemu-devel/2009-07/msg01654.html
1197 * However Gerd Hoffman privately suggested to me using -chardev
1198 * instead, which does work.
1200 else if (qemu_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) {
1201 snprintf (buf, sizeof buf,
1202 "socket,id=guestfsvmc,path=%s,server,nowait", unixsock);
1204 add_cmdline (g, "-chardev");
1205 add_cmdline (g, buf);
1207 snprintf (buf, sizeof buf,
1208 "user,vlan=0,net=10.0.2.0/8,"
1209 "guestfwd=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT
1210 "-chardev:guestfsvmc");
1212 add_cmdline (g, "-net");
1213 add_cmdline (g, buf);
1215 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1218 /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel
1219 * syntax, or if not then we'll get a quick failure.
1222 snprintf (buf, sizeof buf,
1223 "channel," GUESTFWD_PORT ":unix:%s,server,nowait", unixsock);
1225 add_cmdline (g, "-net");
1226 add_cmdline (g, buf);
1227 add_cmdline (g, "-net");
1228 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1230 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1232 add_cmdline (g, "-net");
1233 add_cmdline (g, "nic,model=" NET_IF ",vlan=0");
1235 #define LINUX_CMDLINE \
1236 "panic=1 " /* force kernel to panic if daemon exits */ \
1237 "console=ttyS0 " /* serial console */ \
1238 "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \
1239 "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \
1240 "acpi=off " /* we don't need ACPI, turn it off */ \
1241 "printk.time=1 " /* display timestamp before kernel messages */ \
1242 "cgroup_disable=memory " /* saves us about 5 MB of RAM */
1244 /* Linux kernel command line. */
1245 snprintf (buf, sizeof buf,
1247 "%s " /* (selinux) */
1248 "%s " /* (vmchannel) */
1249 "%s " /* (verbose) */
1250 "%s", /* (append) */
1251 g->selinux ? "selinux=1 enforcing=0" : "selinux=0",
1252 vmchannel ? vmchannel : "",
1253 g->verbose ? "guestfs_verbose=1" : "",
1254 g->append ? g->append : "");
1256 add_cmdline (g, "-kernel");
1257 add_cmdline (g, (char *) kernel);
1258 add_cmdline (g, "-initrd");
1259 add_cmdline (g, (char *) initrd);
1260 add_cmdline (g, "-append");
1261 add_cmdline (g, buf);
1263 /* Finish off the command line. */
1264 incr_cmdline_size (g);
1265 g->cmdline[g->cmdline_size-1] = NULL;
1271 /* Set up stdin, stdout. */
1277 if (dup (wfd[0]) == -1) {
1279 perror ("dup failed");
1282 if (dup (rfd[1]) == -1)
1290 /* Set up a new process group, so we can signal this process
1291 * and all subprocesses (eg. if qemu is really a shell script).
1296 setenv ("LC_ALL", "C", 1);
1298 execv (g->qemu, g->cmdline); /* Run qemu. */
1303 /* Parent (library). */
1311 /* Fork the recovery process off which will kill qemu if the parent
1312 * process fails to do so (eg. if the parent segfaults).
1314 g->recoverypid = -1;
1315 if (g->recovery_proc) {
1318 pid_t qemu_pid = g->pid;
1319 pid_t parent_pid = getppid ();
1321 /* Writing to argv is hideously complicated and error prone. See:
1322 * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain
1325 /* Loop around waiting for one or both of the other processes to
1326 * disappear. It's fair to say this is very hairy. The PIDs that
1327 * we are looking at might be reused by another process. We are
1328 * effectively polling. Is the cure worse than the disease?
1331 if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */
1333 if (kill (parent_pid, 0) == -1) {
1334 /* Parent's gone away, qemu still around, so kill qemu. */
1342 /* Don't worry, if the fork failed, this will be -1. The recovery
1343 * process isn't essential.
1349 /* Close the other ends of the pipe. */
1353 if (fcntl (wfd[1], F_SETFL, O_NONBLOCK) == -1 ||
1354 fcntl (rfd[0], F_SETFL, O_NONBLOCK) == -1) {
1355 perrorf (g, "fcntl");
1359 g->fd[0] = wfd[1]; /* stdin of child */
1360 g->fd[1] = rfd[0]; /* stdout of child */
1362 g->fd[0] = open ("/dev/null", O_RDWR);
1363 if (g->fd[0] == -1) {
1364 perrorf (g, "open /dev/null");
1367 g->fd[1] = dup (g->fd[0]);
1368 if (g->fd[1] == -1) {
1375 if (null_vmchannel_sock) {
1379 /* Null vmchannel implementation: We listen on g->sock for a
1380 * connection. The connection could come from any local process
1381 * so we must check it comes from the appliance (or at least
1382 * from our UID) for security reasons.
1384 while (sock == -1) {
1385 sock = accept_from_daemon (g);
1389 if (check_peer_euid (g, sock, &uid) == -1)
1391 if (uid != geteuid ()) {
1393 "libguestfs: warning: unexpected connection from UID %d to port %d\n",
1394 uid, null_vmchannel_sock);
1400 if (fcntl (sock, F_SETFL, O_NONBLOCK) == -1) {
1401 perrorf (g, "fcntl");
1408 /* Other vmchannel. Open the Unix socket.
1410 * The vmchannel implementation that got merged with qemu sucks in
1411 * a number of ways. Both ends do connect(2), which means that no
1412 * one knows what, if anything, is connected to the other end, or
1413 * if it becomes disconnected. Even worse, we have to wait some
1414 * indeterminate time for qemu to create the socket and connect to
1415 * it (which happens very early in qemu's start-up), so any code
1416 * that uses vmchannel is inherently racy. Hence this silly loop.
1418 g->sock = socket (AF_UNIX, SOCK_STREAM, 0);
1419 if (g->sock == -1) {
1420 perrorf (g, "socket");
1424 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1425 perrorf (g, "fcntl");
1429 addr.sun_family = AF_UNIX;
1430 strncpy (addr.sun_path, unixsock, UNIX_PATH_MAX);
1431 addr.sun_path[UNIX_PATH_MAX-1] = '\0';
1434 /* Always sleep at least once to give qemu a small chance to start up. */
1437 r = connect (g->sock, (struct sockaddr *) &addr, sizeof addr);
1438 if ((r == -1 && errno == EINPROGRESS) || r == 0)
1441 if (errno != ENOENT)
1442 perrorf (g, "connect");
1447 error (g, _("failed to connect to vmchannel socket"));
1453 g->state = LAUNCHING;
1455 /* Wait for qemu to start and to connect back to us via vmchannel and
1456 * send the GUESTFS_LAUNCH_FLAG message.
1460 r = recv_from_daemon (g, &size, &buf);
1463 if (r == -1) return -1;
1465 if (size != GUESTFS_LAUNCH_FLAG) {
1466 error (g, _("guestfs_launch failed, see earlier error messages"));
1471 print_timestamped_message (g, "appliance is up");
1473 /* This is possible in some really strange situations, such as
1474 * guestfsd starts up OK but then qemu immediately exits. Check for
1475 * it because the caller is probably expecting to be able to send
1476 * commands after this function returns.
1478 if (g->state != READY) {
1479 error (g, _("qemu launched and contacted daemon, but state != READY"));
1491 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1492 waitpid (g->pid, NULL, 0);
1493 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1498 memset (&g->launch_t, 0, sizeof g->launch_t);
1511 /* This function is used to print the qemu command line before it gets
1512 * executed, when in verbose mode.
1515 print_cmdline (guestfs_h *g)
1520 while (g->cmdline[i]) {
1521 if (g->cmdline[i][0] == '-') /* -option starts a new line */
1522 fprintf (stderr, " \\\n ");
1524 if (i > 0) fputc (' ', stderr);
1526 /* Does it need shell quoting? This only deals with simple cases. */
1527 needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]);
1529 if (needs_quote) fputc ('\'', stderr);
1530 fprintf (stderr, "%s", g->cmdline[i]);
1531 if (needs_quote) fputc ('\'', stderr);
1535 fputc ('\n', stderr);
1538 /* This function does the hard work of building the supermin appliance
1539 * on the fly. 'path' is the directory containing the control files.
1540 * 'kernel' and 'initrd' are where we will return the names of the
1541 * kernel and initrd (only initrd is built). The work is done by
1542 * an external script. We just tell it where to put the result.
1545 build_supermin_appliance (guestfs_h *g, const char *path,
1546 char **kernel, char **initrd)
1552 print_timestamped_message (g, "begin building supermin appliance");
1554 len = strlen (g->tmpdir);
1555 *kernel = safe_malloc (g, len + 8);
1556 snprintf (*kernel, len+8, "%s/kernel", g->tmpdir);
1557 *initrd = safe_malloc (g, len + 8);
1558 snprintf (*initrd, len+8, "%s/initrd", g->tmpdir);
1560 snprintf (cmd, sizeof cmd,
1562 "libguestfs-supermin-helper%s '%s' " host_cpu " " REPO " %s %s",
1564 g->verbose ? " --verbose" : "",
1565 path, *kernel, *initrd);
1567 print_timestamped_message (g, "%s", cmd);
1570 if (r == -1 || WEXITSTATUS(r) != 0) {
1571 error (g, _("external command failed: %s"), cmd);
1574 *kernel = *initrd = NULL;
1579 print_timestamped_message (g, "finished building supermin appliance");
1584 /* Compute Y - X and return the result in milliseconds.
1585 * Approximately the same as this code:
1586 * http://www.mpp.mpg.de/~huber/util/timevaldiff.c
1589 timeval_diff (const struct timeval *x, const struct timeval *y)
1593 msec = (y->tv_sec - x->tv_sec) * 1000;
1594 msec += (y->tv_usec - x->tv_usec) / 1000;
1599 print_timestamped_message (guestfs_h *g, const char *fs, ...)
1606 va_start (args, fs);
1607 err = vasprintf (&msg, fs, args);
1610 if (err < 0) return;
1612 gettimeofday (&tv, NULL);
1614 fprintf (stderr, "[%05" PRIi64 "ms] %s\n",
1615 timeval_diff (&g->launch_t, &tv), msg);
1620 static int read_all (guestfs_h *g, FILE *fp, char **ret);
1622 /* Test qemu binary (or wrapper) runs, and do 'qemu -help' and
1623 * 'qemu -version' so we know what options this qemu supports and
1627 test_qemu (guestfs_h *g)
1632 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -nographic -help", g->qemu);
1634 fp = popen (cmd, "r");
1635 /* qemu -help should always work (qemu -version OTOH wasn't
1636 * supported by qemu 0.9). If this command doesn't work then it
1637 * probably indicates that the qemu binary is missing.
1640 /* XXX This error is never printed, even if the qemu binary
1641 * doesn't exist. Why?
1644 perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd);
1648 if (read_all (g, fp, &g->qemu_help) == -1)
1651 if (pclose (fp) == -1)
1654 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -nographic -version 2>/dev/null",
1657 fp = popen (cmd, "r");
1659 /* Intentionally ignore errors. */
1660 read_all (g, fp, &g->qemu_version);
1668 read_all (guestfs_h *g, FILE *fp, char **ret)
1675 *ret = safe_realloc (g, *ret, n + 1);
1680 *ret = safe_realloc (g, *ret, n + BUFSIZ);
1682 r = fread (p, 1, BUFSIZ, fp);
1684 perrorf (g, "read");
1691 /* Test if option is supported by qemu command line (just by grepping
1694 * The first time this is used, it has to run the external qemu
1695 * binary. If that fails, it returns -1.
1697 * To just do the first-time run of the qemu binary, call this with
1698 * option == NULL, in which case it will return -1 if there was an
1702 qemu_supports (guestfs_h *g, const char *option)
1704 if (!g->qemu_help) {
1705 if (test_qemu (g) == -1)
1712 return strstr (g->qemu_help, option) != NULL;
1715 /* Check if a file can be opened. */
1717 is_openable (guestfs_h *g, const char *path, int flags)
1719 int fd = open (path, flags);
1729 /* Check the peer effective UID for a TCP socket. Ideally we'd like
1730 * SO_PEERCRED for a loopback TCP socket. This isn't possible on
1731 * Linux (but it is on Solaris!) so we read /proc/net/tcp instead.
1734 check_peer_euid (guestfs_h *g, int sock, uid_t *rtn)
1736 struct sockaddr_in peer;
1737 socklen_t addrlen = sizeof peer;
1739 if (getpeername (sock, (struct sockaddr *) &peer, &addrlen) == -1) {
1740 perrorf (g, "getpeername");
1744 if (peer.sin_family != AF_INET ||
1745 ntohl (peer.sin_addr.s_addr) != INADDR_LOOPBACK) {
1746 error (g, "check_peer_euid: unexpected connection from non-IPv4, non-loopback peer (family = %d, addr = %s)",
1747 peer.sin_family, inet_ntoa (peer.sin_addr));
1751 struct sockaddr_in our;
1752 addrlen = sizeof our;
1753 if (getsockname (sock, (struct sockaddr *) &our, &addrlen) == -1) {
1754 perrorf (g, "getsockname");
1758 FILE *fp = fopen ("/proc/net/tcp", "r");
1760 perrorf (g, "/proc/net/tcp");
1765 if (fgets (line, sizeof line, fp) == NULL) { /* Drop first line. */
1766 error (g, "unexpected end of file in /proc/net/tcp");
1771 while (fgets (line, sizeof line, fp) != NULL) {
1772 unsigned line_our_addr, line_our_port, line_peer_addr, line_peer_port;
1773 int dummy0, dummy1, dummy2, dummy3, dummy4, dummy5, dummy6;
1776 if (sscanf (line, "%d:%08X:%04X %08X:%04X %02X %08X:%08X %02X:%08X %08X %d",
1778 &line_our_addr, &line_our_port,
1779 &line_peer_addr, &line_peer_port,
1780 &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, &dummy6,
1782 /* Note about /proc/net/tcp: local_address and rem_address are
1783 * always in network byte order. However the port part is
1784 * always in host byte order.
1786 * The sockname and peername that we got above are in network
1787 * byte order. So we have to byte swap the port but not the
1790 if (line_our_addr == our.sin_addr.s_addr &&
1791 line_our_port == ntohs (our.sin_port) &&
1792 line_peer_addr == peer.sin_addr.s_addr &&
1793 line_peer_port == ntohs (peer.sin_port)) {
1801 error (g, "check_peer_euid: no matching TCP connection found in /proc/net/tcp");
1806 /* You had to call this function after launch in versions <= 1.0.70,
1807 * but it is now a no-op.
1810 guestfs__wait_ready (guestfs_h *g)
1812 if (g->state != READY) {
1813 error (g, _("qemu has not been launched yet"));
1821 guestfs__kill_subprocess (guestfs_h *g)
1823 if (g->state == CONFIG) {
1824 error (g, _("no subprocess to kill"));
1829 fprintf (stderr, "sending SIGTERM to process %d\n", g->pid);
1831 kill (g->pid, SIGTERM);
1832 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1837 /* Access current state. */
1839 guestfs__is_config (guestfs_h *g)
1841 return g->state == CONFIG;
1845 guestfs__is_launching (guestfs_h *g)
1847 return g->state == LAUNCHING;
1851 guestfs__is_ready (guestfs_h *g)
1853 return g->state == READY;
1857 guestfs__is_busy (guestfs_h *g)
1859 return g->state == BUSY;
1863 guestfs__get_state (guestfs_h *g)
1869 guestfs_set_log_message_callback (guestfs_h *g,
1870 guestfs_log_message_cb cb, void *opaque)
1872 g->log_message_cb = cb;
1873 g->log_message_cb_data = opaque;
1877 guestfs_set_subprocess_quit_callback (guestfs_h *g,
1878 guestfs_subprocess_quit_cb cb, void *opaque)
1880 g->subprocess_quit_cb = cb;
1881 g->subprocess_quit_cb_data = opaque;
1885 guestfs_set_launch_done_callback (guestfs_h *g,
1886 guestfs_launch_done_cb cb, void *opaque)
1888 g->launch_done_cb = cb;
1889 g->launch_done_cb_data = opaque;
1892 /*----------------------------------------------------------------------*/
1894 /* This is the code used to send and receive RPC messages and (for
1895 * certain types of message) to perform file transfers. This code is
1896 * driven from the generated actions (src/guestfs-actions.c). There
1897 * are five different cases to consider:
1899 * (1) A non-daemon function. There is no RPC involved at all, it's
1900 * all handled inside the library.
1902 * (2) A simple RPC (eg. "mount"). We write the request, then read
1903 * the reply. The sequence of calls is:
1905 * guestfs___set_busy
1908 * guestfs___end_busy
1910 * (3) An RPC with FileOut parameters (eg. "upload"). We write the
1911 * request, then write the file(s), then read the reply. The sequence
1914 * guestfs___set_busy
1916 * guestfs___send_file (possibly multiple times)
1918 * guestfs___end_busy
1920 * (4) An RPC with FileIn parameters (eg. "download"). We write the
1921 * request, then read the reply, then read the file(s). The sequence
1924 * guestfs___set_busy
1927 * guestfs___recv_file (possibly multiple times)
1928 * guestfs___end_busy
1930 * (5) Both FileOut and FileIn parameters. There are no calls like
1931 * this in the current API, but they would be implemented as a
1932 * combination of cases (3) and (4).
1934 * During all writes and reads, we also select(2) on qemu stdout
1935 * looking for messages (guestfsd stderr and guest kernel dmesg), and
1936 * anything received is passed up through the log_message_cb. This is
1937 * also the reason why all the sockets are non-blocking. We also have
1938 * to check for EOF (qemu died). All of this is handled by the
1939 * functions send_to_daemon and recv_from_daemon.
1943 guestfs___set_busy (guestfs_h *g)
1945 if (g->state != READY) {
1946 error (g, _("guestfs_set_busy: called when in state %d != READY"),
1955 guestfs___end_busy (guestfs_h *g)
1969 error (g, _("guestfs_end_busy: called when in state %d"), g->state);
1975 /* This is called if we detect EOF, ie. qemu died. */
1977 child_cleanup (guestfs_h *g)
1980 fprintf (stderr, "child_cleanup: %p: child process died\n", g);
1982 /*kill (g->pid, SIGTERM);*/
1983 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1984 waitpid (g->pid, NULL, 0);
1985 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1994 memset (&g->launch_t, 0, sizeof g->launch_t);
1996 if (g->subprocess_quit_cb)
1997 g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
2001 read_log_message_or_eof (guestfs_h *g, int fd)
2009 "read_log_message_or_eof: %p g->state = %d, fd = %d\n",
2013 /* QEMU's console emulates a 16550A serial port. The real 16550A
2014 * device has a small FIFO buffer (16 bytes) which means here we see
2015 * lots of small reads of 1-16 bytes in length, usually single
2018 n = read (fd, buf, sizeof buf);
2020 /* Hopefully this indicates the qemu child process has died. */
2026 if (errno == EINTR || errno == EAGAIN)
2029 perrorf (g, "read");
2033 /* In verbose mode, copy all log messages to stderr. */
2035 ignore_value (write (STDERR_FILENO, buf, n));
2037 /* It's an actual log message, send it upwards if anyone is listening. */
2038 if (g->log_message_cb)
2039 g->log_message_cb (g, g->log_message_cb_data, buf, n);
2045 check_for_daemon_cancellation_or_eof (guestfs_h *g, int fd)
2054 "check_for_daemon_cancellation_or_eof: %p g->state = %d, fd = %d\n",
2057 n = read (fd, buf, 4);
2059 /* Hopefully this indicates the qemu child process has died. */
2065 if (errno == EINTR || errno == EAGAIN)
2068 perrorf (g, "read");
2072 xdrmem_create (&xdr, buf, 4, XDR_DECODE);
2073 xdr_uint32_t (&xdr, &flag);
2076 if (flag != GUESTFS_CANCEL_FLAG) {
2077 error (g, _("check_for_daemon_cancellation_or_eof: read 0x%x from daemon, expected 0x%x\n"),
2078 flag, GUESTFS_CANCEL_FLAG);
2085 /* This writes the whole N bytes of BUF to the daemon socket.
2087 * If the whole write is successful, it returns 0.
2088 * If there was an error, it returns -1.
2089 * If the daemon sent a cancellation message, it returns -2.
2091 * It also checks qemu stdout for log messages and passes those up
2092 * through log_message_cb.
2094 * It also checks for EOF (qemu died) and passes that up through the
2095 * child_cleanup function above.
2098 send_to_daemon (guestfs_h *g, const void *v_buf, size_t n)
2100 const char *buf = v_buf;
2106 "send_to_daemon: %p g->state = %d, n = %zu\n", g, g->state, n);
2111 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2112 FD_SET (g->sock, &rset); /* Read socket for cancellation & EOF. */
2113 FD_SET (g->sock, &wset); /* Write to socket to send the data. */
2115 int max_fd = MAX (g->sock, g->fd[1]);
2120 int r = select (max_fd+1, &rset2, &wset2, NULL, NULL);
2122 if (errno == EINTR || errno == EAGAIN)
2124 perrorf (g, "select");
2128 if (FD_ISSET (g->fd[1], &rset2)) {
2129 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2132 if (FD_ISSET (g->sock, &rset2)) {
2133 r = check_for_daemon_cancellation_or_eof (g, g->sock);
2137 if (FD_ISSET (g->sock, &wset2)) {
2138 r = write (g->sock, buf, n);
2140 if (errno == EINTR || errno == EAGAIN)
2142 perrorf (g, "write");
2143 if (errno == EPIPE) /* Disconnected from guest (RHBZ#508713). */
2155 /* This reads a single message, file chunk, launch flag or
2156 * cancellation flag from the daemon. If something was read, it
2157 * returns 0, otherwise -1.
2159 * Both size_rtn and buf_rtn must be passed by the caller as non-NULL.
2161 * *size_rtn returns the size of the returned message or it may be
2162 * GUESTFS_LAUNCH_FLAG or GUESTFS_CANCEL_FLAG.
2164 * *buf_rtn is returned containing the message (if any) or will be set
2165 * to NULL. *buf_rtn must be freed by the caller.
2167 * It also checks qemu stdout for log messages and passes those up
2168 * through log_message_cb.
2170 * It also checks for EOF (qemu died) and passes that up through the
2171 * child_cleanup function above.
2174 recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn)
2180 "recv_from_daemon: %p g->state = %d, size_rtn = %p, buf_rtn = %p\n",
2181 g, g->state, size_rtn, buf_rtn);
2185 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2186 FD_SET (g->sock, &rset); /* Read socket for data & EOF. */
2188 int max_fd = MAX (g->sock, g->fd[1]);
2194 /* nr is the size of the message, but we prime it as -4 because we
2195 * have to read the message length word first.
2199 while (nr < (ssize_t) *size_rtn) {
2201 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2203 if (errno == EINTR || errno == EAGAIN)
2205 perrorf (g, "select");
2211 if (FD_ISSET (g->fd[1], &rset2)) {
2212 if (read_log_message_or_eof (g, g->fd[1]) == -1) {
2218 if (FD_ISSET (g->sock, &rset2)) {
2219 if (nr < 0) { /* Have we read the message length word yet? */
2220 r = read (g->sock, lenbuf+nr+4, -nr);
2222 if (errno == EINTR || errno == EAGAIN)
2225 perrorf (g, "read");
2226 /* Under some circumstances we see "Connection reset by peer"
2227 * here when the child dies suddenly. Catch this and call
2228 * the cleanup function, same as for EOF.
2230 if (err == ECONNRESET)
2235 error (g, _("unexpected end of file when reading from daemon"));
2241 if (nr < 0) /* Still not got the whole length word. */
2245 xdrmem_create (&xdr, lenbuf, 4, XDR_DECODE);
2246 xdr_uint32_t (&xdr, size_rtn);
2249 if (*size_rtn == GUESTFS_LAUNCH_FLAG) {
2250 if (g->state != LAUNCHING)
2251 error (g, _("received magic signature from guestfsd, but in state %d"),
2255 if (g->launch_done_cb)
2256 g->launch_done_cb (g, g->launch_done_cb_data);
2260 else if (*size_rtn == GUESTFS_CANCEL_FLAG)
2262 /* If this happens, it's pretty bad and we've probably lost
2265 else if (*size_rtn > GUESTFS_MESSAGE_MAX) {
2266 error (g, _("message length (%u) > maximum possible size (%d)"),
2267 (unsigned) *size_rtn, GUESTFS_MESSAGE_MAX);
2271 /* Allocate the complete buffer, size now known. */
2272 *buf_rtn = safe_malloc (g, *size_rtn);
2276 size_t sizetoread = *size_rtn - nr;
2277 if (sizetoread > BUFSIZ) sizetoread = BUFSIZ;
2279 r = read (g->sock, (char *) (*buf_rtn) + nr, sizetoread);
2281 if (errno == EINTR || errno == EAGAIN)
2283 perrorf (g, "read");
2289 error (g, _("unexpected end of file when reading from daemon"));
2299 /* Got the full message, caller can start processing it. */
2300 #ifdef ENABLE_PACKET_DUMP
2304 for (i = 0; i < nr; i += 16) {
2305 printf ("%04zx: ", i);
2306 for (j = i; j < MIN (i+16, nr); ++j)
2307 printf ("%02x ", (*(unsigned char **)buf_rtn)[j]);
2308 for (; j < i+16; ++j)
2311 for (j = i; j < MIN (i+16, nr); ++j)
2312 if (c_isprint ((*(char **)buf_rtn)[j]))
2313 printf ("%c", (*(char **)buf_rtn)[j]);
2316 for (; j < i+16; ++j)
2326 /* This is very much like recv_from_daemon above, but g->sock is
2327 * a listening socket and we are accepting a new connection on
2328 * that socket instead of reading anything. Returns the newly
2332 accept_from_daemon (guestfs_h *g)
2338 "accept_from_daemon: %p g->state = %d\n", g, g->state);
2342 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2343 FD_SET (g->sock, &rset); /* Read socket for accept. */
2345 int max_fd = MAX (g->sock, g->fd[1]);
2348 while (sock == -1) {
2350 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2352 if (errno == EINTR || errno == EAGAIN)
2354 perrorf (g, "select");
2358 if (FD_ISSET (g->fd[1], &rset2)) {
2359 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2362 if (FD_ISSET (g->sock, &rset2)) {
2363 sock = accept (g->sock, NULL, NULL);
2365 if (errno == EINTR || errno == EAGAIN)
2367 perrorf (g, "accept");
2377 guestfs___send (guestfs_h *g, int proc_nr, xdrproc_t xdrp, char *args)
2379 struct guestfs_message_header hdr;
2382 int serial = g->msg_next_serial++;
2385 size_t msg_out_size;
2387 if (g->state != BUSY) {
2388 error (g, _("guestfs___send: state %d != BUSY"), g->state);
2392 /* We have to allocate this message buffer on the heap because
2393 * it is quite large (although will be mostly unused). We
2394 * can't allocate it on the stack because in some environments
2395 * we have quite limited stack space available, notably when
2396 * running in the JVM.
2398 msg_out = safe_malloc (g, GUESTFS_MESSAGE_MAX + 4);
2399 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MESSAGE_MAX, XDR_ENCODE);
2401 /* Serialize the header. */
2402 hdr.prog = GUESTFS_PROGRAM;
2403 hdr.vers = GUESTFS_PROTOCOL_VERSION;
2405 hdr.direction = GUESTFS_DIRECTION_CALL;
2406 hdr.serial = serial;
2407 hdr.status = GUESTFS_STATUS_OK;
2409 if (!xdr_guestfs_message_header (&xdr, &hdr)) {
2410 error (g, _("xdr_guestfs_message_header failed"));
2414 /* Serialize the args. If any, because some message types
2415 * have no parameters.
2418 if (!(*xdrp) (&xdr, args)) {
2419 error (g, _("dispatch failed to marshal args"));
2424 /* Get the actual length of the message, resize the buffer to match
2425 * the actual length, and write the length word at the beginning.
2427 len = xdr_getpos (&xdr);
2430 msg_out = safe_realloc (g, msg_out, len + 4);
2431 msg_out_size = len + 4;
2433 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2434 xdr_uint32_t (&xdr, &len);
2437 r = send_to_daemon (g, msg_out, msg_out_size);
2438 if (r == -2) /* Ignore stray daemon cancellations. */
2451 static int cancel = 0; /* XXX Implement file cancellation. */
2452 static int send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t len);
2453 static int send_file_data (guestfs_h *g, const char *buf, size_t len);
2454 static int send_file_cancellation (guestfs_h *g);
2455 static int send_file_complete (guestfs_h *g);
2461 * -2 daemon cancelled (we must read the error message)
2464 guestfs___send_file (guestfs_h *g, const char *filename)
2466 char buf[GUESTFS_MAX_CHUNK_SIZE];
2469 fd = open (filename, O_RDONLY);
2471 perrorf (g, "open: %s", filename);
2472 send_file_cancellation (g);
2473 /* Daemon sees cancellation and won't reply, so caller can
2479 /* Send file in chunked encoding. */
2481 r = read (fd, buf, sizeof buf);
2482 if (r == -1 && (errno == EINTR || errno == EAGAIN))
2485 err = send_file_data (g, buf, r);
2487 if (err == -2) /* daemon sent cancellation */
2488 send_file_cancellation (g);
2493 if (cancel) { /* cancel from either end */
2494 send_file_cancellation (g);
2499 perrorf (g, "read: %s", filename);
2500 send_file_cancellation (g);
2504 /* End of file, but before we send that, we need to close
2505 * the file and check for errors.
2507 if (close (fd) == -1) {
2508 perrorf (g, "close: %s", filename);
2509 send_file_cancellation (g);
2513 return send_file_complete (g);
2516 /* Send a chunk of file data. */
2518 send_file_data (guestfs_h *g, const char *buf, size_t len)
2520 return send_file_chunk (g, 0, buf, len);
2523 /* Send a cancellation message. */
2525 send_file_cancellation (guestfs_h *g)
2527 return send_file_chunk (g, 1, NULL, 0);
2530 /* Send a file complete chunk. */
2532 send_file_complete (guestfs_h *g)
2535 return send_file_chunk (g, 0, buf, 0);
2539 send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t buflen)
2543 guestfs_chunk chunk;
2546 size_t msg_out_size;
2548 if (g->state != BUSY) {
2549 error (g, _("send_file_chunk: state %d != READY"), g->state);
2553 /* Allocate the chunk buffer. Don't use the stack to avoid
2554 * excessive stack usage and unnecessary copies.
2556 msg_out = safe_malloc (g, GUESTFS_MAX_CHUNK_SIZE + 4 + 48);
2557 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MAX_CHUNK_SIZE + 48, XDR_ENCODE);
2559 /* Serialize the chunk. */
2560 chunk.cancel = cancel;
2561 chunk.data.data_len = buflen;
2562 chunk.data.data_val = (char *) buf;
2564 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2565 error (g, _("xdr_guestfs_chunk failed (buf = %p, buflen = %zu)"),
2571 len = xdr_getpos (&xdr);
2574 /* Reduce the size of the outgoing message buffer to the real length. */
2575 msg_out = safe_realloc (g, msg_out, len + 4);
2576 msg_out_size = len + 4;
2578 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2579 xdr_uint32_t (&xdr, &len);
2581 r = send_to_daemon (g, msg_out, msg_out_size);
2583 /* Did the daemon send a cancellation message? */
2586 fprintf (stderr, "got daemon cancellation\n");
2602 /* Receive a reply. */
2604 guestfs___recv (guestfs_h *g, const char *fn,
2605 guestfs_message_header *hdr,
2606 guestfs_message_error *err,
2607 xdrproc_t xdrp, char *ret)
2615 r = recv_from_daemon (g, &size, &buf);
2619 /* This can happen if a cancellation happens right at the end
2620 * of us sending a FileIn parameter to the daemon. Discard. The
2621 * daemon should send us an error message next.
2623 if (size == GUESTFS_CANCEL_FLAG)
2626 if (size == GUESTFS_LAUNCH_FLAG) {
2627 error (g, "%s: received unexpected launch flag from daemon when expecting reply", fn);
2631 xdrmem_create (&xdr, buf, size, XDR_DECODE);
2633 if (!xdr_guestfs_message_header (&xdr, hdr)) {
2634 error (g, "%s: failed to parse reply header", fn);
2639 if (hdr->status == GUESTFS_STATUS_ERROR) {
2640 if (!xdr_guestfs_message_error (&xdr, err)) {
2641 error (g, "%s: failed to parse reply error", fn);
2647 if (xdrp && ret && !xdrp (&xdr, ret)) {
2648 error (g, "%s: failed to parse reply", fn);
2660 /* Receive a file. */
2662 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2663 static ssize_t receive_file_data (guestfs_h *g, void **buf);
2666 guestfs___recv_file (guestfs_h *g, const char *filename)
2671 fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2673 perrorf (g, "open: %s", filename);
2677 /* Receive the file in chunked encoding. */
2678 while ((r = receive_file_data (g, &buf)) > 0) {
2679 if (xwrite (fd, buf, r) == -1) {
2680 perrorf (g, "%s: write", filename);
2688 error (g, _("%s: error in chunked encoding"), filename);
2692 if (close (fd) == -1) {
2693 perrorf (g, "close: %s", filename);
2700 /* Send cancellation message to daemon, then wait until it
2701 * cancels (just throwing away data).
2705 uint32_t flag = GUESTFS_CANCEL_FLAG;
2708 fprintf (stderr, "%s: waiting for daemon to acknowledge cancellation\n",
2711 xdrmem_create (&xdr, fbuf, sizeof fbuf, XDR_ENCODE);
2712 xdr_uint32_t (&xdr, &flag);
2715 if (xwrite (g->sock, fbuf, sizeof fbuf) == -1) {
2716 perrorf (g, _("write to daemon socket"));
2720 while (receive_file_data (g, NULL) > 0)
2721 ; /* just discard it */
2726 /* Receive a chunk of file data. */
2727 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2729 receive_file_data (guestfs_h *g, void **buf_r)
2735 guestfs_chunk chunk;
2737 r = recv_from_daemon (g, &len, &buf);
2739 error (g, _("receive_file_data: parse error in reply callback"));
2743 if (len == GUESTFS_LAUNCH_FLAG || len == GUESTFS_CANCEL_FLAG) {
2744 error (g, _("receive_file_data: unexpected flag received when reading file chunks"));
2748 memset (&chunk, 0, sizeof chunk);
2750 xdrmem_create (&xdr, buf, len, XDR_DECODE);
2751 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2752 error (g, _("failed to parse file chunk"));
2757 /* After decoding, the original buffer is no longer used. */
2761 error (g, _("file receive cancelled by daemon"));
2762 free (chunk.data.data_val);
2766 if (chunk.data.data_len == 0) { /* end of transfer */
2767 free (chunk.data.data_val);
2771 if (buf_r) *buf_r = chunk.data.data_val;
2772 else free (chunk.data.data_val); /* else caller frees */
2774 return chunk.data.data_len;