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>
37 #include <rpc/types.h>
44 #ifdef HAVE_SYS_TYPES_H
45 #include <sys/types.h>
48 #ifdef HAVE_SYS_WAIT_H
52 #ifdef HAVE_SYS_SOCKET_H
53 #include <sys/socket.h>
60 #include <arpa/inet.h>
61 #include <netinet/in.h>
64 #include "glthread/lock.h"
65 #include "ignore-value.h"
68 #include "guestfs-internal.h"
69 #include "guestfs-internal-actions.h"
70 #include "guestfs_protocol.h"
74 #define _(str) dgettext(PACKAGE, (str))
75 //#define N_(str) dgettext(PACKAGE, (str))
81 #define error guestfs_error
82 #define perrorf guestfs_perrorf
83 #define safe_malloc guestfs_safe_malloc
84 #define safe_realloc guestfs_safe_realloc
85 #define safe_strdup guestfs_safe_strdup
86 //#define safe_memdup guestfs_safe_memdup
88 static void default_error_cb (guestfs_h *g, void *data, const char *msg);
89 static int send_to_daemon (guestfs_h *g, const void *v_buf, size_t n);
90 static int recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn);
91 static int accept_from_daemon (guestfs_h *g);
92 static int check_peer_euid (guestfs_h *g, int sock, uid_t *rtn);
93 static void close_handles (void);
94 static int qemu_supports (guestfs_h *g, const char *option);
96 #define UNIX_PATH_MAX 108
98 /* Also in guestfsd.c */
99 #define GUESTFWD_ADDR "10.0.2.4"
100 #define GUESTFWD_PORT "6666"
102 /* GuestFS handle and connection. */
103 enum state { CONFIG, LAUNCHING, READY, BUSY, NO_HANDLE };
107 struct guestfs_h *next; /* Linked list of open handles. */
109 /* State: see the state machine diagram in the man page guestfs(3). */
112 int fd[2]; /* Stdin/stdout of qemu. */
113 int sock; /* Daemon communications socket. */
114 pid_t pid; /* Qemu PID. */
115 pid_t recoverypid; /* Recovery process PID. */
117 struct timeval launch_t; /* The time that we called guestfs_launch. */
119 char *tmpdir; /* Temporary directory containing socket. */
121 char *qemu_help, *qemu_version; /* Output of qemu -help, qemu -version. */
123 char **cmdline; /* Qemu command line. */
132 char *path; /* Path to kernel, initrd. */
133 char *qemu; /* Qemu binary. */
134 char *append; /* Append to kernel command line. */
136 int memsize; /* Size of RAM (megabytes). */
138 int selinux; /* selinux enabled? */
143 guestfs_abort_cb abort_cb;
144 guestfs_error_handler_cb error_cb;
145 void * error_cb_data;
146 guestfs_log_message_cb log_message_cb;
147 void * log_message_cb_data;
148 guestfs_subprocess_quit_cb subprocess_quit_cb;
149 void * subprocess_quit_cb_data;
150 guestfs_launch_done_cb launch_done_cb;
151 void * launch_done_cb_data;
156 gl_lock_define_initialized (static, handles_lock);
157 static guestfs_h *handles = NULL;
158 static int atexit_handler_set = 0;
161 guestfs_create (void)
166 g = malloc (sizeof (*g));
169 memset (g, 0, sizeof (*g));
178 g->error_cb = default_error_cb;
179 g->error_cb_data = NULL;
181 g->recovery_proc = 1;
183 str = getenv ("LIBGUESTFS_DEBUG");
184 g->verbose = str != NULL && STREQ (str, "1");
186 str = getenv ("LIBGUESTFS_TRACE");
187 g->trace = str != NULL && STREQ (str, "1");
189 str = getenv ("LIBGUESTFS_PATH");
190 g->path = str != NULL ? strdup (str) : strdup (GUESTFS_DEFAULT_PATH);
191 if (!g->path) goto error;
193 str = getenv ("LIBGUESTFS_QEMU");
194 g->qemu = str != NULL ? strdup (str) : strdup (QEMU);
195 if (!g->qemu) goto error;
197 str = getenv ("LIBGUESTFS_APPEND");
199 g->append = strdup (str);
200 if (!g->append) goto error;
203 /* Choose a suitable memory size. Previously we tried to choose
204 * a minimal memory size, but this isn't really necessary since
205 * recent QEMU and KVM don't do anything nasty like locking
206 * memory into core any more. Thus we can safely choose a
207 * large, generous amount of memory, and it'll just get swapped
208 * on smaller systems.
210 str = getenv ("LIBGUESTFS_MEMSIZE");
212 if (sscanf (str, "%d", &g->memsize) != 1 || g->memsize <= 256) {
213 fprintf (stderr, "libguestfs: non-numeric or too small value for LIBGUESTFS_MEMSIZE\n");
219 /* Start with large serial numbers so they are easy to spot
220 * inside the protocol.
222 g->msg_next_serial = 0x00123400;
224 /* Link the handles onto a global list. */
225 gl_lock_lock (handles_lock);
228 if (!atexit_handler_set) {
229 atexit (close_handles);
230 atexit_handler_set = 1;
232 gl_lock_unlock (handles_lock);
235 fprintf (stderr, "new guestfs handle %p\n", g);
248 guestfs_close (guestfs_h *g)
254 if (g->state == NO_HANDLE) {
255 /* Not safe to call 'error' here, so ... */
256 fprintf (stderr, _("guestfs_close: called twice on the same handle\n"));
261 fprintf (stderr, "closing guestfs handle %p (state %d)\n", g, g->state);
263 /* Try to sync if autosync flag is set. */
264 if (g->autosync && g->state == READY) {
265 guestfs_umount_all (g);
269 /* Remove any handlers that might be called back before we kill the
272 g->log_message_cb = NULL;
274 if (g->state != CONFIG)
275 guestfs_kill_subprocess (g);
288 /* Wait for subprocess(es) to exit. */
289 waitpid (g->pid, NULL, 0);
290 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
292 /* Remove tmpfiles. */
294 snprintf (filename, sizeof filename, "%s/sock", g->tmpdir);
297 snprintf (filename, sizeof filename, "%s/initrd", g->tmpdir);
300 snprintf (filename, sizeof filename, "%s/kernel", g->tmpdir);
309 for (i = 0; i < g->cmdline_size; ++i)
310 free (g->cmdline[i]);
314 /* Mark the handle as dead before freeing it. */
315 g->state = NO_HANDLE;
317 gl_lock_lock (handles_lock);
321 for (gg = handles; gg->next != g; gg = gg->next)
325 gl_lock_unlock (handles_lock);
327 free (g->last_error);
332 free (g->qemu_version);
336 /* Close all open handles (called from atexit(3)). */
340 while (handles) guestfs_close (handles);
344 guestfs_last_error (guestfs_h *g)
346 return g->last_error;
350 set_last_error (guestfs_h *g, const char *msg)
352 free (g->last_error);
353 g->last_error = strdup (msg);
357 default_error_cb (guestfs_h *g, void *data, const char *msg)
359 fprintf (stderr, _("libguestfs: error: %s\n"), msg);
363 guestfs_error (guestfs_h *g, const char *fs, ...)
369 int err = vasprintf (&msg, fs, args);
374 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
375 set_last_error (g, msg);
381 guestfs_perrorf (guestfs_h *g, const char *fs, ...)
388 int err = vasprintf (&msg, fs, args);
395 strerror_r (errnum, buf, sizeof buf);
399 buf = strerror_r (errnum, _buf, sizeof _buf);
402 msg = safe_realloc (g, msg, strlen (msg) + 2 + strlen (buf) + 1);
406 if (g->error_cb) g->error_cb (g, g->error_cb_data, msg);
407 set_last_error (g, msg);
413 guestfs_safe_malloc (guestfs_h *g, size_t nbytes)
415 void *ptr = malloc (nbytes);
416 if (nbytes > 0 && !ptr) g->abort_cb ();
420 /* Return 1 if an array of N objects, each of size S, cannot exist due
421 to size arithmetic overflow. S must be positive and N must be
422 nonnegative. This is a macro, not an inline function, so that it
423 works correctly even when SIZE_MAX < N.
425 By gnulib convention, SIZE_MAX represents overflow in size
426 calculations, so the conservative dividend to use here is
427 SIZE_MAX - 1, since SIZE_MAX might represent an overflowed value.
428 However, malloc (SIZE_MAX) fails on all known hosts where
429 sizeof (ptrdiff_t) <= sizeof (size_t), so do not bother to test for
430 exactly-SIZE_MAX allocations on such hosts; this avoids a test and
431 branch when S is known to be 1. */
432 # define xalloc_oversized(n, s) \
433 ((size_t) (sizeof (ptrdiff_t) <= sizeof (size_t) ? -1 : -2) / (s) < (n))
435 /* Technically we should add an autoconf test for this, testing for the desired
436 functionality, like what's done in gnulib, but for now, this is fine. */
437 #define HAVE_GNU_CALLOC (__GLIBC__ >= 2)
439 /* Allocate zeroed memory for N elements of S bytes, with error
440 checking. S must be nonzero. */
442 guestfs_safe_calloc (guestfs_h *g, size_t n, size_t s)
444 /* From gnulib's calloc function in xmalloc.c. */
446 /* Test for overflow, since some calloc implementations don't have
447 proper overflow checks. But omit overflow and size-zero tests if
448 HAVE_GNU_CALLOC, since GNU calloc catches overflow and never
449 returns NULL if successful. */
450 if ((! HAVE_GNU_CALLOC && xalloc_oversized (n, s))
451 || (! (p = calloc (n, s)) && (HAVE_GNU_CALLOC || n != 0)))
457 guestfs_safe_realloc (guestfs_h *g, void *ptr, int nbytes)
459 void *p = realloc (ptr, nbytes);
460 if (nbytes > 0 && !p) g->abort_cb ();
465 guestfs_safe_strdup (guestfs_h *g, const char *str)
467 char *s = strdup (str);
468 if (!s) g->abort_cb ();
473 guestfs_safe_memdup (guestfs_h *g, void *ptr, size_t size)
475 void *p = malloc (size);
476 if (!p) g->abort_cb ();
477 memcpy (p, ptr, size);
482 xwrite (int fd, const void *v_buf, size_t len)
484 const char *buf = v_buf;
488 r = write (fd, buf, len);
500 guestfs_set_out_of_memory_handler (guestfs_h *g, guestfs_abort_cb cb)
506 guestfs_get_out_of_memory_handler (guestfs_h *g)
512 guestfs_set_error_handler (guestfs_h *g, guestfs_error_handler_cb cb, void *data)
515 g->error_cb_data = data;
518 guestfs_error_handler_cb
519 guestfs_get_error_handler (guestfs_h *g, void **data_rtn)
521 if (data_rtn) *data_rtn = g->error_cb_data;
526 guestfs__set_verbose (guestfs_h *g, int v)
533 guestfs__get_verbose (guestfs_h *g)
539 guestfs__set_autosync (guestfs_h *g, int a)
546 guestfs__get_autosync (guestfs_h *g)
552 guestfs__set_path (guestfs_h *g, const char *path)
559 safe_strdup (g, GUESTFS_DEFAULT_PATH) : safe_strdup (g, path);
564 guestfs__get_path (guestfs_h *g)
570 guestfs__set_qemu (guestfs_h *g, const char *qemu)
575 g->qemu = qemu == NULL ? safe_strdup (g, QEMU) : safe_strdup (g, qemu);
580 guestfs__get_qemu (guestfs_h *g)
586 guestfs__set_append (guestfs_h *g, const char *append)
591 g->append = append ? safe_strdup (g, append) : NULL;
596 guestfs__get_append (guestfs_h *g)
602 guestfs__set_memsize (guestfs_h *g, int memsize)
604 g->memsize = memsize;
609 guestfs__get_memsize (guestfs_h *g)
615 guestfs__set_selinux (guestfs_h *g, int selinux)
617 g->selinux = selinux;
622 guestfs__get_selinux (guestfs_h *g)
628 guestfs__get_pid (guestfs_h *g)
633 error (g, "get_pid: no qemu subprocess");
638 struct guestfs_version *
639 guestfs__version (guestfs_h *g)
641 struct guestfs_version *r;
643 r = safe_malloc (g, sizeof *r);
644 r->major = PACKAGE_VERSION_MAJOR;
645 r->minor = PACKAGE_VERSION_MINOR;
646 r->release = PACKAGE_VERSION_RELEASE;
647 r->extra = safe_strdup (g, PACKAGE_VERSION_EXTRA);
652 guestfs__set_trace (guestfs_h *g, int t)
659 guestfs__get_trace (guestfs_h *g)
665 guestfs__set_direct (guestfs_h *g, int d)
672 guestfs__get_direct (guestfs_h *g)
678 guestfs__set_recovery_proc (guestfs_h *g, int f)
680 g->recovery_proc = !!f;
685 guestfs__get_recovery_proc (guestfs_h *g)
687 return g->recovery_proc;
690 /* Add a string to the current command line. */
692 incr_cmdline_size (guestfs_h *g)
694 if (g->cmdline == NULL) {
695 /* g->cmdline[0] is reserved for argv[0], set in guestfs_launch. */
697 g->cmdline = safe_malloc (g, sizeof (char *));
698 g->cmdline[0] = NULL;
702 g->cmdline = safe_realloc (g, g->cmdline, sizeof (char *) * g->cmdline_size);
706 add_cmdline (guestfs_h *g, const char *str)
708 if (g->state != CONFIG) {
710 _("command line cannot be altered after qemu subprocess launched"));
714 incr_cmdline_size (g);
715 g->cmdline[g->cmdline_size-1] = safe_strdup (g, str);
720 guestfs__config (guestfs_h *g,
721 const char *qemu_param, const char *qemu_value)
723 if (qemu_param[0] != '-') {
724 error (g, _("guestfs_config: parameter must begin with '-' character"));
728 /* A bit fascist, but the user will probably break the extra
729 * parameters that we add if they try to set any of these.
731 if (STREQ (qemu_param, "-kernel") ||
732 STREQ (qemu_param, "-initrd") ||
733 STREQ (qemu_param, "-nographic") ||
734 STREQ (qemu_param, "-serial") ||
735 STREQ (qemu_param, "-full-screen") ||
736 STREQ (qemu_param, "-std-vga") ||
737 STREQ (qemu_param, "-vnc")) {
738 error (g, _("guestfs_config: parameter '%s' isn't allowed"), qemu_param);
742 if (add_cmdline (g, qemu_param) != 0) return -1;
744 if (qemu_value != NULL) {
745 if (add_cmdline (g, qemu_value) != 0) return -1;
752 guestfs__add_drive_with_if (guestfs_h *g, const char *filename,
753 const char *drive_if)
755 size_t len = strlen (filename) + 64;
758 if (strchr (filename, ',') != NULL) {
759 error (g, _("filename cannot contain ',' (comma) character"));
763 /* cache=off improves reliability in the event of a host crash.
765 * However this option causes qemu to try to open the file with
766 * O_DIRECT. This fails on some filesystem types (notably tmpfs).
767 * So we check if we can open the file with or without O_DIRECT,
768 * and use cache=off (or not) accordingly.
770 * This test also checks for the presence of the file, which
771 * is a documented semantic of this interface.
773 int fd = open (filename, O_RDONLY|O_DIRECT);
776 snprintf (buf, len, "file=%s,cache=off,if=%s", filename, drive_if);
778 fd = open (filename, O_RDONLY);
781 snprintf (buf, len, "file=%s,if=%s", filename, drive_if);
783 perrorf (g, "%s", filename);
788 return guestfs__config (g, "-drive", buf);
792 guestfs__add_drive_ro_with_if (guestfs_h *g, const char *filename,
793 const char *drive_if)
795 size_t len = strlen (filename) + 64;
798 if (strchr (filename, ',') != NULL) {
799 error (g, _("filename cannot contain ',' (comma) character"));
803 if (access (filename, F_OK) == -1) {
804 perrorf (g, "%s", filename);
808 snprintf (buf, len, "file=%s,snapshot=on,if=%s", filename, drive_if);
810 return guestfs__config (g, "-drive", buf);
814 guestfs__add_drive (guestfs_h *g, const char *filename)
816 return guestfs__add_drive_with_if (g, filename, DRIVE_IF);
820 guestfs__add_drive_ro (guestfs_h *g, const char *filename)
822 return guestfs__add_drive_ro_with_if (g, filename, DRIVE_IF);
826 guestfs__add_cdrom (guestfs_h *g, const char *filename)
828 if (strchr (filename, ',') != NULL) {
829 error (g, _("filename cannot contain ',' (comma) character"));
833 if (access (filename, F_OK) == -1) {
834 perrorf (g, "%s", filename);
838 return guestfs__config (g, "-cdrom", filename);
841 /* Returns true iff file is contained in dir. */
843 dir_contains_file (const char *dir, const char *file)
845 int dirlen = strlen (dir);
846 int filelen = strlen (file);
847 int len = dirlen+filelen+2;
850 snprintf (path, len, "%s/%s", dir, file);
851 return access (path, F_OK) == 0;
854 /* Returns true iff every listed file is contained in 'dir'. */
856 dir_contains_files (const char *dir, ...)
861 va_start (args, dir);
862 while ((file = va_arg (args, const char *)) != NULL) {
863 if (!dir_contains_file (dir, file)) {
872 static void print_timestamped_message (guestfs_h *g, const char *fs, ...);
873 static int build_supermin_appliance (guestfs_h *g, const char *path, char **kernel, char **initrd);
874 static int is_openable (guestfs_h *g, const char *path, int flags);
875 static void print_cmdline (guestfs_h *g);
877 static const char *kernel_name = "vmlinuz." REPO "." host_cpu;
878 static const char *initrd_name = "initramfs." REPO "." host_cpu ".img";
879 static const char *supermin_name =
880 "initramfs." REPO "." host_cpu ".supermin.img";
881 static const char *supermin_hostfiles_name =
882 "initramfs." REPO "." host_cpu ".supermin.hostfiles";
885 guestfs__launch (guestfs_h *g)
888 char dir_template[PATH_MAX];
893 char *path, *pelem, *pend;
894 char *kernel = NULL, *initrd = NULL;
895 int null_vmchannel_sock;
897 struct sockaddr_un addr;
899 /* Start the clock ... */
900 gettimeofday (&g->launch_t, NULL);
908 tmpdir = getenv ("TMPDIR") ? : tmpdir;
909 snprintf (dir_template, sizeof dir_template, "%s/libguestfsXXXXXX", tmpdir);
913 error (g, _("you must call guestfs_add_drive before guestfs_launch"));
917 if (g->state != CONFIG) {
918 error (g, _("qemu has already been launched"));
922 /* Make the temporary directory. */
924 g->tmpdir = safe_strdup (g, dir_template);
925 if (mkdtemp (g->tmpdir) == NULL) {
926 perrorf (g, _("%s: cannot create temporary directory"), dir_template);
931 /* First search g->path for the supermin appliance, and try to
932 * synthesize a kernel and initrd from that. If it fails, we
933 * try the path search again looking for a backup ordinary
936 pelem = path = safe_strdup (g, g->path);
938 pend = strchrnul (pelem, ':');
939 pmore = *pend == ':';
943 /* Empty element of "." means cwd. */
944 if (len == 0 || (len == 1 && *pelem == '.')) {
947 "looking for supermin appliance in current directory\n");
948 if (dir_contains_files (".",
949 supermin_name, supermin_hostfiles_name,
950 "kmod.whitelist", NULL)) {
951 if (build_supermin_appliance (g, ".", &kernel, &initrd) == -1)
956 /* Look at <path>/supermin* etc. */
959 fprintf (stderr, "looking for supermin appliance in %s\n", pelem);
961 if (dir_contains_files (pelem,
962 supermin_name, supermin_hostfiles_name,
963 "kmod.whitelist", NULL)) {
964 if (build_supermin_appliance (g, pelem, &kernel, &initrd) == -1)
975 if (kernel == NULL || initrd == NULL) {
976 /* Search g->path for the kernel and initrd. */
977 pelem = path = safe_strdup (g, g->path);
979 pend = strchrnul (pelem, ':');
980 pmore = *pend == ':';
984 /* Empty element or "." means cwd. */
985 if (len == 0 || (len == 1 && *pelem == '.')) {
988 "looking for appliance in current directory\n");
989 if (dir_contains_files (".", kernel_name, initrd_name, NULL)) {
990 kernel = safe_strdup (g, kernel_name);
991 initrd = safe_strdup (g, initrd_name);
995 /* Look at <path>/kernel etc. */
998 fprintf (stderr, "looking for appliance in %s\n", pelem);
1000 if (dir_contains_files (pelem, kernel_name, initrd_name, NULL)) {
1001 kernel = safe_malloc (g, len + strlen (kernel_name) + 2);
1002 initrd = safe_malloc (g, len + strlen (initrd_name) + 2);
1003 sprintf (kernel, "%s/%s", pelem, kernel_name);
1004 sprintf (initrd, "%s/%s", pelem, initrd_name);
1015 if (kernel == NULL || initrd == NULL) {
1016 error (g, _("cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)"),
1017 kernel_name, initrd_name, g->path);
1022 print_timestamped_message (g, "begin testing qemu features");
1024 /* Get qemu help text and version. */
1025 if (qemu_supports (g, NULL) == -1)
1028 /* Choose which vmchannel implementation to use. */
1029 if (qemu_supports (g, "-net user")) {
1030 /* The "null vmchannel" implementation. Requires SLIRP (user mode
1031 * networking in qemu) but no other vmchannel support. The daemon
1032 * will connect back to a random port number on localhost.
1034 struct sockaddr_in addr;
1035 socklen_t addrlen = sizeof addr;
1037 g->sock = socket (AF_INET, SOCK_STREAM, IPPROTO_TCP);
1038 if (g->sock == -1) {
1039 perrorf (g, "socket");
1042 addr.sin_family = AF_INET;
1043 addr.sin_port = htons (0);
1044 addr.sin_addr.s_addr = htonl (INADDR_LOOPBACK);
1045 if (bind (g->sock, (struct sockaddr *) &addr, addrlen) == -1) {
1046 perrorf (g, "bind");
1050 if (listen (g->sock, 256) == -1) {
1051 perrorf (g, "listen");
1055 if (getsockname (g->sock, (struct sockaddr *) &addr, &addrlen) == -1) {
1056 perrorf (g, "getsockname");
1060 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1061 perrorf (g, "fcntl");
1065 null_vmchannel_sock = ntohs (addr.sin_port);
1067 fprintf (stderr, "null_vmchannel_sock = %d\n", null_vmchannel_sock);
1069 /* Using some vmchannel impl. We need to create a local Unix
1070 * domain socket for qemu to use.
1072 snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
1074 null_vmchannel_sock = 0;
1078 if (pipe (wfd) == -1 || pipe (rfd) == -1) {
1079 perrorf (g, "pipe");
1085 print_timestamped_message (g, "finished testing qemu features");
1089 perrorf (g, "fork");
1099 if (r == 0) { /* Child (qemu). */
1101 const char *vmchannel = NULL;
1103 /* Set up the full command line. Do this in the subprocess so we
1104 * don't need to worry about cleaning up.
1106 g->cmdline[0] = g->qemu;
1108 /* qemu sometimes needs this option to enable hardware
1109 * virtualization, but some versions of 'qemu-kvm' will use KVM
1110 * regardless (even where this option appears in the help text).
1111 * It is rumoured that there are versions of qemu where supplying
1112 * this option when hardware virtualization is not available will
1113 * cause qemu to fail, so we we have to check at least that
1114 * /dev/kvm is openable. That's not reliable, since /dev/kvm
1115 * might be openable by qemu but not by us (think: SELinux) in
1116 * which case the user would not get hardware virtualization,
1117 * although at least shouldn't fail. A giant clusterfuck with the
1118 * qemu command line, again.
1120 if (qemu_supports (g, "-enable-kvm") &&
1121 is_openable (g, "/dev/kvm", O_RDWR))
1122 add_cmdline (g, "-enable-kvm");
1124 /* Newer versions of qemu (from around 2009/12) changed the
1125 * behaviour of monitors so that an implicit '-monitor stdio' is
1126 * assumed if we are in -nographic mode and there is no other
1127 * -monitor option. Only a single stdio device is allowed, so
1128 * this broke the '-serial stdio' option. There is a new flag
1129 * called -nodefaults which gets rid of all this default crud, so
1130 * let's use that to avoid this and any future surprises.
1132 if (qemu_supports (g, "-nodefaults"))
1133 add_cmdline (g, "-nodefaults");
1135 add_cmdline (g, "-nographic");
1136 add_cmdline (g, "-serial");
1137 add_cmdline (g, "stdio");
1139 snprintf (buf, sizeof buf, "%d", g->memsize);
1140 add_cmdline (g, "-m");
1141 add_cmdline (g, buf);
1143 /* Force exit instead of reboot on panic */
1144 add_cmdline (g, "-no-reboot");
1146 /* These options recommended by KVM developers to improve reliability. */
1147 if (qemu_supports (g, "-no-hpet"))
1148 add_cmdline (g, "-no-hpet");
1150 if (qemu_supports (g, "-rtc-td-hack"))
1151 add_cmdline (g, "-rtc-td-hack");
1153 /* If qemu has SLIRP (user mode network) enabled then we can get
1154 * away with "no vmchannel", where we just connect back to a random
1157 if (null_vmchannel_sock) {
1158 add_cmdline (g, "-net");
1159 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1161 snprintf (buf, sizeof buf,
1162 "guestfs_vmchannel=tcp:10.0.2.2:%d", null_vmchannel_sock);
1163 vmchannel = strdup (buf);
1166 /* New-style -net user,guestfwd=... syntax for guestfwd. See:
1168 * http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=c92ef6a22d3c71538fcc48fb61ad353f7ba03b62
1170 * The original suggested format doesn't work, see:
1172 * http://lists.gnu.org/archive/html/qemu-devel/2009-07/msg01654.html
1174 * However Gerd Hoffman privately suggested to me using -chardev
1175 * instead, which does work.
1177 else if (qemu_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) {
1178 snprintf (buf, sizeof buf,
1179 "socket,id=guestfsvmc,path=%s,server,nowait", unixsock);
1181 add_cmdline (g, "-chardev");
1182 add_cmdline (g, buf);
1184 snprintf (buf, sizeof buf,
1185 "user,vlan=0,net=10.0.2.0/8,"
1186 "guestfwd=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT
1187 "-chardev:guestfsvmc");
1189 add_cmdline (g, "-net");
1190 add_cmdline (g, buf);
1192 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1195 /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel
1196 * syntax, or if not then we'll get a quick failure.
1199 snprintf (buf, sizeof buf,
1200 "channel," GUESTFWD_PORT ":unix:%s,server,nowait", unixsock);
1202 add_cmdline (g, "-net");
1203 add_cmdline (g, buf);
1204 add_cmdline (g, "-net");
1205 add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
1207 vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
1209 add_cmdline (g, "-net");
1210 add_cmdline (g, "nic,model=" NET_IF ",vlan=0");
1212 #define LINUX_CMDLINE \
1213 "panic=1 " /* force kernel to panic if daemon exits */ \
1214 "console=ttyS0 " /* serial console */ \
1215 "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \
1216 "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \
1217 "acpi=off " /* we don't need ACPI, turn it off */ \
1218 "printk.time=1 " /* display timestamp before kernel messages */ \
1219 "cgroup_disable=memory " /* saves us about 5 MB of RAM */
1221 /* Linux kernel command line. */
1222 snprintf (buf, sizeof buf,
1224 "%s " /* (selinux) */
1225 "%s " /* (vmchannel) */
1226 "%s " /* (verbose) */
1227 "%s", /* (append) */
1228 g->selinux ? "selinux=1 enforcing=0" : "selinux=0",
1229 vmchannel ? vmchannel : "",
1230 g->verbose ? "guestfs_verbose=1" : "",
1231 g->append ? g->append : "");
1233 add_cmdline (g, "-kernel");
1234 add_cmdline (g, (char *) kernel);
1235 add_cmdline (g, "-initrd");
1236 add_cmdline (g, (char *) initrd);
1237 add_cmdline (g, "-append");
1238 add_cmdline (g, buf);
1240 /* Finish off the command line. */
1241 incr_cmdline_size (g);
1242 g->cmdline[g->cmdline_size-1] = NULL;
1248 /* Set up stdin, stdout. */
1254 if (dup (wfd[0]) == -1) {
1256 perror ("dup failed");
1259 if (dup (rfd[1]) == -1)
1267 /* Set up a new process group, so we can signal this process
1268 * and all subprocesses (eg. if qemu is really a shell script).
1273 setenv ("LC_ALL", "C", 1);
1275 execv (g->qemu, g->cmdline); /* Run qemu. */
1280 /* Parent (library). */
1288 /* Fork the recovery process off which will kill qemu if the parent
1289 * process fails to do so (eg. if the parent segfaults).
1291 g->recoverypid = -1;
1292 if (g->recovery_proc) {
1295 pid_t qemu_pid = g->pid;
1296 pid_t parent_pid = getppid ();
1298 /* Writing to argv is hideously complicated and error prone. See:
1299 * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain
1302 /* Loop around waiting for one or both of the other processes to
1303 * disappear. It's fair to say this is very hairy. The PIDs that
1304 * we are looking at might be reused by another process. We are
1305 * effectively polling. Is the cure worse than the disease?
1308 if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */
1310 if (kill (parent_pid, 0) == -1) {
1311 /* Parent's gone away, qemu still around, so kill qemu. */
1319 /* Don't worry, if the fork failed, this will be -1. The recovery
1320 * process isn't essential.
1326 /* Close the other ends of the pipe. */
1330 if (fcntl (wfd[1], F_SETFL, O_NONBLOCK) == -1 ||
1331 fcntl (rfd[0], F_SETFL, O_NONBLOCK) == -1) {
1332 perrorf (g, "fcntl");
1336 g->fd[0] = wfd[1]; /* stdin of child */
1337 g->fd[1] = rfd[0]; /* stdout of child */
1339 g->fd[0] = open ("/dev/null", O_RDWR);
1340 if (g->fd[0] == -1) {
1341 perrorf (g, "open /dev/null");
1344 g->fd[1] = dup (g->fd[0]);
1345 if (g->fd[1] == -1) {
1352 if (null_vmchannel_sock) {
1356 /* Null vmchannel implementation: We listen on g->sock for a
1357 * connection. The connection could come from any local process
1358 * so we must check it comes from the appliance (or at least
1359 * from our UID) for security reasons.
1361 while (sock == -1) {
1362 sock = accept_from_daemon (g);
1366 if (check_peer_euid (g, sock, &uid) == -1)
1368 if (uid != geteuid ()) {
1370 "libguestfs: warning: unexpected connection from UID %d to port %d\n",
1371 uid, null_vmchannel_sock);
1377 if (fcntl (sock, F_SETFL, O_NONBLOCK) == -1) {
1378 perrorf (g, "fcntl");
1385 /* Other vmchannel. Open the Unix socket.
1387 * The vmchannel implementation that got merged with qemu sucks in
1388 * a number of ways. Both ends do connect(2), which means that no
1389 * one knows what, if anything, is connected to the other end, or
1390 * if it becomes disconnected. Even worse, we have to wait some
1391 * indeterminate time for qemu to create the socket and connect to
1392 * it (which happens very early in qemu's start-up), so any code
1393 * that uses vmchannel is inherently racy. Hence this silly loop.
1395 g->sock = socket (AF_UNIX, SOCK_STREAM, 0);
1396 if (g->sock == -1) {
1397 perrorf (g, "socket");
1401 if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
1402 perrorf (g, "fcntl");
1406 addr.sun_family = AF_UNIX;
1407 strncpy (addr.sun_path, unixsock, UNIX_PATH_MAX);
1408 addr.sun_path[UNIX_PATH_MAX-1] = '\0';
1411 /* Always sleep at least once to give qemu a small chance to start up. */
1414 r = connect (g->sock, (struct sockaddr *) &addr, sizeof addr);
1415 if ((r == -1 && errno == EINPROGRESS) || r == 0)
1418 if (errno != ENOENT)
1419 perrorf (g, "connect");
1424 error (g, _("failed to connect to vmchannel socket"));
1430 g->state = LAUNCHING;
1432 /* Wait for qemu to start and to connect back to us via vmchannel and
1433 * send the GUESTFS_LAUNCH_FLAG message.
1437 r = recv_from_daemon (g, &size, &buf);
1440 if (r == -1) return -1;
1442 if (size != GUESTFS_LAUNCH_FLAG) {
1443 error (g, _("guestfs_launch failed, see earlier error messages"));
1448 print_timestamped_message (g, "appliance is up");
1450 /* This is possible in some really strange situations, such as
1451 * guestfsd starts up OK but then qemu immediately exits. Check for
1452 * it because the caller is probably expecting to be able to send
1453 * commands after this function returns.
1455 if (g->state != READY) {
1456 error (g, _("qemu launched and contacted daemon, but state != READY"));
1468 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1469 waitpid (g->pid, NULL, 0);
1470 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1475 memset (&g->launch_t, 0, sizeof g->launch_t);
1488 /* This function is used to print the qemu command line before it gets
1489 * executed, when in verbose mode.
1492 print_cmdline (guestfs_h *g)
1497 while (g->cmdline[i]) {
1498 if (g->cmdline[i][0] == '-') /* -option starts a new line */
1499 fprintf (stderr, " \\\n ");
1501 if (i > 0) fputc (' ', stderr);
1503 /* Does it need shell quoting? This only deals with simple cases. */
1504 needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]);
1506 if (needs_quote) fputc ('\'', stderr);
1507 fprintf (stderr, "%s", g->cmdline[i]);
1508 if (needs_quote) fputc ('\'', stderr);
1512 fputc ('\n', stderr);
1515 /* This function does the hard work of building the supermin appliance
1516 * on the fly. 'path' is the directory containing the control files.
1517 * 'kernel' and 'initrd' are where we will return the names of the
1518 * kernel and initrd (only initrd is built). The work is done by
1519 * an external script. We just tell it where to put the result.
1522 build_supermin_appliance (guestfs_h *g, const char *path,
1523 char **kernel, char **initrd)
1529 print_timestamped_message (g, "begin building supermin appliance");
1531 len = strlen (g->tmpdir);
1532 *kernel = safe_malloc (g, len + 8);
1533 snprintf (*kernel, len+8, "%s/kernel", g->tmpdir);
1534 *initrd = safe_malloc (g, len + 8);
1535 snprintf (*initrd, len+8, "%s/initrd", g->tmpdir);
1537 snprintf (cmd, sizeof cmd,
1539 "libguestfs-supermin-helper%s '%s' " host_cpu " " REPO " %s %s",
1541 g->verbose ? " --verbose" : "",
1542 path, *kernel, *initrd);
1544 print_timestamped_message (g, "%s", cmd);
1547 if (r == -1 || WEXITSTATUS(r) != 0) {
1548 error (g, _("external command failed: %s"), cmd);
1551 *kernel = *initrd = NULL;
1556 print_timestamped_message (g, "finished building supermin appliance");
1561 /* Compute Y - X and return the result in milliseconds.
1562 * Approximately the same as this code:
1563 * http://www.mpp.mpg.de/~huber/util/timevaldiff.c
1566 timeval_diff (const struct timeval *x, const struct timeval *y)
1570 msec = (y->tv_sec - x->tv_sec) * 1000;
1571 msec += (y->tv_usec - x->tv_usec) / 1000;
1576 print_timestamped_message (guestfs_h *g, const char *fs, ...)
1583 va_start (args, fs);
1584 err = vasprintf (&msg, fs, args);
1587 if (err < 0) return;
1589 gettimeofday (&tv, NULL);
1591 fprintf (stderr, "[%05" PRIi64 "ms] %s\n",
1592 timeval_diff (&g->launch_t, &tv), msg);
1597 static int read_all (guestfs_h *g, FILE *fp, char **ret);
1599 /* Test qemu binary (or wrapper) runs, and do 'qemu -help' and
1600 * 'qemu -version' so we know what options this qemu supports and
1604 test_qemu (guestfs_h *g)
1609 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -help", g->qemu);
1611 fp = popen (cmd, "r");
1612 /* qemu -help should always work (qemu -version OTOH wasn't
1613 * supported by qemu 0.9). If this command doesn't work then it
1614 * probably indicates that the qemu binary is missing.
1617 /* XXX This error is never printed, even if the qemu binary
1618 * doesn't exist. Why?
1621 perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd);
1625 if (read_all (g, fp, &g->qemu_help) == -1)
1628 if (pclose (fp) == -1)
1631 snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -version 2>/dev/null", g->qemu);
1633 fp = popen (cmd, "r");
1635 /* Intentionally ignore errors. */
1636 read_all (g, fp, &g->qemu_version);
1644 read_all (guestfs_h *g, FILE *fp, char **ret)
1651 *ret = safe_realloc (g, *ret, n + 1);
1656 *ret = safe_realloc (g, *ret, n + BUFSIZ);
1658 r = fread (p, 1, BUFSIZ, fp);
1660 perrorf (g, "read");
1667 /* Test if option is supported by qemu command line (just by grepping
1670 * The first time this is used, it has to run the external qemu
1671 * binary. If that fails, it returns -1.
1673 * To just do the first-time run of the qemu binary, call this with
1674 * option == NULL, in which case it will return -1 if there was an
1678 qemu_supports (guestfs_h *g, const char *option)
1680 if (!g->qemu_help) {
1681 if (test_qemu (g) == -1)
1688 return strstr (g->qemu_help, option) != NULL;
1691 /* Check if a file can be opened. */
1693 is_openable (guestfs_h *g, const char *path, int flags)
1695 int fd = open (path, flags);
1705 /* Check the peer effective UID for a TCP socket. Ideally we'd like
1706 * SO_PEERCRED for a loopback TCP socket. This isn't possible on
1707 * Linux (but it is on Solaris!) so we read /proc/net/tcp instead.
1710 check_peer_euid (guestfs_h *g, int sock, uid_t *rtn)
1712 struct sockaddr_in peer;
1713 socklen_t addrlen = sizeof peer;
1715 if (getpeername (sock, (struct sockaddr *) &peer, &addrlen) == -1) {
1716 perrorf (g, "getpeername");
1720 if (peer.sin_family != AF_INET ||
1721 ntohl (peer.sin_addr.s_addr) != INADDR_LOOPBACK) {
1722 error (g, "check_peer_euid: unexpected connection from non-IPv4, non-loopback peer (family = %d, addr = %s)",
1723 peer.sin_family, inet_ntoa (peer.sin_addr));
1727 struct sockaddr_in our;
1728 addrlen = sizeof our;
1729 if (getsockname (sock, (struct sockaddr *) &our, &addrlen) == -1) {
1730 perrorf (g, "getsockname");
1734 FILE *fp = fopen ("/proc/net/tcp", "r");
1736 perrorf (g, "/proc/net/tcp");
1741 if (fgets (line, sizeof line, fp) == NULL) { /* Drop first line. */
1742 error (g, "unexpected end of file in /proc/net/tcp");
1747 while (fgets (line, sizeof line, fp) != NULL) {
1748 unsigned line_our_addr, line_our_port, line_peer_addr, line_peer_port;
1749 int dummy0, dummy1, dummy2, dummy3, dummy4, dummy5, dummy6;
1752 if (sscanf (line, "%d:%08X:%04X %08X:%04X %02X %08X:%08X %02X:%08X %08X %d",
1754 &line_our_addr, &line_our_port,
1755 &line_peer_addr, &line_peer_port,
1756 &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, &dummy6,
1758 /* Note about /proc/net/tcp: local_address and rem_address are
1759 * always in network byte order. However the port part is
1760 * always in host byte order.
1762 * The sockname and peername that we got above are in network
1763 * byte order. So we have to byte swap the port but not the
1766 if (line_our_addr == our.sin_addr.s_addr &&
1767 line_our_port == ntohs (our.sin_port) &&
1768 line_peer_addr == peer.sin_addr.s_addr &&
1769 line_peer_port == ntohs (peer.sin_port)) {
1777 error (g, "check_peer_euid: no matching TCP connection found in /proc/net/tcp");
1782 /* You had to call this function after launch in versions <= 1.0.70,
1783 * but it is now a no-op.
1786 guestfs__wait_ready (guestfs_h *g)
1788 if (g->state != READY) {
1789 error (g, _("qemu has not been launched yet"));
1797 guestfs__kill_subprocess (guestfs_h *g)
1799 if (g->state == CONFIG) {
1800 error (g, _("no subprocess to kill"));
1805 fprintf (stderr, "sending SIGTERM to process %d\n", g->pid);
1807 kill (g->pid, SIGTERM);
1808 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1813 /* Access current state. */
1815 guestfs__is_config (guestfs_h *g)
1817 return g->state == CONFIG;
1821 guestfs__is_launching (guestfs_h *g)
1823 return g->state == LAUNCHING;
1827 guestfs__is_ready (guestfs_h *g)
1829 return g->state == READY;
1833 guestfs__is_busy (guestfs_h *g)
1835 return g->state == BUSY;
1839 guestfs__get_state (guestfs_h *g)
1845 guestfs_set_log_message_callback (guestfs_h *g,
1846 guestfs_log_message_cb cb, void *opaque)
1848 g->log_message_cb = cb;
1849 g->log_message_cb_data = opaque;
1853 guestfs_set_subprocess_quit_callback (guestfs_h *g,
1854 guestfs_subprocess_quit_cb cb, void *opaque)
1856 g->subprocess_quit_cb = cb;
1857 g->subprocess_quit_cb_data = opaque;
1861 guestfs_set_launch_done_callback (guestfs_h *g,
1862 guestfs_launch_done_cb cb, void *opaque)
1864 g->launch_done_cb = cb;
1865 g->launch_done_cb_data = opaque;
1868 /*----------------------------------------------------------------------*/
1870 /* This is the code used to send and receive RPC messages and (for
1871 * certain types of message) to perform file transfers. This code is
1872 * driven from the generated actions (src/guestfs-actions.c). There
1873 * are five different cases to consider:
1875 * (1) A non-daemon function. There is no RPC involved at all, it's
1876 * all handled inside the library.
1878 * (2) A simple RPC (eg. "mount"). We write the request, then read
1879 * the reply. The sequence of calls is:
1881 * guestfs___set_busy
1884 * guestfs___end_busy
1886 * (3) An RPC with FileOut parameters (eg. "upload"). We write the
1887 * request, then write the file(s), then read the reply. The sequence
1890 * guestfs___set_busy
1892 * guestfs___send_file (possibly multiple times)
1894 * guestfs___end_busy
1896 * (4) An RPC with FileIn parameters (eg. "download"). We write the
1897 * request, then read the reply, then read the file(s). The sequence
1900 * guestfs___set_busy
1903 * guestfs___recv_file (possibly multiple times)
1904 * guestfs___end_busy
1906 * (5) Both FileOut and FileIn parameters. There are no calls like
1907 * this in the current API, but they would be implemented as a
1908 * combination of cases (3) and (4).
1910 * During all writes and reads, we also select(2) on qemu stdout
1911 * looking for messages (guestfsd stderr and guest kernel dmesg), and
1912 * anything received is passed up through the log_message_cb. This is
1913 * also the reason why all the sockets are non-blocking. We also have
1914 * to check for EOF (qemu died). All of this is handled by the
1915 * functions send_to_daemon and recv_from_daemon.
1919 guestfs___set_busy (guestfs_h *g)
1921 if (g->state != READY) {
1922 error (g, _("guestfs_set_busy: called when in state %d != READY"),
1931 guestfs___end_busy (guestfs_h *g)
1945 error (g, _("guestfs_end_busy: called when in state %d"), g->state);
1951 /* This is called if we detect EOF, ie. qemu died. */
1953 child_cleanup (guestfs_h *g)
1956 fprintf (stderr, "child_cleanup: %p: child process died\n", g);
1958 /*kill (g->pid, SIGTERM);*/
1959 if (g->recoverypid > 0) kill (g->recoverypid, 9);
1960 waitpid (g->pid, NULL, 0);
1961 if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
1970 memset (&g->launch_t, 0, sizeof g->launch_t);
1972 if (g->subprocess_quit_cb)
1973 g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
1977 read_log_message_or_eof (guestfs_h *g, int fd)
1985 "read_log_message_or_eof: %p g->state = %d, fd = %d\n",
1989 /* QEMU's console emulates a 16550A serial port. The real 16550A
1990 * device has a small FIFO buffer (16 bytes) which means here we see
1991 * lots of small reads of 1-16 bytes in length, usually single
1994 n = read (fd, buf, sizeof buf);
1996 /* Hopefully this indicates the qemu child process has died. */
2002 if (errno == EINTR || errno == EAGAIN)
2005 perrorf (g, "read");
2009 /* In verbose mode, copy all log messages to stderr. */
2011 ignore_value (write (STDERR_FILENO, buf, n));
2013 /* It's an actual log message, send it upwards if anyone is listening. */
2014 if (g->log_message_cb)
2015 g->log_message_cb (g, g->log_message_cb_data, buf, n);
2021 check_for_daemon_cancellation_or_eof (guestfs_h *g, int fd)
2030 "check_for_daemon_cancellation_or_eof: %p g->state = %d, fd = %d\n",
2033 n = read (fd, buf, 4);
2035 /* Hopefully this indicates the qemu child process has died. */
2041 if (errno == EINTR || errno == EAGAIN)
2044 perrorf (g, "read");
2048 xdrmem_create (&xdr, buf, 4, XDR_DECODE);
2049 xdr_uint32_t (&xdr, &flag);
2052 if (flag != GUESTFS_CANCEL_FLAG) {
2053 error (g, _("check_for_daemon_cancellation_or_eof: read 0x%x from daemon, expected 0x%x\n"),
2054 flag, GUESTFS_CANCEL_FLAG);
2061 /* This writes the whole N bytes of BUF to the daemon socket.
2063 * If the whole write is successful, it returns 0.
2064 * If there was an error, it returns -1.
2065 * If the daemon sent a cancellation message, it returns -2.
2067 * It also checks qemu stdout for log messages and passes those up
2068 * through log_message_cb.
2070 * It also checks for EOF (qemu died) and passes that up through the
2071 * child_cleanup function above.
2074 send_to_daemon (guestfs_h *g, const void *v_buf, size_t n)
2076 const char *buf = v_buf;
2082 "send_to_daemon: %p g->state = %d, n = %zu\n", g, g->state, n);
2087 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2088 FD_SET (g->sock, &rset); /* Read socket for cancellation & EOF. */
2089 FD_SET (g->sock, &wset); /* Write to socket to send the data. */
2091 int max_fd = MAX (g->sock, g->fd[1]);
2096 int r = select (max_fd+1, &rset2, &wset2, NULL, NULL);
2098 if (errno == EINTR || errno == EAGAIN)
2100 perrorf (g, "select");
2104 if (FD_ISSET (g->fd[1], &rset2)) {
2105 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2108 if (FD_ISSET (g->sock, &rset2)) {
2109 r = check_for_daemon_cancellation_or_eof (g, g->sock);
2113 if (FD_ISSET (g->sock, &wset2)) {
2114 r = write (g->sock, buf, n);
2116 if (errno == EINTR || errno == EAGAIN)
2118 perrorf (g, "write");
2119 if (errno == EPIPE) /* Disconnected from guest (RHBZ#508713). */
2131 /* This reads a single message, file chunk, launch flag or
2132 * cancellation flag from the daemon. If something was read, it
2133 * returns 0, otherwise -1.
2135 * Both size_rtn and buf_rtn must be passed by the caller as non-NULL.
2137 * *size_rtn returns the size of the returned message or it may be
2138 * GUESTFS_LAUNCH_FLAG or GUESTFS_CANCEL_FLAG.
2140 * *buf_rtn is returned containing the message (if any) or will be set
2141 * to NULL. *buf_rtn must be freed by the caller.
2143 * It also checks qemu stdout for log messages and passes those up
2144 * through log_message_cb.
2146 * It also checks for EOF (qemu died) and passes that up through the
2147 * child_cleanup function above.
2150 recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn)
2156 "recv_from_daemon: %p g->state = %d, size_rtn = %p, buf_rtn = %p\n",
2157 g, g->state, size_rtn, buf_rtn);
2161 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2162 FD_SET (g->sock, &rset); /* Read socket for data & EOF. */
2164 int max_fd = MAX (g->sock, g->fd[1]);
2170 /* nr is the size of the message, but we prime it as -4 because we
2171 * have to read the message length word first.
2175 while (nr < (ssize_t) *size_rtn) {
2177 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2179 if (errno == EINTR || errno == EAGAIN)
2181 perrorf (g, "select");
2187 if (FD_ISSET (g->fd[1], &rset2)) {
2188 if (read_log_message_or_eof (g, g->fd[1]) == -1) {
2194 if (FD_ISSET (g->sock, &rset2)) {
2195 if (nr < 0) { /* Have we read the message length word yet? */
2196 r = read (g->sock, lenbuf+nr+4, -nr);
2198 if (errno == EINTR || errno == EAGAIN)
2201 perrorf (g, "read");
2202 /* Under some circumstances we see "Connection reset by peer"
2203 * here when the child dies suddenly. Catch this and call
2204 * the cleanup function, same as for EOF.
2206 if (err == ECONNRESET)
2211 error (g, _("unexpected end of file when reading from daemon"));
2217 if (nr < 0) /* Still not got the whole length word. */
2221 xdrmem_create (&xdr, lenbuf, 4, XDR_DECODE);
2222 xdr_uint32_t (&xdr, size_rtn);
2225 if (*size_rtn == GUESTFS_LAUNCH_FLAG) {
2226 if (g->state != LAUNCHING)
2227 error (g, _("received magic signature from guestfsd, but in state %d"),
2231 if (g->launch_done_cb)
2232 g->launch_done_cb (g, g->launch_done_cb_data);
2236 else if (*size_rtn == GUESTFS_CANCEL_FLAG)
2238 /* If this happens, it's pretty bad and we've probably lost
2241 else if (*size_rtn > GUESTFS_MESSAGE_MAX) {
2242 error (g, _("message length (%u) > maximum possible size (%d)"),
2243 (unsigned) *size_rtn, GUESTFS_MESSAGE_MAX);
2247 /* Allocate the complete buffer, size now known. */
2248 *buf_rtn = safe_malloc (g, *size_rtn);
2252 size_t sizetoread = *size_rtn - nr;
2253 if (sizetoread > BUFSIZ) sizetoread = BUFSIZ;
2255 r = read (g->sock, (char *) (*buf_rtn) + nr, sizetoread);
2257 if (errno == EINTR || errno == EAGAIN)
2259 perrorf (g, "read");
2265 error (g, _("unexpected end of file when reading from daemon"));
2275 /* Got the full message, caller can start processing it. */
2276 #ifdef ENABLE_PACKET_DUMP
2280 for (i = 0; i < nr; i += 16) {
2281 printf ("%04zx: ", i);
2282 for (j = i; j < MIN (i+16, nr); ++j)
2283 printf ("%02x ", (*(unsigned char **)buf_rtn)[j]);
2284 for (; j < i+16; ++j)
2287 for (j = i; j < MIN (i+16, nr); ++j)
2288 if (c_isprint ((*(char **)buf_rtn)[j]))
2289 printf ("%c", (*(char **)buf_rtn)[j]);
2292 for (; j < i+16; ++j)
2302 /* This is very much like recv_from_daemon above, but g->sock is
2303 * a listening socket and we are accepting a new connection on
2304 * that socket instead of reading anything. Returns the newly
2308 accept_from_daemon (guestfs_h *g)
2314 "accept_from_daemon: %p g->state = %d\n", g, g->state);
2318 FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
2319 FD_SET (g->sock, &rset); /* Read socket for accept. */
2321 int max_fd = MAX (g->sock, g->fd[1]);
2324 while (sock == -1) {
2326 int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
2328 if (errno == EINTR || errno == EAGAIN)
2330 perrorf (g, "select");
2334 if (FD_ISSET (g->fd[1], &rset2)) {
2335 if (read_log_message_or_eof (g, g->fd[1]) == -1)
2338 if (FD_ISSET (g->sock, &rset2)) {
2339 sock = accept (g->sock, NULL, NULL);
2341 if (errno == EINTR || errno == EAGAIN)
2343 perrorf (g, "accept");
2353 guestfs___send (guestfs_h *g, int proc_nr, xdrproc_t xdrp, char *args)
2355 struct guestfs_message_header hdr;
2358 int serial = g->msg_next_serial++;
2361 size_t msg_out_size;
2363 if (g->state != BUSY) {
2364 error (g, _("guestfs___send: state %d != BUSY"), g->state);
2368 /* We have to allocate this message buffer on the heap because
2369 * it is quite large (although will be mostly unused). We
2370 * can't allocate it on the stack because in some environments
2371 * we have quite limited stack space available, notably when
2372 * running in the JVM.
2374 msg_out = safe_malloc (g, GUESTFS_MESSAGE_MAX + 4);
2375 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MESSAGE_MAX, XDR_ENCODE);
2377 /* Serialize the header. */
2378 hdr.prog = GUESTFS_PROGRAM;
2379 hdr.vers = GUESTFS_PROTOCOL_VERSION;
2381 hdr.direction = GUESTFS_DIRECTION_CALL;
2382 hdr.serial = serial;
2383 hdr.status = GUESTFS_STATUS_OK;
2385 if (!xdr_guestfs_message_header (&xdr, &hdr)) {
2386 error (g, _("xdr_guestfs_message_header failed"));
2390 /* Serialize the args. If any, because some message types
2391 * have no parameters.
2394 if (!(*xdrp) (&xdr, args)) {
2395 error (g, _("dispatch failed to marshal args"));
2400 /* Get the actual length of the message, resize the buffer to match
2401 * the actual length, and write the length word at the beginning.
2403 len = xdr_getpos (&xdr);
2406 msg_out = safe_realloc (g, msg_out, len + 4);
2407 msg_out_size = len + 4;
2409 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2410 xdr_uint32_t (&xdr, &len);
2413 r = send_to_daemon (g, msg_out, msg_out_size);
2414 if (r == -2) /* Ignore stray daemon cancellations. */
2427 static int cancel = 0; /* XXX Implement file cancellation. */
2428 static int send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t len);
2429 static int send_file_data (guestfs_h *g, const char *buf, size_t len);
2430 static int send_file_cancellation (guestfs_h *g);
2431 static int send_file_complete (guestfs_h *g);
2437 * -2 daemon cancelled (we must read the error message)
2440 guestfs___send_file (guestfs_h *g, const char *filename)
2442 char buf[GUESTFS_MAX_CHUNK_SIZE];
2445 fd = open (filename, O_RDONLY);
2447 perrorf (g, "open: %s", filename);
2448 send_file_cancellation (g);
2449 /* Daemon sees cancellation and won't reply, so caller can
2455 /* Send file in chunked encoding. */
2457 r = read (fd, buf, sizeof buf);
2458 if (r == -1 && (errno == EINTR || errno == EAGAIN))
2461 err = send_file_data (g, buf, r);
2463 if (err == -2) /* daemon sent cancellation */
2464 send_file_cancellation (g);
2469 if (cancel) { /* cancel from either end */
2470 send_file_cancellation (g);
2475 perrorf (g, "read: %s", filename);
2476 send_file_cancellation (g);
2480 /* End of file, but before we send that, we need to close
2481 * the file and check for errors.
2483 if (close (fd) == -1) {
2484 perrorf (g, "close: %s", filename);
2485 send_file_cancellation (g);
2489 return send_file_complete (g);
2492 /* Send a chunk of file data. */
2494 send_file_data (guestfs_h *g, const char *buf, size_t len)
2496 return send_file_chunk (g, 0, buf, len);
2499 /* Send a cancellation message. */
2501 send_file_cancellation (guestfs_h *g)
2503 return send_file_chunk (g, 1, NULL, 0);
2506 /* Send a file complete chunk. */
2508 send_file_complete (guestfs_h *g)
2511 return send_file_chunk (g, 0, buf, 0);
2515 send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t buflen)
2519 guestfs_chunk chunk;
2522 size_t msg_out_size;
2524 if (g->state != BUSY) {
2525 error (g, _("send_file_chunk: state %d != READY"), g->state);
2529 /* Allocate the chunk buffer. Don't use the stack to avoid
2530 * excessive stack usage and unnecessary copies.
2532 msg_out = safe_malloc (g, GUESTFS_MAX_CHUNK_SIZE + 4 + 48);
2533 xdrmem_create (&xdr, msg_out + 4, GUESTFS_MAX_CHUNK_SIZE + 48, XDR_ENCODE);
2535 /* Serialize the chunk. */
2536 chunk.cancel = cancel;
2537 chunk.data.data_len = buflen;
2538 chunk.data.data_val = (char *) buf;
2540 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2541 error (g, _("xdr_guestfs_chunk failed (buf = %p, buflen = %zu)"),
2547 len = xdr_getpos (&xdr);
2550 /* Reduce the size of the outgoing message buffer to the real length. */
2551 msg_out = safe_realloc (g, msg_out, len + 4);
2552 msg_out_size = len + 4;
2554 xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE);
2555 xdr_uint32_t (&xdr, &len);
2557 r = send_to_daemon (g, msg_out, msg_out_size);
2559 /* Did the daemon send a cancellation message? */
2562 fprintf (stderr, "got daemon cancellation\n");
2578 /* Receive a reply. */
2580 guestfs___recv (guestfs_h *g, const char *fn,
2581 guestfs_message_header *hdr,
2582 guestfs_message_error *err,
2583 xdrproc_t xdrp, char *ret)
2591 r = recv_from_daemon (g, &size, &buf);
2595 /* This can happen if a cancellation happens right at the end
2596 * of us sending a FileIn parameter to the daemon. Discard. The
2597 * daemon should send us an error message next.
2599 if (size == GUESTFS_CANCEL_FLAG)
2602 if (size == GUESTFS_LAUNCH_FLAG) {
2603 error (g, "%s: received unexpected launch flag from daemon when expecting reply", fn);
2607 xdrmem_create (&xdr, buf, size, XDR_DECODE);
2609 if (!xdr_guestfs_message_header (&xdr, hdr)) {
2610 error (g, "%s: failed to parse reply header", fn);
2615 if (hdr->status == GUESTFS_STATUS_ERROR) {
2616 if (!xdr_guestfs_message_error (&xdr, err)) {
2617 error (g, "%s: failed to parse reply error", fn);
2623 if (xdrp && ret && !xdrp (&xdr, ret)) {
2624 error (g, "%s: failed to parse reply", fn);
2636 /* Receive a file. */
2638 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2639 static ssize_t receive_file_data (guestfs_h *g, void **buf);
2642 guestfs___recv_file (guestfs_h *g, const char *filename)
2647 fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2649 perrorf (g, "open: %s", filename);
2653 /* Receive the file in chunked encoding. */
2654 while ((r = receive_file_data (g, &buf)) > 0) {
2655 if (xwrite (fd, buf, r) == -1) {
2656 perrorf (g, "%s: write", filename);
2664 error (g, _("%s: error in chunked encoding"), filename);
2668 if (close (fd) == -1) {
2669 perrorf (g, "close: %s", filename);
2676 /* Send cancellation message to daemon, then wait until it
2677 * cancels (just throwing away data).
2681 uint32_t flag = GUESTFS_CANCEL_FLAG;
2684 fprintf (stderr, "%s: waiting for daemon to acknowledge cancellation\n",
2687 xdrmem_create (&xdr, fbuf, sizeof fbuf, XDR_ENCODE);
2688 xdr_uint32_t (&xdr, &flag);
2691 if (xwrite (g->sock, fbuf, sizeof fbuf) == -1) {
2692 perrorf (g, _("write to daemon socket"));
2696 while (receive_file_data (g, NULL) > 0)
2697 ; /* just discard it */
2702 /* Receive a chunk of file data. */
2703 /* Returns -1 = error, 0 = EOF, > 0 = more data */
2705 receive_file_data (guestfs_h *g, void **buf_r)
2711 guestfs_chunk chunk;
2713 r = recv_from_daemon (g, &len, &buf);
2715 error (g, _("receive_file_data: parse error in reply callback"));
2719 if (len == GUESTFS_LAUNCH_FLAG || len == GUESTFS_CANCEL_FLAG) {
2720 error (g, _("receive_file_data: unexpected flag received when reading file chunks"));
2724 memset (&chunk, 0, sizeof chunk);
2726 xdrmem_create (&xdr, buf, len, XDR_DECODE);
2727 if (!xdr_guestfs_chunk (&xdr, &chunk)) {
2728 error (g, _("failed to parse file chunk"));
2733 /* After decoding, the original buffer is no longer used. */
2737 error (g, _("file receive cancelled by daemon"));
2738 free (chunk.data.data_val);
2742 if (chunk.data.data_len == 0) { /* end of transfer */
2743 free (chunk.data.data_val);
2747 if (buf_r) *buf_r = chunk.data.data_val;
2748 else free (chunk.data.data_val); /* else caller frees */
2750 return chunk.data.data_len;