static int
read_log_message_or_eof (guestfs_h *g, int fd, int error_if_eof)
{
char buf[BUFSIZ];
ssize_t n;
#if 0
debug (g, "read_log_message_or_eof: %p g->state = %d, fd = %d",
g, g->state, fd);
#endif
/* QEMU's console emulates a 16550A serial port. The real 16550A
* device has a small FIFO buffer (16 bytes) which means here we see
* lots of small reads of 1-16 bytes in length, usually single
* bytes. Sleeping here for a very brief period groups reads
* together (so we usually get a few lines of output at once) and
* improves overall throughput, as well as making the event
* interface a bit more sane for callers. With a virtio-serial
* based console (not yet implemented) we may be able to remove
* this. XXX
*/
usleep (1000);
n = read (fd, buf, sizeof buf);
if (n == 0) {
/* Hopefully this indicates the qemu child process has died. */
child_cleanup (g);
if (error_if_eof) {
/* We weren't expecting eof here (called from launch) so place
* something in the error buffer. RHBZ#588851.
*/
error (g, "child process died unexpectedly");
}
return -1;
}
if (n == -1) {
if (errno == EINTR || errno == EAGAIN)
return 0;
perrorf (g, "read");
return -1;
}
/* It's an actual log message, send it upwards if anyone is listening. */
guestfs___call_callbacks_message (g, GUESTFS_EVENT_APPLIANCE, buf, n);
/* This is a gross hack. See the comment above
* guestfs___launch_send_progress.
*/
if (g->state == LAUNCHING) {
const char *sentinel;
size_t len;
sentinel = "Linux version"; /* kernel up */
len = strlen (sentinel);
if (memmem (buf, n, sentinel, len) != NULL)
guestfs___launch_send_progress (g, 6);
sentinel = "Starting /init script"; /* /init running */
len = strlen (sentinel);
if (memmem (buf, n, sentinel, len) != NULL)
guestfs___launch_send_progress (g, 9);
}
return 0;
}
static int
launch_direct (guestfs_h *g, void *datav, const char *arg)
{
struct backend_direct_data *data = datav;
CLEANUP_FREE_STRINGSBUF DECLARE_STRINGSBUF (cmdline);
int daemon_accept_sock = -1, console_sock = -1;
int r;
int flags;
int sv[2];
char guestfsd_sock[256];
struct sockaddr_un addr;
CLEANUP_FREE char *kernel = NULL, *dtb = NULL,
*initrd = NULL, *appliance = NULL;
int has_appliance_drive;
CLEANUP_FREE char *appliance_dev = NULL;
uint32_t size;
CLEANUP_FREE void *buf = NULL;
struct drive *drv;
size_t i;
int virtio_scsi;
struct hv_param *hp;
bool has_kvm;
bool force_tcg;
/* At present you must add drives before starting the appliance. In
* future when we enable hotplugging you won't need to do this.
*/
if (!g->nr_drives) {
error (g, _("you must call guestfs_add_drive before guestfs_launch"));
return -1;
}
force_tcg = guestfs___get_backend_setting_bool (g, "force_tcg");
if (!force_tcg)
debian_kvm_warning (g);
guestfs___launch_send_progress (g, 0);
TRACE0 (launch_build_appliance_start);
/* Locate and/or build the appliance. */
if (guestfs___build_appliance (g, &kernel, &dtb, &initrd, &appliance) == -1)
return -1;
has_appliance_drive = appliance != NULL;
TRACE0 (launch_build_appliance_end);
guestfs___launch_send_progress (g, 3);
if (g->verbose)
guestfs___print_timestamped_message (g, "begin testing qemu features");
/* Get qemu help text and version. */
if (qemu_supports (g, data, NULL) == -1)
goto cleanup0;
/* Using virtio-serial, we need to create a local Unix domain socket
* for qemu to connect to.
*/
snprintf (guestfsd_sock, sizeof guestfsd_sock, "%s/guestfsd.sock", g->tmpdir);
unlink (guestfsd_sock);
daemon_accept_sock = socket (AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC, 0);
if (daemon_accept_sock == -1) {
perrorf (g, "socket");
goto cleanup0;
}
addr.sun_family = AF_UNIX;
strncpy (addr.sun_path, guestfsd_sock, UNIX_PATH_MAX);
addr.sun_path[UNIX_PATH_MAX-1] = '\0';
if (bind (daemon_accept_sock, &addr, sizeof addr) == -1) {
perrorf (g, "bind");
goto cleanup0;
}
if (listen (daemon_accept_sock, 1) == -1) {
perrorf (g, "listen");
goto cleanup0;
}
if (!g->direct_mode) {
if (socketpair (AF_LOCAL, SOCK_STREAM|SOCK_CLOEXEC, 0, sv) == -1) {
perrorf (g, "socketpair");
goto cleanup0;
}
}
if (g->verbose)
guestfs___print_timestamped_message (g, "finished testing qemu features");
/* Construct the qemu command line. We have to do this before
* forking, because after fork we are not allowed to use
* non-signal-safe functions such as malloc.
*/
#define ADD_CMDLINE(str) \
guestfs___add_string (g, &cmdline, (str))
#define ADD_CMDLINE_STRING_NODUP(str) \
guestfs___add_string_nodup (g, &cmdline, (str))
#define ADD_CMDLINE_PRINTF(fs,...) \
guestfs___add_sprintf (g, &cmdline, (fs), ##__VA_ARGS__)
ADD_CMDLINE (g->hv);
/* CVE-2011-4127 mitigation: Disable SCSI ioctls on virtio-blk
* devices. The -global option must exist, but you can pass any
* strings to it so we don't need to check for the specific virtio
* feature.
*/
if (qemu_supports (g, data, "-global")) {
ADD_CMDLINE ("-global");
ADD_CMDLINE (VIRTIO_BLK ".scsi=off");
}
if (qemu_supports (g, data, "-nodefconfig"))
ADD_CMDLINE ("-nodefconfig");
/* This oddly named option doesn't actually enable FIPS. It just
* causes qemu to do the right thing if FIPS is enabled in the
* kernel. So like libvirt, we pass it unconditionally.
*/
if (qemu_supports (g, data, "-enable-fips"))
ADD_CMDLINE ("-enable-fips");
/* Newer versions of qemu (from around 2009/12) changed the
* behaviour of monitors so that an implicit '-monitor stdio' is
* assumed if we are in -nographic mode and there is no other
* -monitor option. Only a single stdio device is allowed, so
* this broke the '-serial stdio' option. There is a new flag
* called -nodefaults which gets rid of all this default crud, so
* let's use that to avoid this and any future surprises.
*/
if (qemu_supports (g, data, "-nodefaults"))
ADD_CMDLINE ("-nodefaults");
ADD_CMDLINE ("-display");
ADD_CMDLINE ("none");
#ifdef MACHINE_TYPE
ADD_CMDLINE ("-M");
ADD_CMDLINE (MACHINE_TYPE);
#endif
/* If this is uncommented, then qemu won't start running the
* appliance immediately. It will wait for you to connect to it
* using gdb:
*
* $ gdb
* (gdb) symbol-file /path/to/vmlinux
* (gdb) target remote tcp::1234
* (gdb) cont
*
* You can then debug the appliance kernel, which is useful to debug
* boot failures (especially ones where there are no debug messages
* printed - tip: look in the kernel log_buf).
*
* On Fedora, install kernel-debuginfo for the vmlinux file
* (containing symbols). Make sure the symbols precisely match the
* kernel being used.
*/
#if 0
ADD_CMDLINE ("-S");
ADD_CMDLINE ("-s");
warning (g, "qemu debugging is enabled, connect gdb to tcp::1234 to begin");
#endif
/* Try to guess if KVM is available. We are just checking that
* /dev/kvm is openable. That's not reliable, since /dev/kvm
* might be openable by qemu but not by us (think: SELinux) in
* which case the user would not get hardware virtualization,
* although at least shouldn't fail.
*/
has_kvm = is_openable (g, "/dev/kvm", O_RDWR|O_CLOEXEC);
/* The qemu -machine option (added 2010-12) is a bit more sane
* since it falls back through various different acceleration
* modes, so try that first (thanks Markus Armbruster).
*/
if (qemu_supports (g, data, "-machine")) {
ADD_CMDLINE ("-machine");
if (!force_tcg)
ADD_CMDLINE ("accel=kvm:tcg");
else
ADD_CMDLINE ("accel=tcg");
} else {
/* qemu sometimes needs this option to enable hardware
* virtualization, but some versions of 'qemu-kvm' will use KVM
* regardless (even where this option appears in the help text).
* It is rumoured that there are versions of qemu where
* supplying this option when hardware virtualization is not
* available will cause qemu to fail. A giant clusterfuck with
* the qemu command line, again.
*/
if (has_kvm && !force_tcg && qemu_supports (g, data, "-enable-kvm"))
ADD_CMDLINE ("-enable-kvm");
}
if (g->smp > 1) {
ADD_CMDLINE ("-smp");
ADD_CMDLINE_PRINTF ("%d", g->smp);
}
ADD_CMDLINE ("-m");
ADD_CMDLINE_PRINTF ("%d", g->memsize);
/* Force exit instead of reboot on panic */
ADD_CMDLINE ("-no-reboot");
/* These are recommended settings, see RHBZ#1053847. */
ADD_CMDLINE ("-rtc");
ADD_CMDLINE ("driftfix=slew");
#ifndef __arm__
/* qemu-system-arm advertises the -no-hpet option but if you try
* to use it, it usefully says:
* "Option no-hpet not supported for this target".
* Cheers qemu developers. How many years have we been asking for
* capabilities? Could be 3 or 4 years, I forget.
*/
ADD_CMDLINE ("-no-hpet");
#endif
ADD_CMDLINE ("-no-kvm-pit-reinjection");
ADD_CMDLINE ("-kernel");
ADD_CMDLINE (kernel);
if (dtb) {
ADD_CMDLINE ("-dtb");
ADD_CMDLINE (dtb);
}
ADD_CMDLINE ("-initrd");
ADD_CMDLINE (initrd);
/* Add drives */
virtio_scsi = qemu_supports_virtio_scsi (g, data);
if (virtio_scsi) {
/* Create the virtio-scsi bus. */
ADD_CMDLINE ("-device");
ADD_CMDLINE (VIRTIO_SCSI ",id=scsi");
}
ITER_DRIVES (g, i, drv) {
CLEANUP_FREE char *file = NULL, *escaped_file = NULL, *param = NULL;
if (!drv->overlay) {
/* Make the file= parameter. */
file = guestfs___drive_source_qemu_param (g, &drv->src);
escaped_file = qemu_escape_param (g, file);
/* Make the first part of the -drive parameter, everything up to
* the if=... at the end.
*/
param = safe_asprintf
(g, "file=%s%s,cache=%s%s%s%s%s,id=hd%zu",
escaped_file,
drv->readonly ? ",snapshot=on" : "",
drv->cachemode ? drv->cachemode : "writeback",
drv->src.format ? ",format=" : "",
drv->src.format ? drv->src.format : "",
drv->disk_label ? ",serial=" : "",
drv->disk_label ? drv->disk_label : "",
i);
}
else {
/* Writable qcow2 overlay on top of read-only drive. */
escaped_file = qemu_escape_param (g, drv->overlay);
param = safe_asprintf
(g, "file=%s,cache=unsafe,format=qcow2%s%s,id=hd%zu",
escaped_file,
drv->disk_label ? ",serial=" : "",
drv->disk_label ? drv->disk_label : "",
i);
}
/* If there's an explicit 'iface', use it. Otherwise default to
* virtio-scsi if available. Otherwise default to virtio-blk.
*/
if (drv->iface && STREQ (drv->iface, "virtio")) /* virtio-blk */
goto virtio_blk;
#if defined(__arm__) || defined(__powerpc__)
else if (drv->iface && STREQ (drv->iface, "ide")) {
error (g, "'ide' interface does not work on ARM or PowerPC");
goto cleanup0;
}
#endif
else if (drv->iface) {
ADD_CMDLINE ("-drive");
ADD_CMDLINE_PRINTF ("%s,if=%s", param, drv->iface);
}
else if (virtio_scsi) {
ADD_CMDLINE ("-drive");
ADD_CMDLINE_PRINTF ("%s,if=none" /* sic */, param);
ADD_CMDLINE ("-device");
ADD_CMDLINE_PRINTF ("scsi-hd,drive=hd%zu", i);
}
else {
virtio_blk:
ADD_CMDLINE ("-drive");
ADD_CMDLINE_PRINTF ("%s,if=none" /* sic */, param);
ADD_CMDLINE ("-device");
ADD_CMDLINE_PRINTF (VIRTIO_BLK ",drive=hd%zu", i);
}
}
