X-Git-Url: http://git.annexia.org/?p=libguestfs.git;a=blobdiff_plain;f=src%2Flaunch.c;h=7b3372c2bc85d43d596ff351eaf62cb15c9ace25;hp=8dfeb3150095d147d577e24b4397b1425f5f9251;hb=3814680423984b3c46c2f99e944c2a71862bde9f;hpb=10167cea98f93a74abe63f0a54d3a662997e7489 diff --git a/src/launch.c b/src/launch.c index 8dfeb31..7b3372c 100644 --- a/src/launch.c +++ b/src/launch.c @@ -63,6 +63,7 @@ #include #include "c-ctype.h" +#include "ignore-value.h" #include "glthread/lock.h" #include "guestfs.h" @@ -72,12 +73,42 @@ static int launch_appliance (guestfs_h *g); static int64_t timeval_diff (const struct timeval *x, const struct timeval *y); +static void print_qemu_command_line (guestfs_h *g, char **argv); static int connect_unix_socket (guestfs_h *g, const char *sock); static int qemu_supports (guestfs_h *g, const char *option); +static char *qemu_drive_param (guestfs_h *g, const struct drive *drv); + +#if 0 +static int qemu_supports_re (guestfs_h *g, const pcre *option_regex); + +static void compile_regexps (void) __attribute__((constructor)); +static void free_regexps (void) __attribute__((destructor)); -/* Add a string to the current command line. */ static void -incr_cmdline_size (guestfs_h *g) +compile_regexps (void) +{ + const char *err; + int offset; + +#define COMPILE(re,pattern,options) \ + do { \ + re = pcre_compile ((pattern), (options), &err, &offset, NULL); \ + if (re == NULL) { \ + ignore_value (write (2, err, strlen (err))); \ + abort (); \ + } \ + } while (0) +} + +static void +free_regexps (void) +{ +} +#endif + +/* Functions to add a string to the current command line. */ +static void +alloc_cmdline (guestfs_h *g) { if (g->cmdline == NULL) { /* g->cmdline[0] is reserved for argv[0], set in guestfs_launch. */ @@ -85,7 +116,12 @@ incr_cmdline_size (guestfs_h *g) g->cmdline = safe_malloc (g, sizeof (char *)); g->cmdline[0] = NULL; } +} +static void +incr_cmdline_size (guestfs_h *g) +{ + alloc_cmdline (g); g->cmdline_size++; g->cmdline = safe_realloc (g, g->cmdline, sizeof (char *) * g->cmdline_size); } @@ -104,23 +140,18 @@ add_cmdline (guestfs_h *g, const char *str) return 0; } -size_t -guestfs___checkpoint_cmdline (guestfs_h *g) +struct drive ** +guestfs___checkpoint_drives (guestfs_h *g) { - return g->cmdline_size; + struct drive **i = &g->drives; + while (*i != NULL) i = &((*i)->next); + return i; } void -guestfs___rollback_cmdline (guestfs_h *g, size_t pos) +guestfs___rollback_drives (guestfs_h *g, struct drive **i) { - size_t i; - - assert (g->cmdline_size >= pos); - - for (i = g->cmdline_size - 1; i >= pos; --i) - free (g->cmdline[i]); - - g->cmdline_size = pos; + guestfs___free_drives(i); } /* Internal command to return the command line. */ @@ -130,11 +161,7 @@ guestfs__debug_cmdline (guestfs_h *g) size_t i; char **r; - if (g->cmdline == NULL) { - r = safe_malloc (g, sizeof (char *) * 1); - r[0] = NULL; - return r; - } + alloc_cmdline (g); r = safe_malloc (g, sizeof (char *) * (g->cmdline_size + 1)); r[0] = safe_strdup (g, g->qemu); /* g->cmdline[0] is always NULL */ @@ -147,6 +174,27 @@ guestfs__debug_cmdline (guestfs_h *g) return r; /* caller frees */ } +/* Internal command to return the list of drives. */ +char ** +guestfs__debug_drives (guestfs_h *g) +{ + size_t i, count; + char **ret; + struct drive *drv; + + for (count = 0, drv = g->drives; drv; count++, drv = drv->next) + ; + + ret = safe_malloc (g, sizeof (char *) * (count + 1)); + + for (i = 0, drv = g->drives; drv; i++, drv = drv->next) + ret[i] = qemu_drive_param (g, drv); + + ret[count] = NULL; + + return ret; /* caller frees */ +} + int guestfs__config (guestfs_h *g, const char *qemu_param, const char *qemu_value) @@ -185,17 +233,21 @@ guestfs__config (guestfs_h *g, * O_DIRECT. This fails on some filesystem types (notably tmpfs). * So we check if we can open the file with or without O_DIRECT, * and use cache=off (or not) accordingly. + * + * NB: This function is only called on the !readonly path. We must + * try to open with O_RDWR to test that the file is readable and + * writable here. */ static int test_cache_off (guestfs_h *g, const char *filename) { - int fd = open (filename, O_RDONLY|O_DIRECT); + int fd = open (filename, O_RDWR|O_DIRECT); if (fd >= 0) { close (fd); return 1; } - fd = open (filename, O_RDONLY); + fd = open (filename, O_RDWR); if (fd >= 0) { close (fd); return 0; @@ -228,8 +280,10 @@ guestfs__add_drive_opts (guestfs_h *g, const char *filename, const struct guestfs_add_drive_opts_argv *optargs) { int readonly; - const char *format; - const char *iface; + char *format; + char *iface; + char *name; + int use_cache_off; if (strchr (filename, ',') != NULL) { error (g, _("filename cannot contain ',' (comma) character")); @@ -239,18 +293,26 @@ guestfs__add_drive_opts (guestfs_h *g, const char *filename, readonly = optargs->bitmask & GUESTFS_ADD_DRIVE_OPTS_READONLY_BITMASK ? optargs->readonly : 0; format = optargs->bitmask & GUESTFS_ADD_DRIVE_OPTS_FORMAT_BITMASK - ? optargs->format : NULL; + ? safe_strdup (g, optargs->format) : NULL; iface = optargs->bitmask & GUESTFS_ADD_DRIVE_OPTS_IFACE_BITMASK - ? optargs->iface : DRIVE_IF; + ? safe_strdup (g, optargs->iface) : safe_strdup (g, DRIVE_IF); + name = optargs->bitmask & GUESTFS_ADD_DRIVE_OPTS_NAME_BITMASK + ? safe_strdup (g, optargs->name) : NULL; if (format && !valid_format_iface (format)) { error (g, _("%s parameter is empty or contains disallowed characters"), "format"); + free (format); + free (iface); + free (name); return -1; } if (!valid_format_iface (iface)) { error (g, _("%s parameter is empty or contains disallowed characters"), "iface"); + free (format); + free (iface); + free (name); return -1; } @@ -258,31 +320,37 @@ guestfs__add_drive_opts (guestfs_h *g, const char *filename, * checks for the existence of the file. For readonly we have * to do the check explicitly. */ - int use_cache_off = readonly ? 0 : test_cache_off (g, filename); - if (use_cache_off == -1) + use_cache_off = readonly ? 0 : test_cache_off (g, filename); + if (use_cache_off == -1) { + free (format); + free (iface); + free (name); return -1; + } if (readonly) { - if (access (filename, F_OK) == -1) { + if (access (filename, R_OK) == -1) { perrorf (g, "%s", filename); + free (format); + free (iface); + free (name); return -1; } } - /* Construct the final -drive parameter. */ - size_t len = 64 + strlen (filename) + strlen (iface); - if (format) len += strlen (format); - char buf[len]; + struct drive **i = &(g->drives); + while (*i != NULL) i = &((*i)->next); - snprintf (buf, len, "file=%s%s%s%s%s,if=%s", - filename, - readonly ? ",snapshot=on" : "", - use_cache_off ? ",cache=off" : "", - format ? ",format=" : "", - format ? format : "", - iface); + *i = safe_malloc (g, sizeof (struct drive)); + (*i)->next = NULL; + (*i)->path = safe_strdup (g, filename); + (*i)->readonly = readonly; + (*i)->format = format; + (*i)->iface = iface; + (*i)->name = name; + (*i)->use_cache_off = use_cache_off; - return guestfs__config (g, "-drive", buf); + return 0; } int @@ -359,6 +427,8 @@ guestfs__launch (guestfs_h *g) return -1; } + TRACE0 (launch_start); + /* Make the temporary directory. */ if (!g->tmpdir) { TMP_TEMPLATE_ON_STACK (dir_template); @@ -400,7 +470,7 @@ launch_appliance (guestfs_h *g) /* 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->cmdline) { + if (!g->drives) { error (g, _("you must call guestfs_add_drive before guestfs_launch")); return -1; } @@ -409,11 +479,15 @@ launch_appliance (guestfs_h *g) gettimeofday (&g->launch_t, NULL); guestfs___launch_send_progress (g, 0); + TRACE0 (launch_build_appliance_start); + /* Locate and/or build the appliance. */ char *kernel = NULL, *initrd = NULL, *appliance = NULL; if (guestfs___build_appliance (g, &kernel, &initrd, &appliance) == -1) return -1; + TRACE0 (launch_build_appliance_end); + guestfs___launch_send_progress (g, 3); if (g->verbose) @@ -482,26 +556,66 @@ launch_appliance (guestfs_h *g) /* Set up the full command line. Do this in the subprocess so we * don't need to worry about cleaning up. */ + + /* Set g->cmdline[0] to the name of the qemu process. However + * it is possible that no g->cmdline has been allocated yet so + * we must do that first. + */ + alloc_cmdline (g); g->cmdline[0] = g->qemu; + /* Add drives */ + struct drive *drv = g->drives; + while (drv != NULL) { + /* Construct the final -drive parameter. */ + char *buf = qemu_drive_param (g, drv); + + add_cmdline (g, "-drive"); + add_cmdline (g, buf); + free (buf); + + drv = drv->next; + } + if (qemu_supports (g, "-nodefconfig")) add_cmdline (g, "-nodefconfig"); - /* 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, so we we have to check at least 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. A giant clusterfuck with the - * qemu command line, again. + /* 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, "-enable-kvm") && - is_openable (g, "/dev/kvm", O_RDWR)) - add_cmdline (g, "-enable-kvm"); + if (qemu_supports (g, "-machine")) { + add_cmdline (g, "-machine"); +#if QEMU_MACHINE_TYPE_IS_BROKEN + /* Workaround for qemu 0.15: We have to add the '[type=]pc' + * since there is no default. This is not a permanent solution + * because this only works on PC-like hardware. Other platforms + * like ppc would need a different machine type. + * + * This bug is fixed in qemu commit 2645c6dcaf6ea2a51a, and was + * not a problem in qemu < 0.15. + */ + add_cmdline (g, "pc,accel=kvm:tcg"); +#else + add_cmdline (g, "accel=kvm:tcg"); +#endif + } 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, so we we have to check at least 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. A giant clusterfuck with the + * qemu command line, again. + */ + if (qemu_supports (g, "-enable-kvm") && + is_openable (g, "/dev/kvm", O_RDWR)) + add_cmdline (g, "-enable-kvm"); + } /* Newer versions of qemu (from around 2009/12) changed the * behaviour of monitors so that an implicit '-monitor stdio' is @@ -516,6 +630,12 @@ launch_appliance (guestfs_h *g) add_cmdline (g, "-nographic"); + if (g->smp > 1) { + snprintf (buf, sizeof buf, "%d", g->smp); + add_cmdline (g, "-smp"); + add_cmdline (g, buf); + } + snprintf (buf, sizeof buf, "%d", g->memsize); add_cmdline (g, "-m"); add_cmdline (g, buf); @@ -567,7 +687,7 @@ launch_appliance (guestfs_h *g) "panic=1 " /* force kernel to panic if daemon exits */ \ "console=ttyS0 " /* serial console */ \ "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \ - "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \ + "no_timer_check " /* fix for RHBZ#502058 */ \ "acpi=off " /* we don't need ACPI, turn it off */ \ "printk.time=1 " /* display timestamp before kernel messages */ \ "cgroup_disable=memory " /* saves us about 5 MB of RAM */ @@ -612,21 +732,28 @@ launch_appliance (guestfs_h *g) incr_cmdline_size (g); g->cmdline[g->cmdline_size-1] = NULL; - if (g->verbose) - guestfs___print_timestamped_argv (g, (const char **)g->cmdline); - if (!g->direct) { - /* Set up stdin, stdout. */ + /* Set up stdin, stdout, stderr. */ close (0); close (1); close (wfd[1]); close (rfd[0]); + /* Stdin. */ if (dup (wfd[0]) == -1) { dup_failed: perror ("dup failed"); _exit (EXIT_FAILURE); } + /* Stdout. */ + if (dup (rfd[1]) == -1) + goto dup_failed; + + /* Particularly since qemu 0.15, qemu spews all sorts of debug + * information on stderr. It is useful to both capture this and + * not confuse casual users, so send stderr to the pipe as well. + */ + close (2); if (dup (rfd[1]) == -1) goto dup_failed; @@ -634,15 +761,18 @@ launch_appliance (guestfs_h *g) close (rfd[1]); } -#if 0 - /* Set up a new process group, so we can signal this process - * and all subprocesses (eg. if qemu is really a shell script). - */ - setpgid (0, 0); -#endif + /* Dump the command line (after setting up stderr above). */ + if (g->verbose) + print_qemu_command_line (g, g->cmdline); + + /* Put qemu in a new process group. */ + if (g->pgroup) + setpgid (0, 0); setenv ("LC_ALL", "C", 1); + TRACE0 (launch_run_qemu); + execv (g->qemu, g->cmdline); /* Run qemu. */ perror (g->qemu); _exit (EXIT_FAILURE); @@ -668,8 +798,16 @@ launch_appliance (guestfs_h *g) pid_t qemu_pid = g->pid; pid_t parent_pid = getppid (); + /* It would be nice to be able to put this in the same process + * group as qemu (ie. setpgid (0, qemu_pid)). However this is + * not possible because we don't have any guarantee here that + * the qemu process has started yet. + */ + if (g->pgroup) + setpgid (0, 0); + /* Writing to argv is hideously complicated and error prone. See: - * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain + * http://git.postgresql.org/gitweb/?p=postgresql.git;a=blob;f=src/backend/utils/misc/ps_status.c;hb=HEAD */ /* Loop around waiting for one or both of the other processes to @@ -731,6 +869,13 @@ launch_appliance (guestfs_h *g) if (r == -1) goto cleanup1; + /* NB: We reach here just because qemu has opened the socket. It + * does not mean the daemon is up until we read the + * GUESTFS_LAUNCH_FLAG below. Failures in qemu startup can still + * happen even if we reach here, even early failures like not being + * able to open a drive. + */ + close (g->sock); /* Close the listening socket. */ g->sock = r; /* This is the accepted data socket. */ @@ -764,6 +909,8 @@ launch_appliance (guestfs_h *g) goto cleanup1; } + TRACE0 (launch_end); + guestfs___launch_send_progress (g, 12); return 0; @@ -949,47 +1096,9 @@ timeval_diff (const struct timeval *x, const struct timeval *y) return msec; } -void -guestfs___print_timestamped_argv (guestfs_h *g, const char * argv[]) -{ - int i = 0; - int needs_quote; - char *buf = NULL; - size_t len; - FILE *fp; - - fp = open_memstream (&buf, &len); - if (fp == NULL) { - warning (g, "open_memstream: %m"); - return; - } - - struct timeval tv; - gettimeofday (&tv, NULL); - fprintf (fp, "[%05" PRIi64 "ms] ", timeval_diff (&g->launch_t, &tv)); - - while (argv[i]) { - if (argv[i][0] == '-') /* -option starts a new line */ - fprintf (fp, " \\\n "); - - if (i > 0) fputc (' ', fp); - - /* Does it need shell quoting? This only deals with simple cases. */ - needs_quote = strcspn (argv[i], " ") != strlen (argv[i]); - - if (needs_quote) fputc ('\'', fp); - fprintf (fp, "%s", argv[i]); - if (needs_quote) fputc ('\'', fp); - i++; - } - - fclose (fp); - - debug (g, "%s", buf); - - free (buf); -} - +/* Note that since this calls 'debug' it should only be called + * from the parent process. + */ void guestfs___print_timestamped_message (guestfs_h *g, const char *fs, ...) { @@ -1011,6 +1120,37 @@ guestfs___print_timestamped_message (guestfs_h *g, const char *fs, ...) free (msg); } +/* This is called from the forked subprocess just before qemu runs, so + * it can just print the message straight to stderr, where it will be + * picked up and funnelled through the usual appliance event API. + */ +static void +print_qemu_command_line (guestfs_h *g, char **argv) +{ + int i = 0; + int needs_quote; + + struct timeval tv; + gettimeofday (&tv, NULL); + fprintf (stderr, "[%05" PRIi64 "ms] ", timeval_diff (&g->launch_t, &tv)); + + while (argv[i]) { + if (argv[i][0] == '-') /* -option starts a new line */ + fprintf (stderr, " \\\n "); + + if (i > 0) fputc (' ', stderr); + + /* Does it need shell quoting? This only deals with simple cases. */ + needs_quote = strcspn (argv[i], " ") != strlen (argv[i]); + + if (needs_quote) fputc ('\'', stderr); + fprintf (stderr, "%s", argv[i]); + if (needs_quote) fputc ('\'', stderr); + i++; + } +} + +static int test_qemu_cmd (guestfs_h *g, const char *cmd, char **ret); static int read_all (guestfs_h *g, FILE *fp, char **ret); /* Test qemu binary (or wrapper) runs, and do 'qemu -help' and @@ -1023,38 +1163,49 @@ test_qemu (guestfs_h *g) char cmd[1024]; FILE *fp; + free (g->qemu_help); + g->qemu_help = NULL; + free (g->qemu_version); + g->qemu_version = NULL; + snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -nographic -help", g->qemu); - fp = popen (cmd, "r"); /* qemu -help should always work (qemu -version OTOH wasn't * supported by qemu 0.9). If this command doesn't work then it * probably indicates that the qemu binary is missing. */ - if (!fp) { - /* XXX This error is never printed, even if the qemu binary - * doesn't exist. Why? - */ - error: - perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd); + if (test_qemu_cmd (g, cmd, &g->qemu_help) == -1) { + error (g, _("command failed: %s\n\nIf qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU\nenvironment variable. There may also be errors printed above."), + cmd); return -1; } - if (read_all (g, fp, &g->qemu_help) == -1) - goto error; - - if (pclose (fp) == -1) - goto error; - snprintf (cmd, sizeof cmd, "LC_ALL=C '%s' -nographic -version 2>/dev/null", g->qemu); + /* Intentionally ignore errors from qemu -version. */ + ignore_value (test_qemu_cmd (g, cmd, &g->qemu_version)); + + return 0; +} + +static int +test_qemu_cmd (guestfs_h *g, const char *cmd, char **ret) +{ + FILE *fp; + fp = popen (cmd, "r"); - if (fp) { - /* Intentionally ignore errors. */ - read_all (g, fp, &g->qemu_version); + if (fp == NULL) + return -1; + + if (read_all (g, fp, ret) == -1) { pclose (fp); + return -1; } + if (pclose (fp) != 0) + return -1; + return 0; } @@ -1106,6 +1257,20 @@ qemu_supports (guestfs_h *g, const char *option) return strstr (g->qemu_help, option) != NULL; } +#if 0 +/* As above but using a regex instead of a fixed string. */ +static int +qemu_supports_re (guestfs_h *g, const pcre *option_regex) +{ + if (!g->qemu_help) { + if (test_qemu (g) == -1) + return -1; + } + + return match (g, g->qemu_help, option_regex); +} +#endif + /* Check if a file can be opened. */ static int is_openable (guestfs_h *g, const char *path, int flags) @@ -1119,6 +1284,30 @@ is_openable (guestfs_h *g, const char *path, int flags) return 1; } +static char * +qemu_drive_param (guestfs_h *g, const struct drive *drv) +{ + size_t len = 64; + char *r; + + len += strlen (drv->path); + len += strlen (drv->iface); + if (drv->format) + len += strlen (drv->format); + + r = safe_malloc (g, len); + + snprintf (r, len, "file=%s%s%s%s%s,if=%s", + drv->path, + drv->readonly ? ",snapshot=on" : "", + drv->use_cache_off ? ",cache=off" : "", + drv->format ? ",format=" : "", + drv->format ? drv->format : "", + drv->iface); + + return r; /* caller frees */ +} + /* You had to call this function after launch in versions <= 1.0.70, * but it is now a no-op. */