X-Git-Url: http://git.annexia.org/?p=libguestfs.git;a=blobdiff_plain;f=src%2Fguestfs.c;h=ec7473ea55e509eb037e64ae66af1734fe00a90b;hp=1b7d61ca96bea0bedf088f6491188f43b09a6788;hb=8b217a87bf9175e7e02a5913e5617e0d12dfd09c;hpb=170f262f0413de843af62b968f6d12c1c476ae7f diff --git a/src/guestfs.c b/src/guestfs.c index 1b7d61c..ec7473e 100644 --- a/src/guestfs.c +++ b/src/guestfs.c @@ -1,5 +1,5 @@ /* libguestfs - * Copyright (C) 2009 Red Hat Inc. + * Copyright (C) 2009 Red Hat Inc. * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public @@ -19,17 +19,20 @@ #include #define _BSD_SOURCE /* for mkdtemp, usleep */ -#define _GNU_SOURCE /* for vasprintf, GNU strerror_r, strchrnul */ #include #include #include +#include #include #include #include #include #include +#include #include +#include + #include #include @@ -54,80 +57,36 @@ #endif #include "guestfs.h" +#include "guestfs-internal-actions.h" #include "guestfs_protocol.h" +#include "ignore-value.h" + +#ifdef HAVE_GETTEXT +#include "gettext.h" +#define _(str) dgettext(PACKAGE, (str)) +//#define N_(str) dgettext(PACKAGE, (str)) +#else +#define _(str) str +//#define N_(str) str +#endif #define error guestfs_error #define perrorf guestfs_perrorf #define safe_malloc guestfs_safe_malloc #define safe_realloc guestfs_safe_realloc #define safe_strdup guestfs_safe_strdup -#define safe_memdup guestfs_safe_memdup +//#define safe_memdup guestfs_safe_memdup static void default_error_cb (guestfs_h *g, void *data, const char *msg); -static void stdout_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data, int watch, int fd, int events); -static void sock_read_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data, int watch, int fd, int events); -static void sock_write_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data, int watch, int fd, int events); - +static int send_to_daemon (guestfs_h *g, const void *v_buf, size_t n); +static int recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn); static void close_handles (void); -static int select_add_handle (guestfs_main_loop *ml, guestfs_h *g, int fd, int events, guestfs_handle_event_cb cb, void *data); -static int select_remove_handle (guestfs_main_loop *ml, guestfs_h *g, int watch); -static int select_add_timeout (guestfs_main_loop *ml, guestfs_h *g, int interval, guestfs_handle_timeout_cb cb, void *data); -static int select_remove_timeout (guestfs_main_loop *ml, guestfs_h *g, int timer); -static int select_main_loop_run (guestfs_main_loop *ml, guestfs_h *g); -static int select_main_loop_quit (guestfs_main_loop *ml, guestfs_h *g); - -/* Default select-based main loop. */ -struct select_handle_cb_data { - guestfs_handle_event_cb cb; - guestfs_h *g; - void *data; -}; - -struct select_main_loop { - /* NB. These fields must be the same as in struct guestfs_main_loop: */ - guestfs_add_handle_cb add_handle; - guestfs_remove_handle_cb remove_handle; - guestfs_add_timeout_cb add_timeout; - guestfs_remove_timeout_cb remove_timeout; - guestfs_main_loop_run_cb main_loop_run; - guestfs_main_loop_quit_cb main_loop_quit; - - /* Additional private data: */ - int is_running; - - fd_set rset; - fd_set wset; - fd_set xset; - - int max_fd; - int nr_fds; - struct select_handle_cb_data *handle_cb_data; -}; - -/* Default main loop. */ -static struct select_main_loop default_main_loop = { - .add_handle = select_add_handle, - .remove_handle = select_remove_handle, - .add_timeout = select_add_timeout, - .remove_timeout = select_remove_timeout, - .main_loop_run = select_main_loop_run, - .main_loop_quit = select_main_loop_quit, - - /* XXX hopefully .rset, .wset, .xset are initialized to the empty - * set by the normal action of everything being initialized to zero. - */ - .is_running = 0, - .max_fd = -1, - .nr_fds = 0, - .handle_cb_data = NULL, -}; - #define UNIX_PATH_MAX 108 /* Also in guestfsd.c */ -#define VMCHANNEL_PORT 6666 -#define VMCHANNEL_ADDR "10.0.2.4" +#define GUESTFWD_ADDR "10.0.2.4" +#define GUESTFWD_PORT "6666" /* GuestFS handle and connection. */ enum state { CONFIG, LAUNCHING, READY, BUSY, NO_HANDLE }; @@ -141,21 +100,28 @@ struct guestfs_h int fd[2]; /* Stdin/stdout of qemu. */ int sock; /* Daemon communications socket. */ - int pid; /* Qemu PID. */ + pid_t pid; /* Qemu PID. */ + pid_t recoverypid; /* Recovery process PID. */ time_t start_t; /* The time when we started qemu. */ - int stdout_watch; /* Watches qemu stdout for log messages. */ - int sock_watch; /* Watches daemon comm socket. */ - char *tmpdir; /* Temporary directory containing socket. */ + char *qemu_help, *qemu_version; /* Output of qemu -help, qemu -version. */ + char **cmdline; /* Qemu command line. */ int cmdline_size; int verbose; + int trace; int autosync; - const char *path; + char *path; /* Path to kernel, initrd. */ + char *qemu; /* Qemu binary. */ + char *append; /* Append to kernel command line. */ + + int memsize; /* Size of RAM (megabytes). */ + + int selinux; /* selinux enabled? */ char *last_error; @@ -163,10 +129,6 @@ struct guestfs_h guestfs_abort_cb abort_cb; guestfs_error_handler_cb error_cb; void * error_cb_data; - guestfs_send_cb send_cb; - void * send_cb_data; - guestfs_reply_cb reply_cb; - void * reply_cb_data; guestfs_log_message_cb log_message_cb; void * log_message_cb_data; guestfs_subprocess_quit_cb subprocess_quit_cb; @@ -174,15 +136,6 @@ struct guestfs_h guestfs_launch_done_cb launch_done_cb; void * launch_done_cb_data; - /* Main loop used by this handle. */ - guestfs_main_loop *main_loop; - - /* Messages sent and received from the daemon. */ - char *msg_in; - int msg_in_size, msg_in_allocated; - char *msg_out; - int msg_out_size, msg_out_pos; - int msg_next_serial; }; @@ -205,8 +158,6 @@ guestfs_create (void) g->fd[0] = -1; g->fd[1] = -1; g->sock = -1; - g->stdout_watch = -1; - g->sock_watch = -1; g->abort_cb = abort; g->error_cb = default_error_cb; @@ -215,11 +166,38 @@ guestfs_create (void) str = getenv ("LIBGUESTFS_DEBUG"); g->verbose = str != NULL && strcmp (str, "1") == 0; + str = getenv ("LIBGUESTFS_TRACE"); + g->trace = str != NULL && strcmp (str, "1") == 0; + str = getenv ("LIBGUESTFS_PATH"); - g->path = str != NULL ? str : GUESTFS_DEFAULT_PATH; - /* XXX We should probably make QEMU configurable as well. */ + g->path = str != NULL ? strdup (str) : strdup (GUESTFS_DEFAULT_PATH); + if (!g->path) goto error; + + str = getenv ("LIBGUESTFS_QEMU"); + g->qemu = str != NULL ? strdup (str) : strdup (QEMU); + if (!g->qemu) goto error; + + str = getenv ("LIBGUESTFS_APPEND"); + if (str) { + g->append = strdup (str); + if (!g->append) goto error; + } - g->main_loop = guestfs_get_default_main_loop (); + /* Choose a suitable memory size. Previously we tried to choose + * a minimal memory size, but this isn't really necessary since + * recent QEMU and KVM don't do anything nasty like locking + * memory into core any more. Thus we can safely choose a + * large, generous amount of memory, and it'll just get swapped + * on smaller systems. + */ + str = getenv ("LIBGUESTFS_MEMSIZE"); + if (str) { + if (sscanf (str, "%d", &g->memsize) != 1 || g->memsize <= 256) { + fprintf (stderr, "libguestfs: non-numeric or too small value for LIBGUESTFS_MEMSIZE\n"); + goto error; + } + } else + g->memsize = 500; /* Start with large serial numbers so they are easy to spot * inside the protocol. @@ -242,6 +220,13 @@ guestfs_create (void) fprintf (stderr, "new guestfs handle %p\n", g); return g; + + error: + free (g->path); + free (g->qemu); + free (g->append); + free (g); + return NULL; } void @@ -253,7 +238,7 @@ guestfs_close (guestfs_h *g) if (g->state == NO_HANDLE) { /* Not safe to call 'error' here, so ... */ - fprintf (stderr, "guestfs_close: called twice on the same handle\n"); + fprintf (stderr, _("guestfs_close: called twice on the same handle\n")); return; } @@ -261,8 +246,10 @@ guestfs_close (guestfs_h *g) fprintf (stderr, "closing guestfs handle %p (state %d)\n", g, g->state); /* Try to sync if autosync flag is set. */ - if (g->autosync && g->state == READY) + if (g->autosync && g->state == READY) { + guestfs_umount_all (g); guestfs_sync (g); + } /* Remove any handlers that might be called back before we kill the * subprocess. @@ -272,10 +259,32 @@ guestfs_close (guestfs_h *g) if (g->state != CONFIG) guestfs_kill_subprocess (g); + /* Close sockets. */ + if (g->fd[0] >= 0) + close (g->fd[0]); + if (g->fd[1] >= 0) + close (g->fd[1]); + if (g->sock >= 0) + close (g->sock); + g->fd[0] = -1; + g->fd[1] = -1; + g->sock = -1; + + /* Wait for subprocess(es) to exit. */ + waitpid (g->pid, NULL, 0); + if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0); + + /* Remove tmpfiles. */ if (g->tmpdir) { snprintf (filename, sizeof filename, "%s/sock", g->tmpdir); unlink (filename); + snprintf (filename, sizeof filename, "%s/initrd", g->tmpdir); + unlink (filename); + + snprintf (filename, sizeof filename, "%s/kernel", g->tmpdir); + unlink (filename); + rmdir (g->tmpdir); free (g->tmpdir); @@ -301,6 +310,11 @@ guestfs_close (guestfs_h *g) /* release mutex (XXX) */ free (g->last_error); + free (g->path); + free (g->qemu); + free (g->append); + free (g->qemu_help); + free (g->qemu_version); free (g); } @@ -327,7 +341,7 @@ set_last_error (guestfs_h *g, const char *msg) static void default_error_cb (guestfs_h *g, void *data, const char *msg) { - fprintf (stderr, "libguestfs: error: %s\n", msg); + fprintf (stderr, _("libguestfs: error: %s\n"), msg); } void @@ -337,9 +351,11 @@ guestfs_error (guestfs_h *g, const char *fs, ...) char *msg; va_start (args, fs); - vasprintf (&msg, fs, args); + int err = vasprintf (&msg, fs, args); va_end (args); + if (err < 0) return; + if (g->error_cb) g->error_cb (g, g->error_cb_data, msg); set_last_error (g, msg); @@ -351,19 +367,21 @@ guestfs_perrorf (guestfs_h *g, const char *fs, ...) { va_list args; char *msg; - int err = errno; + int errnum = errno; va_start (args, fs); - vasprintf (&msg, fs, args); + int err = vasprintf (&msg, fs, args); va_end (args); + if (err < 0) return; + #ifndef _GNU_SOURCE char buf[256]; - strerror_r (err, buf, sizeof buf); + strerror_r (errnum, buf, sizeof buf); #else char _buf[256]; char *buf; - buf = strerror_r (err, _buf, sizeof _buf); + buf = strerror_r (errnum, _buf, sizeof _buf); #endif msg = safe_realloc (g, msg, strlen (msg) + 2 + strlen (buf) + 1); @@ -380,15 +398,51 @@ void * guestfs_safe_malloc (guestfs_h *g, size_t nbytes) { void *ptr = malloc (nbytes); - if (!ptr) g->abort_cb (); + if (nbytes > 0 && !ptr) g->abort_cb (); return ptr; } +/* Return 1 if an array of N objects, each of size S, cannot exist due + to size arithmetic overflow. S must be positive and N must be + nonnegative. This is a macro, not an inline function, so that it + works correctly even when SIZE_MAX < N. + + By gnulib convention, SIZE_MAX represents overflow in size + calculations, so the conservative dividend to use here is + SIZE_MAX - 1, since SIZE_MAX might represent an overflowed value. + However, malloc (SIZE_MAX) fails on all known hosts where + sizeof (ptrdiff_t) <= sizeof (size_t), so do not bother to test for + exactly-SIZE_MAX allocations on such hosts; this avoids a test and + branch when S is known to be 1. */ +# define xalloc_oversized(n, s) \ + ((size_t) (sizeof (ptrdiff_t) <= sizeof (size_t) ? -1 : -2) / (s) < (n)) + +/* Technically we should add an autoconf test for this, testing for the desired + functionality, like what's done in gnulib, but for now, this is fine. */ +#define HAVE_GNU_CALLOC (__GLIBC__ >= 2) + +/* Allocate zeroed memory for N elements of S bytes, with error + checking. S must be nonzero. */ +void * +guestfs_safe_calloc (guestfs_h *g, size_t n, size_t s) +{ + /* From gnulib's calloc function in xmalloc.c. */ + void *p; + /* Test for overflow, since some calloc implementations don't have + proper overflow checks. But omit overflow and size-zero tests if + HAVE_GNU_CALLOC, since GNU calloc catches overflow and never + returns NULL if successful. */ + if ((! HAVE_GNU_CALLOC && xalloc_oversized (n, s)) + || (! (p = calloc (n, s)) && (HAVE_GNU_CALLOC || n != 0))) + g->abort_cb (); + return p; +} + void * guestfs_safe_realloc (guestfs_h *g, void *ptr, int nbytes) { void *p = realloc (ptr, nbytes); - if (!p) g->abort_cb (); + if (nbytes > 0 && !p) g->abort_cb (); return p; } @@ -410,8 +464,9 @@ guestfs_safe_memdup (guestfs_h *g, void *ptr, size_t size) } static int -xwrite (int fd, const void *buf, size_t len) +xwrite (int fd, const void *v_buf, size_t len) { + const char *buf = v_buf; int r; while (len > 0) { @@ -426,23 +481,6 @@ xwrite (int fd, const void *buf, size_t len) return 0; } -static int -xread (int fd, void *buf, size_t len) -{ - int r; - - while (len > 0) { - r = read (fd, buf, len); - if (r == -1) - return -1; - - buf += r; - len -= r; - } - - return 0; -} - void guestfs_set_out_of_memory_handler (guestfs_h *g, guestfs_abort_cb cb) { @@ -470,47 +508,144 @@ guestfs_get_error_handler (guestfs_h *g, void **data_rtn) } int -guestfs_set_verbose (guestfs_h *g, int v) +guestfs__set_verbose (guestfs_h *g, int v) { g->verbose = !!v; return 0; } int -guestfs_get_verbose (guestfs_h *g) +guestfs__get_verbose (guestfs_h *g) { return g->verbose; } int -guestfs_set_autosync (guestfs_h *g, int a) +guestfs__set_autosync (guestfs_h *g, int a) { g->autosync = !!a; return 0; } int -guestfs_get_autosync (guestfs_h *g) +guestfs__get_autosync (guestfs_h *g) { return g->autosync; } int -guestfs_set_path (guestfs_h *g, const char *path) +guestfs__set_path (guestfs_h *g, const char *path) { - if (path == NULL) - g->path = GUESTFS_DEFAULT_PATH; - else - g->path = path; + free (g->path); + g->path = NULL; + + g->path = + path == NULL ? + safe_strdup (g, GUESTFS_DEFAULT_PATH) : safe_strdup (g, path); return 0; } const char * -guestfs_get_path (guestfs_h *g) +guestfs__get_path (guestfs_h *g) { return g->path; } +int +guestfs__set_qemu (guestfs_h *g, const char *qemu) +{ + free (g->qemu); + g->qemu = NULL; + + g->qemu = qemu == NULL ? safe_strdup (g, QEMU) : safe_strdup (g, qemu); + return 0; +} + +const char * +guestfs__get_qemu (guestfs_h *g) +{ + return g->qemu; +} + +int +guestfs__set_append (guestfs_h *g, const char *append) +{ + free (g->append); + g->append = NULL; + + g->append = append ? safe_strdup (g, append) : NULL; + return 0; +} + +const char * +guestfs__get_append (guestfs_h *g) +{ + return g->append; +} + +int +guestfs__set_memsize (guestfs_h *g, int memsize) +{ + g->memsize = memsize; + return 0; +} + +int +guestfs__get_memsize (guestfs_h *g) +{ + return g->memsize; +} + +int +guestfs__set_selinux (guestfs_h *g, int selinux) +{ + g->selinux = selinux; + return 0; +} + +int +guestfs__get_selinux (guestfs_h *g) +{ + return g->selinux; +} + +int +guestfs__get_pid (guestfs_h *g) +{ + if (g->pid > 0) + return g->pid; + else { + error (g, "get_pid: no qemu subprocess"); + return -1; + } +} + +struct guestfs_version * +guestfs__version (guestfs_h *g) +{ + struct guestfs_version *r; + + r = safe_malloc (g, sizeof *r); + r->major = PACKAGE_VERSION_MAJOR; + r->minor = PACKAGE_VERSION_MINOR; + r->release = PACKAGE_VERSION_RELEASE; + r->extra = safe_strdup (g, PACKAGE_VERSION_EXTRA); + return r; +} + +int +guestfs__set_trace (guestfs_h *g, int t) +{ + g->trace = !!t; + return 0; +} + +int +guestfs__get_trace (guestfs_h *g) +{ + return g->trace; +} + /* Add a string to the current command line. */ static void incr_cmdline_size (guestfs_h *g) @@ -530,7 +665,8 @@ static int add_cmdline (guestfs_h *g, const char *str) { if (g->state != CONFIG) { - error (g, "command line cannot be altered after qemu subprocess launched"); + error (g, + _("command line cannot be altered after qemu subprocess launched")); return -1; } @@ -540,11 +676,11 @@ add_cmdline (guestfs_h *g, const char *str) } int -guestfs_config (guestfs_h *g, - const char *qemu_param, const char *qemu_value) +guestfs__config (guestfs_h *g, + const char *qemu_param, const char *qemu_value) { if (qemu_param[0] != '-') { - error (g, "guestfs_config: parameter must begin with '-' character"); + error (g, _("guestfs_config: parameter must begin with '-' character")); return -1; } @@ -555,11 +691,10 @@ guestfs_config (guestfs_h *g, strcmp (qemu_param, "-initrd") == 0 || strcmp (qemu_param, "-nographic") == 0 || strcmp (qemu_param, "-serial") == 0 || - strcmp (qemu_param, "-vnc") == 0 || strcmp (qemu_param, "-full-screen") == 0 || strcmp (qemu_param, "-std-vga") == 0 || strcmp (qemu_param, "-vnc") == 0) { - error (g, "guestfs_config: parameter '%s' isn't allowed", qemu_param); + error (g, _("guestfs_config: parameter '%s' isn't allowed"), qemu_param); return -1; } @@ -573,13 +708,52 @@ guestfs_config (guestfs_h *g, } int -guestfs_add_drive (guestfs_h *g, const char *filename) +guestfs__add_drive (guestfs_h *g, const char *filename) { - int len = strlen (filename) + 64; + size_t len = strlen (filename) + 64; char buf[len]; if (strchr (filename, ',') != NULL) { - error (g, "filename cannot contain ',' (comma) character"); + error (g, _("filename cannot contain ',' (comma) character")); + return -1; + } + + /* cache=off improves reliability in the event of a host crash. + * + * However this option causes qemu to try to open the file with + * 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. + * + * This test also checks for the presence of the file, which + * is a documented semantic of this interface. + */ + int fd = open (filename, O_RDONLY|O_DIRECT); + if (fd >= 0) { + close (fd); + snprintf (buf, len, "file=%s,cache=off,if=" DRIVE_IF, filename); + } else { + fd = open (filename, O_RDONLY); + if (fd >= 0) { + close (fd); + snprintf (buf, len, "file=%s,if=" DRIVE_IF, filename); + } else { + perrorf (g, "%s", filename); + return -1; + } + } + + return guestfs__config (g, "-drive", buf); +} + +int +guestfs__add_drive_ro (guestfs_h *g, const char *filename) +{ + size_t len = strlen (filename) + 64; + char buf[len]; + + if (strchr (filename, ',') != NULL) { + error (g, _("filename cannot contain ',' (comma) character")); return -1; } @@ -588,16 +762,16 @@ guestfs_add_drive (guestfs_h *g, const char *filename) return -1; } - snprintf (buf, len, "file=%s", filename); + snprintf (buf, len, "file=%s,snapshot=on,if=%s", filename, DRIVE_IF); - return guestfs_config (g, "-drive", buf); + return guestfs__config (g, "-drive", buf); } int -guestfs_add_cdrom (guestfs_h *g, const char *filename) +guestfs__add_cdrom (guestfs_h *g, const char *filename) { if (strchr (filename, ',') != NULL) { - error (g, "filename cannot contain ',' (comma) character"); + error (g, _("filename cannot contain ',' (comma) character")); return -1; } @@ -606,35 +780,100 @@ guestfs_add_cdrom (guestfs_h *g, const char *filename) return -1; } - return guestfs_config (g, "-cdrom", filename); + return guestfs__config (g, "-cdrom", filename); +} + +/* Returns true iff file is contained in dir. */ +static int +dir_contains_file (const char *dir, const char *file) +{ + int dirlen = strlen (dir); + int filelen = strlen (file); + int len = dirlen+filelen+2; + char path[len]; + + snprintf (path, len, "%s/%s", dir, file); + return access (path, F_OK) == 0; +} + +/* Returns true iff every listed file is contained in 'dir'. */ +static int +dir_contains_files (const char *dir, ...) +{ + va_list args; + const char *file; + + va_start (args, dir); + while ((file = va_arg (args, const char *)) != NULL) { + if (!dir_contains_file (dir, file)) { + va_end (args); + return 0; + } + } + va_end (args); + return 1; } +static int build_supermin_appliance (guestfs_h *g, const char *path, char **kernel, char **initrd); +static int test_qemu (guestfs_h *g); +static int qemu_supports (guestfs_h *g, const char *option); +static void print_cmdline (guestfs_h *g); + +static const char *kernel_name = "vmlinuz." REPO "." host_cpu; +static const char *initrd_name = "initramfs." REPO "." host_cpu ".img"; +static const char *supermin_name = + "initramfs." REPO "." host_cpu ".supermin.img"; +static const char *supermin_hostfiles_name = + "initramfs." REPO "." host_cpu ".supermin.hostfiles"; + int -guestfs_launch (guestfs_h *g) +guestfs__launch (guestfs_h *g) { - static const char *dir_template = "/tmp/libguestfsXXXXXX"; - int r, i, len, pmore; + const char *tmpdir; + char dir_template[PATH_MAX]; + int r, pmore; + size_t len; int wfd[2], rfd[2]; int tries; - const char *kernel_name = "vmlinuz." REPO "." host_cpu; - const char *initrd_name = "initramfs." REPO "." host_cpu ".img"; char *path, *pelem, *pend; char *kernel = NULL, *initrd = NULL; char unixsock[256]; struct sockaddr_un addr; +#ifdef P_tmpdir + tmpdir = P_tmpdir; +#else + tmpdir = "/tmp"; +#endif + + tmpdir = getenv ("TMPDIR") ? : tmpdir; + snprintf (dir_template, sizeof dir_template, "%s/libguestfsXXXXXX", tmpdir); + /* Configured? */ if (!g->cmdline) { - error (g, "you must call guestfs_add_drive before guestfs_launch"); + error (g, _("you must call guestfs_add_drive before guestfs_launch")); return -1; } if (g->state != CONFIG) { - error (g, "qemu has already been launched"); + error (g, _("qemu has already been launched")); return -1; } - /* Search g->path for the kernel and initrd. */ + /* Make the temporary directory. */ + if (!g->tmpdir) { + g->tmpdir = safe_strdup (g, dir_template); + if (mkdtemp (g->tmpdir) == NULL) { + perrorf (g, _("%s: cannot create temporary directory"), dir_template); + goto cleanup0; + } + } + + /* First search g->path for the supermin appliance, and try to + * synthesize a kernel and initrd from that. If it fails, we + * try the path search again looking for a backup ordinary + * appliance. + */ pelem = path = safe_strdup (g, g->path); do { pend = strchrnul (pelem, ':'); @@ -642,32 +881,31 @@ guestfs_launch (guestfs_h *g) *pend = '\0'; len = pend - pelem; - /* Empty element or "." means cwd. */ + /* Empty element of "." means cwd. */ if (len == 0 || (len == 1 && *pelem == '.')) { if (g->verbose) - fprintf (stderr, - "looking for kernel and initrd in current directory\n"); - if (access (kernel_name, F_OK) == 0 && access (initrd_name, F_OK) == 0) { - kernel = safe_strdup (g, kernel_name); - initrd = safe_strdup (g, initrd_name); - break; + fprintf (stderr, + "looking for supermin appliance in current directory\n"); + if (dir_contains_files (".", + supermin_name, supermin_hostfiles_name, + "kmod.whitelist", NULL)) { + if (build_supermin_appliance (g, ".", &kernel, &initrd) == -1) + return -1; + break; } } - /* Look at /kernel etc. */ + /* Look at /supermin* etc. */ else { - kernel = safe_malloc (g, len + strlen (kernel_name) + 2); - initrd = safe_malloc (g, len + strlen (initrd_name) + 2); - sprintf (kernel, "%s/%s", pelem, kernel_name); - sprintf (initrd, "%s/%s", pelem, initrd_name); - if (g->verbose) - fprintf (stderr, "looking for %s and %s\n", kernel, initrd); - - if (access (kernel, F_OK) == 0 && access (initrd, F_OK) == 0) - break; - free (kernel); - free (initrd); - kernel = initrd = NULL; + fprintf (stderr, "looking for supermin appliance in %s\n", pelem); + + if (dir_contains_files (pelem, + supermin_name, supermin_hostfiles_name, + "kmod.whitelist", NULL)) { + if (build_supermin_appliance (g, pelem, &kernel, &initrd) == -1) + return -1; + break; + } } pelem = pend + 1; @@ -676,20 +914,56 @@ guestfs_launch (guestfs_h *g) free (path); if (kernel == NULL || initrd == NULL) { - error (g, "cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)", - kernel_name, initrd_name, g->path); - goto cleanup0; + /* Search g->path for the kernel and initrd. */ + pelem = path = safe_strdup (g, g->path); + do { + pend = strchrnul (pelem, ':'); + pmore = *pend == ':'; + *pend = '\0'; + len = pend - pelem; + + /* Empty element or "." means cwd. */ + if (len == 0 || (len == 1 && *pelem == '.')) { + if (g->verbose) + fprintf (stderr, + "looking for appliance in current directory\n"); + if (dir_contains_files (".", kernel_name, initrd_name, NULL)) { + kernel = safe_strdup (g, kernel_name); + initrd = safe_strdup (g, initrd_name); + break; + } + } + /* Look at /kernel etc. */ + else { + if (g->verbose) + fprintf (stderr, "looking for appliance in %s\n", pelem); + + if (dir_contains_files (pelem, kernel_name, initrd_name, NULL)) { + kernel = safe_malloc (g, len + strlen (kernel_name) + 2); + initrd = safe_malloc (g, len + strlen (initrd_name) + 2); + sprintf (kernel, "%s/%s", pelem, kernel_name); + sprintf (initrd, "%s/%s", pelem, initrd_name); + break; + } + } + + pelem = pend + 1; + } while (pmore); + + free (path); } - /* Make the temporary directory containing the socket. */ - if (!g->tmpdir) { - g->tmpdir = safe_strdup (g, dir_template); - if (mkdtemp (g->tmpdir) == NULL) { - perrorf (g, "%s: cannot create temporary directory", dir_template); - goto cleanup0; - } + if (kernel == NULL || initrd == NULL) { + error (g, _("cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)"), + kernel_name, initrd_name, g->path); + goto cleanup0; } + /* Get qemu help text and version. */ + if (test_qemu (g) == -1) + goto cleanup0; + + /* Make the vmchannel socket. */ snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir); unlink (unixsock); @@ -709,60 +983,119 @@ guestfs_launch (guestfs_h *g) } if (r == 0) { /* Child (qemu). */ - char vmchannel[256]; - char append[256]; + char buf[256]; + const char *vmchannel = NULL; /* Set up the full command line. Do this in the subprocess so we * don't need to worry about cleaning up. */ - g->cmdline[0] = (char *) QEMU; + g->cmdline[0] = g->qemu; + + snprintf (buf, sizeof buf, "%d", g->memsize); + add_cmdline (g, "-m"); + add_cmdline (g, buf); + + add_cmdline (g, "-no-reboot"); /* Force exit instead of reboot on panic */ + add_cmdline (g, "-nographic"); + add_cmdline (g, "-serial"); + add_cmdline (g, "stdio"); + + /* These options recommended by KVM developers to improve reliability. */ + if (qemu_supports (g, "-no-hpet")) + add_cmdline (g, "-no-hpet"); + + if (qemu_supports (g, "-rtc-td-hack")) + add_cmdline (g, "-rtc-td-hack"); + + if (qemu_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) { + /* New-style -net user,guestfwd=... syntax for guestfwd. See: + * + * http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=c92ef6a22d3c71538fcc48fb61ad353f7ba03b62 + * + * The original suggested format doesn't work, see: + * + * http://lists.gnu.org/archive/html/qemu-devel/2009-07/msg01654.html + * + * However Gerd Hoffman privately suggested to me using -chardev + * instead, which does work. + */ + snprintf (buf, sizeof buf, + "socket,id=guestfsvmc,path=%s,server,nowait", unixsock); + + add_cmdline (g, "-chardev"); + add_cmdline (g, buf); + + snprintf (buf, sizeof buf, + "user,vlan=0,net=10.0.2.0/8," + "guestfwd=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT + "-chardev:guestfsvmc"); + + add_cmdline (g, "-net"); + add_cmdline (g, buf); + } else { + /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel + * syntax, or if not then we'll get a quick failure. + */ + snprintf (buf, sizeof buf, + "channel," GUESTFWD_PORT ":unix:%s,server,nowait", unixsock); + + add_cmdline (g, "-net"); + add_cmdline (g, buf); + add_cmdline (g, "-net"); + add_cmdline (g, "user,vlan=0,net=10.0.2.0/8"); + } + add_cmdline (g, "-net"); + add_cmdline (g, "nic,model=" NET_IF ",vlan=0"); + vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT " "; - /* Construct the -net channel parameter for qemu. */ - snprintf (vmchannel, sizeof vmchannel, - "channel,%d:unix:%s,server,nowait", - VMCHANNEL_PORT, unixsock); +#define LINUX_CMDLINE \ + "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 */ \ + "acpi=off " /* we don't need ACPI, turn it off */ \ + "cgroup_disable=memory " /* saves us about 5 MB of RAM */ /* Linux kernel command line. */ - snprintf (append, sizeof append, - "console=ttyS0 guestfs=%s:%d%s", - VMCHANNEL_ADDR, VMCHANNEL_PORT, - g->verbose ? " guestfs_verbose=1" : ""); + snprintf (buf, sizeof buf, + LINUX_CMDLINE + "%s" /* (selinux) */ + "%s" /* (vmchannel) */ + "%s" /* (verbose) */ + "%s", /* (append) */ + g->selinux ? "selinux=1 enforcing=0 " : "selinux=0 ", + vmchannel ? vmchannel : "", + g->verbose ? "guestfs_verbose=1 " : "", + g->append ? g->append : ""); - add_cmdline (g, "-m"); - add_cmdline (g, "384"); /* XXX Choose best size. */ - add_cmdline (g, "-no-kqemu"); /* Avoids a warning. */ add_cmdline (g, "-kernel"); add_cmdline (g, (char *) kernel); add_cmdline (g, "-initrd"); add_cmdline (g, (char *) initrd); add_cmdline (g, "-append"); - add_cmdline (g, append); - add_cmdline (g, "-nographic"); - add_cmdline (g, "-serial"); - add_cmdline (g, "stdio"); - add_cmdline (g, "-net"); - add_cmdline (g, vmchannel); - add_cmdline (g, "-net"); - add_cmdline (g, "user,vlan=0"); - add_cmdline (g, "-net"); - add_cmdline (g, "nic,vlan=0"); + add_cmdline (g, buf); + + /* Finish off the command line. */ incr_cmdline_size (g); g->cmdline[g->cmdline_size-1] = NULL; - if (g->verbose) { - fprintf (stderr, "%s", QEMU); - for (i = 0; g->cmdline[i]; ++i) - fprintf (stderr, " %s", g->cmdline[i]); - fprintf (stderr, "\n"); - } + if (g->verbose) + print_cmdline (g); /* Set up stdin, stdout. */ close (0); close (1); close (wfd[1]); close (rfd[0]); - dup (wfd[0]); - dup (rfd[1]); + + if (dup (wfd[0]) == -1) { + dup_failed: + perror ("dup failed"); + _exit (1); + } + if (dup (rfd[1]) == -1) + goto dup_failed; + close (wfd[0]); close (rfd[1]); @@ -773,14 +1106,53 @@ guestfs_launch (guestfs_h *g) setpgid (0, 0); #endif - execv (QEMU, g->cmdline); /* Run qemu. */ - perror (QEMU); + execv (g->qemu, g->cmdline); /* Run qemu. */ + perror (g->qemu); _exit (1); } /* Parent (library). */ g->pid = r; + free (kernel); + kernel = NULL; + free (initrd); + initrd = NULL; + + /* Fork the recovery process off which will kill qemu if the parent + * process fails to do so (eg. if the parent segfaults). + */ + r = fork (); + if (r == 0) { + pid_t qemu_pid = g->pid; + pid_t parent_pid = getppid (); + + /* 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 + */ + + /* Loop around waiting for one or both of the other processes to + * disappear. It's fair to say this is very hairy. The PIDs that + * we are looking at might be reused by another process. We are + * effectively polling. Is the cure worse than the disease? + */ + for (;;) { + if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */ + _exit (0); + if (kill (parent_pid, 0) == -1) { + /* Parent's gone away, qemu still around, so kill qemu. */ + kill (qemu_pid, 9); + _exit (0); + } + sleep (2); + } + } + + /* Don't worry, if the fork failed, this will be -1. The recovery + * process isn't essential. + */ + g->recoverypid = r; + /* Start the clock ... */ time (&g->start_t); @@ -822,10 +1194,9 @@ guestfs_launch (guestfs_h *g) addr.sun_path[UNIX_PATH_MAX-1] = '\0'; tries = 100; + /* Always sleep at least once to give qemu a small chance to start up. */ + usleep (10000); while (tries > 0) { - /* Always sleep at least once to give qemu a small chance to start up. */ - usleep (10000); - r = connect (g->sock, (struct sockaddr *) &addr, sizeof addr); if ((r == -1 && errno == EINPROGRESS) || r == 0) goto connected; @@ -833,43 +1204,42 @@ guestfs_launch (guestfs_h *g) if (errno != ENOENT) perrorf (g, "connect"); tries--; + usleep (100000); } - error (g, "failed to connect to vmchannel socket"); + error (g, _("failed to connect to vmchannel socket")); goto cleanup2; connected: - /* Watch the file descriptors. */ - free (g->msg_in); - g->msg_in = NULL; - g->msg_in_size = g->msg_in_allocated = 0; + g->state = LAUNCHING; - free (g->msg_out); - g->msg_out = NULL; - g->msg_out_size = 0; - g->msg_out_pos = 0; + /* Wait for qemu to start and to connect back to us via vmchannel and + * send the GUESTFS_LAUNCH_FLAG message. + */ + uint32_t size; + void *buf = NULL; + r = recv_from_daemon (g, &size, &buf); + free (buf); + + if (r == -1) return -1; - g->stdout_watch = - g->main_loop->add_handle (g->main_loop, g, g->fd[1], - GUESTFS_HANDLE_READABLE, - stdout_event, NULL); - if (g->stdout_watch == -1) { - error (g, "could not watch qemu stdout"); - goto cleanup3; + if (size != GUESTFS_LAUNCH_FLAG) { + error (g, _("guestfs_launch failed, see earlier error messages")); + goto cleanup2; } - if (guestfs__switch_to_receiving (g) == -1) - goto cleanup3; + /* This is possible in some really strange situations, such as + * guestfsd starts up OK but then qemu immediately exits. Check for + * it because the caller is probably expecting to be able to send + * commands after this function returns. + */ + if (g->state != READY) { + error (g, _("qemu launched and contacted daemon, but state != READY")); + goto cleanup2; + } - g->state = LAUNCHING; return 0; - cleanup3: - if (g->stdout_watch >= 0) - g->main_loop->remove_handle (g->main_loop, g, g->stdout_watch); - if (g->sock_watch >= 0) - g->main_loop->remove_handle (g->main_loop, g, g->sock_watch); - cleanup2: close (g->sock); @@ -877,14 +1247,15 @@ guestfs_launch (guestfs_h *g) close (wfd[1]); close (rfd[0]); kill (g->pid, 9); + if (g->recoverypid > 0) kill (g->recoverypid, 9); waitpid (g->pid, NULL, 0); + if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0); g->fd[0] = -1; g->fd[1] = -1; g->sock = -1; g->pid = 0; + g->recoverypid = 0; g->start_t = 0; - g->stdout_watch = -1; - g->sock_watch = -1; cleanup0: free (kernel); @@ -892,53 +1263,162 @@ guestfs_launch (guestfs_h *g) return -1; } +/* This function is used to print the qemu command line before it gets + * executed, when in verbose mode. + */ static void -finish_wait_ready (guestfs_h *g, void *vp) +print_cmdline (guestfs_h *g) { - if (g->verbose) - fprintf (stderr, "finish_wait_ready called, %p, vp = %p\n", g, vp); + int i = 0; + int needs_quote; - *((int *)vp) = 1; - g->main_loop->main_loop_quit (g->main_loop, g); -} + while (g->cmdline[i]) { + if (g->cmdline[i][0] == '-') /* -option starts a new line */ + fprintf (stderr, " \\\n "); -int -guestfs_wait_ready (guestfs_h *g) -{ - int finished = 0, r; + if (i > 0) fputc (' ', stderr); - if (g->state == READY) return 0; + /* Does it need shell quoting? This only deals with simple cases. */ + needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]); - if (g->state == BUSY) { - error (g, "qemu has finished launching already"); - return -1; + if (needs_quote) fputc ('\'', stderr); + fprintf (stderr, "%s", g->cmdline[i]); + if (needs_quote) fputc ('\'', stderr); + i++; } - if (g->state != LAUNCHING) { - error (g, "qemu has not been launched yet"); + fputc ('\n', stderr); +} + +/* This function does the hard work of building the supermin appliance + * on the fly. 'path' is the directory containing the control files. + * 'kernel' and 'initrd' are where we will return the names of the + * kernel and initrd (only initrd is built). The work is done by + * an external script. We just tell it where to put the result. + */ +static int +build_supermin_appliance (guestfs_h *g, const char *path, + char **kernel, char **initrd) +{ + char cmd[4096]; + int r, len; + + len = strlen (g->tmpdir); + *kernel = safe_malloc (g, len + 8); + snprintf (*kernel, len+8, "%s/kernel", g->tmpdir); + *initrd = safe_malloc (g, len + 8); + snprintf (*initrd, len+8, "%s/initrd", g->tmpdir); + + snprintf (cmd, sizeof cmd, + "PATH='%s':$PATH " + "libguestfs-supermin-helper '%s' %s %s", + path, + path, *kernel, *initrd); + + r = system (cmd); + if (r == -1 || WEXITSTATUS(r) != 0) { + error (g, _("external command failed: %s"), cmd); + free (*kernel); + free (*initrd); + *kernel = *initrd = NULL; return -1; } - g->launch_done_cb = finish_wait_ready; - g->launch_done_cb_data = &finished; - r = g->main_loop->main_loop_run (g->main_loop, g); - g->launch_done_cb = NULL; - g->launch_done_cb_data = NULL; + return 0; +} - if (r == -1) return -1; +static int read_all (guestfs_h *g, FILE *fp, char **ret); - if (finished != 1) { - error (g, "guestfs_wait_ready failed, see earlier error messages"); - return -1; - } +/* Test qemu binary (or wrapper) runs, and do 'qemu -help' and + * 'qemu -version' so we know what options this qemu supports and + * the version. + */ +static int +test_qemu (guestfs_h *g) +{ + char cmd[1024]; + FILE *fp; - /* This is possible in some really strange situations, such as - * guestfsd starts up OK but then qemu immediately exits. Check for - * it because the caller is probably expecting to be able to send - * commands after this function returns. + free (g->qemu_help); + free (g->qemu_version); + g->qemu_help = NULL; + g->qemu_version = NULL; + + snprintf (cmd, sizeof cmd, "'%s' -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 (g->state != READY) { - error (g, "qemu launched and contacted daemon, but state != READY"); + 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); + return -1; + } + + if (read_all (g, fp, &g->qemu_help) == -1) + goto error; + + if (pclose (fp) == -1) + goto error; + + snprintf (cmd, sizeof cmd, "'%s' -version 2>/dev/null", g->qemu); + + fp = popen (cmd, "r"); + if (fp) { + /* Intentionally ignore errors. */ + read_all (g, fp, &g->qemu_version); + pclose (fp); + } + + return 0; +} + +static int +read_all (guestfs_h *g, FILE *fp, char **ret) +{ + int r, n = 0; + char *p; + + again: + if (feof (fp)) { + *ret = safe_realloc (g, *ret, n + 1); + (*ret)[n] = '\0'; + return n; + } + + *ret = safe_realloc (g, *ret, n + BUFSIZ); + p = &(*ret)[n]; + r = fread (p, 1, BUFSIZ, fp); + if (ferror (fp)) { + perrorf (g, "read"); + return -1; + } + n += r; + goto again; +} + +/* Test if option is supported by qemu command line (just by grepping + * the help text). + */ +static int +qemu_supports (guestfs_h *g, const char *option) +{ + return g->qemu_help && strstr (g->qemu_help, option) != NULL; +} + +/* You had to call this function after launch in versions <= 1.0.70, + * but it is now a no-op. + */ +int +guestfs__wait_ready (guestfs_h *g) +{ + if (g->state != READY) { + error (g, _("qemu has not been launched yet")); return -1; } @@ -946,10 +1426,10 @@ guestfs_wait_ready (guestfs_h *g) } int -guestfs_kill_subprocess (guestfs_h *g) +guestfs__kill_subprocess (guestfs_h *g) { if (g->state == CONFIG) { - error (g, "no subprocess to kill"); + error (g, _("no subprocess to kill")); return -1; } @@ -957,493 +1437,525 @@ guestfs_kill_subprocess (guestfs_h *g) fprintf (stderr, "sending SIGTERM to process %d\n", g->pid); kill (g->pid, SIGTERM); + if (g->recoverypid > 0) kill (g->recoverypid, 9); return 0; } /* Access current state. */ int -guestfs_is_config (guestfs_h *g) +guestfs__is_config (guestfs_h *g) { return g->state == CONFIG; } int -guestfs_is_launching (guestfs_h *g) +guestfs__is_launching (guestfs_h *g) { return g->state == LAUNCHING; } int -guestfs_is_ready (guestfs_h *g) +guestfs__is_ready (guestfs_h *g) { return g->state == READY; } int -guestfs_is_busy (guestfs_h *g) +guestfs__is_busy (guestfs_h *g) { return g->state == BUSY; } int -guestfs_get_state (guestfs_h *g) +guestfs__get_state (guestfs_h *g) { return g->state; } -int -guestfs_set_ready (guestfs_h *g) +void +guestfs_set_log_message_callback (guestfs_h *g, + guestfs_log_message_cb cb, void *opaque) { - if (g->state != BUSY) { - error (g, "guestfs_set_ready: called when in state %d != BUSY", g->state); - return -1; - } - g->state = READY; - return 0; + g->log_message_cb = cb; + g->log_message_cb_data = opaque; } +void +guestfs_set_subprocess_quit_callback (guestfs_h *g, + guestfs_subprocess_quit_cb cb, void *opaque) +{ + g->subprocess_quit_cb = cb; + g->subprocess_quit_cb_data = opaque; +} + +void +guestfs_set_launch_done_callback (guestfs_h *g, + guestfs_launch_done_cb cb, void *opaque) +{ + g->launch_done_cb = cb; + g->launch_done_cb_data = opaque; +} + +/*----------------------------------------------------------------------*/ + +/* This is the code used to send and receive RPC messages and (for + * certain types of message) to perform file transfers. This code is + * driven from the generated actions (src/guestfs-actions.c). There + * are five different cases to consider: + * + * (1) A non-daemon function. There is no RPC involved at all, it's + * all handled inside the library. + * + * (2) A simple RPC (eg. "mount"). We write the request, then read + * the reply. The sequence of calls is: + * + * guestfs___set_busy + * guestfs___send + * guestfs___recv + * guestfs___end_busy + * + * (3) An RPC with FileOut parameters (eg. "upload"). We write the + * request, then write the file(s), then read the reply. The sequence + * of calls is: + * + * guestfs___set_busy + * guestfs___send + * guestfs___send_file (possibly multiple times) + * guestfs___recv + * guestfs___end_busy + * + * (4) An RPC with FileIn parameters (eg. "download"). We write the + * request, then read the reply, then read the file(s). The sequence + * of calls is: + * + * guestfs___set_busy + * guestfs___send + * guestfs___recv + * guestfs___recv_file (possibly multiple times) + * guestfs___end_busy + * + * (5) Both FileOut and FileIn parameters. There are no calls like + * this in the current API, but they would be implemented as a + * combination of cases (3) and (4). + * + * During all writes and reads, we also select(2) on qemu stdout + * looking for messages (guestfsd stderr and guest kernel dmesg), and + * anything received is passed up through the log_message_cb. This is + * also the reason why all the sockets are non-blocking. We also have + * to check for EOF (qemu died). All of this is handled by the + * functions send_to_daemon and recv_from_daemon. + */ + int -guestfs_set_busy (guestfs_h *g) +guestfs___set_busy (guestfs_h *g) { if (g->state != READY) { - error (g, "guestfs_set_busy: called when in state %d != READY", g->state); + error (g, _("guestfs_set_busy: called when in state %d != READY"), + g->state); return -1; } g->state = BUSY; return 0; } -/* Structure-freeing functions. These rely on the fact that the - * structure format is identical to the XDR format. See note in - * generator.ml. - */ -void -guestfs_free_int_bool (struct guestfs_int_bool *x) +int +guestfs___end_busy (guestfs_h *g) { - free (x); -} + switch (g->state) + { + case BUSY: + g->state = READY; + break; + case CONFIG: + case READY: + break; -void -guestfs_free_lvm_pv_list (struct guestfs_lvm_pv_list *x) -{ - xdr_free ((xdrproc_t) xdr_guestfs_lvm_int_pv_list, (char *) x); - free (x); + case LAUNCHING: + case NO_HANDLE: + default: + error (g, _("guestfs_end_busy: called when in state %d"), g->state); + return -1; + } + return 0; } -void -guestfs_free_lvm_vg_list (struct guestfs_lvm_vg_list *x) +/* This is called if we detect EOF, ie. qemu died. */ +static void +child_cleanup (guestfs_h *g) { - xdr_free ((xdrproc_t) xdr_guestfs_lvm_int_vg_list, (char *) x); - free (x); -} + if (g->verbose) + fprintf (stderr, "child_cleanup: %p: child process died\n", g); -void -guestfs_free_lvm_lv_list (struct guestfs_lvm_lv_list *x) -{ - xdr_free ((xdrproc_t) xdr_guestfs_lvm_int_lv_list, (char *) x); - free (x); + /*kill (g->pid, SIGTERM);*/ + if (g->recoverypid > 0) kill (g->recoverypid, 9); + waitpid (g->pid, NULL, 0); + if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0); + close (g->fd[0]); + close (g->fd[1]); + close (g->sock); + g->fd[0] = -1; + g->fd[1] = -1; + g->sock = -1; + g->pid = 0; + g->recoverypid = 0; + g->start_t = 0; + g->state = CONFIG; + if (g->subprocess_quit_cb) + g->subprocess_quit_cb (g, g->subprocess_quit_cb_data); } -/* This function is called whenever qemu prints something on stdout. - * Qemu's stdout is also connected to the guest's serial console, so - * we see kernel messages here too. - */ -static void -stdout_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data, - int watch, int fd, int events) +static int +read_log_message_or_eof (guestfs_h *g, int fd) { - char buf[4096]; + char buf[BUFSIZ]; int n; #if 0 if (g->verbose) fprintf (stderr, - "stdout_event: %p g->state = %d, fd = %d, events = 0x%x\n", - g, g->state, fd, events); + "read_log_message_or_eof: %p g->state = %d, fd = %d\n", + g, g->state, fd); #endif - if (g->fd[1] != fd) { - error (g, "stdout_event: internal error: %d != %d", g->fd[1], fd); - return; - } - + /* 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. + */ n = read (fd, buf, sizeof buf); if (n == 0) { /* Hopefully this indicates the qemu child process has died. */ - if (g->verbose) - fprintf (stderr, "stdout_event: %p: child process died\n", g); - /*kill (g->pid, SIGTERM);*/ - waitpid (g->pid, NULL, 0); - if (g->stdout_watch >= 0) - g->main_loop->remove_handle (g->main_loop, g, g->stdout_watch); - if (g->sock_watch >= 0) - g->main_loop->remove_handle (g->main_loop, g, g->sock_watch); - close (g->fd[0]); - close (g->fd[1]); - close (g->sock); - g->fd[0] = -1; - g->fd[1] = -1; - g->sock = -1; - g->pid = 0; - g->start_t = 0; - g->stdout_watch = -1; - g->sock_watch = -1; - g->state = CONFIG; - if (g->subprocess_quit_cb) - g->subprocess_quit_cb (g, g->subprocess_quit_cb_data); - return; + child_cleanup (g); + return -1; } if (n == -1) { - if (errno != EAGAIN) - perrorf (g, "read"); - return; + if (errno == EINTR || errno == EAGAIN) + return 0; + + perrorf (g, "read"); + return -1; } /* In verbose mode, copy all log messages to stderr. */ if (g->verbose) - write (2, buf, n); + ignore_value (write (STDERR_FILENO, buf, n)); /* It's an actual log message, send it upwards if anyone is listening. */ if (g->log_message_cb) g->log_message_cb (g, g->log_message_cb_data, buf, n); + + return 0; } -/* The function is called whenever we can read something on the - * guestfsd (daemon inside the guest) communication socket. - */ -static void -sock_read_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data, - int watch, int fd, int events) +static int +check_for_daemon_cancellation_or_eof (guestfs_h *g, int fd) { - XDR xdr; - unsigned len; + char buf[4]; int n; + uint32_t flag; + XDR xdr; if (g->verbose) fprintf (stderr, - "sock_read_event: %p g->state = %d, fd = %d, events = 0x%x\n", - g, g->state, fd, events); + "check_for_daemon_cancellation_or_eof: %p g->state = %d, fd = %d\n", + g, g->state, fd); - if (g->sock != fd) { - error (g, "sock_read_event: internal error: %d != %d", g->sock, fd); - return; - } - - if (g->msg_in_size <= g->msg_in_allocated) { - g->msg_in_allocated += 4096; - g->msg_in = safe_realloc (g, g->msg_in, g->msg_in_allocated); + n = read (fd, buf, 4); + if (n == 0) { + /* Hopefully this indicates the qemu child process has died. */ + child_cleanup (g); + return -1; } - n = read (g->sock, g->msg_in + g->msg_in_size, - g->msg_in_allocated - g->msg_in_size); - if (n == 0) - /* Disconnected? Ignore it because stdout_watch will get called - * and will do the cleanup. - */ - return; if (n == -1) { - if (errno != EAGAIN) - perrorf (g, "read"); - return; - } - - g->msg_in_size += n; - - /* Have we got enough of a message to be able to process it yet? */ - again: - if (g->msg_in_size < 4) return; - - xdrmem_create (&xdr, g->msg_in, g->msg_in_size, XDR_DECODE); - if (!xdr_uint32_t (&xdr, &len)) { - error (g, "can't decode length word"); - goto cleanup; - } - - /* Length is normally the length of the message, but when guestfsd - * starts up it sends a "magic" value (longer than any possible - * message). Check for this. - */ - if (len == GUESTFS_LAUNCH_FLAG) { - if (g->state != LAUNCHING) - error (g, "received magic signature from guestfsd, but in state %d", - g->state); - else if (g->msg_in_size != 4) - error (g, "received magic signature from guestfsd, but msg size is %d", - g->msg_in_size); - else { - g->state = READY; - if (g->launch_done_cb) - g->launch_done_cb (g, g->launch_done_cb_data); - } + if (errno == EINTR || errno == EAGAIN) + return 0; - goto cleanup; - } - - /* This can happen if a cancellation happens right at the end - * of us sending a FileIn parameter to the daemon. Discard. The - * daemon should send us an error message next. - */ - if (len == GUESTFS_CANCEL_FLAG) { - g->msg_in_size -= 4; - memmove (g->msg_in, g->msg_in+4, g->msg_in_size); - goto again; - } - - /* If this happens, it's pretty bad and we've probably lost - * synchronization. - */ - if (len > GUESTFS_MESSAGE_MAX) { - error (g, "message length (%u) > maximum possible size (%d)", - len, GUESTFS_MESSAGE_MAX); - goto cleanup; + perrorf (g, "read"); + return -1; } - if (g->msg_in_size-4 < len) return; /* Need more of this message. */ - - /* Got the full message, begin processing it. */ - if (g->verbose) { - int i, j; + xdrmem_create (&xdr, buf, 4, XDR_DECODE); + xdr_uint32_t (&xdr, &flag); + xdr_destroy (&xdr); - for (i = 0; i < g->msg_in_size; i += 16) { - printf ("%04x: ", i); - for (j = i; j < MIN (i+16, g->msg_in_size); ++j) - printf ("%02x ", (unsigned char) g->msg_in[j]); - for (; j < i+16; ++j) - printf (" "); - printf ("|"); - for (j = i; j < MIN (i+16, g->msg_in_size); ++j) - if (isprint (g->msg_in[j])) - printf ("%c", g->msg_in[j]); - else - printf ("."); - for (; j < i+16; ++j) - printf (" "); - printf ("|\n"); - } + if (flag != GUESTFS_CANCEL_FLAG) { + error (g, _("check_for_daemon_cancellation_or_eof: read 0x%x from daemon, expected 0x%x\n"), + flag, GUESTFS_CANCEL_FLAG); + return -1; } - /* Not in the expected state. */ - if (g->state != BUSY) - error (g, "state %d != BUSY", g->state); - - /* Push the message up to the higher layer. */ - if (g->reply_cb) - g->reply_cb (g, g->reply_cb_data, &xdr); - - g->msg_in_size -= len + 4; - memmove (g->msg_in, g->msg_in+len+4, g->msg_in_size); - if (g->msg_in_size > 0) goto again; - - cleanup: - /* Free the message buffer if it's grown excessively large. */ - if (g->msg_in_allocated > 65536) { - free (g->msg_in); - g->msg_in = NULL; - g->msg_in_size = g->msg_in_allocated = 0; - } else - g->msg_in_size = 0; - - xdr_destroy (&xdr); + return -2; } -/* The function is called whenever we can write something on the - * guestfsd (daemon inside the guest) communication socket. +/* This writes the whole N bytes of BUF to the daemon socket. + * + * If the whole write is successful, it returns 0. + * If there was an error, it returns -1. + * If the daemon sent a cancellation message, it returns -2. + * + * It also checks qemu stdout for log messages and passes those up + * through log_message_cb. + * + * It also checks for EOF (qemu died) and passes that up through the + * child_cleanup function above. */ -static void -sock_write_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data, - int watch, int fd, int events) +static int +send_to_daemon (guestfs_h *g, const void *v_buf, size_t n) { - int n; + const char *buf = v_buf; + fd_set rset, rset2; + fd_set wset, wset2; if (g->verbose) fprintf (stderr, - "sock_write_event: %p g->state = %d, fd = %d, events = 0x%x\n", - g, g->state, fd, events); - - if (g->sock != fd) { - error (g, "sock_write_event: internal error: %d != %d", g->sock, fd); - return; - } - - if (g->state != BUSY) { - error (g, "sock_write_event: state %d != BUSY", g->state); - return; - } - - if (g->verbose) - fprintf (stderr, "sock_write_event: writing %d bytes ...\n", - g->msg_out_size - g->msg_out_pos); + "send_to_daemon: %p g->state = %d, n = %zu\n", g, g->state, n); - n = write (g->sock, g->msg_out + g->msg_out_pos, - g->msg_out_size - g->msg_out_pos); - if (n == -1) { - if (errno != EAGAIN) - perrorf (g, "write"); - return; - } - - if (g->verbose) - fprintf (stderr, "sock_write_event: wrote %d bytes\n", n); + FD_ZERO (&rset); + FD_ZERO (&wset); - g->msg_out_pos += n; + FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */ + FD_SET (g->sock, &rset); /* Read socket for cancellation & EOF. */ + FD_SET (g->sock, &wset); /* Write to socket to send the data. */ - /* More to write? */ - if (g->msg_out_pos < g->msg_out_size) - return; + int max_fd = g->sock > g->fd[1] ? g->sock : g->fd[1]; - if (g->verbose) - fprintf (stderr, "sock_write_event: done writing, calling send_cb\n"); - - free (g->msg_out); - g->msg_out = NULL; - g->msg_out_pos = g->msg_out_size = 0; + while (n > 0) { + rset2 = rset; + wset2 = wset; + int r = select (max_fd+1, &rset2, &wset2, NULL, NULL); + if (r == -1) { + if (errno == EINTR || errno == EAGAIN) + continue; + perrorf (g, "select"); + return -1; + } - /* Done writing, call the higher layer. */ - if (g->send_cb) - g->send_cb (g, g->send_cb_data); -} + if (FD_ISSET (g->fd[1], &rset2)) { + if (read_log_message_or_eof (g, g->fd[1]) == -1) + return -1; + } + if (FD_ISSET (g->sock, &rset2)) { + r = check_for_daemon_cancellation_or_eof (g, g->sock); + if (r < 0) + return r; + } + if (FD_ISSET (g->sock, &wset2)) { + r = write (g->sock, buf, n); + if (r == -1) { + if (errno == EINTR || errno == EAGAIN) + continue; + perrorf (g, "write"); + if (errno == EPIPE) /* Disconnected from guest (RHBZ#508713). */ + child_cleanup (g); + return -1; + } + buf += r; + n -= r; + } + } -void -guestfs_set_send_callback (guestfs_h *g, - guestfs_send_cb cb, void *opaque) -{ - g->send_cb = cb; - g->send_cb_data = opaque; + return 0; } -void -guestfs_set_reply_callback (guestfs_h *g, - guestfs_reply_cb cb, void *opaque) +/* This reads a single message, file chunk, launch flag or + * cancellation flag from the daemon. If something was read, it + * returns 0, otherwise -1. + * + * Both size_rtn and buf_rtn must be passed by the caller as non-NULL. + * + * *size_rtn returns the size of the returned message or it may be + * GUESTFS_LAUNCH_FLAG or GUESTFS_CANCEL_FLAG. + * + * *buf_rtn is returned containing the message (if any) or will be set + * to NULL. *buf_rtn must be freed by the caller. + * + * It also checks qemu stdout for log messages and passes those up + * through log_message_cb. + * + * It also checks for EOF (qemu died) and passes that up through the + * child_cleanup function above. + */ +static int +recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn) { - g->reply_cb = cb; - g->reply_cb_data = opaque; -} + fd_set rset, rset2; -void -guestfs_set_log_message_callback (guestfs_h *g, - guestfs_log_message_cb cb, void *opaque) -{ - g->log_message_cb = cb; - g->log_message_cb_data = opaque; -} + if (g->verbose) + fprintf (stderr, + "recv_from_daemon: %p g->state = %d, size_rtn = %p, buf_rtn = %p\n", + g, g->state, size_rtn, buf_rtn); -void -guestfs_set_subprocess_quit_callback (guestfs_h *g, - guestfs_subprocess_quit_cb cb, void *opaque) -{ - g->subprocess_quit_cb = cb; - g->subprocess_quit_cb_data = opaque; -} + FD_ZERO (&rset); -void -guestfs_set_launch_done_callback (guestfs_h *g, - guestfs_launch_done_cb cb, void *opaque) -{ - g->launch_done_cb = cb; - g->launch_done_cb_data = opaque; -} + FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */ + FD_SET (g->sock, &rset); /* Read socket for data & EOF. */ -/* Access to the handle's main loop and the default main loop. */ -void -guestfs_set_main_loop (guestfs_h *g, guestfs_main_loop *main_loop) -{ - g->main_loop = main_loop; -} + int max_fd = g->sock > g->fd[1] ? g->sock : g->fd[1]; -guestfs_main_loop * -guestfs_get_main_loop (guestfs_h *g) -{ - return g->main_loop; -} + *size_rtn = 0; + *buf_rtn = NULL; -guestfs_main_loop * -guestfs_get_default_main_loop (void) -{ - return (guestfs_main_loop *) &default_main_loop; -} + char lenbuf[4]; + /* nr is the size of the message, but we prime it as -4 because we + * have to read the message length word first. + */ + ssize_t nr = -4; -/* Change the daemon socket handler so that we are now writing. - * This sets the handle to sock_write_event. - */ -int -guestfs__switch_to_sending (guestfs_h *g) -{ - if (g->sock_watch >= 0) { - if (g->main_loop->remove_handle (g->main_loop, g, g->sock_watch) == -1) { - error (g, "remove_handle failed"); - g->sock_watch = -1; + while (nr < (ssize_t) *size_rtn) { + rset2 = rset; + int r = select (max_fd+1, &rset2, NULL, NULL, NULL); + if (r == -1) { + if (errno == EINTR || errno == EAGAIN) + continue; + perrorf (g, "select"); + free (*buf_rtn); + *buf_rtn = NULL; return -1; } - } - g->sock_watch = - g->main_loop->add_handle (g->main_loop, g, g->sock, - GUESTFS_HANDLE_WRITABLE, - sock_write_event, NULL); - if (g->sock_watch == -1) { - error (g, "add_handle failed"); - return -1; - } - - return 0; -} + if (FD_ISSET (g->fd[1], &rset2)) { + if (read_log_message_or_eof (g, g->fd[1]) == -1) { + free (*buf_rtn); + *buf_rtn = NULL; + return -1; + } + } + if (FD_ISSET (g->sock, &rset2)) { + if (nr < 0) { /* Have we read the message length word yet? */ + r = read (g->sock, lenbuf+nr+4, -nr); + if (r == -1) { + if (errno == EINTR || errno == EAGAIN) + continue; + int err = errno; + perrorf (g, "read"); + /* Under some circumstances we see "Connection reset by peer" + * here when the child dies suddenly. Catch this and call + * the cleanup function, same as for EOF. + */ + if (err == ECONNRESET) + child_cleanup (g); + return -1; + } + if (r == 0) { + error (g, _("unexpected end of file when reading from daemon")); + child_cleanup (g); + return -1; + } + nr += r; + + if (nr < 0) /* Still not got the whole length word. */ + continue; + + XDR xdr; + xdrmem_create (&xdr, lenbuf, 4, XDR_DECODE); + xdr_uint32_t (&xdr, size_rtn); + xdr_destroy (&xdr); + + if (*size_rtn == GUESTFS_LAUNCH_FLAG) { + if (g->state != LAUNCHING) + error (g, _("received magic signature from guestfsd, but in state %d"), + g->state); + else { + g->state = READY; + if (g->launch_done_cb) + g->launch_done_cb (g, g->launch_done_cb_data); + } + return 0; + } + else if (*size_rtn == GUESTFS_CANCEL_FLAG) + return 0; + /* If this happens, it's pretty bad and we've probably lost + * synchronization. + */ + else if (*size_rtn > GUESTFS_MESSAGE_MAX) { + error (g, _("message length (%u) > maximum possible size (%d)"), + (unsigned) *size_rtn, GUESTFS_MESSAGE_MAX); + return -1; + } + + /* Allocate the complete buffer, size now known. */ + *buf_rtn = safe_malloc (g, *size_rtn); + /*FALLTHROUGH*/ + } -int -guestfs__switch_to_receiving (guestfs_h *g) -{ - if (g->sock_watch >= 0) { - if (g->main_loop->remove_handle (g->main_loop, g, g->sock_watch) == -1) { - error (g, "remove_handle failed"); - g->sock_watch = -1; - return -1; + size_t sizetoread = *size_rtn - nr; + if (sizetoread > BUFSIZ) sizetoread = BUFSIZ; + + r = read (g->sock, (char *) (*buf_rtn) + nr, sizetoread); + if (r == -1) { + if (errno == EINTR || errno == EAGAIN) + continue; + perrorf (g, "read"); + free (*buf_rtn); + *buf_rtn = NULL; + return -1; + } + if (r == 0) { + error (g, _("unexpected end of file when reading from daemon")); + child_cleanup (g); + free (*buf_rtn); + *buf_rtn = NULL; + return -1; + } + nr += r; } } - g->sock_watch = - g->main_loop->add_handle (g->main_loop, g, g->sock, - GUESTFS_HANDLE_READABLE, - sock_read_event, NULL); - if (g->sock_watch == -1) { - error (g, "add_handle failed"); - return -1; + /* Got the full message, caller can start processing it. */ +#ifdef ENABLE_PACKET_DUMP + if (g->verbose) { + ssize_t i, j; + + for (i = 0; i < nr; i += 16) { + printf ("%04zx: ", i); + for (j = i; j < MIN (i+16, nr); ++j) + printf ("%02x ", (*(unsigned char **)buf_rtn)[j]); + for (; j < i+16; ++j) + printf (" "); + printf ("|"); + for (j = i; j < MIN (i+16, nr); ++j) + if (isprint ((*(char **)buf_rtn)[j])) + printf ("%c", (*(char **)buf_rtn)[j]); + else + printf ("."); + for (; j < i+16; ++j) + printf (" "); + printf ("|\n"); + } } +#endif return 0; } -/* Dispatch a call (len + header + args) to the remote daemon, - * synchronously (ie. using the guest's main loop to wait until - * it has been sent). Returns -1 for error, or the serial - * number of the message. - */ -static void -send_cb (guestfs_h *g, void *data) -{ - guestfs_main_loop *ml = guestfs_get_main_loop (g); - - *((int *)data) = 1; - ml->main_loop_quit (ml, g); -} - int -guestfs__send_sync (guestfs_h *g, int proc_nr, - xdrproc_t xdrp, char *args) +guestfs___send (guestfs_h *g, int proc_nr, xdrproc_t xdrp, char *args) { - char buffer[GUESTFS_MESSAGE_MAX]; struct guestfs_message_header hdr; XDR xdr; - unsigned len; + u_int32_t len; int serial = g->msg_next_serial++; - int sent; - guestfs_main_loop *ml = guestfs_get_main_loop (g); + int r; + char *msg_out; + size_t msg_out_size; if (g->state != BUSY) { - error (g, "guestfs__send_sync: state %d != BUSY", g->state); + error (g, _("guestfs___send: state %d != BUSY"), g->state); return -1; } + /* We have to allocate this message buffer on the heap because + * it is quite large (although will be mostly unused). We + * can't allocate it on the stack because in some environments + * we have quite limited stack space available, notably when + * running in the JVM. + */ + msg_out = safe_malloc (g, GUESTFS_MESSAGE_MAX + 4); + xdrmem_create (&xdr, msg_out + 4, GUESTFS_MESSAGE_MAX, XDR_ENCODE); + /* Serialize the header. */ hdr.prog = GUESTFS_PROGRAM; hdr.vers = GUESTFS_PROTOCOL_VERSION; @@ -1452,10 +1964,9 @@ guestfs__send_sync (guestfs_h *g, int proc_nr, hdr.serial = serial; hdr.status = GUESTFS_STATUS_OK; - xdrmem_create (&xdr, buffer, sizeof buffer, XDR_ENCODE); if (!xdr_guestfs_message_header (&xdr, &hdr)) { - error (g, "xdr_guestfs_message_header failed"); - return -1; + error (g, _("xdr_guestfs_message_header failed")); + goto cleanup1; } /* Serialize the args. If any, because some message types @@ -1463,60 +1974,52 @@ guestfs__send_sync (guestfs_h *g, int proc_nr, */ if (xdrp) { if (!(*xdrp) (&xdr, args)) { - error (g, "dispatch failed to marshal args"); - return -1; + error (g, _("dispatch failed to marshal args")); + goto cleanup1; } } + /* Get the actual length of the message, resize the buffer to match + * the actual length, and write the length word at the beginning. + */ len = xdr_getpos (&xdr); xdr_destroy (&xdr); - /* Allocate the outgoing message buffer. */ - g->msg_out = safe_malloc (g, len + 4); - - g->msg_out_size = len + 4; - g->msg_out_pos = 0; + msg_out = safe_realloc (g, msg_out, len + 4); + msg_out_size = len + 4; - xdrmem_create (&xdr, g->msg_out, 4, XDR_ENCODE); + xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE); xdr_uint32_t (&xdr, &len); - memcpy (g->msg_out + 4, buffer, len); - - if (guestfs__switch_to_sending (g) == -1) - goto cleanup1; - - sent = 0; - guestfs_set_send_callback (g, send_cb, &sent); - if (ml->main_loop_run (ml, g) == -1) - goto cleanup1; - if (sent != 1) { - error (g, "send failed, see earlier error messages"); + again: + r = send_to_daemon (g, msg_out, msg_out_size); + if (r == -2) /* Ignore stray daemon cancellations. */ + goto again; + if (r == -1) goto cleanup1; - } + free (msg_out); return serial; cleanup1: - free (g->msg_out); - g->msg_out = NULL; - g->msg_out_size = 0; + free (msg_out); return -1; } static int cancel = 0; /* XXX Implement file cancellation. */ -static int send_file_chunk_sync (guestfs_h *g, int cancel, const char *buf, size_t len); -static int send_file_data_sync (guestfs_h *g, const char *buf, size_t len); -static int send_file_cancellation_sync (guestfs_h *g); -static int send_file_complete_sync (guestfs_h *g); +static int send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t len); +static int send_file_data (guestfs_h *g, const char *buf, size_t len); +static int send_file_cancellation (guestfs_h *g); +static int send_file_complete (guestfs_h *g); -/* Synchronously send a file. +/* Send a file. * Returns: * 0 OK * -1 error * -2 daemon cancelled (we must read the error message) */ int -guestfs__send_file_sync (guestfs_h *g, const char *filename) +guestfs___send_file (guestfs_h *g, const char *filename) { char buf[GUESTFS_MAX_CHUNK_SIZE]; int fd, r, err; @@ -1524,7 +2027,7 @@ guestfs__send_file_sync (guestfs_h *g, const char *filename) fd = open (filename, O_RDONLY); if (fd == -1) { perrorf (g, "open: %s", filename); - send_file_cancellation_sync (g); + send_file_cancellation (g); /* Daemon sees cancellation and won't reply, so caller can * just return here. */ @@ -1532,20 +2035,27 @@ guestfs__send_file_sync (guestfs_h *g, const char *filename) } /* Send file in chunked encoding. */ - while (!cancel && (r = read (fd, buf, sizeof buf)) > 0) { - err = send_file_data_sync (g, buf, r); - if (err < 0) + while (!cancel) { + r = read (fd, buf, sizeof buf); + if (r == -1 && (errno == EINTR || errno == EAGAIN)) + continue; + if (r <= 0) break; + err = send_file_data (g, buf, r); + if (err < 0) { + if (err == -2) /* daemon sent cancellation */ + send_file_cancellation (g); return err; + } } - if (cancel) { - send_file_cancellation_sync (g); + if (cancel) { /* cancel from either end */ + send_file_cancellation (g); return -1; } if (r == -1) { perrorf (g, "read: %s", filename); - send_file_cancellation_sync (g); + send_file_cancellation (g); return -1; } @@ -1554,160 +2064,168 @@ guestfs__send_file_sync (guestfs_h *g, const char *filename) */ if (close (fd) == -1) { perrorf (g, "close: %s", filename); - send_file_cancellation_sync (g); + send_file_cancellation (g); return -1; } - return send_file_complete_sync (g); + return send_file_complete (g); } /* Send a chunk of file data. */ static int -send_file_data_sync (guestfs_h *g, const char *buf, size_t len) +send_file_data (guestfs_h *g, const char *buf, size_t len) { - return send_file_chunk_sync (g, 0, buf, len); + return send_file_chunk (g, 0, buf, len); } /* Send a cancellation message. */ static int -send_file_cancellation_sync (guestfs_h *g) +send_file_cancellation (guestfs_h *g) { - return send_file_chunk_sync (g, 1, NULL, 0); + return send_file_chunk (g, 1, NULL, 0); } /* Send a file complete chunk. */ static int -send_file_complete_sync (guestfs_h *g) +send_file_complete (guestfs_h *g) { char buf[1]; - return send_file_chunk_sync (g, 0, buf, 0); + return send_file_chunk (g, 0, buf, 0); } -/* Send a chunk, cancellation or end of file, synchronously (ie. wait - * for it to go). - */ -static int check_for_daemon_cancellation (guestfs_h *g); - static int -send_file_chunk_sync (guestfs_h *g, int cancel, const char *buf, size_t len) +send_file_chunk (guestfs_h *g, int cancel, const char *buf, size_t buflen) { - char data[GUESTFS_MAX_CHUNK_SIZE + 48]; - unsigned datalen; - int sent; + u_int32_t len; + int r; guestfs_chunk chunk; XDR xdr; - guestfs_main_loop *ml = guestfs_get_main_loop (g); + char *msg_out; + size_t msg_out_size; if (g->state != BUSY) { - error (g, "send_file_chunk_sync: state %d != READY", g->state); + error (g, _("send_file_chunk: state %d != READY"), g->state); return -1; } - /* Did the daemon send a cancellation message? */ - if (check_for_daemon_cancellation (g)) - return -2; + /* Allocate the chunk buffer. Don't use the stack to avoid + * excessive stack usage and unnecessary copies. + */ + msg_out = safe_malloc (g, GUESTFS_MAX_CHUNK_SIZE + 4 + 48); + xdrmem_create (&xdr, msg_out + 4, GUESTFS_MAX_CHUNK_SIZE + 48, XDR_ENCODE); /* Serialize the chunk. */ chunk.cancel = cancel; - chunk.data.data_len = len; + chunk.data.data_len = buflen; chunk.data.data_val = (char *) buf; - xdrmem_create (&xdr, data, sizeof data, XDR_ENCODE); if (!xdr_guestfs_chunk (&xdr, &chunk)) { - error (g, "xdr_guestfs_chunk failed (buf = %p, len = %zu)", buf, len); + error (g, _("xdr_guestfs_chunk failed (buf = %p, buflen = %zu)"), + buf, buflen); xdr_destroy (&xdr); - return -1; + goto cleanup1; } - datalen = xdr_getpos (&xdr); + len = xdr_getpos (&xdr); xdr_destroy (&xdr); - /* Allocate outgoing message buffer. */ - g->msg_out = safe_malloc (g, datalen + 4); - g->msg_out_size = datalen + 4; - g->msg_out_pos = 0; + /* Reduce the size of the outgoing message buffer to the real length. */ + msg_out = safe_realloc (g, msg_out, len + 4); + msg_out_size = len + 4; - xdrmem_create (&xdr, g->msg_out, 4, XDR_ENCODE); - xdr_uint32_t (&xdr, &datalen); + xdrmem_create (&xdr, msg_out, 4, XDR_ENCODE); + xdr_uint32_t (&xdr, &len); - memcpy (g->msg_out + 4, data, datalen); + r = send_to_daemon (g, msg_out, msg_out_size); - if (guestfs__switch_to_sending (g) == -1) - goto cleanup1; + /* Did the daemon send a cancellation message? */ + if (r == -2) { + if (g->verbose) + fprintf (stderr, "got daemon cancellation\n"); + return -2; + } - sent = 0; - guestfs_set_send_callback (g, send_cb, &sent); - if (ml->main_loop_run (ml, g) == -1) - goto cleanup1; - if (sent != 1) { - error (g, "send file chunk failed, see earlier error messages"); + if (r == -1) goto cleanup1; - } + + free (msg_out); return 0; cleanup1: - free (g->msg_out); - g->msg_out = NULL; - g->msg_out_size = 0; + free (msg_out); return -1; } -/* At this point we are sending FileIn file(s) to the guest, and not - * expecting to read anything, so if we do read anything, it must be - * a cancellation message. This checks for this case without blocking. - */ -static int -check_for_daemon_cancellation (guestfs_h *g) +/* Receive a reply. */ +int +guestfs___recv (guestfs_h *g, const char *fn, + guestfs_message_header *hdr, + guestfs_message_error *err, + xdrproc_t xdrp, char *ret) { - fd_set rset; - struct timeval tv; - int r; - char buf[4]; - uint32_t flag; XDR xdr; + void *buf; + uint32_t size; + int r; - FD_ZERO (&rset); - FD_SET (g->sock, &rset); - tv.tv_sec = 0; - tv.tv_usec = 0; - r = select (g->sock+1, &rset, NULL, NULL, &tv); - if (r == -1) { - perrorf (g, "select"); - return 0; - } - if (r == 0) - return 0; + again: + r = recv_from_daemon (g, &size, &buf); + if (r == -1) + return -1; - /* Read the message from the daemon. */ - r = xread (g->sock, buf, sizeof buf); - if (r == -1) { - perrorf (g, "read"); - return 0; + /* This can happen if a cancellation happens right at the end + * of us sending a FileIn parameter to the daemon. Discard. The + * daemon should send us an error message next. + */ + if (size == GUESTFS_CANCEL_FLAG) + goto again; + + if (size == GUESTFS_LAUNCH_FLAG) { + error (g, "%s: received unexpected launch flag from daemon when expecting reply", fn); + return -1; } - xdrmem_create (&xdr, buf, sizeof buf, XDR_DECODE); - xdr_uint32_t (&xdr, &flag); - xdr_destroy (&xdr); + xdrmem_create (&xdr, buf, size, XDR_DECODE); - if (flag != GUESTFS_CANCEL_FLAG) { - error (g, "check_for_daemon_cancellation: read 0x%x from daemon, expected 0x%x\n", - flag, GUESTFS_CANCEL_FLAG); - return 0; + if (!xdr_guestfs_message_header (&xdr, hdr)) { + error (g, "%s: failed to parse reply header", fn); + xdr_destroy (&xdr); + free (buf); + return -1; } + if (hdr->status == GUESTFS_STATUS_ERROR) { + if (!xdr_guestfs_message_error (&xdr, err)) { + error (g, "%s: failed to parse reply error", fn); + xdr_destroy (&xdr); + free (buf); + return -1; + } + } else { + if (xdrp && ret && !xdrp (&xdr, ret)) { + error (g, "%s: failed to parse reply", fn); + xdr_destroy (&xdr); + free (buf); + return -1; + } + } + xdr_destroy (&xdr); + free (buf); - return 1; + return 0; } -/* Synchronously receive a file. */ +/* Receive a file. */ -static int receive_file_data_sync (guestfs_h *g, void **buf); +/* Returns -1 = error, 0 = EOF, > 0 = more data */ +static ssize_t receive_file_data (guestfs_h *g, void **buf); int -guestfs__receive_file_sync (guestfs_h *g, const char *filename) +guestfs___recv_file (guestfs_h *g, const char *filename) { void *buf; int fd, r; + size_t len; fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666); if (fd == -1) { @@ -1716,7 +2234,7 @@ guestfs__receive_file_sync (guestfs_h *g, const char *filename) } /* Receive the file in chunked encoding. */ - while ((r = receive_file_data_sync (g, &buf)) > 0) { + while ((r = receive_file_data (g, &buf)) > 0) { if (xwrite (fd, buf, r) == -1) { perrorf (g, "%s: write", filename); free (buf); @@ -1726,7 +2244,7 @@ guestfs__receive_file_sync (guestfs_h *g, const char *filename) } if (r == -1) { - error (g, "%s: error in chunked encoding", filename); + error (g, _("%s: error in chunked encoding"), filename); return -1; } @@ -1745,260 +2263,72 @@ guestfs__receive_file_sync (guestfs_h *g, const char *filename) char fbuf[4]; uint32_t flag = GUESTFS_CANCEL_FLAG; + if (g->verbose) + fprintf (stderr, "%s: waiting for daemon to acknowledge cancellation\n", + __func__); + xdrmem_create (&xdr, fbuf, sizeof fbuf, XDR_ENCODE); xdr_uint32_t (&xdr, &flag); xdr_destroy (&xdr); if (xwrite (g->sock, fbuf, sizeof fbuf) == -1) { - perrorf (g, "write to daemon socket"); + perrorf (g, _("write to daemon socket")); return -1; } - while ((r = receive_file_data_sync (g, &buf)) > 0) - free (buf); /* just discard it */ + while (receive_file_data (g, NULL) > 0) + ; /* just discard it */ return -1; } -struct receive_file_ctx { - int code; - void **buf; -}; - -static void -receive_file_cb (guestfs_h *g, void *data, XDR *xdr) -{ - guestfs_main_loop *ml = guestfs_get_main_loop (g); - struct receive_file_ctx *ctx = (struct receive_file_ctx *) data; - guestfs_chunk chunk; - - ml->main_loop_quit (ml, g); - - memset (&chunk, 0, sizeof chunk); - - if (!xdr_guestfs_chunk (xdr, &chunk)) { - error (g, "failed to parse file chunk"); - ctx->code = -1; - return; - } - if (chunk.cancel) { - error (g, "file receive cancelled by daemon"); - ctx->code = -2; - return; - } - if (chunk.data.data_len == 0) { /* end of transfer */ - ctx->code = 0; - return; - } - - ctx->code = chunk.data.data_len; - *ctx->buf = chunk.data.data_val; /* caller frees */ -} - /* Receive a chunk of file data. */ -static int -receive_file_data_sync (guestfs_h *g, void **buf) +/* Returns -1 = error, 0 = EOF, > 0 = more data */ +static ssize_t +receive_file_data (guestfs_h *g, void **buf_r) { - struct receive_file_ctx ctx; - guestfs_main_loop *ml = guestfs_get_main_loop (g); - - ctx.code = -3; - ctx.buf = buf; - - guestfs_set_reply_callback (g, receive_file_cb, &ctx); - (void) ml->main_loop_run (ml, g); - guestfs_set_reply_callback (g, NULL, NULL); - - if (g->verbose) - fprintf (stderr, "receive_file_data_sync: code %d\n", ctx.code); - - switch (ctx.code) { - case 0: /* end of file */ - return 0; - case -1: case -2: - return -1; - case -3: - error (g, "failed to call receive_file_cb"); - return -1; - default: /* received n bytes of data */ - return ctx.code; - } -} - -/* This is the default main loop implementation, using select(2). */ - -static int -select_add_handle (guestfs_main_loop *mlv, guestfs_h *g, int fd, int events, - guestfs_handle_event_cb cb, void *data) -{ - struct select_main_loop *ml = (struct select_main_loop *) mlv; - - if (fd < 0 || fd >= FD_SETSIZE) { - error (g, "fd %d is out of range", fd); - return -1; - } - - if ((events & ~(GUESTFS_HANDLE_READABLE | - GUESTFS_HANDLE_WRITABLE | - GUESTFS_HANDLE_HANGUP | - GUESTFS_HANDLE_ERROR)) != 0) { - error (g, "set of events (0x%x) contains unknown events", events); - return -1; - } - - if (events == 0) { - error (g, "set of events is empty"); - return -1; - } + int r; + void *buf; + uint32_t len; + XDR xdr; + guestfs_chunk chunk; - if (FD_ISSET (fd, &ml->rset) || - FD_ISSET (fd, &ml->wset) || - FD_ISSET (fd, &ml->xset)) { - error (g, "fd %d is already registered", fd); + r = recv_from_daemon (g, &len, &buf); + if (r == -1) { + error (g, _("receive_file_data: parse error in reply callback")); return -1; } - if (cb == NULL) { - error (g, "callback is NULL"); + if (len == GUESTFS_LAUNCH_FLAG || len == GUESTFS_CANCEL_FLAG) { + error (g, _("receive_file_data: unexpected flag received when reading file chunks")); return -1; } - if ((events & GUESTFS_HANDLE_READABLE)) - FD_SET (fd, &ml->rset); - if ((events & GUESTFS_HANDLE_WRITABLE)) - FD_SET (fd, &ml->wset); - if ((events & GUESTFS_HANDLE_HANGUP) || (events & GUESTFS_HANDLE_ERROR)) - FD_SET (fd, &ml->xset); - - if (fd > ml->max_fd) { - ml->max_fd = fd; - ml->handle_cb_data = - safe_realloc (g, ml->handle_cb_data, - sizeof (struct select_handle_cb_data) * (ml->max_fd+1)); - } - ml->handle_cb_data[fd].cb = cb; - ml->handle_cb_data[fd].g = g; - ml->handle_cb_data[fd].data = data; - - ml->nr_fds++; - - /* Any integer >= 0 can be the handle, and this is as good as any ... */ - return fd; -} - -static int -select_remove_handle (guestfs_main_loop *mlv, guestfs_h *g, int fd) -{ - struct select_main_loop *ml = (struct select_main_loop *) mlv; - - if (fd < 0 || fd >= FD_SETSIZE) { - error (g, "fd %d is out of range", fd); - return -1; - } + memset (&chunk, 0, sizeof chunk); - if (!FD_ISSET (fd, &ml->rset) && - !FD_ISSET (fd, &ml->wset) && - !FD_ISSET (fd, &ml->xset)) { - error (g, "fd %d was not registered", fd); + xdrmem_create (&xdr, buf, len, XDR_DECODE); + if (!xdr_guestfs_chunk (&xdr, &chunk)) { + error (g, _("failed to parse file chunk")); + free (buf); return -1; } + xdr_destroy (&xdr); + /* After decoding, the original buffer is no longer used. */ + free (buf); - FD_CLR (fd, &ml->rset); - FD_CLR (fd, &ml->wset); - FD_CLR (fd, &ml->xset); - - if (fd == ml->max_fd) { - ml->max_fd--; - ml->handle_cb_data = - safe_realloc (g, ml->handle_cb_data, - sizeof (struct select_handle_cb_data) * (ml->max_fd+1)); - } - - ml->nr_fds--; - - return 0; -} - -static int -select_add_timeout (guestfs_main_loop *mlv, guestfs_h *g, int interval, - guestfs_handle_timeout_cb cb, void *data) -{ - //struct select_main_loop *ml = (struct select_main_loop *) mlv; - - abort (); /* XXX not implemented yet */ -} - -static int -select_remove_timeout (guestfs_main_loop *mlv, guestfs_h *g, int timer) -{ - //struct select_main_loop *ml = (struct select_main_loop *) mlv; - - abort (); /* XXX not implemented yet */ -} - -/* The 'g' parameter is just used for error reporting. Events - * for multiple handles can be dispatched by running the main - * loop. - */ -static int -select_main_loop_run (guestfs_main_loop *mlv, guestfs_h *g) -{ - struct select_main_loop *ml = (struct select_main_loop *) mlv; - int fd, r, events; - fd_set rset2, wset2, xset2; - - if (ml->is_running) { - error (g, "select_main_loop_run: this cannot be called recursively"); + if (chunk.cancel) { + error (g, _("file receive cancelled by daemon")); + free (chunk.data.data_val); return -1; } - ml->is_running = 1; - - while (ml->is_running) { - if (ml->nr_fds == 0) - break; - - rset2 = ml->rset; - wset2 = ml->wset; - xset2 = ml->xset; - r = select (ml->max_fd+1, &rset2, &wset2, &xset2, NULL); - if (r == -1) { - perrorf (g, "select"); - ml->is_running = 0; - return -1; - } - - for (fd = 0; r > 0 && fd <= ml->max_fd; ++fd) { - events = 0; - if (FD_ISSET (fd, &rset2)) - events |= GUESTFS_HANDLE_READABLE; - if (FD_ISSET (fd, &wset2)) - events |= GUESTFS_HANDLE_WRITABLE; - if (FD_ISSET (fd, &xset2)) - events |= GUESTFS_HANDLE_ERROR | GUESTFS_HANDLE_HANGUP; - if (events) { - r--; - ml->handle_cb_data[fd].cb ((guestfs_main_loop *) ml, - ml->handle_cb_data[fd].g, - ml->handle_cb_data[fd].data, - fd, fd, events); - } - } + if (chunk.data.data_len == 0) { /* end of transfer */ + free (chunk.data.data_val); + return 0; } - ml->is_running = 0; - return 0; -} - -static int -select_main_loop_quit (guestfs_main_loop *mlv, guestfs_h *g) -{ - struct select_main_loop *ml = (struct select_main_loop *) mlv; + if (buf_r) *buf_r = chunk.data.data_val; + else free (chunk.data.data_val); /* else caller frees */ - /* Note that legitimately ml->is_running can be zero when - * this function is called. - */ - - ml->is_running = 0; - return 0; + return chunk.data.data_len; }