]> git.proxmox.com Git - qemu.git/blob - qga/commands-posix.c
qga: distinguish binary modes in "guest_file_open_modes" map
[qemu.git] / qga / commands-posix.c
1 /*
2 * QEMU Guest Agent POSIX-specific command implementations
3 *
4 * Copyright IBM Corp. 2011
5 *
6 * Authors:
7 * Michael Roth <mdroth@linux.vnet.ibm.com>
8 * Michal Privoznik <mprivozn@redhat.com>
9 *
10 * This work is licensed under the terms of the GNU GPL, version 2 or later.
11 * See the COPYING file in the top-level directory.
12 */
13
14 #include <glib.h>
15 #include <sys/types.h>
16 #include <sys/ioctl.h>
17 #include <sys/wait.h>
18 #include <unistd.h>
19 #include <errno.h>
20 #include <fcntl.h>
21 #include <stdio.h>
22 #include <string.h>
23 #include <sys/stat.h>
24 #include <inttypes.h>
25 #include "qga/guest-agent-core.h"
26 #include "qga-qmp-commands.h"
27 #include "qapi/qmp/qerror.h"
28 #include "qemu/queue.h"
29 #include "qemu/host-utils.h"
30
31 #ifndef CONFIG_HAS_ENVIRON
32 #ifdef __APPLE__
33 #include <crt_externs.h>
34 #define environ (*_NSGetEnviron())
35 #else
36 extern char **environ;
37 #endif
38 #endif
39
40 #if defined(__linux__)
41 #include <mntent.h>
42 #include <linux/fs.h>
43 #include <ifaddrs.h>
44 #include <arpa/inet.h>
45 #include <sys/socket.h>
46 #include <net/if.h>
47
48 #ifdef FIFREEZE
49 #define CONFIG_FSFREEZE
50 #endif
51 #ifdef FITRIM
52 #define CONFIG_FSTRIM
53 #endif
54 #endif
55
56 static void ga_wait_child(pid_t pid, int *status, Error **err)
57 {
58 pid_t rpid;
59
60 *status = 0;
61
62 do {
63 rpid = waitpid(pid, status, 0);
64 } while (rpid == -1 && errno == EINTR);
65
66 if (rpid == -1) {
67 error_setg_errno(err, errno, "failed to wait for child (pid: %d)", pid);
68 return;
69 }
70
71 g_assert(rpid == pid);
72 }
73
74 void qmp_guest_shutdown(bool has_mode, const char *mode, Error **err)
75 {
76 const char *shutdown_flag;
77 Error *local_err = NULL;
78 pid_t pid;
79 int status;
80
81 slog("guest-shutdown called, mode: %s", mode);
82 if (!has_mode || strcmp(mode, "powerdown") == 0) {
83 shutdown_flag = "-P";
84 } else if (strcmp(mode, "halt") == 0) {
85 shutdown_flag = "-H";
86 } else if (strcmp(mode, "reboot") == 0) {
87 shutdown_flag = "-r";
88 } else {
89 error_setg(err,
90 "mode is invalid (valid values are: halt|powerdown|reboot");
91 return;
92 }
93
94 pid = fork();
95 if (pid == 0) {
96 /* child, start the shutdown */
97 setsid();
98 reopen_fd_to_null(0);
99 reopen_fd_to_null(1);
100 reopen_fd_to_null(2);
101
102 execle("/sbin/shutdown", "shutdown", shutdown_flag, "+0",
103 "hypervisor initiated shutdown", (char*)NULL, environ);
104 _exit(EXIT_FAILURE);
105 } else if (pid < 0) {
106 error_setg_errno(err, errno, "failed to create child process");
107 return;
108 }
109
110 ga_wait_child(pid, &status, &local_err);
111 if (error_is_set(&local_err)) {
112 error_propagate(err, local_err);
113 return;
114 }
115
116 if (!WIFEXITED(status)) {
117 error_setg(err, "child process has terminated abnormally");
118 return;
119 }
120
121 if (WEXITSTATUS(status)) {
122 error_setg(err, "child process has failed to shutdown");
123 return;
124 }
125
126 /* succeeded */
127 }
128
129 int64_t qmp_guest_get_time(Error **errp)
130 {
131 int ret;
132 qemu_timeval tq;
133 int64_t time_ns;
134
135 ret = qemu_gettimeofday(&tq);
136 if (ret < 0) {
137 error_setg_errno(errp, errno, "Failed to get time");
138 return -1;
139 }
140
141 time_ns = tq.tv_sec * 1000000000LL + tq.tv_usec * 1000;
142 return time_ns;
143 }
144
145 void qmp_guest_set_time(int64_t time_ns, Error **errp)
146 {
147 int ret;
148 int status;
149 pid_t pid;
150 Error *local_err = NULL;
151 struct timeval tv;
152
153 /* year-2038 will overflow in case time_t is 32bit */
154 if (time_ns / 1000000000 != (time_t)(time_ns / 1000000000)) {
155 error_setg(errp, "Time %" PRId64 " is too large", time_ns);
156 return;
157 }
158
159 tv.tv_sec = time_ns / 1000000000;
160 tv.tv_usec = (time_ns % 1000000000) / 1000;
161
162 ret = settimeofday(&tv, NULL);
163 if (ret < 0) {
164 error_setg_errno(errp, errno, "Failed to set time to guest");
165 return;
166 }
167
168 /* Set the Hardware Clock to the current System Time. */
169 pid = fork();
170 if (pid == 0) {
171 setsid();
172 reopen_fd_to_null(0);
173 reopen_fd_to_null(1);
174 reopen_fd_to_null(2);
175
176 execle("/sbin/hwclock", "hwclock", "-w", NULL, environ);
177 _exit(EXIT_FAILURE);
178 } else if (pid < 0) {
179 error_setg_errno(errp, errno, "failed to create child process");
180 return;
181 }
182
183 ga_wait_child(pid, &status, &local_err);
184 if (error_is_set(&local_err)) {
185 error_propagate(errp, local_err);
186 return;
187 }
188
189 if (!WIFEXITED(status)) {
190 error_setg(errp, "child process has terminated abnormally");
191 return;
192 }
193
194 if (WEXITSTATUS(status)) {
195 error_setg(errp, "hwclock failed to set hardware clock to system time");
196 return;
197 }
198 }
199
200 typedef struct GuestFileHandle {
201 uint64_t id;
202 FILE *fh;
203 QTAILQ_ENTRY(GuestFileHandle) next;
204 } GuestFileHandle;
205
206 static struct {
207 QTAILQ_HEAD(, GuestFileHandle) filehandles;
208 } guest_file_state;
209
210 static int64_t guest_file_handle_add(FILE *fh, Error **errp)
211 {
212 GuestFileHandle *gfh;
213 int64_t handle;
214
215 handle = ga_get_fd_handle(ga_state, errp);
216 if (error_is_set(errp)) {
217 return 0;
218 }
219
220 gfh = g_malloc0(sizeof(GuestFileHandle));
221 gfh->id = handle;
222 gfh->fh = fh;
223 QTAILQ_INSERT_TAIL(&guest_file_state.filehandles, gfh, next);
224
225 return handle;
226 }
227
228 static GuestFileHandle *guest_file_handle_find(int64_t id, Error **err)
229 {
230 GuestFileHandle *gfh;
231
232 QTAILQ_FOREACH(gfh, &guest_file_state.filehandles, next)
233 {
234 if (gfh->id == id) {
235 return gfh;
236 }
237 }
238
239 error_setg(err, "handle '%" PRId64 "' has not been found", id);
240 return NULL;
241 }
242
243 typedef const char * const ccpc;
244
245 #ifndef O_BINARY
246 #define O_BINARY 0
247 #endif
248
249 /* http://pubs.opengroup.org/onlinepubs/9699919799/functions/fopen.html */
250 static const struct {
251 ccpc *forms;
252 int oflag_base;
253 } guest_file_open_modes[] = {
254 { (ccpc[]){ "r", NULL }, O_RDONLY },
255 { (ccpc[]){ "rb", NULL }, O_RDONLY | O_BINARY },
256 { (ccpc[]){ "w", NULL }, O_WRONLY | O_CREAT | O_TRUNC },
257 { (ccpc[]){ "wb", NULL }, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY },
258 { (ccpc[]){ "a", NULL }, O_WRONLY | O_CREAT | O_APPEND },
259 { (ccpc[]){ "ab", NULL }, O_WRONLY | O_CREAT | O_APPEND | O_BINARY },
260 { (ccpc[]){ "r+", NULL }, O_RDWR },
261 { (ccpc[]){ "rb+", "r+b", NULL }, O_RDWR | O_BINARY },
262 { (ccpc[]){ "w+", NULL }, O_RDWR | O_CREAT | O_TRUNC },
263 { (ccpc[]){ "wb+", "w+b", NULL }, O_RDWR | O_CREAT | O_TRUNC | O_BINARY },
264 { (ccpc[]){ "a+", NULL }, O_RDWR | O_CREAT | O_APPEND },
265 { (ccpc[]){ "ab+", "a+b", NULL }, O_RDWR | O_CREAT | O_APPEND | O_BINARY }
266 };
267
268 static int
269 find_open_flag(const char *mode_str, Error **err)
270 {
271 unsigned mode;
272
273 for (mode = 0; mode < ARRAY_SIZE(guest_file_open_modes); ++mode) {
274 ccpc *form;
275
276 form = guest_file_open_modes[mode].forms;
277 while (*form != NULL && strcmp(*form, mode_str) != 0) {
278 ++form;
279 }
280 if (*form != NULL) {
281 break;
282 }
283 }
284
285 if (mode == ARRAY_SIZE(guest_file_open_modes)) {
286 error_setg(err, "invalid file open mode '%s'", mode_str);
287 return -1;
288 }
289 return guest_file_open_modes[mode].oflag_base | O_NOCTTY | O_NONBLOCK;
290 }
291
292 #define DEFAULT_NEW_FILE_MODE (S_IRUSR | S_IWUSR | \
293 S_IRGRP | S_IWGRP | \
294 S_IROTH | S_IWOTH)
295
296 static FILE *
297 safe_open_or_create(const char *path, const char *mode, Error **err)
298 {
299 Error *local_err = NULL;
300 int oflag;
301
302 oflag = find_open_flag(mode, &local_err);
303 if (local_err == NULL) {
304 int fd;
305
306 /* If the caller wants / allows creation of a new file, we implement it
307 * with a two step process: open() + (open() / fchmod()).
308 *
309 * First we insist on creating the file exclusively as a new file. If
310 * that succeeds, we're free to set any file-mode bits on it. (The
311 * motivation is that we want to set those file-mode bits independently
312 * of the current umask.)
313 *
314 * If the exclusive creation fails because the file already exists
315 * (EEXIST is not possible for any other reason), we just attempt to
316 * open the file, but in this case we won't be allowed to change the
317 * file-mode bits on the preexistent file.
318 *
319 * The pathname should never disappear between the two open()s in
320 * practice. If it happens, then someone very likely tried to race us.
321 * In this case just go ahead and report the ENOENT from the second
322 * open() to the caller.
323 *
324 * If the caller wants to open a preexistent file, then the first
325 * open() is decisive and its third argument is ignored, and the second
326 * open() and the fchmod() are never called.
327 */
328 fd = open(path, oflag | ((oflag & O_CREAT) ? O_EXCL : 0), 0);
329 if (fd == -1 && errno == EEXIST) {
330 oflag &= ~(unsigned)O_CREAT;
331 fd = open(path, oflag);
332 }
333
334 if (fd == -1) {
335 error_setg_errno(&local_err, errno, "failed to open file '%s' "
336 "(mode: '%s')", path, mode);
337 } else {
338 qemu_set_cloexec(fd);
339
340 if ((oflag & O_CREAT) && fchmod(fd, DEFAULT_NEW_FILE_MODE) == -1) {
341 error_setg_errno(&local_err, errno, "failed to set permission "
342 "0%03o on new file '%s' (mode: '%s')",
343 (unsigned)DEFAULT_NEW_FILE_MODE, path, mode);
344 } else {
345 FILE *f;
346
347 f = fdopen(fd, mode);
348 if (f == NULL) {
349 error_setg_errno(&local_err, errno, "failed to associate "
350 "stdio stream with file descriptor %d, "
351 "file '%s' (mode: '%s')", fd, path, mode);
352 } else {
353 return f;
354 }
355 }
356
357 close(fd);
358 }
359 }
360
361 error_propagate(err, local_err);
362 return NULL;
363 }
364
365 int64_t qmp_guest_file_open(const char *path, bool has_mode, const char *mode, Error **err)
366 {
367 FILE *fh;
368 Error *local_err = NULL;
369 int fd;
370 int64_t ret = -1, handle;
371
372 if (!has_mode) {
373 mode = "r";
374 }
375 slog("guest-file-open called, filepath: %s, mode: %s", path, mode);
376 fh = safe_open_or_create(path, mode, &local_err);
377 if (local_err != NULL) {
378 error_propagate(err, local_err);
379 return -1;
380 }
381
382 /* set fd non-blocking to avoid common use cases (like reading from a
383 * named pipe) from hanging the agent
384 */
385 fd = fileno(fh);
386 ret = fcntl(fd, F_GETFL);
387 ret = fcntl(fd, F_SETFL, ret | O_NONBLOCK);
388 if (ret == -1) {
389 error_setg_errno(err, errno, "failed to make file '%s' non-blocking",
390 path);
391 fclose(fh);
392 return -1;
393 }
394
395 handle = guest_file_handle_add(fh, err);
396 if (error_is_set(err)) {
397 fclose(fh);
398 return -1;
399 }
400
401 slog("guest-file-open, handle: %d", handle);
402 return handle;
403 }
404
405 void qmp_guest_file_close(int64_t handle, Error **err)
406 {
407 GuestFileHandle *gfh = guest_file_handle_find(handle, err);
408 int ret;
409
410 slog("guest-file-close called, handle: %ld", handle);
411 if (!gfh) {
412 return;
413 }
414
415 ret = fclose(gfh->fh);
416 if (ret == EOF) {
417 error_setg_errno(err, errno, "failed to close handle");
418 return;
419 }
420
421 QTAILQ_REMOVE(&guest_file_state.filehandles, gfh, next);
422 g_free(gfh);
423 }
424
425 struct GuestFileRead *qmp_guest_file_read(int64_t handle, bool has_count,
426 int64_t count, Error **err)
427 {
428 GuestFileHandle *gfh = guest_file_handle_find(handle, err);
429 GuestFileRead *read_data = NULL;
430 guchar *buf;
431 FILE *fh;
432 size_t read_count;
433
434 if (!gfh) {
435 return NULL;
436 }
437
438 if (!has_count) {
439 count = QGA_READ_COUNT_DEFAULT;
440 } else if (count < 0) {
441 error_setg(err, "value '%" PRId64 "' is invalid for argument count",
442 count);
443 return NULL;
444 }
445
446 fh = gfh->fh;
447 buf = g_malloc0(count+1);
448 read_count = fread(buf, 1, count, fh);
449 if (ferror(fh)) {
450 error_setg_errno(err, errno, "failed to read file");
451 slog("guest-file-read failed, handle: %ld", handle);
452 } else {
453 buf[read_count] = 0;
454 read_data = g_malloc0(sizeof(GuestFileRead));
455 read_data->count = read_count;
456 read_data->eof = feof(fh);
457 if (read_count) {
458 read_data->buf_b64 = g_base64_encode(buf, read_count);
459 }
460 }
461 g_free(buf);
462 clearerr(fh);
463
464 return read_data;
465 }
466
467 GuestFileWrite *qmp_guest_file_write(int64_t handle, const char *buf_b64,
468 bool has_count, int64_t count, Error **err)
469 {
470 GuestFileWrite *write_data = NULL;
471 guchar *buf;
472 gsize buf_len;
473 int write_count;
474 GuestFileHandle *gfh = guest_file_handle_find(handle, err);
475 FILE *fh;
476
477 if (!gfh) {
478 return NULL;
479 }
480
481 fh = gfh->fh;
482 buf = g_base64_decode(buf_b64, &buf_len);
483
484 if (!has_count) {
485 count = buf_len;
486 } else if (count < 0 || count > buf_len) {
487 error_setg(err, "value '%" PRId64 "' is invalid for argument count",
488 count);
489 g_free(buf);
490 return NULL;
491 }
492
493 write_count = fwrite(buf, 1, count, fh);
494 if (ferror(fh)) {
495 error_setg_errno(err, errno, "failed to write to file");
496 slog("guest-file-write failed, handle: %ld", handle);
497 } else {
498 write_data = g_malloc0(sizeof(GuestFileWrite));
499 write_data->count = write_count;
500 write_data->eof = feof(fh);
501 }
502 g_free(buf);
503 clearerr(fh);
504
505 return write_data;
506 }
507
508 struct GuestFileSeek *qmp_guest_file_seek(int64_t handle, int64_t offset,
509 int64_t whence, Error **err)
510 {
511 GuestFileHandle *gfh = guest_file_handle_find(handle, err);
512 GuestFileSeek *seek_data = NULL;
513 FILE *fh;
514 int ret;
515
516 if (!gfh) {
517 return NULL;
518 }
519
520 fh = gfh->fh;
521 ret = fseek(fh, offset, whence);
522 if (ret == -1) {
523 error_setg_errno(err, errno, "failed to seek file");
524 } else {
525 seek_data = g_malloc0(sizeof(GuestFileRead));
526 seek_data->position = ftell(fh);
527 seek_data->eof = feof(fh);
528 }
529 clearerr(fh);
530
531 return seek_data;
532 }
533
534 void qmp_guest_file_flush(int64_t handle, Error **err)
535 {
536 GuestFileHandle *gfh = guest_file_handle_find(handle, err);
537 FILE *fh;
538 int ret;
539
540 if (!gfh) {
541 return;
542 }
543
544 fh = gfh->fh;
545 ret = fflush(fh);
546 if (ret == EOF) {
547 error_setg_errno(err, errno, "failed to flush file");
548 }
549 }
550
551 static void guest_file_init(void)
552 {
553 QTAILQ_INIT(&guest_file_state.filehandles);
554 }
555
556 /* linux-specific implementations. avoid this if at all possible. */
557 #if defined(__linux__)
558
559 #if defined(CONFIG_FSFREEZE) || defined(CONFIG_FSTRIM)
560 typedef struct FsMount {
561 char *dirname;
562 char *devtype;
563 QTAILQ_ENTRY(FsMount) next;
564 } FsMount;
565
566 typedef QTAILQ_HEAD(, FsMount) FsMountList;
567
568 static void free_fs_mount_list(FsMountList *mounts)
569 {
570 FsMount *mount, *temp;
571
572 if (!mounts) {
573 return;
574 }
575
576 QTAILQ_FOREACH_SAFE(mount, mounts, next, temp) {
577 QTAILQ_REMOVE(mounts, mount, next);
578 g_free(mount->dirname);
579 g_free(mount->devtype);
580 g_free(mount);
581 }
582 }
583
584 /*
585 * Walk the mount table and build a list of local file systems
586 */
587 static void build_fs_mount_list(FsMountList *mounts, Error **err)
588 {
589 struct mntent *ment;
590 FsMount *mount;
591 char const *mtab = "/proc/self/mounts";
592 FILE *fp;
593
594 fp = setmntent(mtab, "r");
595 if (!fp) {
596 error_setg(err, "failed to open mtab file: '%s'", mtab);
597 return;
598 }
599
600 while ((ment = getmntent(fp))) {
601 /*
602 * An entry which device name doesn't start with a '/' is
603 * either a dummy file system or a network file system.
604 * Add special handling for smbfs and cifs as is done by
605 * coreutils as well.
606 */
607 if ((ment->mnt_fsname[0] != '/') ||
608 (strcmp(ment->mnt_type, "smbfs") == 0) ||
609 (strcmp(ment->mnt_type, "cifs") == 0)) {
610 continue;
611 }
612
613 mount = g_malloc0(sizeof(FsMount));
614 mount->dirname = g_strdup(ment->mnt_dir);
615 mount->devtype = g_strdup(ment->mnt_type);
616
617 QTAILQ_INSERT_TAIL(mounts, mount, next);
618 }
619
620 endmntent(fp);
621 }
622 #endif
623
624 #if defined(CONFIG_FSFREEZE)
625
626 typedef enum {
627 FSFREEZE_HOOK_THAW = 0,
628 FSFREEZE_HOOK_FREEZE,
629 } FsfreezeHookArg;
630
631 const char *fsfreeze_hook_arg_string[] = {
632 "thaw",
633 "freeze",
634 };
635
636 static void execute_fsfreeze_hook(FsfreezeHookArg arg, Error **err)
637 {
638 int status;
639 pid_t pid;
640 const char *hook;
641 const char *arg_str = fsfreeze_hook_arg_string[arg];
642 Error *local_err = NULL;
643
644 hook = ga_fsfreeze_hook(ga_state);
645 if (!hook) {
646 return;
647 }
648 if (access(hook, X_OK) != 0) {
649 error_setg_errno(err, errno, "can't access fsfreeze hook '%s'", hook);
650 return;
651 }
652
653 slog("executing fsfreeze hook with arg '%s'", arg_str);
654 pid = fork();
655 if (pid == 0) {
656 setsid();
657 reopen_fd_to_null(0);
658 reopen_fd_to_null(1);
659 reopen_fd_to_null(2);
660
661 execle(hook, hook, arg_str, NULL, environ);
662 _exit(EXIT_FAILURE);
663 } else if (pid < 0) {
664 error_setg_errno(err, errno, "failed to create child process");
665 return;
666 }
667
668 ga_wait_child(pid, &status, &local_err);
669 if (error_is_set(&local_err)) {
670 error_propagate(err, local_err);
671 return;
672 }
673
674 if (!WIFEXITED(status)) {
675 error_setg(err, "fsfreeze hook has terminated abnormally");
676 return;
677 }
678
679 status = WEXITSTATUS(status);
680 if (status) {
681 error_setg(err, "fsfreeze hook has failed with status %d", status);
682 return;
683 }
684 }
685
686 /*
687 * Return status of freeze/thaw
688 */
689 GuestFsfreezeStatus qmp_guest_fsfreeze_status(Error **err)
690 {
691 if (ga_is_frozen(ga_state)) {
692 return GUEST_FSFREEZE_STATUS_FROZEN;
693 }
694
695 return GUEST_FSFREEZE_STATUS_THAWED;
696 }
697
698 /*
699 * Walk list of mounted file systems in the guest, and freeze the ones which
700 * are real local file systems.
701 */
702 int64_t qmp_guest_fsfreeze_freeze(Error **err)
703 {
704 int ret = 0, i = 0;
705 FsMountList mounts;
706 struct FsMount *mount;
707 Error *local_err = NULL;
708 int fd;
709
710 slog("guest-fsfreeze called");
711
712 execute_fsfreeze_hook(FSFREEZE_HOOK_FREEZE, &local_err);
713 if (error_is_set(&local_err)) {
714 error_propagate(err, local_err);
715 return -1;
716 }
717
718 QTAILQ_INIT(&mounts);
719 build_fs_mount_list(&mounts, &local_err);
720 if (error_is_set(&local_err)) {
721 error_propagate(err, local_err);
722 return -1;
723 }
724
725 /* cannot risk guest agent blocking itself on a write in this state */
726 ga_set_frozen(ga_state);
727
728 QTAILQ_FOREACH(mount, &mounts, next) {
729 fd = qemu_open(mount->dirname, O_RDONLY);
730 if (fd == -1) {
731 error_setg_errno(err, errno, "failed to open %s", mount->dirname);
732 goto error;
733 }
734
735 /* we try to cull filesytems we know won't work in advance, but other
736 * filesytems may not implement fsfreeze for less obvious reasons.
737 * these will report EOPNOTSUPP. we simply ignore these when tallying
738 * the number of frozen filesystems.
739 *
740 * any other error means a failure to freeze a filesystem we
741 * expect to be freezable, so return an error in those cases
742 * and return system to thawed state.
743 */
744 ret = ioctl(fd, FIFREEZE);
745 if (ret == -1) {
746 if (errno != EOPNOTSUPP) {
747 error_setg_errno(err, errno, "failed to freeze %s",
748 mount->dirname);
749 close(fd);
750 goto error;
751 }
752 } else {
753 i++;
754 }
755 close(fd);
756 }
757
758 free_fs_mount_list(&mounts);
759 return i;
760
761 error:
762 free_fs_mount_list(&mounts);
763 qmp_guest_fsfreeze_thaw(NULL);
764 return 0;
765 }
766
767 /*
768 * Walk list of frozen file systems in the guest, and thaw them.
769 */
770 int64_t qmp_guest_fsfreeze_thaw(Error **err)
771 {
772 int ret;
773 FsMountList mounts;
774 FsMount *mount;
775 int fd, i = 0, logged;
776 Error *local_err = NULL;
777
778 QTAILQ_INIT(&mounts);
779 build_fs_mount_list(&mounts, &local_err);
780 if (error_is_set(&local_err)) {
781 error_propagate(err, local_err);
782 return 0;
783 }
784
785 QTAILQ_FOREACH(mount, &mounts, next) {
786 logged = false;
787 fd = qemu_open(mount->dirname, O_RDONLY);
788 if (fd == -1) {
789 continue;
790 }
791 /* we have no way of knowing whether a filesystem was actually unfrozen
792 * as a result of a successful call to FITHAW, only that if an error
793 * was returned the filesystem was *not* unfrozen by that particular
794 * call.
795 *
796 * since multiple preceding FIFREEZEs require multiple calls to FITHAW
797 * to unfreeze, continuing issuing FITHAW until an error is returned,
798 * in which case either the filesystem is in an unfreezable state, or,
799 * more likely, it was thawed previously (and remains so afterward).
800 *
801 * also, since the most recent successful call is the one that did
802 * the actual unfreeze, we can use this to provide an accurate count
803 * of the number of filesystems unfrozen by guest-fsfreeze-thaw, which
804 * may * be useful for determining whether a filesystem was unfrozen
805 * during the freeze/thaw phase by a process other than qemu-ga.
806 */
807 do {
808 ret = ioctl(fd, FITHAW);
809 if (ret == 0 && !logged) {
810 i++;
811 logged = true;
812 }
813 } while (ret == 0);
814 close(fd);
815 }
816
817 ga_unset_frozen(ga_state);
818 free_fs_mount_list(&mounts);
819
820 execute_fsfreeze_hook(FSFREEZE_HOOK_THAW, err);
821
822 return i;
823 }
824
825 static void guest_fsfreeze_cleanup(void)
826 {
827 Error *err = NULL;
828
829 if (ga_is_frozen(ga_state) == GUEST_FSFREEZE_STATUS_FROZEN) {
830 qmp_guest_fsfreeze_thaw(&err);
831 if (err) {
832 slog("failed to clean up frozen filesystems: %s",
833 error_get_pretty(err));
834 error_free(err);
835 }
836 }
837 }
838 #endif /* CONFIG_FSFREEZE */
839
840 #if defined(CONFIG_FSTRIM)
841 /*
842 * Walk list of mounted file systems in the guest, and trim them.
843 */
844 void qmp_guest_fstrim(bool has_minimum, int64_t minimum, Error **err)
845 {
846 int ret = 0;
847 FsMountList mounts;
848 struct FsMount *mount;
849 int fd;
850 Error *local_err = NULL;
851 struct fstrim_range r = {
852 .start = 0,
853 .len = -1,
854 .minlen = has_minimum ? minimum : 0,
855 };
856
857 slog("guest-fstrim called");
858
859 QTAILQ_INIT(&mounts);
860 build_fs_mount_list(&mounts, &local_err);
861 if (error_is_set(&local_err)) {
862 error_propagate(err, local_err);
863 return;
864 }
865
866 QTAILQ_FOREACH(mount, &mounts, next) {
867 fd = qemu_open(mount->dirname, O_RDONLY);
868 if (fd == -1) {
869 error_setg_errno(err, errno, "failed to open %s", mount->dirname);
870 goto error;
871 }
872
873 /* We try to cull filesytems we know won't work in advance, but other
874 * filesytems may not implement fstrim for less obvious reasons. These
875 * will report EOPNOTSUPP; we simply ignore these errors. Any other
876 * error means an unexpected error, so return it in those cases. In
877 * some other cases ENOTTY will be reported (e.g. CD-ROMs).
878 */
879 ret = ioctl(fd, FITRIM, &r);
880 if (ret == -1) {
881 if (errno != ENOTTY && errno != EOPNOTSUPP) {
882 error_setg_errno(err, errno, "failed to trim %s",
883 mount->dirname);
884 close(fd);
885 goto error;
886 }
887 }
888 close(fd);
889 }
890
891 error:
892 free_fs_mount_list(&mounts);
893 }
894 #endif /* CONFIG_FSTRIM */
895
896
897 #define LINUX_SYS_STATE_FILE "/sys/power/state"
898 #define SUSPEND_SUPPORTED 0
899 #define SUSPEND_NOT_SUPPORTED 1
900
901 static void bios_supports_mode(const char *pmutils_bin, const char *pmutils_arg,
902 const char *sysfile_str, Error **err)
903 {
904 Error *local_err = NULL;
905 char *pmutils_path;
906 pid_t pid;
907 int status;
908
909 pmutils_path = g_find_program_in_path(pmutils_bin);
910
911 pid = fork();
912 if (!pid) {
913 char buf[32]; /* hopefully big enough */
914 ssize_t ret;
915 int fd;
916
917 setsid();
918 reopen_fd_to_null(0);
919 reopen_fd_to_null(1);
920 reopen_fd_to_null(2);
921
922 if (pmutils_path) {
923 execle(pmutils_path, pmutils_bin, pmutils_arg, NULL, environ);
924 }
925
926 /*
927 * If we get here either pm-utils is not installed or execle() has
928 * failed. Let's try the manual method if the caller wants it.
929 */
930
931 if (!sysfile_str) {
932 _exit(SUSPEND_NOT_SUPPORTED);
933 }
934
935 fd = open(LINUX_SYS_STATE_FILE, O_RDONLY);
936 if (fd < 0) {
937 _exit(SUSPEND_NOT_SUPPORTED);
938 }
939
940 ret = read(fd, buf, sizeof(buf)-1);
941 if (ret <= 0) {
942 _exit(SUSPEND_NOT_SUPPORTED);
943 }
944 buf[ret] = '\0';
945
946 if (strstr(buf, sysfile_str)) {
947 _exit(SUSPEND_SUPPORTED);
948 }
949
950 _exit(SUSPEND_NOT_SUPPORTED);
951 } else if (pid < 0) {
952 error_setg_errno(err, errno, "failed to create child process");
953 goto out;
954 }
955
956 ga_wait_child(pid, &status, &local_err);
957 if (error_is_set(&local_err)) {
958 error_propagate(err, local_err);
959 goto out;
960 }
961
962 if (!WIFEXITED(status)) {
963 error_setg(err, "child process has terminated abnormally");
964 goto out;
965 }
966
967 switch (WEXITSTATUS(status)) {
968 case SUSPEND_SUPPORTED:
969 goto out;
970 case SUSPEND_NOT_SUPPORTED:
971 error_setg(err,
972 "the requested suspend mode is not supported by the guest");
973 goto out;
974 default:
975 error_setg(err,
976 "the helper program '%s' returned an unexpected exit status"
977 " code (%d)", pmutils_path, WEXITSTATUS(status));
978 goto out;
979 }
980
981 out:
982 g_free(pmutils_path);
983 }
984
985 static void guest_suspend(const char *pmutils_bin, const char *sysfile_str,
986 Error **err)
987 {
988 Error *local_err = NULL;
989 char *pmutils_path;
990 pid_t pid;
991 int status;
992
993 pmutils_path = g_find_program_in_path(pmutils_bin);
994
995 pid = fork();
996 if (pid == 0) {
997 /* child */
998 int fd;
999
1000 setsid();
1001 reopen_fd_to_null(0);
1002 reopen_fd_to_null(1);
1003 reopen_fd_to_null(2);
1004
1005 if (pmutils_path) {
1006 execle(pmutils_path, pmutils_bin, NULL, environ);
1007 }
1008
1009 /*
1010 * If we get here either pm-utils is not installed or execle() has
1011 * failed. Let's try the manual method if the caller wants it.
1012 */
1013
1014 if (!sysfile_str) {
1015 _exit(EXIT_FAILURE);
1016 }
1017
1018 fd = open(LINUX_SYS_STATE_FILE, O_WRONLY);
1019 if (fd < 0) {
1020 _exit(EXIT_FAILURE);
1021 }
1022
1023 if (write(fd, sysfile_str, strlen(sysfile_str)) < 0) {
1024 _exit(EXIT_FAILURE);
1025 }
1026
1027 _exit(EXIT_SUCCESS);
1028 } else if (pid < 0) {
1029 error_setg_errno(err, errno, "failed to create child process");
1030 goto out;
1031 }
1032
1033 ga_wait_child(pid, &status, &local_err);
1034 if (error_is_set(&local_err)) {
1035 error_propagate(err, local_err);
1036 goto out;
1037 }
1038
1039 if (!WIFEXITED(status)) {
1040 error_setg(err, "child process has terminated abnormally");
1041 goto out;
1042 }
1043
1044 if (WEXITSTATUS(status)) {
1045 error_setg(err, "child process has failed to suspend");
1046 goto out;
1047 }
1048
1049 out:
1050 g_free(pmutils_path);
1051 }
1052
1053 void qmp_guest_suspend_disk(Error **err)
1054 {
1055 bios_supports_mode("pm-is-supported", "--hibernate", "disk", err);
1056 if (error_is_set(err)) {
1057 return;
1058 }
1059
1060 guest_suspend("pm-hibernate", "disk", err);
1061 }
1062
1063 void qmp_guest_suspend_ram(Error **err)
1064 {
1065 bios_supports_mode("pm-is-supported", "--suspend", "mem", err);
1066 if (error_is_set(err)) {
1067 return;
1068 }
1069
1070 guest_suspend("pm-suspend", "mem", err);
1071 }
1072
1073 void qmp_guest_suspend_hybrid(Error **err)
1074 {
1075 bios_supports_mode("pm-is-supported", "--suspend-hybrid", NULL, err);
1076 if (error_is_set(err)) {
1077 return;
1078 }
1079
1080 guest_suspend("pm-suspend-hybrid", NULL, err);
1081 }
1082
1083 static GuestNetworkInterfaceList *
1084 guest_find_interface(GuestNetworkInterfaceList *head,
1085 const char *name)
1086 {
1087 for (; head; head = head->next) {
1088 if (strcmp(head->value->name, name) == 0) {
1089 break;
1090 }
1091 }
1092
1093 return head;
1094 }
1095
1096 /*
1097 * Build information about guest interfaces
1098 */
1099 GuestNetworkInterfaceList *qmp_guest_network_get_interfaces(Error **errp)
1100 {
1101 GuestNetworkInterfaceList *head = NULL, *cur_item = NULL;
1102 struct ifaddrs *ifap, *ifa;
1103
1104 if (getifaddrs(&ifap) < 0) {
1105 error_setg_errno(errp, errno, "getifaddrs failed");
1106 goto error;
1107 }
1108
1109 for (ifa = ifap; ifa; ifa = ifa->ifa_next) {
1110 GuestNetworkInterfaceList *info;
1111 GuestIpAddressList **address_list = NULL, *address_item = NULL;
1112 char addr4[INET_ADDRSTRLEN];
1113 char addr6[INET6_ADDRSTRLEN];
1114 int sock;
1115 struct ifreq ifr;
1116 unsigned char *mac_addr;
1117 void *p;
1118
1119 g_debug("Processing %s interface", ifa->ifa_name);
1120
1121 info = guest_find_interface(head, ifa->ifa_name);
1122
1123 if (!info) {
1124 info = g_malloc0(sizeof(*info));
1125 info->value = g_malloc0(sizeof(*info->value));
1126 info->value->name = g_strdup(ifa->ifa_name);
1127
1128 if (!cur_item) {
1129 head = cur_item = info;
1130 } else {
1131 cur_item->next = info;
1132 cur_item = info;
1133 }
1134 }
1135
1136 if (!info->value->has_hardware_address &&
1137 ifa->ifa_flags & SIOCGIFHWADDR) {
1138 /* we haven't obtained HW address yet */
1139 sock = socket(PF_INET, SOCK_STREAM, 0);
1140 if (sock == -1) {
1141 error_setg_errno(errp, errno, "failed to create socket");
1142 goto error;
1143 }
1144
1145 memset(&ifr, 0, sizeof(ifr));
1146 pstrcpy(ifr.ifr_name, IF_NAMESIZE, info->value->name);
1147 if (ioctl(sock, SIOCGIFHWADDR, &ifr) == -1) {
1148 error_setg_errno(errp, errno,
1149 "failed to get MAC address of %s",
1150 ifa->ifa_name);
1151 close(sock);
1152 goto error;
1153 }
1154
1155 close(sock);
1156 mac_addr = (unsigned char *) &ifr.ifr_hwaddr.sa_data;
1157
1158 info->value->hardware_address =
1159 g_strdup_printf("%02x:%02x:%02x:%02x:%02x:%02x",
1160 (int) mac_addr[0], (int) mac_addr[1],
1161 (int) mac_addr[2], (int) mac_addr[3],
1162 (int) mac_addr[4], (int) mac_addr[5]);
1163
1164 info->value->has_hardware_address = true;
1165 }
1166
1167 if (ifa->ifa_addr &&
1168 ifa->ifa_addr->sa_family == AF_INET) {
1169 /* interface with IPv4 address */
1170 p = &((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
1171 if (!inet_ntop(AF_INET, p, addr4, sizeof(addr4))) {
1172 error_setg_errno(errp, errno, "inet_ntop failed");
1173 goto error;
1174 }
1175
1176 address_item = g_malloc0(sizeof(*address_item));
1177 address_item->value = g_malloc0(sizeof(*address_item->value));
1178 address_item->value->ip_address = g_strdup(addr4);
1179 address_item->value->ip_address_type = GUEST_IP_ADDRESS_TYPE_IPV4;
1180
1181 if (ifa->ifa_netmask) {
1182 /* Count the number of set bits in netmask.
1183 * This is safe as '1' and '0' cannot be shuffled in netmask. */
1184 p = &((struct sockaddr_in *)ifa->ifa_netmask)->sin_addr;
1185 address_item->value->prefix = ctpop32(((uint32_t *) p)[0]);
1186 }
1187 } else if (ifa->ifa_addr &&
1188 ifa->ifa_addr->sa_family == AF_INET6) {
1189 /* interface with IPv6 address */
1190 p = &((struct sockaddr_in6 *)ifa->ifa_addr)->sin6_addr;
1191 if (!inet_ntop(AF_INET6, p, addr6, sizeof(addr6))) {
1192 error_setg_errno(errp, errno, "inet_ntop failed");
1193 goto error;
1194 }
1195
1196 address_item = g_malloc0(sizeof(*address_item));
1197 address_item->value = g_malloc0(sizeof(*address_item->value));
1198 address_item->value->ip_address = g_strdup(addr6);
1199 address_item->value->ip_address_type = GUEST_IP_ADDRESS_TYPE_IPV6;
1200
1201 if (ifa->ifa_netmask) {
1202 /* Count the number of set bits in netmask.
1203 * This is safe as '1' and '0' cannot be shuffled in netmask. */
1204 p = &((struct sockaddr_in6 *)ifa->ifa_netmask)->sin6_addr;
1205 address_item->value->prefix =
1206 ctpop32(((uint32_t *) p)[0]) +
1207 ctpop32(((uint32_t *) p)[1]) +
1208 ctpop32(((uint32_t *) p)[2]) +
1209 ctpop32(((uint32_t *) p)[3]);
1210 }
1211 }
1212
1213 if (!address_item) {
1214 continue;
1215 }
1216
1217 address_list = &info->value->ip_addresses;
1218
1219 while (*address_list && (*address_list)->next) {
1220 address_list = &(*address_list)->next;
1221 }
1222
1223 if (!*address_list) {
1224 *address_list = address_item;
1225 } else {
1226 (*address_list)->next = address_item;
1227 }
1228
1229 info->value->has_ip_addresses = true;
1230
1231
1232 }
1233
1234 freeifaddrs(ifap);
1235 return head;
1236
1237 error:
1238 freeifaddrs(ifap);
1239 qapi_free_GuestNetworkInterfaceList(head);
1240 return NULL;
1241 }
1242
1243 #define SYSCONF_EXACT(name, err) sysconf_exact((name), #name, (err))
1244
1245 static long sysconf_exact(int name, const char *name_str, Error **err)
1246 {
1247 long ret;
1248
1249 errno = 0;
1250 ret = sysconf(name);
1251 if (ret == -1) {
1252 if (errno == 0) {
1253 error_setg(err, "sysconf(%s): value indefinite", name_str);
1254 } else {
1255 error_setg_errno(err, errno, "sysconf(%s)", name_str);
1256 }
1257 }
1258 return ret;
1259 }
1260
1261 /* Transfer online/offline status between @vcpu and the guest system.
1262 *
1263 * On input either @errp or *@errp must be NULL.
1264 *
1265 * In system-to-@vcpu direction, the following @vcpu fields are accessed:
1266 * - R: vcpu->logical_id
1267 * - W: vcpu->online
1268 * - W: vcpu->can_offline
1269 *
1270 * In @vcpu-to-system direction, the following @vcpu fields are accessed:
1271 * - R: vcpu->logical_id
1272 * - R: vcpu->online
1273 *
1274 * Written members remain unmodified on error.
1275 */
1276 static void transfer_vcpu(GuestLogicalProcessor *vcpu, bool sys2vcpu,
1277 Error **errp)
1278 {
1279 char *dirpath;
1280 int dirfd;
1281
1282 dirpath = g_strdup_printf("/sys/devices/system/cpu/cpu%" PRId64 "/",
1283 vcpu->logical_id);
1284 dirfd = open(dirpath, O_RDONLY | O_DIRECTORY);
1285 if (dirfd == -1) {
1286 error_setg_errno(errp, errno, "open(\"%s\")", dirpath);
1287 } else {
1288 static const char fn[] = "online";
1289 int fd;
1290 int res;
1291
1292 fd = openat(dirfd, fn, sys2vcpu ? O_RDONLY : O_RDWR);
1293 if (fd == -1) {
1294 if (errno != ENOENT) {
1295 error_setg_errno(errp, errno, "open(\"%s/%s\")", dirpath, fn);
1296 } else if (sys2vcpu) {
1297 vcpu->online = true;
1298 vcpu->can_offline = false;
1299 } else if (!vcpu->online) {
1300 error_setg(errp, "logical processor #%" PRId64 " can't be "
1301 "offlined", vcpu->logical_id);
1302 } /* otherwise pretend successful re-onlining */
1303 } else {
1304 unsigned char status;
1305
1306 res = pread(fd, &status, 1, 0);
1307 if (res == -1) {
1308 error_setg_errno(errp, errno, "pread(\"%s/%s\")", dirpath, fn);
1309 } else if (res == 0) {
1310 error_setg(errp, "pread(\"%s/%s\"): unexpected EOF", dirpath,
1311 fn);
1312 } else if (sys2vcpu) {
1313 vcpu->online = (status != '0');
1314 vcpu->can_offline = true;
1315 } else if (vcpu->online != (status != '0')) {
1316 status = '0' + vcpu->online;
1317 if (pwrite(fd, &status, 1, 0) == -1) {
1318 error_setg_errno(errp, errno, "pwrite(\"%s/%s\")", dirpath,
1319 fn);
1320 }
1321 } /* otherwise pretend successful re-(on|off)-lining */
1322
1323 res = close(fd);
1324 g_assert(res == 0);
1325 }
1326
1327 res = close(dirfd);
1328 g_assert(res == 0);
1329 }
1330
1331 g_free(dirpath);
1332 }
1333
1334 GuestLogicalProcessorList *qmp_guest_get_vcpus(Error **errp)
1335 {
1336 int64_t current;
1337 GuestLogicalProcessorList *head, **link;
1338 long sc_max;
1339 Error *local_err = NULL;
1340
1341 current = 0;
1342 head = NULL;
1343 link = &head;
1344 sc_max = SYSCONF_EXACT(_SC_NPROCESSORS_CONF, &local_err);
1345
1346 while (local_err == NULL && current < sc_max) {
1347 GuestLogicalProcessor *vcpu;
1348 GuestLogicalProcessorList *entry;
1349
1350 vcpu = g_malloc0(sizeof *vcpu);
1351 vcpu->logical_id = current++;
1352 vcpu->has_can_offline = true; /* lolspeak ftw */
1353 transfer_vcpu(vcpu, true, &local_err);
1354
1355 entry = g_malloc0(sizeof *entry);
1356 entry->value = vcpu;
1357
1358 *link = entry;
1359 link = &entry->next;
1360 }
1361
1362 if (local_err == NULL) {
1363 /* there's no guest with zero VCPUs */
1364 g_assert(head != NULL);
1365 return head;
1366 }
1367
1368 qapi_free_GuestLogicalProcessorList(head);
1369 error_propagate(errp, local_err);
1370 return NULL;
1371 }
1372
1373 int64_t qmp_guest_set_vcpus(GuestLogicalProcessorList *vcpus, Error **errp)
1374 {
1375 int64_t processed;
1376 Error *local_err = NULL;
1377
1378 processed = 0;
1379 while (vcpus != NULL) {
1380 transfer_vcpu(vcpus->value, false, &local_err);
1381 if (local_err != NULL) {
1382 break;
1383 }
1384 ++processed;
1385 vcpus = vcpus->next;
1386 }
1387
1388 if (local_err != NULL) {
1389 if (processed == 0) {
1390 error_propagate(errp, local_err);
1391 } else {
1392 error_free(local_err);
1393 }
1394 }
1395
1396 return processed;
1397 }
1398
1399 #else /* defined(__linux__) */
1400
1401 void qmp_guest_suspend_disk(Error **err)
1402 {
1403 error_set(err, QERR_UNSUPPORTED);
1404 }
1405
1406 void qmp_guest_suspend_ram(Error **err)
1407 {
1408 error_set(err, QERR_UNSUPPORTED);
1409 }
1410
1411 void qmp_guest_suspend_hybrid(Error **err)
1412 {
1413 error_set(err, QERR_UNSUPPORTED);
1414 }
1415
1416 GuestNetworkInterfaceList *qmp_guest_network_get_interfaces(Error **errp)
1417 {
1418 error_set(errp, QERR_UNSUPPORTED);
1419 return NULL;
1420 }
1421
1422 GuestLogicalProcessorList *qmp_guest_get_vcpus(Error **errp)
1423 {
1424 error_set(errp, QERR_UNSUPPORTED);
1425 return NULL;
1426 }
1427
1428 int64_t qmp_guest_set_vcpus(GuestLogicalProcessorList *vcpus, Error **errp)
1429 {
1430 error_set(errp, QERR_UNSUPPORTED);
1431 return -1;
1432 }
1433
1434 #endif
1435
1436 #if !defined(CONFIG_FSFREEZE)
1437
1438 GuestFsfreezeStatus qmp_guest_fsfreeze_status(Error **err)
1439 {
1440 error_set(err, QERR_UNSUPPORTED);
1441
1442 return 0;
1443 }
1444
1445 int64_t qmp_guest_fsfreeze_freeze(Error **err)
1446 {
1447 error_set(err, QERR_UNSUPPORTED);
1448
1449 return 0;
1450 }
1451
1452 int64_t qmp_guest_fsfreeze_thaw(Error **err)
1453 {
1454 error_set(err, QERR_UNSUPPORTED);
1455
1456 return 0;
1457 }
1458 #endif /* CONFIG_FSFREEZE */
1459
1460 #if !defined(CONFIG_FSTRIM)
1461 void qmp_guest_fstrim(bool has_minimum, int64_t minimum, Error **err)
1462 {
1463 error_set(err, QERR_UNSUPPORTED);
1464 }
1465 #endif
1466
1467 /* register init/cleanup routines for stateful command groups */
1468 void ga_command_state_init(GAState *s, GACommandState *cs)
1469 {
1470 #if defined(CONFIG_FSFREEZE)
1471 ga_command_state_add(cs, NULL, guest_fsfreeze_cleanup);
1472 #endif
1473 ga_command_state_add(cs, guest_file_init, NULL);
1474 }