xref: /openbmc/qemu/qga/commands-posix.c (revision 88f62c2b)
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             if (oflag & O_CREAT) {
359                 unlink(path);
360             }
361         }
362     }
363 
364     error_propagate(err, local_err);
365     return NULL;
366 }
367 
368 int64_t qmp_guest_file_open(const char *path, bool has_mode, const char *mode, Error **err)
369 {
370     FILE *fh;
371     Error *local_err = NULL;
372     int fd;
373     int64_t ret = -1, handle;
374 
375     if (!has_mode) {
376         mode = "r";
377     }
378     slog("guest-file-open called, filepath: %s, mode: %s", path, mode);
379     fh = safe_open_or_create(path, mode, &local_err);
380     if (local_err != NULL) {
381         error_propagate(err, local_err);
382         return -1;
383     }
384 
385     /* set fd non-blocking to avoid common use cases (like reading from a
386      * named pipe) from hanging the agent
387      */
388     fd = fileno(fh);
389     ret = fcntl(fd, F_GETFL);
390     ret = fcntl(fd, F_SETFL, ret | O_NONBLOCK);
391     if (ret == -1) {
392         error_setg_errno(err, errno, "failed to make file '%s' non-blocking",
393                          path);
394         fclose(fh);
395         return -1;
396     }
397 
398     handle = guest_file_handle_add(fh, err);
399     if (error_is_set(err)) {
400         fclose(fh);
401         return -1;
402     }
403 
404     slog("guest-file-open, handle: %d", handle);
405     return handle;
406 }
407 
408 void qmp_guest_file_close(int64_t handle, Error **err)
409 {
410     GuestFileHandle *gfh = guest_file_handle_find(handle, err);
411     int ret;
412 
413     slog("guest-file-close called, handle: %ld", handle);
414     if (!gfh) {
415         return;
416     }
417 
418     ret = fclose(gfh->fh);
419     if (ret == EOF) {
420         error_setg_errno(err, errno, "failed to close handle");
421         return;
422     }
423 
424     QTAILQ_REMOVE(&guest_file_state.filehandles, gfh, next);
425     g_free(gfh);
426 }
427 
428 struct GuestFileRead *qmp_guest_file_read(int64_t handle, bool has_count,
429                                           int64_t count, Error **err)
430 {
431     GuestFileHandle *gfh = guest_file_handle_find(handle, err);
432     GuestFileRead *read_data = NULL;
433     guchar *buf;
434     FILE *fh;
435     size_t read_count;
436 
437     if (!gfh) {
438         return NULL;
439     }
440 
441     if (!has_count) {
442         count = QGA_READ_COUNT_DEFAULT;
443     } else if (count < 0) {
444         error_setg(err, "value '%" PRId64 "' is invalid for argument count",
445                    count);
446         return NULL;
447     }
448 
449     fh = gfh->fh;
450     buf = g_malloc0(count+1);
451     read_count = fread(buf, 1, count, fh);
452     if (ferror(fh)) {
453         error_setg_errno(err, errno, "failed to read file");
454         slog("guest-file-read failed, handle: %ld", handle);
455     } else {
456         buf[read_count] = 0;
457         read_data = g_malloc0(sizeof(GuestFileRead));
458         read_data->count = read_count;
459         read_data->eof = feof(fh);
460         if (read_count) {
461             read_data->buf_b64 = g_base64_encode(buf, read_count);
462         }
463     }
464     g_free(buf);
465     clearerr(fh);
466 
467     return read_data;
468 }
469 
470 GuestFileWrite *qmp_guest_file_write(int64_t handle, const char *buf_b64,
471                                      bool has_count, int64_t count, Error **err)
472 {
473     GuestFileWrite *write_data = NULL;
474     guchar *buf;
475     gsize buf_len;
476     int write_count;
477     GuestFileHandle *gfh = guest_file_handle_find(handle, err);
478     FILE *fh;
479 
480     if (!gfh) {
481         return NULL;
482     }
483 
484     fh = gfh->fh;
485     buf = g_base64_decode(buf_b64, &buf_len);
486 
487     if (!has_count) {
488         count = buf_len;
489     } else if (count < 0 || count > buf_len) {
490         error_setg(err, "value '%" PRId64 "' is invalid for argument count",
491                    count);
492         g_free(buf);
493         return NULL;
494     }
495 
496     write_count = fwrite(buf, 1, count, fh);
497     if (ferror(fh)) {
498         error_setg_errno(err, errno, "failed to write to file");
499         slog("guest-file-write failed, handle: %ld", handle);
500     } else {
501         write_data = g_malloc0(sizeof(GuestFileWrite));
502         write_data->count = write_count;
503         write_data->eof = feof(fh);
504     }
505     g_free(buf);
506     clearerr(fh);
507 
508     return write_data;
509 }
510 
511 struct GuestFileSeek *qmp_guest_file_seek(int64_t handle, int64_t offset,
512                                           int64_t whence, Error **err)
513 {
514     GuestFileHandle *gfh = guest_file_handle_find(handle, err);
515     GuestFileSeek *seek_data = NULL;
516     FILE *fh;
517     int ret;
518 
519     if (!gfh) {
520         return NULL;
521     }
522 
523     fh = gfh->fh;
524     ret = fseek(fh, offset, whence);
525     if (ret == -1) {
526         error_setg_errno(err, errno, "failed to seek file");
527     } else {
528         seek_data = g_malloc0(sizeof(GuestFileRead));
529         seek_data->position = ftell(fh);
530         seek_data->eof = feof(fh);
531     }
532     clearerr(fh);
533 
534     return seek_data;
535 }
536 
537 void qmp_guest_file_flush(int64_t handle, Error **err)
538 {
539     GuestFileHandle *gfh = guest_file_handle_find(handle, err);
540     FILE *fh;
541     int ret;
542 
543     if (!gfh) {
544         return;
545     }
546 
547     fh = gfh->fh;
548     ret = fflush(fh);
549     if (ret == EOF) {
550         error_setg_errno(err, errno, "failed to flush file");
551     }
552 }
553 
554 static void guest_file_init(void)
555 {
556     QTAILQ_INIT(&guest_file_state.filehandles);
557 }
558 
559 /* linux-specific implementations. avoid this if at all possible. */
560 #if defined(__linux__)
561 
562 #if defined(CONFIG_FSFREEZE) || defined(CONFIG_FSTRIM)
563 typedef struct FsMount {
564     char *dirname;
565     char *devtype;
566     QTAILQ_ENTRY(FsMount) next;
567 } FsMount;
568 
569 typedef QTAILQ_HEAD(, FsMount) FsMountList;
570 
571 static void free_fs_mount_list(FsMountList *mounts)
572 {
573      FsMount *mount, *temp;
574 
575      if (!mounts) {
576          return;
577      }
578 
579      QTAILQ_FOREACH_SAFE(mount, mounts, next, temp) {
580          QTAILQ_REMOVE(mounts, mount, next);
581          g_free(mount->dirname);
582          g_free(mount->devtype);
583          g_free(mount);
584      }
585 }
586 
587 /*
588  * Walk the mount table and build a list of local file systems
589  */
590 static void build_fs_mount_list(FsMountList *mounts, Error **err)
591 {
592     struct mntent *ment;
593     FsMount *mount;
594     char const *mtab = "/proc/self/mounts";
595     FILE *fp;
596 
597     fp = setmntent(mtab, "r");
598     if (!fp) {
599         error_setg(err, "failed to open mtab file: '%s'", mtab);
600         return;
601     }
602 
603     while ((ment = getmntent(fp))) {
604         /*
605          * An entry which device name doesn't start with a '/' is
606          * either a dummy file system or a network file system.
607          * Add special handling for smbfs and cifs as is done by
608          * coreutils as well.
609          */
610         if ((ment->mnt_fsname[0] != '/') ||
611             (strcmp(ment->mnt_type, "smbfs") == 0) ||
612             (strcmp(ment->mnt_type, "cifs") == 0)) {
613             continue;
614         }
615 
616         mount = g_malloc0(sizeof(FsMount));
617         mount->dirname = g_strdup(ment->mnt_dir);
618         mount->devtype = g_strdup(ment->mnt_type);
619 
620         QTAILQ_INSERT_TAIL(mounts, mount, next);
621     }
622 
623     endmntent(fp);
624 }
625 #endif
626 
627 #if defined(CONFIG_FSFREEZE)
628 
629 typedef enum {
630     FSFREEZE_HOOK_THAW = 0,
631     FSFREEZE_HOOK_FREEZE,
632 } FsfreezeHookArg;
633 
634 const char *fsfreeze_hook_arg_string[] = {
635     "thaw",
636     "freeze",
637 };
638 
639 static void execute_fsfreeze_hook(FsfreezeHookArg arg, Error **err)
640 {
641     int status;
642     pid_t pid;
643     const char *hook;
644     const char *arg_str = fsfreeze_hook_arg_string[arg];
645     Error *local_err = NULL;
646 
647     hook = ga_fsfreeze_hook(ga_state);
648     if (!hook) {
649         return;
650     }
651     if (access(hook, X_OK) != 0) {
652         error_setg_errno(err, errno, "can't access fsfreeze hook '%s'", hook);
653         return;
654     }
655 
656     slog("executing fsfreeze hook with arg '%s'", arg_str);
657     pid = fork();
658     if (pid == 0) {
659         setsid();
660         reopen_fd_to_null(0);
661         reopen_fd_to_null(1);
662         reopen_fd_to_null(2);
663 
664         execle(hook, hook, arg_str, NULL, environ);
665         _exit(EXIT_FAILURE);
666     } else if (pid < 0) {
667         error_setg_errno(err, errno, "failed to create child process");
668         return;
669     }
670 
671     ga_wait_child(pid, &status, &local_err);
672     if (error_is_set(&local_err)) {
673         error_propagate(err, local_err);
674         return;
675     }
676 
677     if (!WIFEXITED(status)) {
678         error_setg(err, "fsfreeze hook has terminated abnormally");
679         return;
680     }
681 
682     status = WEXITSTATUS(status);
683     if (status) {
684         error_setg(err, "fsfreeze hook has failed with status %d", status);
685         return;
686     }
687 }
688 
689 /*
690  * Return status of freeze/thaw
691  */
692 GuestFsfreezeStatus qmp_guest_fsfreeze_status(Error **err)
693 {
694     if (ga_is_frozen(ga_state)) {
695         return GUEST_FSFREEZE_STATUS_FROZEN;
696     }
697 
698     return GUEST_FSFREEZE_STATUS_THAWED;
699 }
700 
701 /*
702  * Walk list of mounted file systems in the guest, and freeze the ones which
703  * are real local file systems.
704  */
705 int64_t qmp_guest_fsfreeze_freeze(Error **err)
706 {
707     int ret = 0, i = 0;
708     FsMountList mounts;
709     struct FsMount *mount;
710     Error *local_err = NULL;
711     int fd;
712 
713     slog("guest-fsfreeze called");
714 
715     execute_fsfreeze_hook(FSFREEZE_HOOK_FREEZE, &local_err);
716     if (error_is_set(&local_err)) {
717         error_propagate(err, local_err);
718         return -1;
719     }
720 
721     QTAILQ_INIT(&mounts);
722     build_fs_mount_list(&mounts, &local_err);
723     if (error_is_set(&local_err)) {
724         error_propagate(err, local_err);
725         return -1;
726     }
727 
728     /* cannot risk guest agent blocking itself on a write in this state */
729     ga_set_frozen(ga_state);
730 
731     QTAILQ_FOREACH(mount, &mounts, next) {
732         fd = qemu_open(mount->dirname, O_RDONLY);
733         if (fd == -1) {
734             error_setg_errno(err, errno, "failed to open %s", mount->dirname);
735             goto error;
736         }
737 
738         /* we try to cull filesytems we know won't work in advance, but other
739          * filesytems may not implement fsfreeze for less obvious reasons.
740          * these will report EOPNOTSUPP. we simply ignore these when tallying
741          * the number of frozen filesystems.
742          *
743          * any other error means a failure to freeze a filesystem we
744          * expect to be freezable, so return an error in those cases
745          * and return system to thawed state.
746          */
747         ret = ioctl(fd, FIFREEZE);
748         if (ret == -1) {
749             if (errno != EOPNOTSUPP) {
750                 error_setg_errno(err, errno, "failed to freeze %s",
751                                  mount->dirname);
752                 close(fd);
753                 goto error;
754             }
755         } else {
756             i++;
757         }
758         close(fd);
759     }
760 
761     free_fs_mount_list(&mounts);
762     return i;
763 
764 error:
765     free_fs_mount_list(&mounts);
766     qmp_guest_fsfreeze_thaw(NULL);
767     return 0;
768 }
769 
770 /*
771  * Walk list of frozen file systems in the guest, and thaw them.
772  */
773 int64_t qmp_guest_fsfreeze_thaw(Error **err)
774 {
775     int ret;
776     FsMountList mounts;
777     FsMount *mount;
778     int fd, i = 0, logged;
779     Error *local_err = NULL;
780 
781     QTAILQ_INIT(&mounts);
782     build_fs_mount_list(&mounts, &local_err);
783     if (error_is_set(&local_err)) {
784         error_propagate(err, local_err);
785         return 0;
786     }
787 
788     QTAILQ_FOREACH(mount, &mounts, next) {
789         logged = false;
790         fd = qemu_open(mount->dirname, O_RDONLY);
791         if (fd == -1) {
792             continue;
793         }
794         /* we have no way of knowing whether a filesystem was actually unfrozen
795          * as a result of a successful call to FITHAW, only that if an error
796          * was returned the filesystem was *not* unfrozen by that particular
797          * call.
798          *
799          * since multiple preceding FIFREEZEs require multiple calls to FITHAW
800          * to unfreeze, continuing issuing FITHAW until an error is returned,
801          * in which case either the filesystem is in an unfreezable state, or,
802          * more likely, it was thawed previously (and remains so afterward).
803          *
804          * also, since the most recent successful call is the one that did
805          * the actual unfreeze, we can use this to provide an accurate count
806          * of the number of filesystems unfrozen by guest-fsfreeze-thaw, which
807          * may * be useful for determining whether a filesystem was unfrozen
808          * during the freeze/thaw phase by a process other than qemu-ga.
809          */
810         do {
811             ret = ioctl(fd, FITHAW);
812             if (ret == 0 && !logged) {
813                 i++;
814                 logged = true;
815             }
816         } while (ret == 0);
817         close(fd);
818     }
819 
820     ga_unset_frozen(ga_state);
821     free_fs_mount_list(&mounts);
822 
823     execute_fsfreeze_hook(FSFREEZE_HOOK_THAW, err);
824 
825     return i;
826 }
827 
828 static void guest_fsfreeze_cleanup(void)
829 {
830     Error *err = NULL;
831 
832     if (ga_is_frozen(ga_state) == GUEST_FSFREEZE_STATUS_FROZEN) {
833         qmp_guest_fsfreeze_thaw(&err);
834         if (err) {
835             slog("failed to clean up frozen filesystems: %s",
836                  error_get_pretty(err));
837             error_free(err);
838         }
839     }
840 }
841 #endif /* CONFIG_FSFREEZE */
842 
843 #if defined(CONFIG_FSTRIM)
844 /*
845  * Walk list of mounted file systems in the guest, and trim them.
846  */
847 void qmp_guest_fstrim(bool has_minimum, int64_t minimum, Error **err)
848 {
849     int ret = 0;
850     FsMountList mounts;
851     struct FsMount *mount;
852     int fd;
853     Error *local_err = NULL;
854     struct fstrim_range r = {
855         .start = 0,
856         .len = -1,
857         .minlen = has_minimum ? minimum : 0,
858     };
859 
860     slog("guest-fstrim called");
861 
862     QTAILQ_INIT(&mounts);
863     build_fs_mount_list(&mounts, &local_err);
864     if (error_is_set(&local_err)) {
865         error_propagate(err, local_err);
866         return;
867     }
868 
869     QTAILQ_FOREACH(mount, &mounts, next) {
870         fd = qemu_open(mount->dirname, O_RDONLY);
871         if (fd == -1) {
872             error_setg_errno(err, errno, "failed to open %s", mount->dirname);
873             goto error;
874         }
875 
876         /* We try to cull filesytems we know won't work in advance, but other
877          * filesytems may not implement fstrim for less obvious reasons.  These
878          * will report EOPNOTSUPP; we simply ignore these errors.  Any other
879          * error means an unexpected error, so return it in those cases.  In
880          * some other cases ENOTTY will be reported (e.g. CD-ROMs).
881          */
882         ret = ioctl(fd, FITRIM, &r);
883         if (ret == -1) {
884             if (errno != ENOTTY && errno != EOPNOTSUPP) {
885                 error_setg_errno(err, errno, "failed to trim %s",
886                                  mount->dirname);
887                 close(fd);
888                 goto error;
889             }
890         }
891         close(fd);
892     }
893 
894 error:
895     free_fs_mount_list(&mounts);
896 }
897 #endif /* CONFIG_FSTRIM */
898 
899 
900 #define LINUX_SYS_STATE_FILE "/sys/power/state"
901 #define SUSPEND_SUPPORTED 0
902 #define SUSPEND_NOT_SUPPORTED 1
903 
904 static void bios_supports_mode(const char *pmutils_bin, const char *pmutils_arg,
905                                const char *sysfile_str, Error **err)
906 {
907     Error *local_err = NULL;
908     char *pmutils_path;
909     pid_t pid;
910     int status;
911 
912     pmutils_path = g_find_program_in_path(pmutils_bin);
913 
914     pid = fork();
915     if (!pid) {
916         char buf[32]; /* hopefully big enough */
917         ssize_t ret;
918         int fd;
919 
920         setsid();
921         reopen_fd_to_null(0);
922         reopen_fd_to_null(1);
923         reopen_fd_to_null(2);
924 
925         if (pmutils_path) {
926             execle(pmutils_path, pmutils_bin, pmutils_arg, NULL, environ);
927         }
928 
929         /*
930          * If we get here either pm-utils is not installed or execle() has
931          * failed. Let's try the manual method if the caller wants it.
932          */
933 
934         if (!sysfile_str) {
935             _exit(SUSPEND_NOT_SUPPORTED);
936         }
937 
938         fd = open(LINUX_SYS_STATE_FILE, O_RDONLY);
939         if (fd < 0) {
940             _exit(SUSPEND_NOT_SUPPORTED);
941         }
942 
943         ret = read(fd, buf, sizeof(buf)-1);
944         if (ret <= 0) {
945             _exit(SUSPEND_NOT_SUPPORTED);
946         }
947         buf[ret] = '\0';
948 
949         if (strstr(buf, sysfile_str)) {
950             _exit(SUSPEND_SUPPORTED);
951         }
952 
953         _exit(SUSPEND_NOT_SUPPORTED);
954     } else if (pid < 0) {
955         error_setg_errno(err, errno, "failed to create child process");
956         goto out;
957     }
958 
959     ga_wait_child(pid, &status, &local_err);
960     if (error_is_set(&local_err)) {
961         error_propagate(err, local_err);
962         goto out;
963     }
964 
965     if (!WIFEXITED(status)) {
966         error_setg(err, "child process has terminated abnormally");
967         goto out;
968     }
969 
970     switch (WEXITSTATUS(status)) {
971     case SUSPEND_SUPPORTED:
972         goto out;
973     case SUSPEND_NOT_SUPPORTED:
974         error_setg(err,
975                    "the requested suspend mode is not supported by the guest");
976         goto out;
977     default:
978         error_setg(err,
979                    "the helper program '%s' returned an unexpected exit status"
980                    " code (%d)", pmutils_path, WEXITSTATUS(status));
981         goto out;
982     }
983 
984 out:
985     g_free(pmutils_path);
986 }
987 
988 static void guest_suspend(const char *pmutils_bin, const char *sysfile_str,
989                           Error **err)
990 {
991     Error *local_err = NULL;
992     char *pmutils_path;
993     pid_t pid;
994     int status;
995 
996     pmutils_path = g_find_program_in_path(pmutils_bin);
997 
998     pid = fork();
999     if (pid == 0) {
1000         /* child */
1001         int fd;
1002 
1003         setsid();
1004         reopen_fd_to_null(0);
1005         reopen_fd_to_null(1);
1006         reopen_fd_to_null(2);
1007 
1008         if (pmutils_path) {
1009             execle(pmutils_path, pmutils_bin, NULL, environ);
1010         }
1011 
1012         /*
1013          * If we get here either pm-utils is not installed or execle() has
1014          * failed. Let's try the manual method if the caller wants it.
1015          */
1016 
1017         if (!sysfile_str) {
1018             _exit(EXIT_FAILURE);
1019         }
1020 
1021         fd = open(LINUX_SYS_STATE_FILE, O_WRONLY);
1022         if (fd < 0) {
1023             _exit(EXIT_FAILURE);
1024         }
1025 
1026         if (write(fd, sysfile_str, strlen(sysfile_str)) < 0) {
1027             _exit(EXIT_FAILURE);
1028         }
1029 
1030         _exit(EXIT_SUCCESS);
1031     } else if (pid < 0) {
1032         error_setg_errno(err, errno, "failed to create child process");
1033         goto out;
1034     }
1035 
1036     ga_wait_child(pid, &status, &local_err);
1037     if (error_is_set(&local_err)) {
1038         error_propagate(err, local_err);
1039         goto out;
1040     }
1041 
1042     if (!WIFEXITED(status)) {
1043         error_setg(err, "child process has terminated abnormally");
1044         goto out;
1045     }
1046 
1047     if (WEXITSTATUS(status)) {
1048         error_setg(err, "child process has failed to suspend");
1049         goto out;
1050     }
1051 
1052 out:
1053     g_free(pmutils_path);
1054 }
1055 
1056 void qmp_guest_suspend_disk(Error **err)
1057 {
1058     bios_supports_mode("pm-is-supported", "--hibernate", "disk", err);
1059     if (error_is_set(err)) {
1060         return;
1061     }
1062 
1063     guest_suspend("pm-hibernate", "disk", err);
1064 }
1065 
1066 void qmp_guest_suspend_ram(Error **err)
1067 {
1068     bios_supports_mode("pm-is-supported", "--suspend", "mem", err);
1069     if (error_is_set(err)) {
1070         return;
1071     }
1072 
1073     guest_suspend("pm-suspend", "mem", err);
1074 }
1075 
1076 void qmp_guest_suspend_hybrid(Error **err)
1077 {
1078     bios_supports_mode("pm-is-supported", "--suspend-hybrid", NULL, err);
1079     if (error_is_set(err)) {
1080         return;
1081     }
1082 
1083     guest_suspend("pm-suspend-hybrid", NULL, err);
1084 }
1085 
1086 static GuestNetworkInterfaceList *
1087 guest_find_interface(GuestNetworkInterfaceList *head,
1088                      const char *name)
1089 {
1090     for (; head; head = head->next) {
1091         if (strcmp(head->value->name, name) == 0) {
1092             break;
1093         }
1094     }
1095 
1096     return head;
1097 }
1098 
1099 /*
1100  * Build information about guest interfaces
1101  */
1102 GuestNetworkInterfaceList *qmp_guest_network_get_interfaces(Error **errp)
1103 {
1104     GuestNetworkInterfaceList *head = NULL, *cur_item = NULL;
1105     struct ifaddrs *ifap, *ifa;
1106 
1107     if (getifaddrs(&ifap) < 0) {
1108         error_setg_errno(errp, errno, "getifaddrs failed");
1109         goto error;
1110     }
1111 
1112     for (ifa = ifap; ifa; ifa = ifa->ifa_next) {
1113         GuestNetworkInterfaceList *info;
1114         GuestIpAddressList **address_list = NULL, *address_item = NULL;
1115         char addr4[INET_ADDRSTRLEN];
1116         char addr6[INET6_ADDRSTRLEN];
1117         int sock;
1118         struct ifreq ifr;
1119         unsigned char *mac_addr;
1120         void *p;
1121 
1122         g_debug("Processing %s interface", ifa->ifa_name);
1123 
1124         info = guest_find_interface(head, ifa->ifa_name);
1125 
1126         if (!info) {
1127             info = g_malloc0(sizeof(*info));
1128             info->value = g_malloc0(sizeof(*info->value));
1129             info->value->name = g_strdup(ifa->ifa_name);
1130 
1131             if (!cur_item) {
1132                 head = cur_item = info;
1133             } else {
1134                 cur_item->next = info;
1135                 cur_item = info;
1136             }
1137         }
1138 
1139         if (!info->value->has_hardware_address &&
1140             ifa->ifa_flags & SIOCGIFHWADDR) {
1141             /* we haven't obtained HW address yet */
1142             sock = socket(PF_INET, SOCK_STREAM, 0);
1143             if (sock == -1) {
1144                 error_setg_errno(errp, errno, "failed to create socket");
1145                 goto error;
1146             }
1147 
1148             memset(&ifr, 0, sizeof(ifr));
1149             pstrcpy(ifr.ifr_name, IF_NAMESIZE, info->value->name);
1150             if (ioctl(sock, SIOCGIFHWADDR, &ifr) == -1) {
1151                 error_setg_errno(errp, errno,
1152                                  "failed to get MAC address of %s",
1153                                  ifa->ifa_name);
1154                 close(sock);
1155                 goto error;
1156             }
1157 
1158             close(sock);
1159             mac_addr = (unsigned char *) &ifr.ifr_hwaddr.sa_data;
1160 
1161             info->value->hardware_address =
1162                 g_strdup_printf("%02x:%02x:%02x:%02x:%02x:%02x",
1163                                 (int) mac_addr[0], (int) mac_addr[1],
1164                                 (int) mac_addr[2], (int) mac_addr[3],
1165                                 (int) mac_addr[4], (int) mac_addr[5]);
1166 
1167             info->value->has_hardware_address = true;
1168         }
1169 
1170         if (ifa->ifa_addr &&
1171             ifa->ifa_addr->sa_family == AF_INET) {
1172             /* interface with IPv4 address */
1173             p = &((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
1174             if (!inet_ntop(AF_INET, p, addr4, sizeof(addr4))) {
1175                 error_setg_errno(errp, errno, "inet_ntop failed");
1176                 goto error;
1177             }
1178 
1179             address_item = g_malloc0(sizeof(*address_item));
1180             address_item->value = g_malloc0(sizeof(*address_item->value));
1181             address_item->value->ip_address = g_strdup(addr4);
1182             address_item->value->ip_address_type = GUEST_IP_ADDRESS_TYPE_IPV4;
1183 
1184             if (ifa->ifa_netmask) {
1185                 /* Count the number of set bits in netmask.
1186                  * This is safe as '1' and '0' cannot be shuffled in netmask. */
1187                 p = &((struct sockaddr_in *)ifa->ifa_netmask)->sin_addr;
1188                 address_item->value->prefix = ctpop32(((uint32_t *) p)[0]);
1189             }
1190         } else if (ifa->ifa_addr &&
1191                    ifa->ifa_addr->sa_family == AF_INET6) {
1192             /* interface with IPv6 address */
1193             p = &((struct sockaddr_in6 *)ifa->ifa_addr)->sin6_addr;
1194             if (!inet_ntop(AF_INET6, p, addr6, sizeof(addr6))) {
1195                 error_setg_errno(errp, errno, "inet_ntop failed");
1196                 goto error;
1197             }
1198 
1199             address_item = g_malloc0(sizeof(*address_item));
1200             address_item->value = g_malloc0(sizeof(*address_item->value));
1201             address_item->value->ip_address = g_strdup(addr6);
1202             address_item->value->ip_address_type = GUEST_IP_ADDRESS_TYPE_IPV6;
1203 
1204             if (ifa->ifa_netmask) {
1205                 /* Count the number of set bits in netmask.
1206                  * This is safe as '1' and '0' cannot be shuffled in netmask. */
1207                 p = &((struct sockaddr_in6 *)ifa->ifa_netmask)->sin6_addr;
1208                 address_item->value->prefix =
1209                     ctpop32(((uint32_t *) p)[0]) +
1210                     ctpop32(((uint32_t *) p)[1]) +
1211                     ctpop32(((uint32_t *) p)[2]) +
1212                     ctpop32(((uint32_t *) p)[3]);
1213             }
1214         }
1215 
1216         if (!address_item) {
1217             continue;
1218         }
1219 
1220         address_list = &info->value->ip_addresses;
1221 
1222         while (*address_list && (*address_list)->next) {
1223             address_list = &(*address_list)->next;
1224         }
1225 
1226         if (!*address_list) {
1227             *address_list = address_item;
1228         } else {
1229             (*address_list)->next = address_item;
1230         }
1231 
1232         info->value->has_ip_addresses = true;
1233 
1234 
1235     }
1236 
1237     freeifaddrs(ifap);
1238     return head;
1239 
1240 error:
1241     freeifaddrs(ifap);
1242     qapi_free_GuestNetworkInterfaceList(head);
1243     return NULL;
1244 }
1245 
1246 #define SYSCONF_EXACT(name, err) sysconf_exact((name), #name, (err))
1247 
1248 static long sysconf_exact(int name, const char *name_str, Error **err)
1249 {
1250     long ret;
1251 
1252     errno = 0;
1253     ret = sysconf(name);
1254     if (ret == -1) {
1255         if (errno == 0) {
1256             error_setg(err, "sysconf(%s): value indefinite", name_str);
1257         } else {
1258             error_setg_errno(err, errno, "sysconf(%s)", name_str);
1259         }
1260     }
1261     return ret;
1262 }
1263 
1264 /* Transfer online/offline status between @vcpu and the guest system.
1265  *
1266  * On input either @errp or *@errp must be NULL.
1267  *
1268  * In system-to-@vcpu direction, the following @vcpu fields are accessed:
1269  * - R: vcpu->logical_id
1270  * - W: vcpu->online
1271  * - W: vcpu->can_offline
1272  *
1273  * In @vcpu-to-system direction, the following @vcpu fields are accessed:
1274  * - R: vcpu->logical_id
1275  * - R: vcpu->online
1276  *
1277  * Written members remain unmodified on error.
1278  */
1279 static void transfer_vcpu(GuestLogicalProcessor *vcpu, bool sys2vcpu,
1280                           Error **errp)
1281 {
1282     char *dirpath;
1283     int dirfd;
1284 
1285     dirpath = g_strdup_printf("/sys/devices/system/cpu/cpu%" PRId64 "/",
1286                               vcpu->logical_id);
1287     dirfd = open(dirpath, O_RDONLY | O_DIRECTORY);
1288     if (dirfd == -1) {
1289         error_setg_errno(errp, errno, "open(\"%s\")", dirpath);
1290     } else {
1291         static const char fn[] = "online";
1292         int fd;
1293         int res;
1294 
1295         fd = openat(dirfd, fn, sys2vcpu ? O_RDONLY : O_RDWR);
1296         if (fd == -1) {
1297             if (errno != ENOENT) {
1298                 error_setg_errno(errp, errno, "open(\"%s/%s\")", dirpath, fn);
1299             } else if (sys2vcpu) {
1300                 vcpu->online = true;
1301                 vcpu->can_offline = false;
1302             } else if (!vcpu->online) {
1303                 error_setg(errp, "logical processor #%" PRId64 " can't be "
1304                            "offlined", vcpu->logical_id);
1305             } /* otherwise pretend successful re-onlining */
1306         } else {
1307             unsigned char status;
1308 
1309             res = pread(fd, &status, 1, 0);
1310             if (res == -1) {
1311                 error_setg_errno(errp, errno, "pread(\"%s/%s\")", dirpath, fn);
1312             } else if (res == 0) {
1313                 error_setg(errp, "pread(\"%s/%s\"): unexpected EOF", dirpath,
1314                            fn);
1315             } else if (sys2vcpu) {
1316                 vcpu->online = (status != '0');
1317                 vcpu->can_offline = true;
1318             } else if (vcpu->online != (status != '0')) {
1319                 status = '0' + vcpu->online;
1320                 if (pwrite(fd, &status, 1, 0) == -1) {
1321                     error_setg_errno(errp, errno, "pwrite(\"%s/%s\")", dirpath,
1322                                      fn);
1323                 }
1324             } /* otherwise pretend successful re-(on|off)-lining */
1325 
1326             res = close(fd);
1327             g_assert(res == 0);
1328         }
1329 
1330         res = close(dirfd);
1331         g_assert(res == 0);
1332     }
1333 
1334     g_free(dirpath);
1335 }
1336 
1337 GuestLogicalProcessorList *qmp_guest_get_vcpus(Error **errp)
1338 {
1339     int64_t current;
1340     GuestLogicalProcessorList *head, **link;
1341     long sc_max;
1342     Error *local_err = NULL;
1343 
1344     current = 0;
1345     head = NULL;
1346     link = &head;
1347     sc_max = SYSCONF_EXACT(_SC_NPROCESSORS_CONF, &local_err);
1348 
1349     while (local_err == NULL && current < sc_max) {
1350         GuestLogicalProcessor *vcpu;
1351         GuestLogicalProcessorList *entry;
1352 
1353         vcpu = g_malloc0(sizeof *vcpu);
1354         vcpu->logical_id = current++;
1355         vcpu->has_can_offline = true; /* lolspeak ftw */
1356         transfer_vcpu(vcpu, true, &local_err);
1357 
1358         entry = g_malloc0(sizeof *entry);
1359         entry->value = vcpu;
1360 
1361         *link = entry;
1362         link = &entry->next;
1363     }
1364 
1365     if (local_err == NULL) {
1366         /* there's no guest with zero VCPUs */
1367         g_assert(head != NULL);
1368         return head;
1369     }
1370 
1371     qapi_free_GuestLogicalProcessorList(head);
1372     error_propagate(errp, local_err);
1373     return NULL;
1374 }
1375 
1376 int64_t qmp_guest_set_vcpus(GuestLogicalProcessorList *vcpus, Error **errp)
1377 {
1378     int64_t processed;
1379     Error *local_err = NULL;
1380 
1381     processed = 0;
1382     while (vcpus != NULL) {
1383         transfer_vcpu(vcpus->value, false, &local_err);
1384         if (local_err != NULL) {
1385             break;
1386         }
1387         ++processed;
1388         vcpus = vcpus->next;
1389     }
1390 
1391     if (local_err != NULL) {
1392         if (processed == 0) {
1393             error_propagate(errp, local_err);
1394         } else {
1395             error_free(local_err);
1396         }
1397     }
1398 
1399     return processed;
1400 }
1401 
1402 #else /* defined(__linux__) */
1403 
1404 void qmp_guest_suspend_disk(Error **err)
1405 {
1406     error_set(err, QERR_UNSUPPORTED);
1407 }
1408 
1409 void qmp_guest_suspend_ram(Error **err)
1410 {
1411     error_set(err, QERR_UNSUPPORTED);
1412 }
1413 
1414 void qmp_guest_suspend_hybrid(Error **err)
1415 {
1416     error_set(err, QERR_UNSUPPORTED);
1417 }
1418 
1419 GuestNetworkInterfaceList *qmp_guest_network_get_interfaces(Error **errp)
1420 {
1421     error_set(errp, QERR_UNSUPPORTED);
1422     return NULL;
1423 }
1424 
1425 GuestLogicalProcessorList *qmp_guest_get_vcpus(Error **errp)
1426 {
1427     error_set(errp, QERR_UNSUPPORTED);
1428     return NULL;
1429 }
1430 
1431 int64_t qmp_guest_set_vcpus(GuestLogicalProcessorList *vcpus, Error **errp)
1432 {
1433     error_set(errp, QERR_UNSUPPORTED);
1434     return -1;
1435 }
1436 
1437 #endif
1438 
1439 #if !defined(CONFIG_FSFREEZE)
1440 
1441 GuestFsfreezeStatus qmp_guest_fsfreeze_status(Error **err)
1442 {
1443     error_set(err, QERR_UNSUPPORTED);
1444 
1445     return 0;
1446 }
1447 
1448 int64_t qmp_guest_fsfreeze_freeze(Error **err)
1449 {
1450     error_set(err, QERR_UNSUPPORTED);
1451 
1452     return 0;
1453 }
1454 
1455 int64_t qmp_guest_fsfreeze_thaw(Error **err)
1456 {
1457     error_set(err, QERR_UNSUPPORTED);
1458 
1459     return 0;
1460 }
1461 #endif /* CONFIG_FSFREEZE */
1462 
1463 #if !defined(CONFIG_FSTRIM)
1464 void qmp_guest_fstrim(bool has_minimum, int64_t minimum, Error **err)
1465 {
1466     error_set(err, QERR_UNSUPPORTED);
1467 }
1468 #endif
1469 
1470 /* register init/cleanup routines for stateful command groups */
1471 void ga_command_state_init(GAState *s, GACommandState *cs)
1472 {
1473 #if defined(CONFIG_FSFREEZE)
1474     ga_command_state_add(cs, NULL, guest_fsfreeze_cleanup);
1475 #endif
1476     ga_command_state_add(cs, guest_file_init, NULL);
1477 }
1478