xref: /openbmc/qemu/block/rbd.c (revision 40daca54)
1 /*
2  * QEMU Block driver for RADOS (Ceph)
3  *
4  * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>,
5  *                         Josh Durgin <josh.durgin@dreamhost.com>
6  *
7  * This work is licensed under the terms of the GNU GPL, version 2.  See
8  * the COPYING file in the top-level directory.
9  *
10  * Contributions after 2012-01-13 are licensed under the terms of the
11  * GNU GPL, version 2 or (at your option) any later version.
12  */
13 
14 #include <inttypes.h>
15 
16 #include "qemu-common.h"
17 #include "qemu/error-report.h"
18 #include "block/block_int.h"
19 
20 #include <rbd/librbd.h>
21 
22 /*
23  * When specifying the image filename use:
24  *
25  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
26  *
27  * poolname must be the name of an existing rados pool.
28  *
29  * devicename is the name of the rbd image.
30  *
31  * Each option given is used to configure rados, and may be any valid
32  * Ceph option, "id", or "conf".
33  *
34  * The "id" option indicates what user we should authenticate as to
35  * the Ceph cluster.  If it is excluded we will use the Ceph default
36  * (normally 'admin').
37  *
38  * The "conf" option specifies a Ceph configuration file to read.  If
39  * it is not specified, we will read from the default Ceph locations
40  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
41  * file, specify conf=/dev/null.
42  *
43  * Configuration values containing :, @, or = can be escaped with a
44  * leading "\".
45  */
46 
47 /* rbd_aio_discard added in 0.1.2 */
48 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
49 #define LIBRBD_SUPPORTS_DISCARD
50 #else
51 #undef LIBRBD_SUPPORTS_DISCARD
52 #endif
53 
54 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
55 
56 #define RBD_MAX_CONF_NAME_SIZE 128
57 #define RBD_MAX_CONF_VAL_SIZE 512
58 #define RBD_MAX_CONF_SIZE 1024
59 #define RBD_MAX_POOL_NAME_SIZE 128
60 #define RBD_MAX_SNAP_NAME_SIZE 128
61 #define RBD_MAX_SNAPS 100
62 
63 typedef enum {
64     RBD_AIO_READ,
65     RBD_AIO_WRITE,
66     RBD_AIO_DISCARD,
67     RBD_AIO_FLUSH
68 } RBDAIOCmd;
69 
70 typedef struct RBDAIOCB {
71     BlockDriverAIOCB common;
72     QEMUBH *bh;
73     int64_t ret;
74     QEMUIOVector *qiov;
75     char *bounce;
76     RBDAIOCmd cmd;
77     int64_t sector_num;
78     int error;
79     struct BDRVRBDState *s;
80     int cancelled;
81     int status;
82 } RBDAIOCB;
83 
84 typedef struct RADOSCB {
85     int rcbid;
86     RBDAIOCB *acb;
87     struct BDRVRBDState *s;
88     int done;
89     int64_t size;
90     char *buf;
91     int64_t ret;
92 } RADOSCB;
93 
94 #define RBD_FD_READ 0
95 #define RBD_FD_WRITE 1
96 
97 typedef struct BDRVRBDState {
98     int fds[2];
99     rados_t cluster;
100     rados_ioctx_t io_ctx;
101     rbd_image_t image;
102     char name[RBD_MAX_IMAGE_NAME_SIZE];
103     char *snap;
104     int event_reader_pos;
105     RADOSCB *event_rcb;
106 } BDRVRBDState;
107 
108 static void rbd_aio_bh_cb(void *opaque);
109 
110 static int qemu_rbd_next_tok(char *dst, int dst_len,
111                              char *src, char delim,
112                              const char *name,
113                              char **p)
114 {
115     int l;
116     char *end;
117 
118     *p = NULL;
119 
120     if (delim != '\0') {
121         for (end = src; *end; ++end) {
122             if (*end == delim) {
123                 break;
124             }
125             if (*end == '\\' && end[1] != '\0') {
126                 end++;
127             }
128         }
129         if (*end == delim) {
130             *p = end + 1;
131             *end = '\0';
132         }
133     }
134     l = strlen(src);
135     if (l >= dst_len) {
136         error_report("%s too long", name);
137         return -EINVAL;
138     } else if (l == 0) {
139         error_report("%s too short", name);
140         return -EINVAL;
141     }
142 
143     pstrcpy(dst, dst_len, src);
144 
145     return 0;
146 }
147 
148 static void qemu_rbd_unescape(char *src)
149 {
150     char *p;
151 
152     for (p = src; *src; ++src, ++p) {
153         if (*src == '\\' && src[1] != '\0') {
154             src++;
155         }
156         *p = *src;
157     }
158     *p = '\0';
159 }
160 
161 static int qemu_rbd_parsename(const char *filename,
162                               char *pool, int pool_len,
163                               char *snap, int snap_len,
164                               char *name, int name_len,
165                               char *conf, int conf_len)
166 {
167     const char *start;
168     char *p, *buf;
169     int ret;
170 
171     if (!strstart(filename, "rbd:", &start)) {
172         return -EINVAL;
173     }
174 
175     buf = g_strdup(start);
176     p = buf;
177     *snap = '\0';
178     *conf = '\0';
179 
180     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
181     if (ret < 0 || !p) {
182         ret = -EINVAL;
183         goto done;
184     }
185     qemu_rbd_unescape(pool);
186 
187     if (strchr(p, '@')) {
188         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
189         if (ret < 0) {
190             goto done;
191         }
192         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
193         qemu_rbd_unescape(snap);
194     } else {
195         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
196     }
197     qemu_rbd_unescape(name);
198     if (ret < 0 || !p) {
199         goto done;
200     }
201 
202     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
203 
204 done:
205     g_free(buf);
206     return ret;
207 }
208 
209 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
210 {
211     const char *p = conf;
212 
213     while (*p) {
214         int len;
215         const char *end = strchr(p, ':');
216 
217         if (end) {
218             len = end - p;
219         } else {
220             len = strlen(p);
221         }
222 
223         if (strncmp(p, "id=", 3) == 0) {
224             len -= 3;
225             strncpy(clientname, p + 3, len);
226             clientname[len] = '\0';
227             return clientname;
228         }
229         if (end == NULL) {
230             break;
231         }
232         p = end + 1;
233     }
234     return NULL;
235 }
236 
237 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
238 {
239     char *p, *buf;
240     char name[RBD_MAX_CONF_NAME_SIZE];
241     char value[RBD_MAX_CONF_VAL_SIZE];
242     int ret = 0;
243 
244     buf = g_strdup(conf);
245     p = buf;
246 
247     while (p) {
248         ret = qemu_rbd_next_tok(name, sizeof(name), p,
249                                 '=', "conf option name", &p);
250         if (ret < 0) {
251             break;
252         }
253         qemu_rbd_unescape(name);
254 
255         if (!p) {
256             error_report("conf option %s has no value", name);
257             ret = -EINVAL;
258             break;
259         }
260 
261         ret = qemu_rbd_next_tok(value, sizeof(value), p,
262                                 ':', "conf option value", &p);
263         if (ret < 0) {
264             break;
265         }
266         qemu_rbd_unescape(value);
267 
268         if (strcmp(name, "conf") == 0) {
269             ret = rados_conf_read_file(cluster, value);
270             if (ret < 0) {
271                 error_report("error reading conf file %s", value);
272                 break;
273             }
274         } else if (strcmp(name, "id") == 0) {
275             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
276         } else {
277             ret = rados_conf_set(cluster, name, value);
278             if (ret < 0) {
279                 error_report("invalid conf option %s", name);
280                 ret = -EINVAL;
281                 break;
282             }
283         }
284     }
285 
286     g_free(buf);
287     return ret;
288 }
289 
290 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
291 {
292     int64_t bytes = 0;
293     int64_t objsize;
294     int obj_order = 0;
295     char pool[RBD_MAX_POOL_NAME_SIZE];
296     char name[RBD_MAX_IMAGE_NAME_SIZE];
297     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
298     char conf[RBD_MAX_CONF_SIZE];
299     char clientname_buf[RBD_MAX_CONF_SIZE];
300     char *clientname;
301     rados_t cluster;
302     rados_ioctx_t io_ctx;
303     int ret;
304 
305     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
306                            snap_buf, sizeof(snap_buf),
307                            name, sizeof(name),
308                            conf, sizeof(conf)) < 0) {
309         return -EINVAL;
310     }
311 
312     /* Read out options */
313     while (options && options->name) {
314         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
315             bytes = options->value.n;
316         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
317             if (options->value.n) {
318                 objsize = options->value.n;
319                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
320                     error_report("obj size needs to be power of 2");
321                     return -EINVAL;
322                 }
323                 if (objsize < 4096) {
324                     error_report("obj size too small");
325                     return -EINVAL;
326                 }
327                 obj_order = ffs(objsize) - 1;
328             }
329         }
330         options++;
331     }
332 
333     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
334     if (rados_create(&cluster, clientname) < 0) {
335         error_report("error initializing");
336         return -EIO;
337     }
338 
339     if (strstr(conf, "conf=") == NULL) {
340         /* try default location, but ignore failure */
341         rados_conf_read_file(cluster, NULL);
342     }
343 
344     if (conf[0] != '\0' &&
345         qemu_rbd_set_conf(cluster, conf) < 0) {
346         error_report("error setting config options");
347         rados_shutdown(cluster);
348         return -EIO;
349     }
350 
351     if (rados_connect(cluster) < 0) {
352         error_report("error connecting");
353         rados_shutdown(cluster);
354         return -EIO;
355     }
356 
357     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
358         error_report("error opening pool %s", pool);
359         rados_shutdown(cluster);
360         return -EIO;
361     }
362 
363     ret = rbd_create(io_ctx, name, bytes, &obj_order);
364     rados_ioctx_destroy(io_ctx);
365     rados_shutdown(cluster);
366 
367     return ret;
368 }
369 
370 /*
371  * This aio completion is being called from qemu_rbd_aio_event_reader()
372  * and runs in qemu context. It schedules a bh, but just in case the aio
373  * was not cancelled before.
374  */
375 static void qemu_rbd_complete_aio(RADOSCB *rcb)
376 {
377     RBDAIOCB *acb = rcb->acb;
378     int64_t r;
379 
380     r = rcb->ret;
381 
382     if (acb->cmd != RBD_AIO_READ) {
383         if (r < 0) {
384             acb->ret = r;
385             acb->error = 1;
386         } else if (!acb->error) {
387             acb->ret = rcb->size;
388         }
389     } else {
390         if (r < 0) {
391             memset(rcb->buf, 0, rcb->size);
392             acb->ret = r;
393             acb->error = 1;
394         } else if (r < rcb->size) {
395             memset(rcb->buf + r, 0, rcb->size - r);
396             if (!acb->error) {
397                 acb->ret = rcb->size;
398             }
399         } else if (!acb->error) {
400             acb->ret = r;
401         }
402     }
403     /* Note that acb->bh can be NULL in case where the aio was cancelled */
404     acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
405     qemu_bh_schedule(acb->bh);
406     g_free(rcb);
407 }
408 
409 /*
410  * aio fd read handler. It runs in the qemu context and calls the
411  * completion handling of completed rados aio operations.
412  */
413 static void qemu_rbd_aio_event_reader(void *opaque)
414 {
415     BDRVRBDState *s = opaque;
416 
417     ssize_t ret;
418 
419     do {
420         char *p = (char *)&s->event_rcb;
421 
422         /* now read the rcb pointer that was sent from a non qemu thread */
423         ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
424                    sizeof(s->event_rcb) - s->event_reader_pos);
425         if (ret > 0) {
426             s->event_reader_pos += ret;
427             if (s->event_reader_pos == sizeof(s->event_rcb)) {
428                 s->event_reader_pos = 0;
429                 qemu_rbd_complete_aio(s->event_rcb);
430             }
431         }
432     } while (ret < 0 && errno == EINTR);
433 }
434 
435 /* TODO Convert to fine grained options */
436 static QemuOptsList runtime_opts = {
437     .name = "rbd",
438     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
439     .desc = {
440         {
441             .name = "filename",
442             .type = QEMU_OPT_STRING,
443             .help = "Specification of the rbd image",
444         },
445         { /* end of list */ }
446     },
447 };
448 
449 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags)
450 {
451     BDRVRBDState *s = bs->opaque;
452     char pool[RBD_MAX_POOL_NAME_SIZE];
453     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
454     char conf[RBD_MAX_CONF_SIZE];
455     char clientname_buf[RBD_MAX_CONF_SIZE];
456     char *clientname;
457     QemuOpts *opts;
458     Error *local_err = NULL;
459     const char *filename;
460     int r;
461 
462     opts = qemu_opts_create_nofail(&runtime_opts);
463     qemu_opts_absorb_qdict(opts, options, &local_err);
464     if (error_is_set(&local_err)) {
465         qerror_report_err(local_err);
466         error_free(local_err);
467         qemu_opts_del(opts);
468         return -EINVAL;
469     }
470 
471     filename = qemu_opt_get(opts, "filename");
472 
473     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
474                            snap_buf, sizeof(snap_buf),
475                            s->name, sizeof(s->name),
476                            conf, sizeof(conf)) < 0) {
477         r = -EINVAL;
478         goto failed_opts;
479     }
480 
481     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
482     r = rados_create(&s->cluster, clientname);
483     if (r < 0) {
484         error_report("error initializing");
485         goto failed_opts;
486     }
487 
488     s->snap = NULL;
489     if (snap_buf[0] != '\0') {
490         s->snap = g_strdup(snap_buf);
491     }
492 
493     /*
494      * Fallback to more conservative semantics if setting cache
495      * options fails. Ignore errors from setting rbd_cache because the
496      * only possible error is that the option does not exist, and
497      * librbd defaults to no caching. If write through caching cannot
498      * be set up, fall back to no caching.
499      */
500     if (flags & BDRV_O_NOCACHE) {
501         rados_conf_set(s->cluster, "rbd_cache", "false");
502     } else {
503         rados_conf_set(s->cluster, "rbd_cache", "true");
504     }
505 
506     if (strstr(conf, "conf=") == NULL) {
507         /* try default location, but ignore failure */
508         rados_conf_read_file(s->cluster, NULL);
509     }
510 
511     if (conf[0] != '\0') {
512         r = qemu_rbd_set_conf(s->cluster, conf);
513         if (r < 0) {
514             error_report("error setting config options");
515             goto failed_shutdown;
516         }
517     }
518 
519     r = rados_connect(s->cluster);
520     if (r < 0) {
521         error_report("error connecting");
522         goto failed_shutdown;
523     }
524 
525     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
526     if (r < 0) {
527         error_report("error opening pool %s", pool);
528         goto failed_shutdown;
529     }
530 
531     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
532     if (r < 0) {
533         error_report("error reading header from %s", s->name);
534         goto failed_open;
535     }
536 
537     bs->read_only = (s->snap != NULL);
538 
539     s->event_reader_pos = 0;
540     r = qemu_pipe(s->fds);
541     if (r < 0) {
542         error_report("error opening eventfd");
543         goto failed;
544     }
545     fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
546     fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
547     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
548                             NULL, s);
549 
550 
551     qemu_opts_del(opts);
552     return 0;
553 
554 failed:
555     rbd_close(s->image);
556 failed_open:
557     rados_ioctx_destroy(s->io_ctx);
558 failed_shutdown:
559     rados_shutdown(s->cluster);
560     g_free(s->snap);
561 failed_opts:
562     qemu_opts_del(opts);
563     return r;
564 }
565 
566 static void qemu_rbd_close(BlockDriverState *bs)
567 {
568     BDRVRBDState *s = bs->opaque;
569 
570     close(s->fds[0]);
571     close(s->fds[1]);
572     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL);
573 
574     rbd_close(s->image);
575     rados_ioctx_destroy(s->io_ctx);
576     g_free(s->snap);
577     rados_shutdown(s->cluster);
578 }
579 
580 /*
581  * Cancel aio. Since we don't reference acb in a non qemu threads,
582  * it is safe to access it here.
583  */
584 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
585 {
586     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
587     acb->cancelled = 1;
588 
589     while (acb->status == -EINPROGRESS) {
590         qemu_aio_wait();
591     }
592 
593     qemu_aio_release(acb);
594 }
595 
596 static const AIOCBInfo rbd_aiocb_info = {
597     .aiocb_size = sizeof(RBDAIOCB),
598     .cancel = qemu_rbd_aio_cancel,
599 };
600 
601 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
602 {
603     int ret = 0;
604     while (1) {
605         fd_set wfd;
606         int fd = s->fds[RBD_FD_WRITE];
607 
608         /* send the op pointer to the qemu thread that is responsible
609            for the aio/op completion. Must do it in a qemu thread context */
610         ret = write(fd, (void *)&rcb, sizeof(rcb));
611         if (ret >= 0) {
612             break;
613         }
614         if (errno == EINTR) {
615             continue;
616         }
617         if (errno != EAGAIN) {
618             break;
619         }
620 
621         FD_ZERO(&wfd);
622         FD_SET(fd, &wfd);
623         do {
624             ret = select(fd + 1, NULL, &wfd, NULL, NULL);
625         } while (ret < 0 && errno == EINTR);
626     }
627 
628     return ret;
629 }
630 
631 /*
632  * This is the callback function for rbd_aio_read and _write
633  *
634  * Note: this function is being called from a non qemu thread so
635  * we need to be careful about what we do here. Generally we only
636  * write to the block notification pipe, and do the rest of the
637  * io completion handling from qemu_rbd_aio_event_reader() which
638  * runs in a qemu context.
639  */
640 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
641 {
642     int ret;
643     rcb->ret = rbd_aio_get_return_value(c);
644     rbd_aio_release(c);
645     ret = qemu_rbd_send_pipe(rcb->s, rcb);
646     if (ret < 0) {
647         error_report("failed writing to acb->s->fds");
648         g_free(rcb);
649     }
650 }
651 
652 /* Callback when all queued rbd_aio requests are complete */
653 
654 static void rbd_aio_bh_cb(void *opaque)
655 {
656     RBDAIOCB *acb = opaque;
657 
658     if (acb->cmd == RBD_AIO_READ) {
659         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
660     }
661     qemu_vfree(acb->bounce);
662     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
663     qemu_bh_delete(acb->bh);
664     acb->bh = NULL;
665     acb->status = 0;
666 
667     if (!acb->cancelled) {
668         qemu_aio_release(acb);
669     }
670 }
671 
672 static int rbd_aio_discard_wrapper(rbd_image_t image,
673                                    uint64_t off,
674                                    uint64_t len,
675                                    rbd_completion_t comp)
676 {
677 #ifdef LIBRBD_SUPPORTS_DISCARD
678     return rbd_aio_discard(image, off, len, comp);
679 #else
680     return -ENOTSUP;
681 #endif
682 }
683 
684 static int rbd_aio_flush_wrapper(rbd_image_t image,
685                                  rbd_completion_t comp)
686 {
687 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
688     return rbd_aio_flush(image, comp);
689 #else
690     return -ENOTSUP;
691 #endif
692 }
693 
694 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
695                                        int64_t sector_num,
696                                        QEMUIOVector *qiov,
697                                        int nb_sectors,
698                                        BlockDriverCompletionFunc *cb,
699                                        void *opaque,
700                                        RBDAIOCmd cmd)
701 {
702     RBDAIOCB *acb;
703     RADOSCB *rcb;
704     rbd_completion_t c;
705     int64_t off, size;
706     char *buf;
707     int r;
708 
709     BDRVRBDState *s = bs->opaque;
710 
711     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
712     acb->cmd = cmd;
713     acb->qiov = qiov;
714     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
715         acb->bounce = NULL;
716     } else {
717         acb->bounce = qemu_blockalign(bs, qiov->size);
718     }
719     acb->ret = 0;
720     acb->error = 0;
721     acb->s = s;
722     acb->cancelled = 0;
723     acb->bh = NULL;
724     acb->status = -EINPROGRESS;
725 
726     if (cmd == RBD_AIO_WRITE) {
727         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
728     }
729 
730     buf = acb->bounce;
731 
732     off = sector_num * BDRV_SECTOR_SIZE;
733     size = nb_sectors * BDRV_SECTOR_SIZE;
734 
735     rcb = g_malloc(sizeof(RADOSCB));
736     rcb->done = 0;
737     rcb->acb = acb;
738     rcb->buf = buf;
739     rcb->s = acb->s;
740     rcb->size = size;
741     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
742     if (r < 0) {
743         goto failed;
744     }
745 
746     switch (cmd) {
747     case RBD_AIO_WRITE:
748         r = rbd_aio_write(s->image, off, size, buf, c);
749         break;
750     case RBD_AIO_READ:
751         r = rbd_aio_read(s->image, off, size, buf, c);
752         break;
753     case RBD_AIO_DISCARD:
754         r = rbd_aio_discard_wrapper(s->image, off, size, c);
755         break;
756     case RBD_AIO_FLUSH:
757         r = rbd_aio_flush_wrapper(s->image, c);
758         break;
759     default:
760         r = -EINVAL;
761     }
762 
763     if (r < 0) {
764         goto failed;
765     }
766 
767     return &acb->common;
768 
769 failed:
770     g_free(rcb);
771     qemu_aio_release(acb);
772     return NULL;
773 }
774 
775 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
776                                             int64_t sector_num,
777                                             QEMUIOVector *qiov,
778                                             int nb_sectors,
779                                             BlockDriverCompletionFunc *cb,
780                                             void *opaque)
781 {
782     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
783                          RBD_AIO_READ);
784 }
785 
786 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
787                                              int64_t sector_num,
788                                              QEMUIOVector *qiov,
789                                              int nb_sectors,
790                                              BlockDriverCompletionFunc *cb,
791                                              void *opaque)
792 {
793     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
794                          RBD_AIO_WRITE);
795 }
796 
797 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
798 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
799                                             BlockDriverCompletionFunc *cb,
800                                             void *opaque)
801 {
802     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
803 }
804 
805 #else
806 
807 static int qemu_rbd_co_flush(BlockDriverState *bs)
808 {
809 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
810     /* rbd_flush added in 0.1.1 */
811     BDRVRBDState *s = bs->opaque;
812     return rbd_flush(s->image);
813 #else
814     return 0;
815 #endif
816 }
817 #endif
818 
819 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
820 {
821     BDRVRBDState *s = bs->opaque;
822     rbd_image_info_t info;
823     int r;
824 
825     r = rbd_stat(s->image, &info, sizeof(info));
826     if (r < 0) {
827         return r;
828     }
829 
830     bdi->cluster_size = info.obj_size;
831     return 0;
832 }
833 
834 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
835 {
836     BDRVRBDState *s = bs->opaque;
837     rbd_image_info_t info;
838     int r;
839 
840     r = rbd_stat(s->image, &info, sizeof(info));
841     if (r < 0) {
842         return r;
843     }
844 
845     return info.size;
846 }
847 
848 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
849 {
850     BDRVRBDState *s = bs->opaque;
851     int r;
852 
853     r = rbd_resize(s->image, offset);
854     if (r < 0) {
855         return r;
856     }
857 
858     return 0;
859 }
860 
861 static int qemu_rbd_snap_create(BlockDriverState *bs,
862                                 QEMUSnapshotInfo *sn_info)
863 {
864     BDRVRBDState *s = bs->opaque;
865     int r;
866 
867     if (sn_info->name[0] == '\0') {
868         return -EINVAL; /* we need a name for rbd snapshots */
869     }
870 
871     /*
872      * rbd snapshots are using the name as the user controlled unique identifier
873      * we can't use the rbd snapid for that purpose, as it can't be set
874      */
875     if (sn_info->id_str[0] != '\0' &&
876         strcmp(sn_info->id_str, sn_info->name) != 0) {
877         return -EINVAL;
878     }
879 
880     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
881         return -ERANGE;
882     }
883 
884     r = rbd_snap_create(s->image, sn_info->name);
885     if (r < 0) {
886         error_report("failed to create snap: %s", strerror(-r));
887         return r;
888     }
889 
890     return 0;
891 }
892 
893 static int qemu_rbd_snap_remove(BlockDriverState *bs,
894                                 const char *snapshot_name)
895 {
896     BDRVRBDState *s = bs->opaque;
897     int r;
898 
899     r = rbd_snap_remove(s->image, snapshot_name);
900     return r;
901 }
902 
903 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
904                                   const char *snapshot_name)
905 {
906     BDRVRBDState *s = bs->opaque;
907     int r;
908 
909     r = rbd_snap_rollback(s->image, snapshot_name);
910     return r;
911 }
912 
913 static int qemu_rbd_snap_list(BlockDriverState *bs,
914                               QEMUSnapshotInfo **psn_tab)
915 {
916     BDRVRBDState *s = bs->opaque;
917     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
918     int i, snap_count;
919     rbd_snap_info_t *snaps;
920     int max_snaps = RBD_MAX_SNAPS;
921 
922     do {
923         snaps = g_malloc(sizeof(*snaps) * max_snaps);
924         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
925         if (snap_count < 0) {
926             g_free(snaps);
927         }
928     } while (snap_count == -ERANGE);
929 
930     if (snap_count <= 0) {
931         goto done;
932     }
933 
934     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
935 
936     for (i = 0; i < snap_count; i++) {
937         const char *snap_name = snaps[i].name;
938 
939         sn_info = sn_tab + i;
940         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
941         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
942 
943         sn_info->vm_state_size = snaps[i].size;
944         sn_info->date_sec = 0;
945         sn_info->date_nsec = 0;
946         sn_info->vm_clock_nsec = 0;
947     }
948     rbd_snap_list_end(snaps);
949 
950  done:
951     *psn_tab = sn_tab;
952     return snap_count;
953 }
954 
955 #ifdef LIBRBD_SUPPORTS_DISCARD
956 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
957                                               int64_t sector_num,
958                                               int nb_sectors,
959                                               BlockDriverCompletionFunc *cb,
960                                               void *opaque)
961 {
962     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
963                          RBD_AIO_DISCARD);
964 }
965 #endif
966 
967 static QEMUOptionParameter qemu_rbd_create_options[] = {
968     {
969      .name = BLOCK_OPT_SIZE,
970      .type = OPT_SIZE,
971      .help = "Virtual disk size"
972     },
973     {
974      .name = BLOCK_OPT_CLUSTER_SIZE,
975      .type = OPT_SIZE,
976      .help = "RBD object size"
977     },
978     {NULL}
979 };
980 
981 static BlockDriver bdrv_rbd = {
982     .format_name        = "rbd",
983     .instance_size      = sizeof(BDRVRBDState),
984     .bdrv_file_open     = qemu_rbd_open,
985     .bdrv_close         = qemu_rbd_close,
986     .bdrv_create        = qemu_rbd_create,
987     .bdrv_has_zero_init = bdrv_has_zero_init_1,
988     .bdrv_get_info      = qemu_rbd_getinfo,
989     .create_options     = qemu_rbd_create_options,
990     .bdrv_getlength     = qemu_rbd_getlength,
991     .bdrv_truncate      = qemu_rbd_truncate,
992     .protocol_name      = "rbd",
993 
994     .bdrv_aio_readv         = qemu_rbd_aio_readv,
995     .bdrv_aio_writev        = qemu_rbd_aio_writev,
996 
997 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
998     .bdrv_aio_flush         = qemu_rbd_aio_flush,
999 #else
1000     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1001 #endif
1002 
1003 #ifdef LIBRBD_SUPPORTS_DISCARD
1004     .bdrv_aio_discard       = qemu_rbd_aio_discard,
1005 #endif
1006 
1007     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1008     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1009     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1010     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1011 };
1012 
1013 static void bdrv_rbd_init(void)
1014 {
1015     bdrv_register(&bdrv_rbd);
1016 }
1017 
1018 block_init(bdrv_rbd_init);
1019