xref: /openbmc/qemu/block/rbd.c (revision e1fe50dc)
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     int qemu_aio_count;
104     char *snap;
105     int event_reader_pos;
106     RADOSCB *event_rcb;
107 } BDRVRBDState;
108 
109 static void rbd_aio_bh_cb(void *opaque);
110 
111 static int qemu_rbd_next_tok(char *dst, int dst_len,
112                              char *src, char delim,
113                              const char *name,
114                              char **p)
115 {
116     int l;
117     char *end;
118 
119     *p = NULL;
120 
121     if (delim != '\0') {
122         for (end = src; *end; ++end) {
123             if (*end == delim) {
124                 break;
125             }
126             if (*end == '\\' && end[1] != '\0') {
127                 end++;
128             }
129         }
130         if (*end == delim) {
131             *p = end + 1;
132             *end = '\0';
133         }
134     }
135     l = strlen(src);
136     if (l >= dst_len) {
137         error_report("%s too long", name);
138         return -EINVAL;
139     } else if (l == 0) {
140         error_report("%s too short", name);
141         return -EINVAL;
142     }
143 
144     pstrcpy(dst, dst_len, src);
145 
146     return 0;
147 }
148 
149 static void qemu_rbd_unescape(char *src)
150 {
151     char *p;
152 
153     for (p = src; *src; ++src, ++p) {
154         if (*src == '\\' && src[1] != '\0') {
155             src++;
156         }
157         *p = *src;
158     }
159     *p = '\0';
160 }
161 
162 static int qemu_rbd_parsename(const char *filename,
163                               char *pool, int pool_len,
164                               char *snap, int snap_len,
165                               char *name, int name_len,
166                               char *conf, int conf_len)
167 {
168     const char *start;
169     char *p, *buf;
170     int ret;
171 
172     if (!strstart(filename, "rbd:", &start)) {
173         return -EINVAL;
174     }
175 
176     buf = g_strdup(start);
177     p = buf;
178     *snap = '\0';
179     *conf = '\0';
180 
181     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
182     if (ret < 0 || !p) {
183         ret = -EINVAL;
184         goto done;
185     }
186     qemu_rbd_unescape(pool);
187 
188     if (strchr(p, '@')) {
189         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
190         if (ret < 0) {
191             goto done;
192         }
193         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
194         qemu_rbd_unescape(snap);
195     } else {
196         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
197     }
198     qemu_rbd_unescape(name);
199     if (ret < 0 || !p) {
200         goto done;
201     }
202 
203     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
204 
205 done:
206     g_free(buf);
207     return ret;
208 }
209 
210 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
211 {
212     const char *p = conf;
213 
214     while (*p) {
215         int len;
216         const char *end = strchr(p, ':');
217 
218         if (end) {
219             len = end - p;
220         } else {
221             len = strlen(p);
222         }
223 
224         if (strncmp(p, "id=", 3) == 0) {
225             len -= 3;
226             strncpy(clientname, p + 3, len);
227             clientname[len] = '\0';
228             return clientname;
229         }
230         if (end == NULL) {
231             break;
232         }
233         p = end + 1;
234     }
235     return NULL;
236 }
237 
238 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
239 {
240     char *p, *buf;
241     char name[RBD_MAX_CONF_NAME_SIZE];
242     char value[RBD_MAX_CONF_VAL_SIZE];
243     int ret = 0;
244 
245     buf = g_strdup(conf);
246     p = buf;
247 
248     while (p) {
249         ret = qemu_rbd_next_tok(name, sizeof(name), p,
250                                 '=', "conf option name", &p);
251         if (ret < 0) {
252             break;
253         }
254         qemu_rbd_unescape(name);
255 
256         if (!p) {
257             error_report("conf option %s has no value", name);
258             ret = -EINVAL;
259             break;
260         }
261 
262         ret = qemu_rbd_next_tok(value, sizeof(value), p,
263                                 ':', "conf option value", &p);
264         if (ret < 0) {
265             break;
266         }
267         qemu_rbd_unescape(value);
268 
269         if (strcmp(name, "conf") == 0) {
270             ret = rados_conf_read_file(cluster, value);
271             if (ret < 0) {
272                 error_report("error reading conf file %s", value);
273                 break;
274             }
275         } else if (strcmp(name, "id") == 0) {
276             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
277         } else {
278             ret = rados_conf_set(cluster, name, value);
279             if (ret < 0) {
280                 error_report("invalid conf option %s", name);
281                 ret = -EINVAL;
282                 break;
283             }
284         }
285     }
286 
287     g_free(buf);
288     return ret;
289 }
290 
291 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
292 {
293     int64_t bytes = 0;
294     int64_t objsize;
295     int obj_order = 0;
296     char pool[RBD_MAX_POOL_NAME_SIZE];
297     char name[RBD_MAX_IMAGE_NAME_SIZE];
298     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
299     char conf[RBD_MAX_CONF_SIZE];
300     char clientname_buf[RBD_MAX_CONF_SIZE];
301     char *clientname;
302     rados_t cluster;
303     rados_ioctx_t io_ctx;
304     int ret;
305 
306     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
307                            snap_buf, sizeof(snap_buf),
308                            name, sizeof(name),
309                            conf, sizeof(conf)) < 0) {
310         return -EINVAL;
311     }
312 
313     /* Read out options */
314     while (options && options->name) {
315         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
316             bytes = options->value.n;
317         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
318             if (options->value.n) {
319                 objsize = options->value.n;
320                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
321                     error_report("obj size needs to be power of 2");
322                     return -EINVAL;
323                 }
324                 if (objsize < 4096) {
325                     error_report("obj size too small");
326                     return -EINVAL;
327                 }
328                 obj_order = ffs(objsize) - 1;
329             }
330         }
331         options++;
332     }
333 
334     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
335     if (rados_create(&cluster, clientname) < 0) {
336         error_report("error initializing");
337         return -EIO;
338     }
339 
340     if (strstr(conf, "conf=") == NULL) {
341         /* try default location, but ignore failure */
342         rados_conf_read_file(cluster, NULL);
343     }
344 
345     if (conf[0] != '\0' &&
346         qemu_rbd_set_conf(cluster, conf) < 0) {
347         error_report("error setting config options");
348         rados_shutdown(cluster);
349         return -EIO;
350     }
351 
352     if (rados_connect(cluster) < 0) {
353         error_report("error connecting");
354         rados_shutdown(cluster);
355         return -EIO;
356     }
357 
358     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
359         error_report("error opening pool %s", pool);
360         rados_shutdown(cluster);
361         return -EIO;
362     }
363 
364     ret = rbd_create(io_ctx, name, bytes, &obj_order);
365     rados_ioctx_destroy(io_ctx);
366     rados_shutdown(cluster);
367 
368     return ret;
369 }
370 
371 /*
372  * This aio completion is being called from qemu_rbd_aio_event_reader()
373  * and runs in qemu context. It schedules a bh, but just in case the aio
374  * was not cancelled before.
375  */
376 static void qemu_rbd_complete_aio(RADOSCB *rcb)
377 {
378     RBDAIOCB *acb = rcb->acb;
379     int64_t r;
380 
381     r = rcb->ret;
382 
383     if (acb->cmd != RBD_AIO_READ) {
384         if (r < 0) {
385             acb->ret = r;
386             acb->error = 1;
387         } else if (!acb->error) {
388             acb->ret = rcb->size;
389         }
390     } else {
391         if (r < 0) {
392             memset(rcb->buf, 0, rcb->size);
393             acb->ret = r;
394             acb->error = 1;
395         } else if (r < rcb->size) {
396             memset(rcb->buf + r, 0, rcb->size - r);
397             if (!acb->error) {
398                 acb->ret = rcb->size;
399             }
400         } else if (!acb->error) {
401             acb->ret = r;
402         }
403     }
404     /* Note that acb->bh can be NULL in case where the aio was cancelled */
405     acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
406     qemu_bh_schedule(acb->bh);
407     g_free(rcb);
408 }
409 
410 /*
411  * aio fd read handler. It runs in the qemu context and calls the
412  * completion handling of completed rados aio operations.
413  */
414 static void qemu_rbd_aio_event_reader(void *opaque)
415 {
416     BDRVRBDState *s = opaque;
417 
418     ssize_t ret;
419 
420     do {
421         char *p = (char *)&s->event_rcb;
422 
423         /* now read the rcb pointer that was sent from a non qemu thread */
424         ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
425                    sizeof(s->event_rcb) - s->event_reader_pos);
426         if (ret > 0) {
427             s->event_reader_pos += ret;
428             if (s->event_reader_pos == sizeof(s->event_rcb)) {
429                 s->event_reader_pos = 0;
430                 qemu_rbd_complete_aio(s->event_rcb);
431                 s->qemu_aio_count--;
432             }
433         }
434     } while (ret < 0 && errno == EINTR);
435 }
436 
437 static int qemu_rbd_aio_flush_cb(void *opaque)
438 {
439     BDRVRBDState *s = opaque;
440 
441     return (s->qemu_aio_count > 0);
442 }
443 
444 static int qemu_rbd_open(BlockDriverState *bs, const char *filename,
445                          QDict *options, int flags)
446 {
447     BDRVRBDState *s = bs->opaque;
448     char pool[RBD_MAX_POOL_NAME_SIZE];
449     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
450     char conf[RBD_MAX_CONF_SIZE];
451     char clientname_buf[RBD_MAX_CONF_SIZE];
452     char *clientname;
453     int r;
454 
455     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
456                            snap_buf, sizeof(snap_buf),
457                            s->name, sizeof(s->name),
458                            conf, sizeof(conf)) < 0) {
459         return -EINVAL;
460     }
461 
462     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
463     r = rados_create(&s->cluster, clientname);
464     if (r < 0) {
465         error_report("error initializing");
466         return r;
467     }
468 
469     s->snap = NULL;
470     if (snap_buf[0] != '\0') {
471         s->snap = g_strdup(snap_buf);
472     }
473 
474     /*
475      * Fallback to more conservative semantics if setting cache
476      * options fails. Ignore errors from setting rbd_cache because the
477      * only possible error is that the option does not exist, and
478      * librbd defaults to no caching. If write through caching cannot
479      * be set up, fall back to no caching.
480      */
481     if (flags & BDRV_O_NOCACHE) {
482         rados_conf_set(s->cluster, "rbd_cache", "false");
483     } else {
484         rados_conf_set(s->cluster, "rbd_cache", "true");
485     }
486 
487     if (strstr(conf, "conf=") == NULL) {
488         /* try default location, but ignore failure */
489         rados_conf_read_file(s->cluster, NULL);
490     }
491 
492     if (conf[0] != '\0') {
493         r = qemu_rbd_set_conf(s->cluster, conf);
494         if (r < 0) {
495             error_report("error setting config options");
496             goto failed_shutdown;
497         }
498     }
499 
500     r = rados_connect(s->cluster);
501     if (r < 0) {
502         error_report("error connecting");
503         goto failed_shutdown;
504     }
505 
506     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
507     if (r < 0) {
508         error_report("error opening pool %s", pool);
509         goto failed_shutdown;
510     }
511 
512     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
513     if (r < 0) {
514         error_report("error reading header from %s", s->name);
515         goto failed_open;
516     }
517 
518     bs->read_only = (s->snap != NULL);
519 
520     s->event_reader_pos = 0;
521     r = qemu_pipe(s->fds);
522     if (r < 0) {
523         error_report("error opening eventfd");
524         goto failed;
525     }
526     fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
527     fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
528     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
529                             NULL, qemu_rbd_aio_flush_cb, s);
530 
531 
532     return 0;
533 
534 failed:
535     rbd_close(s->image);
536 failed_open:
537     rados_ioctx_destroy(s->io_ctx);
538 failed_shutdown:
539     rados_shutdown(s->cluster);
540     g_free(s->snap);
541     return r;
542 }
543 
544 static void qemu_rbd_close(BlockDriverState *bs)
545 {
546     BDRVRBDState *s = bs->opaque;
547 
548     close(s->fds[0]);
549     close(s->fds[1]);
550     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL);
551 
552     rbd_close(s->image);
553     rados_ioctx_destroy(s->io_ctx);
554     g_free(s->snap);
555     rados_shutdown(s->cluster);
556 }
557 
558 /*
559  * Cancel aio. Since we don't reference acb in a non qemu threads,
560  * it is safe to access it here.
561  */
562 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
563 {
564     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
565     acb->cancelled = 1;
566 
567     while (acb->status == -EINPROGRESS) {
568         qemu_aio_wait();
569     }
570 
571     qemu_aio_release(acb);
572 }
573 
574 static const AIOCBInfo rbd_aiocb_info = {
575     .aiocb_size = sizeof(RBDAIOCB),
576     .cancel = qemu_rbd_aio_cancel,
577 };
578 
579 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
580 {
581     int ret = 0;
582     while (1) {
583         fd_set wfd;
584         int fd = s->fds[RBD_FD_WRITE];
585 
586         /* send the op pointer to the qemu thread that is responsible
587            for the aio/op completion. Must do it in a qemu thread context */
588         ret = write(fd, (void *)&rcb, sizeof(rcb));
589         if (ret >= 0) {
590             break;
591         }
592         if (errno == EINTR) {
593             continue;
594         }
595         if (errno != EAGAIN) {
596             break;
597         }
598 
599         FD_ZERO(&wfd);
600         FD_SET(fd, &wfd);
601         do {
602             ret = select(fd + 1, NULL, &wfd, NULL, NULL);
603         } while (ret < 0 && errno == EINTR);
604     }
605 
606     return ret;
607 }
608 
609 /*
610  * This is the callback function for rbd_aio_read and _write
611  *
612  * Note: this function is being called from a non qemu thread so
613  * we need to be careful about what we do here. Generally we only
614  * write to the block notification pipe, and do the rest of the
615  * io completion handling from qemu_rbd_aio_event_reader() which
616  * runs in a qemu context.
617  */
618 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
619 {
620     int ret;
621     rcb->ret = rbd_aio_get_return_value(c);
622     rbd_aio_release(c);
623     ret = qemu_rbd_send_pipe(rcb->s, rcb);
624     if (ret < 0) {
625         error_report("failed writing to acb->s->fds");
626         g_free(rcb);
627     }
628 }
629 
630 /* Callback when all queued rbd_aio requests are complete */
631 
632 static void rbd_aio_bh_cb(void *opaque)
633 {
634     RBDAIOCB *acb = opaque;
635 
636     if (acb->cmd == RBD_AIO_READ) {
637         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
638     }
639     qemu_vfree(acb->bounce);
640     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
641     qemu_bh_delete(acb->bh);
642     acb->bh = NULL;
643     acb->status = 0;
644 
645     if (!acb->cancelled) {
646         qemu_aio_release(acb);
647     }
648 }
649 
650 static int rbd_aio_discard_wrapper(rbd_image_t image,
651                                    uint64_t off,
652                                    uint64_t len,
653                                    rbd_completion_t comp)
654 {
655 #ifdef LIBRBD_SUPPORTS_DISCARD
656     return rbd_aio_discard(image, off, len, comp);
657 #else
658     return -ENOTSUP;
659 #endif
660 }
661 
662 static int rbd_aio_flush_wrapper(rbd_image_t image,
663                                  rbd_completion_t comp)
664 {
665 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
666     return rbd_aio_flush(image, comp);
667 #else
668     return -ENOTSUP;
669 #endif
670 }
671 
672 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
673                                        int64_t sector_num,
674                                        QEMUIOVector *qiov,
675                                        int nb_sectors,
676                                        BlockDriverCompletionFunc *cb,
677                                        void *opaque,
678                                        RBDAIOCmd cmd)
679 {
680     RBDAIOCB *acb;
681     RADOSCB *rcb;
682     rbd_completion_t c;
683     int64_t off, size;
684     char *buf;
685     int r;
686 
687     BDRVRBDState *s = bs->opaque;
688 
689     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
690     acb->cmd = cmd;
691     acb->qiov = qiov;
692     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
693         acb->bounce = NULL;
694     } else {
695         acb->bounce = qemu_blockalign(bs, qiov->size);
696     }
697     acb->ret = 0;
698     acb->error = 0;
699     acb->s = s;
700     acb->cancelled = 0;
701     acb->bh = NULL;
702     acb->status = -EINPROGRESS;
703 
704     if (cmd == RBD_AIO_WRITE) {
705         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
706     }
707 
708     buf = acb->bounce;
709 
710     off = sector_num * BDRV_SECTOR_SIZE;
711     size = nb_sectors * BDRV_SECTOR_SIZE;
712 
713     s->qemu_aio_count++; /* All the RADOSCB */
714 
715     rcb = g_malloc(sizeof(RADOSCB));
716     rcb->done = 0;
717     rcb->acb = acb;
718     rcb->buf = buf;
719     rcb->s = acb->s;
720     rcb->size = size;
721     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
722     if (r < 0) {
723         goto failed;
724     }
725 
726     switch (cmd) {
727     case RBD_AIO_WRITE:
728         r = rbd_aio_write(s->image, off, size, buf, c);
729         break;
730     case RBD_AIO_READ:
731         r = rbd_aio_read(s->image, off, size, buf, c);
732         break;
733     case RBD_AIO_DISCARD:
734         r = rbd_aio_discard_wrapper(s->image, off, size, c);
735         break;
736     case RBD_AIO_FLUSH:
737         r = rbd_aio_flush_wrapper(s->image, c);
738         break;
739     default:
740         r = -EINVAL;
741     }
742 
743     if (r < 0) {
744         goto failed;
745     }
746 
747     return &acb->common;
748 
749 failed:
750     g_free(rcb);
751     s->qemu_aio_count--;
752     qemu_aio_release(acb);
753     return NULL;
754 }
755 
756 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
757                                             int64_t sector_num,
758                                             QEMUIOVector *qiov,
759                                             int nb_sectors,
760                                             BlockDriverCompletionFunc *cb,
761                                             void *opaque)
762 {
763     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
764                          RBD_AIO_READ);
765 }
766 
767 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
768                                              int64_t sector_num,
769                                              QEMUIOVector *qiov,
770                                              int nb_sectors,
771                                              BlockDriverCompletionFunc *cb,
772                                              void *opaque)
773 {
774     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
775                          RBD_AIO_WRITE);
776 }
777 
778 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
779 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
780                                             BlockDriverCompletionFunc *cb,
781                                             void *opaque)
782 {
783     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
784 }
785 
786 #else
787 
788 static int qemu_rbd_co_flush(BlockDriverState *bs)
789 {
790 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
791     /* rbd_flush added in 0.1.1 */
792     BDRVRBDState *s = bs->opaque;
793     return rbd_flush(s->image);
794 #else
795     return 0;
796 #endif
797 }
798 #endif
799 
800 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
801 {
802     BDRVRBDState *s = bs->opaque;
803     rbd_image_info_t info;
804     int r;
805 
806     r = rbd_stat(s->image, &info, sizeof(info));
807     if (r < 0) {
808         return r;
809     }
810 
811     bdi->cluster_size = info.obj_size;
812     return 0;
813 }
814 
815 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
816 {
817     BDRVRBDState *s = bs->opaque;
818     rbd_image_info_t info;
819     int r;
820 
821     r = rbd_stat(s->image, &info, sizeof(info));
822     if (r < 0) {
823         return r;
824     }
825 
826     return info.size;
827 }
828 
829 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
830 {
831     BDRVRBDState *s = bs->opaque;
832     int r;
833 
834     r = rbd_resize(s->image, offset);
835     if (r < 0) {
836         return r;
837     }
838 
839     return 0;
840 }
841 
842 static int qemu_rbd_snap_create(BlockDriverState *bs,
843                                 QEMUSnapshotInfo *sn_info)
844 {
845     BDRVRBDState *s = bs->opaque;
846     int r;
847 
848     if (sn_info->name[0] == '\0') {
849         return -EINVAL; /* we need a name for rbd snapshots */
850     }
851 
852     /*
853      * rbd snapshots are using the name as the user controlled unique identifier
854      * we can't use the rbd snapid for that purpose, as it can't be set
855      */
856     if (sn_info->id_str[0] != '\0' &&
857         strcmp(sn_info->id_str, sn_info->name) != 0) {
858         return -EINVAL;
859     }
860 
861     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
862         return -ERANGE;
863     }
864 
865     r = rbd_snap_create(s->image, sn_info->name);
866     if (r < 0) {
867         error_report("failed to create snap: %s", strerror(-r));
868         return r;
869     }
870 
871     return 0;
872 }
873 
874 static int qemu_rbd_snap_remove(BlockDriverState *bs,
875                                 const char *snapshot_name)
876 {
877     BDRVRBDState *s = bs->opaque;
878     int r;
879 
880     r = rbd_snap_remove(s->image, snapshot_name);
881     return r;
882 }
883 
884 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
885                                   const char *snapshot_name)
886 {
887     BDRVRBDState *s = bs->opaque;
888     int r;
889 
890     r = rbd_snap_rollback(s->image, snapshot_name);
891     return r;
892 }
893 
894 static int qemu_rbd_snap_list(BlockDriverState *bs,
895                               QEMUSnapshotInfo **psn_tab)
896 {
897     BDRVRBDState *s = bs->opaque;
898     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
899     int i, snap_count;
900     rbd_snap_info_t *snaps;
901     int max_snaps = RBD_MAX_SNAPS;
902 
903     do {
904         snaps = g_malloc(sizeof(*snaps) * max_snaps);
905         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
906         if (snap_count < 0) {
907             g_free(snaps);
908         }
909     } while (snap_count == -ERANGE);
910 
911     if (snap_count <= 0) {
912         goto done;
913     }
914 
915     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
916 
917     for (i = 0; i < snap_count; i++) {
918         const char *snap_name = snaps[i].name;
919 
920         sn_info = sn_tab + i;
921         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
922         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
923 
924         sn_info->vm_state_size = snaps[i].size;
925         sn_info->date_sec = 0;
926         sn_info->date_nsec = 0;
927         sn_info->vm_clock_nsec = 0;
928     }
929     rbd_snap_list_end(snaps);
930 
931  done:
932     *psn_tab = sn_tab;
933     return snap_count;
934 }
935 
936 #ifdef LIBRBD_SUPPORTS_DISCARD
937 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
938                                               int64_t sector_num,
939                                               int nb_sectors,
940                                               BlockDriverCompletionFunc *cb,
941                                               void *opaque)
942 {
943     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
944                          RBD_AIO_DISCARD);
945 }
946 #endif
947 
948 static QEMUOptionParameter qemu_rbd_create_options[] = {
949     {
950      .name = BLOCK_OPT_SIZE,
951      .type = OPT_SIZE,
952      .help = "Virtual disk size"
953     },
954     {
955      .name = BLOCK_OPT_CLUSTER_SIZE,
956      .type = OPT_SIZE,
957      .help = "RBD object size"
958     },
959     {NULL}
960 };
961 
962 static BlockDriver bdrv_rbd = {
963     .format_name        = "rbd",
964     .instance_size      = sizeof(BDRVRBDState),
965     .bdrv_file_open     = qemu_rbd_open,
966     .bdrv_close         = qemu_rbd_close,
967     .bdrv_create        = qemu_rbd_create,
968     .bdrv_get_info      = qemu_rbd_getinfo,
969     .create_options     = qemu_rbd_create_options,
970     .bdrv_getlength     = qemu_rbd_getlength,
971     .bdrv_truncate      = qemu_rbd_truncate,
972     .protocol_name      = "rbd",
973 
974     .bdrv_aio_readv         = qemu_rbd_aio_readv,
975     .bdrv_aio_writev        = qemu_rbd_aio_writev,
976 
977 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
978     .bdrv_aio_flush         = qemu_rbd_aio_flush,
979 #else
980     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
981 #endif
982 
983 #ifdef LIBRBD_SUPPORTS_DISCARD
984     .bdrv_aio_discard       = qemu_rbd_aio_discard,
985 #endif
986 
987     .bdrv_snapshot_create   = qemu_rbd_snap_create,
988     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
989     .bdrv_snapshot_list     = qemu_rbd_snap_list,
990     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
991 };
992 
993 static void bdrv_rbd_init(void)
994 {
995     bdrv_register(&bdrv_rbd);
996 }
997 
998 block_init(bdrv_rbd_init);
999