xref: /openbmc/qemu/block/rbd.c (revision 1f32989d)
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.h"
18 #include "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 } RBDAIOCmd;
68 
69 typedef struct RBDAIOCB {
70     BlockDriverAIOCB common;
71     QEMUBH *bh;
72     int64_t ret;
73     QEMUIOVector *qiov;
74     char *bounce;
75     RBDAIOCmd cmd;
76     int64_t sector_num;
77     int error;
78     struct BDRVRBDState *s;
79     int cancelled;
80 } RBDAIOCB;
81 
82 typedef struct RADOSCB {
83     int rcbid;
84     RBDAIOCB *acb;
85     struct BDRVRBDState *s;
86     int done;
87     int64_t size;
88     char *buf;
89     int64_t ret;
90 } RADOSCB;
91 
92 #define RBD_FD_READ 0
93 #define RBD_FD_WRITE 1
94 
95 typedef struct BDRVRBDState {
96     int fds[2];
97     rados_t cluster;
98     rados_ioctx_t io_ctx;
99     rbd_image_t image;
100     char name[RBD_MAX_IMAGE_NAME_SIZE];
101     int qemu_aio_count;
102     char *snap;
103     int event_reader_pos;
104     RADOSCB *event_rcb;
105 } BDRVRBDState;
106 
107 static void rbd_aio_bh_cb(void *opaque);
108 
109 static int qemu_rbd_next_tok(char *dst, int dst_len,
110                              char *src, char delim,
111                              const char *name,
112                              char **p)
113 {
114     int l;
115     char *end;
116 
117     *p = NULL;
118 
119     if (delim != '\0') {
120         for (end = src; *end; ++end) {
121             if (*end == delim) {
122                 break;
123             }
124             if (*end == '\\' && end[1] != '\0') {
125                 end++;
126             }
127         }
128         if (*end == delim) {
129             *p = end + 1;
130             *end = '\0';
131         }
132     }
133     l = strlen(src);
134     if (l >= dst_len) {
135         error_report("%s too long", name);
136         return -EINVAL;
137     } else if (l == 0) {
138         error_report("%s too short", name);
139         return -EINVAL;
140     }
141 
142     pstrcpy(dst, dst_len, src);
143 
144     return 0;
145 }
146 
147 static void qemu_rbd_unescape(char *src)
148 {
149     char *p;
150 
151     for (p = src; *src; ++src, ++p) {
152         if (*src == '\\' && src[1] != '\0') {
153             src++;
154         }
155         *p = *src;
156     }
157     *p = '\0';
158 }
159 
160 static int qemu_rbd_parsename(const char *filename,
161                               char *pool, int pool_len,
162                               char *snap, int snap_len,
163                               char *name, int name_len,
164                               char *conf, int conf_len)
165 {
166     const char *start;
167     char *p, *buf;
168     int ret;
169 
170     if (!strstart(filename, "rbd:", &start)) {
171         return -EINVAL;
172     }
173 
174     buf = g_strdup(start);
175     p = buf;
176     *snap = '\0';
177     *conf = '\0';
178 
179     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
180     if (ret < 0 || !p) {
181         ret = -EINVAL;
182         goto done;
183     }
184     qemu_rbd_unescape(pool);
185 
186     if (strchr(p, '@')) {
187         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
188         if (ret < 0) {
189             goto done;
190         }
191         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
192         qemu_rbd_unescape(snap);
193     } else {
194         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
195     }
196     qemu_rbd_unescape(name);
197     if (ret < 0 || !p) {
198         goto done;
199     }
200 
201     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
202 
203 done:
204     g_free(buf);
205     return ret;
206 }
207 
208 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
209 {
210     const char *p = conf;
211 
212     while (*p) {
213         int len;
214         const char *end = strchr(p, ':');
215 
216         if (end) {
217             len = end - p;
218         } else {
219             len = strlen(p);
220         }
221 
222         if (strncmp(p, "id=", 3) == 0) {
223             len -= 3;
224             strncpy(clientname, p + 3, len);
225             clientname[len] = '\0';
226             return clientname;
227         }
228         if (end == NULL) {
229             break;
230         }
231         p = end + 1;
232     }
233     return NULL;
234 }
235 
236 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
237 {
238     char *p, *buf;
239     char name[RBD_MAX_CONF_NAME_SIZE];
240     char value[RBD_MAX_CONF_VAL_SIZE];
241     int ret = 0;
242 
243     buf = g_strdup(conf);
244     p = buf;
245 
246     while (p) {
247         ret = qemu_rbd_next_tok(name, sizeof(name), p,
248                                 '=', "conf option name", &p);
249         if (ret < 0) {
250             break;
251         }
252         qemu_rbd_unescape(name);
253 
254         if (!p) {
255             error_report("conf option %s has no value", name);
256             ret = -EINVAL;
257             break;
258         }
259 
260         ret = qemu_rbd_next_tok(value, sizeof(value), p,
261                                 ':', "conf option value", &p);
262         if (ret < 0) {
263             break;
264         }
265         qemu_rbd_unescape(value);
266 
267         if (strcmp(name, "conf") == 0) {
268             ret = rados_conf_read_file(cluster, value);
269             if (ret < 0) {
270                 error_report("error reading conf file %s", value);
271                 break;
272             }
273         } else if (strcmp(name, "id") == 0) {
274             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
275         } else {
276             ret = rados_conf_set(cluster, name, value);
277             if (ret < 0) {
278                 error_report("invalid conf option %s", name);
279                 ret = -EINVAL;
280                 break;
281             }
282         }
283     }
284 
285     g_free(buf);
286     return ret;
287 }
288 
289 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
290 {
291     int64_t bytes = 0;
292     int64_t objsize;
293     int obj_order = 0;
294     char pool[RBD_MAX_POOL_NAME_SIZE];
295     char name[RBD_MAX_IMAGE_NAME_SIZE];
296     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
297     char conf[RBD_MAX_CONF_SIZE];
298     char clientname_buf[RBD_MAX_CONF_SIZE];
299     char *clientname;
300     rados_t cluster;
301     rados_ioctx_t io_ctx;
302     int ret;
303 
304     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
305                            snap_buf, sizeof(snap_buf),
306                            name, sizeof(name),
307                            conf, sizeof(conf)) < 0) {
308         return -EINVAL;
309     }
310 
311     /* Read out options */
312     while (options && options->name) {
313         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
314             bytes = options->value.n;
315         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
316             if (options->value.n) {
317                 objsize = options->value.n;
318                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
319                     error_report("obj size needs to be power of 2");
320                     return -EINVAL;
321                 }
322                 if (objsize < 4096) {
323                     error_report("obj size too small");
324                     return -EINVAL;
325                 }
326                 obj_order = ffs(objsize) - 1;
327             }
328         }
329         options++;
330     }
331 
332     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
333     if (rados_create(&cluster, clientname) < 0) {
334         error_report("error initializing");
335         return -EIO;
336     }
337 
338     if (strstr(conf, "conf=") == NULL) {
339         /* try default location, but ignore failure */
340         rados_conf_read_file(cluster, NULL);
341     }
342 
343     if (conf[0] != '\0' &&
344         qemu_rbd_set_conf(cluster, conf) < 0) {
345         error_report("error setting config options");
346         rados_shutdown(cluster);
347         return -EIO;
348     }
349 
350     if (rados_connect(cluster) < 0) {
351         error_report("error connecting");
352         rados_shutdown(cluster);
353         return -EIO;
354     }
355 
356     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
357         error_report("error opening pool %s", pool);
358         rados_shutdown(cluster);
359         return -EIO;
360     }
361 
362     ret = rbd_create(io_ctx, name, bytes, &obj_order);
363     rados_ioctx_destroy(io_ctx);
364     rados_shutdown(cluster);
365 
366     return ret;
367 }
368 
369 /*
370  * This aio completion is being called from qemu_rbd_aio_event_reader()
371  * and runs in qemu context. It schedules a bh, but just in case the aio
372  * was not cancelled before.
373  */
374 static void qemu_rbd_complete_aio(RADOSCB *rcb)
375 {
376     RBDAIOCB *acb = rcb->acb;
377     int64_t r;
378 
379     if (acb->cancelled) {
380         qemu_vfree(acb->bounce);
381         qemu_aio_release(acb);
382         goto done;
383     }
384 
385     r = rcb->ret;
386 
387     if (acb->cmd == RBD_AIO_WRITE ||
388         acb->cmd == RBD_AIO_DISCARD) {
389         if (r < 0) {
390             acb->ret = r;
391             acb->error = 1;
392         } else if (!acb->error) {
393             acb->ret = rcb->size;
394         }
395     } else {
396         if (r < 0) {
397             memset(rcb->buf, 0, rcb->size);
398             acb->ret = r;
399             acb->error = 1;
400         } else if (r < rcb->size) {
401             memset(rcb->buf + r, 0, rcb->size - r);
402             if (!acb->error) {
403                 acb->ret = rcb->size;
404             }
405         } else if (!acb->error) {
406             acb->ret = r;
407         }
408     }
409     /* Note that acb->bh can be NULL in case where the aio was cancelled */
410     acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
411     qemu_bh_schedule(acb->bh);
412 done:
413     g_free(rcb);
414 }
415 
416 /*
417  * aio fd read handler. It runs in the qemu context and calls the
418  * completion handling of completed rados aio operations.
419  */
420 static void qemu_rbd_aio_event_reader(void *opaque)
421 {
422     BDRVRBDState *s = opaque;
423 
424     ssize_t ret;
425 
426     do {
427         char *p = (char *)&s->event_rcb;
428 
429         /* now read the rcb pointer that was sent from a non qemu thread */
430         ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
431                    sizeof(s->event_rcb) - s->event_reader_pos);
432         if (ret > 0) {
433             s->event_reader_pos += ret;
434             if (s->event_reader_pos == sizeof(s->event_rcb)) {
435                 s->event_reader_pos = 0;
436                 qemu_rbd_complete_aio(s->event_rcb);
437                 s->qemu_aio_count--;
438             }
439         }
440     } while (ret < 0 && errno == EINTR);
441 }
442 
443 static int qemu_rbd_aio_flush_cb(void *opaque)
444 {
445     BDRVRBDState *s = opaque;
446 
447     return (s->qemu_aio_count > 0);
448 }
449 
450 static int qemu_rbd_open(BlockDriverState *bs, const char *filename, int flags)
451 {
452     BDRVRBDState *s = bs->opaque;
453     char pool[RBD_MAX_POOL_NAME_SIZE];
454     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
455     char conf[RBD_MAX_CONF_SIZE];
456     char clientname_buf[RBD_MAX_CONF_SIZE];
457     char *clientname;
458     int r;
459 
460     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
461                            snap_buf, sizeof(snap_buf),
462                            s->name, sizeof(s->name),
463                            conf, sizeof(conf)) < 0) {
464         return -EINVAL;
465     }
466 
467     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
468     r = rados_create(&s->cluster, clientname);
469     if (r < 0) {
470         error_report("error initializing");
471         return r;
472     }
473 
474     s->snap = NULL;
475     if (snap_buf[0] != '\0') {
476         s->snap = g_strdup(snap_buf);
477     }
478 
479     /*
480      * Fallback to more conservative semantics if setting cache
481      * options fails. Ignore errors from setting rbd_cache because the
482      * only possible error is that the option does not exist, and
483      * librbd defaults to no caching. If write through caching cannot
484      * be set up, fall back to no caching.
485      */
486     if (flags & BDRV_O_NOCACHE) {
487         rados_conf_set(s->cluster, "rbd_cache", "false");
488     } else {
489         rados_conf_set(s->cluster, "rbd_cache", "true");
490     }
491 
492     if (strstr(conf, "conf=") == NULL) {
493         /* try default location, but ignore failure */
494         rados_conf_read_file(s->cluster, NULL);
495     }
496 
497     if (conf[0] != '\0') {
498         r = qemu_rbd_set_conf(s->cluster, conf);
499         if (r < 0) {
500             error_report("error setting config options");
501             goto failed_shutdown;
502         }
503     }
504 
505     r = rados_connect(s->cluster);
506     if (r < 0) {
507         error_report("error connecting");
508         goto failed_shutdown;
509     }
510 
511     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
512     if (r < 0) {
513         error_report("error opening pool %s", pool);
514         goto failed_shutdown;
515     }
516 
517     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
518     if (r < 0) {
519         error_report("error reading header from %s", s->name);
520         goto failed_open;
521     }
522 
523     bs->read_only = (s->snap != NULL);
524 
525     s->event_reader_pos = 0;
526     r = qemu_pipe(s->fds);
527     if (r < 0) {
528         error_report("error opening eventfd");
529         goto failed;
530     }
531     fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
532     fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
533     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
534                             NULL, qemu_rbd_aio_flush_cb, s);
535 
536 
537     return 0;
538 
539 failed:
540     rbd_close(s->image);
541 failed_open:
542     rados_ioctx_destroy(s->io_ctx);
543 failed_shutdown:
544     rados_shutdown(s->cluster);
545     g_free(s->snap);
546     return r;
547 }
548 
549 static void qemu_rbd_close(BlockDriverState *bs)
550 {
551     BDRVRBDState *s = bs->opaque;
552 
553     close(s->fds[0]);
554     close(s->fds[1]);
555     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL);
556 
557     rbd_close(s->image);
558     rados_ioctx_destroy(s->io_ctx);
559     g_free(s->snap);
560     rados_shutdown(s->cluster);
561 }
562 
563 /*
564  * Cancel aio. Since we don't reference acb in a non qemu threads,
565  * it is safe to access it here.
566  */
567 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
568 {
569     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
570     acb->cancelled = 1;
571 }
572 
573 static const AIOCBInfo rbd_aiocb_info = {
574     .aiocb_size = sizeof(RBDAIOCB),
575     .cancel = qemu_rbd_aio_cancel,
576 };
577 
578 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
579 {
580     int ret = 0;
581     while (1) {
582         fd_set wfd;
583         int fd = s->fds[RBD_FD_WRITE];
584 
585         /* send the op pointer to the qemu thread that is responsible
586            for the aio/op completion. Must do it in a qemu thread context */
587         ret = write(fd, (void *)&rcb, sizeof(rcb));
588         if (ret >= 0) {
589             break;
590         }
591         if (errno == EINTR) {
592             continue;
593         }
594         if (errno != EAGAIN) {
595             break;
596         }
597 
598         FD_ZERO(&wfd);
599         FD_SET(fd, &wfd);
600         do {
601             ret = select(fd + 1, NULL, &wfd, NULL, NULL);
602         } while (ret < 0 && errno == EINTR);
603     }
604 
605     return ret;
606 }
607 
608 /*
609  * This is the callback function for rbd_aio_read and _write
610  *
611  * Note: this function is being called from a non qemu thread so
612  * we need to be careful about what we do here. Generally we only
613  * write to the block notification pipe, and do the rest of the
614  * io completion handling from qemu_rbd_aio_event_reader() which
615  * runs in a qemu context.
616  */
617 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
618 {
619     int ret;
620     rcb->ret = rbd_aio_get_return_value(c);
621     rbd_aio_release(c);
622     ret = qemu_rbd_send_pipe(rcb->s, rcb);
623     if (ret < 0) {
624         error_report("failed writing to acb->s->fds");
625         g_free(rcb);
626     }
627 }
628 
629 /* Callback when all queued rbd_aio requests are complete */
630 
631 static void rbd_aio_bh_cb(void *opaque)
632 {
633     RBDAIOCB *acb = opaque;
634 
635     if (acb->cmd == RBD_AIO_READ) {
636         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
637     }
638     qemu_vfree(acb->bounce);
639     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
640     qemu_bh_delete(acb->bh);
641     acb->bh = NULL;
642 
643     qemu_aio_release(acb);
644 }
645 
646 static int rbd_aio_discard_wrapper(rbd_image_t image,
647                                    uint64_t off,
648                                    uint64_t len,
649                                    rbd_completion_t comp)
650 {
651 #ifdef LIBRBD_SUPPORTS_DISCARD
652     return rbd_aio_discard(image, off, len, comp);
653 #else
654     return -ENOTSUP;
655 #endif
656 }
657 
658 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
659                                        int64_t sector_num,
660                                        QEMUIOVector *qiov,
661                                        int nb_sectors,
662                                        BlockDriverCompletionFunc *cb,
663                                        void *opaque,
664                                        RBDAIOCmd cmd)
665 {
666     RBDAIOCB *acb;
667     RADOSCB *rcb;
668     rbd_completion_t c;
669     int64_t off, size;
670     char *buf;
671     int r;
672 
673     BDRVRBDState *s = bs->opaque;
674 
675     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
676     acb->cmd = cmd;
677     acb->qiov = qiov;
678     if (cmd == RBD_AIO_DISCARD) {
679         acb->bounce = NULL;
680     } else {
681         acb->bounce = qemu_blockalign(bs, qiov->size);
682     }
683     acb->ret = 0;
684     acb->error = 0;
685     acb->s = s;
686     acb->cancelled = 0;
687     acb->bh = NULL;
688 
689     if (cmd == RBD_AIO_WRITE) {
690         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
691     }
692 
693     buf = acb->bounce;
694 
695     off = sector_num * BDRV_SECTOR_SIZE;
696     size = nb_sectors * BDRV_SECTOR_SIZE;
697 
698     s->qemu_aio_count++; /* All the RADOSCB */
699 
700     rcb = g_malloc(sizeof(RADOSCB));
701     rcb->done = 0;
702     rcb->acb = acb;
703     rcb->buf = buf;
704     rcb->s = acb->s;
705     rcb->size = size;
706     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
707     if (r < 0) {
708         goto failed;
709     }
710 
711     switch (cmd) {
712     case RBD_AIO_WRITE:
713         r = rbd_aio_write(s->image, off, size, buf, c);
714         break;
715     case RBD_AIO_READ:
716         r = rbd_aio_read(s->image, off, size, buf, c);
717         break;
718     case RBD_AIO_DISCARD:
719         r = rbd_aio_discard_wrapper(s->image, off, size, c);
720         break;
721     default:
722         r = -EINVAL;
723     }
724 
725     if (r < 0) {
726         goto failed;
727     }
728 
729     return &acb->common;
730 
731 failed:
732     g_free(rcb);
733     s->qemu_aio_count--;
734     qemu_aio_release(acb);
735     return NULL;
736 }
737 
738 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
739                                             int64_t sector_num,
740                                             QEMUIOVector *qiov,
741                                             int nb_sectors,
742                                             BlockDriverCompletionFunc *cb,
743                                             void *opaque)
744 {
745     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
746                          RBD_AIO_READ);
747 }
748 
749 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
750                                              int64_t sector_num,
751                                              QEMUIOVector *qiov,
752                                              int nb_sectors,
753                                              BlockDriverCompletionFunc *cb,
754                                              void *opaque)
755 {
756     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
757                          RBD_AIO_WRITE);
758 }
759 
760 static int qemu_rbd_co_flush(BlockDriverState *bs)
761 {
762 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
763     /* rbd_flush added in 0.1.1 */
764     BDRVRBDState *s = bs->opaque;
765     return rbd_flush(s->image);
766 #else
767     return 0;
768 #endif
769 }
770 
771 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
772 {
773     BDRVRBDState *s = bs->opaque;
774     rbd_image_info_t info;
775     int r;
776 
777     r = rbd_stat(s->image, &info, sizeof(info));
778     if (r < 0) {
779         return r;
780     }
781 
782     bdi->cluster_size = info.obj_size;
783     return 0;
784 }
785 
786 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
787 {
788     BDRVRBDState *s = bs->opaque;
789     rbd_image_info_t info;
790     int r;
791 
792     r = rbd_stat(s->image, &info, sizeof(info));
793     if (r < 0) {
794         return r;
795     }
796 
797     return info.size;
798 }
799 
800 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
801 {
802     BDRVRBDState *s = bs->opaque;
803     int r;
804 
805     r = rbd_resize(s->image, offset);
806     if (r < 0) {
807         return r;
808     }
809 
810     return 0;
811 }
812 
813 static int qemu_rbd_snap_create(BlockDriverState *bs,
814                                 QEMUSnapshotInfo *sn_info)
815 {
816     BDRVRBDState *s = bs->opaque;
817     int r;
818 
819     if (sn_info->name[0] == '\0') {
820         return -EINVAL; /* we need a name for rbd snapshots */
821     }
822 
823     /*
824      * rbd snapshots are using the name as the user controlled unique identifier
825      * we can't use the rbd snapid for that purpose, as it can't be set
826      */
827     if (sn_info->id_str[0] != '\0' &&
828         strcmp(sn_info->id_str, sn_info->name) != 0) {
829         return -EINVAL;
830     }
831 
832     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
833         return -ERANGE;
834     }
835 
836     r = rbd_snap_create(s->image, sn_info->name);
837     if (r < 0) {
838         error_report("failed to create snap: %s", strerror(-r));
839         return r;
840     }
841 
842     return 0;
843 }
844 
845 static int qemu_rbd_snap_remove(BlockDriverState *bs,
846                                 const char *snapshot_name)
847 {
848     BDRVRBDState *s = bs->opaque;
849     int r;
850 
851     r = rbd_snap_remove(s->image, snapshot_name);
852     return r;
853 }
854 
855 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
856                                   const char *snapshot_name)
857 {
858     BDRVRBDState *s = bs->opaque;
859     int r;
860 
861     r = rbd_snap_rollback(s->image, snapshot_name);
862     return r;
863 }
864 
865 static int qemu_rbd_snap_list(BlockDriverState *bs,
866                               QEMUSnapshotInfo **psn_tab)
867 {
868     BDRVRBDState *s = bs->opaque;
869     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
870     int i, snap_count;
871     rbd_snap_info_t *snaps;
872     int max_snaps = RBD_MAX_SNAPS;
873 
874     do {
875         snaps = g_malloc(sizeof(*snaps) * max_snaps);
876         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
877         if (snap_count < 0) {
878             g_free(snaps);
879         }
880     } while (snap_count == -ERANGE);
881 
882     if (snap_count <= 0) {
883         goto done;
884     }
885 
886     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
887 
888     for (i = 0; i < snap_count; i++) {
889         const char *snap_name = snaps[i].name;
890 
891         sn_info = sn_tab + i;
892         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
893         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
894 
895         sn_info->vm_state_size = snaps[i].size;
896         sn_info->date_sec = 0;
897         sn_info->date_nsec = 0;
898         sn_info->vm_clock_nsec = 0;
899     }
900     rbd_snap_list_end(snaps);
901 
902  done:
903     *psn_tab = sn_tab;
904     return snap_count;
905 }
906 
907 #ifdef LIBRBD_SUPPORTS_DISCARD
908 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
909                                               int64_t sector_num,
910                                               int nb_sectors,
911                                               BlockDriverCompletionFunc *cb,
912                                               void *opaque)
913 {
914     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
915                          RBD_AIO_DISCARD);
916 }
917 #endif
918 
919 static QEMUOptionParameter qemu_rbd_create_options[] = {
920     {
921      .name = BLOCK_OPT_SIZE,
922      .type = OPT_SIZE,
923      .help = "Virtual disk size"
924     },
925     {
926      .name = BLOCK_OPT_CLUSTER_SIZE,
927      .type = OPT_SIZE,
928      .help = "RBD object size"
929     },
930     {NULL}
931 };
932 
933 static BlockDriver bdrv_rbd = {
934     .format_name        = "rbd",
935     .instance_size      = sizeof(BDRVRBDState),
936     .bdrv_file_open     = qemu_rbd_open,
937     .bdrv_close         = qemu_rbd_close,
938     .bdrv_create        = qemu_rbd_create,
939     .bdrv_get_info      = qemu_rbd_getinfo,
940     .create_options     = qemu_rbd_create_options,
941     .bdrv_getlength     = qemu_rbd_getlength,
942     .bdrv_truncate      = qemu_rbd_truncate,
943     .protocol_name      = "rbd",
944 
945     .bdrv_aio_readv         = qemu_rbd_aio_readv,
946     .bdrv_aio_writev        = qemu_rbd_aio_writev,
947     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
948 
949 #ifdef LIBRBD_SUPPORTS_DISCARD
950     .bdrv_aio_discard       = qemu_rbd_aio_discard,
951 #endif
952 
953     .bdrv_snapshot_create   = qemu_rbd_snap_create,
954     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
955     .bdrv_snapshot_list     = qemu_rbd_snap_list,
956     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
957 };
958 
959 static void bdrv_rbd_init(void)
960 {
961     bdrv_register(&bdrv_rbd);
962 }
963 
964 block_init(bdrv_rbd_init);
965