xref: /openbmc/qemu/block/rbd.c (revision 646c5478)
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 "qemu/osdep.h"
15 
16 #include "qapi/error.h"
17 #include "qemu/error-report.h"
18 #include "block/block_int.h"
19 #include "crypto/secret.h"
20 #include "qemu/cutils.h"
21 
22 #include <rbd/librbd.h>
23 
24 /*
25  * When specifying the image filename use:
26  *
27  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
28  *
29  * poolname must be the name of an existing rados pool.
30  *
31  * devicename is the name of the rbd image.
32  *
33  * Each option given is used to configure rados, and may be any valid
34  * Ceph option, "id", or "conf".
35  *
36  * The "id" option indicates what user we should authenticate as to
37  * the Ceph cluster.  If it is excluded we will use the Ceph default
38  * (normally 'admin').
39  *
40  * The "conf" option specifies a Ceph configuration file to read.  If
41  * it is not specified, we will read from the default Ceph locations
42  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
43  * file, specify conf=/dev/null.
44  *
45  * Configuration values containing :, @, or = can be escaped with a
46  * leading "\".
47  */
48 
49 /* rbd_aio_discard added in 0.1.2 */
50 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
51 #define LIBRBD_SUPPORTS_DISCARD
52 #else
53 #undef LIBRBD_SUPPORTS_DISCARD
54 #endif
55 
56 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
57 
58 #define RBD_MAX_CONF_NAME_SIZE 128
59 #define RBD_MAX_CONF_VAL_SIZE 512
60 #define RBD_MAX_CONF_SIZE 1024
61 #define RBD_MAX_POOL_NAME_SIZE 128
62 #define RBD_MAX_SNAP_NAME_SIZE 128
63 #define RBD_MAX_SNAPS 100
64 
65 typedef enum {
66     RBD_AIO_READ,
67     RBD_AIO_WRITE,
68     RBD_AIO_DISCARD,
69     RBD_AIO_FLUSH
70 } RBDAIOCmd;
71 
72 typedef struct RBDAIOCB {
73     BlockAIOCB common;
74     int64_t ret;
75     QEMUIOVector *qiov;
76     char *bounce;
77     RBDAIOCmd cmd;
78     int error;
79     struct BDRVRBDState *s;
80 } RBDAIOCB;
81 
82 typedef struct RADOSCB {
83     RBDAIOCB *acb;
84     struct BDRVRBDState *s;
85     int64_t size;
86     char *buf;
87     int64_t ret;
88 } RADOSCB;
89 
90 typedef struct BDRVRBDState {
91     rados_t cluster;
92     rados_ioctx_t io_ctx;
93     rbd_image_t image;
94     char name[RBD_MAX_IMAGE_NAME_SIZE];
95     char *snap;
96 } BDRVRBDState;
97 
98 static int qemu_rbd_next_tok(char *dst, int dst_len,
99                              char *src, char delim,
100                              const char *name,
101                              char **p, Error **errp)
102 {
103     int l;
104     char *end;
105 
106     *p = NULL;
107 
108     if (delim != '\0') {
109         for (end = src; *end; ++end) {
110             if (*end == delim) {
111                 break;
112             }
113             if (*end == '\\' && end[1] != '\0') {
114                 end++;
115             }
116         }
117         if (*end == delim) {
118             *p = end + 1;
119             *end = '\0';
120         }
121     }
122     l = strlen(src);
123     if (l >= dst_len) {
124         error_setg(errp, "%s too long", name);
125         return -EINVAL;
126     } else if (l == 0) {
127         error_setg(errp, "%s too short", name);
128         return -EINVAL;
129     }
130 
131     pstrcpy(dst, dst_len, src);
132 
133     return 0;
134 }
135 
136 static void qemu_rbd_unescape(char *src)
137 {
138     char *p;
139 
140     for (p = src; *src; ++src, ++p) {
141         if (*src == '\\' && src[1] != '\0') {
142             src++;
143         }
144         *p = *src;
145     }
146     *p = '\0';
147 }
148 
149 static int qemu_rbd_parsename(const char *filename,
150                               char *pool, int pool_len,
151                               char *snap, int snap_len,
152                               char *name, int name_len,
153                               char *conf, int conf_len,
154                               Error **errp)
155 {
156     const char *start;
157     char *p, *buf;
158     int ret;
159 
160     if (!strstart(filename, "rbd:", &start)) {
161         error_setg(errp, "File name must start with 'rbd:'");
162         return -EINVAL;
163     }
164 
165     buf = g_strdup(start);
166     p = buf;
167     *snap = '\0';
168     *conf = '\0';
169 
170     ret = qemu_rbd_next_tok(pool, pool_len, p,
171                             '/', "pool name", &p, errp);
172     if (ret < 0 || !p) {
173         ret = -EINVAL;
174         goto done;
175     }
176     qemu_rbd_unescape(pool);
177 
178     if (strchr(p, '@')) {
179         ret = qemu_rbd_next_tok(name, name_len, p,
180                                 '@', "object name", &p, errp);
181         if (ret < 0) {
182             goto done;
183         }
184         ret = qemu_rbd_next_tok(snap, snap_len, p,
185                                 ':', "snap name", &p, errp);
186         qemu_rbd_unescape(snap);
187     } else {
188         ret = qemu_rbd_next_tok(name, name_len, p,
189                                 ':', "object name", &p, errp);
190     }
191     qemu_rbd_unescape(name);
192     if (ret < 0 || !p) {
193         goto done;
194     }
195 
196     ret = qemu_rbd_next_tok(conf, conf_len, p,
197                             '\0', "configuration", &p, errp);
198 
199 done:
200     g_free(buf);
201     return ret;
202 }
203 
204 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
205 {
206     const char *p = conf;
207 
208     while (*p) {
209         int len;
210         const char *end = strchr(p, ':');
211 
212         if (end) {
213             len = end - p;
214         } else {
215             len = strlen(p);
216         }
217 
218         if (strncmp(p, "id=", 3) == 0) {
219             len -= 3;
220             strncpy(clientname, p + 3, len);
221             clientname[len] = '\0';
222             return clientname;
223         }
224         if (end == NULL) {
225             break;
226         }
227         p = end + 1;
228     }
229     return NULL;
230 }
231 
232 
233 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
234                              Error **errp)
235 {
236     if (secretid == 0) {
237         return 0;
238     }
239 
240     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
241                                                     errp);
242     if (!secret) {
243         return -1;
244     }
245 
246     rados_conf_set(cluster, "key", secret);
247     g_free(secret);
248 
249     return 0;
250 }
251 
252 
253 static int qemu_rbd_set_conf(rados_t cluster, const char *conf,
254                              bool only_read_conf_file,
255                              Error **errp)
256 {
257     char *p, *buf;
258     char name[RBD_MAX_CONF_NAME_SIZE];
259     char value[RBD_MAX_CONF_VAL_SIZE];
260     int ret = 0;
261 
262     buf = g_strdup(conf);
263     p = buf;
264 
265     while (p) {
266         ret = qemu_rbd_next_tok(name, sizeof(name), p,
267                                 '=', "conf option name", &p, errp);
268         if (ret < 0) {
269             break;
270         }
271         qemu_rbd_unescape(name);
272 
273         if (!p) {
274             error_setg(errp, "conf option %s has no value", name);
275             ret = -EINVAL;
276             break;
277         }
278 
279         ret = qemu_rbd_next_tok(value, sizeof(value), p,
280                                 ':', "conf option value", &p, errp);
281         if (ret < 0) {
282             break;
283         }
284         qemu_rbd_unescape(value);
285 
286         if (strcmp(name, "conf") == 0) {
287             /* read the conf file alone, so it doesn't override more
288                specific settings for a particular device */
289             if (only_read_conf_file) {
290                 ret = rados_conf_read_file(cluster, value);
291                 if (ret < 0) {
292                     error_setg_errno(errp, -ret, "error reading conf file %s",
293                                      value);
294                     break;
295                 }
296             }
297         } else if (strcmp(name, "id") == 0) {
298             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
299         } else if (!only_read_conf_file) {
300             ret = rados_conf_set(cluster, name, value);
301             if (ret < 0) {
302                 error_setg_errno(errp, -ret, "invalid conf option %s", name);
303                 ret = -EINVAL;
304                 break;
305             }
306         }
307     }
308 
309     g_free(buf);
310     return ret;
311 }
312 
313 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
314 {
315     Error *local_err = NULL;
316     int64_t bytes = 0;
317     int64_t objsize;
318     int obj_order = 0;
319     char pool[RBD_MAX_POOL_NAME_SIZE];
320     char name[RBD_MAX_IMAGE_NAME_SIZE];
321     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
322     char conf[RBD_MAX_CONF_SIZE];
323     char clientname_buf[RBD_MAX_CONF_SIZE];
324     char *clientname;
325     const char *secretid;
326     rados_t cluster;
327     rados_ioctx_t io_ctx;
328     int ret;
329 
330     secretid = qemu_opt_get(opts, "password-secret");
331 
332     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
333                            snap_buf, sizeof(snap_buf),
334                            name, sizeof(name),
335                            conf, sizeof(conf), &local_err) < 0) {
336         error_propagate(errp, local_err);
337         return -EINVAL;
338     }
339 
340     /* Read out options */
341     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
342                      BDRV_SECTOR_SIZE);
343     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
344     if (objsize) {
345         if ((objsize - 1) & objsize) {    /* not a power of 2? */
346             error_setg(errp, "obj size needs to be power of 2");
347             return -EINVAL;
348         }
349         if (objsize < 4096) {
350             error_setg(errp, "obj size too small");
351             return -EINVAL;
352         }
353         obj_order = ctz32(objsize);
354     }
355 
356     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
357     ret = rados_create(&cluster, clientname);
358     if (ret < 0) {
359         error_setg_errno(errp, -ret, "error initializing");
360         return ret;
361     }
362 
363     if (strstr(conf, "conf=") == NULL) {
364         /* try default location, but ignore failure */
365         rados_conf_read_file(cluster, NULL);
366     } else if (conf[0] != '\0' &&
367                qemu_rbd_set_conf(cluster, conf, true, &local_err) < 0) {
368         error_propagate(errp, local_err);
369         ret = -EIO;
370         goto shutdown;
371     }
372 
373     if (conf[0] != '\0' &&
374         qemu_rbd_set_conf(cluster, conf, false, &local_err) < 0) {
375         error_propagate(errp, local_err);
376         ret = -EIO;
377         goto shutdown;
378     }
379 
380     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
381         ret = -EIO;
382         goto shutdown;
383     }
384 
385     ret = rados_connect(cluster);
386     if (ret < 0) {
387         error_setg_errno(errp, -ret, "error connecting");
388         goto shutdown;
389     }
390 
391     ret = rados_ioctx_create(cluster, pool, &io_ctx);
392     if (ret < 0) {
393         error_setg_errno(errp, -ret, "error opening pool %s", pool);
394         goto shutdown;
395     }
396 
397     ret = rbd_create(io_ctx, name, bytes, &obj_order);
398     if (ret < 0) {
399         error_setg_errno(errp, -ret, "error rbd create");
400     }
401 
402     rados_ioctx_destroy(io_ctx);
403 
404 shutdown:
405     rados_shutdown(cluster);
406     return ret;
407 }
408 
409 /*
410  * This aio completion is being called from rbd_finish_bh() and runs in qemu
411  * BH context.
412  */
413 static void qemu_rbd_complete_aio(RADOSCB *rcb)
414 {
415     RBDAIOCB *acb = rcb->acb;
416     int64_t r;
417 
418     r = rcb->ret;
419 
420     if (acb->cmd != RBD_AIO_READ) {
421         if (r < 0) {
422             acb->ret = r;
423             acb->error = 1;
424         } else if (!acb->error) {
425             acb->ret = rcb->size;
426         }
427     } else {
428         if (r < 0) {
429             memset(rcb->buf, 0, rcb->size);
430             acb->ret = r;
431             acb->error = 1;
432         } else if (r < rcb->size) {
433             memset(rcb->buf + r, 0, rcb->size - r);
434             if (!acb->error) {
435                 acb->ret = rcb->size;
436             }
437         } else if (!acb->error) {
438             acb->ret = r;
439         }
440     }
441 
442     g_free(rcb);
443 
444     if (acb->cmd == RBD_AIO_READ) {
445         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
446     }
447     qemu_vfree(acb->bounce);
448     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
449 
450     qemu_aio_unref(acb);
451 }
452 
453 /* TODO Convert to fine grained options */
454 static QemuOptsList runtime_opts = {
455     .name = "rbd",
456     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
457     .desc = {
458         {
459             .name = "filename",
460             .type = QEMU_OPT_STRING,
461             .help = "Specification of the rbd image",
462         },
463         {
464             .name = "password-secret",
465             .type = QEMU_OPT_STRING,
466             .help = "ID of secret providing the password",
467         },
468         { /* end of list */ }
469     },
470 };
471 
472 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
473                          Error **errp)
474 {
475     BDRVRBDState *s = bs->opaque;
476     char pool[RBD_MAX_POOL_NAME_SIZE];
477     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
478     char conf[RBD_MAX_CONF_SIZE];
479     char clientname_buf[RBD_MAX_CONF_SIZE];
480     char *clientname;
481     const char *secretid;
482     QemuOpts *opts;
483     Error *local_err = NULL;
484     const char *filename;
485     int r;
486 
487     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
488     qemu_opts_absorb_qdict(opts, options, &local_err);
489     if (local_err) {
490         error_propagate(errp, local_err);
491         qemu_opts_del(opts);
492         return -EINVAL;
493     }
494 
495     filename = qemu_opt_get(opts, "filename");
496     secretid = qemu_opt_get(opts, "password-secret");
497 
498     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
499                            snap_buf, sizeof(snap_buf),
500                            s->name, sizeof(s->name),
501                            conf, sizeof(conf), errp) < 0) {
502         r = -EINVAL;
503         goto failed_opts;
504     }
505 
506     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
507     r = rados_create(&s->cluster, clientname);
508     if (r < 0) {
509         error_setg_errno(errp, -r, "error initializing");
510         goto failed_opts;
511     }
512 
513     s->snap = NULL;
514     if (snap_buf[0] != '\0') {
515         s->snap = g_strdup(snap_buf);
516     }
517 
518     if (strstr(conf, "conf=") == NULL) {
519         /* try default location, but ignore failure */
520         rados_conf_read_file(s->cluster, NULL);
521     } else if (conf[0] != '\0') {
522         r = qemu_rbd_set_conf(s->cluster, conf, true, errp);
523         if (r < 0) {
524             goto failed_shutdown;
525         }
526     }
527 
528     if (conf[0] != '\0') {
529         r = qemu_rbd_set_conf(s->cluster, conf, false, errp);
530         if (r < 0) {
531             goto failed_shutdown;
532         }
533     }
534 
535     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
536         r = -EIO;
537         goto failed_shutdown;
538     }
539 
540     /*
541      * Fallback to more conservative semantics if setting cache
542      * options fails. Ignore errors from setting rbd_cache because the
543      * only possible error is that the option does not exist, and
544      * librbd defaults to no caching. If write through caching cannot
545      * be set up, fall back to no caching.
546      */
547     if (flags & BDRV_O_NOCACHE) {
548         rados_conf_set(s->cluster, "rbd_cache", "false");
549     } else {
550         rados_conf_set(s->cluster, "rbd_cache", "true");
551     }
552 
553     r = rados_connect(s->cluster);
554     if (r < 0) {
555         error_setg_errno(errp, -r, "error connecting");
556         goto failed_shutdown;
557     }
558 
559     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
560     if (r < 0) {
561         error_setg_errno(errp, -r, "error opening pool %s", pool);
562         goto failed_shutdown;
563     }
564 
565     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
566     if (r < 0) {
567         error_setg_errno(errp, -r, "error reading header from %s", s->name);
568         goto failed_open;
569     }
570 
571     bs->read_only = (s->snap != NULL);
572 
573     qemu_opts_del(opts);
574     return 0;
575 
576 failed_open:
577     rados_ioctx_destroy(s->io_ctx);
578 failed_shutdown:
579     rados_shutdown(s->cluster);
580     g_free(s->snap);
581 failed_opts:
582     qemu_opts_del(opts);
583     return r;
584 }
585 
586 static void qemu_rbd_close(BlockDriverState *bs)
587 {
588     BDRVRBDState *s = bs->opaque;
589 
590     rbd_close(s->image);
591     rados_ioctx_destroy(s->io_ctx);
592     g_free(s->snap);
593     rados_shutdown(s->cluster);
594 }
595 
596 static const AIOCBInfo rbd_aiocb_info = {
597     .aiocb_size = sizeof(RBDAIOCB),
598 };
599 
600 static void rbd_finish_bh(void *opaque)
601 {
602     RADOSCB *rcb = opaque;
603     qemu_rbd_complete_aio(rcb);
604 }
605 
606 /*
607  * This is the callback function for rbd_aio_read and _write
608  *
609  * Note: this function is being called from a non qemu thread so
610  * we need to be careful about what we do here. Generally we only
611  * schedule a BH, and do the rest of the io completion handling
612  * from rbd_finish_bh() which runs in a qemu context.
613  */
614 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
615 {
616     RBDAIOCB *acb = rcb->acb;
617 
618     rcb->ret = rbd_aio_get_return_value(c);
619     rbd_aio_release(c);
620 
621     aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
622                             rbd_finish_bh, rcb);
623 }
624 
625 static int rbd_aio_discard_wrapper(rbd_image_t image,
626                                    uint64_t off,
627                                    uint64_t len,
628                                    rbd_completion_t comp)
629 {
630 #ifdef LIBRBD_SUPPORTS_DISCARD
631     return rbd_aio_discard(image, off, len, comp);
632 #else
633     return -ENOTSUP;
634 #endif
635 }
636 
637 static int rbd_aio_flush_wrapper(rbd_image_t image,
638                                  rbd_completion_t comp)
639 {
640 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
641     return rbd_aio_flush(image, comp);
642 #else
643     return -ENOTSUP;
644 #endif
645 }
646 
647 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
648                                  int64_t off,
649                                  QEMUIOVector *qiov,
650                                  int64_t size,
651                                  BlockCompletionFunc *cb,
652                                  void *opaque,
653                                  RBDAIOCmd cmd)
654 {
655     RBDAIOCB *acb;
656     RADOSCB *rcb = NULL;
657     rbd_completion_t c;
658     char *buf;
659     int r;
660 
661     BDRVRBDState *s = bs->opaque;
662 
663     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
664     acb->cmd = cmd;
665     acb->qiov = qiov;
666     assert(!qiov || qiov->size == size);
667     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
668         acb->bounce = NULL;
669     } else {
670         acb->bounce = qemu_try_blockalign(bs, qiov->size);
671         if (acb->bounce == NULL) {
672             goto failed;
673         }
674     }
675     acb->ret = 0;
676     acb->error = 0;
677     acb->s = s;
678 
679     if (cmd == RBD_AIO_WRITE) {
680         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
681     }
682 
683     buf = acb->bounce;
684 
685     rcb = g_new(RADOSCB, 1);
686     rcb->acb = acb;
687     rcb->buf = buf;
688     rcb->s = acb->s;
689     rcb->size = size;
690     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
691     if (r < 0) {
692         goto failed;
693     }
694 
695     switch (cmd) {
696     case RBD_AIO_WRITE:
697         r = rbd_aio_write(s->image, off, size, buf, c);
698         break;
699     case RBD_AIO_READ:
700         r = rbd_aio_read(s->image, off, size, buf, c);
701         break;
702     case RBD_AIO_DISCARD:
703         r = rbd_aio_discard_wrapper(s->image, off, size, c);
704         break;
705     case RBD_AIO_FLUSH:
706         r = rbd_aio_flush_wrapper(s->image, c);
707         break;
708     default:
709         r = -EINVAL;
710     }
711 
712     if (r < 0) {
713         goto failed_completion;
714     }
715 
716     return &acb->common;
717 
718 failed_completion:
719     rbd_aio_release(c);
720 failed:
721     g_free(rcb);
722     qemu_vfree(acb->bounce);
723     qemu_aio_unref(acb);
724     return NULL;
725 }
726 
727 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
728                                       int64_t sector_num,
729                                       QEMUIOVector *qiov,
730                                       int nb_sectors,
731                                       BlockCompletionFunc *cb,
732                                       void *opaque)
733 {
734     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
735                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
736                          RBD_AIO_READ);
737 }
738 
739 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
740                                        int64_t sector_num,
741                                        QEMUIOVector *qiov,
742                                        int nb_sectors,
743                                        BlockCompletionFunc *cb,
744                                        void *opaque)
745 {
746     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
747                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
748                          RBD_AIO_WRITE);
749 }
750 
751 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
752 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
753                                       BlockCompletionFunc *cb,
754                                       void *opaque)
755 {
756     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
757 }
758 
759 #else
760 
761 static int qemu_rbd_co_flush(BlockDriverState *bs)
762 {
763 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
764     /* rbd_flush added in 0.1.1 */
765     BDRVRBDState *s = bs->opaque;
766     return rbd_flush(s->image);
767 #else
768     return 0;
769 #endif
770 }
771 #endif
772 
773 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
774 {
775     BDRVRBDState *s = bs->opaque;
776     rbd_image_info_t info;
777     int r;
778 
779     r = rbd_stat(s->image, &info, sizeof(info));
780     if (r < 0) {
781         return r;
782     }
783 
784     bdi->cluster_size = info.obj_size;
785     return 0;
786 }
787 
788 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
789 {
790     BDRVRBDState *s = bs->opaque;
791     rbd_image_info_t info;
792     int r;
793 
794     r = rbd_stat(s->image, &info, sizeof(info));
795     if (r < 0) {
796         return r;
797     }
798 
799     return info.size;
800 }
801 
802 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
803 {
804     BDRVRBDState *s = bs->opaque;
805     int r;
806 
807     r = rbd_resize(s->image, offset);
808     if (r < 0) {
809         return r;
810     }
811 
812     return 0;
813 }
814 
815 static int qemu_rbd_snap_create(BlockDriverState *bs,
816                                 QEMUSnapshotInfo *sn_info)
817 {
818     BDRVRBDState *s = bs->opaque;
819     int r;
820 
821     if (sn_info->name[0] == '\0') {
822         return -EINVAL; /* we need a name for rbd snapshots */
823     }
824 
825     /*
826      * rbd snapshots are using the name as the user controlled unique identifier
827      * we can't use the rbd snapid for that purpose, as it can't be set
828      */
829     if (sn_info->id_str[0] != '\0' &&
830         strcmp(sn_info->id_str, sn_info->name) != 0) {
831         return -EINVAL;
832     }
833 
834     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
835         return -ERANGE;
836     }
837 
838     r = rbd_snap_create(s->image, sn_info->name);
839     if (r < 0) {
840         error_report("failed to create snap: %s", strerror(-r));
841         return r;
842     }
843 
844     return 0;
845 }
846 
847 static int qemu_rbd_snap_remove(BlockDriverState *bs,
848                                 const char *snapshot_id,
849                                 const char *snapshot_name,
850                                 Error **errp)
851 {
852     BDRVRBDState *s = bs->opaque;
853     int r;
854 
855     if (!snapshot_name) {
856         error_setg(errp, "rbd need a valid snapshot name");
857         return -EINVAL;
858     }
859 
860     /* If snapshot_id is specified, it must be equal to name, see
861        qemu_rbd_snap_list() */
862     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
863         error_setg(errp,
864                    "rbd do not support snapshot id, it should be NULL or "
865                    "equal to snapshot name");
866         return -EINVAL;
867     }
868 
869     r = rbd_snap_remove(s->image, snapshot_name);
870     if (r < 0) {
871         error_setg_errno(errp, -r, "Failed to remove the snapshot");
872     }
873     return r;
874 }
875 
876 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
877                                   const char *snapshot_name)
878 {
879     BDRVRBDState *s = bs->opaque;
880 
881     return rbd_snap_rollback(s->image, snapshot_name);
882 }
883 
884 static int qemu_rbd_snap_list(BlockDriverState *bs,
885                               QEMUSnapshotInfo **psn_tab)
886 {
887     BDRVRBDState *s = bs->opaque;
888     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
889     int i, snap_count;
890     rbd_snap_info_t *snaps;
891     int max_snaps = RBD_MAX_SNAPS;
892 
893     do {
894         snaps = g_new(rbd_snap_info_t, max_snaps);
895         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
896         if (snap_count <= 0) {
897             g_free(snaps);
898         }
899     } while (snap_count == -ERANGE);
900 
901     if (snap_count <= 0) {
902         goto done;
903     }
904 
905     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
906 
907     for (i = 0; i < snap_count; i++) {
908         const char *snap_name = snaps[i].name;
909 
910         sn_info = sn_tab + i;
911         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
912         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
913 
914         sn_info->vm_state_size = snaps[i].size;
915         sn_info->date_sec = 0;
916         sn_info->date_nsec = 0;
917         sn_info->vm_clock_nsec = 0;
918     }
919     rbd_snap_list_end(snaps);
920     g_free(snaps);
921 
922  done:
923     *psn_tab = sn_tab;
924     return snap_count;
925 }
926 
927 #ifdef LIBRBD_SUPPORTS_DISCARD
928 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
929                                          int64_t offset,
930                                          int count,
931                                          BlockCompletionFunc *cb,
932                                          void *opaque)
933 {
934     return rbd_start_aio(bs, offset, NULL, count, cb, opaque,
935                          RBD_AIO_DISCARD);
936 }
937 #endif
938 
939 #ifdef LIBRBD_SUPPORTS_INVALIDATE
940 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
941                                       Error **errp)
942 {
943     BDRVRBDState *s = bs->opaque;
944     int r = rbd_invalidate_cache(s->image);
945     if (r < 0) {
946         error_setg_errno(errp, -r, "Failed to invalidate the cache");
947     }
948 }
949 #endif
950 
951 static QemuOptsList qemu_rbd_create_opts = {
952     .name = "rbd-create-opts",
953     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
954     .desc = {
955         {
956             .name = BLOCK_OPT_SIZE,
957             .type = QEMU_OPT_SIZE,
958             .help = "Virtual disk size"
959         },
960         {
961             .name = BLOCK_OPT_CLUSTER_SIZE,
962             .type = QEMU_OPT_SIZE,
963             .help = "RBD object size"
964         },
965         {
966             .name = "password-secret",
967             .type = QEMU_OPT_STRING,
968             .help = "ID of secret providing the password",
969         },
970         { /* end of list */ }
971     }
972 };
973 
974 static BlockDriver bdrv_rbd = {
975     .format_name        = "rbd",
976     .instance_size      = sizeof(BDRVRBDState),
977     .bdrv_needs_filename = true,
978     .bdrv_file_open     = qemu_rbd_open,
979     .bdrv_close         = qemu_rbd_close,
980     .bdrv_create        = qemu_rbd_create,
981     .bdrv_has_zero_init = bdrv_has_zero_init_1,
982     .bdrv_get_info      = qemu_rbd_getinfo,
983     .create_opts        = &qemu_rbd_create_opts,
984     .bdrv_getlength     = qemu_rbd_getlength,
985     .bdrv_truncate      = qemu_rbd_truncate,
986     .protocol_name      = "rbd",
987 
988     .bdrv_aio_readv         = qemu_rbd_aio_readv,
989     .bdrv_aio_writev        = qemu_rbd_aio_writev,
990 
991 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
992     .bdrv_aio_flush         = qemu_rbd_aio_flush,
993 #else
994     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
995 #endif
996 
997 #ifdef LIBRBD_SUPPORTS_DISCARD
998     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
999 #endif
1000 
1001     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1002     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1003     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1004     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1005 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1006     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1007 #endif
1008 };
1009 
1010 static void bdrv_rbd_init(void)
1011 {
1012     bdrv_register(&bdrv_rbd);
1013 }
1014 
1015 block_init(bdrv_rbd_init);
1016