xref: /openbmc/qemu/block/rbd.c (revision e6e03dcf)
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 <rbd/librbd.h>
17 #include "qapi/error.h"
18 #include "qemu/error-report.h"
19 #include "qemu/module.h"
20 #include "qemu/option.h"
21 #include "block/block_int.h"
22 #include "block/qdict.h"
23 #include "crypto/secret.h"
24 #include "qemu/cutils.h"
25 #include "sysemu/replay.h"
26 #include "qapi/qmp/qstring.h"
27 #include "qapi/qmp/qdict.h"
28 #include "qapi/qmp/qjson.h"
29 #include "qapi/qmp/qlist.h"
30 #include "qapi/qobject-input-visitor.h"
31 #include "qapi/qapi-visit-block-core.h"
32 
33 /*
34  * When specifying the image filename use:
35  *
36  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
37  *
38  * poolname must be the name of an existing rados pool.
39  *
40  * devicename is the name of the rbd image.
41  *
42  * Each option given is used to configure rados, and may be any valid
43  * Ceph option, "id", or "conf".
44  *
45  * The "id" option indicates what user we should authenticate as to
46  * the Ceph cluster.  If it is excluded we will use the Ceph default
47  * (normally 'admin').
48  *
49  * The "conf" option specifies a Ceph configuration file to read.  If
50  * it is not specified, we will read from the default Ceph locations
51  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
52  * file, specify conf=/dev/null.
53  *
54  * Configuration values containing :, @, or = can be escaped with a
55  * leading "\".
56  */
57 
58 /* rbd_aio_discard added in 0.1.2 */
59 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
60 #define LIBRBD_SUPPORTS_DISCARD
61 #else
62 #undef LIBRBD_SUPPORTS_DISCARD
63 #endif
64 
65 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
66 
67 #define RBD_MAX_SNAPS 100
68 
69 /* The LIBRBD_SUPPORTS_IOVEC is defined in librbd.h */
70 #ifdef LIBRBD_SUPPORTS_IOVEC
71 #define LIBRBD_USE_IOVEC 1
72 #else
73 #define LIBRBD_USE_IOVEC 0
74 #endif
75 
76 typedef enum {
77     RBD_AIO_READ,
78     RBD_AIO_WRITE,
79     RBD_AIO_DISCARD,
80     RBD_AIO_FLUSH
81 } RBDAIOCmd;
82 
83 typedef struct RBDAIOCB {
84     BlockAIOCB common;
85     int64_t ret;
86     QEMUIOVector *qiov;
87     char *bounce;
88     RBDAIOCmd cmd;
89     int error;
90     struct BDRVRBDState *s;
91 } RBDAIOCB;
92 
93 typedef struct RADOSCB {
94     RBDAIOCB *acb;
95     struct BDRVRBDState *s;
96     int64_t size;
97     char *buf;
98     int64_t ret;
99 } RADOSCB;
100 
101 typedef struct BDRVRBDState {
102     rados_t cluster;
103     rados_ioctx_t io_ctx;
104     rbd_image_t image;
105     char *image_name;
106     char *snap;
107     uint64_t image_size;
108 } BDRVRBDState;
109 
110 static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx,
111                             BlockdevOptionsRbd *opts, bool cache,
112                             const char *keypairs, const char *secretid,
113                             Error **errp);
114 
115 static char *qemu_rbd_next_tok(char *src, char delim, char **p)
116 {
117     char *end;
118 
119     *p = NULL;
120 
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     return src;
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 void qemu_rbd_parse_filename(const char *filename, QDict *options,
150                                     Error **errp)
151 {
152     const char *start;
153     char *p, *buf;
154     QList *keypairs = NULL;
155     char *found_str;
156 
157     if (!strstart(filename, "rbd:", &start)) {
158         error_setg(errp, "File name must start with 'rbd:'");
159         return;
160     }
161 
162     buf = g_strdup(start);
163     p = buf;
164 
165     found_str = qemu_rbd_next_tok(p, '/', &p);
166     if (!p) {
167         error_setg(errp, "Pool name is required");
168         goto done;
169     }
170     qemu_rbd_unescape(found_str);
171     qdict_put_str(options, "pool", found_str);
172 
173     if (strchr(p, '@')) {
174         found_str = qemu_rbd_next_tok(p, '@', &p);
175         qemu_rbd_unescape(found_str);
176         qdict_put_str(options, "image", found_str);
177 
178         found_str = qemu_rbd_next_tok(p, ':', &p);
179         qemu_rbd_unescape(found_str);
180         qdict_put_str(options, "snapshot", found_str);
181     } else {
182         found_str = qemu_rbd_next_tok(p, ':', &p);
183         qemu_rbd_unescape(found_str);
184         qdict_put_str(options, "image", found_str);
185     }
186     if (!p) {
187         goto done;
188     }
189 
190     /* The following are essentially all key/value pairs, and we treat
191      * 'id' and 'conf' a bit special.  Key/value pairs may be in any order. */
192     while (p) {
193         char *name, *value;
194         name = qemu_rbd_next_tok(p, '=', &p);
195         if (!p) {
196             error_setg(errp, "conf option %s has no value", name);
197             break;
198         }
199 
200         qemu_rbd_unescape(name);
201 
202         value = qemu_rbd_next_tok(p, ':', &p);
203         qemu_rbd_unescape(value);
204 
205         if (!strcmp(name, "conf")) {
206             qdict_put_str(options, "conf", value);
207         } else if (!strcmp(name, "id")) {
208             qdict_put_str(options, "user", value);
209         } else {
210             /*
211              * We pass these internally to qemu_rbd_set_keypairs(), so
212              * we can get away with the simpler list of [ "key1",
213              * "value1", "key2", "value2" ] rather than a raw dict
214              * { "key1": "value1", "key2": "value2" } where we can't
215              * guarantee order, or even a more correct but complex
216              * [ { "key1": "value1" }, { "key2": "value2" } ]
217              */
218             if (!keypairs) {
219                 keypairs = qlist_new();
220             }
221             qlist_append_str(keypairs, name);
222             qlist_append_str(keypairs, value);
223         }
224     }
225 
226     if (keypairs) {
227         qdict_put(options, "=keyvalue-pairs",
228                   qobject_to_json(QOBJECT(keypairs)));
229     }
230 
231 done:
232     g_free(buf);
233     qobject_unref(keypairs);
234     return;
235 }
236 
237 
238 static void qemu_rbd_refresh_limits(BlockDriverState *bs, Error **errp)
239 {
240     /* XXX Does RBD support AIO on less than 512-byte alignment? */
241     bs->bl.request_alignment = 512;
242 }
243 
244 
245 static int qemu_rbd_set_auth(rados_t cluster, BlockdevOptionsRbd *opts,
246                              Error **errp)
247 {
248     char *key, *acr;
249     int r;
250     GString *accu;
251     RbdAuthModeList *auth;
252 
253     if (opts->key_secret) {
254         key = qcrypto_secret_lookup_as_base64(opts->key_secret, errp);
255         if (!key) {
256             return -EIO;
257         }
258         r = rados_conf_set(cluster, "key", key);
259         g_free(key);
260         if (r < 0) {
261             error_setg_errno(errp, -r, "Could not set 'key'");
262             return r;
263         }
264     }
265 
266     if (opts->has_auth_client_required) {
267         accu = g_string_new("");
268         for (auth = opts->auth_client_required; auth; auth = auth->next) {
269             if (accu->str[0]) {
270                 g_string_append_c(accu, ';');
271             }
272             g_string_append(accu, RbdAuthMode_str(auth->value));
273         }
274         acr = g_string_free(accu, FALSE);
275         r = rados_conf_set(cluster, "auth_client_required", acr);
276         g_free(acr);
277         if (r < 0) {
278             error_setg_errno(errp, -r,
279                              "Could not set 'auth_client_required'");
280             return r;
281         }
282     }
283 
284     return 0;
285 }
286 
287 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs_json,
288                                  Error **errp)
289 {
290     QList *keypairs;
291     QString *name;
292     QString *value;
293     const char *key;
294     size_t remaining;
295     int ret = 0;
296 
297     if (!keypairs_json) {
298         return ret;
299     }
300     keypairs = qobject_to(QList,
301                           qobject_from_json(keypairs_json, &error_abort));
302     remaining = qlist_size(keypairs) / 2;
303     assert(remaining);
304 
305     while (remaining--) {
306         name = qobject_to(QString, qlist_pop(keypairs));
307         value = qobject_to(QString, qlist_pop(keypairs));
308         assert(name && value);
309         key = qstring_get_str(name);
310 
311         ret = rados_conf_set(cluster, key, qstring_get_str(value));
312         qobject_unref(value);
313         if (ret < 0) {
314             error_setg_errno(errp, -ret, "invalid conf option %s", key);
315             qobject_unref(name);
316             ret = -EINVAL;
317             break;
318         }
319         qobject_unref(name);
320     }
321 
322     qobject_unref(keypairs);
323     return ret;
324 }
325 
326 static void qemu_rbd_memset(RADOSCB *rcb, int64_t offs)
327 {
328     if (LIBRBD_USE_IOVEC) {
329         RBDAIOCB *acb = rcb->acb;
330         iov_memset(acb->qiov->iov, acb->qiov->niov, offs, 0,
331                    acb->qiov->size - offs);
332     } else {
333         memset(rcb->buf + offs, 0, rcb->size - offs);
334     }
335 }
336 
337 static QemuOptsList runtime_opts = {
338     .name = "rbd",
339     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
340     .desc = {
341         {
342             .name = "pool",
343             .type = QEMU_OPT_STRING,
344             .help = "Rados pool name",
345         },
346         {
347             .name = "image",
348             .type = QEMU_OPT_STRING,
349             .help = "Image name in the pool",
350         },
351         {
352             .name = "conf",
353             .type = QEMU_OPT_STRING,
354             .help = "Rados config file location",
355         },
356         {
357             .name = "snapshot",
358             .type = QEMU_OPT_STRING,
359             .help = "Ceph snapshot name",
360         },
361         {
362             /* maps to 'id' in rados_create() */
363             .name = "user",
364             .type = QEMU_OPT_STRING,
365             .help = "Rados id name",
366         },
367         /*
368          * server.* extracted manually, see qemu_rbd_mon_host()
369          */
370         { /* end of list */ }
371     },
372 };
373 
374 /* FIXME Deprecate and remove keypairs or make it available in QMP. */
375 static int qemu_rbd_do_create(BlockdevCreateOptions *options,
376                               const char *keypairs, const char *password_secret,
377                               Error **errp)
378 {
379     BlockdevCreateOptionsRbd *opts = &options->u.rbd;
380     rados_t cluster;
381     rados_ioctx_t io_ctx;
382     int obj_order = 0;
383     int ret;
384 
385     assert(options->driver == BLOCKDEV_DRIVER_RBD);
386     if (opts->location->has_snapshot) {
387         error_setg(errp, "Can't use snapshot name for image creation");
388         return -EINVAL;
389     }
390 
391     if (opts->has_cluster_size) {
392         int64_t objsize = opts->cluster_size;
393         if ((objsize - 1) & objsize) {    /* not a power of 2? */
394             error_setg(errp, "obj size needs to be power of 2");
395             return -EINVAL;
396         }
397         if (objsize < 4096) {
398             error_setg(errp, "obj size too small");
399             return -EINVAL;
400         }
401         obj_order = ctz32(objsize);
402     }
403 
404     ret = qemu_rbd_connect(&cluster, &io_ctx, opts->location, false, keypairs,
405                            password_secret, errp);
406     if (ret < 0) {
407         return ret;
408     }
409 
410     ret = rbd_create(io_ctx, opts->location->image, opts->size, &obj_order);
411     if (ret < 0) {
412         error_setg_errno(errp, -ret, "error rbd create");
413         goto out;
414     }
415 
416     ret = 0;
417 out:
418     rados_ioctx_destroy(io_ctx);
419     rados_shutdown(cluster);
420     return ret;
421 }
422 
423 static int qemu_rbd_co_create(BlockdevCreateOptions *options, Error **errp)
424 {
425     return qemu_rbd_do_create(options, NULL, NULL, errp);
426 }
427 
428 static int coroutine_fn qemu_rbd_co_create_opts(const char *filename,
429                                                 QemuOpts *opts,
430                                                 Error **errp)
431 {
432     BlockdevCreateOptions *create_options;
433     BlockdevCreateOptionsRbd *rbd_opts;
434     BlockdevOptionsRbd *loc;
435     Error *local_err = NULL;
436     const char *keypairs, *password_secret;
437     QDict *options = NULL;
438     int ret = 0;
439 
440     create_options = g_new0(BlockdevCreateOptions, 1);
441     create_options->driver = BLOCKDEV_DRIVER_RBD;
442     rbd_opts = &create_options->u.rbd;
443 
444     rbd_opts->location = g_new0(BlockdevOptionsRbd, 1);
445 
446     password_secret = qemu_opt_get(opts, "password-secret");
447 
448     /* Read out options */
449     rbd_opts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
450                               BDRV_SECTOR_SIZE);
451     rbd_opts->cluster_size = qemu_opt_get_size_del(opts,
452                                                    BLOCK_OPT_CLUSTER_SIZE, 0);
453     rbd_opts->has_cluster_size = (rbd_opts->cluster_size != 0);
454 
455     options = qdict_new();
456     qemu_rbd_parse_filename(filename, options, &local_err);
457     if (local_err) {
458         ret = -EINVAL;
459         error_propagate(errp, local_err);
460         goto exit;
461     }
462 
463     /*
464      * Caution: while qdict_get_try_str() is fine, getting non-string
465      * types would require more care.  When @options come from -blockdev
466      * or blockdev_add, its members are typed according to the QAPI
467      * schema, but when they come from -drive, they're all QString.
468      */
469     loc = rbd_opts->location;
470     loc->pool     = g_strdup(qdict_get_try_str(options, "pool"));
471     loc->conf     = g_strdup(qdict_get_try_str(options, "conf"));
472     loc->has_conf = !!loc->conf;
473     loc->user     = g_strdup(qdict_get_try_str(options, "user"));
474     loc->has_user = !!loc->user;
475     loc->image    = g_strdup(qdict_get_try_str(options, "image"));
476     keypairs      = qdict_get_try_str(options, "=keyvalue-pairs");
477 
478     ret = qemu_rbd_do_create(create_options, keypairs, password_secret, errp);
479     if (ret < 0) {
480         goto exit;
481     }
482 
483 exit:
484     qobject_unref(options);
485     qapi_free_BlockdevCreateOptions(create_options);
486     return ret;
487 }
488 
489 /*
490  * This aio completion is being called from rbd_finish_bh() and runs in qemu
491  * BH context.
492  */
493 static void qemu_rbd_complete_aio(RADOSCB *rcb)
494 {
495     RBDAIOCB *acb = rcb->acb;
496     int64_t r;
497 
498     r = rcb->ret;
499 
500     if (acb->cmd != RBD_AIO_READ) {
501         if (r < 0) {
502             acb->ret = r;
503             acb->error = 1;
504         } else if (!acb->error) {
505             acb->ret = rcb->size;
506         }
507     } else {
508         if (r < 0) {
509             qemu_rbd_memset(rcb, 0);
510             acb->ret = r;
511             acb->error = 1;
512         } else if (r < rcb->size) {
513             qemu_rbd_memset(rcb, r);
514             if (!acb->error) {
515                 acb->ret = rcb->size;
516             }
517         } else if (!acb->error) {
518             acb->ret = r;
519         }
520     }
521 
522     g_free(rcb);
523 
524     if (!LIBRBD_USE_IOVEC) {
525         if (acb->cmd == RBD_AIO_READ) {
526             qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
527         }
528         qemu_vfree(acb->bounce);
529     }
530 
531     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
532 
533     qemu_aio_unref(acb);
534 }
535 
536 static char *qemu_rbd_mon_host(BlockdevOptionsRbd *opts, Error **errp)
537 {
538     const char **vals;
539     const char *host, *port;
540     char *rados_str;
541     InetSocketAddressBaseList *p;
542     int i, cnt;
543 
544     if (!opts->has_server) {
545         return NULL;
546     }
547 
548     for (cnt = 0, p = opts->server; p; p = p->next) {
549         cnt++;
550     }
551 
552     vals = g_new(const char *, cnt + 1);
553 
554     for (i = 0, p = opts->server; p; p = p->next, i++) {
555         host = p->value->host;
556         port = p->value->port;
557 
558         if (strchr(host, ':')) {
559             vals[i] = g_strdup_printf("[%s]:%s", host, port);
560         } else {
561             vals[i] = g_strdup_printf("%s:%s", host, port);
562         }
563     }
564     vals[i] = NULL;
565 
566     rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
567     g_strfreev((char **)vals);
568     return rados_str;
569 }
570 
571 static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx,
572                             BlockdevOptionsRbd *opts, bool cache,
573                             const char *keypairs, const char *secretid,
574                             Error **errp)
575 {
576     char *mon_host = NULL;
577     Error *local_err = NULL;
578     int r;
579 
580     if (secretid) {
581         if (opts->key_secret) {
582             error_setg(errp,
583                        "Legacy 'password-secret' clashes with 'key-secret'");
584             return -EINVAL;
585         }
586         opts->key_secret = g_strdup(secretid);
587         opts->has_key_secret = true;
588     }
589 
590     mon_host = qemu_rbd_mon_host(opts, &local_err);
591     if (local_err) {
592         error_propagate(errp, local_err);
593         r = -EINVAL;
594         goto failed_opts;
595     }
596 
597     r = rados_create(cluster, opts->user);
598     if (r < 0) {
599         error_setg_errno(errp, -r, "error initializing");
600         goto failed_opts;
601     }
602 
603     /* try default location when conf=NULL, but ignore failure */
604     r = rados_conf_read_file(*cluster, opts->conf);
605     if (opts->has_conf && r < 0) {
606         error_setg_errno(errp, -r, "error reading conf file %s", opts->conf);
607         goto failed_shutdown;
608     }
609 
610     r = qemu_rbd_set_keypairs(*cluster, keypairs, errp);
611     if (r < 0) {
612         goto failed_shutdown;
613     }
614 
615     if (mon_host) {
616         r = rados_conf_set(*cluster, "mon_host", mon_host);
617         if (r < 0) {
618             goto failed_shutdown;
619         }
620     }
621 
622     r = qemu_rbd_set_auth(*cluster, opts, errp);
623     if (r < 0) {
624         goto failed_shutdown;
625     }
626 
627     /*
628      * Fallback to more conservative semantics if setting cache
629      * options fails. Ignore errors from setting rbd_cache because the
630      * only possible error is that the option does not exist, and
631      * librbd defaults to no caching. If write through caching cannot
632      * be set up, fall back to no caching.
633      */
634     if (cache) {
635         rados_conf_set(*cluster, "rbd_cache", "true");
636     } else {
637         rados_conf_set(*cluster, "rbd_cache", "false");
638     }
639 
640     r = rados_connect(*cluster);
641     if (r < 0) {
642         error_setg_errno(errp, -r, "error connecting");
643         goto failed_shutdown;
644     }
645 
646     r = rados_ioctx_create(*cluster, opts->pool, io_ctx);
647     if (r < 0) {
648         error_setg_errno(errp, -r, "error opening pool %s", opts->pool);
649         goto failed_shutdown;
650     }
651 
652     return 0;
653 
654 failed_shutdown:
655     rados_shutdown(*cluster);
656 failed_opts:
657     g_free(mon_host);
658     return r;
659 }
660 
661 static int qemu_rbd_convert_options(QDict *options, BlockdevOptionsRbd **opts,
662                                     Error **errp)
663 {
664     Visitor *v;
665     Error *local_err = NULL;
666 
667     /* Convert the remaining options into a QAPI object */
668     v = qobject_input_visitor_new_flat_confused(options, errp);
669     if (!v) {
670         return -EINVAL;
671     }
672 
673     visit_type_BlockdevOptionsRbd(v, NULL, opts, &local_err);
674     visit_free(v);
675 
676     if (local_err) {
677         error_propagate(errp, local_err);
678         return -EINVAL;
679     }
680 
681     return 0;
682 }
683 
684 static int qemu_rbd_attempt_legacy_options(QDict *options,
685                                            BlockdevOptionsRbd **opts,
686                                            char **keypairs)
687 {
688     char *filename;
689     int r;
690 
691     filename = g_strdup(qdict_get_try_str(options, "filename"));
692     if (!filename) {
693         return -EINVAL;
694     }
695     qdict_del(options, "filename");
696 
697     qemu_rbd_parse_filename(filename, options, NULL);
698 
699     /* keypairs freed by caller */
700     *keypairs = g_strdup(qdict_get_try_str(options, "=keyvalue-pairs"));
701     if (*keypairs) {
702         qdict_del(options, "=keyvalue-pairs");
703     }
704 
705     r = qemu_rbd_convert_options(options, opts, NULL);
706 
707     g_free(filename);
708     return r;
709 }
710 
711 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
712                          Error **errp)
713 {
714     BDRVRBDState *s = bs->opaque;
715     BlockdevOptionsRbd *opts = NULL;
716     const QDictEntry *e;
717     Error *local_err = NULL;
718     char *keypairs, *secretid;
719     int r;
720 
721     keypairs = g_strdup(qdict_get_try_str(options, "=keyvalue-pairs"));
722     if (keypairs) {
723         qdict_del(options, "=keyvalue-pairs");
724     }
725 
726     secretid = g_strdup(qdict_get_try_str(options, "password-secret"));
727     if (secretid) {
728         qdict_del(options, "password-secret");
729     }
730 
731     r = qemu_rbd_convert_options(options, &opts, &local_err);
732     if (local_err) {
733         /* If keypairs are present, that means some options are present in
734          * the modern option format.  Don't attempt to parse legacy option
735          * formats, as we won't support mixed usage. */
736         if (keypairs) {
737             error_propagate(errp, local_err);
738             goto out;
739         }
740 
741         /* If the initial attempt to convert and process the options failed,
742          * we may be attempting to open an image file that has the rbd options
743          * specified in the older format consisting of all key/value pairs
744          * encoded in the filename.  Go ahead and attempt to parse the
745          * filename, and see if we can pull out the required options. */
746         r = qemu_rbd_attempt_legacy_options(options, &opts, &keypairs);
747         if (r < 0) {
748             /* Propagate the original error, not the legacy parsing fallback
749              * error, as the latter was just a best-effort attempt. */
750             error_propagate(errp, local_err);
751             goto out;
752         }
753         /* Take care whenever deciding to actually deprecate; once this ability
754          * is removed, we will not be able to open any images with legacy-styled
755          * backing image strings. */
756         warn_report("RBD options encoded in the filename as keyvalue pairs "
757                     "is deprecated");
758     }
759 
760     /* Remove the processed options from the QDict (the visitor processes
761      * _all_ options in the QDict) */
762     while ((e = qdict_first(options))) {
763         qdict_del(options, e->key);
764     }
765 
766     r = qemu_rbd_connect(&s->cluster, &s->io_ctx, opts,
767                          !(flags & BDRV_O_NOCACHE), keypairs, secretid, errp);
768     if (r < 0) {
769         goto out;
770     }
771 
772     s->snap = g_strdup(opts->snapshot);
773     s->image_name = g_strdup(opts->image);
774 
775     /* rbd_open is always r/w */
776     r = rbd_open(s->io_ctx, s->image_name, &s->image, s->snap);
777     if (r < 0) {
778         error_setg_errno(errp, -r, "error reading header from %s",
779                          s->image_name);
780         goto failed_open;
781     }
782 
783     r = rbd_get_size(s->image, &s->image_size);
784     if (r < 0) {
785         error_setg_errno(errp, -r, "error getting image size from %s",
786                          s->image_name);
787         rbd_close(s->image);
788         goto failed_open;
789     }
790 
791     /* If we are using an rbd snapshot, we must be r/o, otherwise
792      * leave as-is */
793     if (s->snap != NULL) {
794         r = bdrv_apply_auto_read_only(bs, "rbd snapshots are read-only", errp);
795         if (r < 0) {
796             rbd_close(s->image);
797             goto failed_open;
798         }
799     }
800 
801     r = 0;
802     goto out;
803 
804 failed_open:
805     rados_ioctx_destroy(s->io_ctx);
806     g_free(s->snap);
807     g_free(s->image_name);
808     rados_shutdown(s->cluster);
809 out:
810     qapi_free_BlockdevOptionsRbd(opts);
811     g_free(keypairs);
812     g_free(secretid);
813     return r;
814 }
815 
816 
817 /* Since RBD is currently always opened R/W via the API,
818  * we just need to check if we are using a snapshot or not, in
819  * order to determine if we will allow it to be R/W */
820 static int qemu_rbd_reopen_prepare(BDRVReopenState *state,
821                                    BlockReopenQueue *queue, Error **errp)
822 {
823     BDRVRBDState *s = state->bs->opaque;
824     int ret = 0;
825 
826     if (s->snap && state->flags & BDRV_O_RDWR) {
827         error_setg(errp,
828                    "Cannot change node '%s' to r/w when using RBD snapshot",
829                    bdrv_get_device_or_node_name(state->bs));
830         ret = -EINVAL;
831     }
832 
833     return ret;
834 }
835 
836 static void qemu_rbd_close(BlockDriverState *bs)
837 {
838     BDRVRBDState *s = bs->opaque;
839 
840     rbd_close(s->image);
841     rados_ioctx_destroy(s->io_ctx);
842     g_free(s->snap);
843     g_free(s->image_name);
844     rados_shutdown(s->cluster);
845 }
846 
847 /* Resize the RBD image and update the 'image_size' with the current size */
848 static int qemu_rbd_resize(BlockDriverState *bs, uint64_t size)
849 {
850     BDRVRBDState *s = bs->opaque;
851     int r;
852 
853     r = rbd_resize(s->image, size);
854     if (r < 0) {
855         return r;
856     }
857 
858     s->image_size = size;
859 
860     return 0;
861 }
862 
863 static const AIOCBInfo rbd_aiocb_info = {
864     .aiocb_size = sizeof(RBDAIOCB),
865 };
866 
867 static void rbd_finish_bh(void *opaque)
868 {
869     RADOSCB *rcb = opaque;
870     qemu_rbd_complete_aio(rcb);
871 }
872 
873 /*
874  * This is the callback function for rbd_aio_read and _write
875  *
876  * Note: this function is being called from a non qemu thread so
877  * we need to be careful about what we do here. Generally we only
878  * schedule a BH, and do the rest of the io completion handling
879  * from rbd_finish_bh() which runs in a qemu context.
880  */
881 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
882 {
883     RBDAIOCB *acb = rcb->acb;
884 
885     rcb->ret = rbd_aio_get_return_value(c);
886     rbd_aio_release(c);
887 
888     replay_bh_schedule_oneshot_event(bdrv_get_aio_context(acb->common.bs),
889                                      rbd_finish_bh, rcb);
890 }
891 
892 static int rbd_aio_discard_wrapper(rbd_image_t image,
893                                    uint64_t off,
894                                    uint64_t len,
895                                    rbd_completion_t comp)
896 {
897 #ifdef LIBRBD_SUPPORTS_DISCARD
898     return rbd_aio_discard(image, off, len, comp);
899 #else
900     return -ENOTSUP;
901 #endif
902 }
903 
904 static int rbd_aio_flush_wrapper(rbd_image_t image,
905                                  rbd_completion_t comp)
906 {
907 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
908     return rbd_aio_flush(image, comp);
909 #else
910     return -ENOTSUP;
911 #endif
912 }
913 
914 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
915                                  int64_t off,
916                                  QEMUIOVector *qiov,
917                                  int64_t size,
918                                  BlockCompletionFunc *cb,
919                                  void *opaque,
920                                  RBDAIOCmd cmd)
921 {
922     RBDAIOCB *acb;
923     RADOSCB *rcb = NULL;
924     rbd_completion_t c;
925     int r;
926 
927     BDRVRBDState *s = bs->opaque;
928 
929     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
930     acb->cmd = cmd;
931     acb->qiov = qiov;
932     assert(!qiov || qiov->size == size);
933 
934     rcb = g_new(RADOSCB, 1);
935 
936     if (!LIBRBD_USE_IOVEC) {
937         if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
938             acb->bounce = NULL;
939         } else {
940             acb->bounce = qemu_try_blockalign(bs, qiov->size);
941             if (acb->bounce == NULL) {
942                 goto failed;
943             }
944         }
945         if (cmd == RBD_AIO_WRITE) {
946             qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
947         }
948         rcb->buf = acb->bounce;
949     }
950 
951     acb->ret = 0;
952     acb->error = 0;
953     acb->s = s;
954 
955     rcb->acb = acb;
956     rcb->s = acb->s;
957     rcb->size = size;
958     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
959     if (r < 0) {
960         goto failed;
961     }
962 
963     switch (cmd) {
964     case RBD_AIO_WRITE: {
965         /*
966          * RBD APIs don't allow us to write more than actual size, so in order
967          * to support growing images, we resize the image before write
968          * operations that exceed the current size.
969          */
970         if (off + size > s->image_size) {
971             r = qemu_rbd_resize(bs, off + size);
972             if (r < 0) {
973                 goto failed_completion;
974             }
975         }
976 #ifdef LIBRBD_SUPPORTS_IOVEC
977             r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
978 #else
979             r = rbd_aio_write(s->image, off, size, rcb->buf, c);
980 #endif
981         break;
982     }
983     case RBD_AIO_READ:
984 #ifdef LIBRBD_SUPPORTS_IOVEC
985             r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
986 #else
987             r = rbd_aio_read(s->image, off, size, rcb->buf, c);
988 #endif
989         break;
990     case RBD_AIO_DISCARD:
991         r = rbd_aio_discard_wrapper(s->image, off, size, c);
992         break;
993     case RBD_AIO_FLUSH:
994         r = rbd_aio_flush_wrapper(s->image, c);
995         break;
996     default:
997         r = -EINVAL;
998     }
999 
1000     if (r < 0) {
1001         goto failed_completion;
1002     }
1003     return &acb->common;
1004 
1005 failed_completion:
1006     rbd_aio_release(c);
1007 failed:
1008     g_free(rcb);
1009     if (!LIBRBD_USE_IOVEC) {
1010         qemu_vfree(acb->bounce);
1011     }
1012 
1013     qemu_aio_unref(acb);
1014     return NULL;
1015 }
1016 
1017 static BlockAIOCB *qemu_rbd_aio_preadv(BlockDriverState *bs,
1018                                        uint64_t offset, uint64_t bytes,
1019                                        QEMUIOVector *qiov, int flags,
1020                                        BlockCompletionFunc *cb,
1021                                        void *opaque)
1022 {
1023     return rbd_start_aio(bs, offset, qiov, bytes, cb, opaque,
1024                          RBD_AIO_READ);
1025 }
1026 
1027 static BlockAIOCB *qemu_rbd_aio_pwritev(BlockDriverState *bs,
1028                                         uint64_t offset, uint64_t bytes,
1029                                         QEMUIOVector *qiov, int flags,
1030                                         BlockCompletionFunc *cb,
1031                                         void *opaque)
1032 {
1033     return rbd_start_aio(bs, offset, qiov, bytes, cb, opaque,
1034                          RBD_AIO_WRITE);
1035 }
1036 
1037 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1038 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
1039                                       BlockCompletionFunc *cb,
1040                                       void *opaque)
1041 {
1042     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
1043 }
1044 
1045 #else
1046 
1047 static int qemu_rbd_co_flush(BlockDriverState *bs)
1048 {
1049 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
1050     /* rbd_flush added in 0.1.1 */
1051     BDRVRBDState *s = bs->opaque;
1052     return rbd_flush(s->image);
1053 #else
1054     return 0;
1055 #endif
1056 }
1057 #endif
1058 
1059 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
1060 {
1061     BDRVRBDState *s = bs->opaque;
1062     rbd_image_info_t info;
1063     int r;
1064 
1065     r = rbd_stat(s->image, &info, sizeof(info));
1066     if (r < 0) {
1067         return r;
1068     }
1069 
1070     bdi->cluster_size = info.obj_size;
1071     return 0;
1072 }
1073 
1074 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
1075 {
1076     BDRVRBDState *s = bs->opaque;
1077     rbd_image_info_t info;
1078     int r;
1079 
1080     r = rbd_stat(s->image, &info, sizeof(info));
1081     if (r < 0) {
1082         return r;
1083     }
1084 
1085     return info.size;
1086 }
1087 
1088 static int coroutine_fn qemu_rbd_co_truncate(BlockDriverState *bs,
1089                                              int64_t offset,
1090                                              PreallocMode prealloc,
1091                                              Error **errp)
1092 {
1093     int r;
1094 
1095     if (prealloc != PREALLOC_MODE_OFF) {
1096         error_setg(errp, "Unsupported preallocation mode '%s'",
1097                    PreallocMode_str(prealloc));
1098         return -ENOTSUP;
1099     }
1100 
1101     r = qemu_rbd_resize(bs, offset);
1102     if (r < 0) {
1103         error_setg_errno(errp, -r, "Failed to resize file");
1104         return r;
1105     }
1106 
1107     return 0;
1108 }
1109 
1110 static int qemu_rbd_snap_create(BlockDriverState *bs,
1111                                 QEMUSnapshotInfo *sn_info)
1112 {
1113     BDRVRBDState *s = bs->opaque;
1114     int r;
1115 
1116     if (sn_info->name[0] == '\0') {
1117         return -EINVAL; /* we need a name for rbd snapshots */
1118     }
1119 
1120     /*
1121      * rbd snapshots are using the name as the user controlled unique identifier
1122      * we can't use the rbd snapid for that purpose, as it can't be set
1123      */
1124     if (sn_info->id_str[0] != '\0' &&
1125         strcmp(sn_info->id_str, sn_info->name) != 0) {
1126         return -EINVAL;
1127     }
1128 
1129     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
1130         return -ERANGE;
1131     }
1132 
1133     r = rbd_snap_create(s->image, sn_info->name);
1134     if (r < 0) {
1135         error_report("failed to create snap: %s", strerror(-r));
1136         return r;
1137     }
1138 
1139     return 0;
1140 }
1141 
1142 static int qemu_rbd_snap_remove(BlockDriverState *bs,
1143                                 const char *snapshot_id,
1144                                 const char *snapshot_name,
1145                                 Error **errp)
1146 {
1147     BDRVRBDState *s = bs->opaque;
1148     int r;
1149 
1150     if (!snapshot_name) {
1151         error_setg(errp, "rbd need a valid snapshot name");
1152         return -EINVAL;
1153     }
1154 
1155     /* If snapshot_id is specified, it must be equal to name, see
1156        qemu_rbd_snap_list() */
1157     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
1158         error_setg(errp,
1159                    "rbd do not support snapshot id, it should be NULL or "
1160                    "equal to snapshot name");
1161         return -EINVAL;
1162     }
1163 
1164     r = rbd_snap_remove(s->image, snapshot_name);
1165     if (r < 0) {
1166         error_setg_errno(errp, -r, "Failed to remove the snapshot");
1167     }
1168     return r;
1169 }
1170 
1171 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
1172                                   const char *snapshot_name)
1173 {
1174     BDRVRBDState *s = bs->opaque;
1175 
1176     return rbd_snap_rollback(s->image, snapshot_name);
1177 }
1178 
1179 static int qemu_rbd_snap_list(BlockDriverState *bs,
1180                               QEMUSnapshotInfo **psn_tab)
1181 {
1182     BDRVRBDState *s = bs->opaque;
1183     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1184     int i, snap_count;
1185     rbd_snap_info_t *snaps;
1186     int max_snaps = RBD_MAX_SNAPS;
1187 
1188     do {
1189         snaps = g_new(rbd_snap_info_t, max_snaps);
1190         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1191         if (snap_count <= 0) {
1192             g_free(snaps);
1193         }
1194     } while (snap_count == -ERANGE);
1195 
1196     if (snap_count <= 0) {
1197         goto done;
1198     }
1199 
1200     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1201 
1202     for (i = 0; i < snap_count; i++) {
1203         const char *snap_name = snaps[i].name;
1204 
1205         sn_info = sn_tab + i;
1206         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1207         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1208 
1209         sn_info->vm_state_size = snaps[i].size;
1210         sn_info->date_sec = 0;
1211         sn_info->date_nsec = 0;
1212         sn_info->vm_clock_nsec = 0;
1213     }
1214     rbd_snap_list_end(snaps);
1215     g_free(snaps);
1216 
1217  done:
1218     *psn_tab = sn_tab;
1219     return snap_count;
1220 }
1221 
1222 #ifdef LIBRBD_SUPPORTS_DISCARD
1223 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1224                                          int64_t offset,
1225                                          int bytes,
1226                                          BlockCompletionFunc *cb,
1227                                          void *opaque)
1228 {
1229     return rbd_start_aio(bs, offset, NULL, bytes, cb, opaque,
1230                          RBD_AIO_DISCARD);
1231 }
1232 #endif
1233 
1234 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1235 static void coroutine_fn qemu_rbd_co_invalidate_cache(BlockDriverState *bs,
1236                                                       Error **errp)
1237 {
1238     BDRVRBDState *s = bs->opaque;
1239     int r = rbd_invalidate_cache(s->image);
1240     if (r < 0) {
1241         error_setg_errno(errp, -r, "Failed to invalidate the cache");
1242     }
1243 }
1244 #endif
1245 
1246 static QemuOptsList qemu_rbd_create_opts = {
1247     .name = "rbd-create-opts",
1248     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1249     .desc = {
1250         {
1251             .name = BLOCK_OPT_SIZE,
1252             .type = QEMU_OPT_SIZE,
1253             .help = "Virtual disk size"
1254         },
1255         {
1256             .name = BLOCK_OPT_CLUSTER_SIZE,
1257             .type = QEMU_OPT_SIZE,
1258             .help = "RBD object size"
1259         },
1260         {
1261             .name = "password-secret",
1262             .type = QEMU_OPT_STRING,
1263             .help = "ID of secret providing the password",
1264         },
1265         { /* end of list */ }
1266     }
1267 };
1268 
1269 static const char *const qemu_rbd_strong_runtime_opts[] = {
1270     "pool",
1271     "image",
1272     "conf",
1273     "snapshot",
1274     "user",
1275     "server.",
1276     "password-secret",
1277 
1278     NULL
1279 };
1280 
1281 static BlockDriver bdrv_rbd = {
1282     .format_name            = "rbd",
1283     .instance_size          = sizeof(BDRVRBDState),
1284     .bdrv_parse_filename    = qemu_rbd_parse_filename,
1285     .bdrv_refresh_limits    = qemu_rbd_refresh_limits,
1286     .bdrv_file_open         = qemu_rbd_open,
1287     .bdrv_close             = qemu_rbd_close,
1288     .bdrv_reopen_prepare    = qemu_rbd_reopen_prepare,
1289     .bdrv_co_create         = qemu_rbd_co_create,
1290     .bdrv_co_create_opts    = qemu_rbd_co_create_opts,
1291     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1292     .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
1293     .bdrv_get_info          = qemu_rbd_getinfo,
1294     .create_opts            = &qemu_rbd_create_opts,
1295     .bdrv_getlength         = qemu_rbd_getlength,
1296     .bdrv_co_truncate       = qemu_rbd_co_truncate,
1297     .protocol_name          = "rbd",
1298 
1299     .bdrv_aio_preadv        = qemu_rbd_aio_preadv,
1300     .bdrv_aio_pwritev       = qemu_rbd_aio_pwritev,
1301 
1302 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1303     .bdrv_aio_flush         = qemu_rbd_aio_flush,
1304 #else
1305     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1306 #endif
1307 
1308 #ifdef LIBRBD_SUPPORTS_DISCARD
1309     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1310 #endif
1311 
1312     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1313     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1314     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1315     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1316 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1317     .bdrv_co_invalidate_cache = qemu_rbd_co_invalidate_cache,
1318 #endif
1319 
1320     .strong_runtime_opts    = qemu_rbd_strong_runtime_opts,
1321 };
1322 
1323 static void bdrv_rbd_init(void)
1324 {
1325     bdrv_register(&bdrv_rbd);
1326 }
1327 
1328 block_init(bdrv_rbd_init);
1329