xref: /openbmc/qemu/block/rbd.c (revision 9af23989)
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/option.h"
20 #include "block/block_int.h"
21 #include "crypto/secret.h"
22 #include "qemu/cutils.h"
23 #include "qapi/qmp/qstring.h"
24 #include "qapi/qmp/qdict.h"
25 #include "qapi/qmp/qjson.h"
26 #include "qapi/qmp/qlist.h"
27 
28 /*
29  * When specifying the image filename use:
30  *
31  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
32  *
33  * poolname must be the name of an existing rados pool.
34  *
35  * devicename is the name of the rbd image.
36  *
37  * Each option given is used to configure rados, and may be any valid
38  * Ceph option, "id", or "conf".
39  *
40  * The "id" option indicates what user we should authenticate as to
41  * the Ceph cluster.  If it is excluded we will use the Ceph default
42  * (normally 'admin').
43  *
44  * The "conf" option specifies a Ceph configuration file to read.  If
45  * it is not specified, we will read from the default Ceph locations
46  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
47  * file, specify conf=/dev/null.
48  *
49  * Configuration values containing :, @, or = can be escaped with a
50  * leading "\".
51  */
52 
53 /* rbd_aio_discard added in 0.1.2 */
54 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
55 #define LIBRBD_SUPPORTS_DISCARD
56 #else
57 #undef LIBRBD_SUPPORTS_DISCARD
58 #endif
59 
60 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
61 
62 #define RBD_MAX_SNAPS 100
63 
64 /* The LIBRBD_SUPPORTS_IOVEC is defined in librbd.h */
65 #ifdef LIBRBD_SUPPORTS_IOVEC
66 #define LIBRBD_USE_IOVEC 1
67 #else
68 #define LIBRBD_USE_IOVEC 0
69 #endif
70 
71 typedef enum {
72     RBD_AIO_READ,
73     RBD_AIO_WRITE,
74     RBD_AIO_DISCARD,
75     RBD_AIO_FLUSH
76 } RBDAIOCmd;
77 
78 typedef struct RBDAIOCB {
79     BlockAIOCB common;
80     int64_t ret;
81     QEMUIOVector *qiov;
82     char *bounce;
83     RBDAIOCmd cmd;
84     int error;
85     struct BDRVRBDState *s;
86 } RBDAIOCB;
87 
88 typedef struct RADOSCB {
89     RBDAIOCB *acb;
90     struct BDRVRBDState *s;
91     int64_t size;
92     char *buf;
93     int64_t ret;
94 } RADOSCB;
95 
96 typedef struct BDRVRBDState {
97     rados_t cluster;
98     rados_ioctx_t io_ctx;
99     rbd_image_t image;
100     char *image_name;
101     char *snap;
102 } BDRVRBDState;
103 
104 static char *qemu_rbd_next_tok(char *src, char delim, char **p)
105 {
106     char *end;
107 
108     *p = NULL;
109 
110     for (end = src; *end; ++end) {
111         if (*end == delim) {
112             break;
113         }
114         if (*end == '\\' && end[1] != '\0') {
115             end++;
116         }
117     }
118     if (*end == delim) {
119         *p = end + 1;
120         *end = '\0';
121     }
122     return src;
123 }
124 
125 static void qemu_rbd_unescape(char *src)
126 {
127     char *p;
128 
129     for (p = src; *src; ++src, ++p) {
130         if (*src == '\\' && src[1] != '\0') {
131             src++;
132         }
133         *p = *src;
134     }
135     *p = '\0';
136 }
137 
138 static void qemu_rbd_parse_filename(const char *filename, QDict *options,
139                                     Error **errp)
140 {
141     const char *start;
142     char *p, *buf;
143     QList *keypairs = NULL;
144     char *found_str;
145 
146     if (!strstart(filename, "rbd:", &start)) {
147         error_setg(errp, "File name must start with 'rbd:'");
148         return;
149     }
150 
151     buf = g_strdup(start);
152     p = buf;
153 
154     found_str = qemu_rbd_next_tok(p, '/', &p);
155     if (!p) {
156         error_setg(errp, "Pool name is required");
157         goto done;
158     }
159     qemu_rbd_unescape(found_str);
160     qdict_put_str(options, "pool", found_str);
161 
162     if (strchr(p, '@')) {
163         found_str = qemu_rbd_next_tok(p, '@', &p);
164         qemu_rbd_unescape(found_str);
165         qdict_put_str(options, "image", found_str);
166 
167         found_str = qemu_rbd_next_tok(p, ':', &p);
168         qemu_rbd_unescape(found_str);
169         qdict_put_str(options, "snapshot", found_str);
170     } else {
171         found_str = qemu_rbd_next_tok(p, ':', &p);
172         qemu_rbd_unescape(found_str);
173         qdict_put_str(options, "image", found_str);
174     }
175     if (!p) {
176         goto done;
177     }
178 
179     /* The following are essentially all key/value pairs, and we treat
180      * 'id' and 'conf' a bit special.  Key/value pairs may be in any order. */
181     while (p) {
182         char *name, *value;
183         name = qemu_rbd_next_tok(p, '=', &p);
184         if (!p) {
185             error_setg(errp, "conf option %s has no value", name);
186             break;
187         }
188 
189         qemu_rbd_unescape(name);
190 
191         value = qemu_rbd_next_tok(p, ':', &p);
192         qemu_rbd_unescape(value);
193 
194         if (!strcmp(name, "conf")) {
195             qdict_put_str(options, "conf", value);
196         } else if (!strcmp(name, "id")) {
197             qdict_put_str(options, "user", value);
198         } else {
199             /*
200              * We pass these internally to qemu_rbd_set_keypairs(), so
201              * we can get away with the simpler list of [ "key1",
202              * "value1", "key2", "value2" ] rather than a raw dict
203              * { "key1": "value1", "key2": "value2" } where we can't
204              * guarantee order, or even a more correct but complex
205              * [ { "key1": "value1" }, { "key2": "value2" } ]
206              */
207             if (!keypairs) {
208                 keypairs = qlist_new();
209             }
210             qlist_append_str(keypairs, name);
211             qlist_append_str(keypairs, value);
212         }
213     }
214 
215     if (keypairs) {
216         qdict_put(options, "=keyvalue-pairs",
217                   qobject_to_json(QOBJECT(keypairs)));
218     }
219 
220 done:
221     g_free(buf);
222     QDECREF(keypairs);
223     return;
224 }
225 
226 
227 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
228                              Error **errp)
229 {
230     if (secretid == 0) {
231         return 0;
232     }
233 
234     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
235                                                     errp);
236     if (!secret) {
237         return -1;
238     }
239 
240     rados_conf_set(cluster, "key", secret);
241     g_free(secret);
242 
243     return 0;
244 }
245 
246 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs_json,
247                                  Error **errp)
248 {
249     QList *keypairs;
250     QString *name;
251     QString *value;
252     const char *key;
253     size_t remaining;
254     int ret = 0;
255 
256     if (!keypairs_json) {
257         return ret;
258     }
259     keypairs = qobject_to_qlist(qobject_from_json(keypairs_json,
260                                                   &error_abort));
261     remaining = qlist_size(keypairs) / 2;
262     assert(remaining);
263 
264     while (remaining--) {
265         name = qobject_to_qstring(qlist_pop(keypairs));
266         value = qobject_to_qstring(qlist_pop(keypairs));
267         assert(name && value);
268         key = qstring_get_str(name);
269 
270         ret = rados_conf_set(cluster, key, qstring_get_str(value));
271         QDECREF(name);
272         QDECREF(value);
273         if (ret < 0) {
274             error_setg_errno(errp, -ret, "invalid conf option %s", key);
275             ret = -EINVAL;
276             break;
277         }
278     }
279 
280     QDECREF(keypairs);
281     return ret;
282 }
283 
284 static void qemu_rbd_memset(RADOSCB *rcb, int64_t offs)
285 {
286     if (LIBRBD_USE_IOVEC) {
287         RBDAIOCB *acb = rcb->acb;
288         iov_memset(acb->qiov->iov, acb->qiov->niov, offs, 0,
289                    acb->qiov->size - offs);
290     } else {
291         memset(rcb->buf + offs, 0, rcb->size - offs);
292     }
293 }
294 
295 static QemuOptsList runtime_opts = {
296     .name = "rbd",
297     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
298     .desc = {
299         {
300             .name = "pool",
301             .type = QEMU_OPT_STRING,
302             .help = "Rados pool name",
303         },
304         {
305             .name = "image",
306             .type = QEMU_OPT_STRING,
307             .help = "Image name in the pool",
308         },
309         {
310             .name = "conf",
311             .type = QEMU_OPT_STRING,
312             .help = "Rados config file location",
313         },
314         {
315             .name = "snapshot",
316             .type = QEMU_OPT_STRING,
317             .help = "Ceph snapshot name",
318         },
319         {
320             /* maps to 'id' in rados_create() */
321             .name = "user",
322             .type = QEMU_OPT_STRING,
323             .help = "Rados id name",
324         },
325         /*
326          * server.* extracted manually, see qemu_rbd_mon_host()
327          */
328         {
329             .name = "password-secret",
330             .type = QEMU_OPT_STRING,
331             .help = "ID of secret providing the password",
332         },
333 
334         /*
335          * Keys for qemu_rbd_parse_filename(), not in the QAPI schema
336          */
337         {
338             /*
339              * HACK: name starts with '=' so that qemu_opts_parse()
340              * can't set it
341              */
342             .name = "=keyvalue-pairs",
343             .type = QEMU_OPT_STRING,
344             .help = "Legacy rados key/value option parameters",
345         },
346         {
347             .name = "filename",
348             .type = QEMU_OPT_STRING,
349         },
350         { /* end of list */ }
351     },
352 };
353 
354 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
355 {
356     Error *local_err = NULL;
357     int64_t bytes = 0;
358     int64_t objsize;
359     int obj_order = 0;
360     const char *pool, *image_name, *conf, *user, *keypairs;
361     const char *secretid;
362     rados_t cluster;
363     rados_ioctx_t io_ctx;
364     QDict *options = NULL;
365     int ret = 0;
366 
367     secretid = qemu_opt_get(opts, "password-secret");
368 
369     /* Read out options */
370     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
371                      BDRV_SECTOR_SIZE);
372     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
373     if (objsize) {
374         if ((objsize - 1) & objsize) {    /* not a power of 2? */
375             error_setg(errp, "obj size needs to be power of 2");
376             ret = -EINVAL;
377             goto exit;
378         }
379         if (objsize < 4096) {
380             error_setg(errp, "obj size too small");
381             ret = -EINVAL;
382             goto exit;
383         }
384         obj_order = ctz32(objsize);
385     }
386 
387     options = qdict_new();
388     qemu_rbd_parse_filename(filename, options, &local_err);
389     if (local_err) {
390         ret = -EINVAL;
391         error_propagate(errp, local_err);
392         goto exit;
393     }
394 
395     /*
396      * Caution: while qdict_get_try_str() is fine, getting non-string
397      * types would require more care.  When @options come from -blockdev
398      * or blockdev_add, its members are typed according to the QAPI
399      * schema, but when they come from -drive, they're all QString.
400      */
401     pool       = qdict_get_try_str(options, "pool");
402     conf       = qdict_get_try_str(options, "conf");
403     user       = qdict_get_try_str(options, "user");
404     image_name = qdict_get_try_str(options, "image");
405     keypairs   = qdict_get_try_str(options, "=keyvalue-pairs");
406 
407     ret = rados_create(&cluster, user);
408     if (ret < 0) {
409         error_setg_errno(errp, -ret, "error initializing");
410         goto exit;
411     }
412 
413     /* try default location when conf=NULL, but ignore failure */
414     ret = rados_conf_read_file(cluster, conf);
415     if (conf && ret < 0) {
416         error_setg_errno(errp, -ret, "error reading conf file %s", conf);
417         ret = -EIO;
418         goto shutdown;
419     }
420 
421     ret = qemu_rbd_set_keypairs(cluster, keypairs, errp);
422     if (ret < 0) {
423         ret = -EIO;
424         goto shutdown;
425     }
426 
427     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
428         ret = -EIO;
429         goto shutdown;
430     }
431 
432     ret = rados_connect(cluster);
433     if (ret < 0) {
434         error_setg_errno(errp, -ret, "error connecting");
435         goto shutdown;
436     }
437 
438     ret = rados_ioctx_create(cluster, pool, &io_ctx);
439     if (ret < 0) {
440         error_setg_errno(errp, -ret, "error opening pool %s", pool);
441         goto shutdown;
442     }
443 
444     ret = rbd_create(io_ctx, image_name, bytes, &obj_order);
445     if (ret < 0) {
446         error_setg_errno(errp, -ret, "error rbd create");
447     }
448 
449     rados_ioctx_destroy(io_ctx);
450 
451 shutdown:
452     rados_shutdown(cluster);
453 
454 exit:
455     QDECREF(options);
456     return ret;
457 }
458 
459 /*
460  * This aio completion is being called from rbd_finish_bh() and runs in qemu
461  * BH context.
462  */
463 static void qemu_rbd_complete_aio(RADOSCB *rcb)
464 {
465     RBDAIOCB *acb = rcb->acb;
466     int64_t r;
467 
468     r = rcb->ret;
469 
470     if (acb->cmd != RBD_AIO_READ) {
471         if (r < 0) {
472             acb->ret = r;
473             acb->error = 1;
474         } else if (!acb->error) {
475             acb->ret = rcb->size;
476         }
477     } else {
478         if (r < 0) {
479             qemu_rbd_memset(rcb, 0);
480             acb->ret = r;
481             acb->error = 1;
482         } else if (r < rcb->size) {
483             qemu_rbd_memset(rcb, r);
484             if (!acb->error) {
485                 acb->ret = rcb->size;
486             }
487         } else if (!acb->error) {
488             acb->ret = r;
489         }
490     }
491 
492     g_free(rcb);
493 
494     if (!LIBRBD_USE_IOVEC) {
495         if (acb->cmd == RBD_AIO_READ) {
496             qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
497         }
498         qemu_vfree(acb->bounce);
499     }
500 
501     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
502 
503     qemu_aio_unref(acb);
504 }
505 
506 static char *qemu_rbd_mon_host(QDict *options, Error **errp)
507 {
508     const char **vals = g_new(const char *, qdict_size(options) + 1);
509     char keybuf[32];
510     const char *host, *port;
511     char *rados_str;
512     int i;
513 
514     for (i = 0;; i++) {
515         sprintf(keybuf, "server.%d.host", i);
516         host = qdict_get_try_str(options, keybuf);
517         qdict_del(options, keybuf);
518         sprintf(keybuf, "server.%d.port", i);
519         port = qdict_get_try_str(options, keybuf);
520         qdict_del(options, keybuf);
521         if (!host && !port) {
522             break;
523         }
524         if (!host) {
525             error_setg(errp, "Parameter server.%d.host is missing", i);
526             rados_str = NULL;
527             goto out;
528         }
529 
530         if (strchr(host, ':')) {
531             vals[i] = port ? g_strdup_printf("[%s]:%s", host, port)
532                 : g_strdup_printf("[%s]", host);
533         } else {
534             vals[i] = port ? g_strdup_printf("%s:%s", host, port)
535                 : g_strdup(host);
536         }
537     }
538     vals[i] = NULL;
539 
540     rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
541 out:
542     g_strfreev((char **)vals);
543     return rados_str;
544 }
545 
546 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
547                          Error **errp)
548 {
549     BDRVRBDState *s = bs->opaque;
550     const char *pool, *snap, *conf, *user, *image_name, *keypairs;
551     const char *secretid, *filename;
552     QemuOpts *opts;
553     Error *local_err = NULL;
554     char *mon_host = NULL;
555     int r;
556 
557     /* If we are given a filename, parse the filename, with precedence given to
558      * filename encoded options */
559     filename = qdict_get_try_str(options, "filename");
560     if (filename) {
561         warn_report("'filename' option specified. "
562                     "This is an unsupported option, and may be deprecated "
563                     "in the future");
564         qemu_rbd_parse_filename(filename, options, &local_err);
565         if (local_err) {
566             r = -EINVAL;
567             error_propagate(errp, local_err);
568             goto exit;
569         }
570     }
571 
572     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
573     qemu_opts_absorb_qdict(opts, options, &local_err);
574     if (local_err) {
575         error_propagate(errp, local_err);
576         r = -EINVAL;
577         goto failed_opts;
578     }
579 
580     mon_host = qemu_rbd_mon_host(options, &local_err);
581     if (local_err) {
582         error_propagate(errp, local_err);
583         r = -EINVAL;
584         goto failed_opts;
585     }
586 
587     secretid = qemu_opt_get(opts, "password-secret");
588 
589     pool           = qemu_opt_get(opts, "pool");
590     conf           = qemu_opt_get(opts, "conf");
591     snap           = qemu_opt_get(opts, "snapshot");
592     user           = qemu_opt_get(opts, "user");
593     image_name     = qemu_opt_get(opts, "image");
594     keypairs       = qemu_opt_get(opts, "=keyvalue-pairs");
595 
596     if (!pool || !image_name) {
597         error_setg(errp, "Parameters 'pool' and 'image' are required");
598         r = -EINVAL;
599         goto failed_opts;
600     }
601 
602     r = rados_create(&s->cluster, user);
603     if (r < 0) {
604         error_setg_errno(errp, -r, "error initializing");
605         goto failed_opts;
606     }
607 
608     s->snap = g_strdup(snap);
609     s->image_name = g_strdup(image_name);
610 
611     /* try default location when conf=NULL, but ignore failure */
612     r = rados_conf_read_file(s->cluster, conf);
613     if (conf && r < 0) {
614         error_setg_errno(errp, -r, "error reading conf file %s", conf);
615         goto failed_shutdown;
616     }
617 
618     r = qemu_rbd_set_keypairs(s->cluster, keypairs, errp);
619     if (r < 0) {
620         goto failed_shutdown;
621     }
622 
623     if (mon_host) {
624         r = rados_conf_set(s->cluster, "mon_host", mon_host);
625         if (r < 0) {
626             goto failed_shutdown;
627         }
628     }
629 
630     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
631         r = -EIO;
632         goto failed_shutdown;
633     }
634 
635     /*
636      * Fallback to more conservative semantics if setting cache
637      * options fails. Ignore errors from setting rbd_cache because the
638      * only possible error is that the option does not exist, and
639      * librbd defaults to no caching. If write through caching cannot
640      * be set up, fall back to no caching.
641      */
642     if (flags & BDRV_O_NOCACHE) {
643         rados_conf_set(s->cluster, "rbd_cache", "false");
644     } else {
645         rados_conf_set(s->cluster, "rbd_cache", "true");
646     }
647 
648     r = rados_connect(s->cluster);
649     if (r < 0) {
650         error_setg_errno(errp, -r, "error connecting");
651         goto failed_shutdown;
652     }
653 
654     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
655     if (r < 0) {
656         error_setg_errno(errp, -r, "error opening pool %s", pool);
657         goto failed_shutdown;
658     }
659 
660     /* rbd_open is always r/w */
661     r = rbd_open(s->io_ctx, s->image_name, &s->image, s->snap);
662     if (r < 0) {
663         error_setg_errno(errp, -r, "error reading header from %s",
664                          s->image_name);
665         goto failed_open;
666     }
667 
668     /* If we are using an rbd snapshot, we must be r/o, otherwise
669      * leave as-is */
670     if (s->snap != NULL) {
671         if (!bdrv_is_read_only(bs)) {
672             error_report("Opening rbd snapshots without an explicit "
673                          "read-only=on option is deprecated. Future versions "
674                          "will refuse to open the image instead of "
675                          "automatically marking the image read-only.");
676             r = bdrv_set_read_only(bs, true, &local_err);
677             if (r < 0) {
678                 error_propagate(errp, local_err);
679                 goto failed_open;
680             }
681         }
682     }
683 
684     qemu_opts_del(opts);
685     return 0;
686 
687 failed_open:
688     rados_ioctx_destroy(s->io_ctx);
689 failed_shutdown:
690     rados_shutdown(s->cluster);
691     g_free(s->snap);
692     g_free(s->image_name);
693 failed_opts:
694     qemu_opts_del(opts);
695     g_free(mon_host);
696 exit:
697     return r;
698 }
699 
700 
701 /* Since RBD is currently always opened R/W via the API,
702  * we just need to check if we are using a snapshot or not, in
703  * order to determine if we will allow it to be R/W */
704 static int qemu_rbd_reopen_prepare(BDRVReopenState *state,
705                                    BlockReopenQueue *queue, Error **errp)
706 {
707     BDRVRBDState *s = state->bs->opaque;
708     int ret = 0;
709 
710     if (s->snap && state->flags & BDRV_O_RDWR) {
711         error_setg(errp,
712                    "Cannot change node '%s' to r/w when using RBD snapshot",
713                    bdrv_get_device_or_node_name(state->bs));
714         ret = -EINVAL;
715     }
716 
717     return ret;
718 }
719 
720 static void qemu_rbd_close(BlockDriverState *bs)
721 {
722     BDRVRBDState *s = bs->opaque;
723 
724     rbd_close(s->image);
725     rados_ioctx_destroy(s->io_ctx);
726     g_free(s->snap);
727     g_free(s->image_name);
728     rados_shutdown(s->cluster);
729 }
730 
731 static const AIOCBInfo rbd_aiocb_info = {
732     .aiocb_size = sizeof(RBDAIOCB),
733 };
734 
735 static void rbd_finish_bh(void *opaque)
736 {
737     RADOSCB *rcb = opaque;
738     qemu_rbd_complete_aio(rcb);
739 }
740 
741 /*
742  * This is the callback function for rbd_aio_read and _write
743  *
744  * Note: this function is being called from a non qemu thread so
745  * we need to be careful about what we do here. Generally we only
746  * schedule a BH, and do the rest of the io completion handling
747  * from rbd_finish_bh() which runs in a qemu context.
748  */
749 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
750 {
751     RBDAIOCB *acb = rcb->acb;
752 
753     rcb->ret = rbd_aio_get_return_value(c);
754     rbd_aio_release(c);
755 
756     aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
757                             rbd_finish_bh, rcb);
758 }
759 
760 static int rbd_aio_discard_wrapper(rbd_image_t image,
761                                    uint64_t off,
762                                    uint64_t len,
763                                    rbd_completion_t comp)
764 {
765 #ifdef LIBRBD_SUPPORTS_DISCARD
766     return rbd_aio_discard(image, off, len, comp);
767 #else
768     return -ENOTSUP;
769 #endif
770 }
771 
772 static int rbd_aio_flush_wrapper(rbd_image_t image,
773                                  rbd_completion_t comp)
774 {
775 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
776     return rbd_aio_flush(image, comp);
777 #else
778     return -ENOTSUP;
779 #endif
780 }
781 
782 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
783                                  int64_t off,
784                                  QEMUIOVector *qiov,
785                                  int64_t size,
786                                  BlockCompletionFunc *cb,
787                                  void *opaque,
788                                  RBDAIOCmd cmd)
789 {
790     RBDAIOCB *acb;
791     RADOSCB *rcb = NULL;
792     rbd_completion_t c;
793     int r;
794 
795     BDRVRBDState *s = bs->opaque;
796 
797     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
798     acb->cmd = cmd;
799     acb->qiov = qiov;
800     assert(!qiov || qiov->size == size);
801 
802     rcb = g_new(RADOSCB, 1);
803 
804     if (!LIBRBD_USE_IOVEC) {
805         if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
806             acb->bounce = NULL;
807         } else {
808             acb->bounce = qemu_try_blockalign(bs, qiov->size);
809             if (acb->bounce == NULL) {
810                 goto failed;
811             }
812         }
813         if (cmd == RBD_AIO_WRITE) {
814             qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
815         }
816         rcb->buf = acb->bounce;
817     }
818 
819     acb->ret = 0;
820     acb->error = 0;
821     acb->s = s;
822 
823     rcb->acb = acb;
824     rcb->s = acb->s;
825     rcb->size = size;
826     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
827     if (r < 0) {
828         goto failed;
829     }
830 
831     switch (cmd) {
832     case RBD_AIO_WRITE:
833 #ifdef LIBRBD_SUPPORTS_IOVEC
834             r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
835 #else
836             r = rbd_aio_write(s->image, off, size, rcb->buf, c);
837 #endif
838         break;
839     case RBD_AIO_READ:
840 #ifdef LIBRBD_SUPPORTS_IOVEC
841             r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
842 #else
843             r = rbd_aio_read(s->image, off, size, rcb->buf, c);
844 #endif
845         break;
846     case RBD_AIO_DISCARD:
847         r = rbd_aio_discard_wrapper(s->image, off, size, c);
848         break;
849     case RBD_AIO_FLUSH:
850         r = rbd_aio_flush_wrapper(s->image, c);
851         break;
852     default:
853         r = -EINVAL;
854     }
855 
856     if (r < 0) {
857         goto failed_completion;
858     }
859     return &acb->common;
860 
861 failed_completion:
862     rbd_aio_release(c);
863 failed:
864     g_free(rcb);
865     if (!LIBRBD_USE_IOVEC) {
866         qemu_vfree(acb->bounce);
867     }
868 
869     qemu_aio_unref(acb);
870     return NULL;
871 }
872 
873 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
874                                       int64_t sector_num,
875                                       QEMUIOVector *qiov,
876                                       int nb_sectors,
877                                       BlockCompletionFunc *cb,
878                                       void *opaque)
879 {
880     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
881                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
882                          RBD_AIO_READ);
883 }
884 
885 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
886                                        int64_t sector_num,
887                                        QEMUIOVector *qiov,
888                                        int nb_sectors,
889                                        BlockCompletionFunc *cb,
890                                        void *opaque)
891 {
892     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
893                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
894                          RBD_AIO_WRITE);
895 }
896 
897 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
898 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
899                                       BlockCompletionFunc *cb,
900                                       void *opaque)
901 {
902     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
903 }
904 
905 #else
906 
907 static int qemu_rbd_co_flush(BlockDriverState *bs)
908 {
909 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
910     /* rbd_flush added in 0.1.1 */
911     BDRVRBDState *s = bs->opaque;
912     return rbd_flush(s->image);
913 #else
914     return 0;
915 #endif
916 }
917 #endif
918 
919 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
920 {
921     BDRVRBDState *s = bs->opaque;
922     rbd_image_info_t info;
923     int r;
924 
925     r = rbd_stat(s->image, &info, sizeof(info));
926     if (r < 0) {
927         return r;
928     }
929 
930     bdi->cluster_size = info.obj_size;
931     return 0;
932 }
933 
934 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
935 {
936     BDRVRBDState *s = bs->opaque;
937     rbd_image_info_t info;
938     int r;
939 
940     r = rbd_stat(s->image, &info, sizeof(info));
941     if (r < 0) {
942         return r;
943     }
944 
945     return info.size;
946 }
947 
948 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset,
949                              PreallocMode prealloc, Error **errp)
950 {
951     BDRVRBDState *s = bs->opaque;
952     int r;
953 
954     if (prealloc != PREALLOC_MODE_OFF) {
955         error_setg(errp, "Unsupported preallocation mode '%s'",
956                    PreallocMode_str(prealloc));
957         return -ENOTSUP;
958     }
959 
960     r = rbd_resize(s->image, offset);
961     if (r < 0) {
962         error_setg_errno(errp, -r, "Failed to resize file");
963         return r;
964     }
965 
966     return 0;
967 }
968 
969 static int qemu_rbd_snap_create(BlockDriverState *bs,
970                                 QEMUSnapshotInfo *sn_info)
971 {
972     BDRVRBDState *s = bs->opaque;
973     int r;
974 
975     if (sn_info->name[0] == '\0') {
976         return -EINVAL; /* we need a name for rbd snapshots */
977     }
978 
979     /*
980      * rbd snapshots are using the name as the user controlled unique identifier
981      * we can't use the rbd snapid for that purpose, as it can't be set
982      */
983     if (sn_info->id_str[0] != '\0' &&
984         strcmp(sn_info->id_str, sn_info->name) != 0) {
985         return -EINVAL;
986     }
987 
988     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
989         return -ERANGE;
990     }
991 
992     r = rbd_snap_create(s->image, sn_info->name);
993     if (r < 0) {
994         error_report("failed to create snap: %s", strerror(-r));
995         return r;
996     }
997 
998     return 0;
999 }
1000 
1001 static int qemu_rbd_snap_remove(BlockDriverState *bs,
1002                                 const char *snapshot_id,
1003                                 const char *snapshot_name,
1004                                 Error **errp)
1005 {
1006     BDRVRBDState *s = bs->opaque;
1007     int r;
1008 
1009     if (!snapshot_name) {
1010         error_setg(errp, "rbd need a valid snapshot name");
1011         return -EINVAL;
1012     }
1013 
1014     /* If snapshot_id is specified, it must be equal to name, see
1015        qemu_rbd_snap_list() */
1016     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
1017         error_setg(errp,
1018                    "rbd do not support snapshot id, it should be NULL or "
1019                    "equal to snapshot name");
1020         return -EINVAL;
1021     }
1022 
1023     r = rbd_snap_remove(s->image, snapshot_name);
1024     if (r < 0) {
1025         error_setg_errno(errp, -r, "Failed to remove the snapshot");
1026     }
1027     return r;
1028 }
1029 
1030 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
1031                                   const char *snapshot_name)
1032 {
1033     BDRVRBDState *s = bs->opaque;
1034 
1035     return rbd_snap_rollback(s->image, snapshot_name);
1036 }
1037 
1038 static int qemu_rbd_snap_list(BlockDriverState *bs,
1039                               QEMUSnapshotInfo **psn_tab)
1040 {
1041     BDRVRBDState *s = bs->opaque;
1042     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1043     int i, snap_count;
1044     rbd_snap_info_t *snaps;
1045     int max_snaps = RBD_MAX_SNAPS;
1046 
1047     do {
1048         snaps = g_new(rbd_snap_info_t, max_snaps);
1049         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1050         if (snap_count <= 0) {
1051             g_free(snaps);
1052         }
1053     } while (snap_count == -ERANGE);
1054 
1055     if (snap_count <= 0) {
1056         goto done;
1057     }
1058 
1059     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1060 
1061     for (i = 0; i < snap_count; i++) {
1062         const char *snap_name = snaps[i].name;
1063 
1064         sn_info = sn_tab + i;
1065         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1066         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1067 
1068         sn_info->vm_state_size = snaps[i].size;
1069         sn_info->date_sec = 0;
1070         sn_info->date_nsec = 0;
1071         sn_info->vm_clock_nsec = 0;
1072     }
1073     rbd_snap_list_end(snaps);
1074     g_free(snaps);
1075 
1076  done:
1077     *psn_tab = sn_tab;
1078     return snap_count;
1079 }
1080 
1081 #ifdef LIBRBD_SUPPORTS_DISCARD
1082 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1083                                          int64_t offset,
1084                                          int bytes,
1085                                          BlockCompletionFunc *cb,
1086                                          void *opaque)
1087 {
1088     return rbd_start_aio(bs, offset, NULL, bytes, cb, opaque,
1089                          RBD_AIO_DISCARD);
1090 }
1091 #endif
1092 
1093 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1094 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
1095                                       Error **errp)
1096 {
1097     BDRVRBDState *s = bs->opaque;
1098     int r = rbd_invalidate_cache(s->image);
1099     if (r < 0) {
1100         error_setg_errno(errp, -r, "Failed to invalidate the cache");
1101     }
1102 }
1103 #endif
1104 
1105 static QemuOptsList qemu_rbd_create_opts = {
1106     .name = "rbd-create-opts",
1107     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1108     .desc = {
1109         {
1110             .name = BLOCK_OPT_SIZE,
1111             .type = QEMU_OPT_SIZE,
1112             .help = "Virtual disk size"
1113         },
1114         {
1115             .name = BLOCK_OPT_CLUSTER_SIZE,
1116             .type = QEMU_OPT_SIZE,
1117             .help = "RBD object size"
1118         },
1119         {
1120             .name = "password-secret",
1121             .type = QEMU_OPT_STRING,
1122             .help = "ID of secret providing the password",
1123         },
1124         { /* end of list */ }
1125     }
1126 };
1127 
1128 static BlockDriver bdrv_rbd = {
1129     .format_name            = "rbd",
1130     .instance_size          = sizeof(BDRVRBDState),
1131     .bdrv_parse_filename    = qemu_rbd_parse_filename,
1132     .bdrv_file_open         = qemu_rbd_open,
1133     .bdrv_close             = qemu_rbd_close,
1134     .bdrv_reopen_prepare    = qemu_rbd_reopen_prepare,
1135     .bdrv_create            = qemu_rbd_create,
1136     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1137     .bdrv_get_info          = qemu_rbd_getinfo,
1138     .create_opts            = &qemu_rbd_create_opts,
1139     .bdrv_getlength         = qemu_rbd_getlength,
1140     .bdrv_truncate          = qemu_rbd_truncate,
1141     .protocol_name          = "rbd",
1142 
1143     .bdrv_aio_readv         = qemu_rbd_aio_readv,
1144     .bdrv_aio_writev        = qemu_rbd_aio_writev,
1145 
1146 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1147     .bdrv_aio_flush         = qemu_rbd_aio_flush,
1148 #else
1149     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1150 #endif
1151 
1152 #ifdef LIBRBD_SUPPORTS_DISCARD
1153     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1154 #endif
1155 
1156     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1157     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1158     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1159     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1160 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1161     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1162 #endif
1163 };
1164 
1165 static void bdrv_rbd_init(void)
1166 {
1167     bdrv_register(&bdrv_rbd);
1168 }
1169 
1170 block_init(bdrv_rbd_init);
1171