crypto: hmac: move crypto/hmac.h into include/crypto/
[qemu/ar7.git] / block / rbd.c
blob9c3aa638e726d61336a3ca8678b8f59954debb23
1 /*
2 * QEMU Block driver for RADOS (Ceph)
4 * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>,
5 * Josh Durgin <josh.durgin@dreamhost.com>
7 * This work is licensed under the terms of the GNU GPL, version 2. See
8 * the COPYING file in the top-level directory.
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.
14 #include "qemu/osdep.h"
16 #include <rbd/librbd.h>
17 #include "qapi/error.h"
18 #include "qemu/error-report.h"
19 #include "block/block_int.h"
20 #include "crypto/secret.h"
21 #include "qemu/cutils.h"
22 #include "qapi/qmp/qstring.h"
23 #include "qapi/qmp/qjson.h"
26 * When specifying the image filename use:
28 * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
30 * poolname must be the name of an existing rados pool.
32 * devicename is the name of the rbd image.
34 * Each option given is used to configure rados, and may be any valid
35 * Ceph option, "id", or "conf".
37 * The "id" option indicates what user we should authenticate as to
38 * the Ceph cluster. If it is excluded we will use the Ceph default
39 * (normally 'admin').
41 * The "conf" option specifies a Ceph configuration file to read. If
42 * it is not specified, we will read from the default Ceph locations
43 * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration
44 * file, specify conf=/dev/null.
46 * Configuration values containing :, @, or = can be escaped with a
47 * leading "\".
50 /* rbd_aio_discard added in 0.1.2 */
51 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
52 #define LIBRBD_SUPPORTS_DISCARD
53 #else
54 #undef LIBRBD_SUPPORTS_DISCARD
55 #endif
57 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
59 #define RBD_MAX_SNAPS 100
61 /* The LIBRBD_SUPPORTS_IOVEC is defined in librbd.h */
62 #ifdef LIBRBD_SUPPORTS_IOVEC
63 #define LIBRBD_USE_IOVEC 1
64 #else
65 #define LIBRBD_USE_IOVEC 0
66 #endif
68 typedef enum {
69 RBD_AIO_READ,
70 RBD_AIO_WRITE,
71 RBD_AIO_DISCARD,
72 RBD_AIO_FLUSH
73 } RBDAIOCmd;
75 typedef struct RBDAIOCB {
76 BlockAIOCB common;
77 int64_t ret;
78 QEMUIOVector *qiov;
79 char *bounce;
80 RBDAIOCmd cmd;
81 int error;
82 struct BDRVRBDState *s;
83 } RBDAIOCB;
85 typedef struct RADOSCB {
86 RBDAIOCB *acb;
87 struct BDRVRBDState *s;
88 int64_t size;
89 char *buf;
90 int64_t ret;
91 } RADOSCB;
93 typedef struct BDRVRBDState {
94 rados_t cluster;
95 rados_ioctx_t io_ctx;
96 rbd_image_t image;
97 char *image_name;
98 char *snap;
99 } BDRVRBDState;
101 static char *qemu_rbd_next_tok(char *src, char delim, char **p)
103 char *end;
105 *p = NULL;
107 for (end = src; *end; ++end) {
108 if (*end == delim) {
109 break;
111 if (*end == '\\' && end[1] != '\0') {
112 end++;
115 if (*end == delim) {
116 *p = end + 1;
117 *end = '\0';
119 return src;
122 static void qemu_rbd_unescape(char *src)
124 char *p;
126 for (p = src; *src; ++src, ++p) {
127 if (*src == '\\' && src[1] != '\0') {
128 src++;
130 *p = *src;
132 *p = '\0';
135 static void qemu_rbd_parse_filename(const char *filename, QDict *options,
136 Error **errp)
138 const char *start;
139 char *p, *buf;
140 QList *keypairs = NULL;
141 char *found_str;
143 if (!strstart(filename, "rbd:", &start)) {
144 error_setg(errp, "File name must start with 'rbd:'");
145 return;
148 buf = g_strdup(start);
149 p = buf;
151 found_str = qemu_rbd_next_tok(p, '/', &p);
152 if (!p) {
153 error_setg(errp, "Pool name is required");
154 goto done;
156 qemu_rbd_unescape(found_str);
157 qdict_put_str(options, "pool", found_str);
159 if (strchr(p, '@')) {
160 found_str = qemu_rbd_next_tok(p, '@', &p);
161 qemu_rbd_unescape(found_str);
162 qdict_put_str(options, "image", found_str);
164 found_str = qemu_rbd_next_tok(p, ':', &p);
165 qemu_rbd_unescape(found_str);
166 qdict_put_str(options, "snapshot", found_str);
167 } else {
168 found_str = qemu_rbd_next_tok(p, ':', &p);
169 qemu_rbd_unescape(found_str);
170 qdict_put_str(options, "image", found_str);
172 if (!p) {
173 goto done;
176 /* The following are essentially all key/value pairs, and we treat
177 * 'id' and 'conf' a bit special. Key/value pairs may be in any order. */
178 while (p) {
179 char *name, *value;
180 name = qemu_rbd_next_tok(p, '=', &p);
181 if (!p) {
182 error_setg(errp, "conf option %s has no value", name);
183 break;
186 qemu_rbd_unescape(name);
188 value = qemu_rbd_next_tok(p, ':', &p);
189 qemu_rbd_unescape(value);
191 if (!strcmp(name, "conf")) {
192 qdict_put_str(options, "conf", value);
193 } else if (!strcmp(name, "id")) {
194 qdict_put_str(options, "user", value);
195 } else {
197 * We pass these internally to qemu_rbd_set_keypairs(), so
198 * we can get away with the simpler list of [ "key1",
199 * "value1", "key2", "value2" ] rather than a raw dict
200 * { "key1": "value1", "key2": "value2" } where we can't
201 * guarantee order, or even a more correct but complex
202 * [ { "key1": "value1" }, { "key2": "value2" } ]
204 if (!keypairs) {
205 keypairs = qlist_new();
207 qlist_append_str(keypairs, name);
208 qlist_append_str(keypairs, value);
212 if (keypairs) {
213 qdict_put(options, "=keyvalue-pairs",
214 qobject_to_json(QOBJECT(keypairs)));
217 done:
218 g_free(buf);
219 QDECREF(keypairs);
220 return;
224 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
225 Error **errp)
227 if (secretid == 0) {
228 return 0;
231 gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
232 errp);
233 if (!secret) {
234 return -1;
237 rados_conf_set(cluster, "key", secret);
238 g_free(secret);
240 return 0;
243 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs_json,
244 Error **errp)
246 QList *keypairs;
247 QString *name;
248 QString *value;
249 const char *key;
250 size_t remaining;
251 int ret = 0;
253 if (!keypairs_json) {
254 return ret;
256 keypairs = qobject_to_qlist(qobject_from_json(keypairs_json,
257 &error_abort));
258 remaining = qlist_size(keypairs) / 2;
259 assert(remaining);
261 while (remaining--) {
262 name = qobject_to_qstring(qlist_pop(keypairs));
263 value = qobject_to_qstring(qlist_pop(keypairs));
264 assert(name && value);
265 key = qstring_get_str(name);
267 ret = rados_conf_set(cluster, key, qstring_get_str(value));
268 QDECREF(name);
269 QDECREF(value);
270 if (ret < 0) {
271 error_setg_errno(errp, -ret, "invalid conf option %s", key);
272 ret = -EINVAL;
273 break;
277 QDECREF(keypairs);
278 return ret;
281 static void qemu_rbd_memset(RADOSCB *rcb, int64_t offs)
283 if (LIBRBD_USE_IOVEC) {
284 RBDAIOCB *acb = rcb->acb;
285 iov_memset(acb->qiov->iov, acb->qiov->niov, offs, 0,
286 acb->qiov->size - offs);
287 } else {
288 memset(rcb->buf + offs, 0, rcb->size - offs);
292 static QemuOptsList runtime_opts = {
293 .name = "rbd",
294 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
295 .desc = {
297 .name = "pool",
298 .type = QEMU_OPT_STRING,
299 .help = "Rados pool name",
302 .name = "image",
303 .type = QEMU_OPT_STRING,
304 .help = "Image name in the pool",
307 .name = "conf",
308 .type = QEMU_OPT_STRING,
309 .help = "Rados config file location",
312 .name = "snapshot",
313 .type = QEMU_OPT_STRING,
314 .help = "Ceph snapshot name",
317 /* maps to 'id' in rados_create() */
318 .name = "user",
319 .type = QEMU_OPT_STRING,
320 .help = "Rados id name",
323 * server.* extracted manually, see qemu_rbd_mon_host()
326 .name = "password-secret",
327 .type = QEMU_OPT_STRING,
328 .help = "ID of secret providing the password",
332 * Keys for qemu_rbd_parse_filename(), not in the QAPI schema
336 * HACK: name starts with '=' so that qemu_opts_parse()
337 * can't set it
339 .name = "=keyvalue-pairs",
340 .type = QEMU_OPT_STRING,
341 .help = "Legacy rados key/value option parameters",
344 .name = "filename",
345 .type = QEMU_OPT_STRING,
347 { /* end of list */ }
351 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
353 Error *local_err = NULL;
354 int64_t bytes = 0;
355 int64_t objsize;
356 int obj_order = 0;
357 const char *pool, *image_name, *conf, *user, *keypairs;
358 const char *secretid;
359 rados_t cluster;
360 rados_ioctx_t io_ctx;
361 QDict *options = NULL;
362 int ret = 0;
364 secretid = qemu_opt_get(opts, "password-secret");
366 /* Read out options */
367 bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
368 BDRV_SECTOR_SIZE);
369 objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
370 if (objsize) {
371 if ((objsize - 1) & objsize) { /* not a power of 2? */
372 error_setg(errp, "obj size needs to be power of 2");
373 ret = -EINVAL;
374 goto exit;
376 if (objsize < 4096) {
377 error_setg(errp, "obj size too small");
378 ret = -EINVAL;
379 goto exit;
381 obj_order = ctz32(objsize);
384 options = qdict_new();
385 qemu_rbd_parse_filename(filename, options, &local_err);
386 if (local_err) {
387 ret = -EINVAL;
388 error_propagate(errp, local_err);
389 goto exit;
393 * Caution: while qdict_get_try_str() is fine, getting non-string
394 * types would require more care. When @options come from -blockdev
395 * or blockdev_add, its members are typed according to the QAPI
396 * schema, but when they come from -drive, they're all QString.
398 pool = qdict_get_try_str(options, "pool");
399 conf = qdict_get_try_str(options, "conf");
400 user = qdict_get_try_str(options, "user");
401 image_name = qdict_get_try_str(options, "image");
402 keypairs = qdict_get_try_str(options, "=keyvalue-pairs");
404 ret = rados_create(&cluster, user);
405 if (ret < 0) {
406 error_setg_errno(errp, -ret, "error initializing");
407 goto exit;
410 /* try default location when conf=NULL, but ignore failure */
411 ret = rados_conf_read_file(cluster, conf);
412 if (conf && ret < 0) {
413 error_setg_errno(errp, -ret, "error reading conf file %s", conf);
414 ret = -EIO;
415 goto shutdown;
418 ret = qemu_rbd_set_keypairs(cluster, keypairs, errp);
419 if (ret < 0) {
420 ret = -EIO;
421 goto shutdown;
424 if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
425 ret = -EIO;
426 goto shutdown;
429 ret = rados_connect(cluster);
430 if (ret < 0) {
431 error_setg_errno(errp, -ret, "error connecting");
432 goto shutdown;
435 ret = rados_ioctx_create(cluster, pool, &io_ctx);
436 if (ret < 0) {
437 error_setg_errno(errp, -ret, "error opening pool %s", pool);
438 goto shutdown;
441 ret = rbd_create(io_ctx, image_name, bytes, &obj_order);
442 if (ret < 0) {
443 error_setg_errno(errp, -ret, "error rbd create");
446 rados_ioctx_destroy(io_ctx);
448 shutdown:
449 rados_shutdown(cluster);
451 exit:
452 QDECREF(options);
453 return ret;
457 * This aio completion is being called from rbd_finish_bh() and runs in qemu
458 * BH context.
460 static void qemu_rbd_complete_aio(RADOSCB *rcb)
462 RBDAIOCB *acb = rcb->acb;
463 int64_t r;
465 r = rcb->ret;
467 if (acb->cmd != RBD_AIO_READ) {
468 if (r < 0) {
469 acb->ret = r;
470 acb->error = 1;
471 } else if (!acb->error) {
472 acb->ret = rcb->size;
474 } else {
475 if (r < 0) {
476 qemu_rbd_memset(rcb, 0);
477 acb->ret = r;
478 acb->error = 1;
479 } else if (r < rcb->size) {
480 qemu_rbd_memset(rcb, r);
481 if (!acb->error) {
482 acb->ret = rcb->size;
484 } else if (!acb->error) {
485 acb->ret = r;
489 g_free(rcb);
491 if (!LIBRBD_USE_IOVEC) {
492 if (acb->cmd == RBD_AIO_READ) {
493 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
495 qemu_vfree(acb->bounce);
498 acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
500 qemu_aio_unref(acb);
503 static char *qemu_rbd_mon_host(QDict *options, Error **errp)
505 const char **vals = g_new(const char *, qdict_size(options) + 1);
506 char keybuf[32];
507 const char *host, *port;
508 char *rados_str;
509 int i;
511 for (i = 0;; i++) {
512 sprintf(keybuf, "server.%d.host", i);
513 host = qdict_get_try_str(options, keybuf);
514 qdict_del(options, keybuf);
515 sprintf(keybuf, "server.%d.port", i);
516 port = qdict_get_try_str(options, keybuf);
517 qdict_del(options, keybuf);
518 if (!host && !port) {
519 break;
521 if (!host) {
522 error_setg(errp, "Parameter server.%d.host is missing", i);
523 rados_str = NULL;
524 goto out;
527 if (strchr(host, ':')) {
528 vals[i] = port ? g_strdup_printf("[%s]:%s", host, port)
529 : g_strdup_printf("[%s]", host);
530 } else {
531 vals[i] = port ? g_strdup_printf("%s:%s", host, port)
532 : g_strdup(host);
535 vals[i] = NULL;
537 rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
538 out:
539 g_strfreev((char **)vals);
540 return rados_str;
543 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
544 Error **errp)
546 BDRVRBDState *s = bs->opaque;
547 const char *pool, *snap, *conf, *user, *image_name, *keypairs;
548 const char *secretid, *filename;
549 QemuOpts *opts;
550 Error *local_err = NULL;
551 char *mon_host = NULL;
552 int r;
554 /* If we are given a filename, parse the filename, with precedence given to
555 * filename encoded options */
556 filename = qdict_get_try_str(options, "filename");
557 if (filename) {
558 warn_report("'filename' option specified. "
559 "This is an unsupported option, and may be deprecated "
560 "in the future");
561 qemu_rbd_parse_filename(filename, options, &local_err);
562 if (local_err) {
563 r = -EINVAL;
564 error_propagate(errp, local_err);
565 goto exit;
569 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
570 qemu_opts_absorb_qdict(opts, options, &local_err);
571 if (local_err) {
572 error_propagate(errp, local_err);
573 r = -EINVAL;
574 goto failed_opts;
577 mon_host = qemu_rbd_mon_host(options, &local_err);
578 if (local_err) {
579 error_propagate(errp, local_err);
580 r = -EINVAL;
581 goto failed_opts;
584 secretid = qemu_opt_get(opts, "password-secret");
586 pool = qemu_opt_get(opts, "pool");
587 conf = qemu_opt_get(opts, "conf");
588 snap = qemu_opt_get(opts, "snapshot");
589 user = qemu_opt_get(opts, "user");
590 image_name = qemu_opt_get(opts, "image");
591 keypairs = qemu_opt_get(opts, "=keyvalue-pairs");
593 if (!pool || !image_name) {
594 error_setg(errp, "Parameters 'pool' and 'image' are required");
595 r = -EINVAL;
596 goto failed_opts;
599 r = rados_create(&s->cluster, user);
600 if (r < 0) {
601 error_setg_errno(errp, -r, "error initializing");
602 goto failed_opts;
605 s->snap = g_strdup(snap);
606 s->image_name = g_strdup(image_name);
608 /* try default location when conf=NULL, but ignore failure */
609 r = rados_conf_read_file(s->cluster, conf);
610 if (conf && r < 0) {
611 error_setg_errno(errp, -r, "error reading conf file %s", conf);
612 goto failed_shutdown;
615 r = qemu_rbd_set_keypairs(s->cluster, keypairs, errp);
616 if (r < 0) {
617 goto failed_shutdown;
620 if (mon_host) {
621 r = rados_conf_set(s->cluster, "mon_host", mon_host);
622 if (r < 0) {
623 goto failed_shutdown;
627 if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
628 r = -EIO;
629 goto failed_shutdown;
633 * Fallback to more conservative semantics if setting cache
634 * options fails. Ignore errors from setting rbd_cache because the
635 * only possible error is that the option does not exist, and
636 * librbd defaults to no caching. If write through caching cannot
637 * be set up, fall back to no caching.
639 if (flags & BDRV_O_NOCACHE) {
640 rados_conf_set(s->cluster, "rbd_cache", "false");
641 } else {
642 rados_conf_set(s->cluster, "rbd_cache", "true");
645 r = rados_connect(s->cluster);
646 if (r < 0) {
647 error_setg_errno(errp, -r, "error connecting");
648 goto failed_shutdown;
651 r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
652 if (r < 0) {
653 error_setg_errno(errp, -r, "error opening pool %s", pool);
654 goto failed_shutdown;
657 /* rbd_open is always r/w */
658 r = rbd_open(s->io_ctx, s->image_name, &s->image, s->snap);
659 if (r < 0) {
660 error_setg_errno(errp, -r, "error reading header from %s",
661 s->image_name);
662 goto failed_open;
665 /* If we are using an rbd snapshot, we must be r/o, otherwise
666 * leave as-is */
667 if (s->snap != NULL) {
668 r = bdrv_set_read_only(bs, true, &local_err);
669 if (r < 0) {
670 error_propagate(errp, local_err);
671 goto failed_open;
675 qemu_opts_del(opts);
676 return 0;
678 failed_open:
679 rados_ioctx_destroy(s->io_ctx);
680 failed_shutdown:
681 rados_shutdown(s->cluster);
682 g_free(s->snap);
683 g_free(s->image_name);
684 failed_opts:
685 qemu_opts_del(opts);
686 g_free(mon_host);
687 exit:
688 return r;
692 /* Since RBD is currently always opened R/W via the API,
693 * we just need to check if we are using a snapshot or not, in
694 * order to determine if we will allow it to be R/W */
695 static int qemu_rbd_reopen_prepare(BDRVReopenState *state,
696 BlockReopenQueue *queue, Error **errp)
698 BDRVRBDState *s = state->bs->opaque;
699 int ret = 0;
701 if (s->snap && state->flags & BDRV_O_RDWR) {
702 error_setg(errp,
703 "Cannot change node '%s' to r/w when using RBD snapshot",
704 bdrv_get_device_or_node_name(state->bs));
705 ret = -EINVAL;
708 return ret;
711 static void qemu_rbd_close(BlockDriverState *bs)
713 BDRVRBDState *s = bs->opaque;
715 rbd_close(s->image);
716 rados_ioctx_destroy(s->io_ctx);
717 g_free(s->snap);
718 g_free(s->image_name);
719 rados_shutdown(s->cluster);
722 static const AIOCBInfo rbd_aiocb_info = {
723 .aiocb_size = sizeof(RBDAIOCB),
726 static void rbd_finish_bh(void *opaque)
728 RADOSCB *rcb = opaque;
729 qemu_rbd_complete_aio(rcb);
733 * This is the callback function for rbd_aio_read and _write
735 * Note: this function is being called from a non qemu thread so
736 * we need to be careful about what we do here. Generally we only
737 * schedule a BH, and do the rest of the io completion handling
738 * from rbd_finish_bh() which runs in a qemu context.
740 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
742 RBDAIOCB *acb = rcb->acb;
744 rcb->ret = rbd_aio_get_return_value(c);
745 rbd_aio_release(c);
747 aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
748 rbd_finish_bh, rcb);
751 static int rbd_aio_discard_wrapper(rbd_image_t image,
752 uint64_t off,
753 uint64_t len,
754 rbd_completion_t comp)
756 #ifdef LIBRBD_SUPPORTS_DISCARD
757 return rbd_aio_discard(image, off, len, comp);
758 #else
759 return -ENOTSUP;
760 #endif
763 static int rbd_aio_flush_wrapper(rbd_image_t image,
764 rbd_completion_t comp)
766 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
767 return rbd_aio_flush(image, comp);
768 #else
769 return -ENOTSUP;
770 #endif
773 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
774 int64_t off,
775 QEMUIOVector *qiov,
776 int64_t size,
777 BlockCompletionFunc *cb,
778 void *opaque,
779 RBDAIOCmd cmd)
781 RBDAIOCB *acb;
782 RADOSCB *rcb = NULL;
783 rbd_completion_t c;
784 int r;
786 BDRVRBDState *s = bs->opaque;
788 acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
789 acb->cmd = cmd;
790 acb->qiov = qiov;
791 assert(!qiov || qiov->size == size);
793 rcb = g_new(RADOSCB, 1);
795 if (!LIBRBD_USE_IOVEC) {
796 if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
797 acb->bounce = NULL;
798 } else {
799 acb->bounce = qemu_try_blockalign(bs, qiov->size);
800 if (acb->bounce == NULL) {
801 goto failed;
804 if (cmd == RBD_AIO_WRITE) {
805 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
807 rcb->buf = acb->bounce;
810 acb->ret = 0;
811 acb->error = 0;
812 acb->s = s;
814 rcb->acb = acb;
815 rcb->s = acb->s;
816 rcb->size = size;
817 r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
818 if (r < 0) {
819 goto failed;
822 switch (cmd) {
823 case RBD_AIO_WRITE:
824 #ifdef LIBRBD_SUPPORTS_IOVEC
825 r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
826 #else
827 r = rbd_aio_write(s->image, off, size, rcb->buf, c);
828 #endif
829 break;
830 case RBD_AIO_READ:
831 #ifdef LIBRBD_SUPPORTS_IOVEC
832 r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
833 #else
834 r = rbd_aio_read(s->image, off, size, rcb->buf, c);
835 #endif
836 break;
837 case RBD_AIO_DISCARD:
838 r = rbd_aio_discard_wrapper(s->image, off, size, c);
839 break;
840 case RBD_AIO_FLUSH:
841 r = rbd_aio_flush_wrapper(s->image, c);
842 break;
843 default:
844 r = -EINVAL;
847 if (r < 0) {
848 goto failed_completion;
850 return &acb->common;
852 failed_completion:
853 rbd_aio_release(c);
854 failed:
855 g_free(rcb);
856 if (!LIBRBD_USE_IOVEC) {
857 qemu_vfree(acb->bounce);
860 qemu_aio_unref(acb);
861 return NULL;
864 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
865 int64_t sector_num,
866 QEMUIOVector *qiov,
867 int nb_sectors,
868 BlockCompletionFunc *cb,
869 void *opaque)
871 return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
872 (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
873 RBD_AIO_READ);
876 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
877 int64_t sector_num,
878 QEMUIOVector *qiov,
879 int nb_sectors,
880 BlockCompletionFunc *cb,
881 void *opaque)
883 return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
884 (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
885 RBD_AIO_WRITE);
888 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
889 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
890 BlockCompletionFunc *cb,
891 void *opaque)
893 return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
896 #else
898 static int qemu_rbd_co_flush(BlockDriverState *bs)
900 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
901 /* rbd_flush added in 0.1.1 */
902 BDRVRBDState *s = bs->opaque;
903 return rbd_flush(s->image);
904 #else
905 return 0;
906 #endif
908 #endif
910 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
912 BDRVRBDState *s = bs->opaque;
913 rbd_image_info_t info;
914 int r;
916 r = rbd_stat(s->image, &info, sizeof(info));
917 if (r < 0) {
918 return r;
921 bdi->cluster_size = info.obj_size;
922 return 0;
925 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
927 BDRVRBDState *s = bs->opaque;
928 rbd_image_info_t info;
929 int r;
931 r = rbd_stat(s->image, &info, sizeof(info));
932 if (r < 0) {
933 return r;
936 return info.size;
939 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset,
940 PreallocMode prealloc, Error **errp)
942 BDRVRBDState *s = bs->opaque;
943 int r;
945 if (prealloc != PREALLOC_MODE_OFF) {
946 error_setg(errp, "Unsupported preallocation mode '%s'",
947 PreallocMode_lookup[prealloc]);
948 return -ENOTSUP;
951 r = rbd_resize(s->image, offset);
952 if (r < 0) {
953 error_setg_errno(errp, -r, "Failed to resize file");
954 return r;
957 return 0;
960 static int qemu_rbd_snap_create(BlockDriverState *bs,
961 QEMUSnapshotInfo *sn_info)
963 BDRVRBDState *s = bs->opaque;
964 int r;
966 if (sn_info->name[0] == '\0') {
967 return -EINVAL; /* we need a name for rbd snapshots */
971 * rbd snapshots are using the name as the user controlled unique identifier
972 * we can't use the rbd snapid for that purpose, as it can't be set
974 if (sn_info->id_str[0] != '\0' &&
975 strcmp(sn_info->id_str, sn_info->name) != 0) {
976 return -EINVAL;
979 if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
980 return -ERANGE;
983 r = rbd_snap_create(s->image, sn_info->name);
984 if (r < 0) {
985 error_report("failed to create snap: %s", strerror(-r));
986 return r;
989 return 0;
992 static int qemu_rbd_snap_remove(BlockDriverState *bs,
993 const char *snapshot_id,
994 const char *snapshot_name,
995 Error **errp)
997 BDRVRBDState *s = bs->opaque;
998 int r;
1000 if (!snapshot_name) {
1001 error_setg(errp, "rbd need a valid snapshot name");
1002 return -EINVAL;
1005 /* If snapshot_id is specified, it must be equal to name, see
1006 qemu_rbd_snap_list() */
1007 if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
1008 error_setg(errp,
1009 "rbd do not support snapshot id, it should be NULL or "
1010 "equal to snapshot name");
1011 return -EINVAL;
1014 r = rbd_snap_remove(s->image, snapshot_name);
1015 if (r < 0) {
1016 error_setg_errno(errp, -r, "Failed to remove the snapshot");
1018 return r;
1021 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
1022 const char *snapshot_name)
1024 BDRVRBDState *s = bs->opaque;
1026 return rbd_snap_rollback(s->image, snapshot_name);
1029 static int qemu_rbd_snap_list(BlockDriverState *bs,
1030 QEMUSnapshotInfo **psn_tab)
1032 BDRVRBDState *s = bs->opaque;
1033 QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1034 int i, snap_count;
1035 rbd_snap_info_t *snaps;
1036 int max_snaps = RBD_MAX_SNAPS;
1038 do {
1039 snaps = g_new(rbd_snap_info_t, max_snaps);
1040 snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1041 if (snap_count <= 0) {
1042 g_free(snaps);
1044 } while (snap_count == -ERANGE);
1046 if (snap_count <= 0) {
1047 goto done;
1050 sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1052 for (i = 0; i < snap_count; i++) {
1053 const char *snap_name = snaps[i].name;
1055 sn_info = sn_tab + i;
1056 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1057 pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1059 sn_info->vm_state_size = snaps[i].size;
1060 sn_info->date_sec = 0;
1061 sn_info->date_nsec = 0;
1062 sn_info->vm_clock_nsec = 0;
1064 rbd_snap_list_end(snaps);
1065 g_free(snaps);
1067 done:
1068 *psn_tab = sn_tab;
1069 return snap_count;
1072 #ifdef LIBRBD_SUPPORTS_DISCARD
1073 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1074 int64_t offset,
1075 int bytes,
1076 BlockCompletionFunc *cb,
1077 void *opaque)
1079 return rbd_start_aio(bs, offset, NULL, bytes, cb, opaque,
1080 RBD_AIO_DISCARD);
1082 #endif
1084 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1085 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
1086 Error **errp)
1088 BDRVRBDState *s = bs->opaque;
1089 int r = rbd_invalidate_cache(s->image);
1090 if (r < 0) {
1091 error_setg_errno(errp, -r, "Failed to invalidate the cache");
1094 #endif
1096 static QemuOptsList qemu_rbd_create_opts = {
1097 .name = "rbd-create-opts",
1098 .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1099 .desc = {
1101 .name = BLOCK_OPT_SIZE,
1102 .type = QEMU_OPT_SIZE,
1103 .help = "Virtual disk size"
1106 .name = BLOCK_OPT_CLUSTER_SIZE,
1107 .type = QEMU_OPT_SIZE,
1108 .help = "RBD object size"
1111 .name = "password-secret",
1112 .type = QEMU_OPT_STRING,
1113 .help = "ID of secret providing the password",
1115 { /* end of list */ }
1119 static BlockDriver bdrv_rbd = {
1120 .format_name = "rbd",
1121 .instance_size = sizeof(BDRVRBDState),
1122 .bdrv_parse_filename = qemu_rbd_parse_filename,
1123 .bdrv_file_open = qemu_rbd_open,
1124 .bdrv_close = qemu_rbd_close,
1125 .bdrv_reopen_prepare = qemu_rbd_reopen_prepare,
1126 .bdrv_create = qemu_rbd_create,
1127 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1128 .bdrv_get_info = qemu_rbd_getinfo,
1129 .create_opts = &qemu_rbd_create_opts,
1130 .bdrv_getlength = qemu_rbd_getlength,
1131 .bdrv_truncate = qemu_rbd_truncate,
1132 .protocol_name = "rbd",
1134 .bdrv_aio_readv = qemu_rbd_aio_readv,
1135 .bdrv_aio_writev = qemu_rbd_aio_writev,
1137 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1138 .bdrv_aio_flush = qemu_rbd_aio_flush,
1139 #else
1140 .bdrv_co_flush_to_disk = qemu_rbd_co_flush,
1141 #endif
1143 #ifdef LIBRBD_SUPPORTS_DISCARD
1144 .bdrv_aio_pdiscard = qemu_rbd_aio_pdiscard,
1145 #endif
1147 .bdrv_snapshot_create = qemu_rbd_snap_create,
1148 .bdrv_snapshot_delete = qemu_rbd_snap_remove,
1149 .bdrv_snapshot_list = qemu_rbd_snap_list,
1150 .bdrv_snapshot_goto = qemu_rbd_snap_rollback,
1151 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1152 .bdrv_invalidate_cache = qemu_rbd_invalidate_cache,
1153 #endif
1156 static void bdrv_rbd_init(void)
1158 bdrv_register(&bdrv_rbd);
1161 block_init(bdrv_rbd_init);