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.
16
#include "qemu-common.h"
17
#include "qemu/error-report.h"
18
#include "block/block_int.h"
20
#include <rbd/librbd.h>
23
* When specifying the image filename use:
25
* rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
27
* poolname must be the name of an existing rados pool.
29
* devicename is the name of the rbd image.
31
* Each option given is used to configure rados, and may be any valid
32
* Ceph option, "id", or "conf".
34
* The "id" option indicates what user we should authenticate as to
35
* the Ceph cluster. If it is excluded we will use the Ceph default
38
* The "conf" option specifies a Ceph configuration file to read. If
39
* it is not specified, we will read from the default Ceph locations
40
* (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration
41
* file, specify conf=/dev/null.
43
* Configuration values containing :, @, or = can be escaped with a
47
/* rbd_aio_discard added in 0.1.2 */
48
#if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
49
#define LIBRBD_SUPPORTS_DISCARD
51
#undef LIBRBD_SUPPORTS_DISCARD
54
#define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
56
#define RBD_MAX_CONF_NAME_SIZE 128
57
#define RBD_MAX_CONF_VAL_SIZE 512
58
#define RBD_MAX_CONF_SIZE 1024
59
#define RBD_MAX_POOL_NAME_SIZE 128
60
#define RBD_MAX_SNAP_NAME_SIZE 128
61
#define RBD_MAX_SNAPS 100
70
typedef struct RBDAIOCB {
71
BlockDriverAIOCB common;
79
struct BDRVRBDState *s;
84
typedef struct RADOSCB {
87
struct BDRVRBDState *s;
95
#define RBD_FD_WRITE 1
97
typedef struct BDRVRBDState {
100
rados_ioctx_t io_ctx;
102
char name[RBD_MAX_IMAGE_NAME_SIZE];
105
int event_reader_pos;
109
static void rbd_aio_bh_cb(void *opaque);
111
static int qemu_rbd_next_tok(char *dst, int dst_len,
112
char *src, char delim,
122
for (end = src; *end; ++end) {
126
if (*end == '\\' && end[1] != '\0') {
137
error_report("%s too long", name);
140
error_report("%s too short", name);
144
pstrcpy(dst, dst_len, src);
149
static void qemu_rbd_unescape(char *src)
153
for (p = src; *src; ++src, ++p) {
154
if (*src == '\\' && src[1] != '\0') {
162
static int qemu_rbd_parsename(const char *filename,
163
char *pool, int pool_len,
164
char *snap, int snap_len,
165
char *name, int name_len,
166
char *conf, int conf_len)
172
if (!strstart(filename, "rbd:", &start)) {
176
buf = g_strdup(start);
181
ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
186
qemu_rbd_unescape(pool);
188
if (strchr(p, '@')) {
189
ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
193
ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
194
qemu_rbd_unescape(snap);
196
ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
198
qemu_rbd_unescape(name);
203
ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
210
static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
212
const char *p = conf;
216
const char *end = strchr(p, ':');
224
if (strncmp(p, "id=", 3) == 0) {
226
strncpy(clientname, p + 3, len);
227
clientname[len] = '\0';
238
static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
241
char name[RBD_MAX_CONF_NAME_SIZE];
242
char value[RBD_MAX_CONF_VAL_SIZE];
245
buf = g_strdup(conf);
249
ret = qemu_rbd_next_tok(name, sizeof(name), p,
250
'=', "conf option name", &p);
254
qemu_rbd_unescape(name);
257
error_report("conf option %s has no value", name);
262
ret = qemu_rbd_next_tok(value, sizeof(value), p,
263
':', "conf option value", &p);
267
qemu_rbd_unescape(value);
269
if (strcmp(name, "conf") == 0) {
270
ret = rados_conf_read_file(cluster, value);
272
error_report("error reading conf file %s", value);
275
} else if (strcmp(name, "id") == 0) {
276
/* ignore, this is parsed by qemu_rbd_parse_clientname() */
278
ret = rados_conf_set(cluster, name, value);
280
error_report("invalid conf option %s", name);
291
static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
296
char pool[RBD_MAX_POOL_NAME_SIZE];
297
char name[RBD_MAX_IMAGE_NAME_SIZE];
298
char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
299
char conf[RBD_MAX_CONF_SIZE];
300
char clientname_buf[RBD_MAX_CONF_SIZE];
303
rados_ioctx_t io_ctx;
306
if (qemu_rbd_parsename(filename, pool, sizeof(pool),
307
snap_buf, sizeof(snap_buf),
309
conf, sizeof(conf)) < 0) {
313
/* Read out options */
314
while (options && options->name) {
315
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
316
bytes = options->value.n;
317
} else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
318
if (options->value.n) {
319
objsize = options->value.n;
320
if ((objsize - 1) & objsize) { /* not a power of 2? */
321
error_report("obj size needs to be power of 2");
324
if (objsize < 4096) {
325
error_report("obj size too small");
328
obj_order = ffs(objsize) - 1;
334
clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
335
if (rados_create(&cluster, clientname) < 0) {
336
error_report("error initializing");
340
if (strstr(conf, "conf=") == NULL) {
341
/* try default location, but ignore failure */
342
rados_conf_read_file(cluster, NULL);
345
if (conf[0] != '\0' &&
346
qemu_rbd_set_conf(cluster, conf) < 0) {
347
error_report("error setting config options");
348
rados_shutdown(cluster);
352
if (rados_connect(cluster) < 0) {
353
error_report("error connecting");
354
rados_shutdown(cluster);
358
if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
359
error_report("error opening pool %s", pool);
360
rados_shutdown(cluster);
364
ret = rbd_create(io_ctx, name, bytes, &obj_order);
365
rados_ioctx_destroy(io_ctx);
366
rados_shutdown(cluster);
372
* This aio completion is being called from qemu_rbd_aio_event_reader()
373
* and runs in qemu context. It schedules a bh, but just in case the aio
374
* was not cancelled before.
376
static void qemu_rbd_complete_aio(RADOSCB *rcb)
378
RBDAIOCB *acb = rcb->acb;
383
if (acb->cmd != RBD_AIO_READ) {
387
} else if (!acb->error) {
388
acb->ret = rcb->size;
392
memset(rcb->buf, 0, rcb->size);
395
} else if (r < rcb->size) {
396
memset(rcb->buf + r, 0, rcb->size - r);
398
acb->ret = rcb->size;
400
} else if (!acb->error) {
404
/* Note that acb->bh can be NULL in case where the aio was cancelled */
405
acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
406
qemu_bh_schedule(acb->bh);
411
* aio fd read handler. It runs in the qemu context and calls the
412
* completion handling of completed rados aio operations.
414
static void qemu_rbd_aio_event_reader(void *opaque)
416
BDRVRBDState *s = opaque;
421
char *p = (char *)&s->event_rcb;
423
/* now read the rcb pointer that was sent from a non qemu thread */
424
ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
425
sizeof(s->event_rcb) - s->event_reader_pos);
427
s->event_reader_pos += ret;
428
if (s->event_reader_pos == sizeof(s->event_rcb)) {
429
s->event_reader_pos = 0;
430
qemu_rbd_complete_aio(s->event_rcb);
434
} while (ret < 0 && errno == EINTR);
437
static int qemu_rbd_aio_flush_cb(void *opaque)
439
BDRVRBDState *s = opaque;
441
return (s->qemu_aio_count > 0);
444
/* TODO Convert to fine grained options */
445
static QemuOptsList runtime_opts = {
447
.head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
451
.type = QEMU_OPT_STRING,
452
.help = "Specification of the rbd image",
454
{ /* end of list */ }
458
static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags)
460
BDRVRBDState *s = bs->opaque;
461
char pool[RBD_MAX_POOL_NAME_SIZE];
462
char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
463
char conf[RBD_MAX_CONF_SIZE];
464
char clientname_buf[RBD_MAX_CONF_SIZE];
467
Error *local_err = NULL;
468
const char *filename;
471
opts = qemu_opts_create_nofail(&runtime_opts);
472
qemu_opts_absorb_qdict(opts, options, &local_err);
473
if (error_is_set(&local_err)) {
474
qerror_report_err(local_err);
475
error_free(local_err);
480
filename = qemu_opt_get(opts, "filename");
482
if (qemu_rbd_parsename(filename, pool, sizeof(pool),
483
snap_buf, sizeof(snap_buf),
484
s->name, sizeof(s->name),
485
conf, sizeof(conf)) < 0) {
490
clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
491
r = rados_create(&s->cluster, clientname);
493
error_report("error initializing");
498
if (snap_buf[0] != '\0') {
499
s->snap = g_strdup(snap_buf);
503
* Fallback to more conservative semantics if setting cache
504
* options fails. Ignore errors from setting rbd_cache because the
505
* only possible error is that the option does not exist, and
506
* librbd defaults to no caching. If write through caching cannot
507
* be set up, fall back to no caching.
509
if (flags & BDRV_O_NOCACHE) {
510
rados_conf_set(s->cluster, "rbd_cache", "false");
512
rados_conf_set(s->cluster, "rbd_cache", "true");
515
if (strstr(conf, "conf=") == NULL) {
516
/* try default location, but ignore failure */
517
rados_conf_read_file(s->cluster, NULL);
520
if (conf[0] != '\0') {
521
r = qemu_rbd_set_conf(s->cluster, conf);
523
error_report("error setting config options");
524
goto failed_shutdown;
528
r = rados_connect(s->cluster);
530
error_report("error connecting");
531
goto failed_shutdown;
534
r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
536
error_report("error opening pool %s", pool);
537
goto failed_shutdown;
540
r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
542
error_report("error reading header from %s", s->name);
546
bs->read_only = (s->snap != NULL);
548
s->event_reader_pos = 0;
549
r = qemu_pipe(s->fds);
551
error_report("error opening eventfd");
554
fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
555
fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
556
qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
557
NULL, qemu_rbd_aio_flush_cb, s);
566
rados_ioctx_destroy(s->io_ctx);
568
rados_shutdown(s->cluster);
575
static void qemu_rbd_close(BlockDriverState *bs)
577
BDRVRBDState *s = bs->opaque;
581
qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL);
584
rados_ioctx_destroy(s->io_ctx);
586
rados_shutdown(s->cluster);
590
* Cancel aio. Since we don't reference acb in a non qemu threads,
591
* it is safe to access it here.
593
static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
595
RBDAIOCB *acb = (RBDAIOCB *) blockacb;
598
while (acb->status == -EINPROGRESS) {
602
qemu_aio_release(acb);
605
static const AIOCBInfo rbd_aiocb_info = {
606
.aiocb_size = sizeof(RBDAIOCB),
607
.cancel = qemu_rbd_aio_cancel,
610
static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
615
int fd = s->fds[RBD_FD_WRITE];
617
/* send the op pointer to the qemu thread that is responsible
618
for the aio/op completion. Must do it in a qemu thread context */
619
ret = write(fd, (void *)&rcb, sizeof(rcb));
623
if (errno == EINTR) {
626
if (errno != EAGAIN) {
633
ret = select(fd + 1, NULL, &wfd, NULL, NULL);
634
} while (ret < 0 && errno == EINTR);
641
* This is the callback function for rbd_aio_read and _write
643
* Note: this function is being called from a non qemu thread so
644
* we need to be careful about what we do here. Generally we only
645
* write to the block notification pipe, and do the rest of the
646
* io completion handling from qemu_rbd_aio_event_reader() which
647
* runs in a qemu context.
649
static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
652
rcb->ret = rbd_aio_get_return_value(c);
654
ret = qemu_rbd_send_pipe(rcb->s, rcb);
656
error_report("failed writing to acb->s->fds");
661
/* Callback when all queued rbd_aio requests are complete */
663
static void rbd_aio_bh_cb(void *opaque)
665
RBDAIOCB *acb = opaque;
667
if (acb->cmd == RBD_AIO_READ) {
668
qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
670
qemu_vfree(acb->bounce);
671
acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
672
qemu_bh_delete(acb->bh);
676
if (!acb->cancelled) {
677
qemu_aio_release(acb);
681
static int rbd_aio_discard_wrapper(rbd_image_t image,
684
rbd_completion_t comp)
686
#ifdef LIBRBD_SUPPORTS_DISCARD
687
return rbd_aio_discard(image, off, len, comp);
693
static int rbd_aio_flush_wrapper(rbd_image_t image,
694
rbd_completion_t comp)
696
#ifdef LIBRBD_SUPPORTS_AIO_FLUSH
697
return rbd_aio_flush(image, comp);
703
static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
707
BlockDriverCompletionFunc *cb,
718
BDRVRBDState *s = bs->opaque;
720
acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
723
if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
726
acb->bounce = qemu_blockalign(bs, qiov->size);
733
acb->status = -EINPROGRESS;
735
if (cmd == RBD_AIO_WRITE) {
736
qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
741
off = sector_num * BDRV_SECTOR_SIZE;
742
size = nb_sectors * BDRV_SECTOR_SIZE;
744
s->qemu_aio_count++; /* All the RADOSCB */
746
rcb = g_malloc(sizeof(RADOSCB));
752
r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
759
r = rbd_aio_write(s->image, off, size, buf, c);
762
r = rbd_aio_read(s->image, off, size, buf, c);
764
case RBD_AIO_DISCARD:
765
r = rbd_aio_discard_wrapper(s->image, off, size, c);
768
r = rbd_aio_flush_wrapper(s->image, c);
783
qemu_aio_release(acb);
787
static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
791
BlockDriverCompletionFunc *cb,
794
return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
798
static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
802
BlockDriverCompletionFunc *cb,
805
return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
809
#ifdef LIBRBD_SUPPORTS_AIO_FLUSH
810
static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
811
BlockDriverCompletionFunc *cb,
814
return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
819
static int qemu_rbd_co_flush(BlockDriverState *bs)
821
#if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
822
/* rbd_flush added in 0.1.1 */
823
BDRVRBDState *s = bs->opaque;
824
return rbd_flush(s->image);
831
static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
833
BDRVRBDState *s = bs->opaque;
834
rbd_image_info_t info;
837
r = rbd_stat(s->image, &info, sizeof(info));
842
bdi->cluster_size = info.obj_size;
846
static int64_t qemu_rbd_getlength(BlockDriverState *bs)
848
BDRVRBDState *s = bs->opaque;
849
rbd_image_info_t info;
852
r = rbd_stat(s->image, &info, sizeof(info));
860
static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
862
BDRVRBDState *s = bs->opaque;
865
r = rbd_resize(s->image, offset);
873
static int qemu_rbd_snap_create(BlockDriverState *bs,
874
QEMUSnapshotInfo *sn_info)
876
BDRVRBDState *s = bs->opaque;
879
if (sn_info->name[0] == '\0') {
880
return -EINVAL; /* we need a name for rbd snapshots */
884
* rbd snapshots are using the name as the user controlled unique identifier
885
* we can't use the rbd snapid for that purpose, as it can't be set
887
if (sn_info->id_str[0] != '\0' &&
888
strcmp(sn_info->id_str, sn_info->name) != 0) {
892
if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
896
r = rbd_snap_create(s->image, sn_info->name);
898
error_report("failed to create snap: %s", strerror(-r));
905
static int qemu_rbd_snap_remove(BlockDriverState *bs,
906
const char *snapshot_name)
908
BDRVRBDState *s = bs->opaque;
911
r = rbd_snap_remove(s->image, snapshot_name);
915
static int qemu_rbd_snap_rollback(BlockDriverState *bs,
916
const char *snapshot_name)
918
BDRVRBDState *s = bs->opaque;
921
r = rbd_snap_rollback(s->image, snapshot_name);
925
static int qemu_rbd_snap_list(BlockDriverState *bs,
926
QEMUSnapshotInfo **psn_tab)
928
BDRVRBDState *s = bs->opaque;
929
QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
931
rbd_snap_info_t *snaps;
932
int max_snaps = RBD_MAX_SNAPS;
935
snaps = g_malloc(sizeof(*snaps) * max_snaps);
936
snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
937
if (snap_count < 0) {
940
} while (snap_count == -ERANGE);
942
if (snap_count <= 0) {
946
sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
948
for (i = 0; i < snap_count; i++) {
949
const char *snap_name = snaps[i].name;
951
sn_info = sn_tab + i;
952
pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
953
pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
955
sn_info->vm_state_size = snaps[i].size;
956
sn_info->date_sec = 0;
957
sn_info->date_nsec = 0;
958
sn_info->vm_clock_nsec = 0;
960
rbd_snap_list_end(snaps);
967
#ifdef LIBRBD_SUPPORTS_DISCARD
968
static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
971
BlockDriverCompletionFunc *cb,
974
return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
979
static QEMUOptionParameter qemu_rbd_create_options[] = {
981
.name = BLOCK_OPT_SIZE,
983
.help = "Virtual disk size"
986
.name = BLOCK_OPT_CLUSTER_SIZE,
988
.help = "RBD object size"
993
static BlockDriver bdrv_rbd = {
994
.format_name = "rbd",
995
.instance_size = sizeof(BDRVRBDState),
996
.bdrv_file_open = qemu_rbd_open,
997
.bdrv_close = qemu_rbd_close,
998
.bdrv_create = qemu_rbd_create,
999
.bdrv_has_zero_init = bdrv_has_zero_init_1,
1000
.bdrv_get_info = qemu_rbd_getinfo,
1001
.create_options = qemu_rbd_create_options,
1002
.bdrv_getlength = qemu_rbd_getlength,
1003
.bdrv_truncate = qemu_rbd_truncate,
1004
.protocol_name = "rbd",
1006
.bdrv_aio_readv = qemu_rbd_aio_readv,
1007
.bdrv_aio_writev = qemu_rbd_aio_writev,
1009
#ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1010
.bdrv_aio_flush = qemu_rbd_aio_flush,
1012
.bdrv_co_flush_to_disk = qemu_rbd_co_flush,
1015
#ifdef LIBRBD_SUPPORTS_DISCARD
1016
.bdrv_aio_discard = qemu_rbd_aio_discard,
1019
.bdrv_snapshot_create = qemu_rbd_snap_create,
1020
.bdrv_snapshot_delete = qemu_rbd_snap_remove,
1021
.bdrv_snapshot_list = qemu_rbd_snap_list,
1022
.bdrv_snapshot_goto = qemu_rbd_snap_rollback,
1025
static void bdrv_rbd_init(void)
1027
bdrv_register(&bdrv_rbd);
1030
block_init(bdrv_rbd_init);