2 * QEMU Block driver for RADOS (Ceph)
7 * This work is licensed under the terms of the GNU GPL, version 2. See
8 * the COPYING file in the top-level directory.
14 #include "qemu-common.h"
15 #include "qemu-error.h"
16 #include "block_int.h"
18 #include <rbd/librbd.h>
21 * When specifying the image filename use:
23 * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
25 * poolname must be the name of an existing rados pool.
27 * devicename is the name of the rbd image.
29 * Each option given is used to configure rados, and may be any valid
30 * Ceph option, "id", or "conf".
32 * The "id" option indicates what user we should authenticate as to
33 * the Ceph cluster. If it is excluded we will use the Ceph default
36 * The "conf" option specifies a Ceph configuration file to read. If
37 * it is not specified, we will read from the default Ceph locations
38 * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration
39 * file, specify conf=/dev/null.
41 * Configuration values containing :, @, or = can be escaped with a
45 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
47 #define RBD_MAX_CONF_NAME_SIZE 128
48 #define RBD_MAX_CONF_VAL_SIZE 512
49 #define RBD_MAX_CONF_SIZE 1024
50 #define RBD_MAX_POOL_NAME_SIZE 128
51 #define RBD_MAX_SNAP_NAME_SIZE 128
52 #define RBD_MAX_SNAPS 100
54 typedef struct RBDAIOCB {
55 BlockDriverAIOCB common;
63 struct BDRVRBDState *s;
67 typedef struct RADOSCB {
70 struct BDRVRBDState *s;
78 #define RBD_FD_WRITE 1
80 typedef struct BDRVRBDState {
85 char name[RBD_MAX_IMAGE_NAME_SIZE];
92 static void rbd_aio_bh_cb(void *opaque);
94 static int qemu_rbd_next_tok(char *dst, int dst_len,
95 char *src, char delim,
105 for (end = src; *end; ++end) {
109 if (*end == '\\' && end[1] != '\0') {
120 error_report("%s too long", name);
123 error_report("%s too short", name);
127 pstrcpy(dst, dst_len, src);
132 static void qemu_rbd_unescape(char *src)
136 for (p = src; *src; ++src, ++p) {
137 if (*src == '\\' && src[1] != '\0') {
145 static int qemu_rbd_parsename(const char *filename,
146 char *pool, int pool_len,
147 char *snap, int snap_len,
148 char *name, int name_len,
149 char *conf, int conf_len)
155 if (!strstart(filename, "rbd:", &start)) {
159 buf = g_strdup(start);
164 ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
169 qemu_rbd_unescape(pool);
171 if (strchr(p, '@')) {
172 ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
176 ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
177 qemu_rbd_unescape(snap);
179 ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
181 qemu_rbd_unescape(name);
186 ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
193 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
195 const char *p = conf;
199 const char *end = strchr(p, ':');
207 if (strncmp(p, "id=", 3) == 0) {
209 strncpy(clientname, p + 3, len);
210 clientname[len] = '\0';
221 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
224 char name[RBD_MAX_CONF_NAME_SIZE];
225 char value[RBD_MAX_CONF_VAL_SIZE];
228 buf = g_strdup(conf);
232 ret = qemu_rbd_next_tok(name, sizeof(name), p,
233 '=', "conf option name", &p);
237 qemu_rbd_unescape(name);
240 error_report("conf option %s has no value", name);
245 ret = qemu_rbd_next_tok(value, sizeof(value), p,
246 ':', "conf option value", &p);
250 qemu_rbd_unescape(value);
252 if (strcmp(name, "conf") == 0) {
253 ret = rados_conf_read_file(cluster, value);
255 error_report("error reading conf file %s", value);
258 } else if (strcmp(name, "id") == 0) {
259 /* ignore, this is parsed by qemu_rbd_parse_clientname() */
261 ret = rados_conf_set(cluster, name, value);
263 error_report("invalid conf option %s", name);
274 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
279 char pool[RBD_MAX_POOL_NAME_SIZE];
280 char name[RBD_MAX_IMAGE_NAME_SIZE];
281 char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
282 char conf[RBD_MAX_CONF_SIZE];
283 char clientname_buf[RBD_MAX_CONF_SIZE];
286 rados_ioctx_t io_ctx;
289 if (qemu_rbd_parsename(filename, pool, sizeof(pool),
290 snap_buf, sizeof(snap_buf),
292 conf, sizeof(conf)) < 0) {
296 /* Read out options */
297 while (options && options->name) {
298 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
299 bytes = options->value.n;
300 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
301 if (options->value.n) {
302 objsize = options->value.n;
303 if ((objsize - 1) & objsize) { /* not a power of 2? */
304 error_report("obj size needs to be power of 2");
307 if (objsize < 4096) {
308 error_report("obj size too small");
311 obj_order = ffs(objsize) - 1;
317 clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
318 if (rados_create(&cluster, clientname) < 0) {
319 error_report("error initializing");
323 if (strstr(conf, "conf=") == NULL) {
324 /* try default location, but ignore failure */
325 rados_conf_read_file(cluster, NULL);
328 if (conf[0] != '\0' &&
329 qemu_rbd_set_conf(cluster, conf) < 0) {
330 error_report("error setting config options");
331 rados_shutdown(cluster);
335 if (rados_connect(cluster) < 0) {
336 error_report("error connecting");
337 rados_shutdown(cluster);
341 if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
342 error_report("error opening pool %s", pool);
343 rados_shutdown(cluster);
347 ret = rbd_create(io_ctx, name, bytes, &obj_order);
348 rados_ioctx_destroy(io_ctx);
349 rados_shutdown(cluster);
355 * This aio completion is being called from qemu_rbd_aio_event_reader()
356 * and runs in qemu context. It schedules a bh, but just in case the aio
357 * was not cancelled before.
359 static void qemu_rbd_complete_aio(RADOSCB *rcb)
361 RBDAIOCB *acb = rcb->acb;
364 if (acb->cancelled) {
365 qemu_vfree(acb->bounce);
366 qemu_aio_release(acb);
376 } else if (!acb->error) {
377 acb->ret = rcb->size;
381 memset(rcb->buf, 0, rcb->size);
384 } else if (r < rcb->size) {
385 memset(rcb->buf + r, 0, rcb->size - r);
387 acb->ret = rcb->size;
389 } else if (!acb->error) {
393 /* Note that acb->bh can be NULL in case where the aio was cancelled */
394 acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
395 qemu_bh_schedule(acb->bh);
401 * aio fd read handler. It runs in the qemu context and calls the
402 * completion handling of completed rados aio operations.
404 static void qemu_rbd_aio_event_reader(void *opaque)
406 BDRVRBDState *s = opaque;
411 char *p = (char *)&s->event_rcb;
413 /* now read the rcb pointer that was sent from a non qemu thread */
414 ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
415 sizeof(s->event_rcb) - s->event_reader_pos);
417 s->event_reader_pos += ret;
418 if (s->event_reader_pos == sizeof(s->event_rcb)) {
419 s->event_reader_pos = 0;
420 qemu_rbd_complete_aio(s->event_rcb);
424 } while (ret < 0 && errno == EINTR);
427 static int qemu_rbd_aio_flush_cb(void *opaque)
429 BDRVRBDState *s = opaque;
431 return (s->qemu_aio_count > 0);
434 static int qemu_rbd_open(BlockDriverState *bs, const char *filename, int flags)
436 BDRVRBDState *s = bs->opaque;
437 char pool[RBD_MAX_POOL_NAME_SIZE];
438 char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
439 char conf[RBD_MAX_CONF_SIZE];
440 char clientname_buf[RBD_MAX_CONF_SIZE];
444 if (qemu_rbd_parsename(filename, pool, sizeof(pool),
445 snap_buf, sizeof(snap_buf),
446 s->name, sizeof(s->name),
447 conf, sizeof(conf)) < 0) {
451 clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
452 r = rados_create(&s->cluster, clientname);
454 error_report("error initializing");
459 if (snap_buf[0] != '\0') {
460 s->snap = g_strdup(snap_buf);
463 if (strstr(conf, "conf=") == NULL) {
464 /* try default location, but ignore failure */
465 rados_conf_read_file(s->cluster, NULL);
468 if (conf[0] != '\0') {
469 r = qemu_rbd_set_conf(s->cluster, conf);
471 error_report("error setting config options");
472 goto failed_shutdown;
476 r = rados_connect(s->cluster);
478 error_report("error connecting");
479 goto failed_shutdown;
482 r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
484 error_report("error opening pool %s", pool);
485 goto failed_shutdown;
488 r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
490 error_report("error reading header from %s", s->name);
494 bs->read_only = (s->snap != NULL);
496 s->event_reader_pos = 0;
497 r = qemu_pipe(s->fds);
499 error_report("error opening eventfd");
502 fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
503 fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
504 qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
505 NULL, qemu_rbd_aio_flush_cb, NULL, s);
513 rados_ioctx_destroy(s->io_ctx);
515 rados_shutdown(s->cluster);
520 static void qemu_rbd_close(BlockDriverState *bs)
522 BDRVRBDState *s = bs->opaque;
526 qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL , NULL, NULL, NULL,
530 rados_ioctx_destroy(s->io_ctx);
532 rados_shutdown(s->cluster);
536 * Cancel aio. Since we don't reference acb in a non qemu threads,
537 * it is safe to access it here.
539 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
541 RBDAIOCB *acb = (RBDAIOCB *) blockacb;
545 static AIOPool rbd_aio_pool = {
546 .aiocb_size = sizeof(RBDAIOCB),
547 .cancel = qemu_rbd_aio_cancel,
550 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
555 int fd = s->fds[RBD_FD_WRITE];
557 /* send the op pointer to the qemu thread that is responsible
558 for the aio/op completion. Must do it in a qemu thread context */
559 ret = write(fd, (void *)&rcb, sizeof(rcb));
563 if (errno == EINTR) {
566 if (errno != EAGAIN) {
573 ret = select(fd + 1, NULL, &wfd, NULL, NULL);
574 } while (ret < 0 && errno == EINTR);
581 * This is the callback function for rbd_aio_read and _write
583 * Note: this function is being called from a non qemu thread so
584 * we need to be careful about what we do here. Generally we only
585 * write to the block notification pipe, and do the rest of the
586 * io completion handling from qemu_rbd_aio_event_reader() which
587 * runs in a qemu context.
589 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
592 rcb->ret = rbd_aio_get_return_value(c);
594 ret = qemu_rbd_send_pipe(rcb->s, rcb);
596 error_report("failed writing to acb->s->fds");
601 /* Callback when all queued rbd_aio requests are complete */
603 static void rbd_aio_bh_cb(void *opaque)
605 RBDAIOCB *acb = opaque;
608 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
610 qemu_vfree(acb->bounce);
611 acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
612 qemu_bh_delete(acb->bh);
615 qemu_aio_release(acb);
618 static BlockDriverAIOCB *rbd_aio_rw_vector(BlockDriverState *bs,
622 BlockDriverCompletionFunc *cb,
623 void *opaque, int write)
632 BDRVRBDState *s = bs->opaque;
634 acb = qemu_aio_get(&rbd_aio_pool, bs, cb, opaque);
640 acb->bounce = qemu_blockalign(bs, qiov->size);
648 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
653 off = sector_num * BDRV_SECTOR_SIZE;
654 size = nb_sectors * BDRV_SECTOR_SIZE;
656 s->qemu_aio_count++; /* All the RADOSCB */
658 rcb = g_malloc(sizeof(RADOSCB));
664 r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
670 r = rbd_aio_write(s->image, off, size, buf, c);
672 r = rbd_aio_read(s->image, off, size, buf, c);
684 qemu_aio_release(acb);
688 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
692 BlockDriverCompletionFunc *cb,
695 return rbd_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
698 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
702 BlockDriverCompletionFunc *cb,
705 return rbd_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
708 static int qemu_rbd_flush(BlockDriverState *bs)
710 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
711 /* rbd_flush added in 0.1.1 */
712 BDRVRBDState *s = bs->opaque;
713 return rbd_flush(s->image);
719 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
721 BDRVRBDState *s = bs->opaque;
722 rbd_image_info_t info;
725 r = rbd_stat(s->image, &info, sizeof(info));
730 bdi->cluster_size = info.obj_size;
734 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
736 BDRVRBDState *s = bs->opaque;
737 rbd_image_info_t info;
740 r = rbd_stat(s->image, &info, sizeof(info));
748 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
750 BDRVRBDState *s = bs->opaque;
753 r = rbd_resize(s->image, offset);
761 static int qemu_rbd_snap_create(BlockDriverState *bs,
762 QEMUSnapshotInfo *sn_info)
764 BDRVRBDState *s = bs->opaque;
767 if (sn_info->name[0] == '\0') {
768 return -EINVAL; /* we need a name for rbd snapshots */
772 * rbd snapshots are using the name as the user controlled unique identifier
773 * we can't use the rbd snapid for that purpose, as it can't be set
775 if (sn_info->id_str[0] != '\0' &&
776 strcmp(sn_info->id_str, sn_info->name) != 0) {
780 if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
784 r = rbd_snap_create(s->image, sn_info->name);
786 error_report("failed to create snap: %s", strerror(-r));
793 static int qemu_rbd_snap_list(BlockDriverState *bs,
794 QEMUSnapshotInfo **psn_tab)
796 BDRVRBDState *s = bs->opaque;
797 QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
799 rbd_snap_info_t *snaps;
800 int max_snaps = RBD_MAX_SNAPS;
803 snaps = g_malloc(sizeof(*snaps) * max_snaps);
804 snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
805 if (snap_count < 0) {
808 } while (snap_count == -ERANGE);
810 if (snap_count <= 0) {
814 sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
816 for (i = 0; i < snap_count; i++) {
817 const char *snap_name = snaps[i].name;
819 sn_info = sn_tab + i;
820 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
821 pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
823 sn_info->vm_state_size = snaps[i].size;
824 sn_info->date_sec = 0;
825 sn_info->date_nsec = 0;
826 sn_info->vm_clock_nsec = 0;
828 rbd_snap_list_end(snaps);
834 static QEMUOptionParameter qemu_rbd_create_options[] = {
836 .name = BLOCK_OPT_SIZE,
838 .help = "Virtual disk size"
841 .name = BLOCK_OPT_CLUSTER_SIZE,
843 .help = "RBD object size"
848 static BlockDriver bdrv_rbd = {
849 .format_name = "rbd",
850 .instance_size = sizeof(BDRVRBDState),
851 .bdrv_file_open = qemu_rbd_open,
852 .bdrv_close = qemu_rbd_close,
853 .bdrv_create = qemu_rbd_create,
854 .bdrv_flush = qemu_rbd_flush,
855 .bdrv_get_info = qemu_rbd_getinfo,
856 .create_options = qemu_rbd_create_options,
857 .bdrv_getlength = qemu_rbd_getlength,
858 .bdrv_truncate = qemu_rbd_truncate,
859 .protocol_name = "rbd",
861 .bdrv_aio_readv = qemu_rbd_aio_readv,
862 .bdrv_aio_writev = qemu_rbd_aio_writev,
864 .bdrv_snapshot_create = qemu_rbd_snap_create,
865 .bdrv_snapshot_list = qemu_rbd_snap_list,
868 static void bdrv_rbd_init(void)
870 bdrv_register(&bdrv_rbd);
873 block_init(bdrv_rbd_init);