2 * Copyright (C) 2009-2010 Nippon Telegraph and Telephone Corporation.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License version
6 * 2 as published by the Free Software Foundation.
8 * You should have received a copy of the GNU General Public License
9 * along with this program. If not, see <http://www.gnu.org/licenses/>.
11 * Contributions after 2012-01-13 are licensed under the terms of the
12 * GNU GPL, version 2 or (at your option) any later version.
15 #include "qemu-common.h"
17 #include "qemu/error-report.h"
18 #include "qemu/sockets.h"
19 #include "block/block_int.h"
20 #include "qemu/bitops.h"
22 #define SD_PROTO_VER 0x01
24 #define SD_DEFAULT_ADDR "localhost"
25 #define SD_DEFAULT_PORT 7000
27 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
28 #define SD_OP_READ_OBJ 0x02
29 #define SD_OP_WRITE_OBJ 0x03
30 /* 0x04 is used internally by Sheepdog */
31 #define SD_OP_DISCARD_OBJ 0x05
33 #define SD_OP_NEW_VDI 0x11
34 #define SD_OP_LOCK_VDI 0x12
35 #define SD_OP_RELEASE_VDI 0x13
36 #define SD_OP_GET_VDI_INFO 0x14
37 #define SD_OP_READ_VDIS 0x15
38 #define SD_OP_FLUSH_VDI 0x16
39 #define SD_OP_DEL_VDI 0x17
41 #define SD_FLAG_CMD_WRITE 0x01
42 #define SD_FLAG_CMD_COW 0x02
43 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
44 #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
46 #define SD_RES_SUCCESS 0x00 /* Success */
47 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
48 #define SD_RES_NO_OBJ 0x02 /* No object found */
49 #define SD_RES_EIO 0x03 /* I/O error */
50 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
51 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
52 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
53 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
54 #define SD_RES_NO_VDI 0x08 /* No vdi found */
55 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
56 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
57 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
58 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
59 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
60 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
61 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
62 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
63 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
64 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
65 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
66 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
67 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
68 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
69 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
70 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
71 #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */
72 #define SD_RES_READONLY 0x1A /* Object is read-only */
77 * 0 - 19 (20 bits): data object space
78 * 20 - 31 (12 bits): reserved data object space
79 * 32 - 55 (24 bits): vdi object space
80 * 56 - 59 ( 4 bits): reserved vdi object space
81 * 60 - 63 ( 4 bits): object type identifier space
84 #define VDI_SPACE_SHIFT 32
85 #define VDI_BIT (UINT64_C(1) << 63)
86 #define VMSTATE_BIT (UINT64_C(1) << 62)
87 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
88 #define MAX_CHILDREN 1024
89 #define SD_MAX_VDI_LEN 256
90 #define SD_MAX_VDI_TAG_LEN 256
91 #define SD_NR_VDIS (1U << 24)
92 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
93 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
95 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
96 * (SD_EC_MAX_STRIP - 1) for parity strips
98 * SD_MAX_COPIES is sum of number of data strips and parity strips.
100 #define SD_EC_MAX_STRIP 16
101 #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
103 #define SD_INODE_SIZE (sizeof(SheepdogInode))
104 #define CURRENT_VDI_ID 0
106 #define LOCK_TYPE_NORMAL 0
107 #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */
109 typedef struct SheepdogReq {
115 uint32_t data_length;
116 uint32_t opcode_specific[8];
119 typedef struct SheepdogRsp {
125 uint32_t data_length;
127 uint32_t opcode_specific[7];
130 typedef struct SheepdogObjReq {
136 uint32_t data_length;
145 typedef struct SheepdogObjRsp {
151 uint32_t data_length;
159 typedef struct SheepdogVdiReq {
165 uint32_t data_length;
167 uint32_t base_vdi_id;
176 typedef struct SheepdogVdiRsp {
182 uint32_t data_length;
189 typedef struct SheepdogInode {
190 char name[SD_MAX_VDI_LEN];
191 char tag[SD_MAX_VDI_TAG_LEN];
194 uint64_t vm_clock_nsec;
196 uint64_t vm_state_size;
197 uint16_t copy_policy;
199 uint8_t block_size_shift;
202 uint32_t parent_vdi_id;
203 uint32_t child_vdi_id[MAX_CHILDREN];
204 uint32_t data_vdi_id[MAX_DATA_OBJS];
207 #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
210 * 64 bit FNV-1a non-zero initial basis
212 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
215 * 64 bit Fowler/Noll/Vo FNV-1a hash code
217 static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
219 unsigned char *bp = buf;
220 unsigned char *be = bp + len;
222 hval ^= (uint64_t) *bp++;
223 hval += (hval << 1) + (hval << 4) + (hval << 5) +
224 (hval << 7) + (hval << 8) + (hval << 40);
229 static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
231 return inode->vdi_id == inode->data_vdi_id[idx];
234 static inline bool is_data_obj(uint64_t oid)
236 return !(VDI_BIT & oid);
239 static inline uint64_t data_oid_to_idx(uint64_t oid)
241 return oid & (MAX_DATA_OBJS - 1);
244 static inline uint32_t oid_to_vid(uint64_t oid)
246 return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT;
249 static inline uint64_t vid_to_vdi_oid(uint32_t vid)
251 return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
254 static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx)
256 return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
259 static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx)
261 return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
264 static inline bool is_snapshot(struct SheepdogInode *inode)
266 return !!inode->snap_ctime;
271 #define DPRINTF(fmt, args...) \
273 fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \
276 #define DPRINTF(fmt, args...)
279 typedef struct SheepdogAIOCB SheepdogAIOCB;
281 typedef struct AIOReq {
282 SheepdogAIOCB *aiocb;
283 unsigned int iov_offset;
288 unsigned int data_len;
293 QLIST_ENTRY(AIOReq) aio_siblings;
303 struct SheepdogAIOCB {
304 BlockDriverAIOCB common;
312 enum AIOCBState aiocb_type;
314 Coroutine *coroutine;
315 void (*aio_done_func)(SheepdogAIOCB *);
321 typedef struct BDRVSheepdogState {
322 BlockDriverState *bs;
323 AioContext *aio_context;
327 uint32_t min_dirty_data_idx;
328 uint32_t max_dirty_data_idx;
330 char name[SD_MAX_VDI_LEN];
332 uint32_t cache_flags;
333 bool discard_supported;
343 uint32_t aioreq_seq_num;
345 /* Every aio request must be linked to either of these queues. */
346 QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head;
347 QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head;
348 QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head;
351 static const char * sd_strerror(int err)
355 static const struct {
359 {SD_RES_SUCCESS, "Success"},
360 {SD_RES_UNKNOWN, "Unknown error"},
361 {SD_RES_NO_OBJ, "No object found"},
362 {SD_RES_EIO, "I/O error"},
363 {SD_RES_VDI_EXIST, "VDI exists already"},
364 {SD_RES_INVALID_PARMS, "Invalid parameters"},
365 {SD_RES_SYSTEM_ERROR, "System error"},
366 {SD_RES_VDI_LOCKED, "VDI is already locked"},
367 {SD_RES_NO_VDI, "No vdi found"},
368 {SD_RES_NO_BASE_VDI, "No base VDI found"},
369 {SD_RES_VDI_READ, "Failed read the requested VDI"},
370 {SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
371 {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
372 {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
373 {SD_RES_NO_TAG, "Failed to find the requested tag"},
374 {SD_RES_STARTUP, "The system is still booting"},
375 {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
376 {SD_RES_SHUTDOWN, "The system is shutting down"},
377 {SD_RES_NO_MEM, "Out of memory on the server"},
378 {SD_RES_FULL_VDI, "We already have the maximum vdis"},
379 {SD_RES_VER_MISMATCH, "Protocol version mismatch"},
380 {SD_RES_NO_SPACE, "Server has no space for new objects"},
381 {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
382 {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
383 {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
384 {SD_RES_HALT, "Sheepdog is stopped serving IO request"},
385 {SD_RES_READONLY, "Object is read-only"},
388 for (i = 0; i < ARRAY_SIZE(errors); ++i) {
389 if (errors[i].err == err) {
390 return errors[i].desc;
394 return "Invalid error code";
398 * Sheepdog I/O handling:
400 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
401 * link the requests to the inflight_list in the
402 * BDRVSheepdogState. The function exits without waiting for
403 * receiving the response.
405 * 2. We receive the response in aio_read_response, the fd handler to
406 * the sheepdog connection. If metadata update is needed, we send
407 * the write request to the vdi object in sd_write_done, the write
408 * completion function. We switch back to sd_co_readv/writev after
409 * all the requests belonging to the AIOCB are finished.
412 static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
413 uint64_t oid, unsigned int data_len,
414 uint64_t offset, uint8_t flags, bool create,
415 uint64_t base_oid, unsigned int iov_offset)
419 aio_req = g_malloc(sizeof(*aio_req));
420 aio_req->aiocb = acb;
421 aio_req->iov_offset = iov_offset;
423 aio_req->base_oid = base_oid;
424 aio_req->offset = offset;
425 aio_req->data_len = data_len;
426 aio_req->flags = flags;
427 aio_req->id = s->aioreq_seq_num++;
428 aio_req->create = create;
434 static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req)
436 SheepdogAIOCB *acb = aio_req->aiocb;
438 acb->cancelable = false;
439 QLIST_REMOVE(aio_req, aio_siblings);
445 static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb)
447 qemu_coroutine_enter(acb->coroutine, NULL);
452 * Check whether the specified acb can be canceled
454 * We can cancel aio when any request belonging to the acb is:
455 * - Not processed by the sheepdog server.
456 * - Not linked to the inflight queue.
458 static bool sd_acb_cancelable(const SheepdogAIOCB *acb)
460 BDRVSheepdogState *s = acb->common.bs->opaque;
463 if (!acb->cancelable) {
467 QLIST_FOREACH(aioreq, &s->inflight_aio_head, aio_siblings) {
468 if (aioreq->aiocb == acb) {
476 static void sd_aio_cancel(BlockDriverAIOCB *blockacb)
478 SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb;
479 BDRVSheepdogState *s = acb->common.bs->opaque;
480 AIOReq *aioreq, *next;
482 if (sd_acb_cancelable(acb)) {
483 /* Remove outstanding requests from pending and failed queues. */
484 QLIST_FOREACH_SAFE(aioreq, &s->pending_aio_head, aio_siblings,
486 if (aioreq->aiocb == acb) {
487 free_aio_req(s, aioreq);
490 QLIST_FOREACH_SAFE(aioreq, &s->failed_aio_head, aio_siblings,
492 if (aioreq->aiocb == acb) {
493 free_aio_req(s, aioreq);
497 assert(acb->nr_pending == 0);
498 if (acb->common.cb) {
499 acb->common.cb(acb->common.opaque, -ECANCELED);
501 sd_finish_aiocb(acb);
505 static const AIOCBInfo sd_aiocb_info = {
506 .aiocb_size = sizeof(SheepdogAIOCB),
507 .cancel_async = sd_aio_cancel,
510 static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov,
511 int64_t sector_num, int nb_sectors)
515 acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL);
519 acb->sector_num = sector_num;
520 acb->nb_sectors = nb_sectors;
522 acb->aio_done_func = NULL;
523 acb->cancelable = true;
524 acb->coroutine = qemu_coroutine_self();
530 static int connect_to_sdog(BDRVSheepdogState *s, Error **errp)
535 fd = unix_connect(s->host_spec, errp);
537 fd = inet_connect(s->host_spec, errp);
540 int ret = socket_set_nodelay(fd);
542 error_report("%s", strerror(errno));
548 qemu_set_nonblock(fd);
554 static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
559 ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
560 if (ret != sizeof(*hdr)) {
561 error_report("failed to send a req, %s", strerror(errno));
565 ret = qemu_co_send(sockfd, data, *wlen);
567 error_report("failed to send a req, %s", strerror(errno));
573 static void restart_co_req(void *opaque)
575 Coroutine *co = opaque;
577 qemu_coroutine_enter(co, NULL);
580 typedef struct SheepdogReqCo {
582 AioContext *aio_context;
591 static coroutine_fn void do_co_req(void *opaque)
595 SheepdogReqCo *srco = opaque;
596 int sockfd = srco->sockfd;
597 SheepdogReq *hdr = srco->hdr;
598 void *data = srco->data;
599 unsigned int *wlen = srco->wlen;
600 unsigned int *rlen = srco->rlen;
602 co = qemu_coroutine_self();
603 aio_set_fd_handler(srco->aio_context, sockfd, NULL, restart_co_req, co);
605 ret = send_co_req(sockfd, hdr, data, wlen);
610 aio_set_fd_handler(srco->aio_context, sockfd, restart_co_req, NULL, co);
612 ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
613 if (ret != sizeof(*hdr)) {
614 error_report("failed to get a rsp, %s", strerror(errno));
619 if (*rlen > hdr->data_length) {
620 *rlen = hdr->data_length;
624 ret = qemu_co_recv(sockfd, data, *rlen);
626 error_report("failed to get the data, %s", strerror(errno));
633 /* there is at most one request for this sockfd, so it is safe to
634 * set each handler to NULL. */
635 aio_set_fd_handler(srco->aio_context, sockfd, NULL, NULL, NULL);
638 srco->finished = true;
641 static int do_req(int sockfd, AioContext *aio_context, SheepdogReq *hdr,
642 void *data, unsigned int *wlen, unsigned int *rlen)
645 SheepdogReqCo srco = {
647 .aio_context = aio_context,
656 if (qemu_in_coroutine()) {
659 co = qemu_coroutine_create(do_co_req);
660 qemu_coroutine_enter(co, &srco);
661 while (!srco.finished) {
662 aio_poll(aio_context, true);
669 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
670 struct iovec *iov, int niov,
671 enum AIOCBState aiocb_type);
672 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req);
673 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag);
674 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp);
675 static void co_write_request(void *opaque);
677 static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid)
681 QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) {
682 if (aio_req->oid == oid) {
691 * This function searchs pending requests to the object `oid', and
694 static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid)
699 while ((aio_req = find_pending_req(s, oid)) != NULL) {
700 acb = aio_req->aiocb;
701 /* move aio_req from pending list to inflight one */
702 QLIST_REMOVE(aio_req, aio_siblings);
703 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
704 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
709 static coroutine_fn void reconnect_to_sdog(void *opaque)
711 BDRVSheepdogState *s = opaque;
712 AIOReq *aio_req, *next;
714 aio_set_fd_handler(s->aio_context, s->fd, NULL, NULL, NULL);
718 /* Wait for outstanding write requests to be completed. */
719 while (s->co_send != NULL) {
720 co_write_request(opaque);
723 /* Try to reconnect the sheepdog server every one second. */
725 Error *local_err = NULL;
726 s->fd = get_sheep_fd(s, &local_err);
728 DPRINTF("Wait for connection to be established\n");
729 error_report("%s", error_get_pretty(local_err));
730 error_free(local_err);
731 co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME,
737 * Now we have to resend all the request in the inflight queue. However,
738 * resend_aioreq() can yield and newly created requests can be added to the
739 * inflight queue before the coroutine is resumed. To avoid mixing them, we
740 * have to move all the inflight requests to the failed queue before
741 * resend_aioreq() is called.
743 QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) {
744 QLIST_REMOVE(aio_req, aio_siblings);
745 QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings);
748 /* Resend all the failed aio requests. */
749 while (!QLIST_EMPTY(&s->failed_aio_head)) {
750 aio_req = QLIST_FIRST(&s->failed_aio_head);
751 QLIST_REMOVE(aio_req, aio_siblings);
752 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
753 resend_aioreq(s, aio_req);
758 * Receive responses of the I/O requests.
760 * This function is registered as a fd handler, and called from the
761 * main loop when s->fd is ready for reading responses.
763 static void coroutine_fn aio_read_response(void *opaque)
766 BDRVSheepdogState *s = opaque;
769 AIOReq *aio_req = NULL;
774 ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
775 if (ret != sizeof(rsp)) {
776 error_report("failed to get the header, %s", strerror(errno));
780 /* find the right aio_req from the inflight aio list */
781 QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
782 if (aio_req->id == rsp.id) {
787 error_report("cannot find aio_req %x", rsp.id);
791 acb = aio_req->aiocb;
793 switch (acb->aiocb_type) {
794 case AIOCB_WRITE_UDATA:
795 /* this coroutine context is no longer suitable for co_recv
796 * because we may send data to update vdi objects */
798 if (!is_data_obj(aio_req->oid)) {
801 idx = data_oid_to_idx(aio_req->oid);
803 if (aio_req->create) {
805 * If the object is newly created one, we need to update
806 * the vdi object (metadata object). min_dirty_data_idx
807 * and max_dirty_data_idx are changed to include updated
808 * index between them.
810 if (rsp.result == SD_RES_SUCCESS) {
811 s->inode.data_vdi_id[idx] = s->inode.vdi_id;
812 s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx);
813 s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx);
816 * Some requests may be blocked because simultaneous
817 * create requests are not allowed, so we search the
818 * pending requests here.
820 send_pending_req(s, aio_req->oid);
823 case AIOCB_READ_UDATA:
824 ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
825 aio_req->iov_offset, rsp.data_length);
826 if (ret != rsp.data_length) {
827 error_report("failed to get the data, %s", strerror(errno));
831 case AIOCB_FLUSH_CACHE:
832 if (rsp.result == SD_RES_INVALID_PARMS) {
833 DPRINTF("disable cache since the server doesn't support it\n");
834 s->cache_flags = SD_FLAG_CMD_DIRECT;
835 rsp.result = SD_RES_SUCCESS;
838 case AIOCB_DISCARD_OBJ:
839 switch (rsp.result) {
840 case SD_RES_INVALID_PARMS:
841 error_report("sheep(%s) doesn't support discard command",
843 rsp.result = SD_RES_SUCCESS;
844 s->discard_supported = false;
847 idx = data_oid_to_idx(aio_req->oid);
848 s->inode.data_vdi_id[idx] = 0;
855 switch (rsp.result) {
858 case SD_RES_READONLY:
859 if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) {
860 ret = reload_inode(s, 0, "");
865 if (is_data_obj(aio_req->oid)) {
866 aio_req->oid = vid_to_data_oid(s->inode.vdi_id,
867 data_oid_to_idx(aio_req->oid));
869 aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id);
871 resend_aioreq(s, aio_req);
875 error_report("%s", sd_strerror(rsp.result));
879 free_aio_req(s, aio_req);
880 if (!acb->nr_pending) {
882 * We've finished all requests which belong to the AIOCB, so
883 * we can switch back to sd_co_readv/writev now.
885 acb->aio_done_func(acb);
892 reconnect_to_sdog(opaque);
895 static void co_read_response(void *opaque)
897 BDRVSheepdogState *s = opaque;
900 s->co_recv = qemu_coroutine_create(aio_read_response);
903 qemu_coroutine_enter(s->co_recv, opaque);
906 static void co_write_request(void *opaque)
908 BDRVSheepdogState *s = opaque;
910 qemu_coroutine_enter(s->co_send, NULL);
914 * Return a socket descriptor to read/write objects.
916 * We cannot use this descriptor for other operations because
917 * the block driver may be on waiting response from the server.
919 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp)
923 fd = connect_to_sdog(s, errp);
928 aio_set_fd_handler(s->aio_context, fd, co_read_response, NULL, s);
932 static int sd_parse_uri(BDRVSheepdogState *s, const char *filename,
933 char *vdi, uint32_t *snapid, char *tag)
936 QueryParams *qp = NULL;
939 uri = uri_parse(filename);
945 if (!strcmp(uri->scheme, "sheepdog")) {
947 } else if (!strcmp(uri->scheme, "sheepdog+tcp")) {
949 } else if (!strcmp(uri->scheme, "sheepdog+unix")) {
956 if (uri->path == NULL || !strcmp(uri->path, "/")) {
960 pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1);
962 qp = query_params_parse(uri->query);
963 if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) {
969 /* sheepdog+unix:///vdiname?socket=path */
970 if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
974 s->host_spec = g_strdup(qp->p[0].value);
976 /* sheepdog[+tcp]://[host:port]/vdiname */
977 s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR,
978 uri->port ?: SD_DEFAULT_PORT);
983 *snapid = strtoul(uri->fragment, NULL, 10);
985 pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment);
988 *snapid = CURRENT_VDI_ID; /* search current vdi */
993 query_params_free(qp);
1000 * Parse a filename (old syntax)
1002 * filename must be one of the following formats:
1004 * 2. [vdiname]:[snapid]
1005 * 3. [vdiname]:[tag]
1006 * 4. [hostname]:[port]:[vdiname]
1007 * 5. [hostname]:[port]:[vdiname]:[snapid]
1008 * 6. [hostname]:[port]:[vdiname]:[tag]
1010 * You can boot from the snapshot images by specifying `snapid` or
1013 * You can run VMs outside the Sheepdog cluster by specifying
1014 * `hostname' and `port' (experimental).
1016 static int parse_vdiname(BDRVSheepdogState *s, const char *filename,
1017 char *vdi, uint32_t *snapid, char *tag)
1020 const char *host_spec, *vdi_spec;
1023 strstart(filename, "sheepdog:", (const char **)&filename);
1024 p = q = g_strdup(filename);
1026 /* count the number of separators */
1036 /* use the first two tokens as host_spec. */
1049 p = strchr(vdi_spec, ':');
1054 uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec);
1056 ret = sd_parse_uri(s, uri, vdi, snapid, tag);
1064 static int find_vdi_name(BDRVSheepdogState *s, const char *filename,
1065 uint32_t snapid, const char *tag, uint32_t *vid,
1066 bool lock, Error **errp)
1070 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1071 unsigned int wlen, rlen = 0;
1072 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
1074 fd = connect_to_sdog(s, errp);
1079 /* This pair of strncpy calls ensures that the buffer is zero-filled,
1080 * which is desirable since we'll soon be sending those bytes, and
1081 * don't want the send_req to read uninitialized data.
1083 strncpy(buf, filename, SD_MAX_VDI_LEN);
1084 strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
1086 memset(&hdr, 0, sizeof(hdr));
1088 hdr.opcode = SD_OP_LOCK_VDI;
1089 hdr.type = LOCK_TYPE_NORMAL;
1091 hdr.opcode = SD_OP_GET_VDI_INFO;
1093 wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
1094 hdr.proto_ver = SD_PROTO_VER;
1095 hdr.data_length = wlen;
1096 hdr.snapid = snapid;
1097 hdr.flags = SD_FLAG_CMD_WRITE;
1099 ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1101 error_setg_errno(errp, -ret, "cannot get vdi info");
1105 if (rsp->result != SD_RES_SUCCESS) {
1106 error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s",
1107 sd_strerror(rsp->result), filename, snapid, tag);
1108 if (rsp->result == SD_RES_NO_VDI) {
1110 } else if (rsp->result == SD_RES_VDI_LOCKED) {
1125 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
1126 struct iovec *iov, int niov,
1127 enum AIOCBState aiocb_type)
1129 int nr_copies = s->inode.nr_copies;
1131 unsigned int wlen = 0;
1133 uint64_t oid = aio_req->oid;
1134 unsigned int datalen = aio_req->data_len;
1135 uint64_t offset = aio_req->offset;
1136 uint8_t flags = aio_req->flags;
1137 uint64_t old_oid = aio_req->base_oid;
1138 bool create = aio_req->create;
1141 error_report("bug");
1144 memset(&hdr, 0, sizeof(hdr));
1146 switch (aiocb_type) {
1147 case AIOCB_FLUSH_CACHE:
1148 hdr.opcode = SD_OP_FLUSH_VDI;
1150 case AIOCB_READ_UDATA:
1151 hdr.opcode = SD_OP_READ_OBJ;
1154 case AIOCB_WRITE_UDATA:
1156 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1158 hdr.opcode = SD_OP_WRITE_OBJ;
1161 hdr.flags = SD_FLAG_CMD_WRITE | flags;
1163 case AIOCB_DISCARD_OBJ:
1164 hdr.opcode = SD_OP_DISCARD_OBJ;
1168 if (s->cache_flags) {
1169 hdr.flags |= s->cache_flags;
1173 hdr.cow_oid = old_oid;
1174 hdr.copies = s->inode.nr_copies;
1176 hdr.data_length = datalen;
1177 hdr.offset = offset;
1179 hdr.id = aio_req->id;
1181 qemu_co_mutex_lock(&s->lock);
1182 s->co_send = qemu_coroutine_self();
1183 aio_set_fd_handler(s->aio_context, s->fd,
1184 co_read_response, co_write_request, s);
1185 socket_set_cork(s->fd, 1);
1188 ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
1189 if (ret != sizeof(hdr)) {
1190 error_report("failed to send a req, %s", strerror(errno));
1195 ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1197 error_report("failed to send a data, %s", strerror(errno));
1201 socket_set_cork(s->fd, 0);
1202 aio_set_fd_handler(s->aio_context, s->fd, co_read_response, NULL, s);
1204 qemu_co_mutex_unlock(&s->lock);
1207 static int read_write_object(int fd, AioContext *aio_context, char *buf,
1208 uint64_t oid, uint8_t copies,
1209 unsigned int datalen, uint64_t offset,
1210 bool write, bool create, uint32_t cache_flags)
1213 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1214 unsigned int wlen, rlen;
1217 memset(&hdr, 0, sizeof(hdr));
1222 hdr.flags = SD_FLAG_CMD_WRITE;
1224 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1226 hdr.opcode = SD_OP_WRITE_OBJ;
1231 hdr.opcode = SD_OP_READ_OBJ;
1234 hdr.flags |= cache_flags;
1237 hdr.data_length = datalen;
1238 hdr.offset = offset;
1239 hdr.copies = copies;
1241 ret = do_req(fd, aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1243 error_report("failed to send a request to the sheep");
1247 switch (rsp->result) {
1248 case SD_RES_SUCCESS:
1251 error_report("%s", sd_strerror(rsp->result));
1256 static int read_object(int fd, AioContext *aio_context, char *buf,
1257 uint64_t oid, uint8_t copies,
1258 unsigned int datalen, uint64_t offset,
1259 uint32_t cache_flags)
1261 return read_write_object(fd, aio_context, buf, oid, copies,
1262 datalen, offset, false,
1263 false, cache_flags);
1266 static int write_object(int fd, AioContext *aio_context, char *buf,
1267 uint64_t oid, uint8_t copies,
1268 unsigned int datalen, uint64_t offset, bool create,
1269 uint32_t cache_flags)
1271 return read_write_object(fd, aio_context, buf, oid, copies,
1272 datalen, offset, true,
1273 create, cache_flags);
1276 /* update inode with the latest state */
1277 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
1279 Error *local_err = NULL;
1280 SheepdogInode *inode;
1284 fd = connect_to_sdog(s, &local_err);
1286 error_report("%s", error_get_pretty(local_err));;
1287 error_free(local_err);
1291 inode = g_malloc(SD_INODE_HEADER_SIZE);
1293 ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err);
1295 error_report("%s", error_get_pretty(local_err));;
1296 error_free(local_err);
1300 ret = read_object(fd, s->aio_context, (char *)inode, vid_to_vdi_oid(vid),
1301 s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0,
1307 if (inode->vdi_id != s->inode.vdi_id) {
1308 memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE);
1318 /* Return true if the specified request is linked to the pending list. */
1319 static bool check_simultaneous_create(BDRVSheepdogState *s, AIOReq *aio_req)
1322 QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) {
1323 if (areq != aio_req && areq->oid == aio_req->oid) {
1325 * Sheepdog cannot handle simultaneous create requests to the same
1326 * object, so we cannot send the request until the previous request
1329 DPRINTF("simultaneous create to %" PRIx64 "\n", aio_req->oid);
1331 aio_req->base_oid = 0;
1332 aio_req->create = false;
1333 QLIST_REMOVE(aio_req, aio_siblings);
1334 QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, aio_siblings);
1342 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
1344 SheepdogAIOCB *acb = aio_req->aiocb;
1346 aio_req->create = false;
1348 /* check whether this request becomes a CoW one */
1349 if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) {
1350 int idx = data_oid_to_idx(aio_req->oid);
1352 if (is_data_obj_writable(&s->inode, idx)) {
1356 if (check_simultaneous_create(s, aio_req)) {
1360 if (s->inode.data_vdi_id[idx]) {
1361 aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx);
1362 aio_req->flags |= SD_FLAG_CMD_COW;
1364 aio_req->create = true;
1367 if (is_data_obj(aio_req->oid)) {
1368 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
1372 iov.iov_base = &s->inode;
1373 iov.iov_len = sizeof(s->inode);
1374 add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
1378 static void sd_detach_aio_context(BlockDriverState *bs)
1380 BDRVSheepdogState *s = bs->opaque;
1382 aio_set_fd_handler(s->aio_context, s->fd, NULL, NULL, NULL);
1385 static void sd_attach_aio_context(BlockDriverState *bs,
1386 AioContext *new_context)
1388 BDRVSheepdogState *s = bs->opaque;
1390 s->aio_context = new_context;
1391 aio_set_fd_handler(new_context, s->fd, co_read_response, NULL, s);
1394 /* TODO Convert to fine grained options */
1395 static QemuOptsList runtime_opts = {
1397 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1401 .type = QEMU_OPT_STRING,
1402 .help = "URL to the sheepdog image",
1404 { /* end of list */ }
1408 static int sd_open(BlockDriverState *bs, QDict *options, int flags,
1413 BDRVSheepdogState *s = bs->opaque;
1414 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1418 Error *local_err = NULL;
1419 const char *filename;
1422 s->aio_context = bdrv_get_aio_context(bs);
1424 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1425 qemu_opts_absorb_qdict(opts, options, &local_err);
1427 error_propagate(errp, local_err);
1432 filename = qemu_opt_get(opts, "filename");
1434 QLIST_INIT(&s->inflight_aio_head);
1435 QLIST_INIT(&s->pending_aio_head);
1436 QLIST_INIT(&s->failed_aio_head);
1439 memset(vdi, 0, sizeof(vdi));
1440 memset(tag, 0, sizeof(tag));
1442 if (strstr(filename, "://")) {
1443 ret = sd_parse_uri(s, filename, vdi, &snapid, tag);
1445 ret = parse_vdiname(s, filename, vdi, &snapid, tag);
1448 error_setg(errp, "Can't parse filename");
1451 s->fd = get_sheep_fd(s, errp);
1457 ret = find_vdi_name(s, vdi, snapid, tag, &vid, true, errp);
1463 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1464 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1466 s->cache_flags = SD_FLAG_CMD_CACHE;
1467 if (flags & BDRV_O_NOCACHE) {
1468 s->cache_flags = SD_FLAG_CMD_DIRECT;
1470 s->discard_supported = true;
1472 if (snapid || tag[0] != '\0') {
1473 DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid);
1474 s->is_snapshot = true;
1477 fd = connect_to_sdog(s, errp);
1483 buf = g_malloc(SD_INODE_SIZE);
1484 ret = read_object(fd, s->aio_context, buf, vid_to_vdi_oid(vid),
1485 0, SD_INODE_SIZE, 0, s->cache_flags);
1490 error_setg(errp, "Can't read snapshot inode");
1494 memcpy(&s->inode, buf, sizeof(s->inode));
1495 s->min_dirty_data_idx = UINT32_MAX;
1496 s->max_dirty_data_idx = 0;
1498 bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE;
1499 pstrcpy(s->name, sizeof(s->name), vdi);
1500 qemu_co_mutex_init(&s->lock);
1501 qemu_opts_del(opts);
1505 aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, NULL, NULL, NULL);
1509 qemu_opts_del(opts);
1514 static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot,
1518 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1520 unsigned int wlen, rlen = 0;
1521 char buf[SD_MAX_VDI_LEN];
1523 fd = connect_to_sdog(s, errp);
1528 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1529 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1531 memset(buf, 0, sizeof(buf));
1532 pstrcpy(buf, sizeof(buf), s->name);
1534 memset(&hdr, 0, sizeof(hdr));
1535 hdr.opcode = SD_OP_NEW_VDI;
1536 hdr.base_vdi_id = s->inode.vdi_id;
1538 wlen = SD_MAX_VDI_LEN;
1540 hdr.flags = SD_FLAG_CMD_WRITE;
1541 hdr.snapid = snapshot;
1543 hdr.data_length = wlen;
1544 hdr.vdi_size = s->inode.vdi_size;
1545 hdr.copy_policy = s->inode.copy_policy;
1546 hdr.copies = s->inode.nr_copies;
1548 ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1553 error_setg_errno(errp, -ret, "create failed");
1557 if (rsp->result != SD_RES_SUCCESS) {
1558 error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name);
1563 *vdi_id = rsp->vdi_id;
1569 static int sd_prealloc(const char *filename, Error **errp)
1571 BlockDriverState *bs = NULL;
1572 uint32_t idx, max_idx;
1574 void *buf = g_malloc0(SD_DATA_OBJ_SIZE);
1577 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1580 goto out_with_err_set;
1583 vdi_size = bdrv_getlength(bs);
1588 max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE);
1590 for (idx = 0; idx < max_idx; idx++) {
1592 * The created image can be a cloned image, so we need to read
1593 * a data from the source image.
1595 ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1599 ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1607 error_setg_errno(errp, -ret, "Can't pre-allocate");
1619 * Sheepdog support two kinds of redundancy, full replication and erasure
1622 * # create a fully replicated vdi with x copies
1623 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1625 * # create a erasure coded vdi with x data strips and y parity strips
1626 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1628 static int parse_redundancy(BDRVSheepdogState *s, const char *opt)
1630 struct SheepdogInode *inode = &s->inode;
1631 const char *n1, *n2;
1635 pstrcpy(p, sizeof(p), opt);
1636 n1 = strtok(p, ":");
1637 n2 = strtok(NULL, ":");
1643 copy = strtol(n1, NULL, 10);
1644 if (copy > SD_MAX_COPIES || copy < 1) {
1648 inode->copy_policy = 0;
1649 inode->nr_copies = copy;
1653 if (copy != 2 && copy != 4 && copy != 8 && copy != 16) {
1657 parity = strtol(n2, NULL, 10);
1658 if (parity >= SD_EC_MAX_STRIP || parity < 1) {
1663 * 4 bits for parity and 4 bits for data.
1664 * We have to compress upper data bits because it can't represent 16
1666 inode->copy_policy = ((copy / 2) << 4) + parity;
1667 inode->nr_copies = copy + parity;
1672 static int sd_create(const char *filename, QemuOpts *opts,
1677 char *backing_file = NULL;
1679 BDRVSheepdogState *s;
1680 char tag[SD_MAX_VDI_TAG_LEN];
1682 bool prealloc = false;
1684 s = g_new0(BDRVSheepdogState, 1);
1686 memset(tag, 0, sizeof(tag));
1687 if (strstr(filename, "://")) {
1688 ret = sd_parse_uri(s, filename, s->name, &snapid, tag);
1690 ret = parse_vdiname(s, filename, s->name, &snapid, tag);
1693 error_setg(errp, "Can't parse filename");
1697 s->inode.vdi_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1699 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1700 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1701 if (!buf || !strcmp(buf, "off")) {
1703 } else if (!strcmp(buf, "full")) {
1706 error_setg(errp, "Invalid preallocation mode: '%s'", buf);
1712 buf = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY);
1714 ret = parse_redundancy(s, buf);
1716 error_setg(errp, "Invalid redundancy mode: '%s'", buf);
1721 if (s->inode.vdi_size > SD_MAX_VDI_SIZE) {
1722 error_setg(errp, "too big image size");
1728 BlockDriverState *bs;
1729 BDRVSheepdogState *base;
1732 /* Currently, only Sheepdog backing image is supported. */
1733 drv = bdrv_find_protocol(backing_file, true);
1734 if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
1735 error_setg(errp, "backing_file must be a sheepdog image");
1741 ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_PROTOCOL, NULL,
1749 if (!is_snapshot(&base->inode)) {
1750 error_setg(errp, "cannot clone from a non snapshot vdi");
1755 s->inode.vdi_id = base->inode.vdi_id;
1759 s->aio_context = qemu_get_aio_context();
1760 ret = do_sd_create(s, &vid, 0, errp);
1766 ret = sd_prealloc(filename, errp);
1769 g_free(backing_file);
1775 static void sd_close(BlockDriverState *bs)
1777 Error *local_err = NULL;
1778 BDRVSheepdogState *s = bs->opaque;
1780 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1781 unsigned int wlen, rlen = 0;
1784 DPRINTF("%s\n", s->name);
1786 fd = connect_to_sdog(s, &local_err);
1788 error_report("%s", error_get_pretty(local_err));;
1789 error_free(local_err);
1793 memset(&hdr, 0, sizeof(hdr));
1795 hdr.opcode = SD_OP_RELEASE_VDI;
1796 hdr.type = LOCK_TYPE_NORMAL;
1797 hdr.base_vdi_id = s->inode.vdi_id;
1798 wlen = strlen(s->name) + 1;
1799 hdr.data_length = wlen;
1800 hdr.flags = SD_FLAG_CMD_WRITE;
1802 ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr,
1803 s->name, &wlen, &rlen);
1807 if (!ret && rsp->result != SD_RES_SUCCESS &&
1808 rsp->result != SD_RES_VDI_NOT_LOCKED) {
1809 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1812 aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, NULL, NULL, NULL);
1814 g_free(s->host_spec);
1817 static int64_t sd_getlength(BlockDriverState *bs)
1819 BDRVSheepdogState *s = bs->opaque;
1821 return s->inode.vdi_size;
1824 static int sd_truncate(BlockDriverState *bs, int64_t offset)
1826 Error *local_err = NULL;
1827 BDRVSheepdogState *s = bs->opaque;
1829 unsigned int datalen;
1831 if (offset < s->inode.vdi_size) {
1832 error_report("shrinking is not supported");
1834 } else if (offset > SD_MAX_VDI_SIZE) {
1835 error_report("too big image size");
1839 fd = connect_to_sdog(s, &local_err);
1841 error_report("%s", error_get_pretty(local_err));;
1842 error_free(local_err);
1846 /* we don't need to update entire object */
1847 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1848 s->inode.vdi_size = offset;
1849 ret = write_object(fd, s->aio_context, (char *)&s->inode,
1850 vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
1851 datalen, 0, false, s->cache_flags);
1855 error_report("failed to update an inode.");
1862 * This function is called after writing data objects. If we need to
1863 * update metadata, this sends a write request to the vdi object.
1864 * Otherwise, this switches back to sd_co_readv/writev.
1866 static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
1868 BDRVSheepdogState *s = acb->common.bs->opaque;
1871 uint32_t offset, data_len, mn, mx;
1873 mn = s->min_dirty_data_idx;
1874 mx = s->max_dirty_data_idx;
1876 /* we need to update the vdi object. */
1877 offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
1878 mn * sizeof(s->inode.data_vdi_id[0]);
1879 data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
1881 s->min_dirty_data_idx = UINT32_MAX;
1882 s->max_dirty_data_idx = 0;
1884 iov.iov_base = &s->inode;
1885 iov.iov_len = sizeof(s->inode);
1886 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
1887 data_len, offset, 0, false, 0, offset);
1888 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1889 add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
1891 acb->aio_done_func = sd_finish_aiocb;
1892 acb->aiocb_type = AIOCB_WRITE_UDATA;
1896 sd_finish_aiocb(acb);
1899 /* Delete current working VDI on the snapshot chain */
1900 static bool sd_delete(BDRVSheepdogState *s)
1902 Error *local_err = NULL;
1903 unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0;
1904 SheepdogVdiReq hdr = {
1905 .opcode = SD_OP_DEL_VDI,
1906 .base_vdi_id = s->inode.vdi_id,
1907 .data_length = wlen,
1908 .flags = SD_FLAG_CMD_WRITE,
1910 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1913 fd = connect_to_sdog(s, &local_err);
1915 error_report("%s", error_get_pretty(local_err));;
1916 error_free(local_err);
1920 ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr,
1921 s->name, &wlen, &rlen);
1926 switch (rsp->result) {
1928 error_report("%s was already deleted", s->name);
1930 case SD_RES_SUCCESS:
1933 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1941 * Create a writable VDI from a snapshot
1943 static int sd_create_branch(BDRVSheepdogState *s)
1945 Error *local_err = NULL;
1951 DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id);
1953 buf = g_malloc(SD_INODE_SIZE);
1956 * Even If deletion fails, we will just create extra snapshot based on
1957 * the working VDI which was supposed to be deleted. So no need to
1960 deleted = sd_delete(s);
1961 ret = do_sd_create(s, &vid, !deleted, &local_err);
1963 error_report("%s", error_get_pretty(local_err));;
1964 error_free(local_err);
1968 DPRINTF("%" PRIx32 " is created.\n", vid);
1970 fd = connect_to_sdog(s, &local_err);
1972 error_report("%s", error_get_pretty(local_err));;
1973 error_free(local_err);
1978 ret = read_object(fd, s->aio_context, buf, vid_to_vdi_oid(vid),
1979 s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags);
1987 memcpy(&s->inode, buf, sizeof(s->inode));
1989 s->is_snapshot = false;
1991 DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id);
2000 * Send I/O requests to the server.
2002 * This function sends requests to the server, links the requests to
2003 * the inflight_list in BDRVSheepdogState, and exits without
2004 * waiting the response. The responses are received in the
2005 * `aio_read_response' function which is called from the main loop as
2008 * Returns 1 when we need to wait a response, 0 when there is no sent
2009 * request and -errno in error cases.
2011 static int coroutine_fn sd_co_rw_vector(void *p)
2013 SheepdogAIOCB *acb = p;
2015 unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE;
2016 unsigned long idx = acb->sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE;
2018 uint64_t offset = (acb->sector_num * BDRV_SECTOR_SIZE) % SD_DATA_OBJ_SIZE;
2019 BDRVSheepdogState *s = acb->common.bs->opaque;
2020 SheepdogInode *inode = &s->inode;
2023 if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
2025 * In the case we open the snapshot VDI, Sheepdog creates the
2026 * writable VDI when we do a write operation first.
2028 ret = sd_create_branch(s);
2036 * Make sure we don't free the aiocb before we are done with all requests.
2037 * This additional reference is dropped at the end of this function.
2041 while (done != total) {
2043 uint64_t old_oid = 0;
2044 bool create = false;
2046 oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
2048 len = MIN(total - done, SD_DATA_OBJ_SIZE - offset);
2050 switch (acb->aiocb_type) {
2051 case AIOCB_READ_UDATA:
2052 if (!inode->data_vdi_id[idx]) {
2053 qemu_iovec_memset(acb->qiov, done, 0, len);
2057 case AIOCB_WRITE_UDATA:
2058 if (!inode->data_vdi_id[idx]) {
2060 } else if (!is_data_obj_writable(inode, idx)) {
2064 flags = SD_FLAG_CMD_COW;
2067 case AIOCB_DISCARD_OBJ:
2069 * We discard the object only when the whole object is
2070 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2072 if (len != SD_DATA_OBJ_SIZE || inode->data_vdi_id[idx] == 0) {
2081 DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n",
2083 vid_to_data_oid(inode->data_vdi_id[idx], idx), idx);
2084 oid = vid_to_data_oid(inode->vdi_id, idx);
2085 DPRINTF("new oid %" PRIx64 "\n", oid);
2088 aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create,
2090 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
2093 if (check_simultaneous_create(s, aio_req)) {
2098 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
2106 if (!--acb->nr_pending) {
2112 static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
2113 int nb_sectors, QEMUIOVector *qiov)
2117 int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE;
2118 BDRVSheepdogState *s = bs->opaque;
2120 if (bs->growable && offset > s->inode.vdi_size) {
2121 ret = sd_truncate(bs, offset);
2127 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors);
2128 acb->aio_done_func = sd_write_done;
2129 acb->aiocb_type = AIOCB_WRITE_UDATA;
2131 ret = sd_co_rw_vector(acb);
2133 qemu_aio_unref(acb);
2137 qemu_coroutine_yield();
2142 static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
2143 int nb_sectors, QEMUIOVector *qiov)
2148 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors);
2149 acb->aiocb_type = AIOCB_READ_UDATA;
2150 acb->aio_done_func = sd_finish_aiocb;
2152 ret = sd_co_rw_vector(acb);
2154 qemu_aio_unref(acb);
2158 qemu_coroutine_yield();
2163 static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
2165 BDRVSheepdogState *s = bs->opaque;
2169 if (s->cache_flags != SD_FLAG_CMD_CACHE) {
2173 acb = sd_aio_setup(bs, NULL, 0, 0);
2174 acb->aiocb_type = AIOCB_FLUSH_CACHE;
2175 acb->aio_done_func = sd_finish_aiocb;
2177 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
2178 0, 0, 0, false, 0, 0);
2179 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
2180 add_aio_request(s, aio_req, NULL, 0, acb->aiocb_type);
2182 qemu_coroutine_yield();
2186 static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
2188 Error *local_err = NULL;
2189 BDRVSheepdogState *s = bs->opaque;
2192 SheepdogInode *inode;
2193 unsigned int datalen;
2195 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " "
2196 "is_snapshot %d\n", sn_info->name, sn_info->id_str,
2197 s->name, sn_info->vm_state_size, s->is_snapshot);
2199 if (s->is_snapshot) {
2200 error_report("You can't create a snapshot of a snapshot VDI, "
2201 "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
2206 DPRINTF("%s %s\n", sn_info->name, sn_info->id_str);
2208 s->inode.vm_state_size = sn_info->vm_state_size;
2209 s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
2210 /* It appears that inode.tag does not require a NUL terminator,
2211 * which means this use of strncpy is ok.
2213 strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
2214 /* we don't need to update entire object */
2215 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
2216 inode = g_malloc(datalen);
2218 /* refresh inode. */
2219 fd = connect_to_sdog(s, &local_err);
2221 error_report("%s", error_get_pretty(local_err));;
2222 error_free(local_err);
2227 ret = write_object(fd, s->aio_context, (char *)&s->inode,
2228 vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
2229 datalen, 0, false, s->cache_flags);
2231 error_report("failed to write snapshot's inode.");
2235 ret = do_sd_create(s, &new_vid, 1, &local_err);
2237 error_report("%s", error_get_pretty(local_err));;
2238 error_free(local_err);
2239 error_report("failed to create inode for snapshot. %s",
2244 ret = read_object(fd, s->aio_context, (char *)inode,
2245 vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0,
2249 error_report("failed to read new inode info. %s", strerror(errno));
2253 memcpy(&s->inode, inode, datalen);
2254 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2255 s->inode.name, s->inode.snap_id, s->inode.vdi_id);
2264 * We implement rollback(loadvm) operation to the specified snapshot by
2265 * 1) switch to the snapshot
2266 * 2) rely on sd_create_branch to delete working VDI and
2267 * 3) create a new working VDI based on the specified snapshot
2269 static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
2271 BDRVSheepdogState *s = bs->opaque;
2272 BDRVSheepdogState *old_s;
2273 char tag[SD_MAX_VDI_TAG_LEN];
2274 uint32_t snapid = 0;
2277 old_s = g_new(BDRVSheepdogState, 1);
2279 memcpy(old_s, s, sizeof(BDRVSheepdogState));
2281 snapid = strtoul(snapshot_id, NULL, 10);
2285 pstrcpy(tag, sizeof(tag), snapshot_id);
2288 ret = reload_inode(s, snapid, tag);
2293 ret = sd_create_branch(s);
2302 /* recover bdrv_sd_state */
2303 memcpy(s, old_s, sizeof(BDRVSheepdogState));
2306 error_report("failed to open. recover old bdrv_sd_state.");
2311 static int sd_snapshot_delete(BlockDriverState *bs,
2312 const char *snapshot_id,
2316 /* FIXME: Delete specified snapshot id. */
2320 static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
2322 Error *local_err = NULL;
2323 BDRVSheepdogState *s = bs->opaque;
2325 int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
2326 QEMUSnapshotInfo *sn_tab = NULL;
2327 unsigned wlen, rlen;
2329 static SheepdogInode inode;
2330 unsigned long *vdi_inuse;
2331 unsigned int start_nr;
2335 vdi_inuse = g_malloc(max);
2337 fd = connect_to_sdog(s, &local_err);
2339 error_report("%s", error_get_pretty(local_err));;
2340 error_free(local_err);
2348 memset(&req, 0, sizeof(req));
2350 req.opcode = SD_OP_READ_VDIS;
2351 req.data_length = max;
2353 ret = do_req(fd, s->aio_context, (SheepdogReq *)&req,
2354 vdi_inuse, &wlen, &rlen);
2361 sn_tab = g_new0(QEMUSnapshotInfo, nr);
2363 /* calculate a vdi id with hash function */
2364 hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
2365 start_nr = hval & (SD_NR_VDIS - 1);
2367 fd = connect_to_sdog(s, &local_err);
2369 error_report("%s", error_get_pretty(local_err));;
2370 error_free(local_err);
2375 for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
2376 if (!test_bit(vid, vdi_inuse)) {
2380 /* we don't need to read entire object */
2381 ret = read_object(fd, s->aio_context, (char *)&inode,
2382 vid_to_vdi_oid(vid),
2383 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0,
2390 if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
2391 sn_tab[found].date_sec = inode.snap_ctime >> 32;
2392 sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
2393 sn_tab[found].vm_state_size = inode.vm_state_size;
2394 sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec;
2396 snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str),
2397 "%" PRIu32, inode.snap_id);
2398 pstrcpy(sn_tab[found].name,
2399 MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)),
2418 static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
2419 int64_t pos, int size, int load)
2421 Error *local_err = NULL;
2423 int fd, ret = 0, remaining = size;
2424 unsigned int data_len;
2425 uint64_t vmstate_oid;
2428 uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id;
2430 fd = connect_to_sdog(s, &local_err);
2432 error_report("%s", error_get_pretty(local_err));;
2433 error_free(local_err);
2438 vdi_index = pos / SD_DATA_OBJ_SIZE;
2439 offset = pos % SD_DATA_OBJ_SIZE;
2441 data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset);
2443 vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index);
2445 create = (offset == 0);
2447 ret = read_object(fd, s->aio_context, (char *)data, vmstate_oid,
2448 s->inode.nr_copies, data_len, offset,
2451 ret = write_object(fd, s->aio_context, (char *)data, vmstate_oid,
2452 s->inode.nr_copies, data_len, offset, create,
2457 error_report("failed to save vmstate %s", strerror(errno));
2463 remaining -= data_len;
2471 static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2474 BDRVSheepdogState *s = bs->opaque;
2478 buf = qemu_blockalign(bs, qiov->size);
2479 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
2480 ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0);
2486 static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data,
2487 int64_t pos, int size)
2489 BDRVSheepdogState *s = bs->opaque;
2491 return do_load_save_vmstate(s, data, pos, size, 1);
2495 static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num,
2500 BDRVSheepdogState *s = bs->opaque;
2503 if (!s->discard_supported) {
2507 acb = sd_aio_setup(bs, &dummy, sector_num, nb_sectors);
2508 acb->aiocb_type = AIOCB_DISCARD_OBJ;
2509 acb->aio_done_func = sd_finish_aiocb;
2511 ret = sd_co_rw_vector(acb);
2513 qemu_aio_unref(acb);
2517 qemu_coroutine_yield();
2522 static coroutine_fn int64_t
2523 sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2526 BDRVSheepdogState *s = bs->opaque;
2527 SheepdogInode *inode = &s->inode;
2528 uint64_t offset = sector_num * BDRV_SECTOR_SIZE;
2529 unsigned long start = offset / SD_DATA_OBJ_SIZE,
2530 end = DIV_ROUND_UP((sector_num + nb_sectors) *
2531 BDRV_SECTOR_SIZE, SD_DATA_OBJ_SIZE);
2533 int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
2535 for (idx = start; idx < end; idx++) {
2536 if (inode->data_vdi_id[idx] == 0) {
2541 /* Get the longest length of unallocated sectors */
2543 for (idx = start + 1; idx < end; idx++) {
2544 if (inode->data_vdi_id[idx] != 0) {
2550 *pnum = (idx - start) * SD_DATA_OBJ_SIZE / BDRV_SECTOR_SIZE;
2551 if (*pnum > nb_sectors) {
2557 static int64_t sd_get_allocated_file_size(BlockDriverState *bs)
2559 BDRVSheepdogState *s = bs->opaque;
2560 SheepdogInode *inode = &s->inode;
2561 unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, SD_DATA_OBJ_SIZE);
2564 for (i = 0; i < last; i++) {
2565 if (inode->data_vdi_id[i] == 0) {
2568 size += SD_DATA_OBJ_SIZE;
2573 static QemuOptsList sd_create_opts = {
2574 .name = "sheepdog-create-opts",
2575 .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head),
2578 .name = BLOCK_OPT_SIZE,
2579 .type = QEMU_OPT_SIZE,
2580 .help = "Virtual disk size"
2583 .name = BLOCK_OPT_BACKING_FILE,
2584 .type = QEMU_OPT_STRING,
2585 .help = "File name of a base image"
2588 .name = BLOCK_OPT_PREALLOC,
2589 .type = QEMU_OPT_STRING,
2590 .help = "Preallocation mode (allowed values: off, full)"
2593 .name = BLOCK_OPT_REDUNDANCY,
2594 .type = QEMU_OPT_STRING,
2595 .help = "Redundancy of the image"
2597 { /* end of list */ }
2601 static BlockDriver bdrv_sheepdog = {
2602 .format_name = "sheepdog",
2603 .protocol_name = "sheepdog",
2604 .instance_size = sizeof(BDRVSheepdogState),
2605 .bdrv_needs_filename = true,
2606 .bdrv_file_open = sd_open,
2607 .bdrv_close = sd_close,
2608 .bdrv_create = sd_create,
2609 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2610 .bdrv_getlength = sd_getlength,
2611 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2612 .bdrv_truncate = sd_truncate,
2614 .bdrv_co_readv = sd_co_readv,
2615 .bdrv_co_writev = sd_co_writev,
2616 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2617 .bdrv_co_discard = sd_co_discard,
2618 .bdrv_co_get_block_status = sd_co_get_block_status,
2620 .bdrv_snapshot_create = sd_snapshot_create,
2621 .bdrv_snapshot_goto = sd_snapshot_goto,
2622 .bdrv_snapshot_delete = sd_snapshot_delete,
2623 .bdrv_snapshot_list = sd_snapshot_list,
2625 .bdrv_save_vmstate = sd_save_vmstate,
2626 .bdrv_load_vmstate = sd_load_vmstate,
2628 .bdrv_detach_aio_context = sd_detach_aio_context,
2629 .bdrv_attach_aio_context = sd_attach_aio_context,
2631 .create_opts = &sd_create_opts,
2634 static BlockDriver bdrv_sheepdog_tcp = {
2635 .format_name = "sheepdog",
2636 .protocol_name = "sheepdog+tcp",
2637 .instance_size = sizeof(BDRVSheepdogState),
2638 .bdrv_needs_filename = true,
2639 .bdrv_file_open = sd_open,
2640 .bdrv_close = sd_close,
2641 .bdrv_create = sd_create,
2642 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2643 .bdrv_getlength = sd_getlength,
2644 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2645 .bdrv_truncate = sd_truncate,
2647 .bdrv_co_readv = sd_co_readv,
2648 .bdrv_co_writev = sd_co_writev,
2649 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2650 .bdrv_co_discard = sd_co_discard,
2651 .bdrv_co_get_block_status = sd_co_get_block_status,
2653 .bdrv_snapshot_create = sd_snapshot_create,
2654 .bdrv_snapshot_goto = sd_snapshot_goto,
2655 .bdrv_snapshot_delete = sd_snapshot_delete,
2656 .bdrv_snapshot_list = sd_snapshot_list,
2658 .bdrv_save_vmstate = sd_save_vmstate,
2659 .bdrv_load_vmstate = sd_load_vmstate,
2661 .bdrv_detach_aio_context = sd_detach_aio_context,
2662 .bdrv_attach_aio_context = sd_attach_aio_context,
2664 .create_opts = &sd_create_opts,
2667 static BlockDriver bdrv_sheepdog_unix = {
2668 .format_name = "sheepdog",
2669 .protocol_name = "sheepdog+unix",
2670 .instance_size = sizeof(BDRVSheepdogState),
2671 .bdrv_needs_filename = true,
2672 .bdrv_file_open = sd_open,
2673 .bdrv_close = sd_close,
2674 .bdrv_create = sd_create,
2675 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2676 .bdrv_getlength = sd_getlength,
2677 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2678 .bdrv_truncate = sd_truncate,
2680 .bdrv_co_readv = sd_co_readv,
2681 .bdrv_co_writev = sd_co_writev,
2682 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2683 .bdrv_co_discard = sd_co_discard,
2684 .bdrv_co_get_block_status = sd_co_get_block_status,
2686 .bdrv_snapshot_create = sd_snapshot_create,
2687 .bdrv_snapshot_goto = sd_snapshot_goto,
2688 .bdrv_snapshot_delete = sd_snapshot_delete,
2689 .bdrv_snapshot_list = sd_snapshot_list,
2691 .bdrv_save_vmstate = sd_save_vmstate,
2692 .bdrv_load_vmstate = sd_load_vmstate,
2694 .bdrv_detach_aio_context = sd_detach_aio_context,
2695 .bdrv_attach_aio_context = sd_attach_aio_context,
2697 .create_opts = &sd_create_opts,
2700 static void bdrv_sheepdog_init(void)
2702 bdrv_register(&bdrv_sheepdog);
2703 bdrv_register(&bdrv_sheepdog_tcp);
2704 bdrv_register(&bdrv_sheepdog_unix);
2706 block_init(bdrv_sheepdog_init);