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"
16 #include "qemu-error.h"
17 #include "qemu_socket.h"
18 #include "block_int.h"
21 #define SD_PROTO_VER 0x01
23 #define SD_DEFAULT_ADDR "localhost"
24 #define SD_DEFAULT_PORT "7000"
26 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
27 #define SD_OP_READ_OBJ 0x02
28 #define SD_OP_WRITE_OBJ 0x03
30 #define SD_OP_NEW_VDI 0x11
31 #define SD_OP_LOCK_VDI 0x12
32 #define SD_OP_RELEASE_VDI 0x13
33 #define SD_OP_GET_VDI_INFO 0x14
34 #define SD_OP_READ_VDIS 0x15
35 #define SD_OP_FLUSH_VDI 0x16
37 #define SD_FLAG_CMD_WRITE 0x01
38 #define SD_FLAG_CMD_COW 0x02
39 #define SD_FLAG_CMD_CACHE 0x04
41 #define SD_RES_SUCCESS 0x00 /* Success */
42 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
43 #define SD_RES_NO_OBJ 0x02 /* No object found */
44 #define SD_RES_EIO 0x03 /* I/O error */
45 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
46 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
47 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
48 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
49 #define SD_RES_NO_VDI 0x08 /* No vdi found */
50 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
51 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
52 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
53 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
54 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
55 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
56 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
57 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
58 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
59 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
60 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
61 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
62 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
63 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
64 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
65 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
70 * 0 - 19 (20 bits): data object space
71 * 20 - 31 (12 bits): reserved data object space
72 * 32 - 55 (24 bits): vdi object space
73 * 56 - 59 ( 4 bits): reserved vdi object space
74 * 60 - 63 ( 4 bits): object type identifier space
77 #define VDI_SPACE_SHIFT 32
78 #define VDI_BIT (UINT64_C(1) << 63)
79 #define VMSTATE_BIT (UINT64_C(1) << 62)
80 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
81 #define MAX_CHILDREN 1024
82 #define SD_MAX_VDI_LEN 256
83 #define SD_MAX_VDI_TAG_LEN 256
84 #define SD_NR_VDIS (1U << 24)
85 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
86 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
87 #define SECTOR_SIZE 512
89 #define SD_INODE_SIZE (sizeof(SheepdogInode))
90 #define CURRENT_VDI_ID 0
92 typedef struct SheepdogReq {
99 uint32_t opcode_specific[8];
102 typedef struct SheepdogRsp {
108 uint32_t data_length;
110 uint32_t opcode_specific[7];
113 typedef struct SheepdogObjReq {
119 uint32_t data_length;
127 typedef struct SheepdogObjRsp {
133 uint32_t data_length;
139 typedef struct SheepdogVdiReq {
145 uint32_t data_length;
147 uint32_t base_vdi_id;
153 typedef struct SheepdogVdiRsp {
159 uint32_t data_length;
166 typedef struct SheepdogInode {
167 char name[SD_MAX_VDI_LEN];
168 char tag[SD_MAX_VDI_TAG_LEN];
171 uint64_t vm_clock_nsec;
173 uint64_t vm_state_size;
174 uint16_t copy_policy;
176 uint8_t block_size_shift;
179 uint32_t parent_vdi_id;
180 uint32_t child_vdi_id[MAX_CHILDREN];
181 uint32_t data_vdi_id[MAX_DATA_OBJS];
185 * 64 bit FNV-1a non-zero initial basis
187 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
190 * 64 bit Fowler/Noll/Vo FNV-1a hash code
192 static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
194 unsigned char *bp = buf;
195 unsigned char *be = bp + len;
197 hval ^= (uint64_t) *bp++;
198 hval += (hval << 1) + (hval << 4) + (hval << 5) +
199 (hval << 7) + (hval << 8) + (hval << 40);
204 static inline int is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
206 return inode->vdi_id == inode->data_vdi_id[idx];
209 static inline int is_data_obj(uint64_t oid)
211 return !(VDI_BIT & oid);
214 static inline uint64_t data_oid_to_idx(uint64_t oid)
216 return oid & (MAX_DATA_OBJS - 1);
219 static inline uint64_t vid_to_vdi_oid(uint32_t vid)
221 return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
224 static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx)
226 return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
229 static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx)
231 return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
234 static inline int is_snapshot(struct SheepdogInode *inode)
236 return !!inode->snap_ctime;
241 #define dprintf(fmt, args...) \
243 fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \
246 #define dprintf(fmt, args...)
249 typedef struct SheepdogAIOCB SheepdogAIOCB;
251 typedef struct AIOReq {
252 SheepdogAIOCB *aiocb;
253 unsigned int iov_offset;
258 unsigned int data_len;
262 QLIST_ENTRY(AIOReq) aio_siblings;
270 struct SheepdogAIOCB {
271 BlockDriverAIOCB common;
279 enum AIOCBState aiocb_type;
281 Coroutine *coroutine;
282 void (*aio_done_func)(SheepdogAIOCB *);
288 typedef struct BDRVSheepdogState {
291 uint32_t min_dirty_data_idx;
292 uint32_t max_dirty_data_idx;
294 char name[SD_MAX_VDI_LEN];
296 uint8_t cache_enabled;
307 uint32_t aioreq_seq_num;
308 QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head;
309 QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head;
312 static const char * sd_strerror(int err)
316 static const struct {
320 {SD_RES_SUCCESS, "Success"},
321 {SD_RES_UNKNOWN, "Unknown error"},
322 {SD_RES_NO_OBJ, "No object found"},
323 {SD_RES_EIO, "I/O error"},
324 {SD_RES_VDI_EXIST, "VDI exists already"},
325 {SD_RES_INVALID_PARMS, "Invalid parameters"},
326 {SD_RES_SYSTEM_ERROR, "System error"},
327 {SD_RES_VDI_LOCKED, "VDI is already locked"},
328 {SD_RES_NO_VDI, "No vdi found"},
329 {SD_RES_NO_BASE_VDI, "No base VDI found"},
330 {SD_RES_VDI_READ, "Failed read the requested VDI"},
331 {SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
332 {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
333 {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
334 {SD_RES_NO_TAG, "Failed to find the requested tag"},
335 {SD_RES_STARTUP, "The system is still booting"},
336 {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
337 {SD_RES_SHUTDOWN, "The system is shutting down"},
338 {SD_RES_NO_MEM, "Out of memory on the server"},
339 {SD_RES_FULL_VDI, "We already have the maximum vdis"},
340 {SD_RES_VER_MISMATCH, "Protocol version mismatch"},
341 {SD_RES_NO_SPACE, "Server has no space for new objects"},
342 {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
343 {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
344 {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
347 for (i = 0; i < ARRAY_SIZE(errors); ++i) {
348 if (errors[i].err == err) {
349 return errors[i].desc;
353 return "Invalid error code";
357 * Sheepdog I/O handling:
359 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
360 * link the requests to the inflight_list in the
361 * BDRVSheepdogState. The function exits without waiting for
362 * receiving the response.
364 * 2. We receive the response in aio_read_response, the fd handler to
365 * the sheepdog connection. If metadata update is needed, we send
366 * the write request to the vdi object in sd_write_done, the write
367 * completion function. We switch back to sd_co_readv/writev after
368 * all the requests belonging to the AIOCB are finished.
371 static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
372 uint64_t oid, unsigned int data_len,
373 uint64_t offset, uint8_t flags,
374 uint64_t base_oid, unsigned int iov_offset)
378 aio_req = g_malloc(sizeof(*aio_req));
379 aio_req->aiocb = acb;
380 aio_req->iov_offset = iov_offset;
382 aio_req->base_oid = base_oid;
383 aio_req->offset = offset;
384 aio_req->data_len = data_len;
385 aio_req->flags = flags;
386 aio_req->id = s->aioreq_seq_num++;
392 static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req)
394 SheepdogAIOCB *acb = aio_req->aiocb;
396 QLIST_REMOVE(aio_req, aio_siblings);
402 static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb)
404 if (!acb->canceled) {
405 qemu_coroutine_enter(acb->coroutine, NULL);
407 qemu_aio_release(acb);
410 static void sd_aio_cancel(BlockDriverAIOCB *blockacb)
412 SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb;
415 * Sheepdog cannot cancel the requests which are already sent to
416 * the servers, so we just complete the request with -EIO here.
419 qemu_coroutine_enter(acb->coroutine, NULL);
423 static AIOPool sd_aio_pool = {
424 .aiocb_size = sizeof(SheepdogAIOCB),
425 .cancel = sd_aio_cancel,
428 static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov,
429 int64_t sector_num, int nb_sectors,
430 BlockDriverCompletionFunc *cb, void *opaque)
434 acb = qemu_aio_get(&sd_aio_pool, bs, cb, opaque);
438 acb->sector_num = sector_num;
439 acb->nb_sectors = nb_sectors;
441 acb->aio_done_func = NULL;
443 acb->coroutine = qemu_coroutine_self();
449 static int connect_to_sdog(const char *addr, const char *port)
451 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
453 struct addrinfo hints, *res, *res0;
456 addr = SD_DEFAULT_ADDR;
457 port = SD_DEFAULT_PORT;
460 memset(&hints, 0, sizeof(hints));
461 hints.ai_socktype = SOCK_STREAM;
463 ret = getaddrinfo(addr, port, &hints, &res0);
465 error_report("unable to get address info %s, %s",
466 addr, strerror(errno));
470 for (res = res0; res; res = res->ai_next) {
471 ret = getnameinfo(res->ai_addr, res->ai_addrlen, hbuf, sizeof(hbuf),
472 sbuf, sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
477 fd = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
483 ret = connect(fd, res->ai_addr, res->ai_addrlen);
485 if (errno == EINTR) {
491 dprintf("connected to %s:%s\n", addr, port);
495 error_report("failed connect to %s:%s", addr, port);
501 static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
506 ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
507 if (ret < sizeof(*hdr)) {
508 error_report("failed to send a req, %s", strerror(errno));
512 ret = qemu_co_send(sockfd, data, *wlen);
514 error_report("failed to send a req, %s", strerror(errno));
520 static void restart_co_req(void *opaque)
522 Coroutine *co = opaque;
524 qemu_coroutine_enter(co, NULL);
527 typedef struct SheepdogReqCo {
537 static coroutine_fn void do_co_req(void *opaque)
541 SheepdogReqCo *srco = opaque;
542 int sockfd = srco->sockfd;
543 SheepdogReq *hdr = srco->hdr;
544 void *data = srco->data;
545 unsigned int *wlen = srco->wlen;
546 unsigned int *rlen = srco->rlen;
548 co = qemu_coroutine_self();
549 qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, NULL, co);
551 socket_set_block(sockfd);
552 ret = send_co_req(sockfd, hdr, data, wlen);
557 qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, NULL, co);
559 ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
560 if (ret < sizeof(*hdr)) {
561 error_report("failed to get a rsp, %s", strerror(errno));
566 if (*rlen > hdr->data_length) {
567 *rlen = hdr->data_length;
571 ret = qemu_co_recv(sockfd, data, *rlen);
573 error_report("failed to get the data, %s", strerror(errno));
580 qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL);
581 socket_set_nonblock(sockfd);
584 srco->finished = true;
587 static int do_req(int sockfd, SheepdogReq *hdr, void *data,
588 unsigned int *wlen, unsigned int *rlen)
591 SheepdogReqCo srco = {
601 if (qemu_in_coroutine()) {
604 co = qemu_coroutine_create(do_co_req);
605 qemu_coroutine_enter(co, &srco);
606 while (!srco.finished) {
614 static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
615 struct iovec *iov, int niov, int create,
616 enum AIOCBState aiocb_type);
619 static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid)
623 QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) {
624 if (aio_req->oid == oid) {
633 * This function searchs pending requests to the object `oid', and
636 static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid)
642 while ((aio_req = find_pending_req(s, oid)) != NULL) {
643 acb = aio_req->aiocb;
644 /* move aio_req from pending list to inflight one */
645 QLIST_REMOVE(aio_req, aio_siblings);
646 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
647 ret = add_aio_request(s, aio_req, acb->qiov->iov,
648 acb->qiov->niov, 0, acb->aiocb_type);
650 error_report("add_aio_request is failed");
651 free_aio_req(s, aio_req);
652 if (!acb->nr_pending) {
653 sd_finish_aiocb(acb);
660 * Receive responses of the I/O requests.
662 * This function is registered as a fd handler, and called from the
663 * main loop when s->fd is ready for reading responses.
665 static void coroutine_fn aio_read_response(void *opaque)
668 BDRVSheepdogState *s = opaque;
671 AIOReq *aio_req = NULL;
675 if (QLIST_EMPTY(&s->inflight_aio_head)) {
680 ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
682 error_report("failed to get the header, %s", strerror(errno));
686 /* find the right aio_req from the inflight aio list */
687 QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
688 if (aio_req->id == rsp.id) {
693 error_report("cannot find aio_req %x", rsp.id);
697 acb = aio_req->aiocb;
699 switch (acb->aiocb_type) {
700 case AIOCB_WRITE_UDATA:
701 /* this coroutine context is no longer suitable for co_recv
702 * because we may send data to update vdi objects */
704 if (!is_data_obj(aio_req->oid)) {
707 idx = data_oid_to_idx(aio_req->oid);
709 if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) {
711 * If the object is newly created one, we need to update
712 * the vdi object (metadata object). min_dirty_data_idx
713 * and max_dirty_data_idx are changed to include updated
714 * index between them.
716 s->inode.data_vdi_id[idx] = s->inode.vdi_id;
717 s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx);
718 s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx);
721 * Some requests may be blocked because simultaneous
722 * create requests are not allowed, so we search the
723 * pending requests here.
725 send_pending_req(s, vid_to_data_oid(s->inode.vdi_id, idx));
728 case AIOCB_READ_UDATA:
729 ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
730 aio_req->iov_offset, rsp.data_length);
732 error_report("failed to get the data, %s", strerror(errno));
738 if (rsp.result != SD_RES_SUCCESS) {
740 error_report("%s", sd_strerror(rsp.result));
743 free_aio_req(s, aio_req);
744 if (!acb->nr_pending) {
746 * We've finished all requests which belong to the AIOCB, so
747 * we can switch back to sd_co_readv/writev now.
749 acb->aio_done_func(acb);
755 static void co_read_response(void *opaque)
757 BDRVSheepdogState *s = opaque;
760 s->co_recv = qemu_coroutine_create(aio_read_response);
763 qemu_coroutine_enter(s->co_recv, opaque);
766 static void co_write_request(void *opaque)
768 BDRVSheepdogState *s = opaque;
770 qemu_coroutine_enter(s->co_send, NULL);
773 static int aio_flush_request(void *opaque)
775 BDRVSheepdogState *s = opaque;
777 return !QLIST_EMPTY(&s->inflight_aio_head) ||
778 !QLIST_EMPTY(&s->pending_aio_head);
781 static int set_nodelay(int fd)
786 ret = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *)&opt, sizeof(opt));
791 * Return a socket discriptor to read/write objects.
793 * We cannot use this discriptor for other operations because
794 * the block driver may be on waiting response from the server.
796 static int get_sheep_fd(BDRVSheepdogState *s)
800 fd = connect_to_sdog(s->addr, s->port);
802 error_report("%s", strerror(errno));
806 socket_set_nonblock(fd);
808 ret = set_nodelay(fd);
810 error_report("%s", strerror(errno));
815 qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s);
822 * filename must be one of the following formats:
824 * 2. [vdiname]:[snapid]
826 * 4. [hostname]:[port]:[vdiname]
827 * 5. [hostname]:[port]:[vdiname]:[snapid]
828 * 6. [hostname]:[port]:[vdiname]:[tag]
830 * You can boot from the snapshot images by specifying `snapid` or
833 * You can run VMs outside the Sheepdog cluster by specifying
834 * `hostname' and `port' (experimental).
836 static int parse_vdiname(BDRVSheepdogState *s, const char *filename,
837 char *vdi, uint32_t *snapid, char *tag)
842 p = q = g_strdup(filename);
844 /* count the number of separators */
854 /* use the first two tokens as hostname and port number. */
868 strncpy(vdi, p, SD_MAX_VDI_LEN);
870 p = strchr(vdi, ':');
873 *snapid = strtoul(p, NULL, 10);
875 strncpy(tag, p, SD_MAX_VDI_TAG_LEN);
878 *snapid = CURRENT_VDI_ID; /* search current vdi */
881 if (s->addr == NULL) {
888 static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid,
889 char *tag, uint32_t *vid, int for_snapshot)
893 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
894 unsigned int wlen, rlen = 0;
895 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
897 fd = connect_to_sdog(s->addr, s->port);
902 memset(buf, 0, sizeof(buf));
903 strncpy(buf, filename, SD_MAX_VDI_LEN);
904 strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
906 memset(&hdr, 0, sizeof(hdr));
908 hdr.opcode = SD_OP_GET_VDI_INFO;
910 hdr.opcode = SD_OP_LOCK_VDI;
912 wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
913 hdr.proto_ver = SD_PROTO_VER;
914 hdr.data_length = wlen;
916 hdr.flags = SD_FLAG_CMD_WRITE;
918 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
923 if (rsp->result != SD_RES_SUCCESS) {
924 error_report("cannot get vdi info, %s, %s %d %s",
925 sd_strerror(rsp->result), filename, snapid, tag);
926 if (rsp->result == SD_RES_NO_VDI) {
941 static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
942 struct iovec *iov, int niov, int create,
943 enum AIOCBState aiocb_type)
945 int nr_copies = s->inode.nr_copies;
949 uint64_t oid = aio_req->oid;
950 unsigned int datalen = aio_req->data_len;
951 uint64_t offset = aio_req->offset;
952 uint8_t flags = aio_req->flags;
953 uint64_t old_oid = aio_req->base_oid;
959 memset(&hdr, 0, sizeof(hdr));
961 if (aiocb_type == AIOCB_READ_UDATA) {
963 hdr.opcode = SD_OP_READ_OBJ;
967 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
968 hdr.flags = SD_FLAG_CMD_WRITE | flags;
971 hdr.opcode = SD_OP_WRITE_OBJ;
972 hdr.flags = SD_FLAG_CMD_WRITE | flags;
975 if (s->cache_enabled) {
976 hdr.flags |= SD_FLAG_CMD_CACHE;
980 hdr.cow_oid = old_oid;
981 hdr.copies = s->inode.nr_copies;
983 hdr.data_length = datalen;
986 hdr.id = aio_req->id;
988 qemu_co_mutex_lock(&s->lock);
989 s->co_send = qemu_coroutine_self();
990 qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request,
991 aio_flush_request, s);
992 socket_set_cork(s->fd, 1);
995 ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
997 qemu_co_mutex_unlock(&s->lock);
998 error_report("failed to send a req, %s", strerror(errno));
1003 ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1005 qemu_co_mutex_unlock(&s->lock);
1006 error_report("failed to send a data, %s", strerror(errno));
1011 socket_set_cork(s->fd, 0);
1012 qemu_aio_set_fd_handler(s->fd, co_read_response, NULL,
1013 aio_flush_request, s);
1014 qemu_co_mutex_unlock(&s->lock);
1019 static int read_write_object(int fd, char *buf, uint64_t oid, int copies,
1020 unsigned int datalen, uint64_t offset,
1021 int write, int create, uint8_t cache)
1024 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1025 unsigned int wlen, rlen;
1028 memset(&hdr, 0, sizeof(hdr));
1033 hdr.flags = SD_FLAG_CMD_WRITE;
1035 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1037 hdr.opcode = SD_OP_WRITE_OBJ;
1042 hdr.opcode = SD_OP_READ_OBJ;
1046 hdr.flags |= SD_FLAG_CMD_CACHE;
1050 hdr.data_length = datalen;
1051 hdr.offset = offset;
1052 hdr.copies = copies;
1054 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1056 error_report("failed to send a request to the sheep");
1060 switch (rsp->result) {
1061 case SD_RES_SUCCESS:
1064 error_report("%s", sd_strerror(rsp->result));
1069 static int read_object(int fd, char *buf, uint64_t oid, int copies,
1070 unsigned int datalen, uint64_t offset, uint8_t cache)
1072 return read_write_object(fd, buf, oid, copies, datalen, offset, 0, 0,
1076 static int write_object(int fd, char *buf, uint64_t oid, int copies,
1077 unsigned int datalen, uint64_t offset, int create,
1080 return read_write_object(fd, buf, oid, copies, datalen, offset, 1, create,
1084 static int sd_open(BlockDriverState *bs, const char *filename, int flags)
1088 BDRVSheepdogState *s = bs->opaque;
1089 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1093 strstart(filename, "sheepdog:", (const char **)&filename);
1095 QLIST_INIT(&s->inflight_aio_head);
1096 QLIST_INIT(&s->pending_aio_head);
1099 memset(vdi, 0, sizeof(vdi));
1100 memset(tag, 0, sizeof(tag));
1101 if (parse_vdiname(s, filename, vdi, &snapid, tag) < 0) {
1105 s->fd = get_sheep_fd(s);
1111 ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0);
1116 if (flags & BDRV_O_CACHE_WB) {
1117 s->cache_enabled = 1;
1118 s->flush_fd = connect_to_sdog(s->addr, s->port);
1119 if (s->flush_fd < 0) {
1120 error_report("failed to connect");
1126 if (snapid || tag[0] != '\0') {
1127 dprintf("%" PRIx32 " snapshot inode was open.\n", vid);
1131 fd = connect_to_sdog(s->addr, s->port);
1133 error_report("failed to connect");
1138 buf = g_malloc(SD_INODE_SIZE);
1139 ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0,
1148 memcpy(&s->inode, buf, sizeof(s->inode));
1149 s->min_dirty_data_idx = UINT32_MAX;
1150 s->max_dirty_data_idx = 0;
1152 bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE;
1153 strncpy(s->name, vdi, sizeof(s->name));
1154 qemu_co_mutex_init(&s->lock);
1158 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL);
1166 static int do_sd_create(char *filename, int64_t vdi_size,
1167 uint32_t base_vid, uint32_t *vdi_id, int snapshot,
1168 const char *addr, const char *port)
1171 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1173 unsigned int wlen, rlen = 0;
1174 char buf[SD_MAX_VDI_LEN];
1176 fd = connect_to_sdog(addr, port);
1181 memset(buf, 0, sizeof(buf));
1182 strncpy(buf, filename, SD_MAX_VDI_LEN);
1184 memset(&hdr, 0, sizeof(hdr));
1185 hdr.opcode = SD_OP_NEW_VDI;
1186 hdr.base_vdi_id = base_vid;
1188 wlen = SD_MAX_VDI_LEN;
1190 hdr.flags = SD_FLAG_CMD_WRITE;
1191 hdr.snapid = snapshot;
1193 hdr.data_length = wlen;
1194 hdr.vdi_size = vdi_size;
1196 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1204 if (rsp->result != SD_RES_SUCCESS) {
1205 error_report("%s, %s", sd_strerror(rsp->result), filename);
1210 *vdi_id = rsp->vdi_id;
1216 static int sd_prealloc(const char *filename)
1218 BlockDriverState *bs = NULL;
1219 uint32_t idx, max_idx;
1221 void *buf = g_malloc0(SD_DATA_OBJ_SIZE);
1224 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
1229 vdi_size = bdrv_getlength(bs);
1234 max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE);
1236 for (idx = 0; idx < max_idx; idx++) {
1238 * The created image can be a cloned image, so we need to read
1239 * a data from the source image.
1241 ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1245 ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1259 static int sd_create(const char *filename, QEMUOptionParameter *options)
1262 uint32_t vid = 0, base_vid = 0;
1263 int64_t vdi_size = 0;
1264 char *backing_file = NULL;
1265 BDRVSheepdogState *s;
1266 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1269 const char *vdiname;
1271 s = g_malloc0(sizeof(BDRVSheepdogState));
1273 strstart(filename, "sheepdog:", &vdiname);
1275 memset(vdi, 0, sizeof(vdi));
1276 memset(tag, 0, sizeof(tag));
1277 if (parse_vdiname(s, vdiname, vdi, &snapid, tag) < 0) {
1278 error_report("invalid filename");
1283 while (options && options->name) {
1284 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1285 vdi_size = options->value.n;
1286 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1287 backing_file = options->value.s;
1288 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1289 if (!options->value.s || !strcmp(options->value.s, "off")) {
1291 } else if (!strcmp(options->value.s, "full")) {
1294 error_report("Invalid preallocation mode: '%s'",
1303 if (vdi_size > SD_MAX_VDI_SIZE) {
1304 error_report("too big image size");
1310 BlockDriverState *bs;
1311 BDRVSheepdogState *s;
1314 /* Currently, only Sheepdog backing image is supported. */
1315 drv = bdrv_find_protocol(backing_file);
1316 if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
1317 error_report("backing_file must be a sheepdog image");
1322 ret = bdrv_file_open(&bs, backing_file, 0);
1329 if (!is_snapshot(&s->inode)) {
1330 error_report("cannot clone from a non snapshot vdi");
1336 base_vid = s->inode.vdi_id;
1340 ret = do_sd_create(vdi, vdi_size, base_vid, &vid, 0, s->addr, s->port);
1341 if (!prealloc || ret) {
1345 ret = sd_prealloc(filename);
1351 static void sd_close(BlockDriverState *bs)
1353 BDRVSheepdogState *s = bs->opaque;
1355 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1356 unsigned int wlen, rlen = 0;
1359 dprintf("%s\n", s->name);
1361 fd = connect_to_sdog(s->addr, s->port);
1366 memset(&hdr, 0, sizeof(hdr));
1368 hdr.opcode = SD_OP_RELEASE_VDI;
1369 wlen = strlen(s->name) + 1;
1370 hdr.data_length = wlen;
1371 hdr.flags = SD_FLAG_CMD_WRITE;
1373 ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen);
1377 if (!ret && rsp->result != SD_RES_SUCCESS &&
1378 rsp->result != SD_RES_VDI_NOT_LOCKED) {
1379 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1382 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL);
1384 if (s->cache_enabled) {
1385 closesocket(s->flush_fd);
1390 static int64_t sd_getlength(BlockDriverState *bs)
1392 BDRVSheepdogState *s = bs->opaque;
1394 return s->inode.vdi_size;
1397 static int sd_truncate(BlockDriverState *bs, int64_t offset)
1399 BDRVSheepdogState *s = bs->opaque;
1401 unsigned int datalen;
1403 if (offset < s->inode.vdi_size) {
1404 error_report("shrinking is not supported");
1406 } else if (offset > SD_MAX_VDI_SIZE) {
1407 error_report("too big image size");
1411 fd = connect_to_sdog(s->addr, s->port);
1416 /* we don't need to update entire object */
1417 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1418 s->inode.vdi_size = offset;
1419 ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
1420 s->inode.nr_copies, datalen, 0, 0, s->cache_enabled);
1424 error_report("failed to update an inode.");
1431 * This function is called after writing data objects. If we need to
1432 * update metadata, this sends a write request to the vdi object.
1433 * Otherwise, this switches back to sd_co_readv/writev.
1435 static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
1438 BDRVSheepdogState *s = acb->common.bs->opaque;
1441 uint32_t offset, data_len, mn, mx;
1443 mn = s->min_dirty_data_idx;
1444 mx = s->max_dirty_data_idx;
1446 /* we need to update the vdi object. */
1447 offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
1448 mn * sizeof(s->inode.data_vdi_id[0]);
1449 data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
1451 s->min_dirty_data_idx = UINT32_MAX;
1452 s->max_dirty_data_idx = 0;
1454 iov.iov_base = &s->inode;
1455 iov.iov_len = sizeof(s->inode);
1456 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
1457 data_len, offset, 0, 0, offset);
1458 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1459 ret = add_aio_request(s, aio_req, &iov, 1, 0, AIOCB_WRITE_UDATA);
1461 free_aio_req(s, aio_req);
1466 acb->aio_done_func = sd_finish_aiocb;
1467 acb->aiocb_type = AIOCB_WRITE_UDATA;
1471 sd_finish_aiocb(acb);
1475 * Create a writable VDI from a snapshot
1477 static int sd_create_branch(BDRVSheepdogState *s)
1483 dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id);
1485 buf = g_malloc(SD_INODE_SIZE);
1487 ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1,
1493 dprintf("%" PRIx32 " is created.\n", vid);
1495 fd = connect_to_sdog(s->addr, s->port);
1497 error_report("failed to connect");
1502 ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
1503 SD_INODE_SIZE, 0, s->cache_enabled);
1511 memcpy(&s->inode, buf, sizeof(s->inode));
1515 dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id);
1524 * Send I/O requests to the server.
1526 * This function sends requests to the server, links the requests to
1527 * the inflight_list in BDRVSheepdogState, and exits without
1528 * waiting the response. The responses are received in the
1529 * `aio_read_response' function which is called from the main loop as
1532 * Returns 1 when we need to wait a response, 0 when there is no sent
1533 * request and -errno in error cases.
1535 static int coroutine_fn sd_co_rw_vector(void *p)
1537 SheepdogAIOCB *acb = p;
1539 unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE;
1540 unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE;
1542 uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE;
1543 BDRVSheepdogState *s = acb->common.bs->opaque;
1544 SheepdogInode *inode = &s->inode;
1547 if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
1549 * In the case we open the snapshot VDI, Sheepdog creates the
1550 * writable VDI when we do a write operation first.
1552 ret = sd_create_branch(s);
1560 * Make sure we don't free the aiocb before we are done with all requests.
1561 * This additional reference is dropped at the end of this function.
1565 while (done != total) {
1567 uint64_t old_oid = 0;
1570 oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
1572 len = MIN(total - done, SD_DATA_OBJ_SIZE - offset);
1574 switch (acb->aiocb_type) {
1575 case AIOCB_READ_UDATA:
1576 if (!inode->data_vdi_id[idx]) {
1577 qemu_iovec_memset(acb->qiov, done, 0, len);
1581 case AIOCB_WRITE_UDATA:
1582 if (!inode->data_vdi_id[idx]) {
1584 } else if (!is_data_obj_writable(inode, idx)) {
1588 flags = SD_FLAG_CMD_COW;
1596 dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n",
1598 vid_to_data_oid(inode->data_vdi_id[idx], idx), idx);
1599 oid = vid_to_data_oid(inode->vdi_id, idx);
1600 dprintf("new oid %" PRIx64 "\n", oid);
1603 aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done);
1607 QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) {
1608 if (areq->oid == oid) {
1610 * Sheepdog cannot handle simultaneous create
1611 * requests to the same object. So we cannot send
1612 * the request until the previous request
1616 aio_req->base_oid = 0;
1617 QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req,
1624 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1625 ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
1626 create, acb->aiocb_type);
1628 error_report("add_aio_request is failed");
1629 free_aio_req(s, aio_req);
1639 if (!--acb->nr_pending) {
1645 static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
1646 int nb_sectors, QEMUIOVector *qiov)
1651 if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
1652 ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE);
1656 bs->total_sectors = sector_num + nb_sectors;
1659 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL);
1660 acb->aio_done_func = sd_write_done;
1661 acb->aiocb_type = AIOCB_WRITE_UDATA;
1663 ret = sd_co_rw_vector(acb);
1665 qemu_aio_release(acb);
1669 qemu_coroutine_yield();
1674 static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
1675 int nb_sectors, QEMUIOVector *qiov)
1680 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL);
1681 acb->aiocb_type = AIOCB_READ_UDATA;
1682 acb->aio_done_func = sd_finish_aiocb;
1684 ret = sd_co_rw_vector(acb);
1686 qemu_aio_release(acb);
1690 qemu_coroutine_yield();
1695 static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
1697 BDRVSheepdogState *s = bs->opaque;
1698 SheepdogObjReq hdr = { 0 };
1699 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1700 SheepdogInode *inode = &s->inode;
1702 unsigned int wlen = 0, rlen = 0;
1704 if (!s->cache_enabled) {
1708 hdr.opcode = SD_OP_FLUSH_VDI;
1709 hdr.oid = vid_to_vdi_oid(inode->vdi_id);
1711 ret = do_req(s->flush_fd, (SheepdogReq *)&hdr, NULL, &wlen, &rlen);
1713 error_report("failed to send a request to the sheep");
1717 if (rsp->result == SD_RES_INVALID_PARMS) {
1718 dprintf("disable write cache since the server doesn't support it\n");
1720 s->cache_enabled = 0;
1721 closesocket(s->flush_fd);
1725 if (rsp->result != SD_RES_SUCCESS) {
1726 error_report("%s", sd_strerror(rsp->result));
1733 static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
1735 BDRVSheepdogState *s = bs->opaque;
1738 SheepdogInode *inode;
1739 unsigned int datalen;
1741 dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " "
1742 "is_snapshot %d\n", sn_info->name, sn_info->id_str,
1743 s->name, sn_info->vm_state_size, s->is_snapshot);
1745 if (s->is_snapshot) {
1746 error_report("You can't create a snapshot of a snapshot VDI, "
1747 "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
1752 dprintf("%s %s\n", sn_info->name, sn_info->id_str);
1754 s->inode.vm_state_size = sn_info->vm_state_size;
1755 s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
1756 strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
1757 /* we don't need to update entire object */
1758 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1760 /* refresh inode. */
1761 fd = connect_to_sdog(s->addr, s->port);
1767 ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
1768 s->inode.nr_copies, datalen, 0, 0, s->cache_enabled);
1770 error_report("failed to write snapshot's inode.");
1774 ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, 1,
1777 error_report("failed to create inode for snapshot. %s",
1782 inode = (SheepdogInode *)g_malloc(datalen);
1784 ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid),
1785 s->inode.nr_copies, datalen, 0, s->cache_enabled);
1788 error_report("failed to read new inode info. %s", strerror(errno));
1792 memcpy(&s->inode, inode, datalen);
1793 dprintf("s->inode: name %s snap_id %x oid %x\n",
1794 s->inode.name, s->inode.snap_id, s->inode.vdi_id);
1801 static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
1803 BDRVSheepdogState *s = bs->opaque;
1804 BDRVSheepdogState *old_s;
1805 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1808 uint32_t snapid = 0;
1811 old_s = g_malloc(sizeof(BDRVSheepdogState));
1813 memcpy(old_s, s, sizeof(BDRVSheepdogState));
1815 memset(vdi, 0, sizeof(vdi));
1816 strncpy(vdi, s->name, sizeof(vdi));
1818 memset(tag, 0, sizeof(tag));
1819 snapid = strtoul(snapshot_id, NULL, 10);
1821 strncpy(tag, s->name, sizeof(tag));
1824 ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1);
1826 error_report("Failed to find_vdi_name");
1830 fd = connect_to_sdog(s->addr, s->port);
1832 error_report("failed to connect");
1837 buf = g_malloc(SD_INODE_SIZE);
1838 ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
1839 SD_INODE_SIZE, 0, s->cache_enabled);
1847 memcpy(&s->inode, buf, sizeof(s->inode));
1849 if (!s->inode.vm_state_size) {
1850 error_report("Invalid snapshot");
1862 /* recover bdrv_sd_state */
1863 memcpy(s, old_s, sizeof(BDRVSheepdogState));
1867 error_report("failed to open. recover old bdrv_sd_state.");
1872 static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1874 /* FIXME: Delete specified snapshot id. */
1878 static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
1880 BDRVSheepdogState *s = bs->opaque;
1882 int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
1883 QEMUSnapshotInfo *sn_tab = NULL;
1884 unsigned wlen, rlen;
1886 static SheepdogInode inode;
1887 unsigned long *vdi_inuse;
1888 unsigned int start_nr;
1892 vdi_inuse = g_malloc(max);
1894 fd = connect_to_sdog(s->addr, s->port);
1903 memset(&req, 0, sizeof(req));
1905 req.opcode = SD_OP_READ_VDIS;
1906 req.data_length = max;
1908 ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen);
1915 sn_tab = g_malloc0(nr * sizeof(*sn_tab));
1917 /* calculate a vdi id with hash function */
1918 hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
1919 start_nr = hval & (SD_NR_VDIS - 1);
1921 fd = connect_to_sdog(s->addr, s->port);
1923 error_report("failed to connect");
1928 for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
1929 if (!test_bit(vid, vdi_inuse)) {
1933 /* we don't need to read entire object */
1934 ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid),
1935 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0,
1942 if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
1943 sn_tab[found].date_sec = inode.snap_ctime >> 32;
1944 sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
1945 sn_tab[found].vm_state_size = inode.vm_state_size;
1946 sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec;
1948 snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u",
1950 strncpy(sn_tab[found].name, inode.tag,
1951 MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)));
1969 static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
1970 int64_t pos, int size, int load)
1973 int ret = 0, remaining = size;
1974 unsigned int data_len;
1975 uint64_t vmstate_oid;
1979 fd = connect_to_sdog(s->addr, s->port);
1985 vdi_index = pos / SD_DATA_OBJ_SIZE;
1986 offset = pos % SD_DATA_OBJ_SIZE;
1988 data_len = MIN(remaining, SD_DATA_OBJ_SIZE);
1990 vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index);
1992 create = (offset == 0);
1994 ret = read_object(fd, (char *)data, vmstate_oid,
1995 s->inode.nr_copies, data_len, offset,
1998 ret = write_object(fd, (char *)data, vmstate_oid,
1999 s->inode.nr_copies, data_len, offset, create,
2004 error_report("failed to save vmstate %s", strerror(errno));
2009 remaining -= data_len;
2017 static int sd_save_vmstate(BlockDriverState *bs, const uint8_t *data,
2018 int64_t pos, int size)
2020 BDRVSheepdogState *s = bs->opaque;
2022 return do_load_save_vmstate(s, (uint8_t *)data, pos, size, 0);
2025 static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data,
2026 int64_t pos, int size)
2028 BDRVSheepdogState *s = bs->opaque;
2030 return do_load_save_vmstate(s, data, pos, size, 1);
2034 static QEMUOptionParameter sd_create_options[] = {
2036 .name = BLOCK_OPT_SIZE,
2038 .help = "Virtual disk size"
2041 .name = BLOCK_OPT_BACKING_FILE,
2043 .help = "File name of a base image"
2046 .name = BLOCK_OPT_PREALLOC,
2048 .help = "Preallocation mode (allowed values: off, full)"
2053 BlockDriver bdrv_sheepdog = {
2054 .format_name = "sheepdog",
2055 .protocol_name = "sheepdog",
2056 .instance_size = sizeof(BDRVSheepdogState),
2057 .bdrv_file_open = sd_open,
2058 .bdrv_close = sd_close,
2059 .bdrv_create = sd_create,
2060 .bdrv_getlength = sd_getlength,
2061 .bdrv_truncate = sd_truncate,
2063 .bdrv_co_readv = sd_co_readv,
2064 .bdrv_co_writev = sd_co_writev,
2065 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2067 .bdrv_snapshot_create = sd_snapshot_create,
2068 .bdrv_snapshot_goto = sd_snapshot_goto,
2069 .bdrv_snapshot_delete = sd_snapshot_delete,
2070 .bdrv_snapshot_list = sd_snapshot_list,
2072 .bdrv_save_vmstate = sd_save_vmstate,
2073 .bdrv_load_vmstate = sd_load_vmstate,
2075 .create_options = sd_create_options,
2078 static void bdrv_sheepdog_init(void)
2080 bdrv_register(&bdrv_sheepdog);
2082 block_init(bdrv_sheepdog_init);