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-report.h"
17 #include "qemu/sockets.h"
18 #include "block/block_int.h"
19 #include "qemu/bitops.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 bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
206 return inode->vdi_id == inode->data_vdi_id[idx];
209 static inline bool 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 bool 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];
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);
420 acb->canceled = true;
423 static const AIOCBInfo sd_aiocb_info = {
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_aiocb_info, bs, cb, opaque);
438 acb->sector_num = sector_num;
439 acb->nb_sectors = nb_sectors;
441 acb->aio_done_func = NULL;
442 acb->canceled = false;
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) {
492 dprintf("connected to %s:%s\n", addr, port);
496 error_report("failed connect to %s:%s", addr, port);
502 static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
507 ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
508 if (ret < sizeof(*hdr)) {
509 error_report("failed to send a req, %s", strerror(errno));
513 ret = qemu_co_send(sockfd, data, *wlen);
515 error_report("failed to send a req, %s", strerror(errno));
521 static void restart_co_req(void *opaque)
523 Coroutine *co = opaque;
525 qemu_coroutine_enter(co, NULL);
528 typedef struct SheepdogReqCo {
538 static coroutine_fn void do_co_req(void *opaque)
542 SheepdogReqCo *srco = opaque;
543 int sockfd = srco->sockfd;
544 SheepdogReq *hdr = srco->hdr;
545 void *data = srco->data;
546 unsigned int *wlen = srco->wlen;
547 unsigned int *rlen = srco->rlen;
549 co = qemu_coroutine_self();
550 qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, NULL, co);
552 socket_set_block(sockfd);
553 ret = send_co_req(sockfd, hdr, data, wlen);
558 qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, NULL, co);
560 ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
561 if (ret < sizeof(*hdr)) {
562 error_report("failed to get a rsp, %s", strerror(errno));
567 if (*rlen > hdr->data_length) {
568 *rlen = hdr->data_length;
572 ret = qemu_co_recv(sockfd, data, *rlen);
574 error_report("failed to get the data, %s", strerror(errno));
581 qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL);
582 socket_set_nonblock(sockfd);
585 srco->finished = true;
588 static int do_req(int sockfd, SheepdogReq *hdr, void *data,
589 unsigned int *wlen, unsigned int *rlen)
592 SheepdogReqCo srco = {
602 if (qemu_in_coroutine()) {
605 co = qemu_coroutine_create(do_co_req);
606 qemu_coroutine_enter(co, &srco);
607 while (!srco.finished) {
615 static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
616 struct iovec *iov, int niov, bool create,
617 enum AIOCBState aiocb_type);
620 static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid)
624 QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) {
625 if (aio_req->oid == oid) {
634 * This function searchs pending requests to the object `oid', and
637 static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid)
643 while ((aio_req = find_pending_req(s, oid)) != NULL) {
644 acb = aio_req->aiocb;
645 /* move aio_req from pending list to inflight one */
646 QLIST_REMOVE(aio_req, aio_siblings);
647 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
648 ret = add_aio_request(s, aio_req, acb->qiov->iov,
649 acb->qiov->niov, false, acb->aiocb_type);
651 error_report("add_aio_request is failed");
652 free_aio_req(s, aio_req);
653 if (!acb->nr_pending) {
654 sd_finish_aiocb(acb);
661 * Receive responses of the I/O requests.
663 * This function is registered as a fd handler, and called from the
664 * main loop when s->fd is ready for reading responses.
666 static void coroutine_fn aio_read_response(void *opaque)
669 BDRVSheepdogState *s = opaque;
672 AIOReq *aio_req = NULL;
676 if (QLIST_EMPTY(&s->inflight_aio_head)) {
681 ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
683 error_report("failed to get the header, %s", strerror(errno));
687 /* find the right aio_req from the inflight aio list */
688 QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
689 if (aio_req->id == rsp.id) {
694 error_report("cannot find aio_req %x", rsp.id);
698 acb = aio_req->aiocb;
700 switch (acb->aiocb_type) {
701 case AIOCB_WRITE_UDATA:
702 /* this coroutine context is no longer suitable for co_recv
703 * because we may send data to update vdi objects */
705 if (!is_data_obj(aio_req->oid)) {
708 idx = data_oid_to_idx(aio_req->oid);
710 if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) {
712 * If the object is newly created one, we need to update
713 * the vdi object (metadata object). min_dirty_data_idx
714 * and max_dirty_data_idx are changed to include updated
715 * index between them.
717 if (rsp.result == SD_RES_SUCCESS) {
718 s->inode.data_vdi_id[idx] = s->inode.vdi_id;
719 s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx);
720 s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx);
723 * Some requests may be blocked because simultaneous
724 * create requests are not allowed, so we search the
725 * pending requests here.
727 send_pending_req(s, aio_req->oid);
730 case AIOCB_READ_UDATA:
731 ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
732 aio_req->iov_offset, rsp.data_length);
734 error_report("failed to get the data, %s", strerror(errno));
740 if (rsp.result != SD_RES_SUCCESS) {
742 error_report("%s", sd_strerror(rsp.result));
745 free_aio_req(s, aio_req);
746 if (!acb->nr_pending) {
748 * We've finished all requests which belong to the AIOCB, so
749 * we can switch back to sd_co_readv/writev now.
751 acb->aio_done_func(acb);
757 static void co_read_response(void *opaque)
759 BDRVSheepdogState *s = opaque;
762 s->co_recv = qemu_coroutine_create(aio_read_response);
765 qemu_coroutine_enter(s->co_recv, opaque);
768 static void co_write_request(void *opaque)
770 BDRVSheepdogState *s = opaque;
772 qemu_coroutine_enter(s->co_send, NULL);
775 static int aio_flush_request(void *opaque)
777 BDRVSheepdogState *s = opaque;
779 return !QLIST_EMPTY(&s->inflight_aio_head) ||
780 !QLIST_EMPTY(&s->pending_aio_head);
783 static int set_nodelay(int fd)
788 ret = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *)&opt, sizeof(opt));
793 * Return a socket discriptor to read/write objects.
795 * We cannot use this discriptor for other operations because
796 * the block driver may be on waiting response from the server.
798 static int get_sheep_fd(BDRVSheepdogState *s)
802 fd = connect_to_sdog(s->addr, s->port);
804 error_report("%s", strerror(errno));
808 socket_set_nonblock(fd);
810 ret = set_nodelay(fd);
812 error_report("%s", strerror(errno));
817 qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s);
824 * filename must be one of the following formats:
826 * 2. [vdiname]:[snapid]
828 * 4. [hostname]:[port]:[vdiname]
829 * 5. [hostname]:[port]:[vdiname]:[snapid]
830 * 6. [hostname]:[port]:[vdiname]:[tag]
832 * You can boot from the snapshot images by specifying `snapid` or
835 * You can run VMs outside the Sheepdog cluster by specifying
836 * `hostname' and `port' (experimental).
838 static int parse_vdiname(BDRVSheepdogState *s, const char *filename,
839 char *vdi, uint32_t *snapid, char *tag)
844 p = q = g_strdup(filename);
846 /* count the number of separators */
856 /* use the first two tokens as hostname and port number. */
870 pstrcpy(vdi, SD_MAX_VDI_LEN, p);
872 p = strchr(vdi, ':');
875 *snapid = strtoul(p, NULL, 10);
877 pstrcpy(tag, SD_MAX_VDI_TAG_LEN, p);
880 *snapid = CURRENT_VDI_ID; /* search current vdi */
883 if (s->addr == NULL) {
890 static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid,
891 char *tag, uint32_t *vid, int for_snapshot)
895 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
896 unsigned int wlen, rlen = 0;
897 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
899 fd = connect_to_sdog(s->addr, s->port);
904 /* This pair of strncpy calls ensures that the buffer is zero-filled,
905 * which is desirable since we'll soon be sending those bytes, and
906 * don't want the send_req to read uninitialized data.
908 strncpy(buf, filename, SD_MAX_VDI_LEN);
909 strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
911 memset(&hdr, 0, sizeof(hdr));
913 hdr.opcode = SD_OP_GET_VDI_INFO;
915 hdr.opcode = SD_OP_LOCK_VDI;
917 wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
918 hdr.proto_ver = SD_PROTO_VER;
919 hdr.data_length = wlen;
921 hdr.flags = SD_FLAG_CMD_WRITE;
923 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
928 if (rsp->result != SD_RES_SUCCESS) {
929 error_report("cannot get vdi info, %s, %s %d %s",
930 sd_strerror(rsp->result), filename, snapid, tag);
931 if (rsp->result == SD_RES_NO_VDI) {
946 static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
947 struct iovec *iov, int niov, bool create,
948 enum AIOCBState aiocb_type)
950 int nr_copies = s->inode.nr_copies;
954 uint64_t oid = aio_req->oid;
955 unsigned int datalen = aio_req->data_len;
956 uint64_t offset = aio_req->offset;
957 uint8_t flags = aio_req->flags;
958 uint64_t old_oid = aio_req->base_oid;
964 memset(&hdr, 0, sizeof(hdr));
966 if (aiocb_type == AIOCB_READ_UDATA) {
968 hdr.opcode = SD_OP_READ_OBJ;
972 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
973 hdr.flags = SD_FLAG_CMD_WRITE | flags;
976 hdr.opcode = SD_OP_WRITE_OBJ;
977 hdr.flags = SD_FLAG_CMD_WRITE | flags;
980 if (s->cache_enabled) {
981 hdr.flags |= SD_FLAG_CMD_CACHE;
985 hdr.cow_oid = old_oid;
986 hdr.copies = s->inode.nr_copies;
988 hdr.data_length = datalen;
991 hdr.id = aio_req->id;
993 qemu_co_mutex_lock(&s->lock);
994 s->co_send = qemu_coroutine_self();
995 qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request,
996 aio_flush_request, s);
997 socket_set_cork(s->fd, 1);
1000 ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
1002 qemu_co_mutex_unlock(&s->lock);
1003 error_report("failed to send a req, %s", strerror(errno));
1008 ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1010 qemu_co_mutex_unlock(&s->lock);
1011 error_report("failed to send a data, %s", strerror(errno));
1016 socket_set_cork(s->fd, 0);
1017 qemu_aio_set_fd_handler(s->fd, co_read_response, NULL,
1018 aio_flush_request, s);
1019 qemu_co_mutex_unlock(&s->lock);
1024 static int read_write_object(int fd, char *buf, uint64_t oid, int copies,
1025 unsigned int datalen, uint64_t offset,
1026 bool write, bool create, bool cache)
1029 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1030 unsigned int wlen, rlen;
1033 memset(&hdr, 0, sizeof(hdr));
1038 hdr.flags = SD_FLAG_CMD_WRITE;
1040 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1042 hdr.opcode = SD_OP_WRITE_OBJ;
1047 hdr.opcode = SD_OP_READ_OBJ;
1051 hdr.flags |= SD_FLAG_CMD_CACHE;
1055 hdr.data_length = datalen;
1056 hdr.offset = offset;
1057 hdr.copies = copies;
1059 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1061 error_report("failed to send a request to the sheep");
1065 switch (rsp->result) {
1066 case SD_RES_SUCCESS:
1069 error_report("%s", sd_strerror(rsp->result));
1074 static int read_object(int fd, char *buf, uint64_t oid, int copies,
1075 unsigned int datalen, uint64_t offset, bool cache)
1077 return read_write_object(fd, buf, oid, copies, datalen, offset, false,
1081 static int write_object(int fd, char *buf, uint64_t oid, int copies,
1082 unsigned int datalen, uint64_t offset, bool create,
1085 return read_write_object(fd, buf, oid, copies, datalen, offset, true,
1089 static int sd_open(BlockDriverState *bs, const char *filename, int flags)
1093 BDRVSheepdogState *s = bs->opaque;
1094 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1098 strstart(filename, "sheepdog:", (const char **)&filename);
1100 QLIST_INIT(&s->inflight_aio_head);
1101 QLIST_INIT(&s->pending_aio_head);
1104 memset(vdi, 0, sizeof(vdi));
1105 memset(tag, 0, sizeof(tag));
1106 if (parse_vdiname(s, filename, vdi, &snapid, tag) < 0) {
1110 s->fd = get_sheep_fd(s);
1116 ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0);
1121 s->cache_enabled = true;
1122 s->flush_fd = connect_to_sdog(s->addr, s->port);
1123 if (s->flush_fd < 0) {
1124 error_report("failed to connect");
1129 if (snapid || tag[0] != '\0') {
1130 dprintf("%" PRIx32 " snapshot inode was open.\n", vid);
1131 s->is_snapshot = true;
1134 fd = connect_to_sdog(s->addr, s->port);
1136 error_report("failed to connect");
1141 buf = g_malloc(SD_INODE_SIZE);
1142 ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0,
1151 memcpy(&s->inode, buf, sizeof(s->inode));
1152 s->min_dirty_data_idx = UINT32_MAX;
1153 s->max_dirty_data_idx = 0;
1155 bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE;
1156 pstrcpy(s->name, sizeof(s->name), vdi);
1157 qemu_co_mutex_init(&s->lock);
1161 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL);
1169 static int do_sd_create(char *filename, int64_t vdi_size,
1170 uint32_t base_vid, uint32_t *vdi_id, int snapshot,
1171 const char *addr, const char *port)
1174 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1176 unsigned int wlen, rlen = 0;
1177 char buf[SD_MAX_VDI_LEN];
1179 fd = connect_to_sdog(addr, port);
1184 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1185 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1187 memset(buf, 0, sizeof(buf));
1188 pstrcpy(buf, sizeof(buf), filename);
1190 memset(&hdr, 0, sizeof(hdr));
1191 hdr.opcode = SD_OP_NEW_VDI;
1192 hdr.base_vdi_id = base_vid;
1194 wlen = SD_MAX_VDI_LEN;
1196 hdr.flags = SD_FLAG_CMD_WRITE;
1197 hdr.snapid = snapshot;
1199 hdr.data_length = wlen;
1200 hdr.vdi_size = vdi_size;
1202 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1210 if (rsp->result != SD_RES_SUCCESS) {
1211 error_report("%s, %s", sd_strerror(rsp->result), filename);
1216 *vdi_id = rsp->vdi_id;
1222 static int sd_prealloc(const char *filename)
1224 BlockDriverState *bs = NULL;
1225 uint32_t idx, max_idx;
1227 void *buf = g_malloc0(SD_DATA_OBJ_SIZE);
1230 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
1235 vdi_size = bdrv_getlength(bs);
1240 max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE);
1242 for (idx = 0; idx < max_idx; idx++) {
1244 * The created image can be a cloned image, so we need to read
1245 * a data from the source image.
1247 ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1251 ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1265 static int sd_create(const char *filename, QEMUOptionParameter *options)
1268 uint32_t vid = 0, base_vid = 0;
1269 int64_t vdi_size = 0;
1270 char *backing_file = NULL;
1271 BDRVSheepdogState *s;
1272 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1274 bool prealloc = false;
1275 const char *vdiname;
1277 s = g_malloc0(sizeof(BDRVSheepdogState));
1279 strstart(filename, "sheepdog:", &vdiname);
1281 memset(vdi, 0, sizeof(vdi));
1282 memset(tag, 0, sizeof(tag));
1283 if (parse_vdiname(s, vdiname, vdi, &snapid, tag) < 0) {
1284 error_report("invalid filename");
1289 while (options && options->name) {
1290 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1291 vdi_size = options->value.n;
1292 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1293 backing_file = options->value.s;
1294 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1295 if (!options->value.s || !strcmp(options->value.s, "off")) {
1297 } else if (!strcmp(options->value.s, "full")) {
1300 error_report("Invalid preallocation mode: '%s'",
1309 if (vdi_size > SD_MAX_VDI_SIZE) {
1310 error_report("too big image size");
1316 BlockDriverState *bs;
1317 BDRVSheepdogState *s;
1320 /* Currently, only Sheepdog backing image is supported. */
1321 drv = bdrv_find_protocol(backing_file);
1322 if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
1323 error_report("backing_file must be a sheepdog image");
1328 ret = bdrv_file_open(&bs, backing_file, 0);
1335 if (!is_snapshot(&s->inode)) {
1336 error_report("cannot clone from a non snapshot vdi");
1342 base_vid = s->inode.vdi_id;
1346 ret = do_sd_create(vdi, vdi_size, base_vid, &vid, 0, s->addr, s->port);
1347 if (!prealloc || ret) {
1351 ret = sd_prealloc(filename);
1357 static void sd_close(BlockDriverState *bs)
1359 BDRVSheepdogState *s = bs->opaque;
1361 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1362 unsigned int wlen, rlen = 0;
1365 dprintf("%s\n", s->name);
1367 fd = connect_to_sdog(s->addr, s->port);
1372 memset(&hdr, 0, sizeof(hdr));
1374 hdr.opcode = SD_OP_RELEASE_VDI;
1375 wlen = strlen(s->name) + 1;
1376 hdr.data_length = wlen;
1377 hdr.flags = SD_FLAG_CMD_WRITE;
1379 ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen);
1383 if (!ret && rsp->result != SD_RES_SUCCESS &&
1384 rsp->result != SD_RES_VDI_NOT_LOCKED) {
1385 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1388 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL);
1390 if (s->cache_enabled) {
1391 closesocket(s->flush_fd);
1396 static int64_t sd_getlength(BlockDriverState *bs)
1398 BDRVSheepdogState *s = bs->opaque;
1400 return s->inode.vdi_size;
1403 static int sd_truncate(BlockDriverState *bs, int64_t offset)
1405 BDRVSheepdogState *s = bs->opaque;
1407 unsigned int datalen;
1409 if (offset < s->inode.vdi_size) {
1410 error_report("shrinking is not supported");
1412 } else if (offset > SD_MAX_VDI_SIZE) {
1413 error_report("too big image size");
1417 fd = connect_to_sdog(s->addr, s->port);
1422 /* we don't need to update entire object */
1423 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1424 s->inode.vdi_size = offset;
1425 ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
1426 s->inode.nr_copies, datalen, 0, false, s->cache_enabled);
1430 error_report("failed to update an inode.");
1437 * This function is called after writing data objects. If we need to
1438 * update metadata, this sends a write request to the vdi object.
1439 * Otherwise, this switches back to sd_co_readv/writev.
1441 static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
1444 BDRVSheepdogState *s = acb->common.bs->opaque;
1447 uint32_t offset, data_len, mn, mx;
1449 mn = s->min_dirty_data_idx;
1450 mx = s->max_dirty_data_idx;
1452 /* we need to update the vdi object. */
1453 offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
1454 mn * sizeof(s->inode.data_vdi_id[0]);
1455 data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
1457 s->min_dirty_data_idx = UINT32_MAX;
1458 s->max_dirty_data_idx = 0;
1460 iov.iov_base = &s->inode;
1461 iov.iov_len = sizeof(s->inode);
1462 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
1463 data_len, offset, 0, 0, offset);
1464 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1465 ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA);
1467 free_aio_req(s, aio_req);
1472 acb->aio_done_func = sd_finish_aiocb;
1473 acb->aiocb_type = AIOCB_WRITE_UDATA;
1477 sd_finish_aiocb(acb);
1481 * Create a writable VDI from a snapshot
1483 static int sd_create_branch(BDRVSheepdogState *s)
1489 dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id);
1491 buf = g_malloc(SD_INODE_SIZE);
1493 ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1,
1499 dprintf("%" PRIx32 " is created.\n", vid);
1501 fd = connect_to_sdog(s->addr, s->port);
1503 error_report("failed to connect");
1508 ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
1509 SD_INODE_SIZE, 0, s->cache_enabled);
1517 memcpy(&s->inode, buf, sizeof(s->inode));
1519 s->is_snapshot = false;
1521 dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id);
1530 * Send I/O requests to the server.
1532 * This function sends requests to the server, links the requests to
1533 * the inflight_list in BDRVSheepdogState, and exits without
1534 * waiting the response. The responses are received in the
1535 * `aio_read_response' function which is called from the main loop as
1538 * Returns 1 when we need to wait a response, 0 when there is no sent
1539 * request and -errno in error cases.
1541 static int coroutine_fn sd_co_rw_vector(void *p)
1543 SheepdogAIOCB *acb = p;
1545 unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE;
1546 unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE;
1548 uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE;
1549 BDRVSheepdogState *s = acb->common.bs->opaque;
1550 SheepdogInode *inode = &s->inode;
1553 if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
1555 * In the case we open the snapshot VDI, Sheepdog creates the
1556 * writable VDI when we do a write operation first.
1558 ret = sd_create_branch(s);
1566 * Make sure we don't free the aiocb before we are done with all requests.
1567 * This additional reference is dropped at the end of this function.
1571 while (done != total) {
1573 uint64_t old_oid = 0;
1574 bool create = false;
1576 oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
1578 len = MIN(total - done, SD_DATA_OBJ_SIZE - offset);
1580 switch (acb->aiocb_type) {
1581 case AIOCB_READ_UDATA:
1582 if (!inode->data_vdi_id[idx]) {
1583 qemu_iovec_memset(acb->qiov, done, 0, len);
1587 case AIOCB_WRITE_UDATA:
1588 if (!inode->data_vdi_id[idx]) {
1590 } else if (!is_data_obj_writable(inode, idx)) {
1594 flags = SD_FLAG_CMD_COW;
1602 dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n",
1604 vid_to_data_oid(inode->data_vdi_id[idx], idx), idx);
1605 oid = vid_to_data_oid(inode->vdi_id, idx);
1606 dprintf("new oid %" PRIx64 "\n", oid);
1609 aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done);
1613 QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) {
1614 if (areq->oid == oid) {
1616 * Sheepdog cannot handle simultaneous create
1617 * requests to the same object. So we cannot send
1618 * the request until the previous request
1622 aio_req->base_oid = 0;
1623 QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req,
1630 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1631 ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
1632 create, acb->aiocb_type);
1634 error_report("add_aio_request is failed");
1635 free_aio_req(s, aio_req);
1645 if (!--acb->nr_pending) {
1651 static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
1652 int nb_sectors, QEMUIOVector *qiov)
1657 if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
1658 ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE);
1662 bs->total_sectors = sector_num + nb_sectors;
1665 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL);
1666 acb->aio_done_func = sd_write_done;
1667 acb->aiocb_type = AIOCB_WRITE_UDATA;
1669 ret = sd_co_rw_vector(acb);
1671 qemu_aio_release(acb);
1675 qemu_coroutine_yield();
1680 static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
1681 int nb_sectors, QEMUIOVector *qiov)
1686 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL);
1687 acb->aiocb_type = AIOCB_READ_UDATA;
1688 acb->aio_done_func = sd_finish_aiocb;
1690 ret = sd_co_rw_vector(acb);
1692 qemu_aio_release(acb);
1696 qemu_coroutine_yield();
1701 static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
1703 BDRVSheepdogState *s = bs->opaque;
1704 SheepdogObjReq hdr = { 0 };
1705 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1706 SheepdogInode *inode = &s->inode;
1708 unsigned int wlen = 0, rlen = 0;
1710 if (!s->cache_enabled) {
1714 hdr.opcode = SD_OP_FLUSH_VDI;
1715 hdr.oid = vid_to_vdi_oid(inode->vdi_id);
1717 ret = do_req(s->flush_fd, (SheepdogReq *)&hdr, NULL, &wlen, &rlen);
1719 error_report("failed to send a request to the sheep");
1723 if (rsp->result == SD_RES_INVALID_PARMS) {
1724 dprintf("disable write cache since the server doesn't support it\n");
1726 s->cache_enabled = false;
1727 closesocket(s->flush_fd);
1731 if (rsp->result != SD_RES_SUCCESS) {
1732 error_report("%s", sd_strerror(rsp->result));
1739 static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
1741 BDRVSheepdogState *s = bs->opaque;
1744 SheepdogInode *inode;
1745 unsigned int datalen;
1747 dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " "
1748 "is_snapshot %d\n", sn_info->name, sn_info->id_str,
1749 s->name, sn_info->vm_state_size, s->is_snapshot);
1751 if (s->is_snapshot) {
1752 error_report("You can't create a snapshot of a snapshot VDI, "
1753 "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
1758 dprintf("%s %s\n", sn_info->name, sn_info->id_str);
1760 s->inode.vm_state_size = sn_info->vm_state_size;
1761 s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
1762 /* It appears that inode.tag does not require a NUL terminator,
1763 * which means this use of strncpy is ok.
1765 strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
1766 /* we don't need to update entire object */
1767 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1769 /* refresh inode. */
1770 fd = connect_to_sdog(s->addr, s->port);
1776 ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
1777 s->inode.nr_copies, datalen, 0, false, s->cache_enabled);
1779 error_report("failed to write snapshot's inode.");
1783 ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, 1,
1786 error_report("failed to create inode for snapshot. %s",
1791 inode = (SheepdogInode *)g_malloc(datalen);
1793 ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid),
1794 s->inode.nr_copies, datalen, 0, s->cache_enabled);
1797 error_report("failed to read new inode info. %s", strerror(errno));
1801 memcpy(&s->inode, inode, datalen);
1802 dprintf("s->inode: name %s snap_id %x oid %x\n",
1803 s->inode.name, s->inode.snap_id, s->inode.vdi_id);
1810 static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
1812 BDRVSheepdogState *s = bs->opaque;
1813 BDRVSheepdogState *old_s;
1814 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1817 uint32_t snapid = 0;
1820 old_s = g_malloc(sizeof(BDRVSheepdogState));
1822 memcpy(old_s, s, sizeof(BDRVSheepdogState));
1824 pstrcpy(vdi, sizeof(vdi), s->name);
1826 snapid = strtoul(snapshot_id, NULL, 10);
1830 pstrcpy(tag, sizeof(tag), s->name);
1833 ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1);
1835 error_report("Failed to find_vdi_name");
1839 fd = connect_to_sdog(s->addr, s->port);
1841 error_report("failed to connect");
1846 buf = g_malloc(SD_INODE_SIZE);
1847 ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
1848 SD_INODE_SIZE, 0, s->cache_enabled);
1856 memcpy(&s->inode, buf, sizeof(s->inode));
1858 if (!s->inode.vm_state_size) {
1859 error_report("Invalid snapshot");
1864 s->is_snapshot = true;
1871 /* recover bdrv_sd_state */
1872 memcpy(s, old_s, sizeof(BDRVSheepdogState));
1876 error_report("failed to open. recover old bdrv_sd_state.");
1881 static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1883 /* FIXME: Delete specified snapshot id. */
1887 static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
1889 BDRVSheepdogState *s = bs->opaque;
1891 int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
1892 QEMUSnapshotInfo *sn_tab = NULL;
1893 unsigned wlen, rlen;
1895 static SheepdogInode inode;
1896 unsigned long *vdi_inuse;
1897 unsigned int start_nr;
1901 vdi_inuse = g_malloc(max);
1903 fd = connect_to_sdog(s->addr, s->port);
1912 memset(&req, 0, sizeof(req));
1914 req.opcode = SD_OP_READ_VDIS;
1915 req.data_length = max;
1917 ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen);
1924 sn_tab = g_malloc0(nr * sizeof(*sn_tab));
1926 /* calculate a vdi id with hash function */
1927 hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
1928 start_nr = hval & (SD_NR_VDIS - 1);
1930 fd = connect_to_sdog(s->addr, s->port);
1932 error_report("failed to connect");
1937 for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
1938 if (!test_bit(vid, vdi_inuse)) {
1942 /* we don't need to read entire object */
1943 ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid),
1944 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0,
1951 if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
1952 sn_tab[found].date_sec = inode.snap_ctime >> 32;
1953 sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
1954 sn_tab[found].vm_state_size = inode.vm_state_size;
1955 sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec;
1957 snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u",
1959 pstrcpy(sn_tab[found].name,
1960 MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)),
1979 static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
1980 int64_t pos, int size, int load)
1983 int fd, ret = 0, remaining = size;
1984 unsigned int data_len;
1985 uint64_t vmstate_oid;
1989 fd = connect_to_sdog(s->addr, s->port);
1995 vdi_index = pos / SD_DATA_OBJ_SIZE;
1996 offset = pos % SD_DATA_OBJ_SIZE;
1998 data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset);
2000 vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index);
2002 create = (offset == 0);
2004 ret = read_object(fd, (char *)data, vmstate_oid,
2005 s->inode.nr_copies, data_len, offset,
2008 ret = write_object(fd, (char *)data, vmstate_oid,
2009 s->inode.nr_copies, data_len, offset, create,
2014 error_report("failed to save vmstate %s", strerror(errno));
2020 remaining -= data_len;
2028 static int sd_save_vmstate(BlockDriverState *bs, const uint8_t *data,
2029 int64_t pos, int size)
2031 BDRVSheepdogState *s = bs->opaque;
2033 return do_load_save_vmstate(s, (uint8_t *)data, pos, size, 0);
2036 static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data,
2037 int64_t pos, int size)
2039 BDRVSheepdogState *s = bs->opaque;
2041 return do_load_save_vmstate(s, data, pos, size, 1);
2045 static QEMUOptionParameter sd_create_options[] = {
2047 .name = BLOCK_OPT_SIZE,
2049 .help = "Virtual disk size"
2052 .name = BLOCK_OPT_BACKING_FILE,
2054 .help = "File name of a base image"
2057 .name = BLOCK_OPT_PREALLOC,
2059 .help = "Preallocation mode (allowed values: off, full)"
2064 BlockDriver bdrv_sheepdog = {
2065 .format_name = "sheepdog",
2066 .protocol_name = "sheepdog",
2067 .instance_size = sizeof(BDRVSheepdogState),
2068 .bdrv_file_open = sd_open,
2069 .bdrv_close = sd_close,
2070 .bdrv_create = sd_create,
2071 .bdrv_getlength = sd_getlength,
2072 .bdrv_truncate = sd_truncate,
2074 .bdrv_co_readv = sd_co_readv,
2075 .bdrv_co_writev = sd_co_writev,
2076 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2078 .bdrv_snapshot_create = sd_snapshot_create,
2079 .bdrv_snapshot_goto = sd_snapshot_goto,
2080 .bdrv_snapshot_delete = sd_snapshot_delete,
2081 .bdrv_snapshot_list = sd_snapshot_list,
2083 .bdrv_save_vmstate = sd_save_vmstate,
2084 .bdrv_load_vmstate = sd_load_vmstate,
2086 .create_options = sd_create_options,
2089 static void bdrv_sheepdog_init(void)
2091 bdrv_register(&bdrv_sheepdog);
2093 block_init(bdrv_sheepdog_init);