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 typedef struct SheepdogReq {
112 uint32_t data_length;
113 uint32_t opcode_specific[8];
116 typedef struct SheepdogRsp {
122 uint32_t data_length;
124 uint32_t opcode_specific[7];
127 typedef struct SheepdogObjReq {
133 uint32_t data_length;
142 typedef struct SheepdogObjRsp {
148 uint32_t data_length;
156 typedef struct SheepdogVdiReq {
162 uint32_t data_length;
164 uint32_t base_vdi_id;
172 typedef struct SheepdogVdiRsp {
178 uint32_t data_length;
185 typedef struct SheepdogInode {
186 char name[SD_MAX_VDI_LEN];
187 char tag[SD_MAX_VDI_TAG_LEN];
190 uint64_t vm_clock_nsec;
192 uint64_t vm_state_size;
193 uint16_t copy_policy;
195 uint8_t block_size_shift;
198 uint32_t parent_vdi_id;
199 uint32_t child_vdi_id[MAX_CHILDREN];
200 uint32_t data_vdi_id[MAX_DATA_OBJS];
204 * 64 bit FNV-1a non-zero initial basis
206 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
209 * 64 bit Fowler/Noll/Vo FNV-1a hash code
211 static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
213 unsigned char *bp = buf;
214 unsigned char *be = bp + len;
216 hval ^= (uint64_t) *bp++;
217 hval += (hval << 1) + (hval << 4) + (hval << 5) +
218 (hval << 7) + (hval << 8) + (hval << 40);
223 static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
225 return inode->vdi_id == inode->data_vdi_id[idx];
228 static inline bool is_data_obj(uint64_t oid)
230 return !(VDI_BIT & oid);
233 static inline uint64_t data_oid_to_idx(uint64_t oid)
235 return oid & (MAX_DATA_OBJS - 1);
238 static inline uint32_t oid_to_vid(uint64_t oid)
240 return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT;
243 static inline uint64_t vid_to_vdi_oid(uint32_t vid)
245 return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
248 static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx)
250 return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
253 static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx)
255 return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
258 static inline bool is_snapshot(struct SheepdogInode *inode)
260 return !!inode->snap_ctime;
265 #define DPRINTF(fmt, args...) \
267 fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \
270 #define DPRINTF(fmt, args...)
273 typedef struct SheepdogAIOCB SheepdogAIOCB;
275 typedef struct AIOReq {
276 SheepdogAIOCB *aiocb;
277 unsigned int iov_offset;
282 unsigned int data_len;
286 QLIST_ENTRY(AIOReq) aio_siblings;
296 struct SheepdogAIOCB {
297 BlockDriverAIOCB common;
305 enum AIOCBState aiocb_type;
307 Coroutine *coroutine;
308 void (*aio_done_func)(SheepdogAIOCB *);
315 typedef struct BDRVSheepdogState {
316 BlockDriverState *bs;
320 uint32_t min_dirty_data_idx;
321 uint32_t max_dirty_data_idx;
323 char name[SD_MAX_VDI_LEN];
325 uint32_t cache_flags;
326 bool discard_supported;
336 uint32_t aioreq_seq_num;
338 /* Every aio request must be linked to either of these queues. */
339 QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head;
340 QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head;
341 QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head;
344 static const char * sd_strerror(int err)
348 static const struct {
352 {SD_RES_SUCCESS, "Success"},
353 {SD_RES_UNKNOWN, "Unknown error"},
354 {SD_RES_NO_OBJ, "No object found"},
355 {SD_RES_EIO, "I/O error"},
356 {SD_RES_VDI_EXIST, "VDI exists already"},
357 {SD_RES_INVALID_PARMS, "Invalid parameters"},
358 {SD_RES_SYSTEM_ERROR, "System error"},
359 {SD_RES_VDI_LOCKED, "VDI is already locked"},
360 {SD_RES_NO_VDI, "No vdi found"},
361 {SD_RES_NO_BASE_VDI, "No base VDI found"},
362 {SD_RES_VDI_READ, "Failed read the requested VDI"},
363 {SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
364 {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
365 {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
366 {SD_RES_NO_TAG, "Failed to find the requested tag"},
367 {SD_RES_STARTUP, "The system is still booting"},
368 {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
369 {SD_RES_SHUTDOWN, "The system is shutting down"},
370 {SD_RES_NO_MEM, "Out of memory on the server"},
371 {SD_RES_FULL_VDI, "We already have the maximum vdis"},
372 {SD_RES_VER_MISMATCH, "Protocol version mismatch"},
373 {SD_RES_NO_SPACE, "Server has no space for new objects"},
374 {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
375 {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
376 {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
377 {SD_RES_HALT, "Sheepdog is stopped serving IO request"},
378 {SD_RES_READONLY, "Object is read-only"},
381 for (i = 0; i < ARRAY_SIZE(errors); ++i) {
382 if (errors[i].err == err) {
383 return errors[i].desc;
387 return "Invalid error code";
391 * Sheepdog I/O handling:
393 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
394 * link the requests to the inflight_list in the
395 * BDRVSheepdogState. The function exits without waiting for
396 * receiving the response.
398 * 2. We receive the response in aio_read_response, the fd handler to
399 * the sheepdog connection. If metadata update is needed, we send
400 * the write request to the vdi object in sd_write_done, the write
401 * completion function. We switch back to sd_co_readv/writev after
402 * all the requests belonging to the AIOCB are finished.
405 static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
406 uint64_t oid, unsigned int data_len,
407 uint64_t offset, uint8_t flags,
408 uint64_t base_oid, unsigned int iov_offset)
412 aio_req = g_malloc(sizeof(*aio_req));
413 aio_req->aiocb = acb;
414 aio_req->iov_offset = iov_offset;
416 aio_req->base_oid = base_oid;
417 aio_req->offset = offset;
418 aio_req->data_len = data_len;
419 aio_req->flags = flags;
420 aio_req->id = s->aioreq_seq_num++;
426 static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req)
428 SheepdogAIOCB *acb = aio_req->aiocb;
430 acb->cancelable = false;
431 QLIST_REMOVE(aio_req, aio_siblings);
437 static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb)
439 qemu_coroutine_enter(acb->coroutine, NULL);
441 *acb->finished = true;
443 qemu_aio_release(acb);
447 * Check whether the specified acb can be canceled
449 * We can cancel aio when any request belonging to the acb is:
450 * - Not processed by the sheepdog server.
451 * - Not linked to the inflight queue.
453 static bool sd_acb_cancelable(const SheepdogAIOCB *acb)
455 BDRVSheepdogState *s = acb->common.bs->opaque;
458 if (!acb->cancelable) {
462 QLIST_FOREACH(aioreq, &s->inflight_aio_head, aio_siblings) {
463 if (aioreq->aiocb == acb) {
471 static void sd_aio_cancel(BlockDriverAIOCB *blockacb)
473 SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb;
474 BDRVSheepdogState *s = acb->common.bs->opaque;
475 AIOReq *aioreq, *next;
476 bool finished = false;
478 acb->finished = &finished;
480 if (sd_acb_cancelable(acb)) {
481 /* Remove outstanding requests from pending and failed queues. */
482 QLIST_FOREACH_SAFE(aioreq, &s->pending_aio_head, aio_siblings,
484 if (aioreq->aiocb == acb) {
485 free_aio_req(s, aioreq);
488 QLIST_FOREACH_SAFE(aioreq, &s->failed_aio_head, aio_siblings,
490 if (aioreq->aiocb == acb) {
491 free_aio_req(s, aioreq);
495 assert(acb->nr_pending == 0);
496 sd_finish_aiocb(acb);
503 static const AIOCBInfo sd_aiocb_info = {
504 .aiocb_size = sizeof(SheepdogAIOCB),
505 .cancel = sd_aio_cancel,
508 static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov,
509 int64_t sector_num, int nb_sectors)
513 acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL);
517 acb->sector_num = sector_num;
518 acb->nb_sectors = nb_sectors;
520 acb->aio_done_func = NULL;
521 acb->cancelable = true;
522 acb->finished = NULL;
523 acb->coroutine = qemu_coroutine_self();
529 static int connect_to_sdog(BDRVSheepdogState *s, Error **errp)
534 fd = unix_connect(s->host_spec, errp);
536 fd = inet_connect(s->host_spec, errp);
539 int ret = socket_set_nodelay(fd);
541 error_report("%s", strerror(errno));
547 qemu_set_nonblock(fd);
553 static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
558 ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
559 if (ret != sizeof(*hdr)) {
560 error_report("failed to send a req, %s", strerror(errno));
564 ret = qemu_co_send(sockfd, data, *wlen);
566 error_report("failed to send a req, %s", strerror(errno));
572 static void restart_co_req(void *opaque)
574 Coroutine *co = opaque;
576 qemu_coroutine_enter(co, NULL);
579 typedef struct SheepdogReqCo {
589 static coroutine_fn void do_co_req(void *opaque)
593 SheepdogReqCo *srco = opaque;
594 int sockfd = srco->sockfd;
595 SheepdogReq *hdr = srco->hdr;
596 void *data = srco->data;
597 unsigned int *wlen = srco->wlen;
598 unsigned int *rlen = srco->rlen;
600 co = qemu_coroutine_self();
601 qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, co);
603 ret = send_co_req(sockfd, hdr, data, wlen);
608 qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, co);
610 ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
611 if (ret != sizeof(*hdr)) {
612 error_report("failed to get a rsp, %s", strerror(errno));
617 if (*rlen > hdr->data_length) {
618 *rlen = hdr->data_length;
622 ret = qemu_co_recv(sockfd, data, *rlen);
624 error_report("failed to get the data, %s", strerror(errno));
631 /* there is at most one request for this sockfd, so it is safe to
632 * set each handler to NULL. */
633 qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL);
636 srco->finished = true;
639 static int do_req(int sockfd, SheepdogReq *hdr, void *data,
640 unsigned int *wlen, unsigned int *rlen)
643 SheepdogReqCo srco = {
653 if (qemu_in_coroutine()) {
656 co = qemu_coroutine_create(do_co_req);
657 qemu_coroutine_enter(co, &srco);
658 while (!srco.finished) {
666 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
667 struct iovec *iov, int niov, bool create,
668 enum AIOCBState aiocb_type);
669 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req);
670 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag);
671 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp);
672 static void co_write_request(void *opaque);
674 static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid)
678 QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) {
679 if (aio_req->oid == oid) {
688 * This function searchs pending requests to the object `oid', and
691 static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid)
696 while ((aio_req = find_pending_req(s, oid)) != NULL) {
697 acb = aio_req->aiocb;
698 /* move aio_req from pending list to inflight one */
699 QLIST_REMOVE(aio_req, aio_siblings);
700 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
701 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, false,
706 static coroutine_fn void reconnect_to_sdog(void *opaque)
708 Error *local_err = NULL;
709 BDRVSheepdogState *s = opaque;
710 AIOReq *aio_req, *next;
712 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL);
716 /* Wait for outstanding write requests to be completed. */
717 while (s->co_send != NULL) {
718 co_write_request(opaque);
721 /* Try to reconnect the sheepdog server every one second. */
723 s->fd = get_sheep_fd(s, &local_err);
725 DPRINTF("Wait for connection to be established\n");
726 qerror_report_err(local_err);
727 error_free(local_err);
728 co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME,
734 * Now we have to resend all the request in the inflight queue. However,
735 * resend_aioreq() can yield and newly created requests can be added to the
736 * inflight queue before the coroutine is resumed. To avoid mixing them, we
737 * have to move all the inflight requests to the failed queue before
738 * resend_aioreq() is called.
740 QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) {
741 QLIST_REMOVE(aio_req, aio_siblings);
742 QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings);
745 /* Resend all the failed aio requests. */
746 while (!QLIST_EMPTY(&s->failed_aio_head)) {
747 aio_req = QLIST_FIRST(&s->failed_aio_head);
748 QLIST_REMOVE(aio_req, aio_siblings);
749 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
750 resend_aioreq(s, aio_req);
755 * Receive responses of the I/O requests.
757 * This function is registered as a fd handler, and called from the
758 * main loop when s->fd is ready for reading responses.
760 static void coroutine_fn aio_read_response(void *opaque)
763 BDRVSheepdogState *s = opaque;
766 AIOReq *aio_req = NULL;
771 ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
772 if (ret != sizeof(rsp)) {
773 error_report("failed to get the header, %s", strerror(errno));
777 /* find the right aio_req from the inflight aio list */
778 QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
779 if (aio_req->id == rsp.id) {
784 error_report("cannot find aio_req %x", rsp.id);
788 acb = aio_req->aiocb;
790 switch (acb->aiocb_type) {
791 case AIOCB_WRITE_UDATA:
792 /* this coroutine context is no longer suitable for co_recv
793 * because we may send data to update vdi objects */
795 if (!is_data_obj(aio_req->oid)) {
798 idx = data_oid_to_idx(aio_req->oid);
800 if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) {
802 * If the object is newly created one, we need to update
803 * the vdi object (metadata object). min_dirty_data_idx
804 * and max_dirty_data_idx are changed to include updated
805 * index between them.
807 if (rsp.result == SD_RES_SUCCESS) {
808 s->inode.data_vdi_id[idx] = s->inode.vdi_id;
809 s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx);
810 s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx);
813 * Some requests may be blocked because simultaneous
814 * create requests are not allowed, so we search the
815 * pending requests here.
817 send_pending_req(s, aio_req->oid);
820 case AIOCB_READ_UDATA:
821 ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
822 aio_req->iov_offset, rsp.data_length);
823 if (ret != rsp.data_length) {
824 error_report("failed to get the data, %s", strerror(errno));
828 case AIOCB_FLUSH_CACHE:
829 if (rsp.result == SD_RES_INVALID_PARMS) {
830 DPRINTF("disable cache since the server doesn't support it\n");
831 s->cache_flags = SD_FLAG_CMD_DIRECT;
832 rsp.result = SD_RES_SUCCESS;
835 case AIOCB_DISCARD_OBJ:
836 switch (rsp.result) {
837 case SD_RES_INVALID_PARMS:
838 error_report("sheep(%s) doesn't support discard command",
840 rsp.result = SD_RES_SUCCESS;
841 s->discard_supported = false;
844 idx = data_oid_to_idx(aio_req->oid);
845 s->inode.data_vdi_id[idx] = 0;
852 switch (rsp.result) {
855 case SD_RES_READONLY:
856 if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) {
857 ret = reload_inode(s, 0, "");
862 if (is_data_obj(aio_req->oid)) {
863 aio_req->oid = vid_to_data_oid(s->inode.vdi_id,
864 data_oid_to_idx(aio_req->oid));
866 aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id);
868 resend_aioreq(s, aio_req);
872 error_report("%s", sd_strerror(rsp.result));
876 free_aio_req(s, aio_req);
877 if (!acb->nr_pending) {
879 * We've finished all requests which belong to the AIOCB, so
880 * we can switch back to sd_co_readv/writev now.
882 acb->aio_done_func(acb);
889 reconnect_to_sdog(opaque);
892 static void co_read_response(void *opaque)
894 BDRVSheepdogState *s = opaque;
897 s->co_recv = qemu_coroutine_create(aio_read_response);
900 qemu_coroutine_enter(s->co_recv, opaque);
903 static void co_write_request(void *opaque)
905 BDRVSheepdogState *s = opaque;
907 qemu_coroutine_enter(s->co_send, NULL);
911 * Return a socket descriptor to read/write objects.
913 * We cannot use this descriptor for other operations because
914 * the block driver may be on waiting response from the server.
916 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp)
920 fd = connect_to_sdog(s, errp);
925 qemu_aio_set_fd_handler(fd, co_read_response, NULL, s);
929 static int sd_parse_uri(BDRVSheepdogState *s, const char *filename,
930 char *vdi, uint32_t *snapid, char *tag)
933 QueryParams *qp = NULL;
936 uri = uri_parse(filename);
942 if (!strcmp(uri->scheme, "sheepdog")) {
944 } else if (!strcmp(uri->scheme, "sheepdog+tcp")) {
946 } else if (!strcmp(uri->scheme, "sheepdog+unix")) {
953 if (uri->path == NULL || !strcmp(uri->path, "/")) {
957 pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1);
959 qp = query_params_parse(uri->query);
960 if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) {
966 /* sheepdog+unix:///vdiname?socket=path */
967 if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
971 s->host_spec = g_strdup(qp->p[0].value);
973 /* sheepdog[+tcp]://[host:port]/vdiname */
974 s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR,
975 uri->port ?: SD_DEFAULT_PORT);
980 *snapid = strtoul(uri->fragment, NULL, 10);
982 pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment);
985 *snapid = CURRENT_VDI_ID; /* search current vdi */
990 query_params_free(qp);
997 * Parse a filename (old syntax)
999 * filename must be one of the following formats:
1001 * 2. [vdiname]:[snapid]
1002 * 3. [vdiname]:[tag]
1003 * 4. [hostname]:[port]:[vdiname]
1004 * 5. [hostname]:[port]:[vdiname]:[snapid]
1005 * 6. [hostname]:[port]:[vdiname]:[tag]
1007 * You can boot from the snapshot images by specifying `snapid` or
1010 * You can run VMs outside the Sheepdog cluster by specifying
1011 * `hostname' and `port' (experimental).
1013 static int parse_vdiname(BDRVSheepdogState *s, const char *filename,
1014 char *vdi, uint32_t *snapid, char *tag)
1017 const char *host_spec, *vdi_spec;
1020 strstart(filename, "sheepdog:", (const char **)&filename);
1021 p = q = g_strdup(filename);
1023 /* count the number of separators */
1033 /* use the first two tokens as host_spec. */
1046 p = strchr(vdi_spec, ':');
1051 uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec);
1053 ret = sd_parse_uri(s, uri, vdi, snapid, tag);
1061 static int find_vdi_name(BDRVSheepdogState *s, const char *filename,
1062 uint32_t snapid, const char *tag, uint32_t *vid,
1065 Error *local_err = NULL;
1068 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1069 unsigned int wlen, rlen = 0;
1070 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
1072 fd = connect_to_sdog(s, &local_err);
1074 qerror_report_err(local_err);
1075 error_free(local_err);
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;
1090 hdr.opcode = SD_OP_GET_VDI_INFO;
1092 wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
1093 hdr.proto_ver = SD_PROTO_VER;
1094 hdr.data_length = wlen;
1095 hdr.snapid = snapid;
1096 hdr.flags = SD_FLAG_CMD_WRITE;
1098 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1103 if (rsp->result != SD_RES_SUCCESS) {
1104 error_report("cannot get vdi info, %s, %s %" PRIu32 " %s",
1105 sd_strerror(rsp->result), filename, snapid, tag);
1106 if (rsp->result == SD_RES_NO_VDI) {
1121 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
1122 struct iovec *iov, int niov, bool create,
1123 enum AIOCBState aiocb_type)
1125 int nr_copies = s->inode.nr_copies;
1127 unsigned int wlen = 0;
1129 uint64_t oid = aio_req->oid;
1130 unsigned int datalen = aio_req->data_len;
1131 uint64_t offset = aio_req->offset;
1132 uint8_t flags = aio_req->flags;
1133 uint64_t old_oid = aio_req->base_oid;
1136 error_report("bug");
1139 memset(&hdr, 0, sizeof(hdr));
1141 switch (aiocb_type) {
1142 case AIOCB_FLUSH_CACHE:
1143 hdr.opcode = SD_OP_FLUSH_VDI;
1145 case AIOCB_READ_UDATA:
1146 hdr.opcode = SD_OP_READ_OBJ;
1149 case AIOCB_WRITE_UDATA:
1151 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1153 hdr.opcode = SD_OP_WRITE_OBJ;
1156 hdr.flags = SD_FLAG_CMD_WRITE | flags;
1158 case AIOCB_DISCARD_OBJ:
1159 hdr.opcode = SD_OP_DISCARD_OBJ;
1163 if (s->cache_flags) {
1164 hdr.flags |= s->cache_flags;
1168 hdr.cow_oid = old_oid;
1169 hdr.copies = s->inode.nr_copies;
1171 hdr.data_length = datalen;
1172 hdr.offset = offset;
1174 hdr.id = aio_req->id;
1176 qemu_co_mutex_lock(&s->lock);
1177 s->co_send = qemu_coroutine_self();
1178 qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, s);
1179 socket_set_cork(s->fd, 1);
1182 ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
1183 if (ret != sizeof(hdr)) {
1184 error_report("failed to send a req, %s", strerror(errno));
1189 ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1191 error_report("failed to send a data, %s", strerror(errno));
1195 socket_set_cork(s->fd, 0);
1196 qemu_aio_set_fd_handler(s->fd, co_read_response, NULL, s);
1198 qemu_co_mutex_unlock(&s->lock);
1201 static int read_write_object(int fd, char *buf, uint64_t oid, uint8_t copies,
1202 unsigned int datalen, uint64_t offset,
1203 bool write, bool create, uint32_t cache_flags)
1206 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1207 unsigned int wlen, rlen;
1210 memset(&hdr, 0, sizeof(hdr));
1215 hdr.flags = SD_FLAG_CMD_WRITE;
1217 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1219 hdr.opcode = SD_OP_WRITE_OBJ;
1224 hdr.opcode = SD_OP_READ_OBJ;
1227 hdr.flags |= cache_flags;
1230 hdr.data_length = datalen;
1231 hdr.offset = offset;
1232 hdr.copies = copies;
1234 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1236 error_report("failed to send a request to the sheep");
1240 switch (rsp->result) {
1241 case SD_RES_SUCCESS:
1244 error_report("%s", sd_strerror(rsp->result));
1249 static int read_object(int fd, char *buf, uint64_t oid, uint8_t copies,
1250 unsigned int datalen, uint64_t offset,
1251 uint32_t cache_flags)
1253 return read_write_object(fd, buf, oid, copies, datalen, offset, false,
1254 false, cache_flags);
1257 static int write_object(int fd, char *buf, uint64_t oid, uint8_t copies,
1258 unsigned int datalen, uint64_t offset, bool create,
1259 uint32_t cache_flags)
1261 return read_write_object(fd, buf, oid, copies, datalen, offset, true,
1262 create, cache_flags);
1265 /* update inode with the latest state */
1266 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
1268 Error *local_err = NULL;
1269 SheepdogInode *inode;
1273 fd = connect_to_sdog(s, &local_err);
1275 qerror_report_err(local_err);
1276 error_free(local_err);
1280 inode = g_malloc(sizeof(s->inode));
1282 ret = find_vdi_name(s, s->name, snapid, tag, &vid, false);
1287 ret = read_object(fd, (char *)inode, vid_to_vdi_oid(vid),
1288 s->inode.nr_copies, sizeof(*inode), 0, s->cache_flags);
1293 if (inode->vdi_id != s->inode.vdi_id) {
1294 memcpy(&s->inode, inode, sizeof(s->inode));
1304 /* Return true if the specified request is linked to the pending list. */
1305 static bool check_simultaneous_create(BDRVSheepdogState *s, AIOReq *aio_req)
1308 QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) {
1309 if (areq != aio_req && areq->oid == aio_req->oid) {
1311 * Sheepdog cannot handle simultaneous create requests to the same
1312 * object, so we cannot send the request until the previous request
1315 DPRINTF("simultaneous create to %" PRIx64 "\n", aio_req->oid);
1317 aio_req->base_oid = 0;
1318 QLIST_REMOVE(aio_req, aio_siblings);
1319 QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, aio_siblings);
1327 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
1329 SheepdogAIOCB *acb = aio_req->aiocb;
1330 bool create = false;
1332 /* check whether this request becomes a CoW one */
1333 if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) {
1334 int idx = data_oid_to_idx(aio_req->oid);
1336 if (is_data_obj_writable(&s->inode, idx)) {
1340 if (check_simultaneous_create(s, aio_req)) {
1344 if (s->inode.data_vdi_id[idx]) {
1345 aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx);
1346 aio_req->flags |= SD_FLAG_CMD_COW;
1351 if (is_data_obj(aio_req->oid)) {
1352 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, create,
1356 iov.iov_base = &s->inode;
1357 iov.iov_len = sizeof(s->inode);
1358 add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA);
1362 /* TODO Convert to fine grained options */
1363 static QemuOptsList runtime_opts = {
1365 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1369 .type = QEMU_OPT_STRING,
1370 .help = "URL to the sheepdog image",
1372 { /* end of list */ }
1376 static int sd_open(BlockDriverState *bs, QDict *options, int flags,
1381 BDRVSheepdogState *s = bs->opaque;
1382 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1386 Error *local_err = NULL;
1387 const char *filename;
1391 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1392 qemu_opts_absorb_qdict(opts, options, &local_err);
1394 qerror_report_err(local_err);
1395 error_free(local_err);
1400 filename = qemu_opt_get(opts, "filename");
1402 QLIST_INIT(&s->inflight_aio_head);
1403 QLIST_INIT(&s->pending_aio_head);
1404 QLIST_INIT(&s->failed_aio_head);
1407 memset(vdi, 0, sizeof(vdi));
1408 memset(tag, 0, sizeof(tag));
1410 if (strstr(filename, "://")) {
1411 ret = sd_parse_uri(s, filename, vdi, &snapid, tag);
1413 ret = parse_vdiname(s, filename, vdi, &snapid, tag);
1418 s->fd = get_sheep_fd(s, &local_err);
1420 qerror_report_err(local_err);
1421 error_free(local_err);
1426 ret = find_vdi_name(s, vdi, snapid, tag, &vid, true);
1432 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1433 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1435 s->cache_flags = SD_FLAG_CMD_CACHE;
1436 if (flags & BDRV_O_NOCACHE) {
1437 s->cache_flags = SD_FLAG_CMD_DIRECT;
1439 s->discard_supported = true;
1441 if (snapid || tag[0] != '\0') {
1442 DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid);
1443 s->is_snapshot = true;
1446 fd = connect_to_sdog(s, &local_err);
1448 qerror_report_err(local_err);
1449 error_free(local_err);
1454 buf = g_malloc(SD_INODE_SIZE);
1455 ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0,
1464 memcpy(&s->inode, buf, sizeof(s->inode));
1465 s->min_dirty_data_idx = UINT32_MAX;
1466 s->max_dirty_data_idx = 0;
1468 bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE;
1469 pstrcpy(s->name, sizeof(s->name), vdi);
1470 qemu_co_mutex_init(&s->lock);
1471 qemu_opts_del(opts);
1475 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL);
1479 qemu_opts_del(opts);
1484 static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot)
1486 Error *local_err = NULL;
1488 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1490 unsigned int wlen, rlen = 0;
1491 char buf[SD_MAX_VDI_LEN];
1493 fd = connect_to_sdog(s, &local_err);
1495 qerror_report_err(local_err);
1496 error_free(local_err);
1500 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1501 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1503 memset(buf, 0, sizeof(buf));
1504 pstrcpy(buf, sizeof(buf), s->name);
1506 memset(&hdr, 0, sizeof(hdr));
1507 hdr.opcode = SD_OP_NEW_VDI;
1508 hdr.base_vdi_id = s->inode.vdi_id;
1510 wlen = SD_MAX_VDI_LEN;
1512 hdr.flags = SD_FLAG_CMD_WRITE;
1513 hdr.snapid = snapshot;
1515 hdr.data_length = wlen;
1516 hdr.vdi_size = s->inode.vdi_size;
1517 hdr.copy_policy = s->inode.copy_policy;
1518 hdr.copies = s->inode.nr_copies;
1520 ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1528 if (rsp->result != SD_RES_SUCCESS) {
1529 error_report("%s, %s", sd_strerror(rsp->result), s->inode.name);
1534 *vdi_id = rsp->vdi_id;
1540 static int sd_prealloc(const char *filename, Error **errp)
1542 BlockDriverState *bs = NULL;
1543 uint32_t idx, max_idx;
1545 void *buf = g_malloc0(SD_DATA_OBJ_SIZE);
1548 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1551 goto out_with_err_set;
1554 vdi_size = bdrv_getlength(bs);
1559 max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE);
1561 for (idx = 0; idx < max_idx; idx++) {
1563 * The created image can be a cloned image, so we need to read
1564 * a data from the source image.
1566 ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1570 ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE);
1578 error_setg_errno(errp, -ret, "Can't pre-allocate");
1590 * Sheepdog support two kinds of redundancy, full replication and erasure
1593 * # create a fully replicated vdi with x copies
1594 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1596 * # create a erasure coded vdi with x data strips and y parity strips
1597 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1599 static int parse_redundancy(BDRVSheepdogState *s, const char *opt)
1601 struct SheepdogInode *inode = &s->inode;
1602 const char *n1, *n2;
1606 pstrcpy(p, sizeof(p), opt);
1607 n1 = strtok(p, ":");
1608 n2 = strtok(NULL, ":");
1614 copy = strtol(n1, NULL, 10);
1615 if (copy > SD_MAX_COPIES || copy < 1) {
1619 inode->copy_policy = 0;
1620 inode->nr_copies = copy;
1624 if (copy != 2 && copy != 4 && copy != 8 && copy != 16) {
1628 parity = strtol(n2, NULL, 10);
1629 if (parity >= SD_EC_MAX_STRIP || parity < 1) {
1634 * 4 bits for parity and 4 bits for data.
1635 * We have to compress upper data bits because it can't represent 16
1637 inode->copy_policy = ((copy / 2) << 4) + parity;
1638 inode->nr_copies = copy + parity;
1643 static int sd_create(const char *filename, QEMUOptionParameter *options,
1648 char *backing_file = NULL;
1649 BDRVSheepdogState *s;
1650 char tag[SD_MAX_VDI_TAG_LEN];
1652 bool prealloc = false;
1653 Error *local_err = NULL;
1655 s = g_malloc0(sizeof(BDRVSheepdogState));
1657 memset(tag, 0, sizeof(tag));
1658 if (strstr(filename, "://")) {
1659 ret = sd_parse_uri(s, filename, s->name, &snapid, tag);
1661 ret = parse_vdiname(s, filename, s->name, &snapid, tag);
1667 while (options && options->name) {
1668 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1669 s->inode.vdi_size = options->value.n;
1670 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1671 backing_file = options->value.s;
1672 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1673 if (!options->value.s || !strcmp(options->value.s, "off")) {
1675 } else if (!strcmp(options->value.s, "full")) {
1678 error_report("Invalid preallocation mode: '%s'",
1683 } else if (!strcmp(options->name, BLOCK_OPT_REDUNDANCY)) {
1684 if (options->value.s) {
1685 ret = parse_redundancy(s, options->value.s);
1694 if (s->inode.vdi_size > SD_MAX_VDI_SIZE) {
1695 error_report("too big image size");
1701 BlockDriverState *bs;
1702 BDRVSheepdogState *base;
1705 /* Currently, only Sheepdog backing image is supported. */
1706 drv = bdrv_find_protocol(backing_file, true);
1707 if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
1708 error_report("backing_file must be a sheepdog image");
1714 ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_PROTOCOL, NULL,
1717 qerror_report_err(local_err);
1718 error_free(local_err);
1724 if (!is_snapshot(&base->inode)) {
1725 error_report("cannot clone from a non snapshot vdi");
1730 s->inode.vdi_id = base->inode.vdi_id;
1734 ret = do_sd_create(s, &vid, 0);
1735 if (!prealloc || ret) {
1739 ret = sd_prealloc(filename, &local_err);
1741 qerror_report_err(local_err);
1742 error_free(local_err);
1749 static void sd_close(BlockDriverState *bs)
1751 Error *local_err = NULL;
1752 BDRVSheepdogState *s = bs->opaque;
1754 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1755 unsigned int wlen, rlen = 0;
1758 DPRINTF("%s\n", s->name);
1760 fd = connect_to_sdog(s, &local_err);
1762 qerror_report_err(local_err);
1763 error_free(local_err);
1767 memset(&hdr, 0, sizeof(hdr));
1769 hdr.opcode = SD_OP_RELEASE_VDI;
1770 hdr.base_vdi_id = s->inode.vdi_id;
1771 wlen = strlen(s->name) + 1;
1772 hdr.data_length = wlen;
1773 hdr.flags = SD_FLAG_CMD_WRITE;
1775 ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen);
1779 if (!ret && rsp->result != SD_RES_SUCCESS &&
1780 rsp->result != SD_RES_VDI_NOT_LOCKED) {
1781 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1784 qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL);
1786 g_free(s->host_spec);
1789 static int64_t sd_getlength(BlockDriverState *bs)
1791 BDRVSheepdogState *s = bs->opaque;
1793 return s->inode.vdi_size;
1796 static int sd_truncate(BlockDriverState *bs, int64_t offset)
1798 Error *local_err = NULL;
1799 BDRVSheepdogState *s = bs->opaque;
1801 unsigned int datalen;
1803 if (offset < s->inode.vdi_size) {
1804 error_report("shrinking is not supported");
1806 } else if (offset > SD_MAX_VDI_SIZE) {
1807 error_report("too big image size");
1811 fd = connect_to_sdog(s, &local_err);
1813 qerror_report_err(local_err);
1814 error_free(local_err);
1818 /* we don't need to update entire object */
1819 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1820 s->inode.vdi_size = offset;
1821 ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
1822 s->inode.nr_copies, datalen, 0, false, s->cache_flags);
1826 error_report("failed to update an inode.");
1833 * This function is called after writing data objects. If we need to
1834 * update metadata, this sends a write request to the vdi object.
1835 * Otherwise, this switches back to sd_co_readv/writev.
1837 static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
1839 BDRVSheepdogState *s = acb->common.bs->opaque;
1842 uint32_t offset, data_len, mn, mx;
1844 mn = s->min_dirty_data_idx;
1845 mx = s->max_dirty_data_idx;
1847 /* we need to update the vdi object. */
1848 offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
1849 mn * sizeof(s->inode.data_vdi_id[0]);
1850 data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
1852 s->min_dirty_data_idx = UINT32_MAX;
1853 s->max_dirty_data_idx = 0;
1855 iov.iov_base = &s->inode;
1856 iov.iov_len = sizeof(s->inode);
1857 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
1858 data_len, offset, 0, 0, offset);
1859 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1860 add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA);
1862 acb->aio_done_func = sd_finish_aiocb;
1863 acb->aiocb_type = AIOCB_WRITE_UDATA;
1867 sd_finish_aiocb(acb);
1870 /* Delete current working VDI on the snapshot chain */
1871 static bool sd_delete(BDRVSheepdogState *s)
1873 Error *local_err = NULL;
1874 unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0;
1875 SheepdogVdiReq hdr = {
1876 .opcode = SD_OP_DEL_VDI,
1877 .base_vdi_id = s->inode.vdi_id,
1878 .data_length = wlen,
1879 .flags = SD_FLAG_CMD_WRITE,
1881 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1884 fd = connect_to_sdog(s, &local_err);
1886 qerror_report_err(local_err);
1887 error_free(local_err);
1891 ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen);
1896 switch (rsp->result) {
1898 error_report("%s was already deleted", s->name);
1900 case SD_RES_SUCCESS:
1903 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1911 * Create a writable VDI from a snapshot
1913 static int sd_create_branch(BDRVSheepdogState *s)
1915 Error *local_err = NULL;
1921 DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id);
1923 buf = g_malloc(SD_INODE_SIZE);
1926 * Even If deletion fails, we will just create extra snapshot based on
1927 * the working VDI which was supposed to be deleted. So no need to
1930 deleted = sd_delete(s);
1931 ret = do_sd_create(s, &vid, !deleted);
1936 DPRINTF("%" PRIx32 " is created.\n", vid);
1938 fd = connect_to_sdog(s, &local_err);
1940 qerror_report_err(local_err);
1941 error_free(local_err);
1946 ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies,
1947 SD_INODE_SIZE, 0, s->cache_flags);
1955 memcpy(&s->inode, buf, sizeof(s->inode));
1957 s->is_snapshot = false;
1959 DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id);
1968 * Send I/O requests to the server.
1970 * This function sends requests to the server, links the requests to
1971 * the inflight_list in BDRVSheepdogState, and exits without
1972 * waiting the response. The responses are received in the
1973 * `aio_read_response' function which is called from the main loop as
1976 * Returns 1 when we need to wait a response, 0 when there is no sent
1977 * request and -errno in error cases.
1979 static int coroutine_fn sd_co_rw_vector(void *p)
1981 SheepdogAIOCB *acb = p;
1983 unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE;
1984 unsigned long idx = acb->sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE;
1986 uint64_t offset = (acb->sector_num * BDRV_SECTOR_SIZE) % SD_DATA_OBJ_SIZE;
1987 BDRVSheepdogState *s = acb->common.bs->opaque;
1988 SheepdogInode *inode = &s->inode;
1991 if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
1993 * In the case we open the snapshot VDI, Sheepdog creates the
1994 * writable VDI when we do a write operation first.
1996 ret = sd_create_branch(s);
2004 * Make sure we don't free the aiocb before we are done with all requests.
2005 * This additional reference is dropped at the end of this function.
2009 while (done != total) {
2011 uint64_t old_oid = 0;
2012 bool create = false;
2014 oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
2016 len = MIN(total - done, SD_DATA_OBJ_SIZE - offset);
2018 switch (acb->aiocb_type) {
2019 case AIOCB_READ_UDATA:
2020 if (!inode->data_vdi_id[idx]) {
2021 qemu_iovec_memset(acb->qiov, done, 0, len);
2025 case AIOCB_WRITE_UDATA:
2026 if (!inode->data_vdi_id[idx]) {
2028 } else if (!is_data_obj_writable(inode, idx)) {
2032 flags = SD_FLAG_CMD_COW;
2035 case AIOCB_DISCARD_OBJ:
2037 * We discard the object only when the whole object is
2038 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2040 if (len != SD_DATA_OBJ_SIZE || inode->data_vdi_id[idx] == 0) {
2049 DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n",
2051 vid_to_data_oid(inode->data_vdi_id[idx], idx), idx);
2052 oid = vid_to_data_oid(inode->vdi_id, idx);
2053 DPRINTF("new oid %" PRIx64 "\n", oid);
2056 aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done);
2057 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
2060 if (check_simultaneous_create(s, aio_req)) {
2065 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, create,
2073 if (!--acb->nr_pending) {
2079 static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
2080 int nb_sectors, QEMUIOVector *qiov)
2084 int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE;
2085 BDRVSheepdogState *s = bs->opaque;
2087 if (bs->growable && offset > s->inode.vdi_size) {
2088 ret = sd_truncate(bs, offset);
2094 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors);
2095 acb->aio_done_func = sd_write_done;
2096 acb->aiocb_type = AIOCB_WRITE_UDATA;
2098 ret = sd_co_rw_vector(acb);
2100 qemu_aio_release(acb);
2104 qemu_coroutine_yield();
2109 static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
2110 int nb_sectors, QEMUIOVector *qiov)
2115 acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors);
2116 acb->aiocb_type = AIOCB_READ_UDATA;
2117 acb->aio_done_func = sd_finish_aiocb;
2119 ret = sd_co_rw_vector(acb);
2121 qemu_aio_release(acb);
2125 qemu_coroutine_yield();
2130 static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
2132 BDRVSheepdogState *s = bs->opaque;
2136 if (s->cache_flags != SD_FLAG_CMD_CACHE) {
2140 acb = sd_aio_setup(bs, NULL, 0, 0);
2141 acb->aiocb_type = AIOCB_FLUSH_CACHE;
2142 acb->aio_done_func = sd_finish_aiocb;
2144 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
2146 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
2147 add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type);
2149 qemu_coroutine_yield();
2153 static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
2155 Error *local_err = NULL;
2156 BDRVSheepdogState *s = bs->opaque;
2159 SheepdogInode *inode;
2160 unsigned int datalen;
2162 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " "
2163 "is_snapshot %d\n", sn_info->name, sn_info->id_str,
2164 s->name, sn_info->vm_state_size, s->is_snapshot);
2166 if (s->is_snapshot) {
2167 error_report("You can't create a snapshot of a snapshot VDI, "
2168 "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
2173 DPRINTF("%s %s\n", sn_info->name, sn_info->id_str);
2175 s->inode.vm_state_size = sn_info->vm_state_size;
2176 s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
2177 /* It appears that inode.tag does not require a NUL terminator,
2178 * which means this use of strncpy is ok.
2180 strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
2181 /* we don't need to update entire object */
2182 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
2184 /* refresh inode. */
2185 fd = connect_to_sdog(s, &local_err);
2187 qerror_report_err(local_err);
2188 error_free(local_err);
2193 ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
2194 s->inode.nr_copies, datalen, 0, false, s->cache_flags);
2196 error_report("failed to write snapshot's inode.");
2200 ret = do_sd_create(s, &new_vid, 1);
2202 error_report("failed to create inode for snapshot. %s",
2207 inode = (SheepdogInode *)g_malloc(datalen);
2209 ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid),
2210 s->inode.nr_copies, datalen, 0, s->cache_flags);
2213 error_report("failed to read new inode info. %s", strerror(errno));
2217 memcpy(&s->inode, inode, datalen);
2218 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2219 s->inode.name, s->inode.snap_id, s->inode.vdi_id);
2227 * We implement rollback(loadvm) operation to the specified snapshot by
2228 * 1) switch to the snapshot
2229 * 2) rely on sd_create_branch to delete working VDI and
2230 * 3) create a new working VDI based on the specified snapshot
2232 static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
2234 BDRVSheepdogState *s = bs->opaque;
2235 BDRVSheepdogState *old_s;
2236 char tag[SD_MAX_VDI_TAG_LEN];
2237 uint32_t snapid = 0;
2240 old_s = g_malloc(sizeof(BDRVSheepdogState));
2242 memcpy(old_s, s, sizeof(BDRVSheepdogState));
2244 snapid = strtoul(snapshot_id, NULL, 10);
2248 pstrcpy(tag, sizeof(tag), snapshot_id);
2251 ret = reload_inode(s, snapid, tag);
2256 ret = sd_create_branch(s);
2265 /* recover bdrv_sd_state */
2266 memcpy(s, old_s, sizeof(BDRVSheepdogState));
2269 error_report("failed to open. recover old bdrv_sd_state.");
2274 static int sd_snapshot_delete(BlockDriverState *bs,
2275 const char *snapshot_id,
2279 /* FIXME: Delete specified snapshot id. */
2283 static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
2285 Error *local_err = NULL;
2286 BDRVSheepdogState *s = bs->opaque;
2288 int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
2289 QEMUSnapshotInfo *sn_tab = NULL;
2290 unsigned wlen, rlen;
2292 static SheepdogInode inode;
2293 unsigned long *vdi_inuse;
2294 unsigned int start_nr;
2298 vdi_inuse = g_malloc(max);
2300 fd = connect_to_sdog(s, &local_err);
2302 qerror_report_err(local_err);
2303 error_free(local_err);
2311 memset(&req, 0, sizeof(req));
2313 req.opcode = SD_OP_READ_VDIS;
2314 req.data_length = max;
2316 ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen);
2323 sn_tab = g_malloc0(nr * sizeof(*sn_tab));
2325 /* calculate a vdi id with hash function */
2326 hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
2327 start_nr = hval & (SD_NR_VDIS - 1);
2329 fd = connect_to_sdog(s, &local_err);
2331 qerror_report_err(local_err);
2332 error_free(local_err);
2337 for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
2338 if (!test_bit(vid, vdi_inuse)) {
2342 /* we don't need to read entire object */
2343 ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid),
2344 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0,
2351 if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
2352 sn_tab[found].date_sec = inode.snap_ctime >> 32;
2353 sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
2354 sn_tab[found].vm_state_size = inode.vm_state_size;
2355 sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec;
2357 snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str),
2358 "%" PRIu32, inode.snap_id);
2359 pstrcpy(sn_tab[found].name,
2360 MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)),
2379 static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
2380 int64_t pos, int size, int load)
2382 Error *local_err = NULL;
2384 int fd, ret = 0, remaining = size;
2385 unsigned int data_len;
2386 uint64_t vmstate_oid;
2389 uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id;
2391 fd = connect_to_sdog(s, &local_err);
2393 qerror_report_err(local_err);
2394 error_free(local_err);
2399 vdi_index = pos / SD_DATA_OBJ_SIZE;
2400 offset = pos % SD_DATA_OBJ_SIZE;
2402 data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset);
2404 vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index);
2406 create = (offset == 0);
2408 ret = read_object(fd, (char *)data, vmstate_oid,
2409 s->inode.nr_copies, data_len, offset,
2412 ret = write_object(fd, (char *)data, vmstate_oid,
2413 s->inode.nr_copies, data_len, offset, create,
2418 error_report("failed to save vmstate %s", strerror(errno));
2424 remaining -= data_len;
2432 static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2435 BDRVSheepdogState *s = bs->opaque;
2439 buf = qemu_blockalign(bs, qiov->size);
2440 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
2441 ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0);
2447 static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data,
2448 int64_t pos, int size)
2450 BDRVSheepdogState *s = bs->opaque;
2452 return do_load_save_vmstate(s, data, pos, size, 1);
2456 static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num,
2461 BDRVSheepdogState *s = bs->opaque;
2464 if (!s->discard_supported) {
2468 acb = sd_aio_setup(bs, &dummy, sector_num, nb_sectors);
2469 acb->aiocb_type = AIOCB_DISCARD_OBJ;
2470 acb->aio_done_func = sd_finish_aiocb;
2472 ret = sd_co_rw_vector(acb);
2474 qemu_aio_release(acb);
2478 qemu_coroutine_yield();
2483 static coroutine_fn int64_t
2484 sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2487 BDRVSheepdogState *s = bs->opaque;
2488 SheepdogInode *inode = &s->inode;
2489 uint64_t offset = sector_num * BDRV_SECTOR_SIZE;
2490 unsigned long start = offset / SD_DATA_OBJ_SIZE,
2491 end = DIV_ROUND_UP((sector_num + nb_sectors) *
2492 BDRV_SECTOR_SIZE, SD_DATA_OBJ_SIZE);
2494 int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
2496 for (idx = start; idx < end; idx++) {
2497 if (inode->data_vdi_id[idx] == 0) {
2502 /* Get the longest length of unallocated sectors */
2504 for (idx = start + 1; idx < end; idx++) {
2505 if (inode->data_vdi_id[idx] != 0) {
2511 *pnum = (idx - start) * SD_DATA_OBJ_SIZE / BDRV_SECTOR_SIZE;
2512 if (*pnum > nb_sectors) {
2518 static int64_t sd_get_allocated_file_size(BlockDriverState *bs)
2520 BDRVSheepdogState *s = bs->opaque;
2521 SheepdogInode *inode = &s->inode;
2522 unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, SD_DATA_OBJ_SIZE);
2525 for (i = 0; i < last; i++) {
2526 if (inode->data_vdi_id[i] == 0) {
2529 size += SD_DATA_OBJ_SIZE;
2534 static QEMUOptionParameter sd_create_options[] = {
2536 .name = BLOCK_OPT_SIZE,
2538 .help = "Virtual disk size"
2541 .name = BLOCK_OPT_BACKING_FILE,
2543 .help = "File name of a base image"
2546 .name = BLOCK_OPT_PREALLOC,
2548 .help = "Preallocation mode (allowed values: off, full)"
2551 .name = BLOCK_OPT_REDUNDANCY,
2553 .help = "Redundancy of the image"
2558 static BlockDriver bdrv_sheepdog = {
2559 .format_name = "sheepdog",
2560 .protocol_name = "sheepdog",
2561 .instance_size = sizeof(BDRVSheepdogState),
2562 .bdrv_needs_filename = true,
2563 .bdrv_file_open = sd_open,
2564 .bdrv_close = sd_close,
2565 .bdrv_create = sd_create,
2566 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2567 .bdrv_getlength = sd_getlength,
2568 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2569 .bdrv_truncate = sd_truncate,
2571 .bdrv_co_readv = sd_co_readv,
2572 .bdrv_co_writev = sd_co_writev,
2573 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2574 .bdrv_co_discard = sd_co_discard,
2575 .bdrv_co_get_block_status = sd_co_get_block_status,
2577 .bdrv_snapshot_create = sd_snapshot_create,
2578 .bdrv_snapshot_goto = sd_snapshot_goto,
2579 .bdrv_snapshot_delete = sd_snapshot_delete,
2580 .bdrv_snapshot_list = sd_snapshot_list,
2582 .bdrv_save_vmstate = sd_save_vmstate,
2583 .bdrv_load_vmstate = sd_load_vmstate,
2585 .create_options = sd_create_options,
2588 static BlockDriver bdrv_sheepdog_tcp = {
2589 .format_name = "sheepdog",
2590 .protocol_name = "sheepdog+tcp",
2591 .instance_size = sizeof(BDRVSheepdogState),
2592 .bdrv_needs_filename = true,
2593 .bdrv_file_open = sd_open,
2594 .bdrv_close = sd_close,
2595 .bdrv_create = sd_create,
2596 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2597 .bdrv_getlength = sd_getlength,
2598 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2599 .bdrv_truncate = sd_truncate,
2601 .bdrv_co_readv = sd_co_readv,
2602 .bdrv_co_writev = sd_co_writev,
2603 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2604 .bdrv_co_discard = sd_co_discard,
2605 .bdrv_co_get_block_status = sd_co_get_block_status,
2607 .bdrv_snapshot_create = sd_snapshot_create,
2608 .bdrv_snapshot_goto = sd_snapshot_goto,
2609 .bdrv_snapshot_delete = sd_snapshot_delete,
2610 .bdrv_snapshot_list = sd_snapshot_list,
2612 .bdrv_save_vmstate = sd_save_vmstate,
2613 .bdrv_load_vmstate = sd_load_vmstate,
2615 .create_options = sd_create_options,
2618 static BlockDriver bdrv_sheepdog_unix = {
2619 .format_name = "sheepdog",
2620 .protocol_name = "sheepdog+unix",
2621 .instance_size = sizeof(BDRVSheepdogState),
2622 .bdrv_needs_filename = true,
2623 .bdrv_file_open = sd_open,
2624 .bdrv_close = sd_close,
2625 .bdrv_create = sd_create,
2626 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2627 .bdrv_getlength = sd_getlength,
2628 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2629 .bdrv_truncate = sd_truncate,
2631 .bdrv_co_readv = sd_co_readv,
2632 .bdrv_co_writev = sd_co_writev,
2633 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2634 .bdrv_co_discard = sd_co_discard,
2635 .bdrv_co_get_block_status = sd_co_get_block_status,
2637 .bdrv_snapshot_create = sd_snapshot_create,
2638 .bdrv_snapshot_goto = sd_snapshot_goto,
2639 .bdrv_snapshot_delete = sd_snapshot_delete,
2640 .bdrv_snapshot_list = sd_snapshot_list,
2642 .bdrv_save_vmstate = sd_save_vmstate,
2643 .bdrv_load_vmstate = sd_load_vmstate,
2645 .create_options = sd_create_options,
2648 static void bdrv_sheepdog_init(void)
2650 bdrv_register(&bdrv_sheepdog);
2651 bdrv_register(&bdrv_sheepdog_tcp);
2652 bdrv_register(&bdrv_sheepdog_unix);
2654 block_init(bdrv_sheepdog_init);