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 #define SD_INODE_SIZE (sizeof(SheepdogInode))
96 #define CURRENT_VDI_ID 0
98 typedef struct SheepdogReq
{
104 uint32_t data_length
;
105 uint32_t opcode_specific
[8];
108 typedef struct SheepdogRsp
{
114 uint32_t data_length
;
116 uint32_t opcode_specific
[7];
119 typedef struct SheepdogObjReq
{
125 uint32_t data_length
;
134 typedef struct SheepdogObjRsp
{
140 uint32_t data_length
;
148 typedef struct SheepdogVdiReq
{
154 uint32_t data_length
;
164 typedef struct SheepdogVdiRsp
{
170 uint32_t data_length
;
177 typedef struct SheepdogInode
{
178 char name
[SD_MAX_VDI_LEN
];
179 char tag
[SD_MAX_VDI_TAG_LEN
];
182 uint64_t vm_clock_nsec
;
184 uint64_t vm_state_size
;
185 uint16_t copy_policy
;
187 uint8_t block_size_shift
;
190 uint32_t parent_vdi_id
;
191 uint32_t child_vdi_id
[MAX_CHILDREN
];
192 uint32_t data_vdi_id
[MAX_DATA_OBJS
];
196 * 64 bit FNV-1a non-zero initial basis
198 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
201 * 64 bit Fowler/Noll/Vo FNV-1a hash code
203 static inline uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
205 unsigned char *bp
= buf
;
206 unsigned char *be
= bp
+ len
;
208 hval
^= (uint64_t) *bp
++;
209 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
210 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
215 static inline bool is_data_obj_writable(SheepdogInode
*inode
, unsigned int idx
)
217 return inode
->vdi_id
== inode
->data_vdi_id
[idx
];
220 static inline bool is_data_obj(uint64_t oid
)
222 return !(VDI_BIT
& oid
);
225 static inline uint64_t data_oid_to_idx(uint64_t oid
)
227 return oid
& (MAX_DATA_OBJS
- 1);
230 static inline uint32_t oid_to_vid(uint64_t oid
)
232 return (oid
& ~VDI_BIT
) >> VDI_SPACE_SHIFT
;
235 static inline uint64_t vid_to_vdi_oid(uint32_t vid
)
237 return VDI_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
);
240 static inline uint64_t vid_to_vmstate_oid(uint32_t vid
, uint32_t idx
)
242 return VMSTATE_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
245 static inline uint64_t vid_to_data_oid(uint32_t vid
, uint32_t idx
)
247 return ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
250 static inline bool is_snapshot(struct SheepdogInode
*inode
)
252 return !!inode
->snap_ctime
;
257 #define DPRINTF(fmt, args...) \
259 fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \
262 #define DPRINTF(fmt, args...)
265 typedef struct SheepdogAIOCB SheepdogAIOCB
;
267 typedef struct AIOReq
{
268 SheepdogAIOCB
*aiocb
;
269 unsigned int iov_offset
;
274 unsigned int data_len
;
278 QLIST_ENTRY(AIOReq
) aio_siblings
;
288 struct SheepdogAIOCB
{
289 BlockDriverAIOCB common
;
297 enum AIOCBState aiocb_type
;
299 Coroutine
*coroutine
;
300 void (*aio_done_func
)(SheepdogAIOCB
*);
306 typedef struct BDRVSheepdogState
{
309 uint32_t min_dirty_data_idx
;
310 uint32_t max_dirty_data_idx
;
312 char name
[SD_MAX_VDI_LEN
];
314 uint32_t cache_flags
;
315 bool discard_supported
;
325 uint32_t aioreq_seq_num
;
326 QLIST_HEAD(inflight_aio_head
, AIOReq
) inflight_aio_head
;
327 QLIST_HEAD(pending_aio_head
, AIOReq
) pending_aio_head
;
330 static const char * sd_strerror(int err
)
334 static const struct {
338 {SD_RES_SUCCESS
, "Success"},
339 {SD_RES_UNKNOWN
, "Unknown error"},
340 {SD_RES_NO_OBJ
, "No object found"},
341 {SD_RES_EIO
, "I/O error"},
342 {SD_RES_VDI_EXIST
, "VDI exists already"},
343 {SD_RES_INVALID_PARMS
, "Invalid parameters"},
344 {SD_RES_SYSTEM_ERROR
, "System error"},
345 {SD_RES_VDI_LOCKED
, "VDI is already locked"},
346 {SD_RES_NO_VDI
, "No vdi found"},
347 {SD_RES_NO_BASE_VDI
, "No base VDI found"},
348 {SD_RES_VDI_READ
, "Failed read the requested VDI"},
349 {SD_RES_VDI_WRITE
, "Failed to write the requested VDI"},
350 {SD_RES_BASE_VDI_READ
, "Failed to read the base VDI"},
351 {SD_RES_BASE_VDI_WRITE
, "Failed to write the base VDI"},
352 {SD_RES_NO_TAG
, "Failed to find the requested tag"},
353 {SD_RES_STARTUP
, "The system is still booting"},
354 {SD_RES_VDI_NOT_LOCKED
, "VDI isn't locked"},
355 {SD_RES_SHUTDOWN
, "The system is shutting down"},
356 {SD_RES_NO_MEM
, "Out of memory on the server"},
357 {SD_RES_FULL_VDI
, "We already have the maximum vdis"},
358 {SD_RES_VER_MISMATCH
, "Protocol version mismatch"},
359 {SD_RES_NO_SPACE
, "Server has no space for new objects"},
360 {SD_RES_WAIT_FOR_FORMAT
, "Sheepdog is waiting for a format operation"},
361 {SD_RES_WAIT_FOR_JOIN
, "Sheepdog is waiting for other nodes joining"},
362 {SD_RES_JOIN_FAILED
, "Target node had failed to join sheepdog"},
363 {SD_RES_HALT
, "Sheepdog is stopped serving IO request"},
364 {SD_RES_READONLY
, "Object is read-only"},
367 for (i
= 0; i
< ARRAY_SIZE(errors
); ++i
) {
368 if (errors
[i
].err
== err
) {
369 return errors
[i
].desc
;
373 return "Invalid error code";
377 * Sheepdog I/O handling:
379 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
380 * link the requests to the inflight_list in the
381 * BDRVSheepdogState. The function exits without waiting for
382 * receiving the response.
384 * 2. We receive the response in aio_read_response, the fd handler to
385 * the sheepdog connection. If metadata update is needed, we send
386 * the write request to the vdi object in sd_write_done, the write
387 * completion function. We switch back to sd_co_readv/writev after
388 * all the requests belonging to the AIOCB are finished.
391 static inline AIOReq
*alloc_aio_req(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
,
392 uint64_t oid
, unsigned int data_len
,
393 uint64_t offset
, uint8_t flags
,
394 uint64_t base_oid
, unsigned int iov_offset
)
398 aio_req
= g_malloc(sizeof(*aio_req
));
399 aio_req
->aiocb
= acb
;
400 aio_req
->iov_offset
= iov_offset
;
402 aio_req
->base_oid
= base_oid
;
403 aio_req
->offset
= offset
;
404 aio_req
->data_len
= data_len
;
405 aio_req
->flags
= flags
;
406 aio_req
->id
= s
->aioreq_seq_num
++;
412 static inline void free_aio_req(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
414 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
416 QLIST_REMOVE(aio_req
, aio_siblings
);
422 static void coroutine_fn
sd_finish_aiocb(SheepdogAIOCB
*acb
)
424 if (!acb
->canceled
) {
425 qemu_coroutine_enter(acb
->coroutine
, NULL
);
427 qemu_aio_release(acb
);
430 static void sd_aio_cancel(BlockDriverAIOCB
*blockacb
)
432 SheepdogAIOCB
*acb
= (SheepdogAIOCB
*)blockacb
;
435 * Sheepdog cannot cancel the requests which are already sent to
436 * the servers, so we just complete the request with -EIO here.
439 qemu_coroutine_enter(acb
->coroutine
, NULL
);
440 acb
->canceled
= true;
443 static const AIOCBInfo sd_aiocb_info
= {
444 .aiocb_size
= sizeof(SheepdogAIOCB
),
445 .cancel
= sd_aio_cancel
,
448 static SheepdogAIOCB
*sd_aio_setup(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
449 int64_t sector_num
, int nb_sectors
)
453 acb
= qemu_aio_get(&sd_aiocb_info
, bs
, NULL
, NULL
);
457 acb
->sector_num
= sector_num
;
458 acb
->nb_sectors
= nb_sectors
;
460 acb
->aio_done_func
= NULL
;
461 acb
->canceled
= false;
462 acb
->coroutine
= qemu_coroutine_self();
468 static int connect_to_sdog(BDRVSheepdogState
*s
)
474 fd
= unix_connect(s
->host_spec
, &err
);
476 fd
= inet_connect(s
->host_spec
, &err
);
479 int ret
= socket_set_nodelay(fd
);
481 error_report("%s", strerror(errno
));
487 qerror_report_err(err
);
490 qemu_set_nonblock(fd
);
496 static coroutine_fn
int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
501 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
502 if (ret
!= sizeof(*hdr
)) {
503 error_report("failed to send a req, %s", strerror(errno
));
507 ret
= qemu_co_send(sockfd
, data
, *wlen
);
509 error_report("failed to send a req, %s", strerror(errno
));
515 static void restart_co_req(void *opaque
)
517 Coroutine
*co
= opaque
;
519 qemu_coroutine_enter(co
, NULL
);
522 typedef struct SheepdogReqCo
{
532 static coroutine_fn
void do_co_req(void *opaque
)
536 SheepdogReqCo
*srco
= opaque
;
537 int sockfd
= srco
->sockfd
;
538 SheepdogReq
*hdr
= srco
->hdr
;
539 void *data
= srco
->data
;
540 unsigned int *wlen
= srco
->wlen
;
541 unsigned int *rlen
= srco
->rlen
;
543 co
= qemu_coroutine_self();
544 qemu_aio_set_fd_handler(sockfd
, NULL
, restart_co_req
, co
);
546 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
551 qemu_aio_set_fd_handler(sockfd
, restart_co_req
, NULL
, co
);
553 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
554 if (ret
!= sizeof(*hdr
)) {
555 error_report("failed to get a rsp, %s", strerror(errno
));
560 if (*rlen
> hdr
->data_length
) {
561 *rlen
= hdr
->data_length
;
565 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
567 error_report("failed to get the data, %s", strerror(errno
));
574 /* there is at most one request for this sockfd, so it is safe to
575 * set each handler to NULL. */
576 qemu_aio_set_fd_handler(sockfd
, NULL
, NULL
, NULL
);
579 srco
->finished
= true;
582 static int do_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
583 unsigned int *wlen
, unsigned int *rlen
)
586 SheepdogReqCo srco
= {
596 if (qemu_in_coroutine()) {
599 co
= qemu_coroutine_create(do_co_req
);
600 qemu_coroutine_enter(co
, &srco
);
601 while (!srco
.finished
) {
609 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
610 struct iovec
*iov
, int niov
, bool create
,
611 enum AIOCBState aiocb_type
);
612 static int coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
);
613 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
);
615 static AIOReq
*find_pending_req(BDRVSheepdogState
*s
, uint64_t oid
)
619 QLIST_FOREACH(aio_req
, &s
->pending_aio_head
, aio_siblings
) {
620 if (aio_req
->oid
== oid
) {
629 * This function searchs pending requests to the object `oid', and
632 static void coroutine_fn
send_pending_req(BDRVSheepdogState
*s
, uint64_t oid
)
638 while ((aio_req
= find_pending_req(s
, oid
)) != NULL
) {
639 acb
= aio_req
->aiocb
;
640 /* move aio_req from pending list to inflight one */
641 QLIST_REMOVE(aio_req
, aio_siblings
);
642 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
643 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
,
644 acb
->qiov
->niov
, false, acb
->aiocb_type
);
646 error_report("add_aio_request is failed");
647 free_aio_req(s
, aio_req
);
648 if (!acb
->nr_pending
) {
649 sd_finish_aiocb(acb
);
656 * Receive responses of the I/O requests.
658 * This function is registered as a fd handler, and called from the
659 * main loop when s->fd is ready for reading responses.
661 static void coroutine_fn
aio_read_response(void *opaque
)
664 BDRVSheepdogState
*s
= opaque
;
667 AIOReq
*aio_req
= NULL
;
671 if (QLIST_EMPTY(&s
->inflight_aio_head
)) {
676 ret
= qemu_co_recv(fd
, &rsp
, sizeof(rsp
));
677 if (ret
!= sizeof(rsp
)) {
678 error_report("failed to get the header, %s", strerror(errno
));
682 /* find the right aio_req from the inflight aio list */
683 QLIST_FOREACH(aio_req
, &s
->inflight_aio_head
, aio_siblings
) {
684 if (aio_req
->id
== rsp
.id
) {
689 error_report("cannot find aio_req %x", rsp
.id
);
693 acb
= aio_req
->aiocb
;
695 switch (acb
->aiocb_type
) {
696 case AIOCB_WRITE_UDATA
:
697 /* this coroutine context is no longer suitable for co_recv
698 * because we may send data to update vdi objects */
700 if (!is_data_obj(aio_req
->oid
)) {
703 idx
= data_oid_to_idx(aio_req
->oid
);
705 if (s
->inode
.data_vdi_id
[idx
] != s
->inode
.vdi_id
) {
707 * If the object is newly created one, we need to update
708 * the vdi object (metadata object). min_dirty_data_idx
709 * and max_dirty_data_idx are changed to include updated
710 * index between them.
712 if (rsp
.result
== SD_RES_SUCCESS
) {
713 s
->inode
.data_vdi_id
[idx
] = s
->inode
.vdi_id
;
714 s
->max_dirty_data_idx
= MAX(idx
, s
->max_dirty_data_idx
);
715 s
->min_dirty_data_idx
= MIN(idx
, s
->min_dirty_data_idx
);
718 * Some requests may be blocked because simultaneous
719 * create requests are not allowed, so we search the
720 * pending requests here.
722 send_pending_req(s
, aio_req
->oid
);
725 case AIOCB_READ_UDATA
:
726 ret
= qemu_co_recvv(fd
, acb
->qiov
->iov
, acb
->qiov
->niov
,
727 aio_req
->iov_offset
, rsp
.data_length
);
728 if (ret
!= rsp
.data_length
) {
729 error_report("failed to get the data, %s", strerror(errno
));
733 case AIOCB_FLUSH_CACHE
:
734 if (rsp
.result
== SD_RES_INVALID_PARMS
) {
735 DPRINTF("disable cache since the server doesn't support it\n");
736 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
737 rsp
.result
= SD_RES_SUCCESS
;
740 case AIOCB_DISCARD_OBJ
:
741 switch (rsp
.result
) {
742 case SD_RES_INVALID_PARMS
:
743 error_report("sheep(%s) doesn't support discard command",
745 rsp
.result
= SD_RES_SUCCESS
;
746 s
->discard_supported
= false;
749 idx
= data_oid_to_idx(aio_req
->oid
);
750 s
->inode
.data_vdi_id
[idx
] = 0;
757 switch (rsp
.result
) {
760 case SD_RES_READONLY
:
761 if (s
->inode
.vdi_id
== oid_to_vid(aio_req
->oid
)) {
762 ret
= reload_inode(s
, 0, "");
768 if (is_data_obj(aio_req
->oid
)) {
769 aio_req
->oid
= vid_to_data_oid(s
->inode
.vdi_id
,
770 data_oid_to_idx(aio_req
->oid
));
772 aio_req
->oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
774 ret
= resend_aioreq(s
, aio_req
);
775 if (ret
== SD_RES_SUCCESS
) {
781 error_report("%s", sd_strerror(rsp
.result
));
785 free_aio_req(s
, aio_req
);
786 if (!acb
->nr_pending
) {
788 * We've finished all requests which belong to the AIOCB, so
789 * we can switch back to sd_co_readv/writev now.
791 acb
->aio_done_func(acb
);
797 static void co_read_response(void *opaque
)
799 BDRVSheepdogState
*s
= opaque
;
802 s
->co_recv
= qemu_coroutine_create(aio_read_response
);
805 qemu_coroutine_enter(s
->co_recv
, opaque
);
808 static void co_write_request(void *opaque
)
810 BDRVSheepdogState
*s
= opaque
;
812 qemu_coroutine_enter(s
->co_send
, NULL
);
816 * Return a socket discriptor to read/write objects.
818 * We cannot use this discriptor for other operations because
819 * the block driver may be on waiting response from the server.
821 static int get_sheep_fd(BDRVSheepdogState
*s
)
825 fd
= connect_to_sdog(s
);
830 qemu_aio_set_fd_handler(fd
, co_read_response
, NULL
, s
);
834 static int sd_parse_uri(BDRVSheepdogState
*s
, const char *filename
,
835 char *vdi
, uint32_t *snapid
, char *tag
)
838 QueryParams
*qp
= NULL
;
841 uri
= uri_parse(filename
);
847 if (!strcmp(uri
->scheme
, "sheepdog")) {
849 } else if (!strcmp(uri
->scheme
, "sheepdog+tcp")) {
851 } else if (!strcmp(uri
->scheme
, "sheepdog+unix")) {
858 if (uri
->path
== NULL
|| !strcmp(uri
->path
, "/")) {
862 pstrcpy(vdi
, SD_MAX_VDI_LEN
, uri
->path
+ 1);
864 qp
= query_params_parse(uri
->query
);
865 if (qp
->n
> 1 || (s
->is_unix
&& !qp
->n
) || (!s
->is_unix
&& qp
->n
)) {
871 /* sheepdog+unix:///vdiname?socket=path */
872 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
876 s
->host_spec
= g_strdup(qp
->p
[0].value
);
878 /* sheepdog[+tcp]://[host:port]/vdiname */
879 s
->host_spec
= g_strdup_printf("%s:%d", uri
->server
?: SD_DEFAULT_ADDR
,
880 uri
->port
?: SD_DEFAULT_PORT
);
885 *snapid
= strtoul(uri
->fragment
, NULL
, 10);
887 pstrcpy(tag
, SD_MAX_VDI_TAG_LEN
, uri
->fragment
);
890 *snapid
= CURRENT_VDI_ID
; /* search current vdi */
895 query_params_free(qp
);
902 * Parse a filename (old syntax)
904 * filename must be one of the following formats:
906 * 2. [vdiname]:[snapid]
908 * 4. [hostname]:[port]:[vdiname]
909 * 5. [hostname]:[port]:[vdiname]:[snapid]
910 * 6. [hostname]:[port]:[vdiname]:[tag]
912 * You can boot from the snapshot images by specifying `snapid` or
915 * You can run VMs outside the Sheepdog cluster by specifying
916 * `hostname' and `port' (experimental).
918 static int parse_vdiname(BDRVSheepdogState
*s
, const char *filename
,
919 char *vdi
, uint32_t *snapid
, char *tag
)
922 const char *host_spec
, *vdi_spec
;
925 strstart(filename
, "sheepdog:", (const char **)&filename
);
926 p
= q
= g_strdup(filename
);
928 /* count the number of separators */
938 /* use the first two tokens as host_spec. */
951 p
= strchr(vdi_spec
, ':');
956 uri
= g_strdup_printf("sheepdog://%s/%s", host_spec
, vdi_spec
);
958 ret
= sd_parse_uri(s
, uri
, vdi
, snapid
, tag
);
966 static int find_vdi_name(BDRVSheepdogState
*s
, const char *filename
,
967 uint32_t snapid
, const char *tag
, uint32_t *vid
,
972 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
973 unsigned int wlen
, rlen
= 0;
974 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
976 fd
= connect_to_sdog(s
);
981 /* This pair of strncpy calls ensures that the buffer is zero-filled,
982 * which is desirable since we'll soon be sending those bytes, and
983 * don't want the send_req to read uninitialized data.
985 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
986 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
988 memset(&hdr
, 0, sizeof(hdr
));
990 hdr
.opcode
= SD_OP_LOCK_VDI
;
992 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
994 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
995 hdr
.proto_ver
= SD_PROTO_VER
;
996 hdr
.data_length
= wlen
;
998 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1000 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1005 if (rsp
->result
!= SD_RES_SUCCESS
) {
1006 error_report("cannot get vdi info, %s, %s %d %s",
1007 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
1008 if (rsp
->result
== SD_RES_NO_VDI
) {
1023 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
1024 struct iovec
*iov
, int niov
, bool create
,
1025 enum AIOCBState aiocb_type
)
1027 int nr_copies
= s
->inode
.nr_copies
;
1029 unsigned int wlen
= 0;
1031 uint64_t oid
= aio_req
->oid
;
1032 unsigned int datalen
= aio_req
->data_len
;
1033 uint64_t offset
= aio_req
->offset
;
1034 uint8_t flags
= aio_req
->flags
;
1035 uint64_t old_oid
= aio_req
->base_oid
;
1038 error_report("bug");
1041 memset(&hdr
, 0, sizeof(hdr
));
1043 switch (aiocb_type
) {
1044 case AIOCB_FLUSH_CACHE
:
1045 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1047 case AIOCB_READ_UDATA
:
1048 hdr
.opcode
= SD_OP_READ_OBJ
;
1051 case AIOCB_WRITE_UDATA
:
1053 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1055 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1058 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1060 case AIOCB_DISCARD_OBJ
:
1061 hdr
.opcode
= SD_OP_DISCARD_OBJ
;
1065 if (s
->cache_flags
) {
1066 hdr
.flags
|= s
->cache_flags
;
1070 hdr
.cow_oid
= old_oid
;
1071 hdr
.copies
= s
->inode
.nr_copies
;
1073 hdr
.data_length
= datalen
;
1074 hdr
.offset
= offset
;
1076 hdr
.id
= aio_req
->id
;
1078 qemu_co_mutex_lock(&s
->lock
);
1079 s
->co_send
= qemu_coroutine_self();
1080 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, co_write_request
, s
);
1081 socket_set_cork(s
->fd
, 1);
1084 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
1085 if (ret
!= sizeof(hdr
)) {
1086 qemu_co_mutex_unlock(&s
->lock
);
1087 error_report("failed to send a req, %s", strerror(errno
));
1092 ret
= qemu_co_sendv(s
->fd
, iov
, niov
, aio_req
->iov_offset
, wlen
);
1094 qemu_co_mutex_unlock(&s
->lock
);
1095 error_report("failed to send a data, %s", strerror(errno
));
1100 socket_set_cork(s
->fd
, 0);
1101 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, NULL
, s
);
1102 qemu_co_mutex_unlock(&s
->lock
);
1107 static int read_write_object(int fd
, char *buf
, uint64_t oid
, uint8_t copies
,
1108 unsigned int datalen
, uint64_t offset
,
1109 bool write
, bool create
, uint32_t cache_flags
)
1112 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1113 unsigned int wlen
, rlen
;
1116 memset(&hdr
, 0, sizeof(hdr
));
1121 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1123 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1125 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1130 hdr
.opcode
= SD_OP_READ_OBJ
;
1133 hdr
.flags
|= cache_flags
;
1136 hdr
.data_length
= datalen
;
1137 hdr
.offset
= offset
;
1138 hdr
.copies
= copies
;
1140 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1142 error_report("failed to send a request to the sheep");
1146 switch (rsp
->result
) {
1147 case SD_RES_SUCCESS
:
1150 error_report("%s", sd_strerror(rsp
->result
));
1155 static int read_object(int fd
, char *buf
, uint64_t oid
, uint8_t copies
,
1156 unsigned int datalen
, uint64_t offset
,
1157 uint32_t cache_flags
)
1159 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, false,
1160 false, cache_flags
);
1163 static int write_object(int fd
, char *buf
, uint64_t oid
, uint8_t copies
,
1164 unsigned int datalen
, uint64_t offset
, bool create
,
1165 uint32_t cache_flags
)
1167 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, true,
1168 create
, cache_flags
);
1171 /* update inode with the latest state */
1172 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
)
1174 SheepdogInode
*inode
;
1178 fd
= connect_to_sdog(s
);
1183 inode
= g_malloc(sizeof(s
->inode
));
1185 ret
= find_vdi_name(s
, s
->name
, snapid
, tag
, &vid
, false);
1190 ret
= read_object(fd
, (char *)inode
, vid_to_vdi_oid(vid
),
1191 s
->inode
.nr_copies
, sizeof(*inode
), 0, s
->cache_flags
);
1196 if (inode
->vdi_id
!= s
->inode
.vdi_id
) {
1197 memcpy(&s
->inode
, inode
, sizeof(s
->inode
));
1207 static int coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
1209 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
1210 bool create
= false;
1212 /* check whether this request becomes a CoW one */
1213 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& is_data_obj(aio_req
->oid
)) {
1214 int idx
= data_oid_to_idx(aio_req
->oid
);
1217 if (s
->inode
.data_vdi_id
[idx
] == 0) {
1221 if (is_data_obj_writable(&s
->inode
, idx
)) {
1225 /* link to the pending list if there is another CoW request to
1226 * the same object */
1227 QLIST_FOREACH(areq
, &s
->inflight_aio_head
, aio_siblings
) {
1228 if (areq
!= aio_req
&& areq
->oid
== aio_req
->oid
) {
1229 DPRINTF("simultaneous CoW to %" PRIx64
"\n", aio_req
->oid
);
1230 QLIST_REMOVE(aio_req
, aio_siblings
);
1231 QLIST_INSERT_HEAD(&s
->pending_aio_head
, aio_req
, aio_siblings
);
1232 return SD_RES_SUCCESS
;
1236 aio_req
->base_oid
= vid_to_data_oid(s
->inode
.data_vdi_id
[idx
], idx
);
1237 aio_req
->flags
|= SD_FLAG_CMD_COW
;
1241 if (is_data_obj(aio_req
->oid
)) {
1242 return add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1243 create
, acb
->aiocb_type
);
1246 iov
.iov_base
= &s
->inode
;
1247 iov
.iov_len
= sizeof(s
->inode
);
1248 return add_aio_request(s
, aio_req
, &iov
, 1, false, AIOCB_WRITE_UDATA
);
1252 /* TODO Convert to fine grained options */
1253 static QemuOptsList runtime_opts
= {
1255 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1259 .type
= QEMU_OPT_STRING
,
1260 .help
= "URL to the sheepdog image",
1262 { /* end of list */ }
1266 static int sd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1271 BDRVSheepdogState
*s
= bs
->opaque
;
1272 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1276 Error
*local_err
= NULL
;
1277 const char *filename
;
1279 opts
= qemu_opts_create_nofail(&runtime_opts
);
1280 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1281 if (error_is_set(&local_err
)) {
1282 qerror_report_err(local_err
);
1283 error_free(local_err
);
1288 filename
= qemu_opt_get(opts
, "filename");
1290 QLIST_INIT(&s
->inflight_aio_head
);
1291 QLIST_INIT(&s
->pending_aio_head
);
1294 memset(vdi
, 0, sizeof(vdi
));
1295 memset(tag
, 0, sizeof(tag
));
1297 if (strstr(filename
, "://")) {
1298 ret
= sd_parse_uri(s
, filename
, vdi
, &snapid
, tag
);
1300 ret
= parse_vdiname(s
, filename
, vdi
, &snapid
, tag
);
1305 s
->fd
= get_sheep_fd(s
);
1311 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, true);
1317 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1318 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1320 s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1321 if (flags
& BDRV_O_NOCACHE
) {
1322 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1324 s
->discard_supported
= true;
1326 if (snapid
|| tag
[0] != '\0') {
1327 DPRINTF("%" PRIx32
" snapshot inode was open.\n", vid
);
1328 s
->is_snapshot
= true;
1331 fd
= connect_to_sdog(s
);
1337 buf
= g_malloc(SD_INODE_SIZE
);
1338 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), 0, SD_INODE_SIZE
, 0,
1347 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1348 s
->min_dirty_data_idx
= UINT32_MAX
;
1349 s
->max_dirty_data_idx
= 0;
1351 bs
->total_sectors
= s
->inode
.vdi_size
/ BDRV_SECTOR_SIZE
;
1352 pstrcpy(s
->name
, sizeof(s
->name
), vdi
);
1353 qemu_co_mutex_init(&s
->lock
);
1354 qemu_opts_del(opts
);
1358 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
);
1362 qemu_opts_del(opts
);
1367 static int do_sd_create(BDRVSheepdogState
*s
, char *filename
, int64_t vdi_size
,
1368 uint32_t base_vid
, uint32_t *vdi_id
, int snapshot
,
1369 uint8_t copy_policy
)
1372 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1374 unsigned int wlen
, rlen
= 0;
1375 char buf
[SD_MAX_VDI_LEN
];
1377 fd
= connect_to_sdog(s
);
1382 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1383 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1385 memset(buf
, 0, sizeof(buf
));
1386 pstrcpy(buf
, sizeof(buf
), filename
);
1388 memset(&hdr
, 0, sizeof(hdr
));
1389 hdr
.opcode
= SD_OP_NEW_VDI
;
1390 hdr
.vdi_id
= base_vid
;
1392 wlen
= SD_MAX_VDI_LEN
;
1394 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1395 hdr
.snapid
= snapshot
;
1397 hdr
.data_length
= wlen
;
1398 hdr
.vdi_size
= vdi_size
;
1399 hdr
.copy_policy
= copy_policy
;
1401 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1409 if (rsp
->result
!= SD_RES_SUCCESS
) {
1410 error_report("%s, %s", sd_strerror(rsp
->result
), filename
);
1415 *vdi_id
= rsp
->vdi_id
;
1421 static int sd_prealloc(const char *filename
)
1423 BlockDriverState
*bs
= NULL
;
1424 uint32_t idx
, max_idx
;
1426 void *buf
= g_malloc0(SD_DATA_OBJ_SIZE
);
1427 Error
*local_err
= NULL
;
1430 ret
= bdrv_file_open(&bs
, filename
, NULL
, BDRV_O_RDWR
, &local_err
);
1432 qerror_report_err(local_err
);
1433 error_free(local_err
);
1437 vdi_size
= bdrv_getlength(bs
);
1442 max_idx
= DIV_ROUND_UP(vdi_size
, SD_DATA_OBJ_SIZE
);
1444 for (idx
= 0; idx
< max_idx
; idx
++) {
1446 * The created image can be a cloned image, so we need to read
1447 * a data from the source image.
1449 ret
= bdrv_pread(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1453 ret
= bdrv_pwrite(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1467 static int sd_create(const char *filename
, QEMUOptionParameter
*options
,
1471 uint32_t vid
= 0, base_vid
= 0;
1472 int64_t vdi_size
= 0;
1473 char *backing_file
= NULL
;
1474 BDRVSheepdogState
*s
;
1475 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1477 bool prealloc
= false;
1478 Error
*local_err
= NULL
;
1480 s
= g_malloc0(sizeof(BDRVSheepdogState
));
1482 memset(vdi
, 0, sizeof(vdi
));
1483 memset(tag
, 0, sizeof(tag
));
1484 if (strstr(filename
, "://")) {
1485 ret
= sd_parse_uri(s
, filename
, vdi
, &snapid
, tag
);
1487 ret
= parse_vdiname(s
, filename
, vdi
, &snapid
, tag
);
1493 while (options
&& options
->name
) {
1494 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1495 vdi_size
= options
->value
.n
;
1496 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1497 backing_file
= options
->value
.s
;
1498 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1499 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1501 } else if (!strcmp(options
->value
.s
, "full")) {
1504 error_report("Invalid preallocation mode: '%s'",
1513 if (vdi_size
> SD_MAX_VDI_SIZE
) {
1514 error_report("too big image size");
1520 BlockDriverState
*bs
;
1521 BDRVSheepdogState
*s
;
1524 /* Currently, only Sheepdog backing image is supported. */
1525 drv
= bdrv_find_protocol(backing_file
, true);
1526 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
1527 error_report("backing_file must be a sheepdog image");
1532 ret
= bdrv_file_open(&bs
, backing_file
, NULL
, 0, &local_err
);
1534 qerror_report_err(local_err
);
1535 error_free(local_err
);
1541 if (!is_snapshot(&s
->inode
)) {
1542 error_report("cannot clone from a non snapshot vdi");
1548 base_vid
= s
->inode
.vdi_id
;
1552 /* TODO: allow users to specify copy number */
1553 ret
= do_sd_create(s
, vdi
, vdi_size
, base_vid
, &vid
, 0, 0);
1554 if (!prealloc
|| ret
) {
1558 ret
= sd_prealloc(filename
);
1564 static void sd_close(BlockDriverState
*bs
)
1566 BDRVSheepdogState
*s
= bs
->opaque
;
1568 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1569 unsigned int wlen
, rlen
= 0;
1572 DPRINTF("%s\n", s
->name
);
1574 fd
= connect_to_sdog(s
);
1579 memset(&hdr
, 0, sizeof(hdr
));
1581 hdr
.opcode
= SD_OP_RELEASE_VDI
;
1582 hdr
.vdi_id
= s
->inode
.vdi_id
;
1583 wlen
= strlen(s
->name
) + 1;
1584 hdr
.data_length
= wlen
;
1585 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1587 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, s
->name
, &wlen
, &rlen
);
1591 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
1592 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
1593 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1596 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
);
1598 g_free(s
->host_spec
);
1601 static int64_t sd_getlength(BlockDriverState
*bs
)
1603 BDRVSheepdogState
*s
= bs
->opaque
;
1605 return s
->inode
.vdi_size
;
1608 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
)
1610 BDRVSheepdogState
*s
= bs
->opaque
;
1612 unsigned int datalen
;
1614 if (offset
< s
->inode
.vdi_size
) {
1615 error_report("shrinking is not supported");
1617 } else if (offset
> SD_MAX_VDI_SIZE
) {
1618 error_report("too big image size");
1622 fd
= connect_to_sdog(s
);
1627 /* we don't need to update entire object */
1628 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1629 s
->inode
.vdi_size
= offset
;
1630 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1631 s
->inode
.nr_copies
, datalen
, 0, false, s
->cache_flags
);
1635 error_report("failed to update an inode.");
1642 * This function is called after writing data objects. If we need to
1643 * update metadata, this sends a write request to the vdi object.
1644 * Otherwise, this switches back to sd_co_readv/writev.
1646 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
1649 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1652 uint32_t offset
, data_len
, mn
, mx
;
1654 mn
= s
->min_dirty_data_idx
;
1655 mx
= s
->max_dirty_data_idx
;
1657 /* we need to update the vdi object. */
1658 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
1659 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
1660 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
1662 s
->min_dirty_data_idx
= UINT32_MAX
;
1663 s
->max_dirty_data_idx
= 0;
1665 iov
.iov_base
= &s
->inode
;
1666 iov
.iov_len
= sizeof(s
->inode
);
1667 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1668 data_len
, offset
, 0, 0, offset
);
1669 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1670 ret
= add_aio_request(s
, aio_req
, &iov
, 1, false, AIOCB_WRITE_UDATA
);
1672 free_aio_req(s
, aio_req
);
1677 acb
->aio_done_func
= sd_finish_aiocb
;
1678 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1682 sd_finish_aiocb(acb
);
1685 /* Delete current working VDI on the snapshot chain */
1686 static bool sd_delete(BDRVSheepdogState
*s
)
1688 unsigned int wlen
= SD_MAX_VDI_LEN
, rlen
= 0;
1689 SheepdogVdiReq hdr
= {
1690 .opcode
= SD_OP_DEL_VDI
,
1691 .vdi_id
= s
->inode
.vdi_id
,
1692 .data_length
= wlen
,
1693 .flags
= SD_FLAG_CMD_WRITE
,
1695 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1698 fd
= connect_to_sdog(s
);
1703 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, s
->name
, &wlen
, &rlen
);
1708 switch (rsp
->result
) {
1710 error_report("%s was already deleted", s
->name
);
1712 case SD_RES_SUCCESS
:
1715 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1723 * Create a writable VDI from a snapshot
1725 static int sd_create_branch(BDRVSheepdogState
*s
)
1732 DPRINTF("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
1734 buf
= g_malloc(SD_INODE_SIZE
);
1737 * Even If deletion fails, we will just create extra snapshot based on
1738 * the workding VDI which was supposed to be deleted. So no need to
1741 deleted
= sd_delete(s
);
1742 ret
= do_sd_create(s
, s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &vid
,
1743 !deleted
, s
->inode
.copy_policy
);
1748 DPRINTF("%" PRIx32
" is created.\n", vid
);
1750 fd
= connect_to_sdog(s
);
1756 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), s
->inode
.nr_copies
,
1757 SD_INODE_SIZE
, 0, s
->cache_flags
);
1765 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1767 s
->is_snapshot
= false;
1769 DPRINTF("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
1778 * Send I/O requests to the server.
1780 * This function sends requests to the server, links the requests to
1781 * the inflight_list in BDRVSheepdogState, and exits without
1782 * waiting the response. The responses are received in the
1783 * `aio_read_response' function which is called from the main loop as
1786 * Returns 1 when we need to wait a response, 0 when there is no sent
1787 * request and -errno in error cases.
1789 static int coroutine_fn
sd_co_rw_vector(void *p
)
1791 SheepdogAIOCB
*acb
= p
;
1793 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* BDRV_SECTOR_SIZE
;
1794 unsigned long idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ SD_DATA_OBJ_SIZE
;
1796 uint64_t offset
= (acb
->sector_num
* BDRV_SECTOR_SIZE
) % SD_DATA_OBJ_SIZE
;
1797 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1798 SheepdogInode
*inode
= &s
->inode
;
1801 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
1803 * In the case we open the snapshot VDI, Sheepdog creates the
1804 * writable VDI when we do a write operation first.
1806 ret
= sd_create_branch(s
);
1814 * Make sure we don't free the aiocb before we are done with all requests.
1815 * This additional reference is dropped at the end of this function.
1819 while (done
!= total
) {
1821 uint64_t old_oid
= 0;
1822 bool create
= false;
1824 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
1826 len
= MIN(total
- done
, SD_DATA_OBJ_SIZE
- offset
);
1828 switch (acb
->aiocb_type
) {
1829 case AIOCB_READ_UDATA
:
1830 if (!inode
->data_vdi_id
[idx
]) {
1831 qemu_iovec_memset(acb
->qiov
, done
, 0, len
);
1835 case AIOCB_WRITE_UDATA
:
1836 if (!inode
->data_vdi_id
[idx
]) {
1838 } else if (!is_data_obj_writable(inode
, idx
)) {
1842 flags
= SD_FLAG_CMD_COW
;
1845 case AIOCB_DISCARD_OBJ
:
1847 * We discard the object only when the whole object is
1848 * 1) allocated 2) trimmed. Otherwise, simply skip it.
1850 if (len
!= SD_DATA_OBJ_SIZE
|| inode
->data_vdi_id
[idx
] == 0) {
1859 DPRINTF("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
" %ld\n",
1861 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
1862 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
1863 DPRINTF("new oid %" PRIx64
"\n", oid
);
1866 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, old_oid
, done
);
1870 QLIST_FOREACH(areq
, &s
->inflight_aio_head
, aio_siblings
) {
1871 if (areq
->oid
== oid
) {
1873 * Sheepdog cannot handle simultaneous create
1874 * requests to the same object. So we cannot send
1875 * the request until the previous request
1879 aio_req
->base_oid
= 0;
1880 QLIST_INSERT_HEAD(&s
->pending_aio_head
, aio_req
,
1887 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1888 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1889 create
, acb
->aiocb_type
);
1891 error_report("add_aio_request is failed");
1892 free_aio_req(s
, aio_req
);
1902 if (!--acb
->nr_pending
) {
1908 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1909 int nb_sectors
, QEMUIOVector
*qiov
)
1914 if (bs
->growable
&& sector_num
+ nb_sectors
> bs
->total_sectors
) {
1915 ret
= sd_truncate(bs
, (sector_num
+ nb_sectors
) * BDRV_SECTOR_SIZE
);
1919 bs
->total_sectors
= sector_num
+ nb_sectors
;
1922 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
);
1923 acb
->aio_done_func
= sd_write_done
;
1924 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1926 ret
= sd_co_rw_vector(acb
);
1928 qemu_aio_release(acb
);
1932 qemu_coroutine_yield();
1937 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1938 int nb_sectors
, QEMUIOVector
*qiov
)
1943 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
);
1944 acb
->aiocb_type
= AIOCB_READ_UDATA
;
1945 acb
->aio_done_func
= sd_finish_aiocb
;
1947 ret
= sd_co_rw_vector(acb
);
1949 qemu_aio_release(acb
);
1953 qemu_coroutine_yield();
1958 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
1960 BDRVSheepdogState
*s
= bs
->opaque
;
1965 if (s
->cache_flags
!= SD_FLAG_CMD_CACHE
) {
1969 acb
= sd_aio_setup(bs
, NULL
, 0, 0);
1970 acb
->aiocb_type
= AIOCB_FLUSH_CACHE
;
1971 acb
->aio_done_func
= sd_finish_aiocb
;
1973 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1975 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1976 ret
= add_aio_request(s
, aio_req
, NULL
, 0, false, acb
->aiocb_type
);
1978 error_report("add_aio_request is failed");
1979 free_aio_req(s
, aio_req
);
1980 qemu_aio_release(acb
);
1984 qemu_coroutine_yield();
1988 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
1990 BDRVSheepdogState
*s
= bs
->opaque
;
1993 SheepdogInode
*inode
;
1994 unsigned int datalen
;
1996 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64
" "
1997 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
1998 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
2000 if (s
->is_snapshot
) {
2001 error_report("You can't create a snapshot of a snapshot VDI, "
2002 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
2007 DPRINTF("%s %s\n", sn_info
->name
, sn_info
->id_str
);
2009 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
2010 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
2011 /* It appears that inode.tag does not require a NUL terminator,
2012 * which means this use of strncpy is ok.
2014 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
2015 /* we don't need to update entire object */
2016 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2018 /* refresh inode. */
2019 fd
= connect_to_sdog(s
);
2025 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2026 s
->inode
.nr_copies
, datalen
, 0, false, s
->cache_flags
);
2028 error_report("failed to write snapshot's inode.");
2032 ret
= do_sd_create(s
, s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &new_vid
,
2033 1, s
->inode
.copy_policy
);
2035 error_report("failed to create inode for snapshot. %s",
2040 inode
= (SheepdogInode
*)g_malloc(datalen
);
2042 ret
= read_object(fd
, (char *)inode
, vid_to_vdi_oid(new_vid
),
2043 s
->inode
.nr_copies
, datalen
, 0, s
->cache_flags
);
2046 error_report("failed to read new inode info. %s", strerror(errno
));
2050 memcpy(&s
->inode
, inode
, datalen
);
2051 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2052 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
2060 * We implement rollback(loadvm) operation to the specified snapshot by
2061 * 1) switch to the snapshot
2062 * 2) rely on sd_create_branch to delete working VDI and
2063 * 3) create a new working VDI based on the speicified snapshot
2065 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
2067 BDRVSheepdogState
*s
= bs
->opaque
;
2068 BDRVSheepdogState
*old_s
;
2069 char tag
[SD_MAX_VDI_TAG_LEN
];
2070 uint32_t snapid
= 0;
2073 old_s
= g_malloc(sizeof(BDRVSheepdogState
));
2075 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
2077 snapid
= strtoul(snapshot_id
, NULL
, 10);
2081 pstrcpy(tag
, sizeof(tag
), snapshot_id
);
2084 ret
= reload_inode(s
, snapid
, tag
);
2089 ret
= sd_create_branch(s
);
2098 /* recover bdrv_sd_state */
2099 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
2102 error_report("failed to open. recover old bdrv_sd_state.");
2107 static int sd_snapshot_delete(BlockDriverState
*bs
,
2108 const char *snapshot_id
,
2112 /* FIXME: Delete specified snapshot id. */
2116 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
2118 BDRVSheepdogState
*s
= bs
->opaque
;
2120 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
2121 QEMUSnapshotInfo
*sn_tab
= NULL
;
2122 unsigned wlen
, rlen
;
2124 static SheepdogInode inode
;
2125 unsigned long *vdi_inuse
;
2126 unsigned int start_nr
;
2130 vdi_inuse
= g_malloc(max
);
2132 fd
= connect_to_sdog(s
);
2141 memset(&req
, 0, sizeof(req
));
2143 req
.opcode
= SD_OP_READ_VDIS
;
2144 req
.data_length
= max
;
2146 ret
= do_req(fd
, (SheepdogReq
*)&req
, vdi_inuse
, &wlen
, &rlen
);
2153 sn_tab
= g_malloc0(nr
* sizeof(*sn_tab
));
2155 /* calculate a vdi id with hash function */
2156 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
2157 start_nr
= hval
& (SD_NR_VDIS
- 1);
2159 fd
= connect_to_sdog(s
);
2165 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
2166 if (!test_bit(vid
, vdi_inuse
)) {
2170 /* we don't need to read entire object */
2171 ret
= read_object(fd
, (char *)&inode
, vid_to_vdi_oid(vid
),
2172 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
2179 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
2180 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
2181 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
2182 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
2183 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
2185 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
), "%u",
2187 pstrcpy(sn_tab
[found
].name
,
2188 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)),
2207 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
2208 int64_t pos
, int size
, int load
)
2211 int fd
, ret
= 0, remaining
= size
;
2212 unsigned int data_len
;
2213 uint64_t vmstate_oid
;
2216 uint32_t vdi_id
= load
? s
->inode
.parent_vdi_id
: s
->inode
.vdi_id
;
2218 fd
= connect_to_sdog(s
);
2224 vdi_index
= pos
/ SD_DATA_OBJ_SIZE
;
2225 offset
= pos
% SD_DATA_OBJ_SIZE
;
2227 data_len
= MIN(remaining
, SD_DATA_OBJ_SIZE
- offset
);
2229 vmstate_oid
= vid_to_vmstate_oid(vdi_id
, vdi_index
);
2231 create
= (offset
== 0);
2233 ret
= read_object(fd
, (char *)data
, vmstate_oid
,
2234 s
->inode
.nr_copies
, data_len
, offset
,
2237 ret
= write_object(fd
, (char *)data
, vmstate_oid
,
2238 s
->inode
.nr_copies
, data_len
, offset
, create
,
2243 error_report("failed to save vmstate %s", strerror(errno
));
2249 remaining
-= data_len
;
2257 static int sd_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2260 BDRVSheepdogState
*s
= bs
->opaque
;
2264 buf
= qemu_blockalign(bs
, qiov
->size
);
2265 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
2266 ret
= do_load_save_vmstate(s
, (uint8_t *) buf
, pos
, qiov
->size
, 0);
2272 static int sd_load_vmstate(BlockDriverState
*bs
, uint8_t *data
,
2273 int64_t pos
, int size
)
2275 BDRVSheepdogState
*s
= bs
->opaque
;
2277 return do_load_save_vmstate(s
, data
, pos
, size
, 1);
2281 static coroutine_fn
int sd_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2286 BDRVSheepdogState
*s
= bs
->opaque
;
2289 if (!s
->discard_supported
) {
2293 acb
= sd_aio_setup(bs
, &dummy
, sector_num
, nb_sectors
);
2294 acb
->aiocb_type
= AIOCB_DISCARD_OBJ
;
2295 acb
->aio_done_func
= sd_finish_aiocb
;
2297 ret
= sd_co_rw_vector(acb
);
2299 qemu_aio_release(acb
);
2303 qemu_coroutine_yield();
2308 static coroutine_fn
int64_t
2309 sd_co_get_block_status(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2312 BDRVSheepdogState
*s
= bs
->opaque
;
2313 SheepdogInode
*inode
= &s
->inode
;
2314 unsigned long start
= sector_num
* BDRV_SECTOR_SIZE
/ SD_DATA_OBJ_SIZE
,
2315 end
= DIV_ROUND_UP((sector_num
+ nb_sectors
) *
2316 BDRV_SECTOR_SIZE
, SD_DATA_OBJ_SIZE
);
2318 int64_t ret
= BDRV_BLOCK_DATA
;
2320 for (idx
= start
; idx
< end
; idx
++) {
2321 if (inode
->data_vdi_id
[idx
] == 0) {
2326 /* Get the longest length of unallocated sectors */
2328 for (idx
= start
+ 1; idx
< end
; idx
++) {
2329 if (inode
->data_vdi_id
[idx
] != 0) {
2335 *pnum
= (idx
- start
) * SD_DATA_OBJ_SIZE
/ BDRV_SECTOR_SIZE
;
2336 if (*pnum
> nb_sectors
) {
2342 static QEMUOptionParameter sd_create_options
[] = {
2344 .name
= BLOCK_OPT_SIZE
,
2346 .help
= "Virtual disk size"
2349 .name
= BLOCK_OPT_BACKING_FILE
,
2351 .help
= "File name of a base image"
2354 .name
= BLOCK_OPT_PREALLOC
,
2356 .help
= "Preallocation mode (allowed values: off, full)"
2361 static BlockDriver bdrv_sheepdog
= {
2362 .format_name
= "sheepdog",
2363 .protocol_name
= "sheepdog",
2364 .instance_size
= sizeof(BDRVSheepdogState
),
2365 .bdrv_needs_filename
= true,
2366 .bdrv_file_open
= sd_open
,
2367 .bdrv_close
= sd_close
,
2368 .bdrv_create
= sd_create
,
2369 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2370 .bdrv_getlength
= sd_getlength
,
2371 .bdrv_truncate
= sd_truncate
,
2373 .bdrv_co_readv
= sd_co_readv
,
2374 .bdrv_co_writev
= sd_co_writev
,
2375 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2376 .bdrv_co_discard
= sd_co_discard
,
2377 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2379 .bdrv_snapshot_create
= sd_snapshot_create
,
2380 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2381 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2382 .bdrv_snapshot_list
= sd_snapshot_list
,
2384 .bdrv_save_vmstate
= sd_save_vmstate
,
2385 .bdrv_load_vmstate
= sd_load_vmstate
,
2387 .create_options
= sd_create_options
,
2390 static BlockDriver bdrv_sheepdog_tcp
= {
2391 .format_name
= "sheepdog",
2392 .protocol_name
= "sheepdog+tcp",
2393 .instance_size
= sizeof(BDRVSheepdogState
),
2394 .bdrv_needs_filename
= true,
2395 .bdrv_file_open
= sd_open
,
2396 .bdrv_close
= sd_close
,
2397 .bdrv_create
= sd_create
,
2398 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2399 .bdrv_getlength
= sd_getlength
,
2400 .bdrv_truncate
= sd_truncate
,
2402 .bdrv_co_readv
= sd_co_readv
,
2403 .bdrv_co_writev
= sd_co_writev
,
2404 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2405 .bdrv_co_discard
= sd_co_discard
,
2406 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2408 .bdrv_snapshot_create
= sd_snapshot_create
,
2409 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2410 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2411 .bdrv_snapshot_list
= sd_snapshot_list
,
2413 .bdrv_save_vmstate
= sd_save_vmstate
,
2414 .bdrv_load_vmstate
= sd_load_vmstate
,
2416 .create_options
= sd_create_options
,
2419 static BlockDriver bdrv_sheepdog_unix
= {
2420 .format_name
= "sheepdog",
2421 .protocol_name
= "sheepdog+unix",
2422 .instance_size
= sizeof(BDRVSheepdogState
),
2423 .bdrv_needs_filename
= true,
2424 .bdrv_file_open
= sd_open
,
2425 .bdrv_close
= sd_close
,
2426 .bdrv_create
= sd_create
,
2427 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2428 .bdrv_getlength
= sd_getlength
,
2429 .bdrv_truncate
= sd_truncate
,
2431 .bdrv_co_readv
= sd_co_readv
,
2432 .bdrv_co_writev
= sd_co_writev
,
2433 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2434 .bdrv_co_discard
= sd_co_discard
,
2435 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2437 .bdrv_snapshot_create
= sd_snapshot_create
,
2438 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2439 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2440 .bdrv_snapshot_list
= sd_snapshot_list
,
2442 .bdrv_save_vmstate
= sd_save_vmstate
,
2443 .bdrv_load_vmstate
= sd_load_vmstate
,
2445 .create_options
= sd_create_options
,
2448 static void bdrv_sheepdog_init(void)
2450 bdrv_register(&bdrv_sheepdog
);
2451 bdrv_register(&bdrv_sheepdog_tcp
);
2452 bdrv_register(&bdrv_sheepdog_unix
);
2454 block_init(bdrv_sheepdog_init
);