2 * Copyright (C) 2009-2010 Nippon Telegraph and Telephone Corporation.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License version
6 * 2 as published by the Free Software Foundation.
8 * You should have received a copy of the GNU General Public License
9 * along with this program. If not, see <http://www.gnu.org/licenses/>.
11 * Contributions after 2012-01-13 are licensed under the terms of the
12 * GNU GPL, version 2 or (at your option) any later version.
15 #include "qemu-common.h"
16 #include "qemu-error.h"
17 #include "qemu_socket.h"
18 #include "block_int.h"
21 #define SD_PROTO_VER 0x01
23 #define SD_DEFAULT_ADDR "localhost"
24 #define SD_DEFAULT_PORT "7000"
26 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
27 #define SD_OP_READ_OBJ 0x02
28 #define SD_OP_WRITE_OBJ 0x03
30 #define SD_OP_NEW_VDI 0x11
31 #define SD_OP_LOCK_VDI 0x12
32 #define SD_OP_RELEASE_VDI 0x13
33 #define SD_OP_GET_VDI_INFO 0x14
34 #define SD_OP_READ_VDIS 0x15
35 #define SD_OP_FLUSH_VDI 0x16
37 #define SD_FLAG_CMD_WRITE 0x01
38 #define SD_FLAG_CMD_COW 0x02
39 #define SD_FLAG_CMD_CACHE 0x04
41 #define SD_RES_SUCCESS 0x00 /* Success */
42 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
43 #define SD_RES_NO_OBJ 0x02 /* No object found */
44 #define SD_RES_EIO 0x03 /* I/O error */
45 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
46 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
47 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
48 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
49 #define SD_RES_NO_VDI 0x08 /* No vdi found */
50 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
51 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
52 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
53 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
54 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
55 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
56 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
57 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
58 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
59 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
60 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
61 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
62 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
63 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
64 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
65 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
70 * 0 - 19 (20 bits): data object space
71 * 20 - 31 (12 bits): reserved data object space
72 * 32 - 55 (24 bits): vdi object space
73 * 56 - 59 ( 4 bits): reserved vdi object space
74 * 60 - 63 ( 4 bits): object type identifier space
77 #define VDI_SPACE_SHIFT 32
78 #define VDI_BIT (UINT64_C(1) << 63)
79 #define VMSTATE_BIT (UINT64_C(1) << 62)
80 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
81 #define MAX_CHILDREN 1024
82 #define SD_MAX_VDI_LEN 256
83 #define SD_MAX_VDI_TAG_LEN 256
84 #define SD_NR_VDIS (1U << 24)
85 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
86 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
87 #define SECTOR_SIZE 512
89 #define SD_INODE_SIZE (sizeof(SheepdogInode))
90 #define CURRENT_VDI_ID 0
92 typedef struct SheepdogReq
{
99 uint32_t opcode_specific
[8];
102 typedef struct SheepdogRsp
{
108 uint32_t data_length
;
110 uint32_t opcode_specific
[7];
113 typedef struct SheepdogObjReq
{
119 uint32_t data_length
;
127 typedef struct SheepdogObjRsp
{
133 uint32_t data_length
;
139 typedef struct SheepdogVdiReq
{
145 uint32_t data_length
;
147 uint32_t base_vdi_id
;
153 typedef struct SheepdogVdiRsp
{
159 uint32_t data_length
;
166 typedef struct SheepdogInode
{
167 char name
[SD_MAX_VDI_LEN
];
168 char tag
[SD_MAX_VDI_TAG_LEN
];
171 uint64_t vm_clock_nsec
;
173 uint64_t vm_state_size
;
174 uint16_t copy_policy
;
176 uint8_t block_size_shift
;
179 uint32_t parent_vdi_id
;
180 uint32_t child_vdi_id
[MAX_CHILDREN
];
181 uint32_t data_vdi_id
[MAX_DATA_OBJS
];
185 * 64 bit FNV-1a non-zero initial basis
187 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
190 * 64 bit Fowler/Noll/Vo FNV-1a hash code
192 static inline uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
194 unsigned char *bp
= buf
;
195 unsigned char *be
= bp
+ len
;
197 hval
^= (uint64_t) *bp
++;
198 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
199 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
204 static inline int is_data_obj_writable(SheepdogInode
*inode
, unsigned int idx
)
206 return inode
->vdi_id
== inode
->data_vdi_id
[idx
];
209 static inline int is_data_obj(uint64_t oid
)
211 return !(VDI_BIT
& oid
);
214 static inline uint64_t data_oid_to_idx(uint64_t oid
)
216 return oid
& (MAX_DATA_OBJS
- 1);
219 static inline uint64_t vid_to_vdi_oid(uint32_t vid
)
221 return VDI_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
);
224 static inline uint64_t vid_to_vmstate_oid(uint32_t vid
, uint32_t idx
)
226 return VMSTATE_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
229 static inline uint64_t vid_to_data_oid(uint32_t vid
, uint32_t idx
)
231 return ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
234 static inline int is_snapshot(struct SheepdogInode
*inode
)
236 return !!inode
->snap_ctime
;
241 #define dprintf(fmt, args...) \
243 fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \
246 #define dprintf(fmt, args...)
249 typedef struct SheepdogAIOCB SheepdogAIOCB
;
251 typedef struct AIOReq
{
252 SheepdogAIOCB
*aiocb
;
253 unsigned int iov_offset
;
258 unsigned int data_len
;
262 QLIST_ENTRY(AIOReq
) outstanding_aio_siblings
;
270 struct SheepdogAIOCB
{
271 BlockDriverAIOCB common
;
279 enum AIOCBState aiocb_type
;
281 Coroutine
*coroutine
;
282 void (*aio_done_func
)(SheepdogAIOCB
*);
288 typedef struct BDRVSheepdogState
{
291 uint32_t min_dirty_data_idx
;
292 uint32_t max_dirty_data_idx
;
294 char name
[SD_MAX_VDI_LEN
];
296 uint8_t cache_enabled
;
307 uint32_t aioreq_seq_num
;
308 QLIST_HEAD(outstanding_aio_head
, AIOReq
) outstanding_aio_head
;
311 static const char * sd_strerror(int err
)
315 static const struct {
319 {SD_RES_SUCCESS
, "Success"},
320 {SD_RES_UNKNOWN
, "Unknown error"},
321 {SD_RES_NO_OBJ
, "No object found"},
322 {SD_RES_EIO
, "I/O error"},
323 {SD_RES_VDI_EXIST
, "VDI exists already"},
324 {SD_RES_INVALID_PARMS
, "Invalid parameters"},
325 {SD_RES_SYSTEM_ERROR
, "System error"},
326 {SD_RES_VDI_LOCKED
, "VDI is already locked"},
327 {SD_RES_NO_VDI
, "No vdi found"},
328 {SD_RES_NO_BASE_VDI
, "No base VDI found"},
329 {SD_RES_VDI_READ
, "Failed read the requested VDI"},
330 {SD_RES_VDI_WRITE
, "Failed to write the requested VDI"},
331 {SD_RES_BASE_VDI_READ
, "Failed to read the base VDI"},
332 {SD_RES_BASE_VDI_WRITE
, "Failed to write the base VDI"},
333 {SD_RES_NO_TAG
, "Failed to find the requested tag"},
334 {SD_RES_STARTUP
, "The system is still booting"},
335 {SD_RES_VDI_NOT_LOCKED
, "VDI isn't locked"},
336 {SD_RES_SHUTDOWN
, "The system is shutting down"},
337 {SD_RES_NO_MEM
, "Out of memory on the server"},
338 {SD_RES_FULL_VDI
, "We already have the maximum vdis"},
339 {SD_RES_VER_MISMATCH
, "Protocol version mismatch"},
340 {SD_RES_NO_SPACE
, "Server has no space for new objects"},
341 {SD_RES_WAIT_FOR_FORMAT
, "Sheepdog is waiting for a format operation"},
342 {SD_RES_WAIT_FOR_JOIN
, "Sheepdog is waiting for other nodes joining"},
343 {SD_RES_JOIN_FAILED
, "Target node had failed to join sheepdog"},
346 for (i
= 0; i
< ARRAY_SIZE(errors
); ++i
) {
347 if (errors
[i
].err
== err
) {
348 return errors
[i
].desc
;
352 return "Invalid error code";
356 * Sheepdog I/O handling:
358 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
359 * link the requests to the outstanding_list in the
360 * BDRVSheepdogState. The function exits without waiting for
361 * receiving the response.
363 * 2. We receive the response in aio_read_response, the fd handler to
364 * the sheepdog connection. If metadata update is needed, we send
365 * the write request to the vdi object in sd_write_done, the write
366 * completion function. We switch back to sd_co_readv/writev after
367 * all the requests belonging to the AIOCB are finished.
370 static inline AIOReq
*alloc_aio_req(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
,
371 uint64_t oid
, unsigned int data_len
,
372 uint64_t offset
, uint8_t flags
,
373 uint64_t base_oid
, unsigned int iov_offset
)
377 aio_req
= g_malloc(sizeof(*aio_req
));
378 aio_req
->aiocb
= acb
;
379 aio_req
->iov_offset
= iov_offset
;
381 aio_req
->base_oid
= base_oid
;
382 aio_req
->offset
= offset
;
383 aio_req
->data_len
= data_len
;
384 aio_req
->flags
= flags
;
385 aio_req
->id
= s
->aioreq_seq_num
++;
387 QLIST_INSERT_HEAD(&s
->outstanding_aio_head
, aio_req
,
388 outstanding_aio_siblings
);
394 static inline void free_aio_req(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
396 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
398 QLIST_REMOVE(aio_req
, outstanding_aio_siblings
);
404 static void coroutine_fn
sd_finish_aiocb(SheepdogAIOCB
*acb
)
406 if (!acb
->canceled
) {
407 qemu_coroutine_enter(acb
->coroutine
, NULL
);
409 qemu_aio_release(acb
);
412 static void sd_aio_cancel(BlockDriverAIOCB
*blockacb
)
414 SheepdogAIOCB
*acb
= (SheepdogAIOCB
*)blockacb
;
417 * Sheepdog cannot cancel the requests which are already sent to
418 * the servers, so we just complete the request with -EIO here.
421 qemu_coroutine_enter(acb
->coroutine
, NULL
);
425 static AIOPool sd_aio_pool
= {
426 .aiocb_size
= sizeof(SheepdogAIOCB
),
427 .cancel
= sd_aio_cancel
,
430 static SheepdogAIOCB
*sd_aio_setup(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
431 int64_t sector_num
, int nb_sectors
,
432 BlockDriverCompletionFunc
*cb
, void *opaque
)
436 acb
= qemu_aio_get(&sd_aio_pool
, bs
, cb
, opaque
);
440 acb
->sector_num
= sector_num
;
441 acb
->nb_sectors
= nb_sectors
;
443 acb
->aio_done_func
= NULL
;
445 acb
->coroutine
= qemu_coroutine_self();
451 static int connect_to_sdog(const char *addr
, const char *port
)
453 char hbuf
[NI_MAXHOST
], sbuf
[NI_MAXSERV
];
455 struct addrinfo hints
, *res
, *res0
;
458 addr
= SD_DEFAULT_ADDR
;
459 port
= SD_DEFAULT_PORT
;
462 memset(&hints
, 0, sizeof(hints
));
463 hints
.ai_socktype
= SOCK_STREAM
;
465 ret
= getaddrinfo(addr
, port
, &hints
, &res0
);
467 error_report("unable to get address info %s, %s",
468 addr
, strerror(errno
));
472 for (res
= res0
; res
; res
= res
->ai_next
) {
473 ret
= getnameinfo(res
->ai_addr
, res
->ai_addrlen
, hbuf
, sizeof(hbuf
),
474 sbuf
, sizeof(sbuf
), NI_NUMERICHOST
| NI_NUMERICSERV
);
479 fd
= socket(res
->ai_family
, res
->ai_socktype
, res
->ai_protocol
);
485 ret
= connect(fd
, res
->ai_addr
, res
->ai_addrlen
);
487 if (errno
== EINTR
) {
493 dprintf("connected to %s:%s\n", addr
, port
);
497 error_report("failed connect to %s:%s", addr
, port
);
503 static int send_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
508 ret
= qemu_send_full(sockfd
, hdr
, sizeof(*hdr
), 0);
509 if (ret
< sizeof(*hdr
)) {
510 error_report("failed to send a req, %s", strerror(errno
));
514 ret
= qemu_send_full(sockfd
, data
, *wlen
, 0);
516 error_report("failed to send a req, %s", strerror(errno
));
523 static coroutine_fn
int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
528 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
529 if (ret
< sizeof(*hdr
)) {
530 error_report("failed to send a req, %s", strerror(errno
));
534 ret
= qemu_co_send(sockfd
, data
, *wlen
);
536 error_report("failed to send a req, %s", strerror(errno
));
542 static coroutine_fn
int do_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
543 unsigned int *wlen
, unsigned int *rlen
);
545 static int do_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
546 unsigned int *wlen
, unsigned int *rlen
)
550 if (qemu_in_coroutine()) {
551 return do_co_req(sockfd
, hdr
, data
, wlen
, rlen
);
554 socket_set_block(sockfd
);
555 ret
= send_req(sockfd
, hdr
, data
, wlen
);
560 ret
= qemu_recv_full(sockfd
, hdr
, sizeof(*hdr
), 0);
561 if (ret
< sizeof(*hdr
)) {
562 error_report("failed to get a rsp, %s", strerror(errno
));
567 if (*rlen
> hdr
->data_length
) {
568 *rlen
= hdr
->data_length
;
572 ret
= qemu_recv_full(sockfd
, data
, *rlen
, 0);
574 error_report("failed to get the data, %s", strerror(errno
));
581 socket_set_nonblock(sockfd
);
585 static void restart_co_req(void *opaque
)
587 Coroutine
*co
= opaque
;
589 qemu_coroutine_enter(co
, NULL
);
592 static coroutine_fn
int do_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
593 unsigned int *wlen
, unsigned int *rlen
)
598 co
= qemu_coroutine_self();
599 qemu_aio_set_fd_handler(sockfd
, NULL
, restart_co_req
, NULL
, co
);
601 socket_set_block(sockfd
);
602 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
607 qemu_aio_set_fd_handler(sockfd
, restart_co_req
, NULL
, NULL
, co
);
609 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
610 if (ret
< sizeof(*hdr
)) {
611 error_report("failed to get a rsp, %s", strerror(errno
));
616 if (*rlen
> hdr
->data_length
) {
617 *rlen
= hdr
->data_length
;
621 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
623 error_report("failed to get the data, %s", strerror(errno
));
630 qemu_aio_set_fd_handler(sockfd
, NULL
, NULL
, NULL
, NULL
);
631 socket_set_nonblock(sockfd
);
635 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
636 struct iovec
*iov
, int niov
, int create
,
637 enum AIOCBState aiocb_type
);
640 * This function searchs pending requests to the object `oid', and
643 static void coroutine_fn
send_pending_req(BDRVSheepdogState
*s
, uint64_t oid
, uint32_t id
)
645 AIOReq
*aio_req
, *next
;
649 QLIST_FOREACH_SAFE(aio_req
, &s
->outstanding_aio_head
,
650 outstanding_aio_siblings
, next
) {
651 if (id
== aio_req
->id
) {
654 if (aio_req
->oid
!= oid
) {
658 acb
= aio_req
->aiocb
;
659 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
,
660 acb
->qiov
->niov
, 0, acb
->aiocb_type
);
662 error_report("add_aio_request is failed");
663 free_aio_req(s
, aio_req
);
664 if (!acb
->nr_pending
) {
665 sd_finish_aiocb(acb
);
672 * Receive responses of the I/O requests.
674 * This function is registered as a fd handler, and called from the
675 * main loop when s->fd is ready for reading responses.
677 static void coroutine_fn
aio_read_response(void *opaque
)
680 BDRVSheepdogState
*s
= opaque
;
683 AIOReq
*aio_req
= NULL
;
687 if (QLIST_EMPTY(&s
->outstanding_aio_head
)) {
692 ret
= qemu_co_recv(fd
, &rsp
, sizeof(rsp
));
694 error_report("failed to get the header, %s", strerror(errno
));
698 /* find the right aio_req from the outstanding_aio list */
699 QLIST_FOREACH(aio_req
, &s
->outstanding_aio_head
, outstanding_aio_siblings
) {
700 if (aio_req
->id
== rsp
.id
) {
705 error_report("cannot find aio_req %x", rsp
.id
);
709 acb
= aio_req
->aiocb
;
711 switch (acb
->aiocb_type
) {
712 case AIOCB_WRITE_UDATA
:
713 /* this coroutine context is no longer suitable for co_recv
714 * because we may send data to update vdi objects */
716 if (!is_data_obj(aio_req
->oid
)) {
719 idx
= data_oid_to_idx(aio_req
->oid
);
721 if (s
->inode
.data_vdi_id
[idx
] != s
->inode
.vdi_id
) {
723 * If the object is newly created one, we need to update
724 * the vdi object (metadata object). min_dirty_data_idx
725 * and max_dirty_data_idx are changed to include updated
726 * index between them.
728 s
->inode
.data_vdi_id
[idx
] = s
->inode
.vdi_id
;
729 s
->max_dirty_data_idx
= MAX(idx
, s
->max_dirty_data_idx
);
730 s
->min_dirty_data_idx
= MIN(idx
, s
->min_dirty_data_idx
);
733 * Some requests may be blocked because simultaneous
734 * create requests are not allowed, so we search the
735 * pending requests here.
737 send_pending_req(s
, vid_to_data_oid(s
->inode
.vdi_id
, idx
), rsp
.id
);
740 case AIOCB_READ_UDATA
:
741 ret
= qemu_co_recvv(fd
, acb
->qiov
->iov
, rsp
.data_length
,
742 aio_req
->iov_offset
);
744 error_report("failed to get the data, %s", strerror(errno
));
750 if (rsp
.result
!= SD_RES_SUCCESS
) {
752 error_report("%s", sd_strerror(rsp
.result
));
755 free_aio_req(s
, aio_req
);
756 if (!acb
->nr_pending
) {
758 * We've finished all requests which belong to the AIOCB, so
759 * we can switch back to sd_co_readv/writev now.
761 acb
->aio_done_func(acb
);
767 static void co_read_response(void *opaque
)
769 BDRVSheepdogState
*s
= opaque
;
772 s
->co_recv
= qemu_coroutine_create(aio_read_response
);
775 qemu_coroutine_enter(s
->co_recv
, opaque
);
778 static void co_write_request(void *opaque
)
780 BDRVSheepdogState
*s
= opaque
;
782 qemu_coroutine_enter(s
->co_send
, NULL
);
785 static int aio_flush_request(void *opaque
)
787 BDRVSheepdogState
*s
= opaque
;
789 return !QLIST_EMPTY(&s
->outstanding_aio_head
);
792 static int set_nodelay(int fd
)
797 ret
= setsockopt(fd
, IPPROTO_TCP
, TCP_NODELAY
, (char *)&opt
, sizeof(opt
));
802 * Return a socket discriptor to read/write objects.
804 * We cannot use this discriptor for other operations because
805 * the block driver may be on waiting response from the server.
807 static int get_sheep_fd(BDRVSheepdogState
*s
)
811 fd
= connect_to_sdog(s
->addr
, s
->port
);
813 error_report("%s", strerror(errno
));
817 socket_set_nonblock(fd
);
819 ret
= set_nodelay(fd
);
821 error_report("%s", strerror(errno
));
826 qemu_aio_set_fd_handler(fd
, co_read_response
, NULL
, aio_flush_request
, s
);
833 * filename must be one of the following formats:
835 * 2. [vdiname]:[snapid]
837 * 4. [hostname]:[port]:[vdiname]
838 * 5. [hostname]:[port]:[vdiname]:[snapid]
839 * 6. [hostname]:[port]:[vdiname]:[tag]
841 * You can boot from the snapshot images by specifying `snapid` or
844 * You can run VMs outside the Sheepdog cluster by specifying
845 * `hostname' and `port' (experimental).
847 static int parse_vdiname(BDRVSheepdogState
*s
, const char *filename
,
848 char *vdi
, uint32_t *snapid
, char *tag
)
853 p
= q
= g_strdup(filename
);
855 /* count the number of separators */
865 /* use the first two tokens as hostname and port number. */
879 strncpy(vdi
, p
, SD_MAX_VDI_LEN
);
881 p
= strchr(vdi
, ':');
884 *snapid
= strtoul(p
, NULL
, 10);
886 strncpy(tag
, p
, SD_MAX_VDI_TAG_LEN
);
889 *snapid
= CURRENT_VDI_ID
; /* search current vdi */
892 if (s
->addr
== NULL
) {
899 static int find_vdi_name(BDRVSheepdogState
*s
, char *filename
, uint32_t snapid
,
900 char *tag
, uint32_t *vid
, int for_snapshot
)
904 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
905 unsigned int wlen
, rlen
= 0;
906 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
908 fd
= connect_to_sdog(s
->addr
, s
->port
);
913 memset(buf
, 0, sizeof(buf
));
914 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
915 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
917 memset(&hdr
, 0, sizeof(hdr
));
919 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
921 hdr
.opcode
= SD_OP_LOCK_VDI
;
923 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
924 hdr
.proto_ver
= SD_PROTO_VER
;
925 hdr
.data_length
= wlen
;
927 hdr
.flags
= SD_FLAG_CMD_WRITE
;
929 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
934 if (rsp
->result
!= SD_RES_SUCCESS
) {
935 error_report("cannot get vdi info, %s, %s %d %s",
936 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
937 if (rsp
->result
== SD_RES_NO_VDI
) {
952 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
953 struct iovec
*iov
, int niov
, int create
,
954 enum AIOCBState aiocb_type
)
956 int nr_copies
= s
->inode
.nr_copies
;
960 uint64_t oid
= aio_req
->oid
;
961 unsigned int datalen
= aio_req
->data_len
;
962 uint64_t offset
= aio_req
->offset
;
963 uint8_t flags
= aio_req
->flags
;
964 uint64_t old_oid
= aio_req
->base_oid
;
970 memset(&hdr
, 0, sizeof(hdr
));
972 if (aiocb_type
== AIOCB_READ_UDATA
) {
974 hdr
.opcode
= SD_OP_READ_OBJ
;
978 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
979 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
982 hdr
.opcode
= SD_OP_WRITE_OBJ
;
983 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
986 if (s
->cache_enabled
) {
987 hdr
.flags
|= SD_FLAG_CMD_CACHE
;
991 hdr
.cow_oid
= old_oid
;
992 hdr
.copies
= s
->inode
.nr_copies
;
994 hdr
.data_length
= datalen
;
997 hdr
.id
= aio_req
->id
;
999 qemu_co_mutex_lock(&s
->lock
);
1000 s
->co_send
= qemu_coroutine_self();
1001 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, co_write_request
,
1002 aio_flush_request
, s
);
1003 socket_set_cork(s
->fd
, 1);
1006 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
1008 qemu_co_mutex_unlock(&s
->lock
);
1009 error_report("failed to send a req, %s", strerror(errno
));
1014 ret
= qemu_co_sendv(s
->fd
, iov
, wlen
, aio_req
->iov_offset
);
1016 qemu_co_mutex_unlock(&s
->lock
);
1017 error_report("failed to send a data, %s", strerror(errno
));
1022 socket_set_cork(s
->fd
, 0);
1023 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, NULL
,
1024 aio_flush_request
, s
);
1025 qemu_co_mutex_unlock(&s
->lock
);
1030 static int read_write_object(int fd
, char *buf
, uint64_t oid
, int copies
,
1031 unsigned int datalen
, uint64_t offset
,
1032 int write
, int create
, uint8_t cache
)
1035 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1036 unsigned int wlen
, rlen
;
1039 memset(&hdr
, 0, sizeof(hdr
));
1044 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1046 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1048 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1053 hdr
.opcode
= SD_OP_READ_OBJ
;
1057 hdr
.flags
|= SD_FLAG_CMD_CACHE
;
1061 hdr
.data_length
= datalen
;
1062 hdr
.offset
= offset
;
1063 hdr
.copies
= copies
;
1065 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1067 error_report("failed to send a request to the sheep");
1071 switch (rsp
->result
) {
1072 case SD_RES_SUCCESS
:
1075 error_report("%s", sd_strerror(rsp
->result
));
1080 static int read_object(int fd
, char *buf
, uint64_t oid
, int copies
,
1081 unsigned int datalen
, uint64_t offset
, uint8_t cache
)
1083 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, 0, 0,
1087 static int write_object(int fd
, char *buf
, uint64_t oid
, int copies
,
1088 unsigned int datalen
, uint64_t offset
, int create
,
1091 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, 1, create
,
1095 static int sd_open(BlockDriverState
*bs
, const char *filename
, int flags
)
1099 BDRVSheepdogState
*s
= bs
->opaque
;
1100 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1104 strstart(filename
, "sheepdog:", (const char **)&filename
);
1106 QLIST_INIT(&s
->outstanding_aio_head
);
1109 memset(vdi
, 0, sizeof(vdi
));
1110 memset(tag
, 0, sizeof(tag
));
1111 if (parse_vdiname(s
, filename
, vdi
, &snapid
, tag
) < 0) {
1115 s
->fd
= get_sheep_fd(s
);
1121 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, 0);
1126 if (flags
& BDRV_O_CACHE_WB
) {
1127 s
->cache_enabled
= 1;
1128 s
->flush_fd
= connect_to_sdog(s
->addr
, s
->port
);
1129 if (s
->flush_fd
< 0) {
1130 error_report("failed to connect");
1136 if (snapid
|| tag
[0] != '\0') {
1137 dprintf("%" PRIx32
" snapshot inode was open.\n", vid
);
1141 fd
= connect_to_sdog(s
->addr
, s
->port
);
1143 error_report("failed to connect");
1148 buf
= g_malloc(SD_INODE_SIZE
);
1149 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), 0, SD_INODE_SIZE
, 0,
1158 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1159 s
->min_dirty_data_idx
= UINT32_MAX
;
1160 s
->max_dirty_data_idx
= 0;
1162 bs
->total_sectors
= s
->inode
.vdi_size
/ SECTOR_SIZE
;
1163 strncpy(s
->name
, vdi
, sizeof(s
->name
));
1164 qemu_co_mutex_init(&s
->lock
);
1168 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
, NULL
);
1176 static int do_sd_create(char *filename
, int64_t vdi_size
,
1177 uint32_t base_vid
, uint32_t *vdi_id
, int snapshot
,
1178 const char *addr
, const char *port
)
1181 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1183 unsigned int wlen
, rlen
= 0;
1184 char buf
[SD_MAX_VDI_LEN
];
1186 fd
= connect_to_sdog(addr
, port
);
1191 memset(buf
, 0, sizeof(buf
));
1192 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
1194 memset(&hdr
, 0, sizeof(hdr
));
1195 hdr
.opcode
= SD_OP_NEW_VDI
;
1196 hdr
.base_vdi_id
= base_vid
;
1198 wlen
= SD_MAX_VDI_LEN
;
1200 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1201 hdr
.snapid
= snapshot
;
1203 hdr
.data_length
= wlen
;
1204 hdr
.vdi_size
= vdi_size
;
1206 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1214 if (rsp
->result
!= SD_RES_SUCCESS
) {
1215 error_report("%s, %s", sd_strerror(rsp
->result
), filename
);
1220 *vdi_id
= rsp
->vdi_id
;
1226 static int sd_prealloc(const char *filename
)
1228 BlockDriverState
*bs
= NULL
;
1229 uint32_t idx
, max_idx
;
1231 void *buf
= g_malloc0(SD_DATA_OBJ_SIZE
);
1234 ret
= bdrv_file_open(&bs
, filename
, BDRV_O_RDWR
);
1239 vdi_size
= bdrv_getlength(bs
);
1244 max_idx
= DIV_ROUND_UP(vdi_size
, SD_DATA_OBJ_SIZE
);
1246 for (idx
= 0; idx
< max_idx
; idx
++) {
1248 * The created image can be a cloned image, so we need to read
1249 * a data from the source image.
1251 ret
= bdrv_pread(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1255 ret
= bdrv_pwrite(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1269 static int sd_create(const char *filename
, QEMUOptionParameter
*options
)
1272 uint32_t vid
= 0, base_vid
= 0;
1273 int64_t vdi_size
= 0;
1274 char *backing_file
= NULL
;
1275 BDRVSheepdogState
*s
;
1276 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1279 const char *vdiname
;
1281 s
= g_malloc0(sizeof(BDRVSheepdogState
));
1283 strstart(filename
, "sheepdog:", &vdiname
);
1285 memset(vdi
, 0, sizeof(vdi
));
1286 memset(tag
, 0, sizeof(tag
));
1287 if (parse_vdiname(s
, vdiname
, vdi
, &snapid
, tag
) < 0) {
1288 error_report("invalid filename");
1293 while (options
&& options
->name
) {
1294 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1295 vdi_size
= options
->value
.n
;
1296 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1297 backing_file
= options
->value
.s
;
1298 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1299 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1301 } else if (!strcmp(options
->value
.s
, "full")) {
1304 error_report("Invalid preallocation mode: '%s'",
1313 if (vdi_size
> SD_MAX_VDI_SIZE
) {
1314 error_report("too big image size");
1320 BlockDriverState
*bs
;
1321 BDRVSheepdogState
*s
;
1324 /* Currently, only Sheepdog backing image is supported. */
1325 drv
= bdrv_find_protocol(backing_file
);
1326 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
1327 error_report("backing_file must be a sheepdog image");
1332 ret
= bdrv_file_open(&bs
, backing_file
, 0);
1339 if (!is_snapshot(&s
->inode
)) {
1340 error_report("cannot clone from a non snapshot vdi");
1346 base_vid
= s
->inode
.vdi_id
;
1350 ret
= do_sd_create(vdi
, vdi_size
, base_vid
, &vid
, 0, s
->addr
, s
->port
);
1351 if (!prealloc
|| ret
) {
1355 ret
= sd_prealloc(filename
);
1361 static void sd_close(BlockDriverState
*bs
)
1363 BDRVSheepdogState
*s
= bs
->opaque
;
1365 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1366 unsigned int wlen
, rlen
= 0;
1369 dprintf("%s\n", s
->name
);
1371 fd
= connect_to_sdog(s
->addr
, s
->port
);
1376 memset(&hdr
, 0, sizeof(hdr
));
1378 hdr
.opcode
= SD_OP_RELEASE_VDI
;
1379 wlen
= strlen(s
->name
) + 1;
1380 hdr
.data_length
= wlen
;
1381 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1383 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, s
->name
, &wlen
, &rlen
);
1387 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
1388 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
1389 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1392 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
, NULL
);
1394 if (s
->cache_enabled
) {
1395 closesocket(s
->flush_fd
);
1400 static int64_t sd_getlength(BlockDriverState
*bs
)
1402 BDRVSheepdogState
*s
= bs
->opaque
;
1404 return s
->inode
.vdi_size
;
1407 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
)
1409 BDRVSheepdogState
*s
= bs
->opaque
;
1411 unsigned int datalen
;
1413 if (offset
< s
->inode
.vdi_size
) {
1414 error_report("shrinking is not supported");
1416 } else if (offset
> SD_MAX_VDI_SIZE
) {
1417 error_report("too big image size");
1421 fd
= connect_to_sdog(s
->addr
, s
->port
);
1426 /* we don't need to update entire object */
1427 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1428 s
->inode
.vdi_size
= offset
;
1429 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1430 s
->inode
.nr_copies
, datalen
, 0, 0, s
->cache_enabled
);
1434 error_report("failed to update an inode.");
1441 * This function is called after writing data objects. If we need to
1442 * update metadata, this sends a write request to the vdi object.
1443 * Otherwise, this switches back to sd_co_readv/writev.
1445 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
1448 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1451 uint32_t offset
, data_len
, mn
, mx
;
1453 mn
= s
->min_dirty_data_idx
;
1454 mx
= s
->max_dirty_data_idx
;
1456 /* we need to update the vdi object. */
1457 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
1458 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
1459 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
1461 s
->min_dirty_data_idx
= UINT32_MAX
;
1462 s
->max_dirty_data_idx
= 0;
1464 iov
.iov_base
= &s
->inode
;
1465 iov
.iov_len
= sizeof(s
->inode
);
1466 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1467 data_len
, offset
, 0, 0, offset
);
1468 ret
= add_aio_request(s
, aio_req
, &iov
, 1, 0, AIOCB_WRITE_UDATA
);
1470 free_aio_req(s
, aio_req
);
1475 acb
->aio_done_func
= sd_finish_aiocb
;
1476 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1480 sd_finish_aiocb(acb
);
1484 * Create a writable VDI from a snapshot
1486 static int sd_create_branch(BDRVSheepdogState
*s
)
1492 dprintf("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
1494 buf
= g_malloc(SD_INODE_SIZE
);
1496 ret
= do_sd_create(s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &vid
, 1,
1502 dprintf("%" PRIx32
" is created.\n", vid
);
1504 fd
= connect_to_sdog(s
->addr
, s
->port
);
1506 error_report("failed to connect");
1511 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), s
->inode
.nr_copies
,
1512 SD_INODE_SIZE
, 0, s
->cache_enabled
);
1520 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1524 dprintf("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
1533 * Send I/O requests to the server.
1535 * This function sends requests to the server, links the requests to
1536 * the outstanding_list in BDRVSheepdogState, and exits without
1537 * waiting the response. The responses are received in the
1538 * `aio_read_response' function which is called from the main loop as
1541 * Returns 1 when we need to wait a response, 0 when there is no sent
1542 * request and -errno in error cases.
1544 static int coroutine_fn
sd_co_rw_vector(void *p
)
1546 SheepdogAIOCB
*acb
= p
;
1548 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* SECTOR_SIZE
;
1549 unsigned long idx
= acb
->sector_num
* SECTOR_SIZE
/ SD_DATA_OBJ_SIZE
;
1551 uint64_t offset
= (acb
->sector_num
* SECTOR_SIZE
) % SD_DATA_OBJ_SIZE
;
1552 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1553 SheepdogInode
*inode
= &s
->inode
;
1556 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
1558 * In the case we open the snapshot VDI, Sheepdog creates the
1559 * writable VDI when we do a write operation first.
1561 ret
= sd_create_branch(s
);
1569 * Make sure we don't free the aiocb before we are done with all requests.
1570 * This additional reference is dropped at the end of this function.
1574 while (done
!= total
) {
1576 uint64_t old_oid
= 0;
1579 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
1581 len
= MIN(total
- done
, SD_DATA_OBJ_SIZE
- offset
);
1583 if (!inode
->data_vdi_id
[idx
]) {
1584 if (acb
->aiocb_type
== AIOCB_READ_UDATA
) {
1589 } else if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
1590 && !is_data_obj_writable(inode
, idx
)) {
1594 flags
= SD_FLAG_CMD_COW
;
1598 dprintf("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
" %ld\n",
1600 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
1601 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
1602 dprintf("new oid %" PRIx64
"\n", oid
);
1605 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, old_oid
, done
);
1609 QLIST_FOREACH(areq
, &s
->outstanding_aio_head
,
1610 outstanding_aio_siblings
) {
1611 if (areq
== aio_req
) {
1614 if (areq
->oid
== oid
) {
1616 * Sheepdog cannot handle simultaneous create
1617 * requests to the same object. So we cannot send
1618 * the request until the previous request
1622 aio_req
->base_oid
= 0;
1628 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1629 create
, acb
->aiocb_type
);
1631 error_report("add_aio_request is failed");
1632 free_aio_req(s
, aio_req
);
1642 if (!--acb
->nr_pending
) {
1648 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1649 int nb_sectors
, QEMUIOVector
*qiov
)
1654 if (bs
->growable
&& sector_num
+ nb_sectors
> bs
->total_sectors
) {
1655 ret
= sd_truncate(bs
, (sector_num
+ nb_sectors
) * SECTOR_SIZE
);
1659 bs
->total_sectors
= sector_num
+ nb_sectors
;
1662 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
, NULL
, NULL
);
1663 acb
->aio_done_func
= sd_write_done
;
1664 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1666 ret
= sd_co_rw_vector(acb
);
1668 qemu_aio_release(acb
);
1672 qemu_coroutine_yield();
1677 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1678 int nb_sectors
, QEMUIOVector
*qiov
)
1683 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
, NULL
, NULL
);
1684 acb
->aiocb_type
= AIOCB_READ_UDATA
;
1685 acb
->aio_done_func
= sd_finish_aiocb
;
1688 * TODO: we can do better; we don't need to initialize
1691 for (i
= 0; i
< qiov
->niov
; i
++) {
1692 memset(qiov
->iov
[i
].iov_base
, 0, qiov
->iov
[i
].iov_len
);
1695 ret
= sd_co_rw_vector(acb
);
1697 qemu_aio_release(acb
);
1701 qemu_coroutine_yield();
1706 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
1708 BDRVSheepdogState
*s
= bs
->opaque
;
1709 SheepdogObjReq hdr
= { 0 };
1710 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1711 SheepdogInode
*inode
= &s
->inode
;
1713 unsigned int wlen
= 0, rlen
= 0;
1715 if (!s
->cache_enabled
) {
1719 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1720 hdr
.oid
= vid_to_vdi_oid(inode
->vdi_id
);
1722 ret
= do_req(s
->flush_fd
, (SheepdogReq
*)&hdr
, NULL
, &wlen
, &rlen
);
1724 error_report("failed to send a request to the sheep");
1728 if (rsp
->result
== SD_RES_INVALID_PARMS
) {
1729 dprintf("disable write cache since the server doesn't support it\n");
1731 s
->cache_enabled
= 0;
1732 closesocket(s
->flush_fd
);
1736 if (rsp
->result
!= SD_RES_SUCCESS
) {
1737 error_report("%s", sd_strerror(rsp
->result
));
1744 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
1746 BDRVSheepdogState
*s
= bs
->opaque
;
1749 SheepdogInode
*inode
;
1750 unsigned int datalen
;
1752 dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64
" "
1753 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
1754 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
1756 if (s
->is_snapshot
) {
1757 error_report("You can't create a snapshot of a snapshot VDI, "
1758 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
1763 dprintf("%s %s\n", sn_info
->name
, sn_info
->id_str
);
1765 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
1766 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
1767 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
1768 /* we don't need to update entire object */
1769 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1771 /* refresh inode. */
1772 fd
= connect_to_sdog(s
->addr
, s
->port
);
1778 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1779 s
->inode
.nr_copies
, datalen
, 0, 0, s
->cache_enabled
);
1781 error_report("failed to write snapshot's inode.");
1785 ret
= do_sd_create(s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &new_vid
, 1,
1788 error_report("failed to create inode for snapshot. %s",
1793 inode
= (SheepdogInode
*)g_malloc(datalen
);
1795 ret
= read_object(fd
, (char *)inode
, vid_to_vdi_oid(new_vid
),
1796 s
->inode
.nr_copies
, datalen
, 0, s
->cache_enabled
);
1799 error_report("failed to read new inode info. %s", strerror(errno
));
1803 memcpy(&s
->inode
, inode
, datalen
);
1804 dprintf("s->inode: name %s snap_id %x oid %x\n",
1805 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
1812 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
1814 BDRVSheepdogState
*s
= bs
->opaque
;
1815 BDRVSheepdogState
*old_s
;
1816 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1819 uint32_t snapid
= 0;
1822 old_s
= g_malloc(sizeof(BDRVSheepdogState
));
1824 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
1826 memset(vdi
, 0, sizeof(vdi
));
1827 strncpy(vdi
, s
->name
, sizeof(vdi
));
1829 memset(tag
, 0, sizeof(tag
));
1830 snapid
= strtoul(snapshot_id
, NULL
, 10);
1832 strncpy(tag
, s
->name
, sizeof(tag
));
1835 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, 1);
1837 error_report("Failed to find_vdi_name");
1841 fd
= connect_to_sdog(s
->addr
, s
->port
);
1843 error_report("failed to connect");
1848 buf
= g_malloc(SD_INODE_SIZE
);
1849 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), s
->inode
.nr_copies
,
1850 SD_INODE_SIZE
, 0, s
->cache_enabled
);
1858 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1860 if (!s
->inode
.vm_state_size
) {
1861 error_report("Invalid snapshot");
1873 /* recover bdrv_sd_state */
1874 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
1878 error_report("failed to open. recover old bdrv_sd_state.");
1883 static int sd_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
1885 /* FIXME: Delete specified snapshot id. */
1889 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
1891 BDRVSheepdogState
*s
= bs
->opaque
;
1893 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
1894 QEMUSnapshotInfo
*sn_tab
= NULL
;
1895 unsigned wlen
, rlen
;
1897 static SheepdogInode inode
;
1898 unsigned long *vdi_inuse
;
1899 unsigned int start_nr
;
1903 vdi_inuse
= g_malloc(max
);
1905 fd
= connect_to_sdog(s
->addr
, s
->port
);
1914 memset(&req
, 0, sizeof(req
));
1916 req
.opcode
= SD_OP_READ_VDIS
;
1917 req
.data_length
= max
;
1919 ret
= do_req(fd
, (SheepdogReq
*)&req
, vdi_inuse
, &wlen
, &rlen
);
1926 sn_tab
= g_malloc0(nr
* sizeof(*sn_tab
));
1928 /* calculate a vdi id with hash function */
1929 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
1930 start_nr
= hval
& (SD_NR_VDIS
- 1);
1932 fd
= connect_to_sdog(s
->addr
, s
->port
);
1934 error_report("failed to connect");
1939 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
1940 if (!test_bit(vid
, vdi_inuse
)) {
1944 /* we don't need to read entire object */
1945 ret
= read_object(fd
, (char *)&inode
, vid_to_vdi_oid(vid
),
1946 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
1953 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
1954 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
1955 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
1956 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
1957 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
1959 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
), "%u",
1961 strncpy(sn_tab
[found
].name
, inode
.tag
,
1962 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)));
1980 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
1981 int64_t pos
, int size
, int load
)
1984 int ret
= 0, remaining
= size
;
1985 unsigned int data_len
;
1986 uint64_t vmstate_oid
;
1990 fd
= connect_to_sdog(s
->addr
, s
->port
);
1996 vdi_index
= pos
/ SD_DATA_OBJ_SIZE
;
1997 offset
= pos
% SD_DATA_OBJ_SIZE
;
1999 data_len
= MIN(remaining
, SD_DATA_OBJ_SIZE
);
2001 vmstate_oid
= vid_to_vmstate_oid(s
->inode
.vdi_id
, vdi_index
);
2003 create
= (offset
== 0);
2005 ret
= read_object(fd
, (char *)data
, vmstate_oid
,
2006 s
->inode
.nr_copies
, data_len
, offset
,
2009 ret
= write_object(fd
, (char *)data
, vmstate_oid
,
2010 s
->inode
.nr_copies
, data_len
, offset
, create
,
2015 error_report("failed to save vmstate %s", strerror(errno
));
2020 remaining
-= data_len
;
2028 static int sd_save_vmstate(BlockDriverState
*bs
, const uint8_t *data
,
2029 int64_t pos
, int size
)
2031 BDRVSheepdogState
*s
= bs
->opaque
;
2033 return do_load_save_vmstate(s
, (uint8_t *)data
, pos
, size
, 0);
2036 static int sd_load_vmstate(BlockDriverState
*bs
, uint8_t *data
,
2037 int64_t pos
, int size
)
2039 BDRVSheepdogState
*s
= bs
->opaque
;
2041 return do_load_save_vmstate(s
, data
, pos
, size
, 1);
2045 static QEMUOptionParameter sd_create_options
[] = {
2047 .name
= BLOCK_OPT_SIZE
,
2049 .help
= "Virtual disk size"
2052 .name
= BLOCK_OPT_BACKING_FILE
,
2054 .help
= "File name of a base image"
2057 .name
= BLOCK_OPT_PREALLOC
,
2059 .help
= "Preallocation mode (allowed values: off, full)"
2064 BlockDriver bdrv_sheepdog
= {
2065 .format_name
= "sheepdog",
2066 .protocol_name
= "sheepdog",
2067 .instance_size
= sizeof(BDRVSheepdogState
),
2068 .bdrv_file_open
= sd_open
,
2069 .bdrv_close
= sd_close
,
2070 .bdrv_create
= sd_create
,
2071 .bdrv_getlength
= sd_getlength
,
2072 .bdrv_truncate
= sd_truncate
,
2074 .bdrv_co_readv
= sd_co_readv
,
2075 .bdrv_co_writev
= sd_co_writev
,
2076 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2078 .bdrv_snapshot_create
= sd_snapshot_create
,
2079 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2080 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2081 .bdrv_snapshot_list
= sd_snapshot_list
,
2083 .bdrv_save_vmstate
= sd_save_vmstate
,
2084 .bdrv_load_vmstate
= sd_load_vmstate
,
2086 .create_options
= sd_create_options
,
2089 static void bdrv_sheepdog_init(void)
2091 bdrv_register(&bdrv_sheepdog
);
2093 block_init(bdrv_sheepdog_init
);