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
;
263 QLIST_ENTRY(AIOReq
) aioreq_siblings
;
271 struct SheepdogAIOCB
{
272 BlockDriverAIOCB common
;
280 enum AIOCBState aiocb_type
;
282 Coroutine
*coroutine
;
283 void (*aio_done_func
)(SheepdogAIOCB
*);
287 QLIST_HEAD(aioreq_head
, AIOReq
) aioreq_head
;
290 typedef struct BDRVSheepdogState
{
293 uint32_t min_dirty_data_idx
;
294 uint32_t max_dirty_data_idx
;
296 char name
[SD_MAX_VDI_LEN
];
298 uint8_t cache_enabled
;
309 uint32_t aioreq_seq_num
;
310 QLIST_HEAD(outstanding_aio_head
, AIOReq
) outstanding_aio_head
;
313 static const char * sd_strerror(int err
)
317 static const struct {
321 {SD_RES_SUCCESS
, "Success"},
322 {SD_RES_UNKNOWN
, "Unknown error"},
323 {SD_RES_NO_OBJ
, "No object found"},
324 {SD_RES_EIO
, "I/O error"},
325 {SD_RES_VDI_EXIST
, "VDI exists already"},
326 {SD_RES_INVALID_PARMS
, "Invalid parameters"},
327 {SD_RES_SYSTEM_ERROR
, "System error"},
328 {SD_RES_VDI_LOCKED
, "VDI is already locked"},
329 {SD_RES_NO_VDI
, "No vdi found"},
330 {SD_RES_NO_BASE_VDI
, "No base VDI found"},
331 {SD_RES_VDI_READ
, "Failed read the requested VDI"},
332 {SD_RES_VDI_WRITE
, "Failed to write the requested VDI"},
333 {SD_RES_BASE_VDI_READ
, "Failed to read the base VDI"},
334 {SD_RES_BASE_VDI_WRITE
, "Failed to write the base VDI"},
335 {SD_RES_NO_TAG
, "Failed to find the requested tag"},
336 {SD_RES_STARTUP
, "The system is still booting"},
337 {SD_RES_VDI_NOT_LOCKED
, "VDI isn't locked"},
338 {SD_RES_SHUTDOWN
, "The system is shutting down"},
339 {SD_RES_NO_MEM
, "Out of memory on the server"},
340 {SD_RES_FULL_VDI
, "We already have the maximum vdis"},
341 {SD_RES_VER_MISMATCH
, "Protocol version mismatch"},
342 {SD_RES_NO_SPACE
, "Server has no space for new objects"},
343 {SD_RES_WAIT_FOR_FORMAT
, "Sheepdog is waiting for a format operation"},
344 {SD_RES_WAIT_FOR_JOIN
, "Sheepdog is waiting for other nodes joining"},
345 {SD_RES_JOIN_FAILED
, "Target node had failed to join sheepdog"},
348 for (i
= 0; i
< ARRAY_SIZE(errors
); ++i
) {
349 if (errors
[i
].err
== err
) {
350 return errors
[i
].desc
;
354 return "Invalid error code";
358 * Sheepdog I/O handling:
360 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
361 * link the requests to the outstanding_list in the
362 * BDRVSheepdogState. The function exits without waiting for
363 * receiving the response.
365 * 2. We receive the response in aio_read_response, the fd handler to
366 * the sheepdog connection. If metadata update is needed, we send
367 * the write request to the vdi object in sd_write_done, the write
368 * completion function. We switch back to sd_co_readv/writev after
369 * all the requests belonging to the AIOCB are finished.
372 static inline AIOReq
*alloc_aio_req(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
,
373 uint64_t oid
, unsigned int data_len
,
374 uint64_t offset
, uint8_t flags
,
375 uint64_t base_oid
, unsigned int iov_offset
)
379 aio_req
= g_malloc(sizeof(*aio_req
));
380 aio_req
->aiocb
= acb
;
381 aio_req
->iov_offset
= iov_offset
;
383 aio_req
->base_oid
= base_oid
;
384 aio_req
->offset
= offset
;
385 aio_req
->data_len
= data_len
;
386 aio_req
->flags
= flags
;
387 aio_req
->id
= s
->aioreq_seq_num
++;
389 QLIST_INSERT_HEAD(&s
->outstanding_aio_head
, aio_req
,
390 outstanding_aio_siblings
);
391 QLIST_INSERT_HEAD(&acb
->aioreq_head
, aio_req
, aioreq_siblings
);
396 static inline int free_aio_req(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
398 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
399 QLIST_REMOVE(aio_req
, outstanding_aio_siblings
);
400 QLIST_REMOVE(aio_req
, aioreq_siblings
);
403 return !QLIST_EMPTY(&acb
->aioreq_head
);
406 static void coroutine_fn
sd_finish_aiocb(SheepdogAIOCB
*acb
)
408 if (!acb
->canceled
) {
409 qemu_coroutine_enter(acb
->coroutine
, NULL
);
411 qemu_aio_release(acb
);
414 static void sd_aio_cancel(BlockDriverAIOCB
*blockacb
)
416 SheepdogAIOCB
*acb
= (SheepdogAIOCB
*)blockacb
;
419 * Sheepdog cannot cancel the requests which are already sent to
420 * the servers, so we just complete the request with -EIO here.
423 qemu_coroutine_enter(acb
->coroutine
, NULL
);
427 static AIOPool sd_aio_pool
= {
428 .aiocb_size
= sizeof(SheepdogAIOCB
),
429 .cancel
= sd_aio_cancel
,
432 static SheepdogAIOCB
*sd_aio_setup(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
433 int64_t sector_num
, int nb_sectors
,
434 BlockDriverCompletionFunc
*cb
, void *opaque
)
438 acb
= qemu_aio_get(&sd_aio_pool
, bs
, cb
, opaque
);
442 acb
->sector_num
= sector_num
;
443 acb
->nb_sectors
= nb_sectors
;
445 acb
->aio_done_func
= NULL
;
447 acb
->coroutine
= qemu_coroutine_self();
449 QLIST_INIT(&acb
->aioreq_head
);
453 static int connect_to_sdog(const char *addr
, const char *port
)
455 char hbuf
[NI_MAXHOST
], sbuf
[NI_MAXSERV
];
457 struct addrinfo hints
, *res
, *res0
;
460 addr
= SD_DEFAULT_ADDR
;
461 port
= SD_DEFAULT_PORT
;
464 memset(&hints
, 0, sizeof(hints
));
465 hints
.ai_socktype
= SOCK_STREAM
;
467 ret
= getaddrinfo(addr
, port
, &hints
, &res0
);
469 error_report("unable to get address info %s, %s",
470 addr
, strerror(errno
));
474 for (res
= res0
; res
; res
= res
->ai_next
) {
475 ret
= getnameinfo(res
->ai_addr
, res
->ai_addrlen
, hbuf
, sizeof(hbuf
),
476 sbuf
, sizeof(sbuf
), NI_NUMERICHOST
| NI_NUMERICSERV
);
481 fd
= socket(res
->ai_family
, res
->ai_socktype
, res
->ai_protocol
);
487 ret
= connect(fd
, res
->ai_addr
, res
->ai_addrlen
);
489 if (errno
== EINTR
) {
495 dprintf("connected to %s:%s\n", addr
, port
);
499 error_report("failed connect to %s:%s", addr
, port
);
505 static int send_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
510 ret
= qemu_send_full(sockfd
, hdr
, sizeof(*hdr
), 0);
511 if (ret
< sizeof(*hdr
)) {
512 error_report("failed to send a req, %s", strerror(errno
));
516 ret
= qemu_send_full(sockfd
, data
, *wlen
, 0);
518 error_report("failed to send a req, %s", strerror(errno
));
524 static int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
529 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
530 if (ret
< sizeof(*hdr
)) {
531 error_report("failed to send a req, %s", strerror(errno
));
535 ret
= qemu_co_send(sockfd
, data
, *wlen
);
537 error_report("failed to send a req, %s", strerror(errno
));
542 static int do_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
543 unsigned int *wlen
, unsigned int *rlen
)
547 socket_set_block(sockfd
);
548 ret
= send_req(sockfd
, hdr
, data
, wlen
);
553 ret
= qemu_recv_full(sockfd
, hdr
, sizeof(*hdr
), 0);
554 if (ret
< sizeof(*hdr
)) {
555 error_report("failed to get a rsp, %s", strerror(errno
));
559 if (*rlen
> hdr
->data_length
) {
560 *rlen
= hdr
->data_length
;
564 ret
= qemu_recv_full(sockfd
, data
, *rlen
, 0);
566 error_report("failed to get the data, %s", strerror(errno
));
572 socket_set_nonblock(sockfd
);
576 static int do_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
577 unsigned int *wlen
, unsigned int *rlen
)
581 socket_set_block(sockfd
);
582 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
587 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
588 if (ret
< sizeof(*hdr
)) {
589 error_report("failed to get a rsp, %s", strerror(errno
));
593 if (*rlen
> hdr
->data_length
) {
594 *rlen
= hdr
->data_length
;
598 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
600 error_report("failed to get the data, %s", strerror(errno
));
606 socket_set_nonblock(sockfd
);
610 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
611 struct iovec
*iov
, int niov
, int create
,
612 enum AIOCBState aiocb_type
);
615 * This function searchs pending requests to the object `oid', and
618 static void coroutine_fn
send_pending_req(BDRVSheepdogState
*s
, uint64_t oid
, uint32_t id
)
620 AIOReq
*aio_req
, *next
;
624 QLIST_FOREACH_SAFE(aio_req
, &s
->outstanding_aio_head
,
625 outstanding_aio_siblings
, next
) {
626 if (id
== aio_req
->id
) {
629 if (aio_req
->oid
!= oid
) {
633 acb
= aio_req
->aiocb
;
634 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
,
635 acb
->qiov
->niov
, 0, acb
->aiocb_type
);
637 error_report("add_aio_request is failed");
638 free_aio_req(s
, aio_req
);
639 if (QLIST_EMPTY(&acb
->aioreq_head
)) {
640 sd_finish_aiocb(acb
);
647 * Receive responses of the I/O requests.
649 * This function is registered as a fd handler, and called from the
650 * main loop when s->fd is ready for reading responses.
652 static void coroutine_fn
aio_read_response(void *opaque
)
655 BDRVSheepdogState
*s
= opaque
;
658 AIOReq
*aio_req
= NULL
;
663 if (QLIST_EMPTY(&s
->outstanding_aio_head
)) {
668 ret
= qemu_co_recv(fd
, &rsp
, sizeof(rsp
));
670 error_report("failed to get the header, %s", strerror(errno
));
674 /* find the right aio_req from the outstanding_aio list */
675 QLIST_FOREACH(aio_req
, &s
->outstanding_aio_head
, outstanding_aio_siblings
) {
676 if (aio_req
->id
== rsp
.id
) {
681 error_report("cannot find aio_req %x", rsp
.id
);
685 acb
= aio_req
->aiocb
;
687 switch (acb
->aiocb_type
) {
688 case AIOCB_WRITE_UDATA
:
689 /* this coroutine context is no longer suitable for co_recv
690 * because we may send data to update vdi objects */
692 if (!is_data_obj(aio_req
->oid
)) {
695 idx
= data_oid_to_idx(aio_req
->oid
);
697 if (s
->inode
.data_vdi_id
[idx
] != s
->inode
.vdi_id
) {
699 * If the object is newly created one, we need to update
700 * the vdi object (metadata object). min_dirty_data_idx
701 * and max_dirty_data_idx are changed to include updated
702 * index between them.
704 s
->inode
.data_vdi_id
[idx
] = s
->inode
.vdi_id
;
705 s
->max_dirty_data_idx
= MAX(idx
, s
->max_dirty_data_idx
);
706 s
->min_dirty_data_idx
= MIN(idx
, s
->min_dirty_data_idx
);
709 * Some requests may be blocked because simultaneous
710 * create requests are not allowed, so we search the
711 * pending requests here.
713 send_pending_req(s
, vid_to_data_oid(s
->inode
.vdi_id
, idx
), rsp
.id
);
716 case AIOCB_READ_UDATA
:
717 ret
= qemu_co_recvv(fd
, acb
->qiov
->iov
, rsp
.data_length
,
718 aio_req
->iov_offset
);
720 error_report("failed to get the data, %s", strerror(errno
));
726 if (rsp
.result
!= SD_RES_SUCCESS
) {
728 error_report("%s", sd_strerror(rsp
.result
));
731 rest
= free_aio_req(s
, aio_req
);
734 * We've finished all requests which belong to the AIOCB, so
735 * we can switch back to sd_co_readv/writev now.
737 acb
->aio_done_func(acb
);
743 static void co_read_response(void *opaque
)
745 BDRVSheepdogState
*s
= opaque
;
748 s
->co_recv
= qemu_coroutine_create(aio_read_response
);
751 qemu_coroutine_enter(s
->co_recv
, opaque
);
754 static void co_write_request(void *opaque
)
756 BDRVSheepdogState
*s
= opaque
;
758 qemu_coroutine_enter(s
->co_send
, NULL
);
761 static int aio_flush_request(void *opaque
)
763 BDRVSheepdogState
*s
= opaque
;
765 return !QLIST_EMPTY(&s
->outstanding_aio_head
);
768 static int set_nodelay(int fd
)
773 ret
= setsockopt(fd
, IPPROTO_TCP
, TCP_NODELAY
, (char *)&opt
, sizeof(opt
));
778 * Return a socket discriptor to read/write objects.
780 * We cannot use this discriptor for other operations because
781 * the block driver may be on waiting response from the server.
783 static int get_sheep_fd(BDRVSheepdogState
*s
)
787 fd
= connect_to_sdog(s
->addr
, s
->port
);
789 error_report("%s", strerror(errno
));
793 socket_set_nonblock(fd
);
795 ret
= set_nodelay(fd
);
797 error_report("%s", strerror(errno
));
802 qemu_aio_set_fd_handler(fd
, co_read_response
, NULL
, aio_flush_request
,
810 * filename must be one of the following formats:
812 * 2. [vdiname]:[snapid]
814 * 4. [hostname]:[port]:[vdiname]
815 * 5. [hostname]:[port]:[vdiname]:[snapid]
816 * 6. [hostname]:[port]:[vdiname]:[tag]
818 * You can boot from the snapshot images by specifying `snapid` or
821 * You can run VMs outside the Sheepdog cluster by specifying
822 * `hostname' and `port' (experimental).
824 static int parse_vdiname(BDRVSheepdogState
*s
, const char *filename
,
825 char *vdi
, uint32_t *snapid
, char *tag
)
830 p
= q
= g_strdup(filename
);
832 /* count the number of separators */
842 /* use the first two tokens as hostname and port number. */
856 strncpy(vdi
, p
, SD_MAX_VDI_LEN
);
858 p
= strchr(vdi
, ':');
861 *snapid
= strtoul(p
, NULL
, 10);
863 strncpy(tag
, p
, SD_MAX_VDI_TAG_LEN
);
866 *snapid
= CURRENT_VDI_ID
; /* search current vdi */
869 if (s
->addr
== NULL
) {
876 static int find_vdi_name(BDRVSheepdogState
*s
, char *filename
, uint32_t snapid
,
877 char *tag
, uint32_t *vid
, int for_snapshot
)
881 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
882 unsigned int wlen
, rlen
= 0;
883 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
885 fd
= connect_to_sdog(s
->addr
, s
->port
);
890 memset(buf
, 0, sizeof(buf
));
891 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
892 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
894 memset(&hdr
, 0, sizeof(hdr
));
896 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
898 hdr
.opcode
= SD_OP_LOCK_VDI
;
900 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
901 hdr
.proto_ver
= SD_PROTO_VER
;
902 hdr
.data_length
= wlen
;
904 hdr
.flags
= SD_FLAG_CMD_WRITE
;
906 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
912 if (rsp
->result
!= SD_RES_SUCCESS
) {
913 error_report("cannot get vdi info, %s, %s %d %s",
914 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
926 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
927 struct iovec
*iov
, int niov
, int create
,
928 enum AIOCBState aiocb_type
)
930 int nr_copies
= s
->inode
.nr_copies
;
934 uint64_t oid
= aio_req
->oid
;
935 unsigned int datalen
= aio_req
->data_len
;
936 uint64_t offset
= aio_req
->offset
;
937 uint8_t flags
= aio_req
->flags
;
938 uint64_t old_oid
= aio_req
->base_oid
;
944 memset(&hdr
, 0, sizeof(hdr
));
946 if (aiocb_type
== AIOCB_READ_UDATA
) {
948 hdr
.opcode
= SD_OP_READ_OBJ
;
952 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
953 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
956 hdr
.opcode
= SD_OP_WRITE_OBJ
;
957 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
960 if (s
->cache_enabled
) {
961 hdr
.flags
|= SD_FLAG_CMD_CACHE
;
965 hdr
.cow_oid
= old_oid
;
966 hdr
.copies
= s
->inode
.nr_copies
;
968 hdr
.data_length
= datalen
;
971 hdr
.id
= aio_req
->id
;
973 qemu_co_mutex_lock(&s
->lock
);
974 s
->co_send
= qemu_coroutine_self();
975 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, co_write_request
,
976 aio_flush_request
, NULL
, s
);
977 socket_set_cork(s
->fd
, 1);
980 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
982 qemu_co_mutex_unlock(&s
->lock
);
983 error_report("failed to send a req, %s", strerror(errno
));
988 ret
= qemu_co_sendv(s
->fd
, iov
, wlen
, aio_req
->iov_offset
);
990 qemu_co_mutex_unlock(&s
->lock
);
991 error_report("failed to send a data, %s", strerror(errno
));
996 socket_set_cork(s
->fd
, 0);
997 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, NULL
,
998 aio_flush_request
, NULL
, s
);
999 qemu_co_mutex_unlock(&s
->lock
);
1004 static int read_write_object(int fd
, char *buf
, uint64_t oid
, int copies
,
1005 unsigned int datalen
, uint64_t offset
,
1006 int write
, int create
, uint8_t cache
)
1009 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1010 unsigned int wlen
, rlen
;
1013 memset(&hdr
, 0, sizeof(hdr
));
1018 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1020 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1022 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1027 hdr
.opcode
= SD_OP_READ_OBJ
;
1031 hdr
.flags
|= SD_FLAG_CMD_CACHE
;
1035 hdr
.data_length
= datalen
;
1036 hdr
.offset
= offset
;
1037 hdr
.copies
= copies
;
1039 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1041 error_report("failed to send a request to the sheep");
1045 switch (rsp
->result
) {
1046 case SD_RES_SUCCESS
:
1049 error_report("%s", sd_strerror(rsp
->result
));
1054 static int read_object(int fd
, char *buf
, uint64_t oid
, int copies
,
1055 unsigned int datalen
, uint64_t offset
, uint8_t cache
)
1057 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, 0, 0,
1061 static int write_object(int fd
, char *buf
, uint64_t oid
, int copies
,
1062 unsigned int datalen
, uint64_t offset
, int create
,
1065 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, 1, create
,
1069 static int sd_open(BlockDriverState
*bs
, const char *filename
, int flags
)
1073 BDRVSheepdogState
*s
= bs
->opaque
;
1074 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1078 strstart(filename
, "sheepdog:", (const char **)&filename
);
1080 QLIST_INIT(&s
->outstanding_aio_head
);
1083 memset(vdi
, 0, sizeof(vdi
));
1084 memset(tag
, 0, sizeof(tag
));
1085 if (parse_vdiname(s
, filename
, vdi
, &snapid
, tag
) < 0) {
1088 s
->fd
= get_sheep_fd(s
);
1093 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, 0);
1098 if (flags
& BDRV_O_CACHE_WB
) {
1099 s
->cache_enabled
= 1;
1100 s
->flush_fd
= connect_to_sdog(s
->addr
, s
->port
);
1101 if (s
->flush_fd
< 0) {
1102 error_report("failed to connect");
1108 dprintf("%" PRIx32
" snapshot inode was open.\n", vid
);
1112 fd
= connect_to_sdog(s
->addr
, s
->port
);
1114 error_report("failed to connect");
1118 buf
= g_malloc(SD_INODE_SIZE
);
1119 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), 0, SD_INODE_SIZE
, 0,
1128 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1129 s
->min_dirty_data_idx
= UINT32_MAX
;
1130 s
->max_dirty_data_idx
= 0;
1132 bs
->total_sectors
= s
->inode
.vdi_size
/ SECTOR_SIZE
;
1133 strncpy(s
->name
, vdi
, sizeof(s
->name
));
1134 qemu_co_mutex_init(&s
->lock
);
1138 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
, NULL
, NULL
);
1146 static int do_sd_create(char *filename
, int64_t vdi_size
,
1147 uint32_t base_vid
, uint32_t *vdi_id
, int snapshot
,
1148 const char *addr
, const char *port
)
1151 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1153 unsigned int wlen
, rlen
= 0;
1154 char buf
[SD_MAX_VDI_LEN
];
1156 fd
= connect_to_sdog(addr
, port
);
1161 memset(buf
, 0, sizeof(buf
));
1162 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
1164 memset(&hdr
, 0, sizeof(hdr
));
1165 hdr
.opcode
= SD_OP_NEW_VDI
;
1166 hdr
.base_vdi_id
= base_vid
;
1168 wlen
= SD_MAX_VDI_LEN
;
1170 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1171 hdr
.snapid
= snapshot
;
1173 hdr
.data_length
= wlen
;
1174 hdr
.vdi_size
= vdi_size
;
1176 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1184 if (rsp
->result
!= SD_RES_SUCCESS
) {
1185 error_report("%s, %s", sd_strerror(rsp
->result
), filename
);
1190 *vdi_id
= rsp
->vdi_id
;
1196 static int sd_prealloc(const char *filename
)
1198 BlockDriverState
*bs
= NULL
;
1199 uint32_t idx
, max_idx
;
1201 void *buf
= g_malloc0(SD_DATA_OBJ_SIZE
);
1204 ret
= bdrv_file_open(&bs
, filename
, BDRV_O_RDWR
);
1209 vdi_size
= bdrv_getlength(bs
);
1214 max_idx
= DIV_ROUND_UP(vdi_size
, SD_DATA_OBJ_SIZE
);
1216 for (idx
= 0; idx
< max_idx
; idx
++) {
1218 * The created image can be a cloned image, so we need to read
1219 * a data from the source image.
1221 ret
= bdrv_pread(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1225 ret
= bdrv_pwrite(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1239 static int sd_create(const char *filename
, QEMUOptionParameter
*options
)
1242 uint32_t vid
= 0, base_vid
= 0;
1243 int64_t vdi_size
= 0;
1244 char *backing_file
= NULL
;
1245 BDRVSheepdogState s
;
1246 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1249 const char *vdiname
;
1251 strstart(filename
, "sheepdog:", &vdiname
);
1253 memset(&s
, 0, sizeof(s
));
1254 memset(vdi
, 0, sizeof(vdi
));
1255 memset(tag
, 0, sizeof(tag
));
1256 if (parse_vdiname(&s
, vdiname
, vdi
, &snapid
, tag
) < 0) {
1257 error_report("invalid filename");
1261 while (options
&& options
->name
) {
1262 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1263 vdi_size
= options
->value
.n
;
1264 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1265 backing_file
= options
->value
.s
;
1266 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1267 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1269 } else if (!strcmp(options
->value
.s
, "full")) {
1272 error_report("Invalid preallocation mode: '%s'",
1280 if (vdi_size
> SD_MAX_VDI_SIZE
) {
1281 error_report("too big image size");
1286 BlockDriverState
*bs
;
1287 BDRVSheepdogState
*s
;
1290 /* Currently, only Sheepdog backing image is supported. */
1291 drv
= bdrv_find_protocol(backing_file
);
1292 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
1293 error_report("backing_file must be a sheepdog image");
1297 ret
= bdrv_file_open(&bs
, backing_file
, 0);
1303 if (!is_snapshot(&s
->inode
)) {
1304 error_report("cannot clone from a non snapshot vdi");
1309 base_vid
= s
->inode
.vdi_id
;
1313 ret
= do_sd_create(vdi
, vdi_size
, base_vid
, &vid
, 0, s
.addr
, s
.port
);
1314 if (!prealloc
|| ret
) {
1318 return sd_prealloc(filename
);
1321 static void sd_close(BlockDriverState
*bs
)
1323 BDRVSheepdogState
*s
= bs
->opaque
;
1325 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1326 unsigned int wlen
, rlen
= 0;
1329 dprintf("%s\n", s
->name
);
1331 fd
= connect_to_sdog(s
->addr
, s
->port
);
1336 memset(&hdr
, 0, sizeof(hdr
));
1338 hdr
.opcode
= SD_OP_RELEASE_VDI
;
1339 wlen
= strlen(s
->name
) + 1;
1340 hdr
.data_length
= wlen
;
1341 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1343 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, s
->name
, &wlen
, &rlen
);
1347 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
1348 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
1349 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1352 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
, NULL
, NULL
);
1354 if (s
->cache_enabled
) {
1355 closesocket(s
->flush_fd
);
1360 static int64_t sd_getlength(BlockDriverState
*bs
)
1362 BDRVSheepdogState
*s
= bs
->opaque
;
1364 return s
->inode
.vdi_size
;
1367 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
)
1369 BDRVSheepdogState
*s
= bs
->opaque
;
1371 unsigned int datalen
;
1373 if (offset
< s
->inode
.vdi_size
) {
1374 error_report("shrinking is not supported");
1376 } else if (offset
> SD_MAX_VDI_SIZE
) {
1377 error_report("too big image size");
1381 fd
= connect_to_sdog(s
->addr
, s
->port
);
1386 /* we don't need to update entire object */
1387 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1388 s
->inode
.vdi_size
= offset
;
1389 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1390 s
->inode
.nr_copies
, datalen
, 0, 0, s
->cache_enabled
);
1394 error_report("failed to update an inode.");
1402 * This function is called after writing data objects. If we need to
1403 * update metadata, this sends a write request to the vdi object.
1404 * Otherwise, this switches back to sd_co_readv/writev.
1406 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
1409 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1412 uint32_t offset
, data_len
, mn
, mx
;
1414 mn
= s
->min_dirty_data_idx
;
1415 mx
= s
->max_dirty_data_idx
;
1417 /* we need to update the vdi object. */
1418 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
1419 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
1420 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
1422 s
->min_dirty_data_idx
= UINT32_MAX
;
1423 s
->max_dirty_data_idx
= 0;
1425 iov
.iov_base
= &s
->inode
;
1426 iov
.iov_len
= sizeof(s
->inode
);
1427 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1428 data_len
, offset
, 0, 0, offset
);
1429 ret
= add_aio_request(s
, aio_req
, &iov
, 1, 0, AIOCB_WRITE_UDATA
);
1431 free_aio_req(s
, aio_req
);
1436 acb
->aio_done_func
= sd_finish_aiocb
;
1437 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1441 sd_finish_aiocb(acb
);
1445 * Create a writable VDI from a snapshot
1447 static int sd_create_branch(BDRVSheepdogState
*s
)
1453 dprintf("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
1455 buf
= g_malloc(SD_INODE_SIZE
);
1457 ret
= do_sd_create(s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &vid
, 1,
1463 dprintf("%" PRIx32
" is created.\n", vid
);
1465 fd
= connect_to_sdog(s
->addr
, s
->port
);
1467 error_report("failed to connect");
1471 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), s
->inode
.nr_copies
,
1472 SD_INODE_SIZE
, 0, s
->cache_enabled
);
1480 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1484 dprintf("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
1493 * Send I/O requests to the server.
1495 * This function sends requests to the server, links the requests to
1496 * the outstanding_list in BDRVSheepdogState, and exits without
1497 * waiting the response. The responses are received in the
1498 * `aio_read_response' function which is called from the main loop as
1501 * Returns 1 when we need to wait a response, 0 when there is no sent
1502 * request and -errno in error cases.
1504 static int coroutine_fn
sd_co_rw_vector(void *p
)
1506 SheepdogAIOCB
*acb
= p
;
1508 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* SECTOR_SIZE
;
1509 unsigned long idx
= acb
->sector_num
* SECTOR_SIZE
/ SD_DATA_OBJ_SIZE
;
1511 uint64_t offset
= (acb
->sector_num
* SECTOR_SIZE
) % SD_DATA_OBJ_SIZE
;
1512 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1513 SheepdogInode
*inode
= &s
->inode
;
1516 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
1518 * In the case we open the snapshot VDI, Sheepdog creates the
1519 * writable VDI when we do a write operation first.
1521 ret
= sd_create_branch(s
);
1528 while (done
!= total
) {
1530 uint64_t old_oid
= 0;
1533 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
1535 len
= MIN(total
- done
, SD_DATA_OBJ_SIZE
- offset
);
1537 if (!inode
->data_vdi_id
[idx
]) {
1538 if (acb
->aiocb_type
== AIOCB_READ_UDATA
) {
1543 } else if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
1544 && !is_data_obj_writable(inode
, idx
)) {
1548 flags
= SD_FLAG_CMD_COW
;
1552 dprintf("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
1553 " %" PRIu64
"\n", inode
->vdi_id
, oid
,
1554 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
1555 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
1556 dprintf("new oid %lx\n", oid
);
1559 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, old_oid
, done
);
1563 QLIST_FOREACH(areq
, &s
->outstanding_aio_head
,
1564 outstanding_aio_siblings
) {
1565 if (areq
== aio_req
) {
1568 if (areq
->oid
== oid
) {
1570 * Sheepdog cannot handle simultaneous create
1571 * requests to the same object. So we cannot send
1572 * the request until the previous request
1576 aio_req
->base_oid
= 0;
1582 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1583 create
, acb
->aiocb_type
);
1585 error_report("add_aio_request is failed");
1586 free_aio_req(s
, aio_req
);
1596 if (QLIST_EMPTY(&acb
->aioreq_head
)) {
1602 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1603 int nb_sectors
, QEMUIOVector
*qiov
)
1608 if (bs
->growable
&& sector_num
+ nb_sectors
> bs
->total_sectors
) {
1609 /* TODO: shouldn't block here */
1610 if (sd_truncate(bs
, (sector_num
+ nb_sectors
) * SECTOR_SIZE
) < 0) {
1613 bs
->total_sectors
= sector_num
+ nb_sectors
;
1616 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
, NULL
, NULL
);
1617 acb
->aio_done_func
= sd_write_done
;
1618 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1620 ret
= sd_co_rw_vector(acb
);
1622 qemu_aio_release(acb
);
1626 qemu_coroutine_yield();
1631 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1632 int nb_sectors
, QEMUIOVector
*qiov
)
1637 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
, NULL
, NULL
);
1638 acb
->aiocb_type
= AIOCB_READ_UDATA
;
1639 acb
->aio_done_func
= sd_finish_aiocb
;
1642 * TODO: we can do better; we don't need to initialize
1645 for (i
= 0; i
< qiov
->niov
; i
++) {
1646 memset(qiov
->iov
[i
].iov_base
, 0, qiov
->iov
[i
].iov_len
);
1649 ret
= sd_co_rw_vector(acb
);
1651 qemu_aio_release(acb
);
1655 qemu_coroutine_yield();
1660 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
1662 BDRVSheepdogState
*s
= bs
->opaque
;
1663 SheepdogObjReq hdr
= { 0 };
1664 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1665 SheepdogInode
*inode
= &s
->inode
;
1667 unsigned int wlen
= 0, rlen
= 0;
1669 if (!s
->cache_enabled
) {
1673 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1674 hdr
.oid
= vid_to_vdi_oid(inode
->vdi_id
);
1676 ret
= do_co_req(s
->flush_fd
, (SheepdogReq
*)&hdr
, NULL
, &wlen
, &rlen
);
1678 error_report("failed to send a request to the sheep");
1682 if (rsp
->result
!= SD_RES_SUCCESS
) {
1683 error_report("%s", sd_strerror(rsp
->result
));
1690 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
1692 BDRVSheepdogState
*s
= bs
->opaque
;
1695 SheepdogInode
*inode
;
1696 unsigned int datalen
;
1698 dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %d "
1699 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
1700 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
1702 if (s
->is_snapshot
) {
1703 error_report("You can't create a snapshot of a snapshot VDI, "
1704 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
1709 dprintf("%s %s\n", sn_info
->name
, sn_info
->id_str
);
1711 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
1712 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
1713 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
1714 /* we don't need to update entire object */
1715 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1717 /* refresh inode. */
1718 fd
= connect_to_sdog(s
->addr
, s
->port
);
1724 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1725 s
->inode
.nr_copies
, datalen
, 0, 0, s
->cache_enabled
);
1727 error_report("failed to write snapshot's inode.");
1732 ret
= do_sd_create(s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &new_vid
, 1,
1735 error_report("failed to create inode for snapshot. %s",
1741 inode
= (SheepdogInode
*)g_malloc(datalen
);
1743 ret
= read_object(fd
, (char *)inode
, vid_to_vdi_oid(new_vid
),
1744 s
->inode
.nr_copies
, datalen
, 0, s
->cache_enabled
);
1747 error_report("failed to read new inode info. %s", strerror(errno
));
1752 memcpy(&s
->inode
, inode
, datalen
);
1753 dprintf("s->inode: name %s snap_id %x oid %x\n",
1754 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
1761 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
1763 BDRVSheepdogState
*s
= bs
->opaque
;
1764 BDRVSheepdogState
*old_s
;
1765 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1768 uint32_t snapid
= 0;
1769 int ret
= -ENOENT
, fd
;
1771 old_s
= g_malloc(sizeof(BDRVSheepdogState
));
1773 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
1775 memset(vdi
, 0, sizeof(vdi
));
1776 strncpy(vdi
, s
->name
, sizeof(vdi
));
1778 memset(tag
, 0, sizeof(tag
));
1779 snapid
= strtoul(snapshot_id
, NULL
, 10);
1781 strncpy(tag
, s
->name
, sizeof(tag
));
1784 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, 1);
1786 error_report("Failed to find_vdi_name");
1791 fd
= connect_to_sdog(s
->addr
, s
->port
);
1793 error_report("failed to connect");
1797 buf
= g_malloc(SD_INODE_SIZE
);
1798 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), s
->inode
.nr_copies
,
1799 SD_INODE_SIZE
, 0, s
->cache_enabled
);
1808 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1810 if (!s
->inode
.vm_state_size
) {
1811 error_report("Invalid snapshot");
1823 /* recover bdrv_sd_state */
1824 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
1828 error_report("failed to open. recover old bdrv_sd_state.");
1833 static int sd_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
1835 /* FIXME: Delete specified snapshot id. */
1839 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
1841 BDRVSheepdogState
*s
= bs
->opaque
;
1843 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
1844 QEMUSnapshotInfo
*sn_tab
= NULL
;
1845 unsigned wlen
, rlen
;
1847 static SheepdogInode inode
;
1848 unsigned long *vdi_inuse
;
1849 unsigned int start_nr
;
1853 vdi_inuse
= g_malloc(max
);
1855 fd
= connect_to_sdog(s
->addr
, s
->port
);
1863 memset(&req
, 0, sizeof(req
));
1865 req
.opcode
= SD_OP_READ_VDIS
;
1866 req
.data_length
= max
;
1868 ret
= do_req(fd
, (SheepdogReq
*)&req
, vdi_inuse
, &wlen
, &rlen
);
1875 sn_tab
= g_malloc0(nr
* sizeof(*sn_tab
));
1877 /* calculate a vdi id with hash function */
1878 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
1879 start_nr
= hval
& (SD_NR_VDIS
- 1);
1881 fd
= connect_to_sdog(s
->addr
, s
->port
);
1883 error_report("failed to connect");
1887 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
1888 if (!test_bit(vid
, vdi_inuse
)) {
1892 /* we don't need to read entire object */
1893 ret
= read_object(fd
, (char *)&inode
, vid_to_vdi_oid(vid
),
1894 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
1901 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
1902 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
1903 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
1904 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
1905 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
1907 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
), "%u",
1909 strncpy(sn_tab
[found
].name
, inode
.tag
,
1910 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)));
1924 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
1925 int64_t pos
, int size
, int load
)
1929 unsigned int data_len
;
1930 uint64_t vmstate_oid
;
1934 fd
= connect_to_sdog(s
->addr
, s
->port
);
1941 vdi_index
= pos
/ SD_DATA_OBJ_SIZE
;
1942 offset
= pos
% SD_DATA_OBJ_SIZE
;
1944 data_len
= MIN(size
, SD_DATA_OBJ_SIZE
);
1946 vmstate_oid
= vid_to_vmstate_oid(s
->inode
.vdi_id
, vdi_index
);
1948 create
= (offset
== 0);
1950 ret
= read_object(fd
, (char *)data
, vmstate_oid
,
1951 s
->inode
.nr_copies
, data_len
, offset
,
1954 ret
= write_object(fd
, (char *)data
, vmstate_oid
,
1955 s
->inode
.nr_copies
, data_len
, offset
, create
,
1960 error_report("failed to save vmstate %s", strerror(errno
));
1974 static int sd_save_vmstate(BlockDriverState
*bs
, const uint8_t *data
,
1975 int64_t pos
, int size
)
1977 BDRVSheepdogState
*s
= bs
->opaque
;
1979 return do_load_save_vmstate(s
, (uint8_t *)data
, pos
, size
, 0);
1982 static int sd_load_vmstate(BlockDriverState
*bs
, uint8_t *data
,
1983 int64_t pos
, int size
)
1985 BDRVSheepdogState
*s
= bs
->opaque
;
1987 return do_load_save_vmstate(s
, data
, pos
, size
, 1);
1991 static QEMUOptionParameter sd_create_options
[] = {
1993 .name
= BLOCK_OPT_SIZE
,
1995 .help
= "Virtual disk size"
1998 .name
= BLOCK_OPT_BACKING_FILE
,
2000 .help
= "File name of a base image"
2003 .name
= BLOCK_OPT_PREALLOC
,
2005 .help
= "Preallocation mode (allowed values: off, full)"
2010 BlockDriver bdrv_sheepdog
= {
2011 .format_name
= "sheepdog",
2012 .protocol_name
= "sheepdog",
2013 .instance_size
= sizeof(BDRVSheepdogState
),
2014 .bdrv_file_open
= sd_open
,
2015 .bdrv_close
= sd_close
,
2016 .bdrv_create
= sd_create
,
2017 .bdrv_getlength
= sd_getlength
,
2018 .bdrv_truncate
= sd_truncate
,
2020 .bdrv_co_readv
= sd_co_readv
,
2021 .bdrv_co_writev
= sd_co_writev
,
2022 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2024 .bdrv_snapshot_create
= sd_snapshot_create
,
2025 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2026 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2027 .bdrv_snapshot_list
= sd_snapshot_list
,
2029 .bdrv_save_vmstate
= sd_save_vmstate
,
2030 .bdrv_load_vmstate
= sd_load_vmstate
,
2032 .create_options
= sd_create_options
,
2035 static void bdrv_sheepdog_init(void)
2037 bdrv_register(&bdrv_sheepdog
);
2039 block_init(bdrv_sheepdog_init
);