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 uint64_t vid_to_vdi_oid(uint32_t vid
)
232 return VDI_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
);
235 static inline uint64_t vid_to_vmstate_oid(uint32_t vid
, uint32_t idx
)
237 return VMSTATE_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
240 static inline uint64_t vid_to_data_oid(uint32_t vid
, uint32_t idx
)
242 return ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
245 static inline bool is_snapshot(struct SheepdogInode
*inode
)
247 return !!inode
->snap_ctime
;
252 #define DPRINTF(fmt, args...) \
254 fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \
257 #define DPRINTF(fmt, args...)
260 typedef struct SheepdogAIOCB SheepdogAIOCB
;
262 typedef struct AIOReq
{
263 SheepdogAIOCB
*aiocb
;
264 unsigned int iov_offset
;
269 unsigned int data_len
;
273 QLIST_ENTRY(AIOReq
) aio_siblings
;
283 struct SheepdogAIOCB
{
284 BlockDriverAIOCB common
;
292 enum AIOCBState aiocb_type
;
294 Coroutine
*coroutine
;
295 void (*aio_done_func
)(SheepdogAIOCB
*);
301 typedef struct BDRVSheepdogState
{
304 uint32_t min_dirty_data_idx
;
305 uint32_t max_dirty_data_idx
;
307 char name
[SD_MAX_VDI_LEN
];
309 uint32_t cache_flags
;
310 bool discard_supported
;
320 uint32_t aioreq_seq_num
;
321 QLIST_HEAD(inflight_aio_head
, AIOReq
) inflight_aio_head
;
322 QLIST_HEAD(pending_aio_head
, AIOReq
) pending_aio_head
;
325 static const char * sd_strerror(int err
)
329 static const struct {
333 {SD_RES_SUCCESS
, "Success"},
334 {SD_RES_UNKNOWN
, "Unknown error"},
335 {SD_RES_NO_OBJ
, "No object found"},
336 {SD_RES_EIO
, "I/O error"},
337 {SD_RES_VDI_EXIST
, "VDI exists already"},
338 {SD_RES_INVALID_PARMS
, "Invalid parameters"},
339 {SD_RES_SYSTEM_ERROR
, "System error"},
340 {SD_RES_VDI_LOCKED
, "VDI is already locked"},
341 {SD_RES_NO_VDI
, "No vdi found"},
342 {SD_RES_NO_BASE_VDI
, "No base VDI found"},
343 {SD_RES_VDI_READ
, "Failed read the requested VDI"},
344 {SD_RES_VDI_WRITE
, "Failed to write the requested VDI"},
345 {SD_RES_BASE_VDI_READ
, "Failed to read the base VDI"},
346 {SD_RES_BASE_VDI_WRITE
, "Failed to write the base VDI"},
347 {SD_RES_NO_TAG
, "Failed to find the requested tag"},
348 {SD_RES_STARTUP
, "The system is still booting"},
349 {SD_RES_VDI_NOT_LOCKED
, "VDI isn't locked"},
350 {SD_RES_SHUTDOWN
, "The system is shutting down"},
351 {SD_RES_NO_MEM
, "Out of memory on the server"},
352 {SD_RES_FULL_VDI
, "We already have the maximum vdis"},
353 {SD_RES_VER_MISMATCH
, "Protocol version mismatch"},
354 {SD_RES_NO_SPACE
, "Server has no space for new objects"},
355 {SD_RES_WAIT_FOR_FORMAT
, "Sheepdog is waiting for a format operation"},
356 {SD_RES_WAIT_FOR_JOIN
, "Sheepdog is waiting for other nodes joining"},
357 {SD_RES_JOIN_FAILED
, "Target node had failed to join sheepdog"},
358 {SD_RES_HALT
, "Sheepdog is stopped serving IO request"},
359 {SD_RES_READONLY
, "Object is read-only"},
362 for (i
= 0; i
< ARRAY_SIZE(errors
); ++i
) {
363 if (errors
[i
].err
== err
) {
364 return errors
[i
].desc
;
368 return "Invalid error code";
372 * Sheepdog I/O handling:
374 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
375 * link the requests to the inflight_list in the
376 * BDRVSheepdogState. The function exits without waiting for
377 * receiving the response.
379 * 2. We receive the response in aio_read_response, the fd handler to
380 * the sheepdog connection. If metadata update is needed, we send
381 * the write request to the vdi object in sd_write_done, the write
382 * completion function. We switch back to sd_co_readv/writev after
383 * all the requests belonging to the AIOCB are finished.
386 static inline AIOReq
*alloc_aio_req(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
,
387 uint64_t oid
, unsigned int data_len
,
388 uint64_t offset
, uint8_t flags
,
389 uint64_t base_oid
, unsigned int iov_offset
)
393 aio_req
= g_malloc(sizeof(*aio_req
));
394 aio_req
->aiocb
= acb
;
395 aio_req
->iov_offset
= iov_offset
;
397 aio_req
->base_oid
= base_oid
;
398 aio_req
->offset
= offset
;
399 aio_req
->data_len
= data_len
;
400 aio_req
->flags
= flags
;
401 aio_req
->id
= s
->aioreq_seq_num
++;
407 static inline void free_aio_req(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
409 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
411 QLIST_REMOVE(aio_req
, aio_siblings
);
417 static void coroutine_fn
sd_finish_aiocb(SheepdogAIOCB
*acb
)
419 if (!acb
->canceled
) {
420 qemu_coroutine_enter(acb
->coroutine
, NULL
);
422 qemu_aio_release(acb
);
425 static void sd_aio_cancel(BlockDriverAIOCB
*blockacb
)
427 SheepdogAIOCB
*acb
= (SheepdogAIOCB
*)blockacb
;
430 * Sheepdog cannot cancel the requests which are already sent to
431 * the servers, so we just complete the request with -EIO here.
434 qemu_coroutine_enter(acb
->coroutine
, NULL
);
435 acb
->canceled
= true;
438 static const AIOCBInfo sd_aiocb_info
= {
439 .aiocb_size
= sizeof(SheepdogAIOCB
),
440 .cancel
= sd_aio_cancel
,
443 static SheepdogAIOCB
*sd_aio_setup(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
444 int64_t sector_num
, int nb_sectors
)
448 acb
= qemu_aio_get(&sd_aiocb_info
, bs
, NULL
, NULL
);
452 acb
->sector_num
= sector_num
;
453 acb
->nb_sectors
= nb_sectors
;
455 acb
->aio_done_func
= NULL
;
456 acb
->canceled
= false;
457 acb
->coroutine
= qemu_coroutine_self();
463 static int connect_to_sdog(BDRVSheepdogState
*s
)
469 fd
= unix_connect(s
->host_spec
, &err
);
471 fd
= inet_connect(s
->host_spec
, &err
);
474 int ret
= socket_set_nodelay(fd
);
476 error_report("%s", strerror(errno
));
482 qerror_report_err(err
);
485 qemu_set_nonblock(fd
);
491 static coroutine_fn
int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
496 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
497 if (ret
!= sizeof(*hdr
)) {
498 error_report("failed to send a req, %s", strerror(errno
));
502 ret
= qemu_co_send(sockfd
, data
, *wlen
);
504 error_report("failed to send a req, %s", strerror(errno
));
510 static void restart_co_req(void *opaque
)
512 Coroutine
*co
= opaque
;
514 qemu_coroutine_enter(co
, NULL
);
517 typedef struct SheepdogReqCo
{
527 static coroutine_fn
void do_co_req(void *opaque
)
531 SheepdogReqCo
*srco
= opaque
;
532 int sockfd
= srco
->sockfd
;
533 SheepdogReq
*hdr
= srco
->hdr
;
534 void *data
= srco
->data
;
535 unsigned int *wlen
= srco
->wlen
;
536 unsigned int *rlen
= srco
->rlen
;
538 co
= qemu_coroutine_self();
539 qemu_aio_set_fd_handler(sockfd
, NULL
, restart_co_req
, co
);
541 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
546 qemu_aio_set_fd_handler(sockfd
, restart_co_req
, NULL
, co
);
548 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
549 if (ret
!= sizeof(*hdr
)) {
550 error_report("failed to get a rsp, %s", strerror(errno
));
555 if (*rlen
> hdr
->data_length
) {
556 *rlen
= hdr
->data_length
;
560 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
562 error_report("failed to get the data, %s", strerror(errno
));
569 /* there is at most one request for this sockfd, so it is safe to
570 * set each handler to NULL. */
571 qemu_aio_set_fd_handler(sockfd
, NULL
, NULL
, NULL
);
574 srco
->finished
= true;
577 static int do_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
578 unsigned int *wlen
, unsigned int *rlen
)
581 SheepdogReqCo srco
= {
591 if (qemu_in_coroutine()) {
594 co
= qemu_coroutine_create(do_co_req
);
595 qemu_coroutine_enter(co
, &srco
);
596 while (!srco
.finished
) {
604 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
605 struct iovec
*iov
, int niov
, bool create
,
606 enum AIOCBState aiocb_type
);
607 static int coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
);
610 static AIOReq
*find_pending_req(BDRVSheepdogState
*s
, uint64_t oid
)
614 QLIST_FOREACH(aio_req
, &s
->pending_aio_head
, aio_siblings
) {
615 if (aio_req
->oid
== oid
) {
624 * This function searchs pending requests to the object `oid', and
627 static void coroutine_fn
send_pending_req(BDRVSheepdogState
*s
, uint64_t oid
)
633 while ((aio_req
= find_pending_req(s
, oid
)) != NULL
) {
634 acb
= aio_req
->aiocb
;
635 /* move aio_req from pending list to inflight one */
636 QLIST_REMOVE(aio_req
, aio_siblings
);
637 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
638 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
,
639 acb
->qiov
->niov
, false, acb
->aiocb_type
);
641 error_report("add_aio_request is failed");
642 free_aio_req(s
, aio_req
);
643 if (!acb
->nr_pending
) {
644 sd_finish_aiocb(acb
);
651 * Receive responses of the I/O requests.
653 * This function is registered as a fd handler, and called from the
654 * main loop when s->fd is ready for reading responses.
656 static void coroutine_fn
aio_read_response(void *opaque
)
659 BDRVSheepdogState
*s
= opaque
;
662 AIOReq
*aio_req
= NULL
;
666 if (QLIST_EMPTY(&s
->inflight_aio_head
)) {
671 ret
= qemu_co_recv(fd
, &rsp
, sizeof(rsp
));
672 if (ret
!= sizeof(rsp
)) {
673 error_report("failed to get the header, %s", strerror(errno
));
677 /* find the right aio_req from the inflight aio list */
678 QLIST_FOREACH(aio_req
, &s
->inflight_aio_head
, aio_siblings
) {
679 if (aio_req
->id
== rsp
.id
) {
684 error_report("cannot find aio_req %x", rsp
.id
);
688 acb
= aio_req
->aiocb
;
690 switch (acb
->aiocb_type
) {
691 case AIOCB_WRITE_UDATA
:
692 /* this coroutine context is no longer suitable for co_recv
693 * because we may send data to update vdi objects */
695 if (!is_data_obj(aio_req
->oid
)) {
698 idx
= data_oid_to_idx(aio_req
->oid
);
700 if (s
->inode
.data_vdi_id
[idx
] != s
->inode
.vdi_id
) {
702 * If the object is newly created one, we need to update
703 * the vdi object (metadata object). min_dirty_data_idx
704 * and max_dirty_data_idx are changed to include updated
705 * index between them.
707 if (rsp
.result
== SD_RES_SUCCESS
) {
708 s
->inode
.data_vdi_id
[idx
] = s
->inode
.vdi_id
;
709 s
->max_dirty_data_idx
= MAX(idx
, s
->max_dirty_data_idx
);
710 s
->min_dirty_data_idx
= MIN(idx
, s
->min_dirty_data_idx
);
713 * Some requests may be blocked because simultaneous
714 * create requests are not allowed, so we search the
715 * pending requests here.
717 send_pending_req(s
, aio_req
->oid
);
720 case AIOCB_READ_UDATA
:
721 ret
= qemu_co_recvv(fd
, acb
->qiov
->iov
, acb
->qiov
->niov
,
722 aio_req
->iov_offset
, rsp
.data_length
);
723 if (ret
!= rsp
.data_length
) {
724 error_report("failed to get the data, %s", strerror(errno
));
728 case AIOCB_FLUSH_CACHE
:
729 if (rsp
.result
== SD_RES_INVALID_PARMS
) {
730 DPRINTF("disable cache since the server doesn't support it\n");
731 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
732 rsp
.result
= SD_RES_SUCCESS
;
735 case AIOCB_DISCARD_OBJ
:
736 switch (rsp
.result
) {
737 case SD_RES_INVALID_PARMS
:
738 error_report("sheep(%s) doesn't support discard command",
740 rsp
.result
= SD_RES_SUCCESS
;
741 s
->discard_supported
= false;
744 idx
= data_oid_to_idx(aio_req
->oid
);
745 s
->inode
.data_vdi_id
[idx
] = 0;
752 switch (rsp
.result
) {
755 case SD_RES_READONLY
:
756 ret
= resend_aioreq(s
, aio_req
);
757 if (ret
== SD_RES_SUCCESS
) {
763 error_report("%s", sd_strerror(rsp
.result
));
767 free_aio_req(s
, aio_req
);
768 if (!acb
->nr_pending
) {
770 * We've finished all requests which belong to the AIOCB, so
771 * we can switch back to sd_co_readv/writev now.
773 acb
->aio_done_func(acb
);
779 static void co_read_response(void *opaque
)
781 BDRVSheepdogState
*s
= opaque
;
784 s
->co_recv
= qemu_coroutine_create(aio_read_response
);
787 qemu_coroutine_enter(s
->co_recv
, opaque
);
790 static void co_write_request(void *opaque
)
792 BDRVSheepdogState
*s
= opaque
;
794 qemu_coroutine_enter(s
->co_send
, NULL
);
798 * Return a socket discriptor to read/write objects.
800 * We cannot use this discriptor for other operations because
801 * the block driver may be on waiting response from the server.
803 static int get_sheep_fd(BDRVSheepdogState
*s
)
807 fd
= connect_to_sdog(s
);
812 qemu_aio_set_fd_handler(fd
, co_read_response
, NULL
, s
);
816 static int sd_parse_uri(BDRVSheepdogState
*s
, const char *filename
,
817 char *vdi
, uint32_t *snapid
, char *tag
)
820 QueryParams
*qp
= NULL
;
823 uri
= uri_parse(filename
);
829 if (!strcmp(uri
->scheme
, "sheepdog")) {
831 } else if (!strcmp(uri
->scheme
, "sheepdog+tcp")) {
833 } else if (!strcmp(uri
->scheme
, "sheepdog+unix")) {
840 if (uri
->path
== NULL
|| !strcmp(uri
->path
, "/")) {
844 pstrcpy(vdi
, SD_MAX_VDI_LEN
, uri
->path
+ 1);
846 qp
= query_params_parse(uri
->query
);
847 if (qp
->n
> 1 || (s
->is_unix
&& !qp
->n
) || (!s
->is_unix
&& qp
->n
)) {
853 /* sheepdog+unix:///vdiname?socket=path */
854 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
858 s
->host_spec
= g_strdup(qp
->p
[0].value
);
860 /* sheepdog[+tcp]://[host:port]/vdiname */
861 s
->host_spec
= g_strdup_printf("%s:%d", uri
->server
?: SD_DEFAULT_ADDR
,
862 uri
->port
?: SD_DEFAULT_PORT
);
867 *snapid
= strtoul(uri
->fragment
, NULL
, 10);
869 pstrcpy(tag
, SD_MAX_VDI_TAG_LEN
, uri
->fragment
);
872 *snapid
= CURRENT_VDI_ID
; /* search current vdi */
877 query_params_free(qp
);
884 * Parse a filename (old syntax)
886 * filename must be one of the following formats:
888 * 2. [vdiname]:[snapid]
890 * 4. [hostname]:[port]:[vdiname]
891 * 5. [hostname]:[port]:[vdiname]:[snapid]
892 * 6. [hostname]:[port]:[vdiname]:[tag]
894 * You can boot from the snapshot images by specifying `snapid` or
897 * You can run VMs outside the Sheepdog cluster by specifying
898 * `hostname' and `port' (experimental).
900 static int parse_vdiname(BDRVSheepdogState
*s
, const char *filename
,
901 char *vdi
, uint32_t *snapid
, char *tag
)
904 const char *host_spec
, *vdi_spec
;
907 strstart(filename
, "sheepdog:", (const char **)&filename
);
908 p
= q
= g_strdup(filename
);
910 /* count the number of separators */
920 /* use the first two tokens as host_spec. */
933 p
= strchr(vdi_spec
, ':');
938 uri
= g_strdup_printf("sheepdog://%s/%s", host_spec
, vdi_spec
);
940 ret
= sd_parse_uri(s
, uri
, vdi
, snapid
, tag
);
948 static int find_vdi_name(BDRVSheepdogState
*s
, const char *filename
,
949 uint32_t snapid
, const char *tag
, uint32_t *vid
,
954 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
955 unsigned int wlen
, rlen
= 0;
956 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
958 fd
= connect_to_sdog(s
);
963 /* This pair of strncpy calls ensures that the buffer is zero-filled,
964 * which is desirable since we'll soon be sending those bytes, and
965 * don't want the send_req to read uninitialized data.
967 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
968 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
970 memset(&hdr
, 0, sizeof(hdr
));
972 hdr
.opcode
= SD_OP_LOCK_VDI
;
974 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
976 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
977 hdr
.proto_ver
= SD_PROTO_VER
;
978 hdr
.data_length
= wlen
;
980 hdr
.flags
= SD_FLAG_CMD_WRITE
;
982 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
987 if (rsp
->result
!= SD_RES_SUCCESS
) {
988 error_report("cannot get vdi info, %s, %s %d %s",
989 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
990 if (rsp
->result
== SD_RES_NO_VDI
) {
1005 static int coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
1006 struct iovec
*iov
, int niov
, bool create
,
1007 enum AIOCBState aiocb_type
)
1009 int nr_copies
= s
->inode
.nr_copies
;
1011 unsigned int wlen
= 0;
1013 uint64_t oid
= aio_req
->oid
;
1014 unsigned int datalen
= aio_req
->data_len
;
1015 uint64_t offset
= aio_req
->offset
;
1016 uint8_t flags
= aio_req
->flags
;
1017 uint64_t old_oid
= aio_req
->base_oid
;
1020 error_report("bug");
1023 memset(&hdr
, 0, sizeof(hdr
));
1025 switch (aiocb_type
) {
1026 case AIOCB_FLUSH_CACHE
:
1027 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1029 case AIOCB_READ_UDATA
:
1030 hdr
.opcode
= SD_OP_READ_OBJ
;
1033 case AIOCB_WRITE_UDATA
:
1035 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1037 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1040 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1042 case AIOCB_DISCARD_OBJ
:
1043 hdr
.opcode
= SD_OP_DISCARD_OBJ
;
1047 if (s
->cache_flags
) {
1048 hdr
.flags
|= s
->cache_flags
;
1052 hdr
.cow_oid
= old_oid
;
1053 hdr
.copies
= s
->inode
.nr_copies
;
1055 hdr
.data_length
= datalen
;
1056 hdr
.offset
= offset
;
1058 hdr
.id
= aio_req
->id
;
1060 qemu_co_mutex_lock(&s
->lock
);
1061 s
->co_send
= qemu_coroutine_self();
1062 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, co_write_request
, s
);
1063 socket_set_cork(s
->fd
, 1);
1066 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
1067 if (ret
!= sizeof(hdr
)) {
1068 qemu_co_mutex_unlock(&s
->lock
);
1069 error_report("failed to send a req, %s", strerror(errno
));
1074 ret
= qemu_co_sendv(s
->fd
, iov
, niov
, aio_req
->iov_offset
, wlen
);
1076 qemu_co_mutex_unlock(&s
->lock
);
1077 error_report("failed to send a data, %s", strerror(errno
));
1082 socket_set_cork(s
->fd
, 0);
1083 qemu_aio_set_fd_handler(s
->fd
, co_read_response
, NULL
, s
);
1084 qemu_co_mutex_unlock(&s
->lock
);
1089 static int read_write_object(int fd
, char *buf
, uint64_t oid
, uint8_t copies
,
1090 unsigned int datalen
, uint64_t offset
,
1091 bool write
, bool create
, uint32_t cache_flags
)
1094 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1095 unsigned int wlen
, rlen
;
1098 memset(&hdr
, 0, sizeof(hdr
));
1103 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1105 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1107 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1112 hdr
.opcode
= SD_OP_READ_OBJ
;
1115 hdr
.flags
|= cache_flags
;
1118 hdr
.data_length
= datalen
;
1119 hdr
.offset
= offset
;
1120 hdr
.copies
= copies
;
1122 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1124 error_report("failed to send a request to the sheep");
1128 switch (rsp
->result
) {
1129 case SD_RES_SUCCESS
:
1132 error_report("%s", sd_strerror(rsp
->result
));
1137 static int read_object(int fd
, char *buf
, uint64_t oid
, uint8_t copies
,
1138 unsigned int datalen
, uint64_t offset
,
1139 uint32_t cache_flags
)
1141 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, false,
1142 false, cache_flags
);
1145 static int write_object(int fd
, char *buf
, uint64_t oid
, uint8_t copies
,
1146 unsigned int datalen
, uint64_t offset
, bool create
,
1147 uint32_t cache_flags
)
1149 return read_write_object(fd
, buf
, oid
, copies
, datalen
, offset
, true,
1150 create
, cache_flags
);
1153 /* update inode with the latest state */
1154 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
)
1156 SheepdogInode
*inode
;
1160 fd
= connect_to_sdog(s
);
1165 inode
= g_malloc(sizeof(s
->inode
));
1167 ret
= find_vdi_name(s
, s
->name
, snapid
, tag
, &vid
, false);
1172 ret
= read_object(fd
, (char *)inode
, vid_to_vdi_oid(vid
),
1173 s
->inode
.nr_copies
, sizeof(*inode
), 0, s
->cache_flags
);
1178 if (inode
->vdi_id
!= s
->inode
.vdi_id
) {
1179 memcpy(&s
->inode
, inode
, sizeof(s
->inode
));
1189 static int coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
1191 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
1192 bool create
= false;
1195 ret
= reload_inode(s
, 0, "");
1200 aio_req
->oid
= vid_to_data_oid(s
->inode
.vdi_id
,
1201 data_oid_to_idx(aio_req
->oid
));
1203 /* check whether this request becomes a CoW one */
1204 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
) {
1205 int idx
= data_oid_to_idx(aio_req
->oid
);
1208 if (s
->inode
.data_vdi_id
[idx
] == 0) {
1212 if (is_data_obj_writable(&s
->inode
, idx
)) {
1216 /* link to the pending list if there is another CoW request to
1217 * the same object */
1218 QLIST_FOREACH(areq
, &s
->inflight_aio_head
, aio_siblings
) {
1219 if (areq
!= aio_req
&& areq
->oid
== aio_req
->oid
) {
1220 DPRINTF("simultaneous CoW to %" PRIx64
"\n", aio_req
->oid
);
1221 QLIST_REMOVE(aio_req
, aio_siblings
);
1222 QLIST_INSERT_HEAD(&s
->pending_aio_head
, aio_req
, aio_siblings
);
1223 return SD_RES_SUCCESS
;
1227 aio_req
->base_oid
= vid_to_data_oid(s
->inode
.data_vdi_id
[idx
], idx
);
1228 aio_req
->flags
|= SD_FLAG_CMD_COW
;
1232 return add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1233 create
, acb
->aiocb_type
);
1236 /* TODO Convert to fine grained options */
1237 static QemuOptsList runtime_opts
= {
1239 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1243 .type
= QEMU_OPT_STRING
,
1244 .help
= "URL to the sheepdog image",
1246 { /* end of list */ }
1250 static int sd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1255 BDRVSheepdogState
*s
= bs
->opaque
;
1256 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1260 Error
*local_err
= NULL
;
1261 const char *filename
;
1263 opts
= qemu_opts_create_nofail(&runtime_opts
);
1264 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1265 if (error_is_set(&local_err
)) {
1266 qerror_report_err(local_err
);
1267 error_free(local_err
);
1272 filename
= qemu_opt_get(opts
, "filename");
1274 QLIST_INIT(&s
->inflight_aio_head
);
1275 QLIST_INIT(&s
->pending_aio_head
);
1278 memset(vdi
, 0, sizeof(vdi
));
1279 memset(tag
, 0, sizeof(tag
));
1281 if (strstr(filename
, "://")) {
1282 ret
= sd_parse_uri(s
, filename
, vdi
, &snapid
, tag
);
1284 ret
= parse_vdiname(s
, filename
, vdi
, &snapid
, tag
);
1289 s
->fd
= get_sheep_fd(s
);
1295 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, true);
1301 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1302 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1304 s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1305 if (flags
& BDRV_O_NOCACHE
) {
1306 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1308 s
->discard_supported
= true;
1310 if (snapid
|| tag
[0] != '\0') {
1311 DPRINTF("%" PRIx32
" snapshot inode was open.\n", vid
);
1312 s
->is_snapshot
= true;
1315 fd
= connect_to_sdog(s
);
1321 buf
= g_malloc(SD_INODE_SIZE
);
1322 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), 0, SD_INODE_SIZE
, 0,
1331 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1332 s
->min_dirty_data_idx
= UINT32_MAX
;
1333 s
->max_dirty_data_idx
= 0;
1335 bs
->total_sectors
= s
->inode
.vdi_size
/ BDRV_SECTOR_SIZE
;
1336 pstrcpy(s
->name
, sizeof(s
->name
), vdi
);
1337 qemu_co_mutex_init(&s
->lock
);
1338 qemu_opts_del(opts
);
1342 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
);
1346 qemu_opts_del(opts
);
1351 static int do_sd_create(BDRVSheepdogState
*s
, char *filename
, int64_t vdi_size
,
1352 uint32_t base_vid
, uint32_t *vdi_id
, int snapshot
,
1353 uint8_t copy_policy
)
1356 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1358 unsigned int wlen
, rlen
= 0;
1359 char buf
[SD_MAX_VDI_LEN
];
1361 fd
= connect_to_sdog(s
);
1366 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1367 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1369 memset(buf
, 0, sizeof(buf
));
1370 pstrcpy(buf
, sizeof(buf
), filename
);
1372 memset(&hdr
, 0, sizeof(hdr
));
1373 hdr
.opcode
= SD_OP_NEW_VDI
;
1374 hdr
.vdi_id
= base_vid
;
1376 wlen
= SD_MAX_VDI_LEN
;
1378 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1379 hdr
.snapid
= snapshot
;
1381 hdr
.data_length
= wlen
;
1382 hdr
.vdi_size
= vdi_size
;
1383 hdr
.copy_policy
= copy_policy
;
1385 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1393 if (rsp
->result
!= SD_RES_SUCCESS
) {
1394 error_report("%s, %s", sd_strerror(rsp
->result
), filename
);
1399 *vdi_id
= rsp
->vdi_id
;
1405 static int sd_prealloc(const char *filename
)
1407 BlockDriverState
*bs
= NULL
;
1408 uint32_t idx
, max_idx
;
1410 void *buf
= g_malloc0(SD_DATA_OBJ_SIZE
);
1411 Error
*local_err
= NULL
;
1414 ret
= bdrv_file_open(&bs
, filename
, NULL
, BDRV_O_RDWR
, &local_err
);
1416 qerror_report_err(local_err
);
1417 error_free(local_err
);
1421 vdi_size
= bdrv_getlength(bs
);
1426 max_idx
= DIV_ROUND_UP(vdi_size
, SD_DATA_OBJ_SIZE
);
1428 for (idx
= 0; idx
< max_idx
; idx
++) {
1430 * The created image can be a cloned image, so we need to read
1431 * a data from the source image.
1433 ret
= bdrv_pread(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1437 ret
= bdrv_pwrite(bs
, idx
* SD_DATA_OBJ_SIZE
, buf
, SD_DATA_OBJ_SIZE
);
1451 static int sd_create(const char *filename
, QEMUOptionParameter
*options
,
1455 uint32_t vid
= 0, base_vid
= 0;
1456 int64_t vdi_size
= 0;
1457 char *backing_file
= NULL
;
1458 BDRVSheepdogState
*s
;
1459 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1461 bool prealloc
= false;
1462 Error
*local_err
= NULL
;
1464 s
= g_malloc0(sizeof(BDRVSheepdogState
));
1466 memset(vdi
, 0, sizeof(vdi
));
1467 memset(tag
, 0, sizeof(tag
));
1468 if (strstr(filename
, "://")) {
1469 ret
= sd_parse_uri(s
, filename
, vdi
, &snapid
, tag
);
1471 ret
= parse_vdiname(s
, filename
, vdi
, &snapid
, tag
);
1477 while (options
&& options
->name
) {
1478 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1479 vdi_size
= options
->value
.n
;
1480 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1481 backing_file
= options
->value
.s
;
1482 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1483 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1485 } else if (!strcmp(options
->value
.s
, "full")) {
1488 error_report("Invalid preallocation mode: '%s'",
1497 if (vdi_size
> SD_MAX_VDI_SIZE
) {
1498 error_report("too big image size");
1504 BlockDriverState
*bs
;
1505 BDRVSheepdogState
*s
;
1508 /* Currently, only Sheepdog backing image is supported. */
1509 drv
= bdrv_find_protocol(backing_file
, true);
1510 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
1511 error_report("backing_file must be a sheepdog image");
1516 ret
= bdrv_file_open(&bs
, backing_file
, NULL
, 0, &local_err
);
1518 qerror_report_err(local_err
);
1519 error_free(local_err
);
1525 if (!is_snapshot(&s
->inode
)) {
1526 error_report("cannot clone from a non snapshot vdi");
1532 base_vid
= s
->inode
.vdi_id
;
1536 /* TODO: allow users to specify copy number */
1537 ret
= do_sd_create(s
, vdi
, vdi_size
, base_vid
, &vid
, 0, 0);
1538 if (!prealloc
|| ret
) {
1542 ret
= sd_prealloc(filename
);
1548 static void sd_close(BlockDriverState
*bs
)
1550 BDRVSheepdogState
*s
= bs
->opaque
;
1552 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1553 unsigned int wlen
, rlen
= 0;
1556 DPRINTF("%s\n", s
->name
);
1558 fd
= connect_to_sdog(s
);
1563 memset(&hdr
, 0, sizeof(hdr
));
1565 hdr
.opcode
= SD_OP_RELEASE_VDI
;
1566 hdr
.vdi_id
= s
->inode
.vdi_id
;
1567 wlen
= strlen(s
->name
) + 1;
1568 hdr
.data_length
= wlen
;
1569 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1571 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, s
->name
, &wlen
, &rlen
);
1575 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
1576 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
1577 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1580 qemu_aio_set_fd_handler(s
->fd
, NULL
, NULL
, NULL
);
1582 g_free(s
->host_spec
);
1585 static int64_t sd_getlength(BlockDriverState
*bs
)
1587 BDRVSheepdogState
*s
= bs
->opaque
;
1589 return s
->inode
.vdi_size
;
1592 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
)
1594 BDRVSheepdogState
*s
= bs
->opaque
;
1596 unsigned int datalen
;
1598 if (offset
< s
->inode
.vdi_size
) {
1599 error_report("shrinking is not supported");
1601 } else if (offset
> SD_MAX_VDI_SIZE
) {
1602 error_report("too big image size");
1606 fd
= connect_to_sdog(s
);
1611 /* we don't need to update entire object */
1612 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1613 s
->inode
.vdi_size
= offset
;
1614 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1615 s
->inode
.nr_copies
, datalen
, 0, false, s
->cache_flags
);
1619 error_report("failed to update an inode.");
1626 * This function is called after writing data objects. If we need to
1627 * update metadata, this sends a write request to the vdi object.
1628 * Otherwise, this switches back to sd_co_readv/writev.
1630 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
1633 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1636 uint32_t offset
, data_len
, mn
, mx
;
1638 mn
= s
->min_dirty_data_idx
;
1639 mx
= s
->max_dirty_data_idx
;
1641 /* we need to update the vdi object. */
1642 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
1643 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
1644 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
1646 s
->min_dirty_data_idx
= UINT32_MAX
;
1647 s
->max_dirty_data_idx
= 0;
1649 iov
.iov_base
= &s
->inode
;
1650 iov
.iov_len
= sizeof(s
->inode
);
1651 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1652 data_len
, offset
, 0, 0, offset
);
1653 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1654 ret
= add_aio_request(s
, aio_req
, &iov
, 1, false, AIOCB_WRITE_UDATA
);
1656 free_aio_req(s
, aio_req
);
1661 acb
->aio_done_func
= sd_finish_aiocb
;
1662 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1666 sd_finish_aiocb(acb
);
1669 /* Delete current working VDI on the snapshot chain */
1670 static bool sd_delete(BDRVSheepdogState
*s
)
1672 unsigned int wlen
= SD_MAX_VDI_LEN
, rlen
= 0;
1673 SheepdogVdiReq hdr
= {
1674 .opcode
= SD_OP_DEL_VDI
,
1675 .vdi_id
= s
->inode
.vdi_id
,
1676 .data_length
= wlen
,
1677 .flags
= SD_FLAG_CMD_WRITE
,
1679 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1682 fd
= connect_to_sdog(s
);
1687 ret
= do_req(fd
, (SheepdogReq
*)&hdr
, s
->name
, &wlen
, &rlen
);
1692 switch (rsp
->result
) {
1694 error_report("%s was already deleted", s
->name
);
1696 case SD_RES_SUCCESS
:
1699 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1707 * Create a writable VDI from a snapshot
1709 static int sd_create_branch(BDRVSheepdogState
*s
)
1716 DPRINTF("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
1718 buf
= g_malloc(SD_INODE_SIZE
);
1721 * Even If deletion fails, we will just create extra snapshot based on
1722 * the workding VDI which was supposed to be deleted. So no need to
1725 deleted
= sd_delete(s
);
1726 ret
= do_sd_create(s
, s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &vid
,
1727 !deleted
, s
->inode
.copy_policy
);
1732 DPRINTF("%" PRIx32
" is created.\n", vid
);
1734 fd
= connect_to_sdog(s
);
1740 ret
= read_object(fd
, buf
, vid_to_vdi_oid(vid
), s
->inode
.nr_copies
,
1741 SD_INODE_SIZE
, 0, s
->cache_flags
);
1749 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1751 s
->is_snapshot
= false;
1753 DPRINTF("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
1762 * Send I/O requests to the server.
1764 * This function sends requests to the server, links the requests to
1765 * the inflight_list in BDRVSheepdogState, and exits without
1766 * waiting the response. The responses are received in the
1767 * `aio_read_response' function which is called from the main loop as
1770 * Returns 1 when we need to wait a response, 0 when there is no sent
1771 * request and -errno in error cases.
1773 static int coroutine_fn
sd_co_rw_vector(void *p
)
1775 SheepdogAIOCB
*acb
= p
;
1777 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* BDRV_SECTOR_SIZE
;
1778 unsigned long idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ SD_DATA_OBJ_SIZE
;
1780 uint64_t offset
= (acb
->sector_num
* BDRV_SECTOR_SIZE
) % SD_DATA_OBJ_SIZE
;
1781 BDRVSheepdogState
*s
= acb
->common
.bs
->opaque
;
1782 SheepdogInode
*inode
= &s
->inode
;
1785 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
1787 * In the case we open the snapshot VDI, Sheepdog creates the
1788 * writable VDI when we do a write operation first.
1790 ret
= sd_create_branch(s
);
1798 * Make sure we don't free the aiocb before we are done with all requests.
1799 * This additional reference is dropped at the end of this function.
1803 while (done
!= total
) {
1805 uint64_t old_oid
= 0;
1806 bool create
= false;
1808 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
1810 len
= MIN(total
- done
, SD_DATA_OBJ_SIZE
- offset
);
1812 switch (acb
->aiocb_type
) {
1813 case AIOCB_READ_UDATA
:
1814 if (!inode
->data_vdi_id
[idx
]) {
1815 qemu_iovec_memset(acb
->qiov
, done
, 0, len
);
1819 case AIOCB_WRITE_UDATA
:
1820 if (!inode
->data_vdi_id
[idx
]) {
1822 } else if (!is_data_obj_writable(inode
, idx
)) {
1826 flags
= SD_FLAG_CMD_COW
;
1829 case AIOCB_DISCARD_OBJ
:
1831 * We discard the object only when the whole object is
1832 * 1) allocated 2) trimmed. Otherwise, simply skip it.
1834 if (len
!= SD_DATA_OBJ_SIZE
|| inode
->data_vdi_id
[idx
] == 0) {
1843 DPRINTF("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
" %ld\n",
1845 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
1846 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
1847 DPRINTF("new oid %" PRIx64
"\n", oid
);
1850 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, old_oid
, done
);
1854 QLIST_FOREACH(areq
, &s
->inflight_aio_head
, aio_siblings
) {
1855 if (areq
->oid
== oid
) {
1857 * Sheepdog cannot handle simultaneous create
1858 * requests to the same object. So we cannot send
1859 * the request until the previous request
1863 aio_req
->base_oid
= 0;
1864 QLIST_INSERT_HEAD(&s
->pending_aio_head
, aio_req
,
1871 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1872 ret
= add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1873 create
, acb
->aiocb_type
);
1875 error_report("add_aio_request is failed");
1876 free_aio_req(s
, aio_req
);
1886 if (!--acb
->nr_pending
) {
1892 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1893 int nb_sectors
, QEMUIOVector
*qiov
)
1898 if (bs
->growable
&& sector_num
+ nb_sectors
> bs
->total_sectors
) {
1899 ret
= sd_truncate(bs
, (sector_num
+ nb_sectors
) * BDRV_SECTOR_SIZE
);
1903 bs
->total_sectors
= sector_num
+ nb_sectors
;
1906 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
);
1907 acb
->aio_done_func
= sd_write_done
;
1908 acb
->aiocb_type
= AIOCB_WRITE_UDATA
;
1910 ret
= sd_co_rw_vector(acb
);
1912 qemu_aio_release(acb
);
1916 qemu_coroutine_yield();
1921 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1922 int nb_sectors
, QEMUIOVector
*qiov
)
1927 acb
= sd_aio_setup(bs
, qiov
, sector_num
, nb_sectors
);
1928 acb
->aiocb_type
= AIOCB_READ_UDATA
;
1929 acb
->aio_done_func
= sd_finish_aiocb
;
1931 ret
= sd_co_rw_vector(acb
);
1933 qemu_aio_release(acb
);
1937 qemu_coroutine_yield();
1942 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
1944 BDRVSheepdogState
*s
= bs
->opaque
;
1949 if (s
->cache_flags
!= SD_FLAG_CMD_CACHE
) {
1953 acb
= sd_aio_setup(bs
, NULL
, 0, 0);
1954 acb
->aiocb_type
= AIOCB_FLUSH_CACHE
;
1955 acb
->aio_done_func
= sd_finish_aiocb
;
1957 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1959 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1960 ret
= add_aio_request(s
, aio_req
, NULL
, 0, false, acb
->aiocb_type
);
1962 error_report("add_aio_request is failed");
1963 free_aio_req(s
, aio_req
);
1964 qemu_aio_release(acb
);
1968 qemu_coroutine_yield();
1972 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
1974 BDRVSheepdogState
*s
= bs
->opaque
;
1977 SheepdogInode
*inode
;
1978 unsigned int datalen
;
1980 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64
" "
1981 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
1982 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
1984 if (s
->is_snapshot
) {
1985 error_report("You can't create a snapshot of a snapshot VDI, "
1986 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
1991 DPRINTF("%s %s\n", sn_info
->name
, sn_info
->id_str
);
1993 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
1994 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
1995 /* It appears that inode.tag does not require a NUL terminator,
1996 * which means this use of strncpy is ok.
1998 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
1999 /* we don't need to update entire object */
2000 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2002 /* refresh inode. */
2003 fd
= connect_to_sdog(s
);
2009 ret
= write_object(fd
, (char *)&s
->inode
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2010 s
->inode
.nr_copies
, datalen
, 0, false, s
->cache_flags
);
2012 error_report("failed to write snapshot's inode.");
2016 ret
= do_sd_create(s
, s
->name
, s
->inode
.vdi_size
, s
->inode
.vdi_id
, &new_vid
,
2017 1, s
->inode
.copy_policy
);
2019 error_report("failed to create inode for snapshot. %s",
2024 inode
= (SheepdogInode
*)g_malloc(datalen
);
2026 ret
= read_object(fd
, (char *)inode
, vid_to_vdi_oid(new_vid
),
2027 s
->inode
.nr_copies
, datalen
, 0, s
->cache_flags
);
2030 error_report("failed to read new inode info. %s", strerror(errno
));
2034 memcpy(&s
->inode
, inode
, datalen
);
2035 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2036 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
2044 * We implement rollback(loadvm) operation to the specified snapshot by
2045 * 1) switch to the snapshot
2046 * 2) rely on sd_create_branch to delete working VDI and
2047 * 3) create a new working VDI based on the speicified snapshot
2049 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
2051 BDRVSheepdogState
*s
= bs
->opaque
;
2052 BDRVSheepdogState
*old_s
;
2053 char tag
[SD_MAX_VDI_TAG_LEN
];
2054 uint32_t snapid
= 0;
2057 old_s
= g_malloc(sizeof(BDRVSheepdogState
));
2059 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
2061 snapid
= strtoul(snapshot_id
, NULL
, 10);
2065 pstrcpy(tag
, sizeof(tag
), snapshot_id
);
2068 ret
= reload_inode(s
, snapid
, tag
);
2073 ret
= sd_create_branch(s
);
2082 /* recover bdrv_sd_state */
2083 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
2086 error_report("failed to open. recover old bdrv_sd_state.");
2091 static int sd_snapshot_delete(BlockDriverState
*bs
,
2092 const char *snapshot_id
,
2096 /* FIXME: Delete specified snapshot id. */
2100 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
2102 BDRVSheepdogState
*s
= bs
->opaque
;
2104 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
2105 QEMUSnapshotInfo
*sn_tab
= NULL
;
2106 unsigned wlen
, rlen
;
2108 static SheepdogInode inode
;
2109 unsigned long *vdi_inuse
;
2110 unsigned int start_nr
;
2114 vdi_inuse
= g_malloc(max
);
2116 fd
= connect_to_sdog(s
);
2125 memset(&req
, 0, sizeof(req
));
2127 req
.opcode
= SD_OP_READ_VDIS
;
2128 req
.data_length
= max
;
2130 ret
= do_req(fd
, (SheepdogReq
*)&req
, vdi_inuse
, &wlen
, &rlen
);
2137 sn_tab
= g_malloc0(nr
* sizeof(*sn_tab
));
2139 /* calculate a vdi id with hash function */
2140 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
2141 start_nr
= hval
& (SD_NR_VDIS
- 1);
2143 fd
= connect_to_sdog(s
);
2149 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
2150 if (!test_bit(vid
, vdi_inuse
)) {
2154 /* we don't need to read entire object */
2155 ret
= read_object(fd
, (char *)&inode
, vid_to_vdi_oid(vid
),
2156 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
2163 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
2164 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
2165 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
2166 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
2167 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
2169 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
), "%u",
2171 pstrcpy(sn_tab
[found
].name
,
2172 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)),
2191 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
2192 int64_t pos
, int size
, int load
)
2195 int fd
, ret
= 0, remaining
= size
;
2196 unsigned int data_len
;
2197 uint64_t vmstate_oid
;
2200 uint32_t vdi_id
= load
? s
->inode
.parent_vdi_id
: s
->inode
.vdi_id
;
2202 fd
= connect_to_sdog(s
);
2208 vdi_index
= pos
/ SD_DATA_OBJ_SIZE
;
2209 offset
= pos
% SD_DATA_OBJ_SIZE
;
2211 data_len
= MIN(remaining
, SD_DATA_OBJ_SIZE
- offset
);
2213 vmstate_oid
= vid_to_vmstate_oid(vdi_id
, vdi_index
);
2215 create
= (offset
== 0);
2217 ret
= read_object(fd
, (char *)data
, vmstate_oid
,
2218 s
->inode
.nr_copies
, data_len
, offset
,
2221 ret
= write_object(fd
, (char *)data
, vmstate_oid
,
2222 s
->inode
.nr_copies
, data_len
, offset
, create
,
2227 error_report("failed to save vmstate %s", strerror(errno
));
2233 remaining
-= data_len
;
2241 static int sd_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2244 BDRVSheepdogState
*s
= bs
->opaque
;
2248 buf
= qemu_blockalign(bs
, qiov
->size
);
2249 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
2250 ret
= do_load_save_vmstate(s
, (uint8_t *) buf
, pos
, qiov
->size
, 0);
2256 static int sd_load_vmstate(BlockDriverState
*bs
, uint8_t *data
,
2257 int64_t pos
, int size
)
2259 BDRVSheepdogState
*s
= bs
->opaque
;
2261 return do_load_save_vmstate(s
, data
, pos
, size
, 1);
2265 static coroutine_fn
int sd_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2270 BDRVSheepdogState
*s
= bs
->opaque
;
2273 if (!s
->discard_supported
) {
2277 acb
= sd_aio_setup(bs
, &dummy
, sector_num
, nb_sectors
);
2278 acb
->aiocb_type
= AIOCB_DISCARD_OBJ
;
2279 acb
->aio_done_func
= sd_finish_aiocb
;
2281 ret
= sd_co_rw_vector(acb
);
2283 qemu_aio_release(acb
);
2287 qemu_coroutine_yield();
2292 static coroutine_fn
int64_t
2293 sd_co_get_block_status(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2296 BDRVSheepdogState
*s
= bs
->opaque
;
2297 SheepdogInode
*inode
= &s
->inode
;
2298 unsigned long start
= sector_num
* BDRV_SECTOR_SIZE
/ SD_DATA_OBJ_SIZE
,
2299 end
= DIV_ROUND_UP((sector_num
+ nb_sectors
) *
2300 BDRV_SECTOR_SIZE
, SD_DATA_OBJ_SIZE
);
2302 int64_t ret
= BDRV_BLOCK_DATA
;
2304 for (idx
= start
; idx
< end
; idx
++) {
2305 if (inode
->data_vdi_id
[idx
] == 0) {
2310 /* Get the longest length of unallocated sectors */
2312 for (idx
= start
+ 1; idx
< end
; idx
++) {
2313 if (inode
->data_vdi_id
[idx
] != 0) {
2319 *pnum
= (idx
- start
) * SD_DATA_OBJ_SIZE
/ BDRV_SECTOR_SIZE
;
2320 if (*pnum
> nb_sectors
) {
2326 static QEMUOptionParameter sd_create_options
[] = {
2328 .name
= BLOCK_OPT_SIZE
,
2330 .help
= "Virtual disk size"
2333 .name
= BLOCK_OPT_BACKING_FILE
,
2335 .help
= "File name of a base image"
2338 .name
= BLOCK_OPT_PREALLOC
,
2340 .help
= "Preallocation mode (allowed values: off, full)"
2345 static BlockDriver bdrv_sheepdog
= {
2346 .format_name
= "sheepdog",
2347 .protocol_name
= "sheepdog",
2348 .instance_size
= sizeof(BDRVSheepdogState
),
2349 .bdrv_needs_filename
= true,
2350 .bdrv_file_open
= sd_open
,
2351 .bdrv_close
= sd_close
,
2352 .bdrv_create
= sd_create
,
2353 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2354 .bdrv_getlength
= sd_getlength
,
2355 .bdrv_truncate
= sd_truncate
,
2357 .bdrv_co_readv
= sd_co_readv
,
2358 .bdrv_co_writev
= sd_co_writev
,
2359 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2360 .bdrv_co_discard
= sd_co_discard
,
2361 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2363 .bdrv_snapshot_create
= sd_snapshot_create
,
2364 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2365 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2366 .bdrv_snapshot_list
= sd_snapshot_list
,
2368 .bdrv_save_vmstate
= sd_save_vmstate
,
2369 .bdrv_load_vmstate
= sd_load_vmstate
,
2371 .create_options
= sd_create_options
,
2374 static BlockDriver bdrv_sheepdog_tcp
= {
2375 .format_name
= "sheepdog",
2376 .protocol_name
= "sheepdog+tcp",
2377 .instance_size
= sizeof(BDRVSheepdogState
),
2378 .bdrv_needs_filename
= true,
2379 .bdrv_file_open
= sd_open
,
2380 .bdrv_close
= sd_close
,
2381 .bdrv_create
= sd_create
,
2382 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2383 .bdrv_getlength
= sd_getlength
,
2384 .bdrv_truncate
= sd_truncate
,
2386 .bdrv_co_readv
= sd_co_readv
,
2387 .bdrv_co_writev
= sd_co_writev
,
2388 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2389 .bdrv_co_discard
= sd_co_discard
,
2390 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2392 .bdrv_snapshot_create
= sd_snapshot_create
,
2393 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2394 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2395 .bdrv_snapshot_list
= sd_snapshot_list
,
2397 .bdrv_save_vmstate
= sd_save_vmstate
,
2398 .bdrv_load_vmstate
= sd_load_vmstate
,
2400 .create_options
= sd_create_options
,
2403 static BlockDriver bdrv_sheepdog_unix
= {
2404 .format_name
= "sheepdog",
2405 .protocol_name
= "sheepdog+unix",
2406 .instance_size
= sizeof(BDRVSheepdogState
),
2407 .bdrv_needs_filename
= true,
2408 .bdrv_file_open
= sd_open
,
2409 .bdrv_close
= sd_close
,
2410 .bdrv_create
= sd_create
,
2411 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2412 .bdrv_getlength
= sd_getlength
,
2413 .bdrv_truncate
= sd_truncate
,
2415 .bdrv_co_readv
= sd_co_readv
,
2416 .bdrv_co_writev
= sd_co_writev
,
2417 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2418 .bdrv_co_discard
= sd_co_discard
,
2419 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2421 .bdrv_snapshot_create
= sd_snapshot_create
,
2422 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2423 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2424 .bdrv_snapshot_list
= sd_snapshot_list
,
2426 .bdrv_save_vmstate
= sd_save_vmstate
,
2427 .bdrv_load_vmstate
= sd_load_vmstate
,
2429 .create_options
= sd_create_options
,
2432 static void bdrv_sheepdog_init(void)
2434 bdrv_register(&bdrv_sheepdog
);
2435 bdrv_register(&bdrv_sheepdog_tcp
);
2436 bdrv_register(&bdrv_sheepdog_unix
);
2438 block_init(bdrv_sheepdog_init
);