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/osdep.h"
16 #include "qapi/error.h"
18 #include "qemu/error-report.h"
19 #include "qemu/sockets.h"
20 #include "block/block_int.h"
21 #include "sysemu/block-backend.h"
22 #include "qemu/bitops.h"
23 #include "qemu/cutils.h"
25 #define SD_PROTO_VER 0x01
27 #define SD_DEFAULT_ADDR "localhost"
28 #define SD_DEFAULT_PORT 7000
30 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
31 #define SD_OP_READ_OBJ 0x02
32 #define SD_OP_WRITE_OBJ 0x03
33 /* 0x04 is used internally by Sheepdog */
35 #define SD_OP_NEW_VDI 0x11
36 #define SD_OP_LOCK_VDI 0x12
37 #define SD_OP_RELEASE_VDI 0x13
38 #define SD_OP_GET_VDI_INFO 0x14
39 #define SD_OP_READ_VDIS 0x15
40 #define SD_OP_FLUSH_VDI 0x16
41 #define SD_OP_DEL_VDI 0x17
42 #define SD_OP_GET_CLUSTER_DEFAULT 0x18
44 #define SD_FLAG_CMD_WRITE 0x01
45 #define SD_FLAG_CMD_COW 0x02
46 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
47 #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
49 #define SD_RES_SUCCESS 0x00 /* Success */
50 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
51 #define SD_RES_NO_OBJ 0x02 /* No object found */
52 #define SD_RES_EIO 0x03 /* I/O error */
53 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
54 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
55 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
56 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
57 #define SD_RES_NO_VDI 0x08 /* No vdi found */
58 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
59 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
60 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
61 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
62 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
63 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
64 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
65 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
66 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
67 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
68 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
69 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
70 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
71 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
72 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
73 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
74 #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */
75 #define SD_RES_READONLY 0x1A /* Object is read-only */
80 * 0 - 19 (20 bits): data object space
81 * 20 - 31 (12 bits): reserved data object space
82 * 32 - 55 (24 bits): vdi object space
83 * 56 - 59 ( 4 bits): reserved vdi object space
84 * 60 - 63 ( 4 bits): object type identifier space
87 #define VDI_SPACE_SHIFT 32
88 #define VDI_BIT (UINT64_C(1) << 63)
89 #define VMSTATE_BIT (UINT64_C(1) << 62)
90 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
91 #define MAX_CHILDREN 1024
92 #define SD_MAX_VDI_LEN 256
93 #define SD_MAX_VDI_TAG_LEN 256
94 #define SD_NR_VDIS (1U << 24)
95 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
96 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
97 #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22
99 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
100 * (SD_EC_MAX_STRIP - 1) for parity strips
102 * SD_MAX_COPIES is sum of number of data strips and parity strips.
104 #define SD_EC_MAX_STRIP 16
105 #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
107 #define SD_INODE_SIZE (sizeof(SheepdogInode))
108 #define CURRENT_VDI_ID 0
110 #define LOCK_TYPE_NORMAL 0
111 #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */
113 typedef struct SheepdogReq
{
119 uint32_t data_length
;
120 uint32_t opcode_specific
[8];
123 typedef struct SheepdogRsp
{
129 uint32_t data_length
;
131 uint32_t opcode_specific
[7];
134 typedef struct SheepdogObjReq
{
140 uint32_t data_length
;
149 typedef struct SheepdogObjRsp
{
155 uint32_t data_length
;
163 typedef struct SheepdogVdiReq
{
169 uint32_t data_length
;
171 uint32_t base_vdi_id
;
174 uint8_t store_policy
;
175 uint8_t block_size_shift
;
181 typedef struct SheepdogVdiRsp
{
187 uint32_t data_length
;
194 typedef struct SheepdogClusterRsp
{
200 uint32_t data_length
;
204 uint8_t block_size_shift
;
207 } SheepdogClusterRsp
;
209 typedef struct SheepdogInode
{
210 char name
[SD_MAX_VDI_LEN
];
211 char tag
[SD_MAX_VDI_TAG_LEN
];
214 uint64_t vm_clock_nsec
;
216 uint64_t vm_state_size
;
217 uint16_t copy_policy
;
219 uint8_t block_size_shift
;
222 uint32_t parent_vdi_id
;
223 uint32_t child_vdi_id
[MAX_CHILDREN
];
224 uint32_t data_vdi_id
[MAX_DATA_OBJS
];
227 #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
230 * 64 bit FNV-1a non-zero initial basis
232 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
235 * 64 bit Fowler/Noll/Vo FNV-1a hash code
237 static inline uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
239 unsigned char *bp
= buf
;
240 unsigned char *be
= bp
+ len
;
242 hval
^= (uint64_t) *bp
++;
243 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
244 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
249 static inline bool is_data_obj_writable(SheepdogInode
*inode
, unsigned int idx
)
251 return inode
->vdi_id
== inode
->data_vdi_id
[idx
];
254 static inline bool is_data_obj(uint64_t oid
)
256 return !(VDI_BIT
& oid
);
259 static inline uint64_t data_oid_to_idx(uint64_t oid
)
261 return oid
& (MAX_DATA_OBJS
- 1);
264 static inline uint32_t oid_to_vid(uint64_t oid
)
266 return (oid
& ~VDI_BIT
) >> VDI_SPACE_SHIFT
;
269 static inline uint64_t vid_to_vdi_oid(uint32_t vid
)
271 return VDI_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
);
274 static inline uint64_t vid_to_vmstate_oid(uint32_t vid
, uint32_t idx
)
276 return VMSTATE_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
279 static inline uint64_t vid_to_data_oid(uint32_t vid
, uint32_t idx
)
281 return ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
284 static inline bool is_snapshot(struct SheepdogInode
*inode
)
286 return !!inode
->snap_ctime
;
289 static inline size_t count_data_objs(const struct SheepdogInode
*inode
)
291 return DIV_ROUND_UP(inode
->vdi_size
,
292 (1UL << inode
->block_size_shift
));
297 #define DEBUG_SDOG_PRINT 1
299 #define DEBUG_SDOG_PRINT 0
301 #define DPRINTF(fmt, args...) \
303 if (DEBUG_SDOG_PRINT) { \
304 fprintf(stderr, "%s %d: " fmt, __func__, __LINE__, ##args); \
308 typedef struct SheepdogAIOCB SheepdogAIOCB
;
309 typedef struct BDRVSheepdogState BDRVSheepdogState
;
311 typedef struct AIOReq
{
312 SheepdogAIOCB
*aiocb
;
313 unsigned int iov_offset
;
318 unsigned int data_len
;
323 QLIST_ENTRY(AIOReq
) aio_siblings
;
333 #define AIOCBOverlapping(x, y) \
334 (!(x->max_affect_data_idx < y->min_affect_data_idx \
335 || y->max_affect_data_idx < x->min_affect_data_idx))
337 struct SheepdogAIOCB
{
338 BDRVSheepdogState
*s
;
346 enum AIOCBState aiocb_type
;
348 Coroutine
*coroutine
;
351 uint32_t min_affect_data_idx
;
352 uint32_t max_affect_data_idx
;
355 * The difference between affect_data_idx and dirty_data_idx:
356 * affect_data_idx represents range of index of all request types.
357 * dirty_data_idx represents range of index updated by COW requests.
358 * dirty_data_idx is used for updating an inode object.
360 uint32_t min_dirty_data_idx
;
361 uint32_t max_dirty_data_idx
;
363 QLIST_ENTRY(SheepdogAIOCB
) aiocb_siblings
;
366 struct BDRVSheepdogState
{
367 BlockDriverState
*bs
;
368 AioContext
*aio_context
;
372 char name
[SD_MAX_VDI_LEN
];
374 uint32_t cache_flags
;
375 bool discard_supported
;
385 uint32_t aioreq_seq_num
;
387 /* Every aio request must be linked to either of these queues. */
388 QLIST_HEAD(inflight_aio_head
, AIOReq
) inflight_aio_head
;
389 QLIST_HEAD(failed_aio_head
, AIOReq
) failed_aio_head
;
391 CoQueue overlapping_queue
;
392 QLIST_HEAD(inflight_aiocb_head
, SheepdogAIOCB
) inflight_aiocb_head
;
395 typedef struct BDRVSheepdogReopenState
{
398 } BDRVSheepdogReopenState
;
400 static const char * sd_strerror(int err
)
404 static const struct {
408 {SD_RES_SUCCESS
, "Success"},
409 {SD_RES_UNKNOWN
, "Unknown error"},
410 {SD_RES_NO_OBJ
, "No object found"},
411 {SD_RES_EIO
, "I/O error"},
412 {SD_RES_VDI_EXIST
, "VDI exists already"},
413 {SD_RES_INVALID_PARMS
, "Invalid parameters"},
414 {SD_RES_SYSTEM_ERROR
, "System error"},
415 {SD_RES_VDI_LOCKED
, "VDI is already locked"},
416 {SD_RES_NO_VDI
, "No vdi found"},
417 {SD_RES_NO_BASE_VDI
, "No base VDI found"},
418 {SD_RES_VDI_READ
, "Failed read the requested VDI"},
419 {SD_RES_VDI_WRITE
, "Failed to write the requested VDI"},
420 {SD_RES_BASE_VDI_READ
, "Failed to read the base VDI"},
421 {SD_RES_BASE_VDI_WRITE
, "Failed to write the base VDI"},
422 {SD_RES_NO_TAG
, "Failed to find the requested tag"},
423 {SD_RES_STARTUP
, "The system is still booting"},
424 {SD_RES_VDI_NOT_LOCKED
, "VDI isn't locked"},
425 {SD_RES_SHUTDOWN
, "The system is shutting down"},
426 {SD_RES_NO_MEM
, "Out of memory on the server"},
427 {SD_RES_FULL_VDI
, "We already have the maximum vdis"},
428 {SD_RES_VER_MISMATCH
, "Protocol version mismatch"},
429 {SD_RES_NO_SPACE
, "Server has no space for new objects"},
430 {SD_RES_WAIT_FOR_FORMAT
, "Sheepdog is waiting for a format operation"},
431 {SD_RES_WAIT_FOR_JOIN
, "Sheepdog is waiting for other nodes joining"},
432 {SD_RES_JOIN_FAILED
, "Target node had failed to join sheepdog"},
433 {SD_RES_HALT
, "Sheepdog is stopped serving IO request"},
434 {SD_RES_READONLY
, "Object is read-only"},
437 for (i
= 0; i
< ARRAY_SIZE(errors
); ++i
) {
438 if (errors
[i
].err
== err
) {
439 return errors
[i
].desc
;
443 return "Invalid error code";
447 * Sheepdog I/O handling:
449 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
450 * link the requests to the inflight_list in the
451 * BDRVSheepdogState. The function yields while waiting for
452 * receiving the response.
454 * 2. We receive the response in aio_read_response, the fd handler to
455 * the sheepdog connection. We switch back to sd_co_readv/sd_writev
456 * after all the requests belonging to the AIOCB are finished. If
457 * needed, sd_co_writev will send another requests for the vdi object.
460 static inline AIOReq
*alloc_aio_req(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
,
461 uint64_t oid
, unsigned int data_len
,
462 uint64_t offset
, uint8_t flags
, bool create
,
463 uint64_t base_oid
, unsigned int iov_offset
)
467 aio_req
= g_malloc(sizeof(*aio_req
));
468 aio_req
->aiocb
= acb
;
469 aio_req
->iov_offset
= iov_offset
;
471 aio_req
->base_oid
= base_oid
;
472 aio_req
->offset
= offset
;
473 aio_req
->data_len
= data_len
;
474 aio_req
->flags
= flags
;
475 aio_req
->id
= s
->aioreq_seq_num
++;
476 aio_req
->create
= create
;
482 static void wait_for_overlapping_aiocb(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
)
487 QLIST_FOREACH(cb
, &s
->inflight_aiocb_head
, aiocb_siblings
) {
488 if (AIOCBOverlapping(acb
, cb
)) {
489 qemu_co_queue_wait(&s
->overlapping_queue
, NULL
);
495 static void sd_aio_setup(SheepdogAIOCB
*acb
, BDRVSheepdogState
*s
,
496 QEMUIOVector
*qiov
, int64_t sector_num
, int nb_sectors
,
499 uint32_t object_size
;
501 object_size
= (UINT32_C(1) << s
->inode
.block_size_shift
);
507 acb
->sector_num
= sector_num
;
508 acb
->nb_sectors
= nb_sectors
;
510 acb
->coroutine
= qemu_coroutine_self();
514 acb
->min_affect_data_idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ object_size
;
515 acb
->max_affect_data_idx
= (acb
->sector_num
* BDRV_SECTOR_SIZE
+
516 acb
->nb_sectors
* BDRV_SECTOR_SIZE
) / object_size
;
518 acb
->min_dirty_data_idx
= UINT32_MAX
;
519 acb
->max_dirty_data_idx
= 0;
520 acb
->aiocb_type
= type
;
522 if (type
== AIOCB_FLUSH_CACHE
) {
526 wait_for_overlapping_aiocb(s
, acb
);
527 QLIST_INSERT_HEAD(&s
->inflight_aiocb_head
, acb
, aiocb_siblings
);
530 /* Return -EIO in case of error, file descriptor on success */
531 static int connect_to_sdog(BDRVSheepdogState
*s
, Error
**errp
)
536 fd
= unix_connect(s
->host_spec
, errp
);
538 fd
= inet_connect(s
->host_spec
, errp
);
541 int ret
= socket_set_nodelay(fd
);
543 error_report("%s", strerror(errno
));
549 qemu_set_nonblock(fd
);
557 /* Return 0 on success and -errno in case of error */
558 static coroutine_fn
int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
563 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
564 if (ret
!= sizeof(*hdr
)) {
565 error_report("failed to send a req, %s", strerror(errno
));
569 ret
= qemu_co_send(sockfd
, data
, *wlen
);
571 error_report("failed to send a req, %s", strerror(errno
));
578 typedef struct SheepdogReqCo
{
580 BlockDriverState
*bs
;
581 AioContext
*aio_context
;
591 static void restart_co_req(void *opaque
)
593 SheepdogReqCo
*srco
= opaque
;
595 aio_co_wake(srco
->co
);
598 static coroutine_fn
void do_co_req(void *opaque
)
601 SheepdogReqCo
*srco
= opaque
;
602 int sockfd
= srco
->sockfd
;
603 SheepdogReq
*hdr
= srco
->hdr
;
604 void *data
= srco
->data
;
605 unsigned int *wlen
= srco
->wlen
;
606 unsigned int *rlen
= srco
->rlen
;
608 srco
->co
= qemu_coroutine_self();
609 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
610 NULL
, restart_co_req
, NULL
, srco
);
612 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
617 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
618 restart_co_req
, NULL
, NULL
, srco
);
620 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
621 if (ret
!= sizeof(*hdr
)) {
622 error_report("failed to get a rsp, %s", strerror(errno
));
627 if (*rlen
> hdr
->data_length
) {
628 *rlen
= hdr
->data_length
;
632 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
634 error_report("failed to get the data, %s", strerror(errno
));
641 /* there is at most one request for this sockfd, so it is safe to
642 * set each handler to NULL. */
643 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
644 NULL
, NULL
, NULL
, NULL
);
648 srco
->finished
= true;
650 bdrv_wakeup(srco
->bs
);
655 * Send the request to the sheep in a synchronous manner.
657 * Return 0 on success, -errno in case of error.
659 static int do_req(int sockfd
, BlockDriverState
*bs
, SheepdogReq
*hdr
,
660 void *data
, unsigned int *wlen
, unsigned int *rlen
)
663 SheepdogReqCo srco
= {
665 .aio_context
= bs
? bdrv_get_aio_context(bs
) : qemu_get_aio_context(),
675 if (qemu_in_coroutine()) {
678 co
= qemu_coroutine_create(do_co_req
, &srco
);
680 qemu_coroutine_enter(co
);
681 BDRV_POLL_WHILE(bs
, !srco
.finished
);
683 qemu_coroutine_enter(co
);
684 while (!srco
.finished
) {
685 aio_poll(qemu_get_aio_context(), true);
693 static void coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
694 struct iovec
*iov
, int niov
,
695 enum AIOCBState aiocb_type
);
696 static void coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
);
697 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
);
698 static int get_sheep_fd(BDRVSheepdogState
*s
, Error
**errp
);
699 static void co_write_request(void *opaque
);
701 static coroutine_fn
void reconnect_to_sdog(void *opaque
)
703 BDRVSheepdogState
*s
= opaque
;
704 AIOReq
*aio_req
, *next
;
706 aio_set_fd_handler(s
->aio_context
, s
->fd
, false, NULL
,
711 /* Wait for outstanding write requests to be completed. */
712 while (s
->co_send
!= NULL
) {
713 co_write_request(opaque
);
716 /* Try to reconnect the sheepdog server every one second. */
718 Error
*local_err
= NULL
;
719 s
->fd
= get_sheep_fd(s
, &local_err
);
721 DPRINTF("Wait for connection to be established\n");
722 error_report_err(local_err
);
723 co_aio_sleep_ns(bdrv_get_aio_context(s
->bs
), QEMU_CLOCK_REALTIME
,
729 * Now we have to resend all the request in the inflight queue. However,
730 * resend_aioreq() can yield and newly created requests can be added to the
731 * inflight queue before the coroutine is resumed. To avoid mixing them, we
732 * have to move all the inflight requests to the failed queue before
733 * resend_aioreq() is called.
735 QLIST_FOREACH_SAFE(aio_req
, &s
->inflight_aio_head
, aio_siblings
, next
) {
736 QLIST_REMOVE(aio_req
, aio_siblings
);
737 QLIST_INSERT_HEAD(&s
->failed_aio_head
, aio_req
, aio_siblings
);
740 /* Resend all the failed aio requests. */
741 while (!QLIST_EMPTY(&s
->failed_aio_head
)) {
742 aio_req
= QLIST_FIRST(&s
->failed_aio_head
);
743 QLIST_REMOVE(aio_req
, aio_siblings
);
744 resend_aioreq(s
, aio_req
);
749 * Receive responses of the I/O requests.
751 * This function is registered as a fd handler, and called from the
752 * main loop when s->fd is ready for reading responses.
754 static void coroutine_fn
aio_read_response(void *opaque
)
757 BDRVSheepdogState
*s
= opaque
;
760 AIOReq
*aio_req
= NULL
;
765 ret
= qemu_co_recv(fd
, &rsp
, sizeof(rsp
));
766 if (ret
!= sizeof(rsp
)) {
767 error_report("failed to get the header, %s", strerror(errno
));
771 /* find the right aio_req from the inflight aio list */
772 QLIST_FOREACH(aio_req
, &s
->inflight_aio_head
, aio_siblings
) {
773 if (aio_req
->id
== rsp
.id
) {
778 error_report("cannot find aio_req %x", rsp
.id
);
782 acb
= aio_req
->aiocb
;
784 switch (acb
->aiocb_type
) {
785 case AIOCB_WRITE_UDATA
:
786 if (!is_data_obj(aio_req
->oid
)) {
789 idx
= data_oid_to_idx(aio_req
->oid
);
791 if (aio_req
->create
) {
793 * If the object is newly created one, we need to update
794 * the vdi object (metadata object). min_dirty_data_idx
795 * and max_dirty_data_idx are changed to include updated
796 * index between them.
798 if (rsp
.result
== SD_RES_SUCCESS
) {
799 s
->inode
.data_vdi_id
[idx
] = s
->inode
.vdi_id
;
800 acb
->max_dirty_data_idx
= MAX(idx
, acb
->max_dirty_data_idx
);
801 acb
->min_dirty_data_idx
= MIN(idx
, acb
->min_dirty_data_idx
);
805 case AIOCB_READ_UDATA
:
806 ret
= qemu_co_recvv(fd
, acb
->qiov
->iov
, acb
->qiov
->niov
,
807 aio_req
->iov_offset
, rsp
.data_length
);
808 if (ret
!= rsp
.data_length
) {
809 error_report("failed to get the data, %s", strerror(errno
));
813 case AIOCB_FLUSH_CACHE
:
814 if (rsp
.result
== SD_RES_INVALID_PARMS
) {
815 DPRINTF("disable cache since the server doesn't support it\n");
816 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
817 rsp
.result
= SD_RES_SUCCESS
;
820 case AIOCB_DISCARD_OBJ
:
821 switch (rsp
.result
) {
822 case SD_RES_INVALID_PARMS
:
823 error_report("sheep(%s) doesn't support discard command",
825 rsp
.result
= SD_RES_SUCCESS
;
826 s
->discard_supported
= false;
833 /* No more data for this aio_req (reload_inode below uses its own file
834 * descriptor handler which doesn't use co_recv).
838 QLIST_REMOVE(aio_req
, aio_siblings
);
839 switch (rsp
.result
) {
842 case SD_RES_READONLY
:
843 if (s
->inode
.vdi_id
== oid_to_vid(aio_req
->oid
)) {
844 ret
= reload_inode(s
, 0, "");
849 if (is_data_obj(aio_req
->oid
)) {
850 aio_req
->oid
= vid_to_data_oid(s
->inode
.vdi_id
,
851 data_oid_to_idx(aio_req
->oid
));
853 aio_req
->oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
855 resend_aioreq(s
, aio_req
);
859 error_report("%s", sd_strerror(rsp
.result
));
865 if (!--acb
->nr_pending
) {
867 * We've finished all requests which belong to the AIOCB, so
868 * we can switch back to sd_co_readv/writev now.
870 aio_co_wake(acb
->coroutine
);
876 reconnect_to_sdog(opaque
);
879 static void co_read_response(void *opaque
)
881 BDRVSheepdogState
*s
= opaque
;
884 s
->co_recv
= qemu_coroutine_create(aio_read_response
, opaque
);
887 aio_co_wake(s
->co_recv
);
890 static void co_write_request(void *opaque
)
892 BDRVSheepdogState
*s
= opaque
;
894 aio_co_wake(s
->co_send
);
898 * Return a socket descriptor to read/write objects.
900 * We cannot use this descriptor for other operations because
901 * the block driver may be on waiting response from the server.
903 static int get_sheep_fd(BDRVSheepdogState
*s
, Error
**errp
)
907 fd
= connect_to_sdog(s
, errp
);
912 aio_set_fd_handler(s
->aio_context
, fd
, false,
913 co_read_response
, NULL
, NULL
, s
);
917 static int sd_parse_uri(BDRVSheepdogState
*s
, const char *filename
,
918 char *vdi
, uint32_t *snapid
, char *tag
)
921 QueryParams
*qp
= NULL
;
924 uri
= uri_parse(filename
);
930 if (!strcmp(uri
->scheme
, "sheepdog")) {
932 } else if (!strcmp(uri
->scheme
, "sheepdog+tcp")) {
934 } else if (!strcmp(uri
->scheme
, "sheepdog+unix")) {
941 if (uri
->path
== NULL
|| !strcmp(uri
->path
, "/")) {
945 pstrcpy(vdi
, SD_MAX_VDI_LEN
, uri
->path
+ 1);
947 qp
= query_params_parse(uri
->query
);
948 if (qp
->n
> 1 || (s
->is_unix
&& !qp
->n
) || (!s
->is_unix
&& qp
->n
)) {
954 /* sheepdog+unix:///vdiname?socket=path */
955 if (uri
->server
|| uri
->port
|| strcmp(qp
->p
[0].name
, "socket")) {
959 s
->host_spec
= g_strdup(qp
->p
[0].value
);
961 /* sheepdog[+tcp]://[host:port]/vdiname */
962 s
->host_spec
= g_strdup_printf("%s:%d", uri
->server
?: SD_DEFAULT_ADDR
,
963 uri
->port
?: SD_DEFAULT_PORT
);
968 *snapid
= strtoul(uri
->fragment
, NULL
, 10);
970 pstrcpy(tag
, SD_MAX_VDI_TAG_LEN
, uri
->fragment
);
973 *snapid
= CURRENT_VDI_ID
; /* search current vdi */
978 query_params_free(qp
);
985 * Parse a filename (old syntax)
987 * filename must be one of the following formats:
989 * 2. [vdiname]:[snapid]
991 * 4. [hostname]:[port]:[vdiname]
992 * 5. [hostname]:[port]:[vdiname]:[snapid]
993 * 6. [hostname]:[port]:[vdiname]:[tag]
995 * You can boot from the snapshot images by specifying `snapid` or
998 * You can run VMs outside the Sheepdog cluster by specifying
999 * `hostname' and `port' (experimental).
1001 static int parse_vdiname(BDRVSheepdogState
*s
, const char *filename
,
1002 char *vdi
, uint32_t *snapid
, char *tag
)
1005 const char *host_spec
, *vdi_spec
;
1008 strstart(filename
, "sheepdog:", &filename
);
1009 p
= q
= g_strdup(filename
);
1011 /* count the number of separators */
1021 /* use the first two tokens as host_spec. */
1034 p
= strchr(vdi_spec
, ':');
1039 uri
= g_strdup_printf("sheepdog://%s/%s", host_spec
, vdi_spec
);
1041 ret
= sd_parse_uri(s
, uri
, vdi
, snapid
, tag
);
1049 static int find_vdi_name(BDRVSheepdogState
*s
, const char *filename
,
1050 uint32_t snapid
, const char *tag
, uint32_t *vid
,
1051 bool lock
, Error
**errp
)
1055 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1056 unsigned int wlen
, rlen
= 0;
1057 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
1059 fd
= connect_to_sdog(s
, errp
);
1064 /* This pair of strncpy calls ensures that the buffer is zero-filled,
1065 * which is desirable since we'll soon be sending those bytes, and
1066 * don't want the send_req to read uninitialized data.
1068 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
1069 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
1071 memset(&hdr
, 0, sizeof(hdr
));
1073 hdr
.opcode
= SD_OP_LOCK_VDI
;
1074 hdr
.type
= LOCK_TYPE_NORMAL
;
1076 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
1078 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
1079 hdr
.proto_ver
= SD_PROTO_VER
;
1080 hdr
.data_length
= wlen
;
1081 hdr
.snapid
= snapid
;
1082 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1084 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1086 error_setg_errno(errp
, -ret
, "cannot get vdi info");
1090 if (rsp
->result
!= SD_RES_SUCCESS
) {
1091 error_setg(errp
, "cannot get vdi info, %s, %s %" PRIu32
" %s",
1092 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
1093 if (rsp
->result
== SD_RES_NO_VDI
) {
1095 } else if (rsp
->result
== SD_RES_VDI_LOCKED
) {
1110 static void coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
1111 struct iovec
*iov
, int niov
,
1112 enum AIOCBState aiocb_type
)
1114 int nr_copies
= s
->inode
.nr_copies
;
1116 unsigned int wlen
= 0;
1118 uint64_t oid
= aio_req
->oid
;
1119 unsigned int datalen
= aio_req
->data_len
;
1120 uint64_t offset
= aio_req
->offset
;
1121 uint8_t flags
= aio_req
->flags
;
1122 uint64_t old_oid
= aio_req
->base_oid
;
1123 bool create
= aio_req
->create
;
1125 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1128 error_report("bug");
1131 memset(&hdr
, 0, sizeof(hdr
));
1133 switch (aiocb_type
) {
1134 case AIOCB_FLUSH_CACHE
:
1135 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1137 case AIOCB_READ_UDATA
:
1138 hdr
.opcode
= SD_OP_READ_OBJ
;
1141 case AIOCB_WRITE_UDATA
:
1143 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1145 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1148 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1150 case AIOCB_DISCARD_OBJ
:
1151 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1152 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1153 s
->inode
.data_vdi_id
[data_oid_to_idx(oid
)] = 0;
1154 offset
= offsetof(SheepdogInode
,
1155 data_vdi_id
[data_oid_to_idx(oid
)]);
1156 oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
1157 wlen
= datalen
= sizeof(uint32_t);
1161 if (s
->cache_flags
) {
1162 hdr
.flags
|= s
->cache_flags
;
1166 hdr
.cow_oid
= old_oid
;
1167 hdr
.copies
= s
->inode
.nr_copies
;
1169 hdr
.data_length
= datalen
;
1170 hdr
.offset
= offset
;
1172 hdr
.id
= aio_req
->id
;
1174 qemu_co_mutex_lock(&s
->lock
);
1175 s
->co_send
= qemu_coroutine_self();
1176 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1177 co_read_response
, co_write_request
, NULL
, s
);
1178 socket_set_cork(s
->fd
, 1);
1181 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
1182 if (ret
!= sizeof(hdr
)) {
1183 error_report("failed to send a req, %s", strerror(errno
));
1188 ret
= qemu_co_sendv(s
->fd
, iov
, niov
, aio_req
->iov_offset
, wlen
);
1190 error_report("failed to send a data, %s", strerror(errno
));
1194 socket_set_cork(s
->fd
, 0);
1195 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1196 co_read_response
, NULL
, NULL
, s
);
1198 qemu_co_mutex_unlock(&s
->lock
);
1201 static int read_write_object(int fd
, BlockDriverState
*bs
, char *buf
,
1202 uint64_t oid
, uint8_t copies
,
1203 unsigned int datalen
, uint64_t offset
,
1204 bool write
, bool create
, uint32_t cache_flags
)
1207 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1208 unsigned int wlen
, rlen
;
1211 memset(&hdr
, 0, sizeof(hdr
));
1216 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1218 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1220 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1225 hdr
.opcode
= SD_OP_READ_OBJ
;
1228 hdr
.flags
|= cache_flags
;
1231 hdr
.data_length
= datalen
;
1232 hdr
.offset
= offset
;
1233 hdr
.copies
= copies
;
1235 ret
= do_req(fd
, bs
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1237 error_report("failed to send a request to the sheep");
1241 switch (rsp
->result
) {
1242 case SD_RES_SUCCESS
:
1245 error_report("%s", sd_strerror(rsp
->result
));
1250 static int read_object(int fd
, BlockDriverState
*bs
, char *buf
,
1251 uint64_t oid
, uint8_t copies
,
1252 unsigned int datalen
, uint64_t offset
,
1253 uint32_t cache_flags
)
1255 return read_write_object(fd
, bs
, buf
, oid
, copies
,
1256 datalen
, offset
, false,
1257 false, cache_flags
);
1260 static int write_object(int fd
, BlockDriverState
*bs
, char *buf
,
1261 uint64_t oid
, uint8_t copies
,
1262 unsigned int datalen
, uint64_t offset
, bool create
,
1263 uint32_t cache_flags
)
1265 return read_write_object(fd
, bs
, buf
, oid
, copies
,
1266 datalen
, offset
, true,
1267 create
, cache_flags
);
1270 /* update inode with the latest state */
1271 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
)
1273 Error
*local_err
= NULL
;
1274 SheepdogInode
*inode
;
1278 fd
= connect_to_sdog(s
, &local_err
);
1280 error_report_err(local_err
);
1284 inode
= g_malloc(SD_INODE_HEADER_SIZE
);
1286 ret
= find_vdi_name(s
, s
->name
, snapid
, tag
, &vid
, false, &local_err
);
1288 error_report_err(local_err
);
1292 ret
= read_object(fd
, s
->bs
, (char *)inode
, vid_to_vdi_oid(vid
),
1293 s
->inode
.nr_copies
, SD_INODE_HEADER_SIZE
, 0,
1299 if (inode
->vdi_id
!= s
->inode
.vdi_id
) {
1300 memcpy(&s
->inode
, inode
, SD_INODE_HEADER_SIZE
);
1310 static void coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
1312 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
1314 aio_req
->create
= false;
1316 /* check whether this request becomes a CoW one */
1317 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& is_data_obj(aio_req
->oid
)) {
1318 int idx
= data_oid_to_idx(aio_req
->oid
);
1320 if (is_data_obj_writable(&s
->inode
, idx
)) {
1324 if (s
->inode
.data_vdi_id
[idx
]) {
1325 aio_req
->base_oid
= vid_to_data_oid(s
->inode
.data_vdi_id
[idx
], idx
);
1326 aio_req
->flags
|= SD_FLAG_CMD_COW
;
1328 aio_req
->create
= true;
1331 if (is_data_obj(aio_req
->oid
)) {
1332 add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1336 iov
.iov_base
= &s
->inode
;
1337 iov
.iov_len
= sizeof(s
->inode
);
1338 add_aio_request(s
, aio_req
, &iov
, 1, AIOCB_WRITE_UDATA
);
1342 static void sd_detach_aio_context(BlockDriverState
*bs
)
1344 BDRVSheepdogState
*s
= bs
->opaque
;
1346 aio_set_fd_handler(s
->aio_context
, s
->fd
, false, NULL
,
1350 static void sd_attach_aio_context(BlockDriverState
*bs
,
1351 AioContext
*new_context
)
1353 BDRVSheepdogState
*s
= bs
->opaque
;
1355 s
->aio_context
= new_context
;
1356 aio_set_fd_handler(new_context
, s
->fd
, false,
1357 co_read_response
, NULL
, NULL
, s
);
1360 /* TODO Convert to fine grained options */
1361 static QemuOptsList runtime_opts
= {
1363 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1367 .type
= QEMU_OPT_STRING
,
1368 .help
= "URL to the sheepdog image",
1370 { /* end of list */ }
1374 static int sd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1379 BDRVSheepdogState
*s
= bs
->opaque
;
1380 char vdi
[SD_MAX_VDI_LEN
], tag
[SD_MAX_VDI_TAG_LEN
];
1384 Error
*local_err
= NULL
;
1385 const char *filename
;
1388 s
->aio_context
= bdrv_get_aio_context(bs
);
1390 opts
= qemu_opts_create(&runtime_opts
, NULL
, 0, &error_abort
);
1391 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1393 error_propagate(errp
, local_err
);
1398 filename
= qemu_opt_get(opts
, "filename");
1400 QLIST_INIT(&s
->inflight_aio_head
);
1401 QLIST_INIT(&s
->failed_aio_head
);
1402 QLIST_INIT(&s
->inflight_aiocb_head
);
1405 memset(vdi
, 0, sizeof(vdi
));
1406 memset(tag
, 0, sizeof(tag
));
1408 if (strstr(filename
, "://")) {
1409 ret
= sd_parse_uri(s
, filename
, vdi
, &snapid
, tag
);
1411 ret
= parse_vdiname(s
, filename
, vdi
, &snapid
, tag
);
1414 error_setg(errp
, "Can't parse filename");
1417 s
->fd
= get_sheep_fd(s
, errp
);
1423 ret
= find_vdi_name(s
, vdi
, snapid
, tag
, &vid
, true, errp
);
1429 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1430 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1432 s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1433 if (flags
& BDRV_O_NOCACHE
) {
1434 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1436 s
->discard_supported
= true;
1438 if (snapid
|| tag
[0] != '\0') {
1439 DPRINTF("%" PRIx32
" snapshot inode was open.\n", vid
);
1440 s
->is_snapshot
= true;
1443 fd
= connect_to_sdog(s
, errp
);
1449 buf
= g_malloc(SD_INODE_SIZE
);
1450 ret
= read_object(fd
, s
->bs
, buf
, vid_to_vdi_oid(vid
),
1451 0, SD_INODE_SIZE
, 0, s
->cache_flags
);
1456 error_setg(errp
, "Can't read snapshot inode");
1460 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1462 bs
->total_sectors
= s
->inode
.vdi_size
/ BDRV_SECTOR_SIZE
;
1463 pstrcpy(s
->name
, sizeof(s
->name
), vdi
);
1464 qemu_co_mutex_init(&s
->lock
);
1465 qemu_co_queue_init(&s
->overlapping_queue
);
1466 qemu_opts_del(opts
);
1470 aio_set_fd_handler(bdrv_get_aio_context(bs
), s
->fd
,
1471 false, NULL
, NULL
, NULL
, NULL
);
1475 qemu_opts_del(opts
);
1480 static int sd_reopen_prepare(BDRVReopenState
*state
, BlockReopenQueue
*queue
,
1483 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1484 BDRVSheepdogReopenState
*re_s
;
1487 re_s
= state
->opaque
= g_new0(BDRVSheepdogReopenState
, 1);
1489 re_s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1490 if (state
->flags
& BDRV_O_NOCACHE
) {
1491 re_s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1494 re_s
->fd
= get_sheep_fd(s
, errp
);
1503 static void sd_reopen_commit(BDRVReopenState
*state
)
1505 BDRVSheepdogReopenState
*re_s
= state
->opaque
;
1506 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1509 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1510 NULL
, NULL
, NULL
, NULL
);
1515 s
->cache_flags
= re_s
->cache_flags
;
1517 g_free(state
->opaque
);
1518 state
->opaque
= NULL
;
1523 static void sd_reopen_abort(BDRVReopenState
*state
)
1525 BDRVSheepdogReopenState
*re_s
= state
->opaque
;
1526 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1533 aio_set_fd_handler(s
->aio_context
, re_s
->fd
, false,
1534 NULL
, NULL
, NULL
, NULL
);
1535 closesocket(re_s
->fd
);
1538 g_free(state
->opaque
);
1539 state
->opaque
= NULL
;
1544 static int do_sd_create(BDRVSheepdogState
*s
, uint32_t *vdi_id
, int snapshot
,
1548 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1550 unsigned int wlen
, rlen
= 0;
1551 char buf
[SD_MAX_VDI_LEN
];
1553 fd
= connect_to_sdog(s
, errp
);
1558 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1559 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1561 memset(buf
, 0, sizeof(buf
));
1562 pstrcpy(buf
, sizeof(buf
), s
->name
);
1564 memset(&hdr
, 0, sizeof(hdr
));
1565 hdr
.opcode
= SD_OP_NEW_VDI
;
1566 hdr
.base_vdi_id
= s
->inode
.vdi_id
;
1568 wlen
= SD_MAX_VDI_LEN
;
1570 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1571 hdr
.snapid
= snapshot
;
1573 hdr
.data_length
= wlen
;
1574 hdr
.vdi_size
= s
->inode
.vdi_size
;
1575 hdr
.copy_policy
= s
->inode
.copy_policy
;
1576 hdr
.copies
= s
->inode
.nr_copies
;
1577 hdr
.block_size_shift
= s
->inode
.block_size_shift
;
1579 ret
= do_req(fd
, NULL
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1584 error_setg_errno(errp
, -ret
, "create failed");
1588 if (rsp
->result
!= SD_RES_SUCCESS
) {
1589 error_setg(errp
, "%s, %s", sd_strerror(rsp
->result
), s
->inode
.name
);
1594 *vdi_id
= rsp
->vdi_id
;
1600 static int sd_prealloc(const char *filename
, Error
**errp
)
1602 BlockBackend
*blk
= NULL
;
1603 BDRVSheepdogState
*base
= NULL
;
1604 unsigned long buf_size
;
1605 uint32_t idx
, max_idx
;
1606 uint32_t object_size
;
1611 blk
= blk_new_open(filename
, NULL
, NULL
,
1612 BDRV_O_RDWR
| BDRV_O_PROTOCOL
, errp
);
1615 goto out_with_err_set
;
1618 blk_set_allow_write_beyond_eof(blk
, true);
1620 vdi_size
= blk_getlength(blk
);
1626 base
= blk_bs(blk
)->opaque
;
1627 object_size
= (UINT32_C(1) << base
->inode
.block_size_shift
);
1628 buf_size
= MIN(object_size
, SD_DATA_OBJ_SIZE
);
1629 buf
= g_malloc0(buf_size
);
1631 max_idx
= DIV_ROUND_UP(vdi_size
, buf_size
);
1633 for (idx
= 0; idx
< max_idx
; idx
++) {
1635 * The created image can be a cloned image, so we need to read
1636 * a data from the source image.
1638 ret
= blk_pread(blk
, idx
* buf_size
, buf
, buf_size
);
1642 ret
= blk_pwrite(blk
, idx
* buf_size
, buf
, buf_size
, 0);
1651 error_setg_errno(errp
, -ret
, "Can't pre-allocate");
1663 * Sheepdog support two kinds of redundancy, full replication and erasure
1666 * # create a fully replicated vdi with x copies
1667 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1669 * # create a erasure coded vdi with x data strips and y parity strips
1670 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1672 static int parse_redundancy(BDRVSheepdogState
*s
, const char *opt
)
1674 struct SheepdogInode
*inode
= &s
->inode
;
1675 const char *n1
, *n2
;
1679 pstrcpy(p
, sizeof(p
), opt
);
1680 n1
= strtok(p
, ":");
1681 n2
= strtok(NULL
, ":");
1687 copy
= strtol(n1
, NULL
, 10);
1688 if (copy
> SD_MAX_COPIES
|| copy
< 1) {
1692 inode
->copy_policy
= 0;
1693 inode
->nr_copies
= copy
;
1697 if (copy
!= 2 && copy
!= 4 && copy
!= 8 && copy
!= 16) {
1701 parity
= strtol(n2
, NULL
, 10);
1702 if (parity
>= SD_EC_MAX_STRIP
|| parity
< 1) {
1707 * 4 bits for parity and 4 bits for data.
1708 * We have to compress upper data bits because it can't represent 16
1710 inode
->copy_policy
= ((copy
/ 2) << 4) + parity
;
1711 inode
->nr_copies
= copy
+ parity
;
1716 static int parse_block_size_shift(BDRVSheepdogState
*s
, QemuOpts
*opt
)
1718 struct SheepdogInode
*inode
= &s
->inode
;
1719 uint64_t object_size
;
1722 object_size
= qemu_opt_get_size_del(opt
, BLOCK_OPT_OBJECT_SIZE
, 0);
1724 if ((object_size
- 1) & object_size
) { /* not a power of 2? */
1727 obj_order
= ctz32(object_size
);
1728 if (obj_order
< 20 || obj_order
> 31) {
1731 inode
->block_size_shift
= (uint8_t)obj_order
;
1737 static int sd_create(const char *filename
, QemuOpts
*opts
,
1742 char *backing_file
= NULL
;
1744 BDRVSheepdogState
*s
;
1745 char tag
[SD_MAX_VDI_TAG_LEN
];
1747 uint64_t max_vdi_size
;
1748 bool prealloc
= false;
1750 s
= g_new0(BDRVSheepdogState
, 1);
1752 memset(tag
, 0, sizeof(tag
));
1753 if (strstr(filename
, "://")) {
1754 ret
= sd_parse_uri(s
, filename
, s
->name
, &snapid
, tag
);
1756 ret
= parse_vdiname(s
, filename
, s
->name
, &snapid
, tag
);
1759 error_setg(errp
, "Can't parse filename");
1763 s
->inode
.vdi_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1765 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
1766 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
1767 if (!buf
|| !strcmp(buf
, "off")) {
1769 } else if (!strcmp(buf
, "full")) {
1772 error_setg(errp
, "Invalid preallocation mode: '%s'", buf
);
1778 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_REDUNDANCY
);
1780 ret
= parse_redundancy(s
, buf
);
1782 error_setg(errp
, "Invalid redundancy mode: '%s'", buf
);
1786 ret
= parse_block_size_shift(s
, opts
);
1788 error_setg(errp
, "Invalid object_size."
1789 " obect_size needs to be power of 2"
1790 " and be limited from 2^20 to 2^31");
1796 BDRVSheepdogState
*base
;
1799 /* Currently, only Sheepdog backing image is supported. */
1800 drv
= bdrv_find_protocol(backing_file
, true, NULL
);
1801 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
1802 error_setg(errp
, "backing_file must be a sheepdog image");
1807 blk
= blk_new_open(backing_file
, NULL
, NULL
,
1808 BDRV_O_PROTOCOL
, errp
);
1814 base
= blk_bs(blk
)->opaque
;
1816 if (!is_snapshot(&base
->inode
)) {
1817 error_setg(errp
, "cannot clone from a non snapshot vdi");
1822 s
->inode
.vdi_id
= base
->inode
.vdi_id
;
1826 s
->aio_context
= qemu_get_aio_context();
1828 /* if block_size_shift is not specified, get cluster default value */
1829 if (s
->inode
.block_size_shift
== 0) {
1831 SheepdogClusterRsp
*rsp
= (SheepdogClusterRsp
*)&hdr
;
1832 Error
*local_err
= NULL
;
1834 unsigned int wlen
= 0, rlen
= 0;
1836 fd
= connect_to_sdog(s
, &local_err
);
1838 error_report_err(local_err
);
1843 memset(&hdr
, 0, sizeof(hdr
));
1844 hdr
.opcode
= SD_OP_GET_CLUSTER_DEFAULT
;
1845 hdr
.proto_ver
= SD_PROTO_VER
;
1847 ret
= do_req(fd
, NULL
, (SheepdogReq
*)&hdr
,
1848 NULL
, &wlen
, &rlen
);
1851 error_setg_errno(errp
, -ret
, "failed to get cluster default");
1854 if (rsp
->result
== SD_RES_SUCCESS
) {
1855 s
->inode
.block_size_shift
= rsp
->block_size_shift
;
1857 s
->inode
.block_size_shift
= SD_DEFAULT_BLOCK_SIZE_SHIFT
;
1861 max_vdi_size
= (UINT64_C(1) << s
->inode
.block_size_shift
) * MAX_DATA_OBJS
;
1863 if (s
->inode
.vdi_size
> max_vdi_size
) {
1864 error_setg(errp
, "An image is too large."
1865 " The maximum image size is %"PRIu64
"GB",
1866 max_vdi_size
/ 1024 / 1024 / 1024);
1871 ret
= do_sd_create(s
, &vid
, 0, errp
);
1877 ret
= sd_prealloc(filename
, errp
);
1880 g_free(backing_file
);
1886 static void sd_close(BlockDriverState
*bs
)
1888 Error
*local_err
= NULL
;
1889 BDRVSheepdogState
*s
= bs
->opaque
;
1891 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1892 unsigned int wlen
, rlen
= 0;
1895 DPRINTF("%s\n", s
->name
);
1897 fd
= connect_to_sdog(s
, &local_err
);
1899 error_report_err(local_err
);
1903 memset(&hdr
, 0, sizeof(hdr
));
1905 hdr
.opcode
= SD_OP_RELEASE_VDI
;
1906 hdr
.type
= LOCK_TYPE_NORMAL
;
1907 hdr
.base_vdi_id
= s
->inode
.vdi_id
;
1908 wlen
= strlen(s
->name
) + 1;
1909 hdr
.data_length
= wlen
;
1910 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1912 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
1913 s
->name
, &wlen
, &rlen
);
1917 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
1918 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
1919 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
1922 aio_set_fd_handler(bdrv_get_aio_context(bs
), s
->fd
,
1923 false, NULL
, NULL
, NULL
, NULL
);
1925 g_free(s
->host_spec
);
1928 static int64_t sd_getlength(BlockDriverState
*bs
)
1930 BDRVSheepdogState
*s
= bs
->opaque
;
1932 return s
->inode
.vdi_size
;
1935 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
)
1937 Error
*local_err
= NULL
;
1938 BDRVSheepdogState
*s
= bs
->opaque
;
1940 unsigned int datalen
;
1941 uint64_t max_vdi_size
;
1943 max_vdi_size
= (UINT64_C(1) << s
->inode
.block_size_shift
) * MAX_DATA_OBJS
;
1944 if (offset
< s
->inode
.vdi_size
) {
1945 error_report("shrinking is not supported");
1947 } else if (offset
> max_vdi_size
) {
1948 error_report("too big image size");
1952 fd
= connect_to_sdog(s
, &local_err
);
1954 error_report_err(local_err
);
1958 /* we don't need to update entire object */
1959 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
1960 s
->inode
.vdi_size
= offset
;
1961 ret
= write_object(fd
, s
->bs
, (char *)&s
->inode
,
1962 vid_to_vdi_oid(s
->inode
.vdi_id
), s
->inode
.nr_copies
,
1963 datalen
, 0, false, s
->cache_flags
);
1967 error_report("failed to update an inode.");
1974 * This function is called after writing data objects. If we need to
1975 * update metadata, this sends a write request to the vdi object.
1977 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
1979 BDRVSheepdogState
*s
= acb
->s
;
1982 uint32_t offset
, data_len
, mn
, mx
;
1984 mn
= acb
->min_dirty_data_idx
;
1985 mx
= acb
->max_dirty_data_idx
;
1987 /* we need to update the vdi object. */
1989 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
1990 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
1991 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
1993 acb
->min_dirty_data_idx
= UINT32_MAX
;
1994 acb
->max_dirty_data_idx
= 0;
1996 iov
.iov_base
= &s
->inode
;
1997 iov
.iov_len
= sizeof(s
->inode
);
1998 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
1999 data_len
, offset
, 0, false, 0, offset
);
2000 add_aio_request(s
, aio_req
, &iov
, 1, AIOCB_WRITE_UDATA
);
2001 if (--acb
->nr_pending
) {
2002 qemu_coroutine_yield();
2007 /* Delete current working VDI on the snapshot chain */
2008 static bool sd_delete(BDRVSheepdogState
*s
)
2010 Error
*local_err
= NULL
;
2011 unsigned int wlen
= SD_MAX_VDI_LEN
, rlen
= 0;
2012 SheepdogVdiReq hdr
= {
2013 .opcode
= SD_OP_DEL_VDI
,
2014 .base_vdi_id
= s
->inode
.vdi_id
,
2015 .data_length
= wlen
,
2016 .flags
= SD_FLAG_CMD_WRITE
,
2018 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2021 fd
= connect_to_sdog(s
, &local_err
);
2023 error_report_err(local_err
);
2027 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2028 s
->name
, &wlen
, &rlen
);
2033 switch (rsp
->result
) {
2035 error_report("%s was already deleted", s
->name
);
2037 case SD_RES_SUCCESS
:
2040 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
2048 * Create a writable VDI from a snapshot
2050 static int sd_create_branch(BDRVSheepdogState
*s
)
2052 Error
*local_err
= NULL
;
2058 DPRINTF("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
2060 buf
= g_malloc(SD_INODE_SIZE
);
2063 * Even If deletion fails, we will just create extra snapshot based on
2064 * the working VDI which was supposed to be deleted. So no need to
2067 deleted
= sd_delete(s
);
2068 ret
= do_sd_create(s
, &vid
, !deleted
, &local_err
);
2070 error_report_err(local_err
);
2074 DPRINTF("%" PRIx32
" is created.\n", vid
);
2076 fd
= connect_to_sdog(s
, &local_err
);
2078 error_report_err(local_err
);
2083 ret
= read_object(fd
, s
->bs
, buf
, vid_to_vdi_oid(vid
),
2084 s
->inode
.nr_copies
, SD_INODE_SIZE
, 0, s
->cache_flags
);
2092 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
2094 s
->is_snapshot
= false;
2096 DPRINTF("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
2105 * Send I/O requests to the server.
2107 * This function sends requests to the server, links the requests to
2108 * the inflight_list in BDRVSheepdogState, and exits without
2109 * waiting the response. The responses are received in the
2110 * `aio_read_response' function which is called from the main loop as
2113 * Returns 1 when we need to wait a response, 0 when there is no sent
2114 * request and -errno in error cases.
2116 static void coroutine_fn
sd_co_rw_vector(SheepdogAIOCB
*acb
)
2119 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* BDRV_SECTOR_SIZE
;
2121 uint32_t object_size
;
2124 BDRVSheepdogState
*s
= acb
->s
;
2125 SheepdogInode
*inode
= &s
->inode
;
2128 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
2130 * In the case we open the snapshot VDI, Sheepdog creates the
2131 * writable VDI when we do a write operation first.
2133 ret
= sd_create_branch(s
);
2140 object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2141 idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ object_size
;
2142 offset
= (acb
->sector_num
* BDRV_SECTOR_SIZE
) % object_size
;
2145 * Make sure we don't free the aiocb before we are done with all requests.
2146 * This additional reference is dropped at the end of this function.
2150 while (done
!= total
) {
2152 uint64_t old_oid
= 0;
2153 bool create
= false;
2155 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
2157 len
= MIN(total
- done
, object_size
- offset
);
2159 switch (acb
->aiocb_type
) {
2160 case AIOCB_READ_UDATA
:
2161 if (!inode
->data_vdi_id
[idx
]) {
2162 qemu_iovec_memset(acb
->qiov
, done
, 0, len
);
2166 case AIOCB_WRITE_UDATA
:
2167 if (!inode
->data_vdi_id
[idx
]) {
2169 } else if (!is_data_obj_writable(inode
, idx
)) {
2173 flags
= SD_FLAG_CMD_COW
;
2176 case AIOCB_DISCARD_OBJ
:
2178 * We discard the object only when the whole object is
2179 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2181 if (len
!= object_size
|| inode
->data_vdi_id
[idx
] == 0) {
2190 DPRINTF("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
" %ld\n",
2192 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
2193 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
2194 DPRINTF("new oid %" PRIx64
"\n", oid
);
2197 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, create
,
2199 acb
->aiocb_type
== AIOCB_DISCARD_OBJ
?
2201 add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
2208 if (--acb
->nr_pending
) {
2209 qemu_coroutine_yield();
2213 static void sd_aio_complete(SheepdogAIOCB
*acb
)
2215 if (acb
->aiocb_type
== AIOCB_FLUSH_CACHE
) {
2219 QLIST_REMOVE(acb
, aiocb_siblings
);
2220 qemu_co_queue_restart_all(&acb
->s
->overlapping_queue
);
2223 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2224 int nb_sectors
, QEMUIOVector
*qiov
)
2228 int64_t offset
= (sector_num
+ nb_sectors
) * BDRV_SECTOR_SIZE
;
2229 BDRVSheepdogState
*s
= bs
->opaque
;
2231 if (offset
> s
->inode
.vdi_size
) {
2232 ret
= sd_truncate(bs
, offset
);
2238 sd_aio_setup(&acb
, s
, qiov
, sector_num
, nb_sectors
, AIOCB_WRITE_UDATA
);
2239 sd_co_rw_vector(&acb
);
2240 sd_write_done(&acb
);
2241 sd_aio_complete(&acb
);
2246 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2247 int nb_sectors
, QEMUIOVector
*qiov
)
2250 BDRVSheepdogState
*s
= bs
->opaque
;
2252 sd_aio_setup(&acb
, s
, qiov
, sector_num
, nb_sectors
, AIOCB_READ_UDATA
);
2253 sd_co_rw_vector(&acb
);
2254 sd_aio_complete(&acb
);
2259 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
2261 BDRVSheepdogState
*s
= bs
->opaque
;
2265 if (s
->cache_flags
!= SD_FLAG_CMD_CACHE
) {
2269 sd_aio_setup(&acb
, s
, NULL
, 0, 0, AIOCB_FLUSH_CACHE
);
2272 aio_req
= alloc_aio_req(s
, &acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2273 0, 0, 0, false, 0, 0);
2274 add_aio_request(s
, aio_req
, NULL
, 0, acb
.aiocb_type
);
2276 if (--acb
.nr_pending
) {
2277 qemu_coroutine_yield();
2280 sd_aio_complete(&acb
);
2284 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
2286 Error
*local_err
= NULL
;
2287 BDRVSheepdogState
*s
= bs
->opaque
;
2290 SheepdogInode
*inode
;
2291 unsigned int datalen
;
2293 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64
" "
2294 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
2295 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
2297 if (s
->is_snapshot
) {
2298 error_report("You can't create a snapshot of a snapshot VDI, "
2299 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
2304 DPRINTF("%s %s\n", sn_info
->name
, sn_info
->id_str
);
2306 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
2307 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
2308 /* It appears that inode.tag does not require a NUL terminator,
2309 * which means this use of strncpy is ok.
2311 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
2312 /* we don't need to update entire object */
2313 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2314 inode
= g_malloc(datalen
);
2316 /* refresh inode. */
2317 fd
= connect_to_sdog(s
, &local_err
);
2319 error_report_err(local_err
);
2324 ret
= write_object(fd
, s
->bs
, (char *)&s
->inode
,
2325 vid_to_vdi_oid(s
->inode
.vdi_id
), s
->inode
.nr_copies
,
2326 datalen
, 0, false, s
->cache_flags
);
2328 error_report("failed to write snapshot's inode.");
2332 ret
= do_sd_create(s
, &new_vid
, 1, &local_err
);
2334 error_reportf_err(local_err
,
2335 "failed to create inode for snapshot: ");
2339 ret
= read_object(fd
, s
->bs
, (char *)inode
,
2340 vid_to_vdi_oid(new_vid
), s
->inode
.nr_copies
, datalen
, 0,
2344 error_report("failed to read new inode info. %s", strerror(errno
));
2348 memcpy(&s
->inode
, inode
, datalen
);
2349 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2350 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
2359 * We implement rollback(loadvm) operation to the specified snapshot by
2360 * 1) switch to the snapshot
2361 * 2) rely on sd_create_branch to delete working VDI and
2362 * 3) create a new working VDI based on the specified snapshot
2364 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
2366 BDRVSheepdogState
*s
= bs
->opaque
;
2367 BDRVSheepdogState
*old_s
;
2368 char tag
[SD_MAX_VDI_TAG_LEN
];
2369 uint32_t snapid
= 0;
2372 old_s
= g_new(BDRVSheepdogState
, 1);
2374 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
2376 snapid
= strtoul(snapshot_id
, NULL
, 10);
2380 pstrcpy(tag
, sizeof(tag
), snapshot_id
);
2383 ret
= reload_inode(s
, snapid
, tag
);
2388 ret
= sd_create_branch(s
);
2397 /* recover bdrv_sd_state */
2398 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
2401 error_report("failed to open. recover old bdrv_sd_state.");
2406 #define NR_BATCHED_DISCARD 128
2408 static bool remove_objects(BDRVSheepdogState
*s
)
2410 int fd
, i
= 0, nr_objs
= 0;
2411 Error
*local_err
= NULL
;
2414 SheepdogInode
*inode
= &s
->inode
;
2416 fd
= connect_to_sdog(s
, &local_err
);
2418 error_report_err(local_err
);
2422 nr_objs
= count_data_objs(inode
);
2423 while (i
< nr_objs
) {
2424 int start_idx
, nr_filled_idx
;
2426 while (i
< nr_objs
&& !inode
->data_vdi_id
[i
]) {
2432 while (i
< nr_objs
&& nr_filled_idx
< NR_BATCHED_DISCARD
) {
2433 if (inode
->data_vdi_id
[i
]) {
2434 inode
->data_vdi_id
[i
] = 0;
2441 ret
= write_object(fd
, s
->bs
,
2442 (char *)&inode
->data_vdi_id
[start_idx
],
2443 vid_to_vdi_oid(s
->inode
.vdi_id
), inode
->nr_copies
,
2444 (i
- start_idx
) * sizeof(uint32_t),
2445 offsetof(struct SheepdogInode
,
2446 data_vdi_id
[start_idx
]),
2447 false, s
->cache_flags
);
2449 error_report("failed to discard snapshot inode.");
2460 static int sd_snapshot_delete(BlockDriverState
*bs
,
2461 const char *snapshot_id
,
2465 unsigned long snap_id
= 0;
2466 char snap_tag
[SD_MAX_VDI_TAG_LEN
];
2467 Error
*local_err
= NULL
;
2469 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
2470 BDRVSheepdogState
*s
= bs
->opaque
;
2471 unsigned int wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
, rlen
= 0;
2473 SheepdogVdiReq hdr
= {
2474 .opcode
= SD_OP_DEL_VDI
,
2475 .data_length
= wlen
,
2476 .flags
= SD_FLAG_CMD_WRITE
,
2478 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2480 if (!remove_objects(s
)) {
2484 memset(buf
, 0, sizeof(buf
));
2485 memset(snap_tag
, 0, sizeof(snap_tag
));
2486 pstrcpy(buf
, SD_MAX_VDI_LEN
, s
->name
);
2487 ret
= qemu_strtoul(snapshot_id
, NULL
, 10, &snap_id
);
2488 if (ret
|| snap_id
> UINT32_MAX
) {
2489 error_setg(errp
, "Invalid snapshot ID: %s",
2490 snapshot_id
? snapshot_id
: "<null>");
2495 hdr
.snapid
= (uint32_t) snap_id
;
2497 pstrcpy(snap_tag
, sizeof(snap_tag
), snapshot_id
);
2498 pstrcpy(buf
+ SD_MAX_VDI_LEN
, SD_MAX_VDI_TAG_LEN
, snap_tag
);
2501 ret
= find_vdi_name(s
, s
->name
, snap_id
, snap_tag
, &vid
, true,
2507 fd
= connect_to_sdog(s
, &local_err
);
2509 error_report_err(local_err
);
2513 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2520 switch (rsp
->result
) {
2522 error_report("%s was already deleted", s
->name
);
2523 case SD_RES_SUCCESS
:
2526 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
2533 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
2535 Error
*local_err
= NULL
;
2536 BDRVSheepdogState
*s
= bs
->opaque
;
2538 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
2539 QEMUSnapshotInfo
*sn_tab
= NULL
;
2540 unsigned wlen
, rlen
;
2542 static SheepdogInode inode
;
2543 unsigned long *vdi_inuse
;
2544 unsigned int start_nr
;
2548 vdi_inuse
= g_malloc(max
);
2550 fd
= connect_to_sdog(s
, &local_err
);
2552 error_report_err(local_err
);
2560 memset(&req
, 0, sizeof(req
));
2562 req
.opcode
= SD_OP_READ_VDIS
;
2563 req
.data_length
= max
;
2565 ret
= do_req(fd
, s
->bs
, &req
, vdi_inuse
, &wlen
, &rlen
);
2572 sn_tab
= g_new0(QEMUSnapshotInfo
, nr
);
2574 /* calculate a vdi id with hash function */
2575 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
2576 start_nr
= hval
& (SD_NR_VDIS
- 1);
2578 fd
= connect_to_sdog(s
, &local_err
);
2580 error_report_err(local_err
);
2585 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
2586 if (!test_bit(vid
, vdi_inuse
)) {
2590 /* we don't need to read entire object */
2591 ret
= read_object(fd
, s
->bs
, (char *)&inode
,
2592 vid_to_vdi_oid(vid
),
2593 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
2600 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
2601 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
2602 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
2603 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
2604 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
2606 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
),
2607 "%" PRIu32
, inode
.snap_id
);
2608 pstrcpy(sn_tab
[found
].name
,
2609 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)),
2628 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
2629 int64_t pos
, int size
, int load
)
2631 Error
*local_err
= NULL
;
2633 int fd
, ret
= 0, remaining
= size
;
2634 unsigned int data_len
;
2635 uint64_t vmstate_oid
;
2638 uint32_t vdi_id
= load
? s
->inode
.parent_vdi_id
: s
->inode
.vdi_id
;
2639 uint32_t object_size
= (UINT32_C(1) << s
->inode
.block_size_shift
);
2641 fd
= connect_to_sdog(s
, &local_err
);
2643 error_report_err(local_err
);
2648 vdi_index
= pos
/ object_size
;
2649 offset
= pos
% object_size
;
2651 data_len
= MIN(remaining
, object_size
- offset
);
2653 vmstate_oid
= vid_to_vmstate_oid(vdi_id
, vdi_index
);
2655 create
= (offset
== 0);
2657 ret
= read_object(fd
, s
->bs
, (char *)data
, vmstate_oid
,
2658 s
->inode
.nr_copies
, data_len
, offset
,
2661 ret
= write_object(fd
, s
->bs
, (char *)data
, vmstate_oid
,
2662 s
->inode
.nr_copies
, data_len
, offset
, create
,
2667 error_report("failed to save vmstate %s", strerror(errno
));
2673 remaining
-= data_len
;
2681 static int sd_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2684 BDRVSheepdogState
*s
= bs
->opaque
;
2688 buf
= qemu_blockalign(bs
, qiov
->size
);
2689 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
2690 ret
= do_load_save_vmstate(s
, (uint8_t *) buf
, pos
, qiov
->size
, 0);
2696 static int sd_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2699 BDRVSheepdogState
*s
= bs
->opaque
;
2703 buf
= qemu_blockalign(bs
, qiov
->size
);
2704 ret
= do_load_save_vmstate(s
, buf
, pos
, qiov
->size
, 1);
2705 qemu_iovec_from_buf(qiov
, 0, buf
, qiov
->size
);
2712 static coroutine_fn
int sd_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
2716 BDRVSheepdogState
*s
= bs
->opaque
;
2717 QEMUIOVector discard_iov
;
2721 if (!s
->discard_supported
) {
2725 memset(&discard_iov
, 0, sizeof(discard_iov
));
2726 memset(&iov
, 0, sizeof(iov
));
2727 iov
.iov_base
= &zero
;
2728 iov
.iov_len
= sizeof(zero
);
2729 discard_iov
.iov
= &iov
;
2730 discard_iov
.niov
= 1;
2731 if (!QEMU_IS_ALIGNED(offset
| count
, BDRV_SECTOR_SIZE
)) {
2734 sd_aio_setup(&acb
, s
, &discard_iov
, offset
>> BDRV_SECTOR_BITS
,
2735 count
>> BDRV_SECTOR_BITS
, AIOCB_DISCARD_OBJ
);
2736 sd_co_rw_vector(&acb
);
2737 sd_aio_complete(&acb
);
2742 static coroutine_fn
int64_t
2743 sd_co_get_block_status(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2744 int *pnum
, BlockDriverState
**file
)
2746 BDRVSheepdogState
*s
= bs
->opaque
;
2747 SheepdogInode
*inode
= &s
->inode
;
2748 uint32_t object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2749 uint64_t offset
= sector_num
* BDRV_SECTOR_SIZE
;
2750 unsigned long start
= offset
/ object_size
,
2751 end
= DIV_ROUND_UP((sector_num
+ nb_sectors
) *
2752 BDRV_SECTOR_SIZE
, object_size
);
2754 int64_t ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| offset
;
2756 for (idx
= start
; idx
< end
; idx
++) {
2757 if (inode
->data_vdi_id
[idx
] == 0) {
2762 /* Get the longest length of unallocated sectors */
2764 for (idx
= start
+ 1; idx
< end
; idx
++) {
2765 if (inode
->data_vdi_id
[idx
] != 0) {
2771 *pnum
= (idx
- start
) * object_size
/ BDRV_SECTOR_SIZE
;
2772 if (*pnum
> nb_sectors
) {
2775 if (ret
> 0 && ret
& BDRV_BLOCK_OFFSET_VALID
) {
2781 static int64_t sd_get_allocated_file_size(BlockDriverState
*bs
)
2783 BDRVSheepdogState
*s
= bs
->opaque
;
2784 SheepdogInode
*inode
= &s
->inode
;
2785 uint32_t object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2786 unsigned long i
, last
= DIV_ROUND_UP(inode
->vdi_size
, object_size
);
2789 for (i
= 0; i
< last
; i
++) {
2790 if (inode
->data_vdi_id
[i
] == 0) {
2793 size
+= object_size
;
2798 static QemuOptsList sd_create_opts
= {
2799 .name
= "sheepdog-create-opts",
2800 .head
= QTAILQ_HEAD_INITIALIZER(sd_create_opts
.head
),
2803 .name
= BLOCK_OPT_SIZE
,
2804 .type
= QEMU_OPT_SIZE
,
2805 .help
= "Virtual disk size"
2808 .name
= BLOCK_OPT_BACKING_FILE
,
2809 .type
= QEMU_OPT_STRING
,
2810 .help
= "File name of a base image"
2813 .name
= BLOCK_OPT_PREALLOC
,
2814 .type
= QEMU_OPT_STRING
,
2815 .help
= "Preallocation mode (allowed values: off, full)"
2818 .name
= BLOCK_OPT_REDUNDANCY
,
2819 .type
= QEMU_OPT_STRING
,
2820 .help
= "Redundancy of the image"
2823 .name
= BLOCK_OPT_OBJECT_SIZE
,
2824 .type
= QEMU_OPT_SIZE
,
2825 .help
= "Object size of the image"
2827 { /* end of list */ }
2831 static BlockDriver bdrv_sheepdog
= {
2832 .format_name
= "sheepdog",
2833 .protocol_name
= "sheepdog",
2834 .instance_size
= sizeof(BDRVSheepdogState
),
2835 .bdrv_needs_filename
= true,
2836 .bdrv_file_open
= sd_open
,
2837 .bdrv_reopen_prepare
= sd_reopen_prepare
,
2838 .bdrv_reopen_commit
= sd_reopen_commit
,
2839 .bdrv_reopen_abort
= sd_reopen_abort
,
2840 .bdrv_close
= sd_close
,
2841 .bdrv_create
= sd_create
,
2842 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2843 .bdrv_getlength
= sd_getlength
,
2844 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
2845 .bdrv_truncate
= sd_truncate
,
2847 .bdrv_co_readv
= sd_co_readv
,
2848 .bdrv_co_writev
= sd_co_writev
,
2849 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2850 .bdrv_co_pdiscard
= sd_co_pdiscard
,
2851 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2853 .bdrv_snapshot_create
= sd_snapshot_create
,
2854 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2855 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2856 .bdrv_snapshot_list
= sd_snapshot_list
,
2858 .bdrv_save_vmstate
= sd_save_vmstate
,
2859 .bdrv_load_vmstate
= sd_load_vmstate
,
2861 .bdrv_detach_aio_context
= sd_detach_aio_context
,
2862 .bdrv_attach_aio_context
= sd_attach_aio_context
,
2864 .create_opts
= &sd_create_opts
,
2867 static BlockDriver bdrv_sheepdog_tcp
= {
2868 .format_name
= "sheepdog",
2869 .protocol_name
= "sheepdog+tcp",
2870 .instance_size
= sizeof(BDRVSheepdogState
),
2871 .bdrv_needs_filename
= true,
2872 .bdrv_file_open
= sd_open
,
2873 .bdrv_reopen_prepare
= sd_reopen_prepare
,
2874 .bdrv_reopen_commit
= sd_reopen_commit
,
2875 .bdrv_reopen_abort
= sd_reopen_abort
,
2876 .bdrv_close
= sd_close
,
2877 .bdrv_create
= sd_create
,
2878 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2879 .bdrv_getlength
= sd_getlength
,
2880 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
2881 .bdrv_truncate
= sd_truncate
,
2883 .bdrv_co_readv
= sd_co_readv
,
2884 .bdrv_co_writev
= sd_co_writev
,
2885 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2886 .bdrv_co_pdiscard
= sd_co_pdiscard
,
2887 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2889 .bdrv_snapshot_create
= sd_snapshot_create
,
2890 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2891 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2892 .bdrv_snapshot_list
= sd_snapshot_list
,
2894 .bdrv_save_vmstate
= sd_save_vmstate
,
2895 .bdrv_load_vmstate
= sd_load_vmstate
,
2897 .bdrv_detach_aio_context
= sd_detach_aio_context
,
2898 .bdrv_attach_aio_context
= sd_attach_aio_context
,
2900 .create_opts
= &sd_create_opts
,
2903 static BlockDriver bdrv_sheepdog_unix
= {
2904 .format_name
= "sheepdog",
2905 .protocol_name
= "sheepdog+unix",
2906 .instance_size
= sizeof(BDRVSheepdogState
),
2907 .bdrv_needs_filename
= true,
2908 .bdrv_file_open
= sd_open
,
2909 .bdrv_reopen_prepare
= sd_reopen_prepare
,
2910 .bdrv_reopen_commit
= sd_reopen_commit
,
2911 .bdrv_reopen_abort
= sd_reopen_abort
,
2912 .bdrv_close
= sd_close
,
2913 .bdrv_create
= sd_create
,
2914 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2915 .bdrv_getlength
= sd_getlength
,
2916 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
2917 .bdrv_truncate
= sd_truncate
,
2919 .bdrv_co_readv
= sd_co_readv
,
2920 .bdrv_co_writev
= sd_co_writev
,
2921 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
2922 .bdrv_co_pdiscard
= sd_co_pdiscard
,
2923 .bdrv_co_get_block_status
= sd_co_get_block_status
,
2925 .bdrv_snapshot_create
= sd_snapshot_create
,
2926 .bdrv_snapshot_goto
= sd_snapshot_goto
,
2927 .bdrv_snapshot_delete
= sd_snapshot_delete
,
2928 .bdrv_snapshot_list
= sd_snapshot_list
,
2930 .bdrv_save_vmstate
= sd_save_vmstate
,
2931 .bdrv_load_vmstate
= sd_load_vmstate
,
2933 .bdrv_detach_aio_context
= sd_detach_aio_context
,
2934 .bdrv_attach_aio_context
= sd_attach_aio_context
,
2936 .create_opts
= &sd_create_opts
,
2939 static void bdrv_sheepdog_init(void)
2941 bdrv_register(&bdrv_sheepdog
);
2942 bdrv_register(&bdrv_sheepdog_tcp
);
2943 bdrv_register(&bdrv_sheepdog_unix
);
2945 block_init(bdrv_sheepdog_init
);