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-visit.h"
17 #include "qapi/error.h"
18 #include "qapi/qmp/qdict.h"
19 #include "qapi/qmp/qint.h"
20 #include "qapi/qobject-input-visitor.h"
22 #include "qemu/error-report.h"
23 #include "qemu/sockets.h"
24 #include "block/block_int.h"
25 #include "sysemu/block-backend.h"
26 #include "qemu/bitops.h"
27 #include "qemu/cutils.h"
29 #define SD_PROTO_VER 0x01
31 #define SD_DEFAULT_ADDR "localhost"
32 #define SD_DEFAULT_PORT 7000
34 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
35 #define SD_OP_READ_OBJ 0x02
36 #define SD_OP_WRITE_OBJ 0x03
37 /* 0x04 is used internally by Sheepdog */
39 #define SD_OP_NEW_VDI 0x11
40 #define SD_OP_LOCK_VDI 0x12
41 #define SD_OP_RELEASE_VDI 0x13
42 #define SD_OP_GET_VDI_INFO 0x14
43 #define SD_OP_READ_VDIS 0x15
44 #define SD_OP_FLUSH_VDI 0x16
45 #define SD_OP_DEL_VDI 0x17
46 #define SD_OP_GET_CLUSTER_DEFAULT 0x18
48 #define SD_FLAG_CMD_WRITE 0x01
49 #define SD_FLAG_CMD_COW 0x02
50 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
51 #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
53 #define SD_RES_SUCCESS 0x00 /* Success */
54 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
55 #define SD_RES_NO_OBJ 0x02 /* No object found */
56 #define SD_RES_EIO 0x03 /* I/O error */
57 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
58 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
59 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
60 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
61 #define SD_RES_NO_VDI 0x08 /* No vdi found */
62 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
63 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
64 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
65 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
66 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
67 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
68 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
69 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
70 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
71 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
72 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
73 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
74 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
75 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
76 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
77 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
78 #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */
79 #define SD_RES_READONLY 0x1A /* Object is read-only */
84 * 0 - 19 (20 bits): data object space
85 * 20 - 31 (12 bits): reserved data object space
86 * 32 - 55 (24 bits): vdi object space
87 * 56 - 59 ( 4 bits): reserved vdi object space
88 * 60 - 63 ( 4 bits): object type identifier space
91 #define VDI_SPACE_SHIFT 32
92 #define VDI_BIT (UINT64_C(1) << 63)
93 #define VMSTATE_BIT (UINT64_C(1) << 62)
94 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
95 #define MAX_CHILDREN 1024
96 #define SD_MAX_VDI_LEN 256
97 #define SD_MAX_VDI_TAG_LEN 256
98 #define SD_NR_VDIS (1U << 24)
99 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
100 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
101 #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22
103 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
104 * (SD_EC_MAX_STRIP - 1) for parity strips
106 * SD_MAX_COPIES is sum of number of data strips and parity strips.
108 #define SD_EC_MAX_STRIP 16
109 #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
111 #define SD_INODE_SIZE (sizeof(SheepdogInode))
112 #define CURRENT_VDI_ID 0
114 #define LOCK_TYPE_NORMAL 0
115 #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */
117 typedef struct SheepdogReq
{
123 uint32_t data_length
;
124 uint32_t opcode_specific
[8];
127 typedef struct SheepdogRsp
{
133 uint32_t data_length
;
135 uint32_t opcode_specific
[7];
138 typedef struct SheepdogObjReq
{
144 uint32_t data_length
;
153 typedef struct SheepdogObjRsp
{
159 uint32_t data_length
;
167 typedef struct SheepdogVdiReq
{
173 uint32_t data_length
;
175 uint32_t base_vdi_id
;
178 uint8_t store_policy
;
179 uint8_t block_size_shift
;
185 typedef struct SheepdogVdiRsp
{
191 uint32_t data_length
;
198 typedef struct SheepdogClusterRsp
{
204 uint32_t data_length
;
208 uint8_t block_size_shift
;
211 } SheepdogClusterRsp
;
213 typedef struct SheepdogInode
{
214 char name
[SD_MAX_VDI_LEN
];
215 char tag
[SD_MAX_VDI_TAG_LEN
];
218 uint64_t vm_clock_nsec
;
220 uint64_t vm_state_size
;
221 uint16_t copy_policy
;
223 uint8_t block_size_shift
;
226 uint32_t parent_vdi_id
;
227 uint32_t child_vdi_id
[MAX_CHILDREN
];
228 uint32_t data_vdi_id
[MAX_DATA_OBJS
];
231 #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
234 * 64 bit FNV-1a non-zero initial basis
236 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
239 * 64 bit Fowler/Noll/Vo FNV-1a hash code
241 static inline uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
243 unsigned char *bp
= buf
;
244 unsigned char *be
= bp
+ len
;
246 hval
^= (uint64_t) *bp
++;
247 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
248 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
253 static inline bool is_data_obj_writable(SheepdogInode
*inode
, unsigned int idx
)
255 return inode
->vdi_id
== inode
->data_vdi_id
[idx
];
258 static inline bool is_data_obj(uint64_t oid
)
260 return !(VDI_BIT
& oid
);
263 static inline uint64_t data_oid_to_idx(uint64_t oid
)
265 return oid
& (MAX_DATA_OBJS
- 1);
268 static inline uint32_t oid_to_vid(uint64_t oid
)
270 return (oid
& ~VDI_BIT
) >> VDI_SPACE_SHIFT
;
273 static inline uint64_t vid_to_vdi_oid(uint32_t vid
)
275 return VDI_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
);
278 static inline uint64_t vid_to_vmstate_oid(uint32_t vid
, uint32_t idx
)
280 return VMSTATE_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
283 static inline uint64_t vid_to_data_oid(uint32_t vid
, uint32_t idx
)
285 return ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
288 static inline bool is_snapshot(struct SheepdogInode
*inode
)
290 return !!inode
->snap_ctime
;
293 static inline size_t count_data_objs(const struct SheepdogInode
*inode
)
295 return DIV_ROUND_UP(inode
->vdi_size
,
296 (1UL << inode
->block_size_shift
));
301 #define DEBUG_SDOG_PRINT 1
303 #define DEBUG_SDOG_PRINT 0
305 #define DPRINTF(fmt, args...) \
307 if (DEBUG_SDOG_PRINT) { \
308 fprintf(stderr, "%s %d: " fmt, __func__, __LINE__, ##args); \
312 typedef struct SheepdogAIOCB SheepdogAIOCB
;
313 typedef struct BDRVSheepdogState BDRVSheepdogState
;
315 typedef struct AIOReq
{
316 SheepdogAIOCB
*aiocb
;
317 unsigned int iov_offset
;
322 unsigned int data_len
;
327 QLIST_ENTRY(AIOReq
) aio_siblings
;
337 #define AIOCBOverlapping(x, y) \
338 (!(x->max_affect_data_idx < y->min_affect_data_idx \
339 || y->max_affect_data_idx < x->min_affect_data_idx))
341 struct SheepdogAIOCB
{
342 BDRVSheepdogState
*s
;
350 enum AIOCBState aiocb_type
;
352 Coroutine
*coroutine
;
355 uint32_t min_affect_data_idx
;
356 uint32_t max_affect_data_idx
;
359 * The difference between affect_data_idx and dirty_data_idx:
360 * affect_data_idx represents range of index of all request types.
361 * dirty_data_idx represents range of index updated by COW requests.
362 * dirty_data_idx is used for updating an inode object.
364 uint32_t min_dirty_data_idx
;
365 uint32_t max_dirty_data_idx
;
367 QLIST_ENTRY(SheepdogAIOCB
) aiocb_siblings
;
370 struct BDRVSheepdogState
{
371 BlockDriverState
*bs
;
372 AioContext
*aio_context
;
376 char name
[SD_MAX_VDI_LEN
];
378 uint32_t cache_flags
;
379 bool discard_supported
;
388 uint32_t aioreq_seq_num
;
390 /* Every aio request must be linked to either of these queues. */
391 QLIST_HEAD(inflight_aio_head
, AIOReq
) inflight_aio_head
;
392 QLIST_HEAD(failed_aio_head
, AIOReq
) failed_aio_head
;
394 CoQueue overlapping_queue
;
395 QLIST_HEAD(inflight_aiocb_head
, SheepdogAIOCB
) inflight_aiocb_head
;
398 typedef struct BDRVSheepdogReopenState
{
401 } BDRVSheepdogReopenState
;
403 static const char * sd_strerror(int err
)
407 static const struct {
411 {SD_RES_SUCCESS
, "Success"},
412 {SD_RES_UNKNOWN
, "Unknown error"},
413 {SD_RES_NO_OBJ
, "No object found"},
414 {SD_RES_EIO
, "I/O error"},
415 {SD_RES_VDI_EXIST
, "VDI exists already"},
416 {SD_RES_INVALID_PARMS
, "Invalid parameters"},
417 {SD_RES_SYSTEM_ERROR
, "System error"},
418 {SD_RES_VDI_LOCKED
, "VDI is already locked"},
419 {SD_RES_NO_VDI
, "No vdi found"},
420 {SD_RES_NO_BASE_VDI
, "No base VDI found"},
421 {SD_RES_VDI_READ
, "Failed read the requested VDI"},
422 {SD_RES_VDI_WRITE
, "Failed to write the requested VDI"},
423 {SD_RES_BASE_VDI_READ
, "Failed to read the base VDI"},
424 {SD_RES_BASE_VDI_WRITE
, "Failed to write the base VDI"},
425 {SD_RES_NO_TAG
, "Failed to find the requested tag"},
426 {SD_RES_STARTUP
, "The system is still booting"},
427 {SD_RES_VDI_NOT_LOCKED
, "VDI isn't locked"},
428 {SD_RES_SHUTDOWN
, "The system is shutting down"},
429 {SD_RES_NO_MEM
, "Out of memory on the server"},
430 {SD_RES_FULL_VDI
, "We already have the maximum vdis"},
431 {SD_RES_VER_MISMATCH
, "Protocol version mismatch"},
432 {SD_RES_NO_SPACE
, "Server has no space for new objects"},
433 {SD_RES_WAIT_FOR_FORMAT
, "Sheepdog is waiting for a format operation"},
434 {SD_RES_WAIT_FOR_JOIN
, "Sheepdog is waiting for other nodes joining"},
435 {SD_RES_JOIN_FAILED
, "Target node had failed to join sheepdog"},
436 {SD_RES_HALT
, "Sheepdog is stopped serving IO request"},
437 {SD_RES_READONLY
, "Object is read-only"},
440 for (i
= 0; i
< ARRAY_SIZE(errors
); ++i
) {
441 if (errors
[i
].err
== err
) {
442 return errors
[i
].desc
;
446 return "Invalid error code";
450 * Sheepdog I/O handling:
452 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
453 * link the requests to the inflight_list in the
454 * BDRVSheepdogState. The function yields while waiting for
455 * receiving the response.
457 * 2. We receive the response in aio_read_response, the fd handler to
458 * the sheepdog connection. We switch back to sd_co_readv/sd_writev
459 * after all the requests belonging to the AIOCB are finished. If
460 * needed, sd_co_writev will send another requests for the vdi object.
463 static inline AIOReq
*alloc_aio_req(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
,
464 uint64_t oid
, unsigned int data_len
,
465 uint64_t offset
, uint8_t flags
, bool create
,
466 uint64_t base_oid
, unsigned int iov_offset
)
470 aio_req
= g_malloc(sizeof(*aio_req
));
471 aio_req
->aiocb
= acb
;
472 aio_req
->iov_offset
= iov_offset
;
474 aio_req
->base_oid
= base_oid
;
475 aio_req
->offset
= offset
;
476 aio_req
->data_len
= data_len
;
477 aio_req
->flags
= flags
;
478 aio_req
->id
= s
->aioreq_seq_num
++;
479 aio_req
->create
= create
;
485 static void wait_for_overlapping_aiocb(BDRVSheepdogState
*s
, SheepdogAIOCB
*acb
)
490 QLIST_FOREACH(cb
, &s
->inflight_aiocb_head
, aiocb_siblings
) {
491 if (AIOCBOverlapping(acb
, cb
)) {
492 qemu_co_queue_wait(&s
->overlapping_queue
, NULL
);
498 static void sd_aio_setup(SheepdogAIOCB
*acb
, BDRVSheepdogState
*s
,
499 QEMUIOVector
*qiov
, int64_t sector_num
, int nb_sectors
,
502 uint32_t object_size
;
504 object_size
= (UINT32_C(1) << s
->inode
.block_size_shift
);
510 acb
->sector_num
= sector_num
;
511 acb
->nb_sectors
= nb_sectors
;
513 acb
->coroutine
= qemu_coroutine_self();
517 acb
->min_affect_data_idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ object_size
;
518 acb
->max_affect_data_idx
= (acb
->sector_num
* BDRV_SECTOR_SIZE
+
519 acb
->nb_sectors
* BDRV_SECTOR_SIZE
) / object_size
;
521 acb
->min_dirty_data_idx
= UINT32_MAX
;
522 acb
->max_dirty_data_idx
= 0;
523 acb
->aiocb_type
= type
;
525 if (type
== AIOCB_FLUSH_CACHE
) {
529 wait_for_overlapping_aiocb(s
, acb
);
530 QLIST_INSERT_HEAD(&s
->inflight_aiocb_head
, acb
, aiocb_siblings
);
533 static SocketAddress
*sd_socket_address(const char *path
,
534 const char *host
, const char *port
)
536 SocketAddress
*addr
= g_new0(SocketAddress
, 1);
539 addr
->type
= SOCKET_ADDRESS_KIND_UNIX
;
540 addr
->u
.q_unix
.data
= g_new0(UnixSocketAddress
, 1);
541 addr
->u
.q_unix
.data
->path
= g_strdup(path
);
543 addr
->type
= SOCKET_ADDRESS_KIND_INET
;
544 addr
->u
.inet
.data
= g_new0(InetSocketAddress
, 1);
545 addr
->u
.inet
.data
->host
= g_strdup(host
?: SD_DEFAULT_ADDR
);
546 addr
->u
.inet
.data
->port
= g_strdup(port
?: stringify(SD_DEFAULT_PORT
));
552 static SocketAddress
*sd_server_config(QDict
*options
, Error
**errp
)
554 QDict
*server
= NULL
;
555 QObject
*crumpled_server
= NULL
;
557 SocketAddressFlat
*saddr_flat
= NULL
;
558 SocketAddress
*saddr
= NULL
;
559 Error
*local_err
= NULL
;
561 qdict_extract_subqdict(options
, &server
, "server.");
563 crumpled_server
= qdict_crumple(server
, errp
);
564 if (!crumpled_server
) {
569 * FIXME .numeric, .to, .ipv4 or .ipv6 don't work with -drive
570 * server.type=inet. .to doesn't matter, it's ignored anyway.
571 * That's because when @options come from -blockdev or
572 * blockdev_add, members are typed according to the QAPI schema,
573 * but when they come from -drive, they're all QString. The
574 * visitor expects the former.
576 iv
= qobject_input_visitor_new(crumpled_server
);
577 visit_type_SocketAddressFlat(iv
, NULL
, &saddr_flat
, &local_err
);
579 error_propagate(errp
, local_err
);
583 saddr
= socket_address_crumple(saddr_flat
);
586 qapi_free_SocketAddressFlat(saddr_flat
);
588 qobject_decref(crumpled_server
);
593 /* Return -EIO in case of error, file descriptor on success */
594 static int connect_to_sdog(BDRVSheepdogState
*s
, Error
**errp
)
598 fd
= socket_connect(s
->addr
, NULL
, NULL
, errp
);
600 if (s
->addr
->type
== SOCKET_ADDRESS_KIND_INET
&& fd
>= 0) {
601 int ret
= socket_set_nodelay(fd
);
603 error_report("%s", strerror(errno
));
608 qemu_set_nonblock(fd
);
616 /* Return 0 on success and -errno in case of error */
617 static coroutine_fn
int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
622 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
623 if (ret
!= sizeof(*hdr
)) {
624 error_report("failed to send a req, %s", strerror(errno
));
628 ret
= qemu_co_send(sockfd
, data
, *wlen
);
630 error_report("failed to send a req, %s", strerror(errno
));
637 typedef struct SheepdogReqCo
{
639 BlockDriverState
*bs
;
640 AioContext
*aio_context
;
650 static void restart_co_req(void *opaque
)
652 SheepdogReqCo
*srco
= opaque
;
654 aio_co_wake(srco
->co
);
657 static coroutine_fn
void do_co_req(void *opaque
)
660 SheepdogReqCo
*srco
= opaque
;
661 int sockfd
= srco
->sockfd
;
662 SheepdogReq
*hdr
= srco
->hdr
;
663 void *data
= srco
->data
;
664 unsigned int *wlen
= srco
->wlen
;
665 unsigned int *rlen
= srco
->rlen
;
667 srco
->co
= qemu_coroutine_self();
668 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
669 NULL
, restart_co_req
, NULL
, srco
);
671 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
676 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
677 restart_co_req
, NULL
, NULL
, srco
);
679 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
680 if (ret
!= sizeof(*hdr
)) {
681 error_report("failed to get a rsp, %s", strerror(errno
));
686 if (*rlen
> hdr
->data_length
) {
687 *rlen
= hdr
->data_length
;
691 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
693 error_report("failed to get the data, %s", strerror(errno
));
700 /* there is at most one request for this sockfd, so it is safe to
701 * set each handler to NULL. */
702 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
703 NULL
, NULL
, NULL
, NULL
);
707 srco
->finished
= true;
709 bdrv_wakeup(srco
->bs
);
714 * Send the request to the sheep in a synchronous manner.
716 * Return 0 on success, -errno in case of error.
718 static int do_req(int sockfd
, BlockDriverState
*bs
, SheepdogReq
*hdr
,
719 void *data
, unsigned int *wlen
, unsigned int *rlen
)
722 SheepdogReqCo srco
= {
724 .aio_context
= bs
? bdrv_get_aio_context(bs
) : qemu_get_aio_context(),
734 if (qemu_in_coroutine()) {
737 co
= qemu_coroutine_create(do_co_req
, &srco
);
739 bdrv_coroutine_enter(bs
, co
);
740 BDRV_POLL_WHILE(bs
, !srco
.finished
);
742 qemu_coroutine_enter(co
);
743 while (!srco
.finished
) {
744 aio_poll(qemu_get_aio_context(), true);
752 static void coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
753 struct iovec
*iov
, int niov
,
754 enum AIOCBState aiocb_type
);
755 static void coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
);
756 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
);
757 static int get_sheep_fd(BDRVSheepdogState
*s
, Error
**errp
);
758 static void co_write_request(void *opaque
);
760 static coroutine_fn
void reconnect_to_sdog(void *opaque
)
762 BDRVSheepdogState
*s
= opaque
;
763 AIOReq
*aio_req
, *next
;
765 aio_set_fd_handler(s
->aio_context
, s
->fd
, false, NULL
,
770 /* Wait for outstanding write requests to be completed. */
771 while (s
->co_send
!= NULL
) {
772 co_write_request(opaque
);
775 /* Try to reconnect the sheepdog server every one second. */
777 Error
*local_err
= NULL
;
778 s
->fd
= get_sheep_fd(s
, &local_err
);
780 DPRINTF("Wait for connection to be established\n");
781 error_report_err(local_err
);
782 co_aio_sleep_ns(bdrv_get_aio_context(s
->bs
), QEMU_CLOCK_REALTIME
,
788 * Now we have to resend all the request in the inflight queue. However,
789 * resend_aioreq() can yield and newly created requests can be added to the
790 * inflight queue before the coroutine is resumed. To avoid mixing them, we
791 * have to move all the inflight requests to the failed queue before
792 * resend_aioreq() is called.
794 QLIST_FOREACH_SAFE(aio_req
, &s
->inflight_aio_head
, aio_siblings
, next
) {
795 QLIST_REMOVE(aio_req
, aio_siblings
);
796 QLIST_INSERT_HEAD(&s
->failed_aio_head
, aio_req
, aio_siblings
);
799 /* Resend all the failed aio requests. */
800 while (!QLIST_EMPTY(&s
->failed_aio_head
)) {
801 aio_req
= QLIST_FIRST(&s
->failed_aio_head
);
802 QLIST_REMOVE(aio_req
, aio_siblings
);
803 resend_aioreq(s
, aio_req
);
808 * Receive responses of the I/O requests.
810 * This function is registered as a fd handler, and called from the
811 * main loop when s->fd is ready for reading responses.
813 static void coroutine_fn
aio_read_response(void *opaque
)
816 BDRVSheepdogState
*s
= opaque
;
819 AIOReq
*aio_req
= NULL
;
824 ret
= qemu_co_recv(fd
, &rsp
, sizeof(rsp
));
825 if (ret
!= sizeof(rsp
)) {
826 error_report("failed to get the header, %s", strerror(errno
));
830 /* find the right aio_req from the inflight aio list */
831 QLIST_FOREACH(aio_req
, &s
->inflight_aio_head
, aio_siblings
) {
832 if (aio_req
->id
== rsp
.id
) {
837 error_report("cannot find aio_req %x", rsp
.id
);
841 acb
= aio_req
->aiocb
;
843 switch (acb
->aiocb_type
) {
844 case AIOCB_WRITE_UDATA
:
845 if (!is_data_obj(aio_req
->oid
)) {
848 idx
= data_oid_to_idx(aio_req
->oid
);
850 if (aio_req
->create
) {
852 * If the object is newly created one, we need to update
853 * the vdi object (metadata object). min_dirty_data_idx
854 * and max_dirty_data_idx are changed to include updated
855 * index between them.
857 if (rsp
.result
== SD_RES_SUCCESS
) {
858 s
->inode
.data_vdi_id
[idx
] = s
->inode
.vdi_id
;
859 acb
->max_dirty_data_idx
= MAX(idx
, acb
->max_dirty_data_idx
);
860 acb
->min_dirty_data_idx
= MIN(idx
, acb
->min_dirty_data_idx
);
864 case AIOCB_READ_UDATA
:
865 ret
= qemu_co_recvv(fd
, acb
->qiov
->iov
, acb
->qiov
->niov
,
866 aio_req
->iov_offset
, rsp
.data_length
);
867 if (ret
!= rsp
.data_length
) {
868 error_report("failed to get the data, %s", strerror(errno
));
872 case AIOCB_FLUSH_CACHE
:
873 if (rsp
.result
== SD_RES_INVALID_PARMS
) {
874 DPRINTF("disable cache since the server doesn't support it\n");
875 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
876 rsp
.result
= SD_RES_SUCCESS
;
879 case AIOCB_DISCARD_OBJ
:
880 switch (rsp
.result
) {
881 case SD_RES_INVALID_PARMS
:
882 error_report("server doesn't support discard command");
883 rsp
.result
= SD_RES_SUCCESS
;
884 s
->discard_supported
= false;
891 /* No more data for this aio_req (reload_inode below uses its own file
892 * descriptor handler which doesn't use co_recv).
896 QLIST_REMOVE(aio_req
, aio_siblings
);
897 switch (rsp
.result
) {
900 case SD_RES_READONLY
:
901 if (s
->inode
.vdi_id
== oid_to_vid(aio_req
->oid
)) {
902 ret
= reload_inode(s
, 0, "");
907 if (is_data_obj(aio_req
->oid
)) {
908 aio_req
->oid
= vid_to_data_oid(s
->inode
.vdi_id
,
909 data_oid_to_idx(aio_req
->oid
));
911 aio_req
->oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
913 resend_aioreq(s
, aio_req
);
917 error_report("%s", sd_strerror(rsp
.result
));
923 if (!--acb
->nr_pending
) {
925 * We've finished all requests which belong to the AIOCB, so
926 * we can switch back to sd_co_readv/writev now.
928 aio_co_wake(acb
->coroutine
);
934 reconnect_to_sdog(opaque
);
937 static void co_read_response(void *opaque
)
939 BDRVSheepdogState
*s
= opaque
;
942 s
->co_recv
= qemu_coroutine_create(aio_read_response
, opaque
);
945 aio_co_enter(s
->aio_context
, s
->co_recv
);
948 static void co_write_request(void *opaque
)
950 BDRVSheepdogState
*s
= opaque
;
952 aio_co_wake(s
->co_send
);
956 * Return a socket descriptor to read/write objects.
958 * We cannot use this descriptor for other operations because
959 * the block driver may be on waiting response from the server.
961 static int get_sheep_fd(BDRVSheepdogState
*s
, Error
**errp
)
965 fd
= connect_to_sdog(s
, errp
);
970 aio_set_fd_handler(s
->aio_context
, fd
, false,
971 co_read_response
, NULL
, NULL
, s
);
976 * Parse numeric snapshot ID in @str
977 * If @str can't be parsed as number, return false.
978 * Else, if the number is zero or too large, set *@snapid to zero and
980 * Else, set *@snapid to the number and return true.
982 static bool sd_parse_snapid(const char *str
, uint32_t *snapid
)
987 ret
= qemu_strtoul(str
, NULL
, 10, &ul
);
988 if (ret
== -ERANGE
) {
994 if (ul
> UINT32_MAX
) {
1002 static bool sd_parse_snapid_or_tag(const char *str
,
1003 uint32_t *snapid
, char tag
[])
1005 if (!sd_parse_snapid(str
, snapid
)) {
1007 if (g_strlcpy(tag
, str
, SD_MAX_VDI_TAG_LEN
) >= SD_MAX_VDI_TAG_LEN
) {
1010 } else if (!*snapid
) {
1019 const char *path
; /* non-null iff transport is tcp */
1020 const char *host
; /* valid when transport is tcp */
1021 int port
; /* valid when transport is tcp */
1022 char vdi
[SD_MAX_VDI_LEN
];
1023 char tag
[SD_MAX_VDI_TAG_LEN
];
1025 /* Remainder is only for sd_config_done() */
1030 static void sd_config_done(SheepdogConfig
*cfg
)
1033 query_params_free(cfg
->qp
);
1038 static void sd_parse_uri(SheepdogConfig
*cfg
, const char *filename
,
1042 QueryParams
*qp
= NULL
;
1046 memset(cfg
, 0, sizeof(*cfg
));
1048 cfg
->uri
= uri
= uri_parse(filename
);
1050 error_setg(&err
, "invalid URI");
1055 if (!strcmp(uri
->scheme
, "sheepdog")) {
1057 } else if (!strcmp(uri
->scheme
, "sheepdog+tcp")) {
1059 } else if (!strcmp(uri
->scheme
, "sheepdog+unix")) {
1062 error_setg(&err
, "URI scheme must be 'sheepdog', 'sheepdog+tcp',"
1063 " or 'sheepdog+unix'");
1067 if (uri
->path
== NULL
|| !strcmp(uri
->path
, "/")) {
1068 error_setg(&err
, "missing file path in URI");
1071 if (g_strlcpy(cfg
->vdi
, uri
->path
+ 1, SD_MAX_VDI_LEN
)
1072 >= SD_MAX_VDI_LEN
) {
1073 error_setg(&err
, "VDI name is too long");
1077 cfg
->qp
= qp
= query_params_parse(uri
->query
);
1080 /* sheepdog+unix:///vdiname?socket=path */
1081 if (uri
->server
|| uri
->port
) {
1082 error_setg(&err
, "URI scheme %s doesn't accept a server address",
1088 "URI scheme %s requires query parameter 'socket'",
1092 if (qp
->n
!= 1 || strcmp(qp
->p
[0].name
, "socket")) {
1093 error_setg(&err
, "unexpected query parameters");
1096 cfg
->path
= qp
->p
[0].value
;
1098 /* sheepdog[+tcp]://[host:port]/vdiname */
1100 error_setg(&err
, "unexpected query parameters");
1103 cfg
->host
= uri
->server
;
1104 cfg
->port
= uri
->port
;
1108 if (uri
->fragment
) {
1109 if (!sd_parse_snapid_or_tag(uri
->fragment
,
1110 &cfg
->snap_id
, cfg
->tag
)) {
1111 error_setg(&err
, "'%s' is not a valid snapshot ID",
1116 cfg
->snap_id
= CURRENT_VDI_ID
; /* search current vdi */
1121 error_propagate(errp
, err
);
1122 sd_config_done(cfg
);
1127 * Parse a filename (old syntax)
1129 * filename must be one of the following formats:
1131 * 2. [vdiname]:[snapid]
1132 * 3. [vdiname]:[tag]
1133 * 4. [hostname]:[port]:[vdiname]
1134 * 5. [hostname]:[port]:[vdiname]:[snapid]
1135 * 6. [hostname]:[port]:[vdiname]:[tag]
1137 * You can boot from the snapshot images by specifying `snapid` or
1140 * You can run VMs outside the Sheepdog cluster by specifying
1141 * `hostname' and `port' (experimental).
1143 static void parse_vdiname(SheepdogConfig
*cfg
, const char *filename
,
1148 const char *host_spec
, *vdi_spec
;
1151 strstart(filename
, "sheepdog:", &filename
);
1152 p
= q
= g_strdup(filename
);
1154 /* count the number of separators */
1164 /* use the first two tokens as host_spec. */
1177 p
= strchr(vdi_spec
, ':');
1182 uri
= g_strdup_printf("sheepdog://%s/%s", host_spec
, vdi_spec
);
1185 * FIXME We to escape URI meta-characters, e.g. "x?y=z"
1186 * produces "sheepdog://x?y=z". Because of that ...
1188 sd_parse_uri(cfg
, uri
, &err
);
1191 * ... this can fail, but the error message is misleading.
1192 * Replace it by the traditional useless one until the
1193 * escaping is fixed.
1196 error_setg(errp
, "Can't parse filename");
1203 static void sd_parse_filename(const char *filename
, QDict
*options
,
1210 if (strstr(filename
, "://")) {
1211 sd_parse_uri(&cfg
, filename
, &err
);
1213 parse_vdiname(&cfg
, filename
, &err
);
1216 error_propagate(errp
, err
);
1221 qdict_set_default_str(options
, "server.path", cfg
.path
);
1222 qdict_set_default_str(options
, "server.type", "unix");
1224 qdict_set_default_str(options
, "server.type", "inet");
1225 qdict_set_default_str(options
, "server.host",
1226 cfg
.host
?: SD_DEFAULT_ADDR
);
1227 snprintf(buf
, sizeof(buf
), "%d", cfg
.port
?: SD_DEFAULT_PORT
);
1228 qdict_set_default_str(options
, "server.port", buf
);
1230 qdict_set_default_str(options
, "vdi", cfg
.vdi
);
1231 qdict_set_default_str(options
, "tag", cfg
.tag
);
1233 snprintf(buf
, sizeof(buf
), "%d", cfg
.snap_id
);
1234 qdict_set_default_str(options
, "snap-id", buf
);
1237 sd_config_done(&cfg
);
1240 static int find_vdi_name(BDRVSheepdogState
*s
, const char *filename
,
1241 uint32_t snapid
, const char *tag
, uint32_t *vid
,
1242 bool lock
, Error
**errp
)
1246 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1247 unsigned int wlen
, rlen
= 0;
1248 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
1250 fd
= connect_to_sdog(s
, errp
);
1255 /* This pair of strncpy calls ensures that the buffer is zero-filled,
1256 * which is desirable since we'll soon be sending those bytes, and
1257 * don't want the send_req to read uninitialized data.
1259 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
1260 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
1262 memset(&hdr
, 0, sizeof(hdr
));
1264 hdr
.opcode
= SD_OP_LOCK_VDI
;
1265 hdr
.type
= LOCK_TYPE_NORMAL
;
1267 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
1269 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
1270 hdr
.proto_ver
= SD_PROTO_VER
;
1271 hdr
.data_length
= wlen
;
1272 hdr
.snapid
= snapid
;
1273 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1275 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1277 error_setg_errno(errp
, -ret
, "cannot get vdi info");
1281 if (rsp
->result
!= SD_RES_SUCCESS
) {
1282 error_setg(errp
, "cannot get vdi info, %s, %s %" PRIu32
" %s",
1283 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
1284 if (rsp
->result
== SD_RES_NO_VDI
) {
1286 } else if (rsp
->result
== SD_RES_VDI_LOCKED
) {
1301 static void coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
1302 struct iovec
*iov
, int niov
,
1303 enum AIOCBState aiocb_type
)
1305 int nr_copies
= s
->inode
.nr_copies
;
1307 unsigned int wlen
= 0;
1309 uint64_t oid
= aio_req
->oid
;
1310 unsigned int datalen
= aio_req
->data_len
;
1311 uint64_t offset
= aio_req
->offset
;
1312 uint8_t flags
= aio_req
->flags
;
1313 uint64_t old_oid
= aio_req
->base_oid
;
1314 bool create
= aio_req
->create
;
1316 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1319 error_report("bug");
1322 memset(&hdr
, 0, sizeof(hdr
));
1324 switch (aiocb_type
) {
1325 case AIOCB_FLUSH_CACHE
:
1326 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1328 case AIOCB_READ_UDATA
:
1329 hdr
.opcode
= SD_OP_READ_OBJ
;
1332 case AIOCB_WRITE_UDATA
:
1334 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1336 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1339 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1341 case AIOCB_DISCARD_OBJ
:
1342 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1343 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1344 s
->inode
.data_vdi_id
[data_oid_to_idx(oid
)] = 0;
1345 offset
= offsetof(SheepdogInode
,
1346 data_vdi_id
[data_oid_to_idx(oid
)]);
1347 oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
1348 wlen
= datalen
= sizeof(uint32_t);
1352 if (s
->cache_flags
) {
1353 hdr
.flags
|= s
->cache_flags
;
1357 hdr
.cow_oid
= old_oid
;
1358 hdr
.copies
= s
->inode
.nr_copies
;
1360 hdr
.data_length
= datalen
;
1361 hdr
.offset
= offset
;
1363 hdr
.id
= aio_req
->id
;
1365 qemu_co_mutex_lock(&s
->lock
);
1366 s
->co_send
= qemu_coroutine_self();
1367 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1368 co_read_response
, co_write_request
, NULL
, s
);
1369 socket_set_cork(s
->fd
, 1);
1372 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
1373 if (ret
!= sizeof(hdr
)) {
1374 error_report("failed to send a req, %s", strerror(errno
));
1379 ret
= qemu_co_sendv(s
->fd
, iov
, niov
, aio_req
->iov_offset
, wlen
);
1381 error_report("failed to send a data, %s", strerror(errno
));
1385 socket_set_cork(s
->fd
, 0);
1386 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1387 co_read_response
, NULL
, NULL
, s
);
1389 qemu_co_mutex_unlock(&s
->lock
);
1392 static int read_write_object(int fd
, BlockDriverState
*bs
, char *buf
,
1393 uint64_t oid
, uint8_t copies
,
1394 unsigned int datalen
, uint64_t offset
,
1395 bool write
, bool create
, uint32_t cache_flags
)
1398 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1399 unsigned int wlen
, rlen
;
1402 memset(&hdr
, 0, sizeof(hdr
));
1407 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1409 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1411 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1416 hdr
.opcode
= SD_OP_READ_OBJ
;
1419 hdr
.flags
|= cache_flags
;
1422 hdr
.data_length
= datalen
;
1423 hdr
.offset
= offset
;
1424 hdr
.copies
= copies
;
1426 ret
= do_req(fd
, bs
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1428 error_report("failed to send a request to the sheep");
1432 switch (rsp
->result
) {
1433 case SD_RES_SUCCESS
:
1436 error_report("%s", sd_strerror(rsp
->result
));
1441 static int read_object(int fd
, BlockDriverState
*bs
, char *buf
,
1442 uint64_t oid
, uint8_t copies
,
1443 unsigned int datalen
, uint64_t offset
,
1444 uint32_t cache_flags
)
1446 return read_write_object(fd
, bs
, buf
, oid
, copies
,
1447 datalen
, offset
, false,
1448 false, cache_flags
);
1451 static int write_object(int fd
, BlockDriverState
*bs
, char *buf
,
1452 uint64_t oid
, uint8_t copies
,
1453 unsigned int datalen
, uint64_t offset
, bool create
,
1454 uint32_t cache_flags
)
1456 return read_write_object(fd
, bs
, buf
, oid
, copies
,
1457 datalen
, offset
, true,
1458 create
, cache_flags
);
1461 /* update inode with the latest state */
1462 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
)
1464 Error
*local_err
= NULL
;
1465 SheepdogInode
*inode
;
1469 fd
= connect_to_sdog(s
, &local_err
);
1471 error_report_err(local_err
);
1475 inode
= g_malloc(SD_INODE_HEADER_SIZE
);
1477 ret
= find_vdi_name(s
, s
->name
, snapid
, tag
, &vid
, false, &local_err
);
1479 error_report_err(local_err
);
1483 ret
= read_object(fd
, s
->bs
, (char *)inode
, vid_to_vdi_oid(vid
),
1484 s
->inode
.nr_copies
, SD_INODE_HEADER_SIZE
, 0,
1490 if (inode
->vdi_id
!= s
->inode
.vdi_id
) {
1491 memcpy(&s
->inode
, inode
, SD_INODE_HEADER_SIZE
);
1501 static void coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
1503 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
1505 aio_req
->create
= false;
1507 /* check whether this request becomes a CoW one */
1508 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& is_data_obj(aio_req
->oid
)) {
1509 int idx
= data_oid_to_idx(aio_req
->oid
);
1511 if (is_data_obj_writable(&s
->inode
, idx
)) {
1515 if (s
->inode
.data_vdi_id
[idx
]) {
1516 aio_req
->base_oid
= vid_to_data_oid(s
->inode
.data_vdi_id
[idx
], idx
);
1517 aio_req
->flags
|= SD_FLAG_CMD_COW
;
1519 aio_req
->create
= true;
1522 if (is_data_obj(aio_req
->oid
)) {
1523 add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1527 iov
.iov_base
= &s
->inode
;
1528 iov
.iov_len
= sizeof(s
->inode
);
1529 add_aio_request(s
, aio_req
, &iov
, 1, AIOCB_WRITE_UDATA
);
1533 static void sd_detach_aio_context(BlockDriverState
*bs
)
1535 BDRVSheepdogState
*s
= bs
->opaque
;
1537 aio_set_fd_handler(s
->aio_context
, s
->fd
, false, NULL
,
1541 static void sd_attach_aio_context(BlockDriverState
*bs
,
1542 AioContext
*new_context
)
1544 BDRVSheepdogState
*s
= bs
->opaque
;
1546 s
->aio_context
= new_context
;
1547 aio_set_fd_handler(new_context
, s
->fd
, false,
1548 co_read_response
, NULL
, NULL
, s
);
1551 static QemuOptsList runtime_opts
= {
1553 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1557 .type
= QEMU_OPT_STRING
,
1561 .type
= QEMU_OPT_NUMBER
,
1565 .type
= QEMU_OPT_STRING
,
1567 { /* end of list */ }
1571 static int sd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1576 BDRVSheepdogState
*s
= bs
->opaque
;
1577 const char *vdi
, *snap_id_str
, *tag
;
1581 Error
*local_err
= NULL
;
1584 s
->aio_context
= bdrv_get_aio_context(bs
);
1586 opts
= qemu_opts_create(&runtime_opts
, NULL
, 0, &error_abort
);
1587 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1589 error_propagate(errp
, local_err
);
1594 s
->addr
= sd_server_config(options
, errp
);
1600 vdi
= qemu_opt_get(opts
, "vdi");
1601 snap_id_str
= qemu_opt_get(opts
, "snap-id");
1602 snap_id
= qemu_opt_get_number(opts
, "snap-id", CURRENT_VDI_ID
);
1603 tag
= qemu_opt_get(opts
, "tag");
1606 error_setg(errp
, "parameter 'vdi' is missing");
1610 if (strlen(vdi
) >= SD_MAX_VDI_LEN
) {
1611 error_setg(errp
, "value of parameter 'vdi' is too long");
1616 if (snap_id
> UINT32_MAX
) {
1619 if (snap_id_str
&& !snap_id
) {
1620 error_setg(errp
, "'snap-id=%s' is not a valid snapshot ID",
1629 if (tag
&& strlen(tag
) >= SD_MAX_VDI_TAG_LEN
) {
1630 error_setg(errp
, "value of parameter 'tag' is too long");
1635 QLIST_INIT(&s
->inflight_aio_head
);
1636 QLIST_INIT(&s
->failed_aio_head
);
1637 QLIST_INIT(&s
->inflight_aiocb_head
);
1639 s
->fd
= get_sheep_fd(s
, errp
);
1645 ret
= find_vdi_name(s
, vdi
, (uint32_t)snap_id
, tag
, &vid
, true, errp
);
1651 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1652 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1654 s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1655 if (flags
& BDRV_O_NOCACHE
) {
1656 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1658 s
->discard_supported
= true;
1660 if (snap_id
|| tag
[0]) {
1661 DPRINTF("%" PRIx32
" snapshot inode was open.\n", vid
);
1662 s
->is_snapshot
= true;
1665 fd
= connect_to_sdog(s
, errp
);
1671 buf
= g_malloc(SD_INODE_SIZE
);
1672 ret
= read_object(fd
, s
->bs
, buf
, vid_to_vdi_oid(vid
),
1673 0, SD_INODE_SIZE
, 0, s
->cache_flags
);
1678 error_setg(errp
, "Can't read snapshot inode");
1682 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1684 bs
->total_sectors
= s
->inode
.vdi_size
/ BDRV_SECTOR_SIZE
;
1685 pstrcpy(s
->name
, sizeof(s
->name
), vdi
);
1686 qemu_co_mutex_init(&s
->lock
);
1687 qemu_co_queue_init(&s
->overlapping_queue
);
1688 qemu_opts_del(opts
);
1693 aio_set_fd_handler(bdrv_get_aio_context(bs
), s
->fd
,
1694 false, NULL
, NULL
, NULL
, NULL
);
1697 qemu_opts_del(opts
);
1702 static int sd_reopen_prepare(BDRVReopenState
*state
, BlockReopenQueue
*queue
,
1705 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1706 BDRVSheepdogReopenState
*re_s
;
1709 re_s
= state
->opaque
= g_new0(BDRVSheepdogReopenState
, 1);
1711 re_s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1712 if (state
->flags
& BDRV_O_NOCACHE
) {
1713 re_s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1716 re_s
->fd
= get_sheep_fd(s
, errp
);
1725 static void sd_reopen_commit(BDRVReopenState
*state
)
1727 BDRVSheepdogReopenState
*re_s
= state
->opaque
;
1728 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1731 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1732 NULL
, NULL
, NULL
, NULL
);
1737 s
->cache_flags
= re_s
->cache_flags
;
1739 g_free(state
->opaque
);
1740 state
->opaque
= NULL
;
1745 static void sd_reopen_abort(BDRVReopenState
*state
)
1747 BDRVSheepdogReopenState
*re_s
= state
->opaque
;
1748 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1755 aio_set_fd_handler(s
->aio_context
, re_s
->fd
, false,
1756 NULL
, NULL
, NULL
, NULL
);
1757 closesocket(re_s
->fd
);
1760 g_free(state
->opaque
);
1761 state
->opaque
= NULL
;
1766 static int do_sd_create(BDRVSheepdogState
*s
, uint32_t *vdi_id
, int snapshot
,
1770 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1772 unsigned int wlen
, rlen
= 0;
1773 char buf
[SD_MAX_VDI_LEN
];
1775 fd
= connect_to_sdog(s
, errp
);
1780 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1781 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1783 memset(buf
, 0, sizeof(buf
));
1784 pstrcpy(buf
, sizeof(buf
), s
->name
);
1786 memset(&hdr
, 0, sizeof(hdr
));
1787 hdr
.opcode
= SD_OP_NEW_VDI
;
1788 hdr
.base_vdi_id
= s
->inode
.vdi_id
;
1790 wlen
= SD_MAX_VDI_LEN
;
1792 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1793 hdr
.snapid
= snapshot
;
1795 hdr
.data_length
= wlen
;
1796 hdr
.vdi_size
= s
->inode
.vdi_size
;
1797 hdr
.copy_policy
= s
->inode
.copy_policy
;
1798 hdr
.copies
= s
->inode
.nr_copies
;
1799 hdr
.block_size_shift
= s
->inode
.block_size_shift
;
1801 ret
= do_req(fd
, NULL
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1806 error_setg_errno(errp
, -ret
, "create failed");
1810 if (rsp
->result
!= SD_RES_SUCCESS
) {
1811 error_setg(errp
, "%s, %s", sd_strerror(rsp
->result
), s
->inode
.name
);
1816 *vdi_id
= rsp
->vdi_id
;
1822 static int sd_prealloc(const char *filename
, Error
**errp
)
1824 BlockBackend
*blk
= NULL
;
1825 BDRVSheepdogState
*base
= NULL
;
1826 unsigned long buf_size
;
1827 uint32_t idx
, max_idx
;
1828 uint32_t object_size
;
1833 blk
= blk_new_open(filename
, NULL
, NULL
,
1834 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
1837 goto out_with_err_set
;
1840 blk_set_allow_write_beyond_eof(blk
, true);
1842 vdi_size
= blk_getlength(blk
);
1848 base
= blk_bs(blk
)->opaque
;
1849 object_size
= (UINT32_C(1) << base
->inode
.block_size_shift
);
1850 buf_size
= MIN(object_size
, SD_DATA_OBJ_SIZE
);
1851 buf
= g_malloc0(buf_size
);
1853 max_idx
= DIV_ROUND_UP(vdi_size
, buf_size
);
1855 for (idx
= 0; idx
< max_idx
; idx
++) {
1857 * The created image can be a cloned image, so we need to read
1858 * a data from the source image.
1860 ret
= blk_pread(blk
, idx
* buf_size
, buf
, buf_size
);
1864 ret
= blk_pwrite(blk
, idx
* buf_size
, buf
, buf_size
, 0);
1873 error_setg_errno(errp
, -ret
, "Can't pre-allocate");
1885 * Sheepdog support two kinds of redundancy, full replication and erasure
1888 * # create a fully replicated vdi with x copies
1889 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1891 * # create a erasure coded vdi with x data strips and y parity strips
1892 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1894 static int parse_redundancy(BDRVSheepdogState
*s
, const char *opt
)
1896 struct SheepdogInode
*inode
= &s
->inode
;
1897 const char *n1
, *n2
;
1901 pstrcpy(p
, sizeof(p
), opt
);
1902 n1
= strtok(p
, ":");
1903 n2
= strtok(NULL
, ":");
1909 copy
= strtol(n1
, NULL
, 10);
1910 /* FIXME fix error checking by switching to qemu_strtol() */
1911 if (copy
> SD_MAX_COPIES
|| copy
< 1) {
1915 inode
->copy_policy
= 0;
1916 inode
->nr_copies
= copy
;
1920 if (copy
!= 2 && copy
!= 4 && copy
!= 8 && copy
!= 16) {
1924 parity
= strtol(n2
, NULL
, 10);
1925 /* FIXME fix error checking by switching to qemu_strtol() */
1926 if (parity
>= SD_EC_MAX_STRIP
|| parity
< 1) {
1931 * 4 bits for parity and 4 bits for data.
1932 * We have to compress upper data bits because it can't represent 16
1934 inode
->copy_policy
= ((copy
/ 2) << 4) + parity
;
1935 inode
->nr_copies
= copy
+ parity
;
1940 static int parse_block_size_shift(BDRVSheepdogState
*s
, QemuOpts
*opt
)
1942 struct SheepdogInode
*inode
= &s
->inode
;
1943 uint64_t object_size
;
1946 object_size
= qemu_opt_get_size_del(opt
, BLOCK_OPT_OBJECT_SIZE
, 0);
1948 if ((object_size
- 1) & object_size
) { /* not a power of 2? */
1951 obj_order
= ctz32(object_size
);
1952 if (obj_order
< 20 || obj_order
> 31) {
1955 inode
->block_size_shift
= (uint8_t)obj_order
;
1961 static int sd_create(const char *filename
, QemuOpts
*opts
,
1967 char *backing_file
= NULL
;
1969 BDRVSheepdogState
*s
;
1971 uint64_t max_vdi_size
;
1972 bool prealloc
= false;
1974 s
= g_new0(BDRVSheepdogState
, 1);
1976 if (strstr(filename
, "://")) {
1977 sd_parse_uri(&cfg
, filename
, &err
);
1979 parse_vdiname(&cfg
, filename
, &err
);
1982 error_propagate(errp
, err
);
1986 buf
= cfg
.port
? g_strdup_printf("%d", cfg
.port
) : NULL
;
1987 s
->addr
= sd_socket_address(cfg
.path
, cfg
.host
, buf
);
1989 strcpy(s
->name
, cfg
.vdi
);
1990 sd_config_done(&cfg
);
1992 s
->inode
.vdi_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1994 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
1995 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
1996 if (!buf
|| !strcmp(buf
, "off")) {
1998 } else if (!strcmp(buf
, "full")) {
2001 error_setg(errp
, "Invalid preallocation mode: '%s'", buf
);
2007 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_REDUNDANCY
);
2009 ret
= parse_redundancy(s
, buf
);
2011 error_setg(errp
, "Invalid redundancy mode: '%s'", buf
);
2015 ret
= parse_block_size_shift(s
, opts
);
2017 error_setg(errp
, "Invalid object_size."
2018 " obect_size needs to be power of 2"
2019 " and be limited from 2^20 to 2^31");
2025 BDRVSheepdogState
*base
;
2028 /* Currently, only Sheepdog backing image is supported. */
2029 drv
= bdrv_find_protocol(backing_file
, true, NULL
);
2030 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
2031 error_setg(errp
, "backing_file must be a sheepdog image");
2036 blk
= blk_new_open(backing_file
, NULL
, NULL
,
2037 BDRV_O_PROTOCOL
, errp
);
2043 base
= blk_bs(blk
)->opaque
;
2045 if (!is_snapshot(&base
->inode
)) {
2046 error_setg(errp
, "cannot clone from a non snapshot vdi");
2051 s
->inode
.vdi_id
= base
->inode
.vdi_id
;
2055 s
->aio_context
= qemu_get_aio_context();
2057 /* if block_size_shift is not specified, get cluster default value */
2058 if (s
->inode
.block_size_shift
== 0) {
2060 SheepdogClusterRsp
*rsp
= (SheepdogClusterRsp
*)&hdr
;
2062 unsigned int wlen
= 0, rlen
= 0;
2064 fd
= connect_to_sdog(s
, errp
);
2070 memset(&hdr
, 0, sizeof(hdr
));
2071 hdr
.opcode
= SD_OP_GET_CLUSTER_DEFAULT
;
2072 hdr
.proto_ver
= SD_PROTO_VER
;
2074 ret
= do_req(fd
, NULL
, (SheepdogReq
*)&hdr
,
2075 NULL
, &wlen
, &rlen
);
2078 error_setg_errno(errp
, -ret
, "failed to get cluster default");
2081 if (rsp
->result
== SD_RES_SUCCESS
) {
2082 s
->inode
.block_size_shift
= rsp
->block_size_shift
;
2084 s
->inode
.block_size_shift
= SD_DEFAULT_BLOCK_SIZE_SHIFT
;
2088 max_vdi_size
= (UINT64_C(1) << s
->inode
.block_size_shift
) * MAX_DATA_OBJS
;
2090 if (s
->inode
.vdi_size
> max_vdi_size
) {
2091 error_setg(errp
, "An image is too large."
2092 " The maximum image size is %"PRIu64
"GB",
2093 max_vdi_size
/ 1024 / 1024 / 1024);
2098 ret
= do_sd_create(s
, &vid
, 0, errp
);
2104 ret
= sd_prealloc(filename
, errp
);
2107 g_free(backing_file
);
2113 static void sd_close(BlockDriverState
*bs
)
2115 Error
*local_err
= NULL
;
2116 BDRVSheepdogState
*s
= bs
->opaque
;
2118 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2119 unsigned int wlen
, rlen
= 0;
2122 DPRINTF("%s\n", s
->name
);
2124 fd
= connect_to_sdog(s
, &local_err
);
2126 error_report_err(local_err
);
2130 memset(&hdr
, 0, sizeof(hdr
));
2132 hdr
.opcode
= SD_OP_RELEASE_VDI
;
2133 hdr
.type
= LOCK_TYPE_NORMAL
;
2134 hdr
.base_vdi_id
= s
->inode
.vdi_id
;
2135 wlen
= strlen(s
->name
) + 1;
2136 hdr
.data_length
= wlen
;
2137 hdr
.flags
= SD_FLAG_CMD_WRITE
;
2139 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2140 s
->name
, &wlen
, &rlen
);
2144 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
2145 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
2146 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
2149 aio_set_fd_handler(bdrv_get_aio_context(bs
), s
->fd
,
2150 false, NULL
, NULL
, NULL
, NULL
);
2152 qapi_free_SocketAddress(s
->addr
);
2155 static int64_t sd_getlength(BlockDriverState
*bs
)
2157 BDRVSheepdogState
*s
= bs
->opaque
;
2159 return s
->inode
.vdi_size
;
2162 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
)
2164 Error
*local_err
= NULL
;
2165 BDRVSheepdogState
*s
= bs
->opaque
;
2167 unsigned int datalen
;
2168 uint64_t max_vdi_size
;
2170 max_vdi_size
= (UINT64_C(1) << s
->inode
.block_size_shift
) * MAX_DATA_OBJS
;
2171 if (offset
< s
->inode
.vdi_size
) {
2172 error_report("shrinking is not supported");
2174 } else if (offset
> max_vdi_size
) {
2175 error_report("too big image size");
2179 fd
= connect_to_sdog(s
, &local_err
);
2181 error_report_err(local_err
);
2185 /* we don't need to update entire object */
2186 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2187 s
->inode
.vdi_size
= offset
;
2188 ret
= write_object(fd
, s
->bs
, (char *)&s
->inode
,
2189 vid_to_vdi_oid(s
->inode
.vdi_id
), s
->inode
.nr_copies
,
2190 datalen
, 0, false, s
->cache_flags
);
2194 error_report("failed to update an inode.");
2201 * This function is called after writing data objects. If we need to
2202 * update metadata, this sends a write request to the vdi object.
2204 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
2206 BDRVSheepdogState
*s
= acb
->s
;
2209 uint32_t offset
, data_len
, mn
, mx
;
2211 mn
= acb
->min_dirty_data_idx
;
2212 mx
= acb
->max_dirty_data_idx
;
2214 /* we need to update the vdi object. */
2216 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
2217 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
2218 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
2220 acb
->min_dirty_data_idx
= UINT32_MAX
;
2221 acb
->max_dirty_data_idx
= 0;
2223 iov
.iov_base
= &s
->inode
;
2224 iov
.iov_len
= sizeof(s
->inode
);
2225 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2226 data_len
, offset
, 0, false, 0, offset
);
2227 add_aio_request(s
, aio_req
, &iov
, 1, AIOCB_WRITE_UDATA
);
2228 if (--acb
->nr_pending
) {
2229 qemu_coroutine_yield();
2234 /* Delete current working VDI on the snapshot chain */
2235 static bool sd_delete(BDRVSheepdogState
*s
)
2237 Error
*local_err
= NULL
;
2238 unsigned int wlen
= SD_MAX_VDI_LEN
, rlen
= 0;
2239 SheepdogVdiReq hdr
= {
2240 .opcode
= SD_OP_DEL_VDI
,
2241 .base_vdi_id
= s
->inode
.vdi_id
,
2242 .data_length
= wlen
,
2243 .flags
= SD_FLAG_CMD_WRITE
,
2245 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2248 fd
= connect_to_sdog(s
, &local_err
);
2250 error_report_err(local_err
);
2254 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2255 s
->name
, &wlen
, &rlen
);
2260 switch (rsp
->result
) {
2262 error_report("%s was already deleted", s
->name
);
2264 case SD_RES_SUCCESS
:
2267 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
2275 * Create a writable VDI from a snapshot
2277 static int sd_create_branch(BDRVSheepdogState
*s
)
2279 Error
*local_err
= NULL
;
2285 DPRINTF("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
2287 buf
= g_malloc(SD_INODE_SIZE
);
2290 * Even If deletion fails, we will just create extra snapshot based on
2291 * the working VDI which was supposed to be deleted. So no need to
2294 deleted
= sd_delete(s
);
2295 ret
= do_sd_create(s
, &vid
, !deleted
, &local_err
);
2297 error_report_err(local_err
);
2301 DPRINTF("%" PRIx32
" is created.\n", vid
);
2303 fd
= connect_to_sdog(s
, &local_err
);
2305 error_report_err(local_err
);
2310 ret
= read_object(fd
, s
->bs
, buf
, vid_to_vdi_oid(vid
),
2311 s
->inode
.nr_copies
, SD_INODE_SIZE
, 0, s
->cache_flags
);
2319 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
2321 s
->is_snapshot
= false;
2323 DPRINTF("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
2332 * Send I/O requests to the server.
2334 * This function sends requests to the server, links the requests to
2335 * the inflight_list in BDRVSheepdogState, and exits without
2336 * waiting the response. The responses are received in the
2337 * `aio_read_response' function which is called from the main loop as
2340 * Returns 1 when we need to wait a response, 0 when there is no sent
2341 * request and -errno in error cases.
2343 static void coroutine_fn
sd_co_rw_vector(SheepdogAIOCB
*acb
)
2346 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* BDRV_SECTOR_SIZE
;
2348 uint32_t object_size
;
2351 BDRVSheepdogState
*s
= acb
->s
;
2352 SheepdogInode
*inode
= &s
->inode
;
2355 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
2357 * In the case we open the snapshot VDI, Sheepdog creates the
2358 * writable VDI when we do a write operation first.
2360 ret
= sd_create_branch(s
);
2367 object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2368 idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ object_size
;
2369 offset
= (acb
->sector_num
* BDRV_SECTOR_SIZE
) % object_size
;
2372 * Make sure we don't free the aiocb before we are done with all requests.
2373 * This additional reference is dropped at the end of this function.
2377 while (done
!= total
) {
2379 uint64_t old_oid
= 0;
2380 bool create
= false;
2382 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
2384 len
= MIN(total
- done
, object_size
- offset
);
2386 switch (acb
->aiocb_type
) {
2387 case AIOCB_READ_UDATA
:
2388 if (!inode
->data_vdi_id
[idx
]) {
2389 qemu_iovec_memset(acb
->qiov
, done
, 0, len
);
2393 case AIOCB_WRITE_UDATA
:
2394 if (!inode
->data_vdi_id
[idx
]) {
2396 } else if (!is_data_obj_writable(inode
, idx
)) {
2400 flags
= SD_FLAG_CMD_COW
;
2403 case AIOCB_DISCARD_OBJ
:
2405 * We discard the object only when the whole object is
2406 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2408 if (len
!= object_size
|| inode
->data_vdi_id
[idx
] == 0) {
2417 DPRINTF("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
" %ld\n",
2419 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
2420 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
2421 DPRINTF("new oid %" PRIx64
"\n", oid
);
2424 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, create
,
2426 acb
->aiocb_type
== AIOCB_DISCARD_OBJ
?
2428 add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
2435 if (--acb
->nr_pending
) {
2436 qemu_coroutine_yield();
2440 static void sd_aio_complete(SheepdogAIOCB
*acb
)
2442 if (acb
->aiocb_type
== AIOCB_FLUSH_CACHE
) {
2446 QLIST_REMOVE(acb
, aiocb_siblings
);
2447 qemu_co_queue_restart_all(&acb
->s
->overlapping_queue
);
2450 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2451 int nb_sectors
, QEMUIOVector
*qiov
)
2455 int64_t offset
= (sector_num
+ nb_sectors
) * BDRV_SECTOR_SIZE
;
2456 BDRVSheepdogState
*s
= bs
->opaque
;
2458 if (offset
> s
->inode
.vdi_size
) {
2459 ret
= sd_truncate(bs
, offset
);
2465 sd_aio_setup(&acb
, s
, qiov
, sector_num
, nb_sectors
, AIOCB_WRITE_UDATA
);
2466 sd_co_rw_vector(&acb
);
2467 sd_write_done(&acb
);
2468 sd_aio_complete(&acb
);
2473 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2474 int nb_sectors
, QEMUIOVector
*qiov
)
2477 BDRVSheepdogState
*s
= bs
->opaque
;
2479 sd_aio_setup(&acb
, s
, qiov
, sector_num
, nb_sectors
, AIOCB_READ_UDATA
);
2480 sd_co_rw_vector(&acb
);
2481 sd_aio_complete(&acb
);
2486 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
2488 BDRVSheepdogState
*s
= bs
->opaque
;
2492 if (s
->cache_flags
!= SD_FLAG_CMD_CACHE
) {
2496 sd_aio_setup(&acb
, s
, NULL
, 0, 0, AIOCB_FLUSH_CACHE
);
2499 aio_req
= alloc_aio_req(s
, &acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2500 0, 0, 0, false, 0, 0);
2501 add_aio_request(s
, aio_req
, NULL
, 0, acb
.aiocb_type
);
2503 if (--acb
.nr_pending
) {
2504 qemu_coroutine_yield();
2507 sd_aio_complete(&acb
);
2511 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
2513 Error
*local_err
= NULL
;
2514 BDRVSheepdogState
*s
= bs
->opaque
;
2517 SheepdogInode
*inode
;
2518 unsigned int datalen
;
2520 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64
" "
2521 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
2522 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
2524 if (s
->is_snapshot
) {
2525 error_report("You can't create a snapshot of a snapshot VDI, "
2526 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
2531 DPRINTF("%s %s\n", sn_info
->name
, sn_info
->id_str
);
2533 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
2534 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
2535 /* It appears that inode.tag does not require a NUL terminator,
2536 * which means this use of strncpy is ok.
2538 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
2539 /* we don't need to update entire object */
2540 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2541 inode
= g_malloc(datalen
);
2543 /* refresh inode. */
2544 fd
= connect_to_sdog(s
, &local_err
);
2546 error_report_err(local_err
);
2551 ret
= write_object(fd
, s
->bs
, (char *)&s
->inode
,
2552 vid_to_vdi_oid(s
->inode
.vdi_id
), s
->inode
.nr_copies
,
2553 datalen
, 0, false, s
->cache_flags
);
2555 error_report("failed to write snapshot's inode.");
2559 ret
= do_sd_create(s
, &new_vid
, 1, &local_err
);
2561 error_reportf_err(local_err
,
2562 "failed to create inode for snapshot: ");
2566 ret
= read_object(fd
, s
->bs
, (char *)inode
,
2567 vid_to_vdi_oid(new_vid
), s
->inode
.nr_copies
, datalen
, 0,
2571 error_report("failed to read new inode info. %s", strerror(errno
));
2575 memcpy(&s
->inode
, inode
, datalen
);
2576 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2577 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
2586 * We implement rollback(loadvm) operation to the specified snapshot by
2587 * 1) switch to the snapshot
2588 * 2) rely on sd_create_branch to delete working VDI and
2589 * 3) create a new working VDI based on the specified snapshot
2591 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
2593 BDRVSheepdogState
*s
= bs
->opaque
;
2594 BDRVSheepdogState
*old_s
;
2595 char tag
[SD_MAX_VDI_TAG_LEN
];
2596 uint32_t snapid
= 0;
2599 if (!sd_parse_snapid_or_tag(snapshot_id
, &snapid
, tag
)) {
2603 old_s
= g_new(BDRVSheepdogState
, 1);
2605 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
2607 ret
= reload_inode(s
, snapid
, tag
);
2612 ret
= sd_create_branch(s
);
2621 /* recover bdrv_sd_state */
2622 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
2625 error_report("failed to open. recover old bdrv_sd_state.");
2630 #define NR_BATCHED_DISCARD 128
2632 static int remove_objects(BDRVSheepdogState
*s
, Error
**errp
)
2634 int fd
, i
= 0, nr_objs
= 0;
2636 SheepdogInode
*inode
= &s
->inode
;
2638 fd
= connect_to_sdog(s
, errp
);
2643 nr_objs
= count_data_objs(inode
);
2644 while (i
< nr_objs
) {
2645 int start_idx
, nr_filled_idx
;
2647 while (i
< nr_objs
&& !inode
->data_vdi_id
[i
]) {
2653 while (i
< nr_objs
&& nr_filled_idx
< NR_BATCHED_DISCARD
) {
2654 if (inode
->data_vdi_id
[i
]) {
2655 inode
->data_vdi_id
[i
] = 0;
2662 ret
= write_object(fd
, s
->bs
,
2663 (char *)&inode
->data_vdi_id
[start_idx
],
2664 vid_to_vdi_oid(s
->inode
.vdi_id
), inode
->nr_copies
,
2665 (i
- start_idx
) * sizeof(uint32_t),
2666 offsetof(struct SheepdogInode
,
2667 data_vdi_id
[start_idx
]),
2668 false, s
->cache_flags
);
2670 error_setg(errp
, "Failed to discard snapshot inode");
2681 static int sd_snapshot_delete(BlockDriverState
*bs
,
2682 const char *snapshot_id
,
2687 * FIXME should delete the snapshot matching both @snapshot_id and
2688 * @name, but @name not used here
2690 unsigned long snap_id
= 0;
2691 char snap_tag
[SD_MAX_VDI_TAG_LEN
];
2693 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
2694 BDRVSheepdogState
*s
= bs
->opaque
;
2695 unsigned int wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
, rlen
= 0;
2697 SheepdogVdiReq hdr
= {
2698 .opcode
= SD_OP_DEL_VDI
,
2699 .data_length
= wlen
,
2700 .flags
= SD_FLAG_CMD_WRITE
,
2702 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2704 ret
= remove_objects(s
, errp
);
2709 memset(buf
, 0, sizeof(buf
));
2710 memset(snap_tag
, 0, sizeof(snap_tag
));
2711 pstrcpy(buf
, SD_MAX_VDI_LEN
, s
->name
);
2712 /* TODO Use sd_parse_snapid() once this mess is cleaned up */
2713 ret
= qemu_strtoul(snapshot_id
, NULL
, 10, &snap_id
);
2714 if (ret
|| snap_id
> UINT32_MAX
) {
2716 * FIXME Since qemu_strtoul() returns -EINVAL when
2717 * @snapshot_id is null, @snapshot_id is mandatory. Correct
2718 * would be to require at least one of @snapshot_id and @name.
2720 error_setg(errp
, "Invalid snapshot ID: %s",
2721 snapshot_id
? snapshot_id
: "<null>");
2726 hdr
.snapid
= (uint32_t) snap_id
;
2728 /* FIXME I suspect we should use @name here */
2729 /* FIXME don't truncate silently */
2730 pstrcpy(snap_tag
, sizeof(snap_tag
), snapshot_id
);
2731 pstrcpy(buf
+ SD_MAX_VDI_LEN
, SD_MAX_VDI_TAG_LEN
, snap_tag
);
2734 ret
= find_vdi_name(s
, s
->name
, snap_id
, snap_tag
, &vid
, true, errp
);
2739 fd
= connect_to_sdog(s
, errp
);
2744 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2748 error_setg_errno(errp
, -ret
, "Couldn't send request to server");
2752 switch (rsp
->result
) {
2754 error_setg(errp
, "Can't find the snapshot");
2756 case SD_RES_SUCCESS
:
2759 error_setg(errp
, "%s", sd_strerror(rsp
->result
));
2766 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
2768 Error
*local_err
= NULL
;
2769 BDRVSheepdogState
*s
= bs
->opaque
;
2771 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
2772 QEMUSnapshotInfo
*sn_tab
= NULL
;
2773 unsigned wlen
, rlen
;
2775 static SheepdogInode inode
;
2776 unsigned long *vdi_inuse
;
2777 unsigned int start_nr
;
2781 vdi_inuse
= g_malloc(max
);
2783 fd
= connect_to_sdog(s
, &local_err
);
2785 error_report_err(local_err
);
2793 memset(&req
, 0, sizeof(req
));
2795 req
.opcode
= SD_OP_READ_VDIS
;
2796 req
.data_length
= max
;
2798 ret
= do_req(fd
, s
->bs
, &req
, vdi_inuse
, &wlen
, &rlen
);
2805 sn_tab
= g_new0(QEMUSnapshotInfo
, nr
);
2807 /* calculate a vdi id with hash function */
2808 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
2809 start_nr
= hval
& (SD_NR_VDIS
- 1);
2811 fd
= connect_to_sdog(s
, &local_err
);
2813 error_report_err(local_err
);
2818 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
2819 if (!test_bit(vid
, vdi_inuse
)) {
2823 /* we don't need to read entire object */
2824 ret
= read_object(fd
, s
->bs
, (char *)&inode
,
2825 vid_to_vdi_oid(vid
),
2826 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
2833 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
2834 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
2835 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
2836 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
2837 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
2839 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
),
2840 "%" PRIu32
, inode
.snap_id
);
2841 pstrcpy(sn_tab
[found
].name
,
2842 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)),
2861 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
2862 int64_t pos
, int size
, int load
)
2864 Error
*local_err
= NULL
;
2866 int fd
, ret
= 0, remaining
= size
;
2867 unsigned int data_len
;
2868 uint64_t vmstate_oid
;
2871 uint32_t vdi_id
= load
? s
->inode
.parent_vdi_id
: s
->inode
.vdi_id
;
2872 uint32_t object_size
= (UINT32_C(1) << s
->inode
.block_size_shift
);
2874 fd
= connect_to_sdog(s
, &local_err
);
2876 error_report_err(local_err
);
2881 vdi_index
= pos
/ object_size
;
2882 offset
= pos
% object_size
;
2884 data_len
= MIN(remaining
, object_size
- offset
);
2886 vmstate_oid
= vid_to_vmstate_oid(vdi_id
, vdi_index
);
2888 create
= (offset
== 0);
2890 ret
= read_object(fd
, s
->bs
, (char *)data
, vmstate_oid
,
2891 s
->inode
.nr_copies
, data_len
, offset
,
2894 ret
= write_object(fd
, s
->bs
, (char *)data
, vmstate_oid
,
2895 s
->inode
.nr_copies
, data_len
, offset
, create
,
2900 error_report("failed to save vmstate %s", strerror(errno
));
2906 remaining
-= data_len
;
2914 static int sd_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2917 BDRVSheepdogState
*s
= bs
->opaque
;
2921 buf
= qemu_blockalign(bs
, qiov
->size
);
2922 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
2923 ret
= do_load_save_vmstate(s
, (uint8_t *) buf
, pos
, qiov
->size
, 0);
2929 static int sd_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2932 BDRVSheepdogState
*s
= bs
->opaque
;
2936 buf
= qemu_blockalign(bs
, qiov
->size
);
2937 ret
= do_load_save_vmstate(s
, buf
, pos
, qiov
->size
, 1);
2938 qemu_iovec_from_buf(qiov
, 0, buf
, qiov
->size
);
2945 static coroutine_fn
int sd_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
2949 BDRVSheepdogState
*s
= bs
->opaque
;
2950 QEMUIOVector discard_iov
;
2954 if (!s
->discard_supported
) {
2958 memset(&discard_iov
, 0, sizeof(discard_iov
));
2959 memset(&iov
, 0, sizeof(iov
));
2960 iov
.iov_base
= &zero
;
2961 iov
.iov_len
= sizeof(zero
);
2962 discard_iov
.iov
= &iov
;
2963 discard_iov
.niov
= 1;
2964 if (!QEMU_IS_ALIGNED(offset
| count
, BDRV_SECTOR_SIZE
)) {
2967 sd_aio_setup(&acb
, s
, &discard_iov
, offset
>> BDRV_SECTOR_BITS
,
2968 count
>> BDRV_SECTOR_BITS
, AIOCB_DISCARD_OBJ
);
2969 sd_co_rw_vector(&acb
);
2970 sd_aio_complete(&acb
);
2975 static coroutine_fn
int64_t
2976 sd_co_get_block_status(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2977 int *pnum
, BlockDriverState
**file
)
2979 BDRVSheepdogState
*s
= bs
->opaque
;
2980 SheepdogInode
*inode
= &s
->inode
;
2981 uint32_t object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2982 uint64_t offset
= sector_num
* BDRV_SECTOR_SIZE
;
2983 unsigned long start
= offset
/ object_size
,
2984 end
= DIV_ROUND_UP((sector_num
+ nb_sectors
) *
2985 BDRV_SECTOR_SIZE
, object_size
);
2987 int64_t ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| offset
;
2989 for (idx
= start
; idx
< end
; idx
++) {
2990 if (inode
->data_vdi_id
[idx
] == 0) {
2995 /* Get the longest length of unallocated sectors */
2997 for (idx
= start
+ 1; idx
< end
; idx
++) {
2998 if (inode
->data_vdi_id
[idx
] != 0) {
3004 *pnum
= (idx
- start
) * object_size
/ BDRV_SECTOR_SIZE
;
3005 if (*pnum
> nb_sectors
) {
3008 if (ret
> 0 && ret
& BDRV_BLOCK_OFFSET_VALID
) {
3014 static int64_t sd_get_allocated_file_size(BlockDriverState
*bs
)
3016 BDRVSheepdogState
*s
= bs
->opaque
;
3017 SheepdogInode
*inode
= &s
->inode
;
3018 uint32_t object_size
= (UINT32_C(1) << inode
->block_size_shift
);
3019 unsigned long i
, last
= DIV_ROUND_UP(inode
->vdi_size
, object_size
);
3022 for (i
= 0; i
< last
; i
++) {
3023 if (inode
->data_vdi_id
[i
] == 0) {
3026 size
+= object_size
;
3031 static QemuOptsList sd_create_opts
= {
3032 .name
= "sheepdog-create-opts",
3033 .head
= QTAILQ_HEAD_INITIALIZER(sd_create_opts
.head
),
3036 .name
= BLOCK_OPT_SIZE
,
3037 .type
= QEMU_OPT_SIZE
,
3038 .help
= "Virtual disk size"
3041 .name
= BLOCK_OPT_BACKING_FILE
,
3042 .type
= QEMU_OPT_STRING
,
3043 .help
= "File name of a base image"
3046 .name
= BLOCK_OPT_PREALLOC
,
3047 .type
= QEMU_OPT_STRING
,
3048 .help
= "Preallocation mode (allowed values: off, full)"
3051 .name
= BLOCK_OPT_REDUNDANCY
,
3052 .type
= QEMU_OPT_STRING
,
3053 .help
= "Redundancy of the image"
3056 .name
= BLOCK_OPT_OBJECT_SIZE
,
3057 .type
= QEMU_OPT_SIZE
,
3058 .help
= "Object size of the image"
3060 { /* end of list */ }
3064 static BlockDriver bdrv_sheepdog
= {
3065 .format_name
= "sheepdog",
3066 .protocol_name
= "sheepdog",
3067 .instance_size
= sizeof(BDRVSheepdogState
),
3068 .bdrv_parse_filename
= sd_parse_filename
,
3069 .bdrv_file_open
= sd_open
,
3070 .bdrv_reopen_prepare
= sd_reopen_prepare
,
3071 .bdrv_reopen_commit
= sd_reopen_commit
,
3072 .bdrv_reopen_abort
= sd_reopen_abort
,
3073 .bdrv_close
= sd_close
,
3074 .bdrv_create
= sd_create
,
3075 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3076 .bdrv_getlength
= sd_getlength
,
3077 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
3078 .bdrv_truncate
= sd_truncate
,
3080 .bdrv_co_readv
= sd_co_readv
,
3081 .bdrv_co_writev
= sd_co_writev
,
3082 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
3083 .bdrv_co_pdiscard
= sd_co_pdiscard
,
3084 .bdrv_co_get_block_status
= sd_co_get_block_status
,
3086 .bdrv_snapshot_create
= sd_snapshot_create
,
3087 .bdrv_snapshot_goto
= sd_snapshot_goto
,
3088 .bdrv_snapshot_delete
= sd_snapshot_delete
,
3089 .bdrv_snapshot_list
= sd_snapshot_list
,
3091 .bdrv_save_vmstate
= sd_save_vmstate
,
3092 .bdrv_load_vmstate
= sd_load_vmstate
,
3094 .bdrv_detach_aio_context
= sd_detach_aio_context
,
3095 .bdrv_attach_aio_context
= sd_attach_aio_context
,
3097 .create_opts
= &sd_create_opts
,
3100 static BlockDriver bdrv_sheepdog_tcp
= {
3101 .format_name
= "sheepdog",
3102 .protocol_name
= "sheepdog+tcp",
3103 .instance_size
= sizeof(BDRVSheepdogState
),
3104 .bdrv_parse_filename
= sd_parse_filename
,
3105 .bdrv_file_open
= sd_open
,
3106 .bdrv_reopen_prepare
= sd_reopen_prepare
,
3107 .bdrv_reopen_commit
= sd_reopen_commit
,
3108 .bdrv_reopen_abort
= sd_reopen_abort
,
3109 .bdrv_close
= sd_close
,
3110 .bdrv_create
= sd_create
,
3111 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3112 .bdrv_getlength
= sd_getlength
,
3113 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
3114 .bdrv_truncate
= sd_truncate
,
3116 .bdrv_co_readv
= sd_co_readv
,
3117 .bdrv_co_writev
= sd_co_writev
,
3118 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
3119 .bdrv_co_pdiscard
= sd_co_pdiscard
,
3120 .bdrv_co_get_block_status
= sd_co_get_block_status
,
3122 .bdrv_snapshot_create
= sd_snapshot_create
,
3123 .bdrv_snapshot_goto
= sd_snapshot_goto
,
3124 .bdrv_snapshot_delete
= sd_snapshot_delete
,
3125 .bdrv_snapshot_list
= sd_snapshot_list
,
3127 .bdrv_save_vmstate
= sd_save_vmstate
,
3128 .bdrv_load_vmstate
= sd_load_vmstate
,
3130 .bdrv_detach_aio_context
= sd_detach_aio_context
,
3131 .bdrv_attach_aio_context
= sd_attach_aio_context
,
3133 .create_opts
= &sd_create_opts
,
3136 static BlockDriver bdrv_sheepdog_unix
= {
3137 .format_name
= "sheepdog",
3138 .protocol_name
= "sheepdog+unix",
3139 .instance_size
= sizeof(BDRVSheepdogState
),
3140 .bdrv_parse_filename
= sd_parse_filename
,
3141 .bdrv_file_open
= sd_open
,
3142 .bdrv_reopen_prepare
= sd_reopen_prepare
,
3143 .bdrv_reopen_commit
= sd_reopen_commit
,
3144 .bdrv_reopen_abort
= sd_reopen_abort
,
3145 .bdrv_close
= sd_close
,
3146 .bdrv_create
= sd_create
,
3147 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3148 .bdrv_getlength
= sd_getlength
,
3149 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
3150 .bdrv_truncate
= sd_truncate
,
3152 .bdrv_co_readv
= sd_co_readv
,
3153 .bdrv_co_writev
= sd_co_writev
,
3154 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
3155 .bdrv_co_pdiscard
= sd_co_pdiscard
,
3156 .bdrv_co_get_block_status
= sd_co_get_block_status
,
3158 .bdrv_snapshot_create
= sd_snapshot_create
,
3159 .bdrv_snapshot_goto
= sd_snapshot_goto
,
3160 .bdrv_snapshot_delete
= sd_snapshot_delete
,
3161 .bdrv_snapshot_list
= sd_snapshot_list
,
3163 .bdrv_save_vmstate
= sd_save_vmstate
,
3164 .bdrv_load_vmstate
= sd_load_vmstate
,
3166 .bdrv_detach_aio_context
= sd_detach_aio_context
,
3167 .bdrv_attach_aio_context
= sd_attach_aio_context
,
3169 .create_opts
= &sd_create_opts
,
3172 static void bdrv_sheepdog_init(void)
3174 bdrv_register(&bdrv_sheepdog
);
3175 bdrv_register(&bdrv_sheepdog_tcp
);
3176 bdrv_register(&bdrv_sheepdog_unix
);
3178 block_init(bdrv_sheepdog_init
);