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/qobject-input-visitor.h"
21 #include "qemu/error-report.h"
22 #include "qemu/sockets.h"
23 #include "block/block_int.h"
24 #include "sysemu/block-backend.h"
25 #include "qemu/bitops.h"
26 #include "qemu/cutils.h"
28 #define SD_PROTO_VER 0x01
30 #define SD_DEFAULT_ADDR "localhost"
31 #define SD_DEFAULT_PORT 7000
33 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
34 #define SD_OP_READ_OBJ 0x02
35 #define SD_OP_WRITE_OBJ 0x03
36 /* 0x04 is used internally by Sheepdog */
38 #define SD_OP_NEW_VDI 0x11
39 #define SD_OP_LOCK_VDI 0x12
40 #define SD_OP_RELEASE_VDI 0x13
41 #define SD_OP_GET_VDI_INFO 0x14
42 #define SD_OP_READ_VDIS 0x15
43 #define SD_OP_FLUSH_VDI 0x16
44 #define SD_OP_DEL_VDI 0x17
45 #define SD_OP_GET_CLUSTER_DEFAULT 0x18
47 #define SD_FLAG_CMD_WRITE 0x01
48 #define SD_FLAG_CMD_COW 0x02
49 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
50 #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
52 #define SD_RES_SUCCESS 0x00 /* Success */
53 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
54 #define SD_RES_NO_OBJ 0x02 /* No object found */
55 #define SD_RES_EIO 0x03 /* I/O error */
56 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
57 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
58 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
59 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
60 #define SD_RES_NO_VDI 0x08 /* No vdi found */
61 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
62 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
63 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
64 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
65 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
66 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
67 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
68 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
69 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
70 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
71 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
72 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
73 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
74 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
75 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
76 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
77 #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */
78 #define SD_RES_READONLY 0x1A /* Object is read-only */
83 * 0 - 19 (20 bits): data object space
84 * 20 - 31 (12 bits): reserved data object space
85 * 32 - 55 (24 bits): vdi object space
86 * 56 - 59 ( 4 bits): reserved vdi object space
87 * 60 - 63 ( 4 bits): object type identifier space
90 #define VDI_SPACE_SHIFT 32
91 #define VDI_BIT (UINT64_C(1) << 63)
92 #define VMSTATE_BIT (UINT64_C(1) << 62)
93 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
94 #define MAX_CHILDREN 1024
95 #define SD_MAX_VDI_LEN 256
96 #define SD_MAX_VDI_TAG_LEN 256
97 #define SD_NR_VDIS (1U << 24)
98 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
99 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
100 #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22
102 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
103 * (SD_EC_MAX_STRIP - 1) for parity strips
105 * SD_MAX_COPIES is sum of number of data strips and parity strips.
107 #define SD_EC_MAX_STRIP 16
108 #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
110 #define SD_INODE_SIZE (sizeof(SheepdogInode))
111 #define CURRENT_VDI_ID 0
113 #define LOCK_TYPE_NORMAL 0
114 #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */
116 typedef struct SheepdogReq
{
122 uint32_t data_length
;
123 uint32_t opcode_specific
[8];
126 typedef struct SheepdogRsp
{
132 uint32_t data_length
;
134 uint32_t opcode_specific
[7];
137 typedef struct SheepdogObjReq
{
143 uint32_t data_length
;
152 typedef struct SheepdogObjRsp
{
158 uint32_t data_length
;
166 typedef struct SheepdogVdiReq
{
172 uint32_t data_length
;
174 uint32_t base_vdi_id
;
177 uint8_t store_policy
;
178 uint8_t block_size_shift
;
184 typedef struct SheepdogVdiRsp
{
190 uint32_t data_length
;
197 typedef struct SheepdogClusterRsp
{
203 uint32_t data_length
;
207 uint8_t block_size_shift
;
210 } SheepdogClusterRsp
;
212 typedef struct SheepdogInode
{
213 char name
[SD_MAX_VDI_LEN
];
214 char tag
[SD_MAX_VDI_TAG_LEN
];
217 uint64_t vm_clock_nsec
;
219 uint64_t vm_state_size
;
220 uint16_t copy_policy
;
222 uint8_t block_size_shift
;
225 uint32_t parent_vdi_id
;
226 uint32_t child_vdi_id
[MAX_CHILDREN
];
227 uint32_t data_vdi_id
[MAX_DATA_OBJS
];
230 #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
233 * 64 bit FNV-1a non-zero initial basis
235 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
238 * 64 bit Fowler/Noll/Vo FNV-1a hash code
240 static inline uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
242 unsigned char *bp
= buf
;
243 unsigned char *be
= bp
+ len
;
245 hval
^= (uint64_t) *bp
++;
246 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
247 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
252 static inline bool is_data_obj_writable(SheepdogInode
*inode
, unsigned int idx
)
254 return inode
->vdi_id
== inode
->data_vdi_id
[idx
];
257 static inline bool is_data_obj(uint64_t oid
)
259 return !(VDI_BIT
& oid
);
262 static inline uint64_t data_oid_to_idx(uint64_t oid
)
264 return oid
& (MAX_DATA_OBJS
- 1);
267 static inline uint32_t oid_to_vid(uint64_t oid
)
269 return (oid
& ~VDI_BIT
) >> VDI_SPACE_SHIFT
;
272 static inline uint64_t vid_to_vdi_oid(uint32_t vid
)
274 return VDI_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
);
277 static inline uint64_t vid_to_vmstate_oid(uint32_t vid
, uint32_t idx
)
279 return VMSTATE_BIT
| ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
282 static inline uint64_t vid_to_data_oid(uint32_t vid
, uint32_t idx
)
284 return ((uint64_t)vid
<< VDI_SPACE_SHIFT
) | idx
;
287 static inline bool is_snapshot(struct SheepdogInode
*inode
)
289 return !!inode
->snap_ctime
;
292 static inline size_t count_data_objs(const struct SheepdogInode
*inode
)
294 return DIV_ROUND_UP(inode
->vdi_size
,
295 (1UL << inode
->block_size_shift
));
300 #define DEBUG_SDOG_PRINT 1
302 #define DEBUG_SDOG_PRINT 0
304 #define DPRINTF(fmt, args...) \
306 if (DEBUG_SDOG_PRINT) { \
307 fprintf(stderr, "%s %d: " fmt, __func__, __LINE__, ##args); \
311 typedef struct SheepdogAIOCB SheepdogAIOCB
;
312 typedef struct BDRVSheepdogState BDRVSheepdogState
;
314 typedef struct AIOReq
{
315 SheepdogAIOCB
*aiocb
;
316 unsigned int iov_offset
;
321 unsigned int data_len
;
326 QLIST_ENTRY(AIOReq
) aio_siblings
;
336 #define AIOCBOverlapping(x, y) \
337 (!(x->max_affect_data_idx < y->min_affect_data_idx \
338 || y->max_affect_data_idx < x->min_affect_data_idx))
340 struct SheepdogAIOCB
{
341 BDRVSheepdogState
*s
;
349 enum AIOCBState aiocb_type
;
351 Coroutine
*coroutine
;
354 uint32_t min_affect_data_idx
;
355 uint32_t max_affect_data_idx
;
358 * The difference between affect_data_idx and dirty_data_idx:
359 * affect_data_idx represents range of index of all request types.
360 * dirty_data_idx represents range of index updated by COW requests.
361 * dirty_data_idx is used for updating an inode object.
363 uint32_t min_dirty_data_idx
;
364 uint32_t max_dirty_data_idx
;
366 QLIST_ENTRY(SheepdogAIOCB
) aiocb_siblings
;
369 struct BDRVSheepdogState
{
370 BlockDriverState
*bs
;
371 AioContext
*aio_context
;
375 char name
[SD_MAX_VDI_LEN
];
377 uint32_t cache_flags
;
378 bool discard_supported
;
387 uint32_t aioreq_seq_num
;
389 /* Every aio request must be linked to either of these queues. */
390 QLIST_HEAD(inflight_aio_head
, AIOReq
) inflight_aio_head
;
391 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
, &s
->queue_lock
);
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 qemu_co_mutex_lock(&s
->queue_lock
);
530 wait_for_overlapping_aiocb(s
, acb
);
531 QLIST_INSERT_HEAD(&s
->inflight_aiocb_head
, acb
, aiocb_siblings
);
532 qemu_co_mutex_unlock(&s
->queue_lock
);
535 static SocketAddress
*sd_socket_address(const char *path
,
536 const char *host
, const char *port
)
538 SocketAddress
*addr
= g_new0(SocketAddress
, 1);
541 addr
->type
= SOCKET_ADDRESS_TYPE_UNIX
;
542 addr
->u
.q_unix
.path
= g_strdup(path
);
544 addr
->type
= SOCKET_ADDRESS_TYPE_INET
;
545 addr
->u
.inet
.host
= g_strdup(host
?: SD_DEFAULT_ADDR
);
546 addr
->u
.inet
.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 SocketAddress
*saddr
= NULL
;
558 Error
*local_err
= NULL
;
560 qdict_extract_subqdict(options
, &server
, "server.");
562 crumpled_server
= qdict_crumple(server
, errp
);
563 if (!crumpled_server
) {
568 * FIXME .numeric, .to, .ipv4 or .ipv6 don't work with -drive
569 * server.type=inet. .to doesn't matter, it's ignored anyway.
570 * That's because when @options come from -blockdev or
571 * blockdev_add, members are typed according to the QAPI schema,
572 * but when they come from -drive, they're all QString. The
573 * visitor expects the former.
575 iv
= qobject_input_visitor_new(crumpled_server
);
576 visit_type_SocketAddress(iv
, NULL
, &saddr
, &local_err
);
578 error_propagate(errp
, local_err
);
584 qobject_decref(crumpled_server
);
589 /* Return -EIO in case of error, file descriptor on success */
590 static int connect_to_sdog(BDRVSheepdogState
*s
, Error
**errp
)
594 fd
= socket_connect(s
->addr
, errp
);
596 if (s
->addr
->type
== SOCKET_ADDRESS_TYPE_INET
&& fd
>= 0) {
597 int ret
= socket_set_nodelay(fd
);
599 error_report("%s", strerror(errno
));
604 qemu_set_nonblock(fd
);
612 /* Return 0 on success and -errno in case of error */
613 static coroutine_fn
int send_co_req(int sockfd
, SheepdogReq
*hdr
, void *data
,
618 ret
= qemu_co_send(sockfd
, hdr
, sizeof(*hdr
));
619 if (ret
!= sizeof(*hdr
)) {
620 error_report("failed to send a req, %s", strerror(errno
));
624 ret
= qemu_co_send(sockfd
, data
, *wlen
);
626 error_report("failed to send a req, %s", strerror(errno
));
633 typedef struct SheepdogReqCo
{
635 BlockDriverState
*bs
;
636 AioContext
*aio_context
;
646 static void restart_co_req(void *opaque
)
648 SheepdogReqCo
*srco
= opaque
;
650 aio_co_wake(srco
->co
);
653 static coroutine_fn
void do_co_req(void *opaque
)
656 SheepdogReqCo
*srco
= opaque
;
657 int sockfd
= srco
->sockfd
;
658 SheepdogReq
*hdr
= srco
->hdr
;
659 void *data
= srco
->data
;
660 unsigned int *wlen
= srco
->wlen
;
661 unsigned int *rlen
= srco
->rlen
;
663 srco
->co
= qemu_coroutine_self();
664 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
665 NULL
, restart_co_req
, NULL
, srco
);
667 ret
= send_co_req(sockfd
, hdr
, data
, wlen
);
672 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
673 restart_co_req
, NULL
, NULL
, srco
);
675 ret
= qemu_co_recv(sockfd
, hdr
, sizeof(*hdr
));
676 if (ret
!= sizeof(*hdr
)) {
677 error_report("failed to get a rsp, %s", strerror(errno
));
682 if (*rlen
> hdr
->data_length
) {
683 *rlen
= hdr
->data_length
;
687 ret
= qemu_co_recv(sockfd
, data
, *rlen
);
689 error_report("failed to get the data, %s", strerror(errno
));
696 /* there is at most one request for this sockfd, so it is safe to
697 * set each handler to NULL. */
698 aio_set_fd_handler(srco
->aio_context
, sockfd
, false,
699 NULL
, NULL
, NULL
, NULL
);
703 /* Set srco->finished before reading bs->wakeup. */
704 atomic_mb_set(&srco
->finished
, true);
706 bdrv_wakeup(srco
->bs
);
711 * Send the request to the sheep in a synchronous manner.
713 * Return 0 on success, -errno in case of error.
715 static int do_req(int sockfd
, BlockDriverState
*bs
, SheepdogReq
*hdr
,
716 void *data
, unsigned int *wlen
, unsigned int *rlen
)
719 SheepdogReqCo srco
= {
721 .aio_context
= bs
? bdrv_get_aio_context(bs
) : qemu_get_aio_context(),
731 if (qemu_in_coroutine()) {
734 co
= qemu_coroutine_create(do_co_req
, &srco
);
736 bdrv_coroutine_enter(bs
, co
);
737 BDRV_POLL_WHILE(bs
, !srco
.finished
);
739 qemu_coroutine_enter(co
);
740 while (!srco
.finished
) {
741 aio_poll(qemu_get_aio_context(), true);
749 static void coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
750 struct iovec
*iov
, int niov
,
751 enum AIOCBState aiocb_type
);
752 static void coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
);
753 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
);
754 static int get_sheep_fd(BDRVSheepdogState
*s
, Error
**errp
);
755 static void co_write_request(void *opaque
);
757 static coroutine_fn
void reconnect_to_sdog(void *opaque
)
759 BDRVSheepdogState
*s
= opaque
;
760 AIOReq
*aio_req
, *next
;
762 aio_set_fd_handler(s
->aio_context
, s
->fd
, false, NULL
,
767 /* Wait for outstanding write requests to be completed. */
768 while (s
->co_send
!= NULL
) {
769 co_write_request(opaque
);
772 /* Try to reconnect the sheepdog server every one second. */
774 Error
*local_err
= NULL
;
775 s
->fd
= get_sheep_fd(s
, &local_err
);
777 DPRINTF("Wait for connection to be established\n");
778 error_report_err(local_err
);
779 qemu_co_sleep_ns(QEMU_CLOCK_REALTIME
, 1000000000ULL);
784 * Now we have to resend all the request in the inflight queue. However,
785 * resend_aioreq() can yield and newly created requests can be added to the
786 * inflight queue before the coroutine is resumed. To avoid mixing them, we
787 * have to move all the inflight requests to the failed queue before
788 * resend_aioreq() is called.
790 qemu_co_mutex_lock(&s
->queue_lock
);
791 QLIST_FOREACH_SAFE(aio_req
, &s
->inflight_aio_head
, aio_siblings
, next
) {
792 QLIST_REMOVE(aio_req
, aio_siblings
);
793 QLIST_INSERT_HEAD(&s
->failed_aio_head
, aio_req
, aio_siblings
);
796 /* Resend all the failed aio requests. */
797 while (!QLIST_EMPTY(&s
->failed_aio_head
)) {
798 aio_req
= QLIST_FIRST(&s
->failed_aio_head
);
799 QLIST_REMOVE(aio_req
, aio_siblings
);
800 qemu_co_mutex_unlock(&s
->queue_lock
);
801 resend_aioreq(s
, aio_req
);
802 qemu_co_mutex_lock(&s
->queue_lock
);
804 qemu_co_mutex_unlock(&s
->queue_lock
);
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 qemu_co_mutex_lock(&s
->queue_lock
);
897 QLIST_REMOVE(aio_req
, aio_siblings
);
898 qemu_co_mutex_unlock(&s
->queue_lock
);
900 switch (rsp
.result
) {
903 case SD_RES_READONLY
:
904 if (s
->inode
.vdi_id
== oid_to_vid(aio_req
->oid
)) {
905 ret
= reload_inode(s
, 0, "");
910 if (is_data_obj(aio_req
->oid
)) {
911 aio_req
->oid
= vid_to_data_oid(s
->inode
.vdi_id
,
912 data_oid_to_idx(aio_req
->oid
));
914 aio_req
->oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
916 resend_aioreq(s
, aio_req
);
920 error_report("%s", sd_strerror(rsp
.result
));
926 if (!--acb
->nr_pending
) {
928 * We've finished all requests which belong to the AIOCB, so
929 * we can switch back to sd_co_readv/writev now.
931 aio_co_wake(acb
->coroutine
);
937 reconnect_to_sdog(opaque
);
940 static void co_read_response(void *opaque
)
942 BDRVSheepdogState
*s
= opaque
;
945 s
->co_recv
= qemu_coroutine_create(aio_read_response
, opaque
);
948 aio_co_enter(s
->aio_context
, s
->co_recv
);
951 static void co_write_request(void *opaque
)
953 BDRVSheepdogState
*s
= opaque
;
955 aio_co_wake(s
->co_send
);
959 * Return a socket descriptor to read/write objects.
961 * We cannot use this descriptor for other operations because
962 * the block driver may be on waiting response from the server.
964 static int get_sheep_fd(BDRVSheepdogState
*s
, Error
**errp
)
968 fd
= connect_to_sdog(s
, errp
);
973 aio_set_fd_handler(s
->aio_context
, fd
, false,
974 co_read_response
, NULL
, NULL
, s
);
979 * Parse numeric snapshot ID in @str
980 * If @str can't be parsed as number, return false.
981 * Else, if the number is zero or too large, set *@snapid to zero and
983 * Else, set *@snapid to the number and return true.
985 static bool sd_parse_snapid(const char *str
, uint32_t *snapid
)
990 ret
= qemu_strtoul(str
, NULL
, 10, &ul
);
991 if (ret
== -ERANGE
) {
997 if (ul
> UINT32_MAX
) {
1005 static bool sd_parse_snapid_or_tag(const char *str
,
1006 uint32_t *snapid
, char tag
[])
1008 if (!sd_parse_snapid(str
, snapid
)) {
1010 if (g_strlcpy(tag
, str
, SD_MAX_VDI_TAG_LEN
) >= SD_MAX_VDI_TAG_LEN
) {
1013 } else if (!*snapid
) {
1022 const char *path
; /* non-null iff transport is tcp */
1023 const char *host
; /* valid when transport is tcp */
1024 int port
; /* valid when transport is tcp */
1025 char vdi
[SD_MAX_VDI_LEN
];
1026 char tag
[SD_MAX_VDI_TAG_LEN
];
1028 /* Remainder is only for sd_config_done() */
1033 static void sd_config_done(SheepdogConfig
*cfg
)
1036 query_params_free(cfg
->qp
);
1041 static void sd_parse_uri(SheepdogConfig
*cfg
, const char *filename
,
1045 QueryParams
*qp
= NULL
;
1049 memset(cfg
, 0, sizeof(*cfg
));
1051 cfg
->uri
= uri
= uri_parse(filename
);
1053 error_setg(&err
, "invalid URI");
1058 if (!g_strcmp0(uri
->scheme
, "sheepdog")) {
1060 } else if (!g_strcmp0(uri
->scheme
, "sheepdog+tcp")) {
1062 } else if (!g_strcmp0(uri
->scheme
, "sheepdog+unix")) {
1065 error_setg(&err
, "URI scheme must be 'sheepdog', 'sheepdog+tcp',"
1066 " or 'sheepdog+unix'");
1070 if (uri
->path
== NULL
|| !strcmp(uri
->path
, "/")) {
1071 error_setg(&err
, "missing file path in URI");
1074 if (g_strlcpy(cfg
->vdi
, uri
->path
+ 1, SD_MAX_VDI_LEN
)
1075 >= SD_MAX_VDI_LEN
) {
1076 error_setg(&err
, "VDI name is too long");
1080 cfg
->qp
= qp
= query_params_parse(uri
->query
);
1083 /* sheepdog+unix:///vdiname?socket=path */
1084 if (uri
->server
|| uri
->port
) {
1085 error_setg(&err
, "URI scheme %s doesn't accept a server address",
1091 "URI scheme %s requires query parameter 'socket'",
1095 if (qp
->n
!= 1 || strcmp(qp
->p
[0].name
, "socket")) {
1096 error_setg(&err
, "unexpected query parameters");
1099 cfg
->path
= qp
->p
[0].value
;
1101 /* sheepdog[+tcp]://[host:port]/vdiname */
1103 error_setg(&err
, "unexpected query parameters");
1106 cfg
->host
= uri
->server
;
1107 cfg
->port
= uri
->port
;
1111 if (uri
->fragment
) {
1112 if (!sd_parse_snapid_or_tag(uri
->fragment
,
1113 &cfg
->snap_id
, cfg
->tag
)) {
1114 error_setg(&err
, "'%s' is not a valid snapshot ID",
1119 cfg
->snap_id
= CURRENT_VDI_ID
; /* search current vdi */
1124 error_propagate(errp
, err
);
1125 sd_config_done(cfg
);
1130 * Parse a filename (old syntax)
1132 * filename must be one of the following formats:
1134 * 2. [vdiname]:[snapid]
1135 * 3. [vdiname]:[tag]
1136 * 4. [hostname]:[port]:[vdiname]
1137 * 5. [hostname]:[port]:[vdiname]:[snapid]
1138 * 6. [hostname]:[port]:[vdiname]:[tag]
1140 * You can boot from the snapshot images by specifying `snapid` or
1143 * You can run VMs outside the Sheepdog cluster by specifying
1144 * `hostname' and `port' (experimental).
1146 static void parse_vdiname(SheepdogConfig
*cfg
, const char *filename
,
1151 const char *host_spec
, *vdi_spec
;
1154 strstart(filename
, "sheepdog:", &filename
);
1155 p
= q
= g_strdup(filename
);
1157 /* count the number of separators */
1167 /* use the first two tokens as host_spec. */
1180 p
= strchr(vdi_spec
, ':');
1185 uri
= g_strdup_printf("sheepdog://%s/%s", host_spec
, vdi_spec
);
1188 * FIXME We to escape URI meta-characters, e.g. "x?y=z"
1189 * produces "sheepdog://x?y=z". Because of that ...
1191 sd_parse_uri(cfg
, uri
, &err
);
1194 * ... this can fail, but the error message is misleading.
1195 * Replace it by the traditional useless one until the
1196 * escaping is fixed.
1199 error_setg(errp
, "Can't parse filename");
1206 static void sd_parse_filename(const char *filename
, QDict
*options
,
1213 if (strstr(filename
, "://")) {
1214 sd_parse_uri(&cfg
, filename
, &err
);
1216 parse_vdiname(&cfg
, filename
, &err
);
1219 error_propagate(errp
, err
);
1224 qdict_set_default_str(options
, "server.path", cfg
.path
);
1225 qdict_set_default_str(options
, "server.type", "unix");
1227 qdict_set_default_str(options
, "server.type", "inet");
1228 qdict_set_default_str(options
, "server.host",
1229 cfg
.host
?: SD_DEFAULT_ADDR
);
1230 snprintf(buf
, sizeof(buf
), "%d", cfg
.port
?: SD_DEFAULT_PORT
);
1231 qdict_set_default_str(options
, "server.port", buf
);
1233 qdict_set_default_str(options
, "vdi", cfg
.vdi
);
1234 qdict_set_default_str(options
, "tag", cfg
.tag
);
1236 snprintf(buf
, sizeof(buf
), "%d", cfg
.snap_id
);
1237 qdict_set_default_str(options
, "snap-id", buf
);
1240 sd_config_done(&cfg
);
1243 static int find_vdi_name(BDRVSheepdogState
*s
, const char *filename
,
1244 uint32_t snapid
, const char *tag
, uint32_t *vid
,
1245 bool lock
, Error
**errp
)
1249 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1250 unsigned int wlen
, rlen
= 0;
1251 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
1253 fd
= connect_to_sdog(s
, errp
);
1258 /* This pair of strncpy calls ensures that the buffer is zero-filled,
1259 * which is desirable since we'll soon be sending those bytes, and
1260 * don't want the send_req to read uninitialized data.
1262 strncpy(buf
, filename
, SD_MAX_VDI_LEN
);
1263 strncpy(buf
+ SD_MAX_VDI_LEN
, tag
, SD_MAX_VDI_TAG_LEN
);
1265 memset(&hdr
, 0, sizeof(hdr
));
1267 hdr
.opcode
= SD_OP_LOCK_VDI
;
1268 hdr
.type
= LOCK_TYPE_NORMAL
;
1270 hdr
.opcode
= SD_OP_GET_VDI_INFO
;
1272 wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
;
1273 hdr
.proto_ver
= SD_PROTO_VER
;
1274 hdr
.data_length
= wlen
;
1275 hdr
.snapid
= snapid
;
1276 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1278 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1280 error_setg_errno(errp
, -ret
, "cannot get vdi info");
1284 if (rsp
->result
!= SD_RES_SUCCESS
) {
1285 error_setg(errp
, "cannot get vdi info, %s, %s %" PRIu32
" %s",
1286 sd_strerror(rsp
->result
), filename
, snapid
, tag
);
1287 if (rsp
->result
== SD_RES_NO_VDI
) {
1289 } else if (rsp
->result
== SD_RES_VDI_LOCKED
) {
1304 static void coroutine_fn
add_aio_request(BDRVSheepdogState
*s
, AIOReq
*aio_req
,
1305 struct iovec
*iov
, int niov
,
1306 enum AIOCBState aiocb_type
)
1308 int nr_copies
= s
->inode
.nr_copies
;
1310 unsigned int wlen
= 0;
1312 uint64_t oid
= aio_req
->oid
;
1313 unsigned int datalen
= aio_req
->data_len
;
1314 uint64_t offset
= aio_req
->offset
;
1315 uint8_t flags
= aio_req
->flags
;
1316 uint64_t old_oid
= aio_req
->base_oid
;
1317 bool create
= aio_req
->create
;
1319 qemu_co_mutex_lock(&s
->queue_lock
);
1320 QLIST_INSERT_HEAD(&s
->inflight_aio_head
, aio_req
, aio_siblings
);
1321 qemu_co_mutex_unlock(&s
->queue_lock
);
1324 error_report("bug");
1327 memset(&hdr
, 0, sizeof(hdr
));
1329 switch (aiocb_type
) {
1330 case AIOCB_FLUSH_CACHE
:
1331 hdr
.opcode
= SD_OP_FLUSH_VDI
;
1333 case AIOCB_READ_UDATA
:
1334 hdr
.opcode
= SD_OP_READ_OBJ
;
1337 case AIOCB_WRITE_UDATA
:
1339 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1341 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1344 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1346 case AIOCB_DISCARD_OBJ
:
1347 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1348 hdr
.flags
= SD_FLAG_CMD_WRITE
| flags
;
1349 s
->inode
.data_vdi_id
[data_oid_to_idx(oid
)] = 0;
1350 offset
= offsetof(SheepdogInode
,
1351 data_vdi_id
[data_oid_to_idx(oid
)]);
1352 oid
= vid_to_vdi_oid(s
->inode
.vdi_id
);
1353 wlen
= datalen
= sizeof(uint32_t);
1357 if (s
->cache_flags
) {
1358 hdr
.flags
|= s
->cache_flags
;
1362 hdr
.cow_oid
= old_oid
;
1363 hdr
.copies
= s
->inode
.nr_copies
;
1365 hdr
.data_length
= datalen
;
1366 hdr
.offset
= offset
;
1368 hdr
.id
= aio_req
->id
;
1370 qemu_co_mutex_lock(&s
->lock
);
1371 s
->co_send
= qemu_coroutine_self();
1372 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1373 co_read_response
, co_write_request
, NULL
, s
);
1374 socket_set_cork(s
->fd
, 1);
1377 ret
= qemu_co_send(s
->fd
, &hdr
, sizeof(hdr
));
1378 if (ret
!= sizeof(hdr
)) {
1379 error_report("failed to send a req, %s", strerror(errno
));
1384 ret
= qemu_co_sendv(s
->fd
, iov
, niov
, aio_req
->iov_offset
, wlen
);
1386 error_report("failed to send a data, %s", strerror(errno
));
1390 socket_set_cork(s
->fd
, 0);
1391 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1392 co_read_response
, NULL
, NULL
, s
);
1394 qemu_co_mutex_unlock(&s
->lock
);
1397 static int read_write_object(int fd
, BlockDriverState
*bs
, char *buf
,
1398 uint64_t oid
, uint8_t copies
,
1399 unsigned int datalen
, uint64_t offset
,
1400 bool write
, bool create
, uint32_t cache_flags
)
1403 SheepdogObjRsp
*rsp
= (SheepdogObjRsp
*)&hdr
;
1404 unsigned int wlen
, rlen
;
1407 memset(&hdr
, 0, sizeof(hdr
));
1412 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1414 hdr
.opcode
= SD_OP_CREATE_AND_WRITE_OBJ
;
1416 hdr
.opcode
= SD_OP_WRITE_OBJ
;
1421 hdr
.opcode
= SD_OP_READ_OBJ
;
1424 hdr
.flags
|= cache_flags
;
1427 hdr
.data_length
= datalen
;
1428 hdr
.offset
= offset
;
1429 hdr
.copies
= copies
;
1431 ret
= do_req(fd
, bs
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1433 error_report("failed to send a request to the sheep");
1437 switch (rsp
->result
) {
1438 case SD_RES_SUCCESS
:
1441 error_report("%s", sd_strerror(rsp
->result
));
1446 static int read_object(int fd
, BlockDriverState
*bs
, char *buf
,
1447 uint64_t oid
, uint8_t copies
,
1448 unsigned int datalen
, uint64_t offset
,
1449 uint32_t cache_flags
)
1451 return read_write_object(fd
, bs
, buf
, oid
, copies
,
1452 datalen
, offset
, false,
1453 false, cache_flags
);
1456 static int write_object(int fd
, BlockDriverState
*bs
, char *buf
,
1457 uint64_t oid
, uint8_t copies
,
1458 unsigned int datalen
, uint64_t offset
, bool create
,
1459 uint32_t cache_flags
)
1461 return read_write_object(fd
, bs
, buf
, oid
, copies
,
1462 datalen
, offset
, true,
1463 create
, cache_flags
);
1466 /* update inode with the latest state */
1467 static int reload_inode(BDRVSheepdogState
*s
, uint32_t snapid
, const char *tag
)
1469 Error
*local_err
= NULL
;
1470 SheepdogInode
*inode
;
1474 fd
= connect_to_sdog(s
, &local_err
);
1476 error_report_err(local_err
);
1480 inode
= g_malloc(SD_INODE_HEADER_SIZE
);
1482 ret
= find_vdi_name(s
, s
->name
, snapid
, tag
, &vid
, false, &local_err
);
1484 error_report_err(local_err
);
1488 ret
= read_object(fd
, s
->bs
, (char *)inode
, vid_to_vdi_oid(vid
),
1489 s
->inode
.nr_copies
, SD_INODE_HEADER_SIZE
, 0,
1495 if (inode
->vdi_id
!= s
->inode
.vdi_id
) {
1496 memcpy(&s
->inode
, inode
, SD_INODE_HEADER_SIZE
);
1506 static void coroutine_fn
resend_aioreq(BDRVSheepdogState
*s
, AIOReq
*aio_req
)
1508 SheepdogAIOCB
*acb
= aio_req
->aiocb
;
1510 aio_req
->create
= false;
1512 /* check whether this request becomes a CoW one */
1513 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& is_data_obj(aio_req
->oid
)) {
1514 int idx
= data_oid_to_idx(aio_req
->oid
);
1516 if (is_data_obj_writable(&s
->inode
, idx
)) {
1520 if (s
->inode
.data_vdi_id
[idx
]) {
1521 aio_req
->base_oid
= vid_to_data_oid(s
->inode
.data_vdi_id
[idx
], idx
);
1522 aio_req
->flags
|= SD_FLAG_CMD_COW
;
1524 aio_req
->create
= true;
1527 if (is_data_obj(aio_req
->oid
)) {
1528 add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
1532 iov
.iov_base
= &s
->inode
;
1533 iov
.iov_len
= sizeof(s
->inode
);
1534 add_aio_request(s
, aio_req
, &iov
, 1, AIOCB_WRITE_UDATA
);
1538 static void sd_detach_aio_context(BlockDriverState
*bs
)
1540 BDRVSheepdogState
*s
= bs
->opaque
;
1542 aio_set_fd_handler(s
->aio_context
, s
->fd
, false, NULL
,
1546 static void sd_attach_aio_context(BlockDriverState
*bs
,
1547 AioContext
*new_context
)
1549 BDRVSheepdogState
*s
= bs
->opaque
;
1551 s
->aio_context
= new_context
;
1552 aio_set_fd_handler(new_context
, s
->fd
, false,
1553 co_read_response
, NULL
, NULL
, s
);
1556 static QemuOptsList runtime_opts
= {
1558 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1562 .type
= QEMU_OPT_STRING
,
1566 .type
= QEMU_OPT_NUMBER
,
1570 .type
= QEMU_OPT_STRING
,
1572 { /* end of list */ }
1576 static int sd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1581 BDRVSheepdogState
*s
= bs
->opaque
;
1582 const char *vdi
, *snap_id_str
, *tag
;
1586 Error
*local_err
= NULL
;
1589 s
->aio_context
= bdrv_get_aio_context(bs
);
1591 opts
= qemu_opts_create(&runtime_opts
, NULL
, 0, &error_abort
);
1592 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1594 error_propagate(errp
, local_err
);
1599 s
->addr
= sd_server_config(options
, errp
);
1605 vdi
= qemu_opt_get(opts
, "vdi");
1606 snap_id_str
= qemu_opt_get(opts
, "snap-id");
1607 snap_id
= qemu_opt_get_number(opts
, "snap-id", CURRENT_VDI_ID
);
1608 tag
= qemu_opt_get(opts
, "tag");
1611 error_setg(errp
, "parameter 'vdi' is missing");
1615 if (strlen(vdi
) >= SD_MAX_VDI_LEN
) {
1616 error_setg(errp
, "value of parameter 'vdi' is too long");
1621 if (snap_id
> UINT32_MAX
) {
1624 if (snap_id_str
&& !snap_id
) {
1625 error_setg(errp
, "'snap-id=%s' is not a valid snapshot ID",
1634 if (strlen(tag
) >= SD_MAX_VDI_TAG_LEN
) {
1635 error_setg(errp
, "value of parameter 'tag' is too long");
1640 QLIST_INIT(&s
->inflight_aio_head
);
1641 QLIST_INIT(&s
->failed_aio_head
);
1642 QLIST_INIT(&s
->inflight_aiocb_head
);
1644 s
->fd
= get_sheep_fd(s
, errp
);
1650 ret
= find_vdi_name(s
, vdi
, (uint32_t)snap_id
, tag
, &vid
, true, errp
);
1656 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1657 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1659 s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1660 if (flags
& BDRV_O_NOCACHE
) {
1661 s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1663 s
->discard_supported
= true;
1665 if (snap_id
|| tag
[0]) {
1666 DPRINTF("%" PRIx32
" snapshot inode was open.\n", vid
);
1667 s
->is_snapshot
= true;
1670 fd
= connect_to_sdog(s
, errp
);
1676 buf
= g_malloc(SD_INODE_SIZE
);
1677 ret
= read_object(fd
, s
->bs
, buf
, vid_to_vdi_oid(vid
),
1678 0, SD_INODE_SIZE
, 0, s
->cache_flags
);
1683 error_setg(errp
, "Can't read snapshot inode");
1687 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
1689 bs
->total_sectors
= s
->inode
.vdi_size
/ BDRV_SECTOR_SIZE
;
1690 pstrcpy(s
->name
, sizeof(s
->name
), vdi
);
1691 qemu_co_mutex_init(&s
->lock
);
1692 qemu_co_mutex_init(&s
->queue_lock
);
1693 qemu_co_queue_init(&s
->overlapping_queue
);
1694 qemu_opts_del(opts
);
1699 aio_set_fd_handler(bdrv_get_aio_context(bs
), s
->fd
,
1700 false, NULL
, NULL
, NULL
, NULL
);
1703 qemu_opts_del(opts
);
1708 static int sd_reopen_prepare(BDRVReopenState
*state
, BlockReopenQueue
*queue
,
1711 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1712 BDRVSheepdogReopenState
*re_s
;
1715 re_s
= state
->opaque
= g_new0(BDRVSheepdogReopenState
, 1);
1717 re_s
->cache_flags
= SD_FLAG_CMD_CACHE
;
1718 if (state
->flags
& BDRV_O_NOCACHE
) {
1719 re_s
->cache_flags
= SD_FLAG_CMD_DIRECT
;
1722 re_s
->fd
= get_sheep_fd(s
, errp
);
1731 static void sd_reopen_commit(BDRVReopenState
*state
)
1733 BDRVSheepdogReopenState
*re_s
= state
->opaque
;
1734 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1737 aio_set_fd_handler(s
->aio_context
, s
->fd
, false,
1738 NULL
, NULL
, NULL
, NULL
);
1743 s
->cache_flags
= re_s
->cache_flags
;
1745 g_free(state
->opaque
);
1746 state
->opaque
= NULL
;
1751 static void sd_reopen_abort(BDRVReopenState
*state
)
1753 BDRVSheepdogReopenState
*re_s
= state
->opaque
;
1754 BDRVSheepdogState
*s
= state
->bs
->opaque
;
1761 aio_set_fd_handler(s
->aio_context
, re_s
->fd
, false,
1762 NULL
, NULL
, NULL
, NULL
);
1763 closesocket(re_s
->fd
);
1766 g_free(state
->opaque
);
1767 state
->opaque
= NULL
;
1772 static int do_sd_create(BDRVSheepdogState
*s
, uint32_t *vdi_id
, int snapshot
,
1776 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
1778 unsigned int wlen
, rlen
= 0;
1779 char buf
[SD_MAX_VDI_LEN
];
1781 fd
= connect_to_sdog(s
, errp
);
1786 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1787 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1789 memset(buf
, 0, sizeof(buf
));
1790 pstrcpy(buf
, sizeof(buf
), s
->name
);
1792 memset(&hdr
, 0, sizeof(hdr
));
1793 hdr
.opcode
= SD_OP_NEW_VDI
;
1794 hdr
.base_vdi_id
= s
->inode
.vdi_id
;
1796 wlen
= SD_MAX_VDI_LEN
;
1798 hdr
.flags
= SD_FLAG_CMD_WRITE
;
1799 hdr
.snapid
= snapshot
;
1801 hdr
.data_length
= wlen
;
1802 hdr
.vdi_size
= s
->inode
.vdi_size
;
1803 hdr
.copy_policy
= s
->inode
.copy_policy
;
1804 hdr
.copies
= s
->inode
.nr_copies
;
1805 hdr
.block_size_shift
= s
->inode
.block_size_shift
;
1807 ret
= do_req(fd
, NULL
, (SheepdogReq
*)&hdr
, buf
, &wlen
, &rlen
);
1812 error_setg_errno(errp
, -ret
, "create failed");
1816 if (rsp
->result
!= SD_RES_SUCCESS
) {
1817 error_setg(errp
, "%s, %s", sd_strerror(rsp
->result
), s
->inode
.name
);
1822 *vdi_id
= rsp
->vdi_id
;
1828 static int sd_prealloc(const char *filename
, Error
**errp
)
1830 BlockBackend
*blk
= NULL
;
1831 BDRVSheepdogState
*base
= NULL
;
1832 unsigned long buf_size
;
1833 uint32_t idx
, max_idx
;
1834 uint32_t object_size
;
1839 blk
= blk_new_open(filename
, NULL
, NULL
,
1840 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
1843 goto out_with_err_set
;
1846 blk_set_allow_write_beyond_eof(blk
, true);
1848 vdi_size
= blk_getlength(blk
);
1854 base
= blk_bs(blk
)->opaque
;
1855 object_size
= (UINT32_C(1) << base
->inode
.block_size_shift
);
1856 buf_size
= MIN(object_size
, SD_DATA_OBJ_SIZE
);
1857 buf
= g_malloc0(buf_size
);
1859 max_idx
= DIV_ROUND_UP(vdi_size
, buf_size
);
1861 for (idx
= 0; idx
< max_idx
; idx
++) {
1863 * The created image can be a cloned image, so we need to read
1864 * a data from the source image.
1866 ret
= blk_pread(blk
, idx
* buf_size
, buf
, buf_size
);
1870 ret
= blk_pwrite(blk
, idx
* buf_size
, buf
, buf_size
, 0);
1879 error_setg_errno(errp
, -ret
, "Can't pre-allocate");
1891 * Sheepdog support two kinds of redundancy, full replication and erasure
1894 * # create a fully replicated vdi with x copies
1895 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1897 * # create a erasure coded vdi with x data strips and y parity strips
1898 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1900 static int parse_redundancy(BDRVSheepdogState
*s
, const char *opt
)
1902 struct SheepdogInode
*inode
= &s
->inode
;
1903 const char *n1
, *n2
;
1907 pstrcpy(p
, sizeof(p
), opt
);
1908 n1
= strtok(p
, ":");
1909 n2
= strtok(NULL
, ":");
1915 copy
= strtol(n1
, NULL
, 10);
1916 /* FIXME fix error checking by switching to qemu_strtol() */
1917 if (copy
> SD_MAX_COPIES
|| copy
< 1) {
1921 inode
->copy_policy
= 0;
1922 inode
->nr_copies
= copy
;
1926 if (copy
!= 2 && copy
!= 4 && copy
!= 8 && copy
!= 16) {
1930 parity
= strtol(n2
, NULL
, 10);
1931 /* FIXME fix error checking by switching to qemu_strtol() */
1932 if (parity
>= SD_EC_MAX_STRIP
|| parity
< 1) {
1937 * 4 bits for parity and 4 bits for data.
1938 * We have to compress upper data bits because it can't represent 16
1940 inode
->copy_policy
= ((copy
/ 2) << 4) + parity
;
1941 inode
->nr_copies
= copy
+ parity
;
1946 static int parse_block_size_shift(BDRVSheepdogState
*s
, QemuOpts
*opt
)
1948 struct SheepdogInode
*inode
= &s
->inode
;
1949 uint64_t object_size
;
1952 object_size
= qemu_opt_get_size_del(opt
, BLOCK_OPT_OBJECT_SIZE
, 0);
1954 if ((object_size
- 1) & object_size
) { /* not a power of 2? */
1957 obj_order
= ctz32(object_size
);
1958 if (obj_order
< 20 || obj_order
> 31) {
1961 inode
->block_size_shift
= (uint8_t)obj_order
;
1967 static int sd_create(const char *filename
, QemuOpts
*opts
,
1973 char *backing_file
= NULL
;
1975 BDRVSheepdogState
*s
;
1977 uint64_t max_vdi_size
;
1978 bool prealloc
= false;
1980 s
= g_new0(BDRVSheepdogState
, 1);
1982 if (strstr(filename
, "://")) {
1983 sd_parse_uri(&cfg
, filename
, &err
);
1985 parse_vdiname(&cfg
, filename
, &err
);
1988 error_propagate(errp
, err
);
1992 buf
= cfg
.port
? g_strdup_printf("%d", cfg
.port
) : NULL
;
1993 s
->addr
= sd_socket_address(cfg
.path
, cfg
.host
, buf
);
1995 strcpy(s
->name
, cfg
.vdi
);
1996 sd_config_done(&cfg
);
1998 s
->inode
.vdi_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2000 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
2001 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2002 if (!buf
|| !strcmp(buf
, "off")) {
2004 } else if (!strcmp(buf
, "full")) {
2007 error_setg(errp
, "Invalid preallocation mode: '%s'", buf
);
2013 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_REDUNDANCY
);
2015 ret
= parse_redundancy(s
, buf
);
2017 error_setg(errp
, "Invalid redundancy mode: '%s'", buf
);
2021 ret
= parse_block_size_shift(s
, opts
);
2023 error_setg(errp
, "Invalid object_size."
2024 " obect_size needs to be power of 2"
2025 " and be limited from 2^20 to 2^31");
2031 BDRVSheepdogState
*base
;
2034 /* Currently, only Sheepdog backing image is supported. */
2035 drv
= bdrv_find_protocol(backing_file
, true, NULL
);
2036 if (!drv
|| strcmp(drv
->protocol_name
, "sheepdog") != 0) {
2037 error_setg(errp
, "backing_file must be a sheepdog image");
2042 blk
= blk_new_open(backing_file
, NULL
, NULL
,
2043 BDRV_O_PROTOCOL
, errp
);
2049 base
= blk_bs(blk
)->opaque
;
2051 if (!is_snapshot(&base
->inode
)) {
2052 error_setg(errp
, "cannot clone from a non snapshot vdi");
2057 s
->inode
.vdi_id
= base
->inode
.vdi_id
;
2061 s
->aio_context
= qemu_get_aio_context();
2063 /* if block_size_shift is not specified, get cluster default value */
2064 if (s
->inode
.block_size_shift
== 0) {
2066 SheepdogClusterRsp
*rsp
= (SheepdogClusterRsp
*)&hdr
;
2068 unsigned int wlen
= 0, rlen
= 0;
2070 fd
= connect_to_sdog(s
, errp
);
2076 memset(&hdr
, 0, sizeof(hdr
));
2077 hdr
.opcode
= SD_OP_GET_CLUSTER_DEFAULT
;
2078 hdr
.proto_ver
= SD_PROTO_VER
;
2080 ret
= do_req(fd
, NULL
, (SheepdogReq
*)&hdr
,
2081 NULL
, &wlen
, &rlen
);
2084 error_setg_errno(errp
, -ret
, "failed to get cluster default");
2087 if (rsp
->result
== SD_RES_SUCCESS
) {
2088 s
->inode
.block_size_shift
= rsp
->block_size_shift
;
2090 s
->inode
.block_size_shift
= SD_DEFAULT_BLOCK_SIZE_SHIFT
;
2094 max_vdi_size
= (UINT64_C(1) << s
->inode
.block_size_shift
) * MAX_DATA_OBJS
;
2096 if (s
->inode
.vdi_size
> max_vdi_size
) {
2097 error_setg(errp
, "An image is too large."
2098 " The maximum image size is %"PRIu64
"GB",
2099 max_vdi_size
/ 1024 / 1024 / 1024);
2104 ret
= do_sd_create(s
, &vid
, 0, errp
);
2110 ret
= sd_prealloc(filename
, errp
);
2113 g_free(backing_file
);
2119 static void sd_close(BlockDriverState
*bs
)
2121 Error
*local_err
= NULL
;
2122 BDRVSheepdogState
*s
= bs
->opaque
;
2124 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2125 unsigned int wlen
, rlen
= 0;
2128 DPRINTF("%s\n", s
->name
);
2130 fd
= connect_to_sdog(s
, &local_err
);
2132 error_report_err(local_err
);
2136 memset(&hdr
, 0, sizeof(hdr
));
2138 hdr
.opcode
= SD_OP_RELEASE_VDI
;
2139 hdr
.type
= LOCK_TYPE_NORMAL
;
2140 hdr
.base_vdi_id
= s
->inode
.vdi_id
;
2141 wlen
= strlen(s
->name
) + 1;
2142 hdr
.data_length
= wlen
;
2143 hdr
.flags
= SD_FLAG_CMD_WRITE
;
2145 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2146 s
->name
, &wlen
, &rlen
);
2150 if (!ret
&& rsp
->result
!= SD_RES_SUCCESS
&&
2151 rsp
->result
!= SD_RES_VDI_NOT_LOCKED
) {
2152 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
2155 aio_set_fd_handler(bdrv_get_aio_context(bs
), s
->fd
,
2156 false, NULL
, NULL
, NULL
, NULL
);
2158 qapi_free_SocketAddress(s
->addr
);
2161 static int64_t sd_getlength(BlockDriverState
*bs
)
2163 BDRVSheepdogState
*s
= bs
->opaque
;
2165 return s
->inode
.vdi_size
;
2168 static int sd_truncate(BlockDriverState
*bs
, int64_t offset
,
2169 PreallocMode prealloc
, Error
**errp
)
2171 BDRVSheepdogState
*s
= bs
->opaque
;
2173 unsigned int datalen
;
2174 uint64_t max_vdi_size
;
2176 if (prealloc
!= PREALLOC_MODE_OFF
) {
2177 error_setg(errp
, "Unsupported preallocation mode '%s'",
2178 PreallocMode_str(prealloc
));
2182 max_vdi_size
= (UINT64_C(1) << s
->inode
.block_size_shift
) * MAX_DATA_OBJS
;
2183 if (offset
< s
->inode
.vdi_size
) {
2184 error_setg(errp
, "shrinking is not supported");
2186 } else if (offset
> max_vdi_size
) {
2187 error_setg(errp
, "too big image size");
2191 fd
= connect_to_sdog(s
, errp
);
2196 /* we don't need to update entire object */
2197 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2198 s
->inode
.vdi_size
= offset
;
2199 ret
= write_object(fd
, s
->bs
, (char *)&s
->inode
,
2200 vid_to_vdi_oid(s
->inode
.vdi_id
), s
->inode
.nr_copies
,
2201 datalen
, 0, false, s
->cache_flags
);
2205 error_setg_errno(errp
, -ret
, "failed to update an inode");
2212 * This function is called after writing data objects. If we need to
2213 * update metadata, this sends a write request to the vdi object.
2215 static void coroutine_fn
sd_write_done(SheepdogAIOCB
*acb
)
2217 BDRVSheepdogState
*s
= acb
->s
;
2220 uint32_t offset
, data_len
, mn
, mx
;
2222 mn
= acb
->min_dirty_data_idx
;
2223 mx
= acb
->max_dirty_data_idx
;
2225 /* we need to update the vdi object. */
2227 offset
= sizeof(s
->inode
) - sizeof(s
->inode
.data_vdi_id
) +
2228 mn
* sizeof(s
->inode
.data_vdi_id
[0]);
2229 data_len
= (mx
- mn
+ 1) * sizeof(s
->inode
.data_vdi_id
[0]);
2231 acb
->min_dirty_data_idx
= UINT32_MAX
;
2232 acb
->max_dirty_data_idx
= 0;
2234 iov
.iov_base
= &s
->inode
;
2235 iov
.iov_len
= sizeof(s
->inode
);
2236 aio_req
= alloc_aio_req(s
, acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2237 data_len
, offset
, 0, false, 0, offset
);
2238 add_aio_request(s
, aio_req
, &iov
, 1, AIOCB_WRITE_UDATA
);
2239 if (--acb
->nr_pending
) {
2240 qemu_coroutine_yield();
2245 /* Delete current working VDI on the snapshot chain */
2246 static bool sd_delete(BDRVSheepdogState
*s
)
2248 Error
*local_err
= NULL
;
2249 unsigned int wlen
= SD_MAX_VDI_LEN
, rlen
= 0;
2250 SheepdogVdiReq hdr
= {
2251 .opcode
= SD_OP_DEL_VDI
,
2252 .base_vdi_id
= s
->inode
.vdi_id
,
2253 .data_length
= wlen
,
2254 .flags
= SD_FLAG_CMD_WRITE
,
2256 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2259 fd
= connect_to_sdog(s
, &local_err
);
2261 error_report_err(local_err
);
2265 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2266 s
->name
, &wlen
, &rlen
);
2271 switch (rsp
->result
) {
2273 error_report("%s was already deleted", s
->name
);
2275 case SD_RES_SUCCESS
:
2278 error_report("%s, %s", sd_strerror(rsp
->result
), s
->name
);
2286 * Create a writable VDI from a snapshot
2288 static int sd_create_branch(BDRVSheepdogState
*s
)
2290 Error
*local_err
= NULL
;
2296 DPRINTF("%" PRIx32
" is snapshot.\n", s
->inode
.vdi_id
);
2298 buf
= g_malloc(SD_INODE_SIZE
);
2301 * Even If deletion fails, we will just create extra snapshot based on
2302 * the working VDI which was supposed to be deleted. So no need to
2305 deleted
= sd_delete(s
);
2306 ret
= do_sd_create(s
, &vid
, !deleted
, &local_err
);
2308 error_report_err(local_err
);
2312 DPRINTF("%" PRIx32
" is created.\n", vid
);
2314 fd
= connect_to_sdog(s
, &local_err
);
2316 error_report_err(local_err
);
2321 ret
= read_object(fd
, s
->bs
, buf
, vid_to_vdi_oid(vid
),
2322 s
->inode
.nr_copies
, SD_INODE_SIZE
, 0, s
->cache_flags
);
2330 memcpy(&s
->inode
, buf
, sizeof(s
->inode
));
2332 s
->is_snapshot
= false;
2334 DPRINTF("%" PRIx32
" was newly created.\n", s
->inode
.vdi_id
);
2343 * Send I/O requests to the server.
2345 * This function sends requests to the server, links the requests to
2346 * the inflight_list in BDRVSheepdogState, and exits without
2347 * waiting the response. The responses are received in the
2348 * `aio_read_response' function which is called from the main loop as
2351 * Returns 1 when we need to wait a response, 0 when there is no sent
2352 * request and -errno in error cases.
2354 static void coroutine_fn
sd_co_rw_vector(SheepdogAIOCB
*acb
)
2357 unsigned long len
, done
= 0, total
= acb
->nb_sectors
* BDRV_SECTOR_SIZE
;
2359 uint32_t object_size
;
2362 BDRVSheepdogState
*s
= acb
->s
;
2363 SheepdogInode
*inode
= &s
->inode
;
2366 if (acb
->aiocb_type
== AIOCB_WRITE_UDATA
&& s
->is_snapshot
) {
2368 * In the case we open the snapshot VDI, Sheepdog creates the
2369 * writable VDI when we do a write operation first.
2371 ret
= sd_create_branch(s
);
2378 object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2379 idx
= acb
->sector_num
* BDRV_SECTOR_SIZE
/ object_size
;
2380 offset
= (acb
->sector_num
* BDRV_SECTOR_SIZE
) % object_size
;
2383 * Make sure we don't free the aiocb before we are done with all requests.
2384 * This additional reference is dropped at the end of this function.
2388 while (done
!= total
) {
2390 uint64_t old_oid
= 0;
2391 bool create
= false;
2393 oid
= vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
);
2395 len
= MIN(total
- done
, object_size
- offset
);
2397 switch (acb
->aiocb_type
) {
2398 case AIOCB_READ_UDATA
:
2399 if (!inode
->data_vdi_id
[idx
]) {
2400 qemu_iovec_memset(acb
->qiov
, done
, 0, len
);
2404 case AIOCB_WRITE_UDATA
:
2405 if (!inode
->data_vdi_id
[idx
]) {
2407 } else if (!is_data_obj_writable(inode
, idx
)) {
2411 flags
= SD_FLAG_CMD_COW
;
2414 case AIOCB_DISCARD_OBJ
:
2416 * We discard the object only when the whole object is
2417 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2419 if (len
!= object_size
|| inode
->data_vdi_id
[idx
] == 0) {
2428 DPRINTF("update ino (%" PRIu32
") %" PRIu64
" %" PRIu64
" %ld\n",
2430 vid_to_data_oid(inode
->data_vdi_id
[idx
], idx
), idx
);
2431 oid
= vid_to_data_oid(inode
->vdi_id
, idx
);
2432 DPRINTF("new oid %" PRIx64
"\n", oid
);
2435 aio_req
= alloc_aio_req(s
, acb
, oid
, len
, offset
, flags
, create
,
2437 acb
->aiocb_type
== AIOCB_DISCARD_OBJ
?
2439 add_aio_request(s
, aio_req
, acb
->qiov
->iov
, acb
->qiov
->niov
,
2446 if (--acb
->nr_pending
) {
2447 qemu_coroutine_yield();
2451 static void sd_aio_complete(SheepdogAIOCB
*acb
)
2453 BDRVSheepdogState
*s
;
2454 if (acb
->aiocb_type
== AIOCB_FLUSH_CACHE
) {
2459 qemu_co_mutex_lock(&s
->queue_lock
);
2460 QLIST_REMOVE(acb
, aiocb_siblings
);
2461 qemu_co_queue_restart_all(&s
->overlapping_queue
);
2462 qemu_co_mutex_unlock(&s
->queue_lock
);
2465 static coroutine_fn
int sd_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2466 int nb_sectors
, QEMUIOVector
*qiov
)
2470 int64_t offset
= (sector_num
+ nb_sectors
) * BDRV_SECTOR_SIZE
;
2471 BDRVSheepdogState
*s
= bs
->opaque
;
2473 if (offset
> s
->inode
.vdi_size
) {
2474 ret
= sd_truncate(bs
, offset
, PREALLOC_MODE_OFF
, NULL
);
2480 sd_aio_setup(&acb
, s
, qiov
, sector_num
, nb_sectors
, AIOCB_WRITE_UDATA
);
2481 sd_co_rw_vector(&acb
);
2482 sd_write_done(&acb
);
2483 sd_aio_complete(&acb
);
2488 static coroutine_fn
int sd_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2489 int nb_sectors
, QEMUIOVector
*qiov
)
2492 BDRVSheepdogState
*s
= bs
->opaque
;
2494 sd_aio_setup(&acb
, s
, qiov
, sector_num
, nb_sectors
, AIOCB_READ_UDATA
);
2495 sd_co_rw_vector(&acb
);
2496 sd_aio_complete(&acb
);
2501 static int coroutine_fn
sd_co_flush_to_disk(BlockDriverState
*bs
)
2503 BDRVSheepdogState
*s
= bs
->opaque
;
2507 if (s
->cache_flags
!= SD_FLAG_CMD_CACHE
) {
2511 sd_aio_setup(&acb
, s
, NULL
, 0, 0, AIOCB_FLUSH_CACHE
);
2514 aio_req
= alloc_aio_req(s
, &acb
, vid_to_vdi_oid(s
->inode
.vdi_id
),
2515 0, 0, 0, false, 0, 0);
2516 add_aio_request(s
, aio_req
, NULL
, 0, acb
.aiocb_type
);
2518 if (--acb
.nr_pending
) {
2519 qemu_coroutine_yield();
2522 sd_aio_complete(&acb
);
2526 static int sd_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
2528 Error
*local_err
= NULL
;
2529 BDRVSheepdogState
*s
= bs
->opaque
;
2532 SheepdogInode
*inode
;
2533 unsigned int datalen
;
2535 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64
" "
2536 "is_snapshot %d\n", sn_info
->name
, sn_info
->id_str
,
2537 s
->name
, sn_info
->vm_state_size
, s
->is_snapshot
);
2539 if (s
->is_snapshot
) {
2540 error_report("You can't create a snapshot of a snapshot VDI, "
2541 "%s (%" PRIu32
").", s
->name
, s
->inode
.vdi_id
);
2546 DPRINTF("%s %s\n", sn_info
->name
, sn_info
->id_str
);
2548 s
->inode
.vm_state_size
= sn_info
->vm_state_size
;
2549 s
->inode
.vm_clock_nsec
= sn_info
->vm_clock_nsec
;
2550 /* It appears that inode.tag does not require a NUL terminator,
2551 * which means this use of strncpy is ok.
2553 strncpy(s
->inode
.tag
, sn_info
->name
, sizeof(s
->inode
.tag
));
2554 /* we don't need to update entire object */
2555 datalen
= SD_INODE_SIZE
- sizeof(s
->inode
.data_vdi_id
);
2556 inode
= g_malloc(datalen
);
2558 /* refresh inode. */
2559 fd
= connect_to_sdog(s
, &local_err
);
2561 error_report_err(local_err
);
2566 ret
= write_object(fd
, s
->bs
, (char *)&s
->inode
,
2567 vid_to_vdi_oid(s
->inode
.vdi_id
), s
->inode
.nr_copies
,
2568 datalen
, 0, false, s
->cache_flags
);
2570 error_report("failed to write snapshot's inode.");
2574 ret
= do_sd_create(s
, &new_vid
, 1, &local_err
);
2576 error_reportf_err(local_err
,
2577 "failed to create inode for snapshot: ");
2581 ret
= read_object(fd
, s
->bs
, (char *)inode
,
2582 vid_to_vdi_oid(new_vid
), s
->inode
.nr_copies
, datalen
, 0,
2586 error_report("failed to read new inode info. %s", strerror(errno
));
2590 memcpy(&s
->inode
, inode
, datalen
);
2591 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2592 s
->inode
.name
, s
->inode
.snap_id
, s
->inode
.vdi_id
);
2601 * We implement rollback(loadvm) operation to the specified snapshot by
2602 * 1) switch to the snapshot
2603 * 2) rely on sd_create_branch to delete working VDI and
2604 * 3) create a new working VDI based on the specified snapshot
2606 static int sd_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
2608 BDRVSheepdogState
*s
= bs
->opaque
;
2609 BDRVSheepdogState
*old_s
;
2610 char tag
[SD_MAX_VDI_TAG_LEN
];
2611 uint32_t snapid
= 0;
2614 if (!sd_parse_snapid_or_tag(snapshot_id
, &snapid
, tag
)) {
2618 old_s
= g_new(BDRVSheepdogState
, 1);
2620 memcpy(old_s
, s
, sizeof(BDRVSheepdogState
));
2622 ret
= reload_inode(s
, snapid
, tag
);
2627 ret
= sd_create_branch(s
);
2636 /* recover bdrv_sd_state */
2637 memcpy(s
, old_s
, sizeof(BDRVSheepdogState
));
2640 error_report("failed to open. recover old bdrv_sd_state.");
2645 #define NR_BATCHED_DISCARD 128
2647 static int remove_objects(BDRVSheepdogState
*s
, Error
**errp
)
2649 int fd
, i
= 0, nr_objs
= 0;
2651 SheepdogInode
*inode
= &s
->inode
;
2653 fd
= connect_to_sdog(s
, errp
);
2658 nr_objs
= count_data_objs(inode
);
2659 while (i
< nr_objs
) {
2660 int start_idx
, nr_filled_idx
;
2662 while (i
< nr_objs
&& !inode
->data_vdi_id
[i
]) {
2668 while (i
< nr_objs
&& nr_filled_idx
< NR_BATCHED_DISCARD
) {
2669 if (inode
->data_vdi_id
[i
]) {
2670 inode
->data_vdi_id
[i
] = 0;
2677 ret
= write_object(fd
, s
->bs
,
2678 (char *)&inode
->data_vdi_id
[start_idx
],
2679 vid_to_vdi_oid(s
->inode
.vdi_id
), inode
->nr_copies
,
2680 (i
- start_idx
) * sizeof(uint32_t),
2681 offsetof(struct SheepdogInode
,
2682 data_vdi_id
[start_idx
]),
2683 false, s
->cache_flags
);
2685 error_setg(errp
, "Failed to discard snapshot inode");
2696 static int sd_snapshot_delete(BlockDriverState
*bs
,
2697 const char *snapshot_id
,
2702 * FIXME should delete the snapshot matching both @snapshot_id and
2703 * @name, but @name not used here
2705 unsigned long snap_id
= 0;
2706 char snap_tag
[SD_MAX_VDI_TAG_LEN
];
2708 char buf
[SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
];
2709 BDRVSheepdogState
*s
= bs
->opaque
;
2710 unsigned int wlen
= SD_MAX_VDI_LEN
+ SD_MAX_VDI_TAG_LEN
, rlen
= 0;
2712 SheepdogVdiReq hdr
= {
2713 .opcode
= SD_OP_DEL_VDI
,
2714 .data_length
= wlen
,
2715 .flags
= SD_FLAG_CMD_WRITE
,
2717 SheepdogVdiRsp
*rsp
= (SheepdogVdiRsp
*)&hdr
;
2719 ret
= remove_objects(s
, errp
);
2724 memset(buf
, 0, sizeof(buf
));
2725 memset(snap_tag
, 0, sizeof(snap_tag
));
2726 pstrcpy(buf
, SD_MAX_VDI_LEN
, s
->name
);
2727 /* TODO Use sd_parse_snapid() once this mess is cleaned up */
2728 ret
= qemu_strtoul(snapshot_id
, NULL
, 10, &snap_id
);
2729 if (ret
|| snap_id
> UINT32_MAX
) {
2731 * FIXME Since qemu_strtoul() returns -EINVAL when
2732 * @snapshot_id is null, @snapshot_id is mandatory. Correct
2733 * would be to require at least one of @snapshot_id and @name.
2735 error_setg(errp
, "Invalid snapshot ID: %s",
2736 snapshot_id
? snapshot_id
: "<null>");
2741 hdr
.snapid
= (uint32_t) snap_id
;
2743 /* FIXME I suspect we should use @name here */
2744 /* FIXME don't truncate silently */
2745 pstrcpy(snap_tag
, sizeof(snap_tag
), snapshot_id
);
2746 pstrcpy(buf
+ SD_MAX_VDI_LEN
, SD_MAX_VDI_TAG_LEN
, snap_tag
);
2749 ret
= find_vdi_name(s
, s
->name
, snap_id
, snap_tag
, &vid
, true, errp
);
2754 fd
= connect_to_sdog(s
, errp
);
2759 ret
= do_req(fd
, s
->bs
, (SheepdogReq
*)&hdr
,
2763 error_setg_errno(errp
, -ret
, "Couldn't send request to server");
2767 switch (rsp
->result
) {
2769 error_setg(errp
, "Can't find the snapshot");
2771 case SD_RES_SUCCESS
:
2774 error_setg(errp
, "%s", sd_strerror(rsp
->result
));
2781 static int sd_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
2783 Error
*local_err
= NULL
;
2784 BDRVSheepdogState
*s
= bs
->opaque
;
2786 int fd
, nr
= 1024, ret
, max
= BITS_TO_LONGS(SD_NR_VDIS
) * sizeof(long);
2787 QEMUSnapshotInfo
*sn_tab
= NULL
;
2788 unsigned wlen
, rlen
;
2790 static SheepdogInode inode
;
2791 unsigned long *vdi_inuse
;
2792 unsigned int start_nr
;
2796 vdi_inuse
= g_malloc(max
);
2798 fd
= connect_to_sdog(s
, &local_err
);
2800 error_report_err(local_err
);
2808 memset(&req
, 0, sizeof(req
));
2810 req
.opcode
= SD_OP_READ_VDIS
;
2811 req
.data_length
= max
;
2813 ret
= do_req(fd
, s
->bs
, &req
, vdi_inuse
, &wlen
, &rlen
);
2820 sn_tab
= g_new0(QEMUSnapshotInfo
, nr
);
2822 /* calculate a vdi id with hash function */
2823 hval
= fnv_64a_buf(s
->name
, strlen(s
->name
), FNV1A_64_INIT
);
2824 start_nr
= hval
& (SD_NR_VDIS
- 1);
2826 fd
= connect_to_sdog(s
, &local_err
);
2828 error_report_err(local_err
);
2833 for (vid
= start_nr
; found
< nr
; vid
= (vid
+ 1) % SD_NR_VDIS
) {
2834 if (!test_bit(vid
, vdi_inuse
)) {
2838 /* we don't need to read entire object */
2839 ret
= read_object(fd
, s
->bs
, (char *)&inode
,
2840 vid_to_vdi_oid(vid
),
2841 0, SD_INODE_SIZE
- sizeof(inode
.data_vdi_id
), 0,
2848 if (!strcmp(inode
.name
, s
->name
) && is_snapshot(&inode
)) {
2849 sn_tab
[found
].date_sec
= inode
.snap_ctime
>> 32;
2850 sn_tab
[found
].date_nsec
= inode
.snap_ctime
& 0xffffffff;
2851 sn_tab
[found
].vm_state_size
= inode
.vm_state_size
;
2852 sn_tab
[found
].vm_clock_nsec
= inode
.vm_clock_nsec
;
2854 snprintf(sn_tab
[found
].id_str
, sizeof(sn_tab
[found
].id_str
),
2855 "%" PRIu32
, inode
.snap_id
);
2856 pstrcpy(sn_tab
[found
].name
,
2857 MIN(sizeof(sn_tab
[found
].name
), sizeof(inode
.tag
)),
2876 static int do_load_save_vmstate(BDRVSheepdogState
*s
, uint8_t *data
,
2877 int64_t pos
, int size
, int load
)
2879 Error
*local_err
= NULL
;
2881 int fd
, ret
= 0, remaining
= size
;
2882 unsigned int data_len
;
2883 uint64_t vmstate_oid
;
2886 uint32_t vdi_id
= load
? s
->inode
.parent_vdi_id
: s
->inode
.vdi_id
;
2887 uint32_t object_size
= (UINT32_C(1) << s
->inode
.block_size_shift
);
2889 fd
= connect_to_sdog(s
, &local_err
);
2891 error_report_err(local_err
);
2896 vdi_index
= pos
/ object_size
;
2897 offset
= pos
% object_size
;
2899 data_len
= MIN(remaining
, object_size
- offset
);
2901 vmstate_oid
= vid_to_vmstate_oid(vdi_id
, vdi_index
);
2903 create
= (offset
== 0);
2905 ret
= read_object(fd
, s
->bs
, (char *)data
, vmstate_oid
,
2906 s
->inode
.nr_copies
, data_len
, offset
,
2909 ret
= write_object(fd
, s
->bs
, (char *)data
, vmstate_oid
,
2910 s
->inode
.nr_copies
, data_len
, offset
, create
,
2915 error_report("failed to save vmstate %s", strerror(errno
));
2921 remaining
-= data_len
;
2929 static int sd_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2932 BDRVSheepdogState
*s
= bs
->opaque
;
2936 buf
= qemu_blockalign(bs
, qiov
->size
);
2937 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
2938 ret
= do_load_save_vmstate(s
, (uint8_t *) buf
, pos
, qiov
->size
, 0);
2944 static int sd_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2947 BDRVSheepdogState
*s
= bs
->opaque
;
2951 buf
= qemu_blockalign(bs
, qiov
->size
);
2952 ret
= do_load_save_vmstate(s
, buf
, pos
, qiov
->size
, 1);
2953 qemu_iovec_from_buf(qiov
, 0, buf
, qiov
->size
);
2960 static coroutine_fn
int sd_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
2964 BDRVSheepdogState
*s
= bs
->opaque
;
2965 QEMUIOVector discard_iov
;
2969 if (!s
->discard_supported
) {
2973 memset(&discard_iov
, 0, sizeof(discard_iov
));
2974 memset(&iov
, 0, sizeof(iov
));
2975 iov
.iov_base
= &zero
;
2976 iov
.iov_len
= sizeof(zero
);
2977 discard_iov
.iov
= &iov
;
2978 discard_iov
.niov
= 1;
2979 if (!QEMU_IS_ALIGNED(offset
| bytes
, BDRV_SECTOR_SIZE
)) {
2982 sd_aio_setup(&acb
, s
, &discard_iov
, offset
>> BDRV_SECTOR_BITS
,
2983 bytes
>> BDRV_SECTOR_BITS
, AIOCB_DISCARD_OBJ
);
2984 sd_co_rw_vector(&acb
);
2985 sd_aio_complete(&acb
);
2990 static coroutine_fn
int64_t
2991 sd_co_get_block_status(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2992 int *pnum
, BlockDriverState
**file
)
2994 BDRVSheepdogState
*s
= bs
->opaque
;
2995 SheepdogInode
*inode
= &s
->inode
;
2996 uint32_t object_size
= (UINT32_C(1) << inode
->block_size_shift
);
2997 uint64_t offset
= sector_num
* BDRV_SECTOR_SIZE
;
2998 unsigned long start
= offset
/ object_size
,
2999 end
= DIV_ROUND_UP((sector_num
+ nb_sectors
) *
3000 BDRV_SECTOR_SIZE
, object_size
);
3002 int64_t ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| offset
;
3004 for (idx
= start
; idx
< end
; idx
++) {
3005 if (inode
->data_vdi_id
[idx
] == 0) {
3010 /* Get the longest length of unallocated sectors */
3012 for (idx
= start
+ 1; idx
< end
; idx
++) {
3013 if (inode
->data_vdi_id
[idx
] != 0) {
3019 *pnum
= (idx
- start
) * object_size
/ BDRV_SECTOR_SIZE
;
3020 if (*pnum
> nb_sectors
) {
3023 if (ret
> 0 && ret
& BDRV_BLOCK_OFFSET_VALID
) {
3029 static int64_t sd_get_allocated_file_size(BlockDriverState
*bs
)
3031 BDRVSheepdogState
*s
= bs
->opaque
;
3032 SheepdogInode
*inode
= &s
->inode
;
3033 uint32_t object_size
= (UINT32_C(1) << inode
->block_size_shift
);
3034 unsigned long i
, last
= DIV_ROUND_UP(inode
->vdi_size
, object_size
);
3037 for (i
= 0; i
< last
; i
++) {
3038 if (inode
->data_vdi_id
[i
] == 0) {
3041 size
+= object_size
;
3046 static QemuOptsList sd_create_opts
= {
3047 .name
= "sheepdog-create-opts",
3048 .head
= QTAILQ_HEAD_INITIALIZER(sd_create_opts
.head
),
3051 .name
= BLOCK_OPT_SIZE
,
3052 .type
= QEMU_OPT_SIZE
,
3053 .help
= "Virtual disk size"
3056 .name
= BLOCK_OPT_BACKING_FILE
,
3057 .type
= QEMU_OPT_STRING
,
3058 .help
= "File name of a base image"
3061 .name
= BLOCK_OPT_PREALLOC
,
3062 .type
= QEMU_OPT_STRING
,
3063 .help
= "Preallocation mode (allowed values: off, full)"
3066 .name
= BLOCK_OPT_REDUNDANCY
,
3067 .type
= QEMU_OPT_STRING
,
3068 .help
= "Redundancy of the image"
3071 .name
= BLOCK_OPT_OBJECT_SIZE
,
3072 .type
= QEMU_OPT_SIZE
,
3073 .help
= "Object size of the image"
3075 { /* end of list */ }
3079 static BlockDriver bdrv_sheepdog
= {
3080 .format_name
= "sheepdog",
3081 .protocol_name
= "sheepdog",
3082 .instance_size
= sizeof(BDRVSheepdogState
),
3083 .bdrv_parse_filename
= sd_parse_filename
,
3084 .bdrv_file_open
= sd_open
,
3085 .bdrv_reopen_prepare
= sd_reopen_prepare
,
3086 .bdrv_reopen_commit
= sd_reopen_commit
,
3087 .bdrv_reopen_abort
= sd_reopen_abort
,
3088 .bdrv_close
= sd_close
,
3089 .bdrv_create
= sd_create
,
3090 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3091 .bdrv_getlength
= sd_getlength
,
3092 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
3093 .bdrv_truncate
= sd_truncate
,
3095 .bdrv_co_readv
= sd_co_readv
,
3096 .bdrv_co_writev
= sd_co_writev
,
3097 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
3098 .bdrv_co_pdiscard
= sd_co_pdiscard
,
3099 .bdrv_co_get_block_status
= sd_co_get_block_status
,
3101 .bdrv_snapshot_create
= sd_snapshot_create
,
3102 .bdrv_snapshot_goto
= sd_snapshot_goto
,
3103 .bdrv_snapshot_delete
= sd_snapshot_delete
,
3104 .bdrv_snapshot_list
= sd_snapshot_list
,
3106 .bdrv_save_vmstate
= sd_save_vmstate
,
3107 .bdrv_load_vmstate
= sd_load_vmstate
,
3109 .bdrv_detach_aio_context
= sd_detach_aio_context
,
3110 .bdrv_attach_aio_context
= sd_attach_aio_context
,
3112 .create_opts
= &sd_create_opts
,
3115 static BlockDriver bdrv_sheepdog_tcp
= {
3116 .format_name
= "sheepdog",
3117 .protocol_name
= "sheepdog+tcp",
3118 .instance_size
= sizeof(BDRVSheepdogState
),
3119 .bdrv_parse_filename
= sd_parse_filename
,
3120 .bdrv_file_open
= sd_open
,
3121 .bdrv_reopen_prepare
= sd_reopen_prepare
,
3122 .bdrv_reopen_commit
= sd_reopen_commit
,
3123 .bdrv_reopen_abort
= sd_reopen_abort
,
3124 .bdrv_close
= sd_close
,
3125 .bdrv_create
= sd_create
,
3126 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3127 .bdrv_getlength
= sd_getlength
,
3128 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
3129 .bdrv_truncate
= sd_truncate
,
3131 .bdrv_co_readv
= sd_co_readv
,
3132 .bdrv_co_writev
= sd_co_writev
,
3133 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
3134 .bdrv_co_pdiscard
= sd_co_pdiscard
,
3135 .bdrv_co_get_block_status
= sd_co_get_block_status
,
3137 .bdrv_snapshot_create
= sd_snapshot_create
,
3138 .bdrv_snapshot_goto
= sd_snapshot_goto
,
3139 .bdrv_snapshot_delete
= sd_snapshot_delete
,
3140 .bdrv_snapshot_list
= sd_snapshot_list
,
3142 .bdrv_save_vmstate
= sd_save_vmstate
,
3143 .bdrv_load_vmstate
= sd_load_vmstate
,
3145 .bdrv_detach_aio_context
= sd_detach_aio_context
,
3146 .bdrv_attach_aio_context
= sd_attach_aio_context
,
3148 .create_opts
= &sd_create_opts
,
3151 static BlockDriver bdrv_sheepdog_unix
= {
3152 .format_name
= "sheepdog",
3153 .protocol_name
= "sheepdog+unix",
3154 .instance_size
= sizeof(BDRVSheepdogState
),
3155 .bdrv_parse_filename
= sd_parse_filename
,
3156 .bdrv_file_open
= sd_open
,
3157 .bdrv_reopen_prepare
= sd_reopen_prepare
,
3158 .bdrv_reopen_commit
= sd_reopen_commit
,
3159 .bdrv_reopen_abort
= sd_reopen_abort
,
3160 .bdrv_close
= sd_close
,
3161 .bdrv_create
= sd_create
,
3162 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3163 .bdrv_getlength
= sd_getlength
,
3164 .bdrv_get_allocated_file_size
= sd_get_allocated_file_size
,
3165 .bdrv_truncate
= sd_truncate
,
3167 .bdrv_co_readv
= sd_co_readv
,
3168 .bdrv_co_writev
= sd_co_writev
,
3169 .bdrv_co_flush_to_disk
= sd_co_flush_to_disk
,
3170 .bdrv_co_pdiscard
= sd_co_pdiscard
,
3171 .bdrv_co_get_block_status
= sd_co_get_block_status
,
3173 .bdrv_snapshot_create
= sd_snapshot_create
,
3174 .bdrv_snapshot_goto
= sd_snapshot_goto
,
3175 .bdrv_snapshot_delete
= sd_snapshot_delete
,
3176 .bdrv_snapshot_list
= sd_snapshot_list
,
3178 .bdrv_save_vmstate
= sd_save_vmstate
,
3179 .bdrv_load_vmstate
= sd_load_vmstate
,
3181 .bdrv_detach_aio_context
= sd_detach_aio_context
,
3182 .bdrv_attach_aio_context
= sd_attach_aio_context
,
3184 .create_opts
= &sd_create_opts
,
3187 static void bdrv_sheepdog_init(void)
3189 bdrv_register(&bdrv_sheepdog
);
3190 bdrv_register(&bdrv_sheepdog_tcp
);
3191 bdrv_register(&bdrv_sheepdog_unix
);
3193 block_init(bdrv_sheepdog_init
);