sheepdog: Mark sd_snapshot_delete() lossage FIXME
[qemu.git] / block / sheepdog.c
1 /*
2 * Copyright (C) 2009-2010 Nippon Telegraph and Telephone Corporation.
3 *
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.
7 *
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/>.
10 *
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.
13 */
14
15 #include "qemu/osdep.h"
16 #include "qapi/error.h"
17 #include "qemu/uri.h"
18 #include "qemu/error-report.h"
19 #include "qemu/sockets.h"
20 #include "block/block_int.h"
21 #include "sysemu/block-backend.h"
22 #include "qemu/bitops.h"
23 #include "qemu/cutils.h"
24
25 #define SD_PROTO_VER 0x01
26
27 #define SD_DEFAULT_ADDR "localhost"
28 #define SD_DEFAULT_PORT 7000
29
30 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
31 #define SD_OP_READ_OBJ 0x02
32 #define SD_OP_WRITE_OBJ 0x03
33 /* 0x04 is used internally by Sheepdog */
34
35 #define SD_OP_NEW_VDI 0x11
36 #define SD_OP_LOCK_VDI 0x12
37 #define SD_OP_RELEASE_VDI 0x13
38 #define SD_OP_GET_VDI_INFO 0x14
39 #define SD_OP_READ_VDIS 0x15
40 #define SD_OP_FLUSH_VDI 0x16
41 #define SD_OP_DEL_VDI 0x17
42 #define SD_OP_GET_CLUSTER_DEFAULT 0x18
43
44 #define SD_FLAG_CMD_WRITE 0x01
45 #define SD_FLAG_CMD_COW 0x02
46 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
47 #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
48
49 #define SD_RES_SUCCESS 0x00 /* Success */
50 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
51 #define SD_RES_NO_OBJ 0x02 /* No object found */
52 #define SD_RES_EIO 0x03 /* I/O error */
53 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
54 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
55 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
56 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
57 #define SD_RES_NO_VDI 0x08 /* No vdi found */
58 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
59 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
60 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
61 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
62 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
63 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
64 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
65 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
66 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
67 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
68 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
69 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
70 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
71 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
72 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
73 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
74 #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */
75 #define SD_RES_READONLY 0x1A /* Object is read-only */
76
77 /*
78 * Object ID rules
79 *
80 * 0 - 19 (20 bits): data object space
81 * 20 - 31 (12 bits): reserved data object space
82 * 32 - 55 (24 bits): vdi object space
83 * 56 - 59 ( 4 bits): reserved vdi object space
84 * 60 - 63 ( 4 bits): object type identifier space
85 */
86
87 #define VDI_SPACE_SHIFT 32
88 #define VDI_BIT (UINT64_C(1) << 63)
89 #define VMSTATE_BIT (UINT64_C(1) << 62)
90 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
91 #define MAX_CHILDREN 1024
92 #define SD_MAX_VDI_LEN 256
93 #define SD_MAX_VDI_TAG_LEN 256
94 #define SD_NR_VDIS (1U << 24)
95 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
96 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
97 #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22
98 /*
99 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
100 * (SD_EC_MAX_STRIP - 1) for parity strips
101 *
102 * SD_MAX_COPIES is sum of number of data strips and parity strips.
103 */
104 #define SD_EC_MAX_STRIP 16
105 #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
106
107 #define SD_INODE_SIZE (sizeof(SheepdogInode))
108 #define CURRENT_VDI_ID 0
109
110 #define LOCK_TYPE_NORMAL 0
111 #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */
112
113 typedef struct SheepdogReq {
114 uint8_t proto_ver;
115 uint8_t opcode;
116 uint16_t flags;
117 uint32_t epoch;
118 uint32_t id;
119 uint32_t data_length;
120 uint32_t opcode_specific[8];
121 } SheepdogReq;
122
123 typedef struct SheepdogRsp {
124 uint8_t proto_ver;
125 uint8_t opcode;
126 uint16_t flags;
127 uint32_t epoch;
128 uint32_t id;
129 uint32_t data_length;
130 uint32_t result;
131 uint32_t opcode_specific[7];
132 } SheepdogRsp;
133
134 typedef struct SheepdogObjReq {
135 uint8_t proto_ver;
136 uint8_t opcode;
137 uint16_t flags;
138 uint32_t epoch;
139 uint32_t id;
140 uint32_t data_length;
141 uint64_t oid;
142 uint64_t cow_oid;
143 uint8_t copies;
144 uint8_t copy_policy;
145 uint8_t reserved[6];
146 uint64_t offset;
147 } SheepdogObjReq;
148
149 typedef struct SheepdogObjRsp {
150 uint8_t proto_ver;
151 uint8_t opcode;
152 uint16_t flags;
153 uint32_t epoch;
154 uint32_t id;
155 uint32_t data_length;
156 uint32_t result;
157 uint8_t copies;
158 uint8_t copy_policy;
159 uint8_t reserved[2];
160 uint32_t pad[6];
161 } SheepdogObjRsp;
162
163 typedef struct SheepdogVdiReq {
164 uint8_t proto_ver;
165 uint8_t opcode;
166 uint16_t flags;
167 uint32_t epoch;
168 uint32_t id;
169 uint32_t data_length;
170 uint64_t vdi_size;
171 uint32_t base_vdi_id;
172 uint8_t copies;
173 uint8_t copy_policy;
174 uint8_t store_policy;
175 uint8_t block_size_shift;
176 uint32_t snapid;
177 uint32_t type;
178 uint32_t pad[2];
179 } SheepdogVdiReq;
180
181 typedef struct SheepdogVdiRsp {
182 uint8_t proto_ver;
183 uint8_t opcode;
184 uint16_t flags;
185 uint32_t epoch;
186 uint32_t id;
187 uint32_t data_length;
188 uint32_t result;
189 uint32_t rsvd;
190 uint32_t vdi_id;
191 uint32_t pad[5];
192 } SheepdogVdiRsp;
193
194 typedef struct SheepdogClusterRsp {
195 uint8_t proto_ver;
196 uint8_t opcode;
197 uint16_t flags;
198 uint32_t epoch;
199 uint32_t id;
200 uint32_t data_length;
201 uint32_t result;
202 uint8_t nr_copies;
203 uint8_t copy_policy;
204 uint8_t block_size_shift;
205 uint8_t __pad1;
206 uint32_t __pad2[6];
207 } SheepdogClusterRsp;
208
209 typedef struct SheepdogInode {
210 char name[SD_MAX_VDI_LEN];
211 char tag[SD_MAX_VDI_TAG_LEN];
212 uint64_t ctime;
213 uint64_t snap_ctime;
214 uint64_t vm_clock_nsec;
215 uint64_t vdi_size;
216 uint64_t vm_state_size;
217 uint16_t copy_policy;
218 uint8_t nr_copies;
219 uint8_t block_size_shift;
220 uint32_t snap_id;
221 uint32_t vdi_id;
222 uint32_t parent_vdi_id;
223 uint32_t child_vdi_id[MAX_CHILDREN];
224 uint32_t data_vdi_id[MAX_DATA_OBJS];
225 } SheepdogInode;
226
227 #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
228
229 /*
230 * 64 bit FNV-1a non-zero initial basis
231 */
232 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
233
234 /*
235 * 64 bit Fowler/Noll/Vo FNV-1a hash code
236 */
237 static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
238 {
239 unsigned char *bp = buf;
240 unsigned char *be = bp + len;
241 while (bp < be) {
242 hval ^= (uint64_t) *bp++;
243 hval += (hval << 1) + (hval << 4) + (hval << 5) +
244 (hval << 7) + (hval << 8) + (hval << 40);
245 }
246 return hval;
247 }
248
249 static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
250 {
251 return inode->vdi_id == inode->data_vdi_id[idx];
252 }
253
254 static inline bool is_data_obj(uint64_t oid)
255 {
256 return !(VDI_BIT & oid);
257 }
258
259 static inline uint64_t data_oid_to_idx(uint64_t oid)
260 {
261 return oid & (MAX_DATA_OBJS - 1);
262 }
263
264 static inline uint32_t oid_to_vid(uint64_t oid)
265 {
266 return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT;
267 }
268
269 static inline uint64_t vid_to_vdi_oid(uint32_t vid)
270 {
271 return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
272 }
273
274 static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx)
275 {
276 return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
277 }
278
279 static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx)
280 {
281 return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
282 }
283
284 static inline bool is_snapshot(struct SheepdogInode *inode)
285 {
286 return !!inode->snap_ctime;
287 }
288
289 static inline size_t count_data_objs(const struct SheepdogInode *inode)
290 {
291 return DIV_ROUND_UP(inode->vdi_size,
292 (1UL << inode->block_size_shift));
293 }
294
295 #undef DPRINTF
296 #ifdef DEBUG_SDOG
297 #define DEBUG_SDOG_PRINT 1
298 #else
299 #define DEBUG_SDOG_PRINT 0
300 #endif
301 #define DPRINTF(fmt, args...) \
302 do { \
303 if (DEBUG_SDOG_PRINT) { \
304 fprintf(stderr, "%s %d: " fmt, __func__, __LINE__, ##args); \
305 } \
306 } while (0)
307
308 typedef struct SheepdogAIOCB SheepdogAIOCB;
309 typedef struct BDRVSheepdogState BDRVSheepdogState;
310
311 typedef struct AIOReq {
312 SheepdogAIOCB *aiocb;
313 unsigned int iov_offset;
314
315 uint64_t oid;
316 uint64_t base_oid;
317 uint64_t offset;
318 unsigned int data_len;
319 uint8_t flags;
320 uint32_t id;
321 bool create;
322
323 QLIST_ENTRY(AIOReq) aio_siblings;
324 } AIOReq;
325
326 enum AIOCBState {
327 AIOCB_WRITE_UDATA,
328 AIOCB_READ_UDATA,
329 AIOCB_FLUSH_CACHE,
330 AIOCB_DISCARD_OBJ,
331 };
332
333 #define AIOCBOverlapping(x, y) \
334 (!(x->max_affect_data_idx < y->min_affect_data_idx \
335 || y->max_affect_data_idx < x->min_affect_data_idx))
336
337 struct SheepdogAIOCB {
338 BDRVSheepdogState *s;
339
340 QEMUIOVector *qiov;
341
342 int64_t sector_num;
343 int nb_sectors;
344
345 int ret;
346 enum AIOCBState aiocb_type;
347
348 Coroutine *coroutine;
349 int nr_pending;
350
351 uint32_t min_affect_data_idx;
352 uint32_t max_affect_data_idx;
353
354 /*
355 * The difference between affect_data_idx and dirty_data_idx:
356 * affect_data_idx represents range of index of all request types.
357 * dirty_data_idx represents range of index updated by COW requests.
358 * dirty_data_idx is used for updating an inode object.
359 */
360 uint32_t min_dirty_data_idx;
361 uint32_t max_dirty_data_idx;
362
363 QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings;
364 };
365
366 struct BDRVSheepdogState {
367 BlockDriverState *bs;
368 AioContext *aio_context;
369
370 SheepdogInode inode;
371
372 char name[SD_MAX_VDI_LEN];
373 bool is_snapshot;
374 uint32_t cache_flags;
375 bool discard_supported;
376
377 char *host_spec;
378 bool is_unix;
379 int fd;
380
381 CoMutex lock;
382 Coroutine *co_send;
383 Coroutine *co_recv;
384
385 uint32_t aioreq_seq_num;
386
387 /* Every aio request must be linked to either of these queues. */
388 QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head;
389 QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head;
390
391 CoQueue overlapping_queue;
392 QLIST_HEAD(inflight_aiocb_head, SheepdogAIOCB) inflight_aiocb_head;
393 };
394
395 typedef struct BDRVSheepdogReopenState {
396 int fd;
397 int cache_flags;
398 } BDRVSheepdogReopenState;
399
400 static const char * sd_strerror(int err)
401 {
402 int i;
403
404 static const struct {
405 int err;
406 const char *desc;
407 } errors[] = {
408 {SD_RES_SUCCESS, "Success"},
409 {SD_RES_UNKNOWN, "Unknown error"},
410 {SD_RES_NO_OBJ, "No object found"},
411 {SD_RES_EIO, "I/O error"},
412 {SD_RES_VDI_EXIST, "VDI exists already"},
413 {SD_RES_INVALID_PARMS, "Invalid parameters"},
414 {SD_RES_SYSTEM_ERROR, "System error"},
415 {SD_RES_VDI_LOCKED, "VDI is already locked"},
416 {SD_RES_NO_VDI, "No vdi found"},
417 {SD_RES_NO_BASE_VDI, "No base VDI found"},
418 {SD_RES_VDI_READ, "Failed read the requested VDI"},
419 {SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
420 {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
421 {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
422 {SD_RES_NO_TAG, "Failed to find the requested tag"},
423 {SD_RES_STARTUP, "The system is still booting"},
424 {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
425 {SD_RES_SHUTDOWN, "The system is shutting down"},
426 {SD_RES_NO_MEM, "Out of memory on the server"},
427 {SD_RES_FULL_VDI, "We already have the maximum vdis"},
428 {SD_RES_VER_MISMATCH, "Protocol version mismatch"},
429 {SD_RES_NO_SPACE, "Server has no space for new objects"},
430 {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
431 {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
432 {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
433 {SD_RES_HALT, "Sheepdog is stopped serving IO request"},
434 {SD_RES_READONLY, "Object is read-only"},
435 };
436
437 for (i = 0; i < ARRAY_SIZE(errors); ++i) {
438 if (errors[i].err == err) {
439 return errors[i].desc;
440 }
441 }
442
443 return "Invalid error code";
444 }
445
446 /*
447 * Sheepdog I/O handling:
448 *
449 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
450 * link the requests to the inflight_list in the
451 * BDRVSheepdogState. The function yields while waiting for
452 * receiving the response.
453 *
454 * 2. We receive the response in aio_read_response, the fd handler to
455 * the sheepdog connection. We switch back to sd_co_readv/sd_writev
456 * after all the requests belonging to the AIOCB are finished. If
457 * needed, sd_co_writev will send another requests for the vdi object.
458 */
459
460 static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
461 uint64_t oid, unsigned int data_len,
462 uint64_t offset, uint8_t flags, bool create,
463 uint64_t base_oid, unsigned int iov_offset)
464 {
465 AIOReq *aio_req;
466
467 aio_req = g_malloc(sizeof(*aio_req));
468 aio_req->aiocb = acb;
469 aio_req->iov_offset = iov_offset;
470 aio_req->oid = oid;
471 aio_req->base_oid = base_oid;
472 aio_req->offset = offset;
473 aio_req->data_len = data_len;
474 aio_req->flags = flags;
475 aio_req->id = s->aioreq_seq_num++;
476 aio_req->create = create;
477
478 acb->nr_pending++;
479 return aio_req;
480 }
481
482 static void wait_for_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *acb)
483 {
484 SheepdogAIOCB *cb;
485
486 retry:
487 QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) {
488 if (AIOCBOverlapping(acb, cb)) {
489 qemu_co_queue_wait(&s->overlapping_queue, NULL);
490 goto retry;
491 }
492 }
493 }
494
495 static void sd_aio_setup(SheepdogAIOCB *acb, BDRVSheepdogState *s,
496 QEMUIOVector *qiov, int64_t sector_num, int nb_sectors,
497 int type)
498 {
499 uint32_t object_size;
500
501 object_size = (UINT32_C(1) << s->inode.block_size_shift);
502
503 acb->s = s;
504
505 acb->qiov = qiov;
506
507 acb->sector_num = sector_num;
508 acb->nb_sectors = nb_sectors;
509
510 acb->coroutine = qemu_coroutine_self();
511 acb->ret = 0;
512 acb->nr_pending = 0;
513
514 acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size;
515 acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE +
516 acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size;
517
518 acb->min_dirty_data_idx = UINT32_MAX;
519 acb->max_dirty_data_idx = 0;
520 acb->aiocb_type = type;
521
522 if (type == AIOCB_FLUSH_CACHE) {
523 return;
524 }
525
526 wait_for_overlapping_aiocb(s, acb);
527 QLIST_INSERT_HEAD(&s->inflight_aiocb_head, acb, aiocb_siblings);
528 }
529
530 /* Return -EIO in case of error, file descriptor on success */
531 static int connect_to_sdog(BDRVSheepdogState *s, Error **errp)
532 {
533 int fd;
534
535 if (s->is_unix) {
536 fd = unix_connect(s->host_spec, errp);
537 } else {
538 fd = inet_connect(s->host_spec, errp);
539
540 if (fd >= 0) {
541 int ret = socket_set_nodelay(fd);
542 if (ret < 0) {
543 error_report("%s", strerror(errno));
544 }
545 }
546 }
547
548 if (fd >= 0) {
549 qemu_set_nonblock(fd);
550 } else {
551 fd = -EIO;
552 }
553
554 return fd;
555 }
556
557 /* Return 0 on success and -errno in case of error */
558 static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
559 unsigned int *wlen)
560 {
561 int ret;
562
563 ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
564 if (ret != sizeof(*hdr)) {
565 error_report("failed to send a req, %s", strerror(errno));
566 return -errno;
567 }
568
569 ret = qemu_co_send(sockfd, data, *wlen);
570 if (ret != *wlen) {
571 error_report("failed to send a req, %s", strerror(errno));
572 return -errno;
573 }
574
575 return ret;
576 }
577
578 typedef struct SheepdogReqCo {
579 int sockfd;
580 BlockDriverState *bs;
581 AioContext *aio_context;
582 SheepdogReq *hdr;
583 void *data;
584 unsigned int *wlen;
585 unsigned int *rlen;
586 int ret;
587 bool finished;
588 Coroutine *co;
589 } SheepdogReqCo;
590
591 static void restart_co_req(void *opaque)
592 {
593 SheepdogReqCo *srco = opaque;
594
595 aio_co_wake(srco->co);
596 }
597
598 static coroutine_fn void do_co_req(void *opaque)
599 {
600 int ret;
601 SheepdogReqCo *srco = opaque;
602 int sockfd = srco->sockfd;
603 SheepdogReq *hdr = srco->hdr;
604 void *data = srco->data;
605 unsigned int *wlen = srco->wlen;
606 unsigned int *rlen = srco->rlen;
607
608 srco->co = qemu_coroutine_self();
609 aio_set_fd_handler(srco->aio_context, sockfd, false,
610 NULL, restart_co_req, NULL, srco);
611
612 ret = send_co_req(sockfd, hdr, data, wlen);
613 if (ret < 0) {
614 goto out;
615 }
616
617 aio_set_fd_handler(srco->aio_context, sockfd, false,
618 restart_co_req, NULL, NULL, srco);
619
620 ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
621 if (ret != sizeof(*hdr)) {
622 error_report("failed to get a rsp, %s", strerror(errno));
623 ret = -errno;
624 goto out;
625 }
626
627 if (*rlen > hdr->data_length) {
628 *rlen = hdr->data_length;
629 }
630
631 if (*rlen) {
632 ret = qemu_co_recv(sockfd, data, *rlen);
633 if (ret != *rlen) {
634 error_report("failed to get the data, %s", strerror(errno));
635 ret = -errno;
636 goto out;
637 }
638 }
639 ret = 0;
640 out:
641 /* there is at most one request for this sockfd, so it is safe to
642 * set each handler to NULL. */
643 aio_set_fd_handler(srco->aio_context, sockfd, false,
644 NULL, NULL, NULL, NULL);
645
646 srco->co = NULL;
647 srco->ret = ret;
648 srco->finished = true;
649 if (srco->bs) {
650 bdrv_wakeup(srco->bs);
651 }
652 }
653
654 /*
655 * Send the request to the sheep in a synchronous manner.
656 *
657 * Return 0 on success, -errno in case of error.
658 */
659 static int do_req(int sockfd, BlockDriverState *bs, SheepdogReq *hdr,
660 void *data, unsigned int *wlen, unsigned int *rlen)
661 {
662 Coroutine *co;
663 SheepdogReqCo srco = {
664 .sockfd = sockfd,
665 .aio_context = bs ? bdrv_get_aio_context(bs) : qemu_get_aio_context(),
666 .bs = bs,
667 .hdr = hdr,
668 .data = data,
669 .wlen = wlen,
670 .rlen = rlen,
671 .ret = 0,
672 .finished = false,
673 };
674
675 if (qemu_in_coroutine()) {
676 do_co_req(&srco);
677 } else {
678 co = qemu_coroutine_create(do_co_req, &srco);
679 if (bs) {
680 qemu_coroutine_enter(co);
681 BDRV_POLL_WHILE(bs, !srco.finished);
682 } else {
683 qemu_coroutine_enter(co);
684 while (!srco.finished) {
685 aio_poll(qemu_get_aio_context(), true);
686 }
687 }
688 }
689
690 return srco.ret;
691 }
692
693 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
694 struct iovec *iov, int niov,
695 enum AIOCBState aiocb_type);
696 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req);
697 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag);
698 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp);
699 static void co_write_request(void *opaque);
700
701 static coroutine_fn void reconnect_to_sdog(void *opaque)
702 {
703 BDRVSheepdogState *s = opaque;
704 AIOReq *aio_req, *next;
705
706 aio_set_fd_handler(s->aio_context, s->fd, false, NULL,
707 NULL, NULL, NULL);
708 close(s->fd);
709 s->fd = -1;
710
711 /* Wait for outstanding write requests to be completed. */
712 while (s->co_send != NULL) {
713 co_write_request(opaque);
714 }
715
716 /* Try to reconnect the sheepdog server every one second. */
717 while (s->fd < 0) {
718 Error *local_err = NULL;
719 s->fd = get_sheep_fd(s, &local_err);
720 if (s->fd < 0) {
721 DPRINTF("Wait for connection to be established\n");
722 error_report_err(local_err);
723 co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME,
724 1000000000ULL);
725 }
726 };
727
728 /*
729 * Now we have to resend all the request in the inflight queue. However,
730 * resend_aioreq() can yield and newly created requests can be added to the
731 * inflight queue before the coroutine is resumed. To avoid mixing them, we
732 * have to move all the inflight requests to the failed queue before
733 * resend_aioreq() is called.
734 */
735 QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) {
736 QLIST_REMOVE(aio_req, aio_siblings);
737 QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings);
738 }
739
740 /* Resend all the failed aio requests. */
741 while (!QLIST_EMPTY(&s->failed_aio_head)) {
742 aio_req = QLIST_FIRST(&s->failed_aio_head);
743 QLIST_REMOVE(aio_req, aio_siblings);
744 resend_aioreq(s, aio_req);
745 }
746 }
747
748 /*
749 * Receive responses of the I/O requests.
750 *
751 * This function is registered as a fd handler, and called from the
752 * main loop when s->fd is ready for reading responses.
753 */
754 static void coroutine_fn aio_read_response(void *opaque)
755 {
756 SheepdogObjRsp rsp;
757 BDRVSheepdogState *s = opaque;
758 int fd = s->fd;
759 int ret;
760 AIOReq *aio_req = NULL;
761 SheepdogAIOCB *acb;
762 uint64_t idx;
763
764 /* read a header */
765 ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
766 if (ret != sizeof(rsp)) {
767 error_report("failed to get the header, %s", strerror(errno));
768 goto err;
769 }
770
771 /* find the right aio_req from the inflight aio list */
772 QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
773 if (aio_req->id == rsp.id) {
774 break;
775 }
776 }
777 if (!aio_req) {
778 error_report("cannot find aio_req %x", rsp.id);
779 goto err;
780 }
781
782 acb = aio_req->aiocb;
783
784 switch (acb->aiocb_type) {
785 case AIOCB_WRITE_UDATA:
786 if (!is_data_obj(aio_req->oid)) {
787 break;
788 }
789 idx = data_oid_to_idx(aio_req->oid);
790
791 if (aio_req->create) {
792 /*
793 * If the object is newly created one, we need to update
794 * the vdi object (metadata object). min_dirty_data_idx
795 * and max_dirty_data_idx are changed to include updated
796 * index between them.
797 */
798 if (rsp.result == SD_RES_SUCCESS) {
799 s->inode.data_vdi_id[idx] = s->inode.vdi_id;
800 acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx);
801 acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx);
802 }
803 }
804 break;
805 case AIOCB_READ_UDATA:
806 ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
807 aio_req->iov_offset, rsp.data_length);
808 if (ret != rsp.data_length) {
809 error_report("failed to get the data, %s", strerror(errno));
810 goto err;
811 }
812 break;
813 case AIOCB_FLUSH_CACHE:
814 if (rsp.result == SD_RES_INVALID_PARMS) {
815 DPRINTF("disable cache since the server doesn't support it\n");
816 s->cache_flags = SD_FLAG_CMD_DIRECT;
817 rsp.result = SD_RES_SUCCESS;
818 }
819 break;
820 case AIOCB_DISCARD_OBJ:
821 switch (rsp.result) {
822 case SD_RES_INVALID_PARMS:
823 error_report("sheep(%s) doesn't support discard command",
824 s->host_spec);
825 rsp.result = SD_RES_SUCCESS;
826 s->discard_supported = false;
827 break;
828 default:
829 break;
830 }
831 }
832
833 /* No more data for this aio_req (reload_inode below uses its own file
834 * descriptor handler which doesn't use co_recv).
835 */
836 s->co_recv = NULL;
837
838 QLIST_REMOVE(aio_req, aio_siblings);
839 switch (rsp.result) {
840 case SD_RES_SUCCESS:
841 break;
842 case SD_RES_READONLY:
843 if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) {
844 ret = reload_inode(s, 0, "");
845 if (ret < 0) {
846 goto err;
847 }
848 }
849 if (is_data_obj(aio_req->oid)) {
850 aio_req->oid = vid_to_data_oid(s->inode.vdi_id,
851 data_oid_to_idx(aio_req->oid));
852 } else {
853 aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id);
854 }
855 resend_aioreq(s, aio_req);
856 return;
857 default:
858 acb->ret = -EIO;
859 error_report("%s", sd_strerror(rsp.result));
860 break;
861 }
862
863 g_free(aio_req);
864
865 if (!--acb->nr_pending) {
866 /*
867 * We've finished all requests which belong to the AIOCB, so
868 * we can switch back to sd_co_readv/writev now.
869 */
870 aio_co_wake(acb->coroutine);
871 }
872
873 return;
874
875 err:
876 reconnect_to_sdog(opaque);
877 }
878
879 static void co_read_response(void *opaque)
880 {
881 BDRVSheepdogState *s = opaque;
882
883 if (!s->co_recv) {
884 s->co_recv = qemu_coroutine_create(aio_read_response, opaque);
885 }
886
887 aio_co_wake(s->co_recv);
888 }
889
890 static void co_write_request(void *opaque)
891 {
892 BDRVSheepdogState *s = opaque;
893
894 aio_co_wake(s->co_send);
895 }
896
897 /*
898 * Return a socket descriptor to read/write objects.
899 *
900 * We cannot use this descriptor for other operations because
901 * the block driver may be on waiting response from the server.
902 */
903 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp)
904 {
905 int fd;
906
907 fd = connect_to_sdog(s, errp);
908 if (fd < 0) {
909 return fd;
910 }
911
912 aio_set_fd_handler(s->aio_context, fd, false,
913 co_read_response, NULL, NULL, s);
914 return fd;
915 }
916
917 static int sd_parse_uri(BDRVSheepdogState *s, const char *filename,
918 char *vdi, uint32_t *snapid, char *tag)
919 {
920 URI *uri;
921 QueryParams *qp = NULL;
922 int ret = 0;
923
924 uri = uri_parse(filename);
925 if (!uri) {
926 return -EINVAL;
927 }
928
929 /* transport */
930 if (!strcmp(uri->scheme, "sheepdog")) {
931 s->is_unix = false;
932 } else if (!strcmp(uri->scheme, "sheepdog+tcp")) {
933 s->is_unix = false;
934 } else if (!strcmp(uri->scheme, "sheepdog+unix")) {
935 s->is_unix = true;
936 } else {
937 ret = -EINVAL;
938 goto out;
939 }
940
941 if (uri->path == NULL || !strcmp(uri->path, "/")) {
942 ret = -EINVAL;
943 goto out;
944 }
945 pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1);
946
947 qp = query_params_parse(uri->query);
948 if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) {
949 ret = -EINVAL;
950 goto out;
951 }
952
953 if (s->is_unix) {
954 /* sheepdog+unix:///vdiname?socket=path */
955 if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
956 ret = -EINVAL;
957 goto out;
958 }
959 s->host_spec = g_strdup(qp->p[0].value);
960 } else {
961 /* sheepdog[+tcp]://[host:port]/vdiname */
962 s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR,
963 uri->port ?: SD_DEFAULT_PORT);
964 }
965
966 /* snapshot tag */
967 if (uri->fragment) {
968 *snapid = strtoul(uri->fragment, NULL, 10);
969 if (*snapid == 0) {
970 pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment);
971 }
972 } else {
973 *snapid = CURRENT_VDI_ID; /* search current vdi */
974 }
975
976 out:
977 if (qp) {
978 query_params_free(qp);
979 }
980 uri_free(uri);
981 return ret;
982 }
983
984 /*
985 * Parse a filename (old syntax)
986 *
987 * filename must be one of the following formats:
988 * 1. [vdiname]
989 * 2. [vdiname]:[snapid]
990 * 3. [vdiname]:[tag]
991 * 4. [hostname]:[port]:[vdiname]
992 * 5. [hostname]:[port]:[vdiname]:[snapid]
993 * 6. [hostname]:[port]:[vdiname]:[tag]
994 *
995 * You can boot from the snapshot images by specifying `snapid` or
996 * `tag'.
997 *
998 * You can run VMs outside the Sheepdog cluster by specifying
999 * `hostname' and `port' (experimental).
1000 */
1001 static int parse_vdiname(BDRVSheepdogState *s, const char *filename,
1002 char *vdi, uint32_t *snapid, char *tag)
1003 {
1004 char *p, *q, *uri;
1005 const char *host_spec, *vdi_spec;
1006 int nr_sep, ret;
1007
1008 strstart(filename, "sheepdog:", &filename);
1009 p = q = g_strdup(filename);
1010
1011 /* count the number of separators */
1012 nr_sep = 0;
1013 while (*p) {
1014 if (*p == ':') {
1015 nr_sep++;
1016 }
1017 p++;
1018 }
1019 p = q;
1020
1021 /* use the first two tokens as host_spec. */
1022 if (nr_sep >= 2) {
1023 host_spec = p;
1024 p = strchr(p, ':');
1025 p++;
1026 p = strchr(p, ':');
1027 *p++ = '\0';
1028 } else {
1029 host_spec = "";
1030 }
1031
1032 vdi_spec = p;
1033
1034 p = strchr(vdi_spec, ':');
1035 if (p) {
1036 *p++ = '#';
1037 }
1038
1039 uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec);
1040
1041 ret = sd_parse_uri(s, uri, vdi, snapid, tag);
1042
1043 g_free(q);
1044 g_free(uri);
1045
1046 return ret;
1047 }
1048
1049 static int find_vdi_name(BDRVSheepdogState *s, const char *filename,
1050 uint32_t snapid, const char *tag, uint32_t *vid,
1051 bool lock, Error **errp)
1052 {
1053 int ret, fd;
1054 SheepdogVdiReq hdr;
1055 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1056 unsigned int wlen, rlen = 0;
1057 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
1058
1059 fd = connect_to_sdog(s, errp);
1060 if (fd < 0) {
1061 return fd;
1062 }
1063
1064 /* This pair of strncpy calls ensures that the buffer is zero-filled,
1065 * which is desirable since we'll soon be sending those bytes, and
1066 * don't want the send_req to read uninitialized data.
1067 */
1068 strncpy(buf, filename, SD_MAX_VDI_LEN);
1069 strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
1070
1071 memset(&hdr, 0, sizeof(hdr));
1072 if (lock) {
1073 hdr.opcode = SD_OP_LOCK_VDI;
1074 hdr.type = LOCK_TYPE_NORMAL;
1075 } else {
1076 hdr.opcode = SD_OP_GET_VDI_INFO;
1077 }
1078 wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
1079 hdr.proto_ver = SD_PROTO_VER;
1080 hdr.data_length = wlen;
1081 hdr.snapid = snapid;
1082 hdr.flags = SD_FLAG_CMD_WRITE;
1083
1084 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1085 if (ret) {
1086 error_setg_errno(errp, -ret, "cannot get vdi info");
1087 goto out;
1088 }
1089
1090 if (rsp->result != SD_RES_SUCCESS) {
1091 error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s",
1092 sd_strerror(rsp->result), filename, snapid, tag);
1093 if (rsp->result == SD_RES_NO_VDI) {
1094 ret = -ENOENT;
1095 } else if (rsp->result == SD_RES_VDI_LOCKED) {
1096 ret = -EBUSY;
1097 } else {
1098 ret = -EIO;
1099 }
1100 goto out;
1101 }
1102 *vid = rsp->vdi_id;
1103
1104 ret = 0;
1105 out:
1106 closesocket(fd);
1107 return ret;
1108 }
1109
1110 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
1111 struct iovec *iov, int niov,
1112 enum AIOCBState aiocb_type)
1113 {
1114 int nr_copies = s->inode.nr_copies;
1115 SheepdogObjReq hdr;
1116 unsigned int wlen = 0;
1117 int ret;
1118 uint64_t oid = aio_req->oid;
1119 unsigned int datalen = aio_req->data_len;
1120 uint64_t offset = aio_req->offset;
1121 uint8_t flags = aio_req->flags;
1122 uint64_t old_oid = aio_req->base_oid;
1123 bool create = aio_req->create;
1124
1125 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1126
1127 if (!nr_copies) {
1128 error_report("bug");
1129 }
1130
1131 memset(&hdr, 0, sizeof(hdr));
1132
1133 switch (aiocb_type) {
1134 case AIOCB_FLUSH_CACHE:
1135 hdr.opcode = SD_OP_FLUSH_VDI;
1136 break;
1137 case AIOCB_READ_UDATA:
1138 hdr.opcode = SD_OP_READ_OBJ;
1139 hdr.flags = flags;
1140 break;
1141 case AIOCB_WRITE_UDATA:
1142 if (create) {
1143 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1144 } else {
1145 hdr.opcode = SD_OP_WRITE_OBJ;
1146 }
1147 wlen = datalen;
1148 hdr.flags = SD_FLAG_CMD_WRITE | flags;
1149 break;
1150 case AIOCB_DISCARD_OBJ:
1151 hdr.opcode = SD_OP_WRITE_OBJ;
1152 hdr.flags = SD_FLAG_CMD_WRITE | flags;
1153 s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0;
1154 offset = offsetof(SheepdogInode,
1155 data_vdi_id[data_oid_to_idx(oid)]);
1156 oid = vid_to_vdi_oid(s->inode.vdi_id);
1157 wlen = datalen = sizeof(uint32_t);
1158 break;
1159 }
1160
1161 if (s->cache_flags) {
1162 hdr.flags |= s->cache_flags;
1163 }
1164
1165 hdr.oid = oid;
1166 hdr.cow_oid = old_oid;
1167 hdr.copies = s->inode.nr_copies;
1168
1169 hdr.data_length = datalen;
1170 hdr.offset = offset;
1171
1172 hdr.id = aio_req->id;
1173
1174 qemu_co_mutex_lock(&s->lock);
1175 s->co_send = qemu_coroutine_self();
1176 aio_set_fd_handler(s->aio_context, s->fd, false,
1177 co_read_response, co_write_request, NULL, s);
1178 socket_set_cork(s->fd, 1);
1179
1180 /* send a header */
1181 ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
1182 if (ret != sizeof(hdr)) {
1183 error_report("failed to send a req, %s", strerror(errno));
1184 goto out;
1185 }
1186
1187 if (wlen) {
1188 ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1189 if (ret != wlen) {
1190 error_report("failed to send a data, %s", strerror(errno));
1191 }
1192 }
1193 out:
1194 socket_set_cork(s->fd, 0);
1195 aio_set_fd_handler(s->aio_context, s->fd, false,
1196 co_read_response, NULL, NULL, s);
1197 s->co_send = NULL;
1198 qemu_co_mutex_unlock(&s->lock);
1199 }
1200
1201 static int read_write_object(int fd, BlockDriverState *bs, char *buf,
1202 uint64_t oid, uint8_t copies,
1203 unsigned int datalen, uint64_t offset,
1204 bool write, bool create, uint32_t cache_flags)
1205 {
1206 SheepdogObjReq hdr;
1207 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1208 unsigned int wlen, rlen;
1209 int ret;
1210
1211 memset(&hdr, 0, sizeof(hdr));
1212
1213 if (write) {
1214 wlen = datalen;
1215 rlen = 0;
1216 hdr.flags = SD_FLAG_CMD_WRITE;
1217 if (create) {
1218 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1219 } else {
1220 hdr.opcode = SD_OP_WRITE_OBJ;
1221 }
1222 } else {
1223 wlen = 0;
1224 rlen = datalen;
1225 hdr.opcode = SD_OP_READ_OBJ;
1226 }
1227
1228 hdr.flags |= cache_flags;
1229
1230 hdr.oid = oid;
1231 hdr.data_length = datalen;
1232 hdr.offset = offset;
1233 hdr.copies = copies;
1234
1235 ret = do_req(fd, bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1236 if (ret) {
1237 error_report("failed to send a request to the sheep");
1238 return ret;
1239 }
1240
1241 switch (rsp->result) {
1242 case SD_RES_SUCCESS:
1243 return 0;
1244 default:
1245 error_report("%s", sd_strerror(rsp->result));
1246 return -EIO;
1247 }
1248 }
1249
1250 static int read_object(int fd, BlockDriverState *bs, char *buf,
1251 uint64_t oid, uint8_t copies,
1252 unsigned int datalen, uint64_t offset,
1253 uint32_t cache_flags)
1254 {
1255 return read_write_object(fd, bs, buf, oid, copies,
1256 datalen, offset, false,
1257 false, cache_flags);
1258 }
1259
1260 static int write_object(int fd, BlockDriverState *bs, char *buf,
1261 uint64_t oid, uint8_t copies,
1262 unsigned int datalen, uint64_t offset, bool create,
1263 uint32_t cache_flags)
1264 {
1265 return read_write_object(fd, bs, buf, oid, copies,
1266 datalen, offset, true,
1267 create, cache_flags);
1268 }
1269
1270 /* update inode with the latest state */
1271 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
1272 {
1273 Error *local_err = NULL;
1274 SheepdogInode *inode;
1275 int ret = 0, fd;
1276 uint32_t vid = 0;
1277
1278 fd = connect_to_sdog(s, &local_err);
1279 if (fd < 0) {
1280 error_report_err(local_err);
1281 return -EIO;
1282 }
1283
1284 inode = g_malloc(SD_INODE_HEADER_SIZE);
1285
1286 ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err);
1287 if (ret) {
1288 error_report_err(local_err);
1289 goto out;
1290 }
1291
1292 ret = read_object(fd, s->bs, (char *)inode, vid_to_vdi_oid(vid),
1293 s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0,
1294 s->cache_flags);
1295 if (ret < 0) {
1296 goto out;
1297 }
1298
1299 if (inode->vdi_id != s->inode.vdi_id) {
1300 memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE);
1301 }
1302
1303 out:
1304 g_free(inode);
1305 closesocket(fd);
1306
1307 return ret;
1308 }
1309
1310 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
1311 {
1312 SheepdogAIOCB *acb = aio_req->aiocb;
1313
1314 aio_req->create = false;
1315
1316 /* check whether this request becomes a CoW one */
1317 if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) {
1318 int idx = data_oid_to_idx(aio_req->oid);
1319
1320 if (is_data_obj_writable(&s->inode, idx)) {
1321 goto out;
1322 }
1323
1324 if (s->inode.data_vdi_id[idx]) {
1325 aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx);
1326 aio_req->flags |= SD_FLAG_CMD_COW;
1327 }
1328 aio_req->create = true;
1329 }
1330 out:
1331 if (is_data_obj(aio_req->oid)) {
1332 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
1333 acb->aiocb_type);
1334 } else {
1335 struct iovec iov;
1336 iov.iov_base = &s->inode;
1337 iov.iov_len = sizeof(s->inode);
1338 add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
1339 }
1340 }
1341
1342 static void sd_detach_aio_context(BlockDriverState *bs)
1343 {
1344 BDRVSheepdogState *s = bs->opaque;
1345
1346 aio_set_fd_handler(s->aio_context, s->fd, false, NULL,
1347 NULL, NULL, NULL);
1348 }
1349
1350 static void sd_attach_aio_context(BlockDriverState *bs,
1351 AioContext *new_context)
1352 {
1353 BDRVSheepdogState *s = bs->opaque;
1354
1355 s->aio_context = new_context;
1356 aio_set_fd_handler(new_context, s->fd, false,
1357 co_read_response, NULL, NULL, s);
1358 }
1359
1360 /* TODO Convert to fine grained options */
1361 static QemuOptsList runtime_opts = {
1362 .name = "sheepdog",
1363 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1364 .desc = {
1365 {
1366 .name = "filename",
1367 .type = QEMU_OPT_STRING,
1368 .help = "URL to the sheepdog image",
1369 },
1370 { /* end of list */ }
1371 },
1372 };
1373
1374 static int sd_open(BlockDriverState *bs, QDict *options, int flags,
1375 Error **errp)
1376 {
1377 int ret, fd;
1378 uint32_t vid = 0;
1379 BDRVSheepdogState *s = bs->opaque;
1380 char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
1381 uint32_t snapid;
1382 char *buf = NULL;
1383 QemuOpts *opts;
1384 Error *local_err = NULL;
1385 const char *filename;
1386
1387 s->bs = bs;
1388 s->aio_context = bdrv_get_aio_context(bs);
1389
1390 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1391 qemu_opts_absorb_qdict(opts, options, &local_err);
1392 if (local_err) {
1393 error_propagate(errp, local_err);
1394 ret = -EINVAL;
1395 goto err_no_fd;
1396 }
1397
1398 filename = qemu_opt_get(opts, "filename");
1399
1400 QLIST_INIT(&s->inflight_aio_head);
1401 QLIST_INIT(&s->failed_aio_head);
1402 QLIST_INIT(&s->inflight_aiocb_head);
1403 s->fd = -1;
1404
1405 memset(vdi, 0, sizeof(vdi));
1406 memset(tag, 0, sizeof(tag));
1407
1408 if (strstr(filename, "://")) {
1409 ret = sd_parse_uri(s, filename, vdi, &snapid, tag);
1410 } else {
1411 ret = parse_vdiname(s, filename, vdi, &snapid, tag);
1412 }
1413 if (ret < 0) {
1414 error_setg(errp, "Can't parse filename");
1415 goto err_no_fd;
1416 }
1417 s->fd = get_sheep_fd(s, errp);
1418 if (s->fd < 0) {
1419 ret = s->fd;
1420 goto err_no_fd;
1421 }
1422
1423 ret = find_vdi_name(s, vdi, snapid, tag, &vid, true, errp);
1424 if (ret) {
1425 goto err;
1426 }
1427
1428 /*
1429 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1430 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1431 */
1432 s->cache_flags = SD_FLAG_CMD_CACHE;
1433 if (flags & BDRV_O_NOCACHE) {
1434 s->cache_flags = SD_FLAG_CMD_DIRECT;
1435 }
1436 s->discard_supported = true;
1437
1438 if (snapid || tag[0] != '\0') {
1439 DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid);
1440 s->is_snapshot = true;
1441 }
1442
1443 fd = connect_to_sdog(s, errp);
1444 if (fd < 0) {
1445 ret = fd;
1446 goto err;
1447 }
1448
1449 buf = g_malloc(SD_INODE_SIZE);
1450 ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid),
1451 0, SD_INODE_SIZE, 0, s->cache_flags);
1452
1453 closesocket(fd);
1454
1455 if (ret) {
1456 error_setg(errp, "Can't read snapshot inode");
1457 goto err;
1458 }
1459
1460 memcpy(&s->inode, buf, sizeof(s->inode));
1461
1462 bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE;
1463 pstrcpy(s->name, sizeof(s->name), vdi);
1464 qemu_co_mutex_init(&s->lock);
1465 qemu_co_queue_init(&s->overlapping_queue);
1466 qemu_opts_del(opts);
1467 g_free(buf);
1468 return 0;
1469
1470 err:
1471 aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd,
1472 false, NULL, NULL, NULL, NULL);
1473 closesocket(s->fd);
1474 err_no_fd:
1475 qemu_opts_del(opts);
1476 g_free(buf);
1477 return ret;
1478 }
1479
1480 static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue,
1481 Error **errp)
1482 {
1483 BDRVSheepdogState *s = state->bs->opaque;
1484 BDRVSheepdogReopenState *re_s;
1485 int ret = 0;
1486
1487 re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1);
1488
1489 re_s->cache_flags = SD_FLAG_CMD_CACHE;
1490 if (state->flags & BDRV_O_NOCACHE) {
1491 re_s->cache_flags = SD_FLAG_CMD_DIRECT;
1492 }
1493
1494 re_s->fd = get_sheep_fd(s, errp);
1495 if (re_s->fd < 0) {
1496 ret = re_s->fd;
1497 return ret;
1498 }
1499
1500 return ret;
1501 }
1502
1503 static void sd_reopen_commit(BDRVReopenState *state)
1504 {
1505 BDRVSheepdogReopenState *re_s = state->opaque;
1506 BDRVSheepdogState *s = state->bs->opaque;
1507
1508 if (s->fd) {
1509 aio_set_fd_handler(s->aio_context, s->fd, false,
1510 NULL, NULL, NULL, NULL);
1511 closesocket(s->fd);
1512 }
1513
1514 s->fd = re_s->fd;
1515 s->cache_flags = re_s->cache_flags;
1516
1517 g_free(state->opaque);
1518 state->opaque = NULL;
1519
1520 return;
1521 }
1522
1523 static void sd_reopen_abort(BDRVReopenState *state)
1524 {
1525 BDRVSheepdogReopenState *re_s = state->opaque;
1526 BDRVSheepdogState *s = state->bs->opaque;
1527
1528 if (re_s == NULL) {
1529 return;
1530 }
1531
1532 if (re_s->fd) {
1533 aio_set_fd_handler(s->aio_context, re_s->fd, false,
1534 NULL, NULL, NULL, NULL);
1535 closesocket(re_s->fd);
1536 }
1537
1538 g_free(state->opaque);
1539 state->opaque = NULL;
1540
1541 return;
1542 }
1543
1544 static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot,
1545 Error **errp)
1546 {
1547 SheepdogVdiReq hdr;
1548 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1549 int fd, ret;
1550 unsigned int wlen, rlen = 0;
1551 char buf[SD_MAX_VDI_LEN];
1552
1553 fd = connect_to_sdog(s, errp);
1554 if (fd < 0) {
1555 return fd;
1556 }
1557
1558 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1559 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1560 */
1561 memset(buf, 0, sizeof(buf));
1562 pstrcpy(buf, sizeof(buf), s->name);
1563
1564 memset(&hdr, 0, sizeof(hdr));
1565 hdr.opcode = SD_OP_NEW_VDI;
1566 hdr.base_vdi_id = s->inode.vdi_id;
1567
1568 wlen = SD_MAX_VDI_LEN;
1569
1570 hdr.flags = SD_FLAG_CMD_WRITE;
1571 hdr.snapid = snapshot;
1572
1573 hdr.data_length = wlen;
1574 hdr.vdi_size = s->inode.vdi_size;
1575 hdr.copy_policy = s->inode.copy_policy;
1576 hdr.copies = s->inode.nr_copies;
1577 hdr.block_size_shift = s->inode.block_size_shift;
1578
1579 ret = do_req(fd, NULL, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1580
1581 closesocket(fd);
1582
1583 if (ret) {
1584 error_setg_errno(errp, -ret, "create failed");
1585 return ret;
1586 }
1587
1588 if (rsp->result != SD_RES_SUCCESS) {
1589 error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name);
1590 return -EIO;
1591 }
1592
1593 if (vdi_id) {
1594 *vdi_id = rsp->vdi_id;
1595 }
1596
1597 return 0;
1598 }
1599
1600 static int sd_prealloc(const char *filename, Error **errp)
1601 {
1602 BlockBackend *blk = NULL;
1603 BDRVSheepdogState *base = NULL;
1604 unsigned long buf_size;
1605 uint32_t idx, max_idx;
1606 uint32_t object_size;
1607 int64_t vdi_size;
1608 void *buf = NULL;
1609 int ret;
1610
1611 blk = blk_new_open(filename, NULL, NULL,
1612 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
1613 if (blk == NULL) {
1614 ret = -EIO;
1615 goto out_with_err_set;
1616 }
1617
1618 blk_set_allow_write_beyond_eof(blk, true);
1619
1620 vdi_size = blk_getlength(blk);
1621 if (vdi_size < 0) {
1622 ret = vdi_size;
1623 goto out;
1624 }
1625
1626 base = blk_bs(blk)->opaque;
1627 object_size = (UINT32_C(1) << base->inode.block_size_shift);
1628 buf_size = MIN(object_size, SD_DATA_OBJ_SIZE);
1629 buf = g_malloc0(buf_size);
1630
1631 max_idx = DIV_ROUND_UP(vdi_size, buf_size);
1632
1633 for (idx = 0; idx < max_idx; idx++) {
1634 /*
1635 * The created image can be a cloned image, so we need to read
1636 * a data from the source image.
1637 */
1638 ret = blk_pread(blk, idx * buf_size, buf, buf_size);
1639 if (ret < 0) {
1640 goto out;
1641 }
1642 ret = blk_pwrite(blk, idx * buf_size, buf, buf_size, 0);
1643 if (ret < 0) {
1644 goto out;
1645 }
1646 }
1647
1648 ret = 0;
1649 out:
1650 if (ret < 0) {
1651 error_setg_errno(errp, -ret, "Can't pre-allocate");
1652 }
1653 out_with_err_set:
1654 if (blk) {
1655 blk_unref(blk);
1656 }
1657 g_free(buf);
1658
1659 return ret;
1660 }
1661
1662 /*
1663 * Sheepdog support two kinds of redundancy, full replication and erasure
1664 * coding.
1665 *
1666 * # create a fully replicated vdi with x copies
1667 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1668 *
1669 * # create a erasure coded vdi with x data strips and y parity strips
1670 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1671 */
1672 static int parse_redundancy(BDRVSheepdogState *s, const char *opt)
1673 {
1674 struct SheepdogInode *inode = &s->inode;
1675 const char *n1, *n2;
1676 long copy, parity;
1677 char p[10];
1678
1679 pstrcpy(p, sizeof(p), opt);
1680 n1 = strtok(p, ":");
1681 n2 = strtok(NULL, ":");
1682
1683 if (!n1) {
1684 return -EINVAL;
1685 }
1686
1687 copy = strtol(n1, NULL, 10);
1688 if (copy > SD_MAX_COPIES || copy < 1) {
1689 return -EINVAL;
1690 }
1691 if (!n2) {
1692 inode->copy_policy = 0;
1693 inode->nr_copies = copy;
1694 return 0;
1695 }
1696
1697 if (copy != 2 && copy != 4 && copy != 8 && copy != 16) {
1698 return -EINVAL;
1699 }
1700
1701 parity = strtol(n2, NULL, 10);
1702 if (parity >= SD_EC_MAX_STRIP || parity < 1) {
1703 return -EINVAL;
1704 }
1705
1706 /*
1707 * 4 bits for parity and 4 bits for data.
1708 * We have to compress upper data bits because it can't represent 16
1709 */
1710 inode->copy_policy = ((copy / 2) << 4) + parity;
1711 inode->nr_copies = copy + parity;
1712
1713 return 0;
1714 }
1715
1716 static int parse_block_size_shift(BDRVSheepdogState *s, QemuOpts *opt)
1717 {
1718 struct SheepdogInode *inode = &s->inode;
1719 uint64_t object_size;
1720 int obj_order;
1721
1722 object_size = qemu_opt_get_size_del(opt, BLOCK_OPT_OBJECT_SIZE, 0);
1723 if (object_size) {
1724 if ((object_size - 1) & object_size) { /* not a power of 2? */
1725 return -EINVAL;
1726 }
1727 obj_order = ctz32(object_size);
1728 if (obj_order < 20 || obj_order > 31) {
1729 return -EINVAL;
1730 }
1731 inode->block_size_shift = (uint8_t)obj_order;
1732 }
1733
1734 return 0;
1735 }
1736
1737 static int sd_create(const char *filename, QemuOpts *opts,
1738 Error **errp)
1739 {
1740 int ret = 0;
1741 uint32_t vid = 0;
1742 char *backing_file = NULL;
1743 char *buf = NULL;
1744 BDRVSheepdogState *s;
1745 char tag[SD_MAX_VDI_TAG_LEN];
1746 uint32_t snapid;
1747 uint64_t max_vdi_size;
1748 bool prealloc = false;
1749
1750 s = g_new0(BDRVSheepdogState, 1);
1751
1752 memset(tag, 0, sizeof(tag));
1753 if (strstr(filename, "://")) {
1754 ret = sd_parse_uri(s, filename, s->name, &snapid, tag);
1755 } else {
1756 ret = parse_vdiname(s, filename, s->name, &snapid, tag);
1757 }
1758 if (ret < 0) {
1759 error_setg(errp, "Can't parse filename");
1760 goto out;
1761 }
1762
1763 s->inode.vdi_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1764 BDRV_SECTOR_SIZE);
1765 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1766 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1767 if (!buf || !strcmp(buf, "off")) {
1768 prealloc = false;
1769 } else if (!strcmp(buf, "full")) {
1770 prealloc = true;
1771 } else {
1772 error_setg(errp, "Invalid preallocation mode: '%s'", buf);
1773 ret = -EINVAL;
1774 goto out;
1775 }
1776
1777 g_free(buf);
1778 buf = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY);
1779 if (buf) {
1780 ret = parse_redundancy(s, buf);
1781 if (ret < 0) {
1782 error_setg(errp, "Invalid redundancy mode: '%s'", buf);
1783 goto out;
1784 }
1785 }
1786 ret = parse_block_size_shift(s, opts);
1787 if (ret < 0) {
1788 error_setg(errp, "Invalid object_size."
1789 " obect_size needs to be power of 2"
1790 " and be limited from 2^20 to 2^31");
1791 goto out;
1792 }
1793
1794 if (backing_file) {
1795 BlockBackend *blk;
1796 BDRVSheepdogState *base;
1797 BlockDriver *drv;
1798
1799 /* Currently, only Sheepdog backing image is supported. */
1800 drv = bdrv_find_protocol(backing_file, true, NULL);
1801 if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
1802 error_setg(errp, "backing_file must be a sheepdog image");
1803 ret = -EINVAL;
1804 goto out;
1805 }
1806
1807 blk = blk_new_open(backing_file, NULL, NULL,
1808 BDRV_O_PROTOCOL, errp);
1809 if (blk == NULL) {
1810 ret = -EIO;
1811 goto out;
1812 }
1813
1814 base = blk_bs(blk)->opaque;
1815
1816 if (!is_snapshot(&base->inode)) {
1817 error_setg(errp, "cannot clone from a non snapshot vdi");
1818 blk_unref(blk);
1819 ret = -EINVAL;
1820 goto out;
1821 }
1822 s->inode.vdi_id = base->inode.vdi_id;
1823 blk_unref(blk);
1824 }
1825
1826 s->aio_context = qemu_get_aio_context();
1827
1828 /* if block_size_shift is not specified, get cluster default value */
1829 if (s->inode.block_size_shift == 0) {
1830 SheepdogVdiReq hdr;
1831 SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr;
1832 int fd;
1833 unsigned int wlen = 0, rlen = 0;
1834
1835 fd = connect_to_sdog(s, errp);
1836 if (fd < 0) {
1837 ret = fd;
1838 goto out;
1839 }
1840
1841 memset(&hdr, 0, sizeof(hdr));
1842 hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT;
1843 hdr.proto_ver = SD_PROTO_VER;
1844
1845 ret = do_req(fd, NULL, (SheepdogReq *)&hdr,
1846 NULL, &wlen, &rlen);
1847 closesocket(fd);
1848 if (ret) {
1849 error_setg_errno(errp, -ret, "failed to get cluster default");
1850 goto out;
1851 }
1852 if (rsp->result == SD_RES_SUCCESS) {
1853 s->inode.block_size_shift = rsp->block_size_shift;
1854 } else {
1855 s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT;
1856 }
1857 }
1858
1859 max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS;
1860
1861 if (s->inode.vdi_size > max_vdi_size) {
1862 error_setg(errp, "An image is too large."
1863 " The maximum image size is %"PRIu64 "GB",
1864 max_vdi_size / 1024 / 1024 / 1024);
1865 ret = -EINVAL;
1866 goto out;
1867 }
1868
1869 ret = do_sd_create(s, &vid, 0, errp);
1870 if (ret) {
1871 goto out;
1872 }
1873
1874 if (prealloc) {
1875 ret = sd_prealloc(filename, errp);
1876 }
1877 out:
1878 g_free(backing_file);
1879 g_free(buf);
1880 g_free(s);
1881 return ret;
1882 }
1883
1884 static void sd_close(BlockDriverState *bs)
1885 {
1886 Error *local_err = NULL;
1887 BDRVSheepdogState *s = bs->opaque;
1888 SheepdogVdiReq hdr;
1889 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1890 unsigned int wlen, rlen = 0;
1891 int fd, ret;
1892
1893 DPRINTF("%s\n", s->name);
1894
1895 fd = connect_to_sdog(s, &local_err);
1896 if (fd < 0) {
1897 error_report_err(local_err);
1898 return;
1899 }
1900
1901 memset(&hdr, 0, sizeof(hdr));
1902
1903 hdr.opcode = SD_OP_RELEASE_VDI;
1904 hdr.type = LOCK_TYPE_NORMAL;
1905 hdr.base_vdi_id = s->inode.vdi_id;
1906 wlen = strlen(s->name) + 1;
1907 hdr.data_length = wlen;
1908 hdr.flags = SD_FLAG_CMD_WRITE;
1909
1910 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
1911 s->name, &wlen, &rlen);
1912
1913 closesocket(fd);
1914
1915 if (!ret && rsp->result != SD_RES_SUCCESS &&
1916 rsp->result != SD_RES_VDI_NOT_LOCKED) {
1917 error_report("%s, %s", sd_strerror(rsp->result), s->name);
1918 }
1919
1920 aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd,
1921 false, NULL, NULL, NULL, NULL);
1922 closesocket(s->fd);
1923 g_free(s->host_spec);
1924 }
1925
1926 static int64_t sd_getlength(BlockDriverState *bs)
1927 {
1928 BDRVSheepdogState *s = bs->opaque;
1929
1930 return s->inode.vdi_size;
1931 }
1932
1933 static int sd_truncate(BlockDriverState *bs, int64_t offset)
1934 {
1935 Error *local_err = NULL;
1936 BDRVSheepdogState *s = bs->opaque;
1937 int ret, fd;
1938 unsigned int datalen;
1939 uint64_t max_vdi_size;
1940
1941 max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS;
1942 if (offset < s->inode.vdi_size) {
1943 error_report("shrinking is not supported");
1944 return -EINVAL;
1945 } else if (offset > max_vdi_size) {
1946 error_report("too big image size");
1947 return -EINVAL;
1948 }
1949
1950 fd = connect_to_sdog(s, &local_err);
1951 if (fd < 0) {
1952 error_report_err(local_err);
1953 return fd;
1954 }
1955
1956 /* we don't need to update entire object */
1957 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
1958 s->inode.vdi_size = offset;
1959 ret = write_object(fd, s->bs, (char *)&s->inode,
1960 vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
1961 datalen, 0, false, s->cache_flags);
1962 close(fd);
1963
1964 if (ret < 0) {
1965 error_report("failed to update an inode.");
1966 }
1967
1968 return ret;
1969 }
1970
1971 /*
1972 * This function is called after writing data objects. If we need to
1973 * update metadata, this sends a write request to the vdi object.
1974 */
1975 static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
1976 {
1977 BDRVSheepdogState *s = acb->s;
1978 struct iovec iov;
1979 AIOReq *aio_req;
1980 uint32_t offset, data_len, mn, mx;
1981
1982 mn = acb->min_dirty_data_idx;
1983 mx = acb->max_dirty_data_idx;
1984 if (mn <= mx) {
1985 /* we need to update the vdi object. */
1986 ++acb->nr_pending;
1987 offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
1988 mn * sizeof(s->inode.data_vdi_id[0]);
1989 data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
1990
1991 acb->min_dirty_data_idx = UINT32_MAX;
1992 acb->max_dirty_data_idx = 0;
1993
1994 iov.iov_base = &s->inode;
1995 iov.iov_len = sizeof(s->inode);
1996 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
1997 data_len, offset, 0, false, 0, offset);
1998 add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
1999 if (--acb->nr_pending) {
2000 qemu_coroutine_yield();
2001 }
2002 }
2003 }
2004
2005 /* Delete current working VDI on the snapshot chain */
2006 static bool sd_delete(BDRVSheepdogState *s)
2007 {
2008 Error *local_err = NULL;
2009 unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0;
2010 SheepdogVdiReq hdr = {
2011 .opcode = SD_OP_DEL_VDI,
2012 .base_vdi_id = s->inode.vdi_id,
2013 .data_length = wlen,
2014 .flags = SD_FLAG_CMD_WRITE,
2015 };
2016 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2017 int fd, ret;
2018
2019 fd = connect_to_sdog(s, &local_err);
2020 if (fd < 0) {
2021 error_report_err(local_err);
2022 return false;
2023 }
2024
2025 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2026 s->name, &wlen, &rlen);
2027 closesocket(fd);
2028 if (ret) {
2029 return false;
2030 }
2031 switch (rsp->result) {
2032 case SD_RES_NO_VDI:
2033 error_report("%s was already deleted", s->name);
2034 /* fall through */
2035 case SD_RES_SUCCESS:
2036 break;
2037 default:
2038 error_report("%s, %s", sd_strerror(rsp->result), s->name);
2039 return false;
2040 }
2041
2042 return true;
2043 }
2044
2045 /*
2046 * Create a writable VDI from a snapshot
2047 */
2048 static int sd_create_branch(BDRVSheepdogState *s)
2049 {
2050 Error *local_err = NULL;
2051 int ret, fd;
2052 uint32_t vid;
2053 char *buf;
2054 bool deleted;
2055
2056 DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id);
2057
2058 buf = g_malloc(SD_INODE_SIZE);
2059
2060 /*
2061 * Even If deletion fails, we will just create extra snapshot based on
2062 * the working VDI which was supposed to be deleted. So no need to
2063 * false bail out.
2064 */
2065 deleted = sd_delete(s);
2066 ret = do_sd_create(s, &vid, !deleted, &local_err);
2067 if (ret) {
2068 error_report_err(local_err);
2069 goto out;
2070 }
2071
2072 DPRINTF("%" PRIx32 " is created.\n", vid);
2073
2074 fd = connect_to_sdog(s, &local_err);
2075 if (fd < 0) {
2076 error_report_err(local_err);
2077 ret = fd;
2078 goto out;
2079 }
2080
2081 ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid),
2082 s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags);
2083
2084 closesocket(fd);
2085
2086 if (ret < 0) {
2087 goto out;
2088 }
2089
2090 memcpy(&s->inode, buf, sizeof(s->inode));
2091
2092 s->is_snapshot = false;
2093 ret = 0;
2094 DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id);
2095
2096 out:
2097 g_free(buf);
2098
2099 return ret;
2100 }
2101
2102 /*
2103 * Send I/O requests to the server.
2104 *
2105 * This function sends requests to the server, links the requests to
2106 * the inflight_list in BDRVSheepdogState, and exits without
2107 * waiting the response. The responses are received in the
2108 * `aio_read_response' function which is called from the main loop as
2109 * a fd handler.
2110 *
2111 * Returns 1 when we need to wait a response, 0 when there is no sent
2112 * request and -errno in error cases.
2113 */
2114 static void coroutine_fn sd_co_rw_vector(SheepdogAIOCB *acb)
2115 {
2116 int ret = 0;
2117 unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE;
2118 unsigned long idx;
2119 uint32_t object_size;
2120 uint64_t oid;
2121 uint64_t offset;
2122 BDRVSheepdogState *s = acb->s;
2123 SheepdogInode *inode = &s->inode;
2124 AIOReq *aio_req;
2125
2126 if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
2127 /*
2128 * In the case we open the snapshot VDI, Sheepdog creates the
2129 * writable VDI when we do a write operation first.
2130 */
2131 ret = sd_create_branch(s);
2132 if (ret) {
2133 acb->ret = -EIO;
2134 return;
2135 }
2136 }
2137
2138 object_size = (UINT32_C(1) << inode->block_size_shift);
2139 idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size;
2140 offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size;
2141
2142 /*
2143 * Make sure we don't free the aiocb before we are done with all requests.
2144 * This additional reference is dropped at the end of this function.
2145 */
2146 acb->nr_pending++;
2147
2148 while (done != total) {
2149 uint8_t flags = 0;
2150 uint64_t old_oid = 0;
2151 bool create = false;
2152
2153 oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
2154
2155 len = MIN(total - done, object_size - offset);
2156
2157 switch (acb->aiocb_type) {
2158 case AIOCB_READ_UDATA:
2159 if (!inode->data_vdi_id[idx]) {
2160 qemu_iovec_memset(acb->qiov, done, 0, len);
2161 goto done;
2162 }
2163 break;
2164 case AIOCB_WRITE_UDATA:
2165 if (!inode->data_vdi_id[idx]) {
2166 create = true;
2167 } else if (!is_data_obj_writable(inode, idx)) {
2168 /* Copy-On-Write */
2169 create = true;
2170 old_oid = oid;
2171 flags = SD_FLAG_CMD_COW;
2172 }
2173 break;
2174 case AIOCB_DISCARD_OBJ:
2175 /*
2176 * We discard the object only when the whole object is
2177 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2178 */
2179 if (len != object_size || inode->data_vdi_id[idx] == 0) {
2180 goto done;
2181 }
2182 break;
2183 default:
2184 break;
2185 }
2186
2187 if (create) {
2188 DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n",
2189 inode->vdi_id, oid,
2190 vid_to_data_oid(inode->data_vdi_id[idx], idx), idx);
2191 oid = vid_to_data_oid(inode->vdi_id, idx);
2192 DPRINTF("new oid %" PRIx64 "\n", oid);
2193 }
2194
2195 aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create,
2196 old_oid,
2197 acb->aiocb_type == AIOCB_DISCARD_OBJ ?
2198 0 : done);
2199 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
2200 acb->aiocb_type);
2201 done:
2202 offset = 0;
2203 idx++;
2204 done += len;
2205 }
2206 if (--acb->nr_pending) {
2207 qemu_coroutine_yield();
2208 }
2209 }
2210
2211 static void sd_aio_complete(SheepdogAIOCB *acb)
2212 {
2213 if (acb->aiocb_type == AIOCB_FLUSH_CACHE) {
2214 return;
2215 }
2216
2217 QLIST_REMOVE(acb, aiocb_siblings);
2218 qemu_co_queue_restart_all(&acb->s->overlapping_queue);
2219 }
2220
2221 static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
2222 int nb_sectors, QEMUIOVector *qiov)
2223 {
2224 SheepdogAIOCB acb;
2225 int ret;
2226 int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE;
2227 BDRVSheepdogState *s = bs->opaque;
2228
2229 if (offset > s->inode.vdi_size) {
2230 ret = sd_truncate(bs, offset);
2231 if (ret < 0) {
2232 return ret;
2233 }
2234 }
2235
2236 sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_WRITE_UDATA);
2237 sd_co_rw_vector(&acb);
2238 sd_write_done(&acb);
2239 sd_aio_complete(&acb);
2240
2241 return acb.ret;
2242 }
2243
2244 static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
2245 int nb_sectors, QEMUIOVector *qiov)
2246 {
2247 SheepdogAIOCB acb;
2248 BDRVSheepdogState *s = bs->opaque;
2249
2250 sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_READ_UDATA);
2251 sd_co_rw_vector(&acb);
2252 sd_aio_complete(&acb);
2253
2254 return acb.ret;
2255 }
2256
2257 static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
2258 {
2259 BDRVSheepdogState *s = bs->opaque;
2260 SheepdogAIOCB acb;
2261 AIOReq *aio_req;
2262
2263 if (s->cache_flags != SD_FLAG_CMD_CACHE) {
2264 return 0;
2265 }
2266
2267 sd_aio_setup(&acb, s, NULL, 0, 0, AIOCB_FLUSH_CACHE);
2268
2269 acb.nr_pending++;
2270 aio_req = alloc_aio_req(s, &acb, vid_to_vdi_oid(s->inode.vdi_id),
2271 0, 0, 0, false, 0, 0);
2272 add_aio_request(s, aio_req, NULL, 0, acb.aiocb_type);
2273
2274 if (--acb.nr_pending) {
2275 qemu_coroutine_yield();
2276 }
2277
2278 sd_aio_complete(&acb);
2279 return acb.ret;
2280 }
2281
2282 static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
2283 {
2284 Error *local_err = NULL;
2285 BDRVSheepdogState *s = bs->opaque;
2286 int ret, fd;
2287 uint32_t new_vid;
2288 SheepdogInode *inode;
2289 unsigned int datalen;
2290
2291 DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " "
2292 "is_snapshot %d\n", sn_info->name, sn_info->id_str,
2293 s->name, sn_info->vm_state_size, s->is_snapshot);
2294
2295 if (s->is_snapshot) {
2296 error_report("You can't create a snapshot of a snapshot VDI, "
2297 "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
2298
2299 return -EINVAL;
2300 }
2301
2302 DPRINTF("%s %s\n", sn_info->name, sn_info->id_str);
2303
2304 s->inode.vm_state_size = sn_info->vm_state_size;
2305 s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
2306 /* It appears that inode.tag does not require a NUL terminator,
2307 * which means this use of strncpy is ok.
2308 */
2309 strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
2310 /* we don't need to update entire object */
2311 datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
2312 inode = g_malloc(datalen);
2313
2314 /* refresh inode. */
2315 fd = connect_to_sdog(s, &local_err);
2316 if (fd < 0) {
2317 error_report_err(local_err);
2318 ret = fd;
2319 goto cleanup;
2320 }
2321
2322 ret = write_object(fd, s->bs, (char *)&s->inode,
2323 vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
2324 datalen, 0, false, s->cache_flags);
2325 if (ret < 0) {
2326 error_report("failed to write snapshot's inode.");
2327 goto cleanup;
2328 }
2329
2330 ret = do_sd_create(s, &new_vid, 1, &local_err);
2331 if (ret < 0) {
2332 error_reportf_err(local_err,
2333 "failed to create inode for snapshot: ");
2334 goto cleanup;
2335 }
2336
2337 ret = read_object(fd, s->bs, (char *)inode,
2338 vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0,
2339 s->cache_flags);
2340
2341 if (ret < 0) {
2342 error_report("failed to read new inode info. %s", strerror(errno));
2343 goto cleanup;
2344 }
2345
2346 memcpy(&s->inode, inode, datalen);
2347 DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2348 s->inode.name, s->inode.snap_id, s->inode.vdi_id);
2349
2350 cleanup:
2351 g_free(inode);
2352 closesocket(fd);
2353 return ret;
2354 }
2355
2356 /*
2357 * We implement rollback(loadvm) operation to the specified snapshot by
2358 * 1) switch to the snapshot
2359 * 2) rely on sd_create_branch to delete working VDI and
2360 * 3) create a new working VDI based on the specified snapshot
2361 */
2362 static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
2363 {
2364 BDRVSheepdogState *s = bs->opaque;
2365 BDRVSheepdogState *old_s;
2366 char tag[SD_MAX_VDI_TAG_LEN];
2367 uint32_t snapid = 0;
2368 int ret = 0;
2369
2370 old_s = g_new(BDRVSheepdogState, 1);
2371
2372 memcpy(old_s, s, sizeof(BDRVSheepdogState));
2373
2374 snapid = strtoul(snapshot_id, NULL, 10);
2375 if (snapid) {
2376 tag[0] = 0;
2377 } else {
2378 pstrcpy(tag, sizeof(tag), snapshot_id);
2379 }
2380
2381 ret = reload_inode(s, snapid, tag);
2382 if (ret) {
2383 goto out;
2384 }
2385
2386 ret = sd_create_branch(s);
2387 if (ret) {
2388 goto out;
2389 }
2390
2391 g_free(old_s);
2392
2393 return 0;
2394 out:
2395 /* recover bdrv_sd_state */
2396 memcpy(s, old_s, sizeof(BDRVSheepdogState));
2397 g_free(old_s);
2398
2399 error_report("failed to open. recover old bdrv_sd_state.");
2400
2401 return ret;
2402 }
2403
2404 #define NR_BATCHED_DISCARD 128
2405
2406 static int remove_objects(BDRVSheepdogState *s, Error **errp)
2407 {
2408 int fd, i = 0, nr_objs = 0;
2409 int ret;
2410 SheepdogInode *inode = &s->inode;
2411
2412 fd = connect_to_sdog(s, errp);
2413 if (fd < 0) {
2414 return fd;
2415 }
2416
2417 nr_objs = count_data_objs(inode);
2418 while (i < nr_objs) {
2419 int start_idx, nr_filled_idx;
2420
2421 while (i < nr_objs && !inode->data_vdi_id[i]) {
2422 i++;
2423 }
2424 start_idx = i;
2425
2426 nr_filled_idx = 0;
2427 while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) {
2428 if (inode->data_vdi_id[i]) {
2429 inode->data_vdi_id[i] = 0;
2430 nr_filled_idx++;
2431 }
2432
2433 i++;
2434 }
2435
2436 ret = write_object(fd, s->bs,
2437 (char *)&inode->data_vdi_id[start_idx],
2438 vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies,
2439 (i - start_idx) * sizeof(uint32_t),
2440 offsetof(struct SheepdogInode,
2441 data_vdi_id[start_idx]),
2442 false, s->cache_flags);
2443 if (ret < 0) {
2444 error_setg(errp, "Failed to discard snapshot inode");
2445 goto out;
2446 }
2447 }
2448
2449 ret = 0;
2450 out:
2451 closesocket(fd);
2452 return ret;
2453 }
2454
2455 static int sd_snapshot_delete(BlockDriverState *bs,
2456 const char *snapshot_id,
2457 const char *name,
2458 Error **errp)
2459 {
2460 /*
2461 * FIXME should delete the snapshot matching both @snapshot_id and
2462 * @name, but @name not used here
2463 */
2464 unsigned long snap_id = 0;
2465 char snap_tag[SD_MAX_VDI_TAG_LEN];
2466 int fd, ret;
2467 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
2468 BDRVSheepdogState *s = bs->opaque;
2469 unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0;
2470 uint32_t vid;
2471 SheepdogVdiReq hdr = {
2472 .opcode = SD_OP_DEL_VDI,
2473 .data_length = wlen,
2474 .flags = SD_FLAG_CMD_WRITE,
2475 };
2476 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2477
2478 ret = remove_objects(s, errp);
2479 if (ret) {
2480 return ret;
2481 }
2482
2483 memset(buf, 0, sizeof(buf));
2484 memset(snap_tag, 0, sizeof(snap_tag));
2485 pstrcpy(buf, SD_MAX_VDI_LEN, s->name);
2486 ret = qemu_strtoul(snapshot_id, NULL, 10, &snap_id);
2487 if (ret || snap_id > UINT32_MAX) {
2488 /*
2489 * FIXME Since qemu_strtoul() returns -EINVAL when
2490 * @snapshot_id is null, @snapshot_id is mandatory. Correct
2491 * would be to require at least one of @snapshot_id and @name.
2492 */
2493 error_setg(errp, "Invalid snapshot ID: %s",
2494 snapshot_id ? snapshot_id : "<null>");
2495 return -EINVAL;
2496 }
2497
2498 if (snap_id) {
2499 hdr.snapid = (uint32_t) snap_id;
2500 } else {
2501 /* FIXME I suspect we should use @name here */
2502 pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id);
2503 pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag);
2504 }
2505
2506 ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, errp);
2507 if (ret) {
2508 return ret;
2509 }
2510
2511 fd = connect_to_sdog(s, errp);
2512 if (fd < 0) {
2513 return fd;
2514 }
2515
2516 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2517 buf, &wlen, &rlen);
2518 closesocket(fd);
2519 if (ret) {
2520 error_setg_errno(errp, -ret, "Couldn't send request to server");
2521 return ret;
2522 }
2523
2524 switch (rsp->result) {
2525 case SD_RES_NO_VDI:
2526 error_setg(errp, "Can't find the snapshot");
2527 return -ENOENT;
2528 case SD_RES_SUCCESS:
2529 break;
2530 default:
2531 error_setg(errp, "%s", sd_strerror(rsp->result));
2532 return -EIO;
2533 }
2534
2535 return 0;
2536 }
2537
2538 static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
2539 {
2540 Error *local_err = NULL;
2541 BDRVSheepdogState *s = bs->opaque;
2542 SheepdogReq req;
2543 int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
2544 QEMUSnapshotInfo *sn_tab = NULL;
2545 unsigned wlen, rlen;
2546 int found = 0;
2547 static SheepdogInode inode;
2548 unsigned long *vdi_inuse;
2549 unsigned int start_nr;
2550 uint64_t hval;
2551 uint32_t vid;
2552
2553 vdi_inuse = g_malloc(max);
2554
2555 fd = connect_to_sdog(s, &local_err);
2556 if (fd < 0) {
2557 error_report_err(local_err);
2558 ret = fd;
2559 goto out;
2560 }
2561
2562 rlen = max;
2563 wlen = 0;
2564
2565 memset(&req, 0, sizeof(req));
2566
2567 req.opcode = SD_OP_READ_VDIS;
2568 req.data_length = max;
2569
2570 ret = do_req(fd, s->bs, &req, vdi_inuse, &wlen, &rlen);
2571
2572 closesocket(fd);
2573 if (ret) {
2574 goto out;
2575 }
2576
2577 sn_tab = g_new0(QEMUSnapshotInfo, nr);
2578
2579 /* calculate a vdi id with hash function */
2580 hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
2581 start_nr = hval & (SD_NR_VDIS - 1);
2582
2583 fd = connect_to_sdog(s, &local_err);
2584 if (fd < 0) {
2585 error_report_err(local_err);
2586 ret = fd;
2587 goto out;
2588 }
2589
2590 for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
2591 if (!test_bit(vid, vdi_inuse)) {
2592 break;
2593 }
2594
2595 /* we don't need to read entire object */
2596 ret = read_object(fd, s->bs, (char *)&inode,
2597 vid_to_vdi_oid(vid),
2598 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0,
2599 s->cache_flags);
2600
2601 if (ret) {
2602 continue;
2603 }
2604
2605 if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
2606 sn_tab[found].date_sec = inode.snap_ctime >> 32;
2607 sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
2608 sn_tab[found].vm_state_size = inode.vm_state_size;
2609 sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec;
2610
2611 snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str),
2612 "%" PRIu32, inode.snap_id);
2613 pstrcpy(sn_tab[found].name,
2614 MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)),
2615 inode.tag);
2616 found++;
2617 }
2618 }
2619
2620 closesocket(fd);
2621 out:
2622 *psn_tab = sn_tab;
2623
2624 g_free(vdi_inuse);
2625
2626 if (ret < 0) {
2627 return ret;
2628 }
2629
2630 return found;
2631 }
2632
2633 static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
2634 int64_t pos, int size, int load)
2635 {
2636 Error *local_err = NULL;
2637 bool create;
2638 int fd, ret = 0, remaining = size;
2639 unsigned int data_len;
2640 uint64_t vmstate_oid;
2641 uint64_t offset;
2642 uint32_t vdi_index;
2643 uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id;
2644 uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift);
2645
2646 fd = connect_to_sdog(s, &local_err);
2647 if (fd < 0) {
2648 error_report_err(local_err);
2649 return fd;
2650 }
2651
2652 while (remaining) {
2653 vdi_index = pos / object_size;
2654 offset = pos % object_size;
2655
2656 data_len = MIN(remaining, object_size - offset);
2657
2658 vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index);
2659
2660 create = (offset == 0);
2661 if (load) {
2662 ret = read_object(fd, s->bs, (char *)data, vmstate_oid,
2663 s->inode.nr_copies, data_len, offset,
2664 s->cache_flags);
2665 } else {
2666 ret = write_object(fd, s->bs, (char *)data, vmstate_oid,
2667 s->inode.nr_copies, data_len, offset, create,
2668 s->cache_flags);
2669 }
2670
2671 if (ret < 0) {
2672 error_report("failed to save vmstate %s", strerror(errno));
2673 goto cleanup;
2674 }
2675
2676 pos += data_len;
2677 data += data_len;
2678 remaining -= data_len;
2679 }
2680 ret = size;
2681 cleanup:
2682 closesocket(fd);
2683 return ret;
2684 }
2685
2686 static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2687 int64_t pos)
2688 {
2689 BDRVSheepdogState *s = bs->opaque;
2690 void *buf;
2691 int ret;
2692
2693 buf = qemu_blockalign(bs, qiov->size);
2694 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
2695 ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0);
2696 qemu_vfree(buf);
2697
2698 return ret;
2699 }
2700
2701 static int sd_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2702 int64_t pos)
2703 {
2704 BDRVSheepdogState *s = bs->opaque;
2705 void *buf;
2706 int ret;
2707
2708 buf = qemu_blockalign(bs, qiov->size);
2709 ret = do_load_save_vmstate(s, buf, pos, qiov->size, 1);
2710 qemu_iovec_from_buf(qiov, 0, buf, qiov->size);
2711 qemu_vfree(buf);
2712
2713 return ret;
2714 }
2715
2716
2717 static coroutine_fn int sd_co_pdiscard(BlockDriverState *bs, int64_t offset,
2718 int count)
2719 {
2720 SheepdogAIOCB acb;
2721 BDRVSheepdogState *s = bs->opaque;
2722 QEMUIOVector discard_iov;
2723 struct iovec iov;
2724 uint32_t zero = 0;
2725
2726 if (!s->discard_supported) {
2727 return 0;
2728 }
2729
2730 memset(&discard_iov, 0, sizeof(discard_iov));
2731 memset(&iov, 0, sizeof(iov));
2732 iov.iov_base = &zero;
2733 iov.iov_len = sizeof(zero);
2734 discard_iov.iov = &iov;
2735 discard_iov.niov = 1;
2736 if (!QEMU_IS_ALIGNED(offset | count, BDRV_SECTOR_SIZE)) {
2737 return -ENOTSUP;
2738 }
2739 sd_aio_setup(&acb, s, &discard_iov, offset >> BDRV_SECTOR_BITS,
2740 count >> BDRV_SECTOR_BITS, AIOCB_DISCARD_OBJ);
2741 sd_co_rw_vector(&acb);
2742 sd_aio_complete(&acb);
2743
2744 return acb.ret;
2745 }
2746
2747 static coroutine_fn int64_t
2748 sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2749 int *pnum, BlockDriverState **file)
2750 {
2751 BDRVSheepdogState *s = bs->opaque;
2752 SheepdogInode *inode = &s->inode;
2753 uint32_t object_size = (UINT32_C(1) << inode->block_size_shift);
2754 uint64_t offset = sector_num * BDRV_SECTOR_SIZE;
2755 unsigned long start = offset / object_size,
2756 end = DIV_ROUND_UP((sector_num + nb_sectors) *
2757 BDRV_SECTOR_SIZE, object_size);
2758 unsigned long idx;
2759 int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
2760
2761 for (idx = start; idx < end; idx++) {
2762 if (inode->data_vdi_id[idx] == 0) {
2763 break;
2764 }
2765 }
2766 if (idx == start) {
2767 /* Get the longest length of unallocated sectors */
2768 ret = 0;
2769 for (idx = start + 1; idx < end; idx++) {
2770 if (inode->data_vdi_id[idx] != 0) {
2771 break;
2772 }
2773 }
2774 }
2775
2776 *pnum = (idx - start) * object_size / BDRV_SECTOR_SIZE;
2777 if (*pnum > nb_sectors) {
2778 *pnum = nb_sectors;
2779 }
2780 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
2781 *file = bs;
2782 }
2783 return ret;
2784 }
2785
2786 static int64_t sd_get_allocated_file_size(BlockDriverState *bs)
2787 {
2788 BDRVSheepdogState *s = bs->opaque;
2789 SheepdogInode *inode = &s->inode;
2790 uint32_t object_size = (UINT32_C(1) << inode->block_size_shift);
2791 unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size);
2792 uint64_t size = 0;
2793
2794 for (i = 0; i < last; i++) {
2795 if (inode->data_vdi_id[i] == 0) {
2796 continue;
2797 }
2798 size += object_size;
2799 }
2800 return size;
2801 }
2802
2803 static QemuOptsList sd_create_opts = {
2804 .name = "sheepdog-create-opts",
2805 .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head),
2806 .desc = {
2807 {
2808 .name = BLOCK_OPT_SIZE,
2809 .type = QEMU_OPT_SIZE,
2810 .help = "Virtual disk size"
2811 },
2812 {
2813 .name = BLOCK_OPT_BACKING_FILE,
2814 .type = QEMU_OPT_STRING,
2815 .help = "File name of a base image"
2816 },
2817 {
2818 .name = BLOCK_OPT_PREALLOC,
2819 .type = QEMU_OPT_STRING,
2820 .help = "Preallocation mode (allowed values: off, full)"
2821 },
2822 {
2823 .name = BLOCK_OPT_REDUNDANCY,
2824 .type = QEMU_OPT_STRING,
2825 .help = "Redundancy of the image"
2826 },
2827 {
2828 .name = BLOCK_OPT_OBJECT_SIZE,
2829 .type = QEMU_OPT_SIZE,
2830 .help = "Object size of the image"
2831 },
2832 { /* end of list */ }
2833 }
2834 };
2835
2836 static BlockDriver bdrv_sheepdog = {
2837 .format_name = "sheepdog",
2838 .protocol_name = "sheepdog",
2839 .instance_size = sizeof(BDRVSheepdogState),
2840 .bdrv_needs_filename = true,
2841 .bdrv_file_open = sd_open,
2842 .bdrv_reopen_prepare = sd_reopen_prepare,
2843 .bdrv_reopen_commit = sd_reopen_commit,
2844 .bdrv_reopen_abort = sd_reopen_abort,
2845 .bdrv_close = sd_close,
2846 .bdrv_create = sd_create,
2847 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2848 .bdrv_getlength = sd_getlength,
2849 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2850 .bdrv_truncate = sd_truncate,
2851
2852 .bdrv_co_readv = sd_co_readv,
2853 .bdrv_co_writev = sd_co_writev,
2854 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2855 .bdrv_co_pdiscard = sd_co_pdiscard,
2856 .bdrv_co_get_block_status = sd_co_get_block_status,
2857
2858 .bdrv_snapshot_create = sd_snapshot_create,
2859 .bdrv_snapshot_goto = sd_snapshot_goto,
2860 .bdrv_snapshot_delete = sd_snapshot_delete,
2861 .bdrv_snapshot_list = sd_snapshot_list,
2862
2863 .bdrv_save_vmstate = sd_save_vmstate,
2864 .bdrv_load_vmstate = sd_load_vmstate,
2865
2866 .bdrv_detach_aio_context = sd_detach_aio_context,
2867 .bdrv_attach_aio_context = sd_attach_aio_context,
2868
2869 .create_opts = &sd_create_opts,
2870 };
2871
2872 static BlockDriver bdrv_sheepdog_tcp = {
2873 .format_name = "sheepdog",
2874 .protocol_name = "sheepdog+tcp",
2875 .instance_size = sizeof(BDRVSheepdogState),
2876 .bdrv_needs_filename = true,
2877 .bdrv_file_open = sd_open,
2878 .bdrv_reopen_prepare = sd_reopen_prepare,
2879 .bdrv_reopen_commit = sd_reopen_commit,
2880 .bdrv_reopen_abort = sd_reopen_abort,
2881 .bdrv_close = sd_close,
2882 .bdrv_create = sd_create,
2883 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2884 .bdrv_getlength = sd_getlength,
2885 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2886 .bdrv_truncate = sd_truncate,
2887
2888 .bdrv_co_readv = sd_co_readv,
2889 .bdrv_co_writev = sd_co_writev,
2890 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2891 .bdrv_co_pdiscard = sd_co_pdiscard,
2892 .bdrv_co_get_block_status = sd_co_get_block_status,
2893
2894 .bdrv_snapshot_create = sd_snapshot_create,
2895 .bdrv_snapshot_goto = sd_snapshot_goto,
2896 .bdrv_snapshot_delete = sd_snapshot_delete,
2897 .bdrv_snapshot_list = sd_snapshot_list,
2898
2899 .bdrv_save_vmstate = sd_save_vmstate,
2900 .bdrv_load_vmstate = sd_load_vmstate,
2901
2902 .bdrv_detach_aio_context = sd_detach_aio_context,
2903 .bdrv_attach_aio_context = sd_attach_aio_context,
2904
2905 .create_opts = &sd_create_opts,
2906 };
2907
2908 static BlockDriver bdrv_sheepdog_unix = {
2909 .format_name = "sheepdog",
2910 .protocol_name = "sheepdog+unix",
2911 .instance_size = sizeof(BDRVSheepdogState),
2912 .bdrv_needs_filename = true,
2913 .bdrv_file_open = sd_open,
2914 .bdrv_reopen_prepare = sd_reopen_prepare,
2915 .bdrv_reopen_commit = sd_reopen_commit,
2916 .bdrv_reopen_abort = sd_reopen_abort,
2917 .bdrv_close = sd_close,
2918 .bdrv_create = sd_create,
2919 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2920 .bdrv_getlength = sd_getlength,
2921 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
2922 .bdrv_truncate = sd_truncate,
2923
2924 .bdrv_co_readv = sd_co_readv,
2925 .bdrv_co_writev = sd_co_writev,
2926 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
2927 .bdrv_co_pdiscard = sd_co_pdiscard,
2928 .bdrv_co_get_block_status = sd_co_get_block_status,
2929
2930 .bdrv_snapshot_create = sd_snapshot_create,
2931 .bdrv_snapshot_goto = sd_snapshot_goto,
2932 .bdrv_snapshot_delete = sd_snapshot_delete,
2933 .bdrv_snapshot_list = sd_snapshot_list,
2934
2935 .bdrv_save_vmstate = sd_save_vmstate,
2936 .bdrv_load_vmstate = sd_load_vmstate,
2937
2938 .bdrv_detach_aio_context = sd_detach_aio_context,
2939 .bdrv_attach_aio_context = sd_attach_aio_context,
2940
2941 .create_opts = &sd_create_opts,
2942 };
2943
2944 static void bdrv_sheepdog_init(void)
2945 {
2946 bdrv_register(&bdrv_sheepdog);
2947 bdrv_register(&bdrv_sheepdog_tcp);
2948 bdrv_register(&bdrv_sheepdog_unix);
2949 }
2950 block_init(bdrv_sheepdog_init);