block/sheepdog: Replace magic val by NANOSECONDS_PER_SECOND definition
[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 "qemu-common.h"
17 #include "qapi/error.h"
18 #include "qapi/qapi-visit-sockets.h"
19 #include "qapi/qapi-visit-block-core.h"
20 #include "qapi/qmp/qdict.h"
21 #include "qapi/qobject-input-visitor.h"
22 #include "qapi/qobject-output-visitor.h"
23 #include "qemu/uri.h"
24 #include "qemu/error-report.h"
25 #include "qemu/main-loop.h"
26 #include "qemu/module.h"
27 #include "qemu/option.h"
28 #include "qemu/sockets.h"
29 #include "block/block_int.h"
30 #include "block/qdict.h"
31 #include "sysemu/block-backend.h"
32 #include "qemu/bitops.h"
33 #include "qemu/cutils.h"
34 #include "trace.h"
35
36 #define SD_PROTO_VER 0x01
37
38 #define SD_DEFAULT_ADDR "localhost"
39 #define SD_DEFAULT_PORT 7000
40
41 #define SD_OP_CREATE_AND_WRITE_OBJ 0x01
42 #define SD_OP_READ_OBJ 0x02
43 #define SD_OP_WRITE_OBJ 0x03
44 /* 0x04 is used internally by Sheepdog */
45
46 #define SD_OP_NEW_VDI 0x11
47 #define SD_OP_LOCK_VDI 0x12
48 #define SD_OP_RELEASE_VDI 0x13
49 #define SD_OP_GET_VDI_INFO 0x14
50 #define SD_OP_READ_VDIS 0x15
51 #define SD_OP_FLUSH_VDI 0x16
52 #define SD_OP_DEL_VDI 0x17
53 #define SD_OP_GET_CLUSTER_DEFAULT 0x18
54
55 #define SD_FLAG_CMD_WRITE 0x01
56 #define SD_FLAG_CMD_COW 0x02
57 #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */
58 #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */
59
60 #define SD_RES_SUCCESS 0x00 /* Success */
61 #define SD_RES_UNKNOWN 0x01 /* Unknown error */
62 #define SD_RES_NO_OBJ 0x02 /* No object found */
63 #define SD_RES_EIO 0x03 /* I/O error */
64 #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */
65 #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
66 #define SD_RES_SYSTEM_ERROR 0x06 /* System error */
67 #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */
68 #define SD_RES_NO_VDI 0x08 /* No vdi found */
69 #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */
70 #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */
71 #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */
72 #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
73 #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */
74 #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */
75 #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */
76 #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */
77 #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */
78 #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */
79 #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */
80 #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */
81 #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */
82 #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */
83 #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */
84 #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */
85 #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */
86 #define SD_RES_READONLY 0x1A /* Object is read-only */
87
88 /*
89 * Object ID rules
90 *
91 * 0 - 19 (20 bits): data object space
92 * 20 - 31 (12 bits): reserved data object space
93 * 32 - 55 (24 bits): vdi object space
94 * 56 - 59 ( 4 bits): reserved vdi object space
95 * 60 - 63 ( 4 bits): object type identifier space
96 */
97
98 #define VDI_SPACE_SHIFT 32
99 #define VDI_BIT (UINT64_C(1) << 63)
100 #define VMSTATE_BIT (UINT64_C(1) << 62)
101 #define MAX_DATA_OBJS (UINT64_C(1) << 20)
102 #define MAX_CHILDREN 1024
103 #define SD_MAX_VDI_LEN 256
104 #define SD_MAX_VDI_TAG_LEN 256
105 #define SD_NR_VDIS (1U << 24)
106 #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
107 #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
108 #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22
109 /*
110 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
111 * (SD_EC_MAX_STRIP - 1) for parity strips
112 *
113 * SD_MAX_COPIES is sum of number of data strips and parity strips.
114 */
115 #define SD_EC_MAX_STRIP 16
116 #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
117
118 #define SD_INODE_SIZE (sizeof(SheepdogInode))
119 #define CURRENT_VDI_ID 0
120
121 #define LOCK_TYPE_NORMAL 0
122 #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */
123
124 typedef struct SheepdogReq {
125 uint8_t proto_ver;
126 uint8_t opcode;
127 uint16_t flags;
128 uint32_t epoch;
129 uint32_t id;
130 uint32_t data_length;
131 uint32_t opcode_specific[8];
132 } SheepdogReq;
133
134 typedef struct SheepdogRsp {
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 uint32_t result;
142 uint32_t opcode_specific[7];
143 } SheepdogRsp;
144
145 typedef struct SheepdogObjReq {
146 uint8_t proto_ver;
147 uint8_t opcode;
148 uint16_t flags;
149 uint32_t epoch;
150 uint32_t id;
151 uint32_t data_length;
152 uint64_t oid;
153 uint64_t cow_oid;
154 uint8_t copies;
155 uint8_t copy_policy;
156 uint8_t reserved[6];
157 uint64_t offset;
158 } SheepdogObjReq;
159
160 typedef struct SheepdogObjRsp {
161 uint8_t proto_ver;
162 uint8_t opcode;
163 uint16_t flags;
164 uint32_t epoch;
165 uint32_t id;
166 uint32_t data_length;
167 uint32_t result;
168 uint8_t copies;
169 uint8_t copy_policy;
170 uint8_t reserved[2];
171 uint32_t pad[6];
172 } SheepdogObjRsp;
173
174 typedef struct SheepdogVdiReq {
175 uint8_t proto_ver;
176 uint8_t opcode;
177 uint16_t flags;
178 uint32_t epoch;
179 uint32_t id;
180 uint32_t data_length;
181 uint64_t vdi_size;
182 uint32_t base_vdi_id;
183 uint8_t copies;
184 uint8_t copy_policy;
185 uint8_t store_policy;
186 uint8_t block_size_shift;
187 uint32_t snapid;
188 uint32_t type;
189 uint32_t pad[2];
190 } SheepdogVdiReq;
191
192 typedef struct SheepdogVdiRsp {
193 uint8_t proto_ver;
194 uint8_t opcode;
195 uint16_t flags;
196 uint32_t epoch;
197 uint32_t id;
198 uint32_t data_length;
199 uint32_t result;
200 uint32_t rsvd;
201 uint32_t vdi_id;
202 uint32_t pad[5];
203 } SheepdogVdiRsp;
204
205 typedef struct SheepdogClusterRsp {
206 uint8_t proto_ver;
207 uint8_t opcode;
208 uint16_t flags;
209 uint32_t epoch;
210 uint32_t id;
211 uint32_t data_length;
212 uint32_t result;
213 uint8_t nr_copies;
214 uint8_t copy_policy;
215 uint8_t block_size_shift;
216 uint8_t __pad1;
217 uint32_t __pad2[6];
218 } SheepdogClusterRsp;
219
220 typedef struct SheepdogInode {
221 char name[SD_MAX_VDI_LEN];
222 char tag[SD_MAX_VDI_TAG_LEN];
223 uint64_t ctime;
224 uint64_t snap_ctime;
225 uint64_t vm_clock_nsec;
226 uint64_t vdi_size;
227 uint64_t vm_state_size;
228 uint16_t copy_policy;
229 uint8_t nr_copies;
230 uint8_t block_size_shift;
231 uint32_t snap_id;
232 uint32_t vdi_id;
233 uint32_t parent_vdi_id;
234 uint32_t child_vdi_id[MAX_CHILDREN];
235 uint32_t data_vdi_id[MAX_DATA_OBJS];
236 } SheepdogInode;
237
238 #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
239
240 /*
241 * 64 bit FNV-1a non-zero initial basis
242 */
243 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
244
245 /*
246 * 64 bit Fowler/Noll/Vo FNV-1a hash code
247 */
248 static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
249 {
250 unsigned char *bp = buf;
251 unsigned char *be = bp + len;
252 while (bp < be) {
253 hval ^= (uint64_t) *bp++;
254 hval += (hval << 1) + (hval << 4) + (hval << 5) +
255 (hval << 7) + (hval << 8) + (hval << 40);
256 }
257 return hval;
258 }
259
260 static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
261 {
262 return inode->vdi_id == inode->data_vdi_id[idx];
263 }
264
265 static inline bool is_data_obj(uint64_t oid)
266 {
267 return !(VDI_BIT & oid);
268 }
269
270 static inline uint64_t data_oid_to_idx(uint64_t oid)
271 {
272 return oid & (MAX_DATA_OBJS - 1);
273 }
274
275 static inline uint32_t oid_to_vid(uint64_t oid)
276 {
277 return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT;
278 }
279
280 static inline uint64_t vid_to_vdi_oid(uint32_t vid)
281 {
282 return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
283 }
284
285 static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx)
286 {
287 return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
288 }
289
290 static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx)
291 {
292 return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
293 }
294
295 static inline bool is_snapshot(struct SheepdogInode *inode)
296 {
297 return !!inode->snap_ctime;
298 }
299
300 static inline size_t count_data_objs(const struct SheepdogInode *inode)
301 {
302 return DIV_ROUND_UP(inode->vdi_size,
303 (1UL << inode->block_size_shift));
304 }
305
306 typedef struct SheepdogAIOCB SheepdogAIOCB;
307 typedef struct BDRVSheepdogState BDRVSheepdogState;
308
309 typedef struct AIOReq {
310 SheepdogAIOCB *aiocb;
311 unsigned int iov_offset;
312
313 uint64_t oid;
314 uint64_t base_oid;
315 uint64_t offset;
316 unsigned int data_len;
317 uint8_t flags;
318 uint32_t id;
319 bool create;
320
321 QLIST_ENTRY(AIOReq) aio_siblings;
322 } AIOReq;
323
324 enum AIOCBState {
325 AIOCB_WRITE_UDATA,
326 AIOCB_READ_UDATA,
327 AIOCB_FLUSH_CACHE,
328 AIOCB_DISCARD_OBJ,
329 };
330
331 #define AIOCBOverlapping(x, y) \
332 (!(x->max_affect_data_idx < y->min_affect_data_idx \
333 || y->max_affect_data_idx < x->min_affect_data_idx))
334
335 struct SheepdogAIOCB {
336 BDRVSheepdogState *s;
337
338 QEMUIOVector *qiov;
339
340 int64_t sector_num;
341 int nb_sectors;
342
343 int ret;
344 enum AIOCBState aiocb_type;
345
346 Coroutine *coroutine;
347 int nr_pending;
348
349 uint32_t min_affect_data_idx;
350 uint32_t max_affect_data_idx;
351
352 /*
353 * The difference between affect_data_idx and dirty_data_idx:
354 * affect_data_idx represents range of index of all request types.
355 * dirty_data_idx represents range of index updated by COW requests.
356 * dirty_data_idx is used for updating an inode object.
357 */
358 uint32_t min_dirty_data_idx;
359 uint32_t max_dirty_data_idx;
360
361 QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings;
362 };
363
364 struct BDRVSheepdogState {
365 BlockDriverState *bs;
366 AioContext *aio_context;
367
368 SheepdogInode inode;
369
370 char name[SD_MAX_VDI_LEN];
371 bool is_snapshot;
372 uint32_t cache_flags;
373 bool discard_supported;
374
375 SocketAddress *addr;
376 int fd;
377
378 CoMutex lock;
379 Coroutine *co_send;
380 Coroutine *co_recv;
381
382 uint32_t aioreq_seq_num;
383
384 /* Every aio request must be linked to either of these queues. */
385 QLIST_HEAD(, AIOReq) inflight_aio_head;
386 QLIST_HEAD(, AIOReq) failed_aio_head;
387
388 CoMutex queue_lock;
389 CoQueue overlapping_queue;
390 QLIST_HEAD(, SheepdogAIOCB) inflight_aiocb_head;
391 };
392
393 typedef struct BDRVSheepdogReopenState {
394 int fd;
395 int cache_flags;
396 } BDRVSheepdogReopenState;
397
398 static const char *sd_strerror(int err)
399 {
400 int i;
401
402 static const struct {
403 int err;
404 const char *desc;
405 } errors[] = {
406 {SD_RES_SUCCESS, "Success"},
407 {SD_RES_UNKNOWN, "Unknown error"},
408 {SD_RES_NO_OBJ, "No object found"},
409 {SD_RES_EIO, "I/O error"},
410 {SD_RES_VDI_EXIST, "VDI exists already"},
411 {SD_RES_INVALID_PARMS, "Invalid parameters"},
412 {SD_RES_SYSTEM_ERROR, "System error"},
413 {SD_RES_VDI_LOCKED, "VDI is already locked"},
414 {SD_RES_NO_VDI, "No vdi found"},
415 {SD_RES_NO_BASE_VDI, "No base VDI found"},
416 {SD_RES_VDI_READ, "Failed read the requested VDI"},
417 {SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
418 {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
419 {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
420 {SD_RES_NO_TAG, "Failed to find the requested tag"},
421 {SD_RES_STARTUP, "The system is still booting"},
422 {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
423 {SD_RES_SHUTDOWN, "The system is shutting down"},
424 {SD_RES_NO_MEM, "Out of memory on the server"},
425 {SD_RES_FULL_VDI, "We already have the maximum vdis"},
426 {SD_RES_VER_MISMATCH, "Protocol version mismatch"},
427 {SD_RES_NO_SPACE, "Server has no space for new objects"},
428 {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
429 {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
430 {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
431 {SD_RES_HALT, "Sheepdog is stopped serving IO request"},
432 {SD_RES_READONLY, "Object is read-only"},
433 };
434
435 for (i = 0; i < ARRAY_SIZE(errors); ++i) {
436 if (errors[i].err == err) {
437 return errors[i].desc;
438 }
439 }
440
441 return "Invalid error code";
442 }
443
444 /*
445 * Sheepdog I/O handling:
446 *
447 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
448 * link the requests to the inflight_list in the
449 * BDRVSheepdogState. The function yields while waiting for
450 * receiving the response.
451 *
452 * 2. We receive the response in aio_read_response, the fd handler to
453 * the sheepdog connection. We switch back to sd_co_readv/sd_writev
454 * after all the requests belonging to the AIOCB are finished. If
455 * needed, sd_co_writev will send another requests for the vdi object.
456 */
457
458 static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
459 uint64_t oid, unsigned int data_len,
460 uint64_t offset, uint8_t flags, bool create,
461 uint64_t base_oid, unsigned int iov_offset)
462 {
463 AIOReq *aio_req;
464
465 aio_req = g_malloc(sizeof(*aio_req));
466 aio_req->aiocb = acb;
467 aio_req->iov_offset = iov_offset;
468 aio_req->oid = oid;
469 aio_req->base_oid = base_oid;
470 aio_req->offset = offset;
471 aio_req->data_len = data_len;
472 aio_req->flags = flags;
473 aio_req->id = s->aioreq_seq_num++;
474 aio_req->create = create;
475
476 acb->nr_pending++;
477 return aio_req;
478 }
479
480 static void wait_for_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *acb)
481 {
482 SheepdogAIOCB *cb;
483
484 retry:
485 QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) {
486 if (AIOCBOverlapping(acb, cb)) {
487 qemu_co_queue_wait(&s->overlapping_queue, &s->queue_lock);
488 goto retry;
489 }
490 }
491 }
492
493 static void sd_aio_setup(SheepdogAIOCB *acb, BDRVSheepdogState *s,
494 QEMUIOVector *qiov, int64_t sector_num, int nb_sectors,
495 int type)
496 {
497 uint32_t object_size;
498
499 object_size = (UINT32_C(1) << s->inode.block_size_shift);
500
501 acb->s = s;
502
503 acb->qiov = qiov;
504
505 acb->sector_num = sector_num;
506 acb->nb_sectors = nb_sectors;
507
508 acb->coroutine = qemu_coroutine_self();
509 acb->ret = 0;
510 acb->nr_pending = 0;
511
512 acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size;
513 acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE +
514 acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size;
515
516 acb->min_dirty_data_idx = UINT32_MAX;
517 acb->max_dirty_data_idx = 0;
518 acb->aiocb_type = type;
519
520 if (type == AIOCB_FLUSH_CACHE) {
521 return;
522 }
523
524 qemu_co_mutex_lock(&s->queue_lock);
525 wait_for_overlapping_aiocb(s, acb);
526 QLIST_INSERT_HEAD(&s->inflight_aiocb_head, acb, aiocb_siblings);
527 qemu_co_mutex_unlock(&s->queue_lock);
528 }
529
530 static SocketAddress *sd_server_config(QDict *options, Error **errp)
531 {
532 QDict *server = NULL;
533 Visitor *iv = NULL;
534 SocketAddress *saddr = NULL;
535
536 qdict_extract_subqdict(options, &server, "server.");
537
538 iv = qobject_input_visitor_new_flat_confused(server, errp);
539 if (!iv) {
540 goto done;
541 }
542
543 if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) {
544 goto done;
545 }
546
547 done:
548 visit_free(iv);
549 qobject_unref(server);
550 return saddr;
551 }
552
553 /* Return -EIO in case of error, file descriptor on success */
554 static int connect_to_sdog(BDRVSheepdogState *s, Error **errp)
555 {
556 int fd;
557
558 fd = socket_connect(s->addr, errp);
559
560 if (s->addr->type == SOCKET_ADDRESS_TYPE_INET && fd >= 0) {
561 int ret = socket_set_nodelay(fd);
562 if (ret < 0) {
563 warn_report("can't set TCP_NODELAY: %s", strerror(errno));
564 }
565 }
566
567 if (fd >= 0) {
568 qemu_set_nonblock(fd);
569 } else {
570 fd = -EIO;
571 }
572
573 return fd;
574 }
575
576 /* Return 0 on success and -errno in case of error */
577 static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
578 unsigned int *wlen)
579 {
580 int ret;
581
582 ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
583 if (ret != sizeof(*hdr)) {
584 error_report("failed to send a req, %s", strerror(errno));
585 return -errno;
586 }
587
588 ret = qemu_co_send(sockfd, data, *wlen);
589 if (ret != *wlen) {
590 error_report("failed to send a req, %s", strerror(errno));
591 return -errno;
592 }
593
594 return ret;
595 }
596
597 typedef struct SheepdogReqCo {
598 int sockfd;
599 BlockDriverState *bs;
600 AioContext *aio_context;
601 SheepdogReq *hdr;
602 void *data;
603 unsigned int *wlen;
604 unsigned int *rlen;
605 int ret;
606 bool finished;
607 Coroutine *co;
608 } SheepdogReqCo;
609
610 static void restart_co_req(void *opaque)
611 {
612 SheepdogReqCo *srco = opaque;
613
614 aio_co_wake(srco->co);
615 }
616
617 static coroutine_fn void do_co_req(void *opaque)
618 {
619 int ret;
620 SheepdogReqCo *srco = opaque;
621 int sockfd = srco->sockfd;
622 SheepdogReq *hdr = srco->hdr;
623 void *data = srco->data;
624 unsigned int *wlen = srco->wlen;
625 unsigned int *rlen = srco->rlen;
626
627 srco->co = qemu_coroutine_self();
628 aio_set_fd_handler(srco->aio_context, sockfd, false,
629 NULL, restart_co_req, NULL, srco);
630
631 ret = send_co_req(sockfd, hdr, data, wlen);
632 if (ret < 0) {
633 goto out;
634 }
635
636 aio_set_fd_handler(srco->aio_context, sockfd, false,
637 restart_co_req, NULL, NULL, srco);
638
639 ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
640 if (ret != sizeof(*hdr)) {
641 error_report("failed to get a rsp, %s", strerror(errno));
642 ret = -errno;
643 goto out;
644 }
645
646 if (*rlen > hdr->data_length) {
647 *rlen = hdr->data_length;
648 }
649
650 if (*rlen) {
651 ret = qemu_co_recv(sockfd, data, *rlen);
652 if (ret != *rlen) {
653 error_report("failed to get the data, %s", strerror(errno));
654 ret = -errno;
655 goto out;
656 }
657 }
658 ret = 0;
659 out:
660 /* there is at most one request for this sockfd, so it is safe to
661 * set each handler to NULL. */
662 aio_set_fd_handler(srco->aio_context, sockfd, false,
663 NULL, NULL, NULL, NULL);
664
665 srco->co = NULL;
666 srco->ret = ret;
667 /* Set srco->finished before reading bs->wakeup. */
668 qatomic_mb_set(&srco->finished, true);
669 if (srco->bs) {
670 bdrv_wakeup(srco->bs);
671 }
672 }
673
674 /*
675 * Send the request to the sheep in a synchronous manner.
676 *
677 * Return 0 on success, -errno in case of error.
678 */
679 static int do_req(int sockfd, BlockDriverState *bs, SheepdogReq *hdr,
680 void *data, unsigned int *wlen, unsigned int *rlen)
681 {
682 Coroutine *co;
683 SheepdogReqCo srco = {
684 .sockfd = sockfd,
685 .aio_context = bs ? bdrv_get_aio_context(bs) : qemu_get_aio_context(),
686 .bs = bs,
687 .hdr = hdr,
688 .data = data,
689 .wlen = wlen,
690 .rlen = rlen,
691 .ret = 0,
692 .finished = false,
693 };
694
695 if (qemu_in_coroutine()) {
696 do_co_req(&srco);
697 } else {
698 co = qemu_coroutine_create(do_co_req, &srco);
699 if (bs) {
700 bdrv_coroutine_enter(bs, co);
701 BDRV_POLL_WHILE(bs, !srco.finished);
702 } else {
703 qemu_coroutine_enter(co);
704 while (!srco.finished) {
705 aio_poll(qemu_get_aio_context(), true);
706 }
707 }
708 }
709
710 return srco.ret;
711 }
712
713 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
714 struct iovec *iov, int niov,
715 enum AIOCBState aiocb_type);
716 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req);
717 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag);
718 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp);
719 static void co_write_request(void *opaque);
720
721 static coroutine_fn void reconnect_to_sdog(void *opaque)
722 {
723 BDRVSheepdogState *s = opaque;
724 AIOReq *aio_req, *next;
725
726 aio_set_fd_handler(s->aio_context, s->fd, false, NULL,
727 NULL, NULL, NULL);
728 close(s->fd);
729 s->fd = -1;
730
731 /* Wait for outstanding write requests to be completed. */
732 while (s->co_send != NULL) {
733 co_write_request(opaque);
734 }
735
736 /* Try to reconnect the sheepdog server every one second. */
737 while (s->fd < 0) {
738 Error *local_err = NULL;
739 s->fd = get_sheep_fd(s, &local_err);
740 if (s->fd < 0) {
741 trace_sheepdog_reconnect_to_sdog();
742 error_report_err(local_err);
743 qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, NANOSECONDS_PER_SECOND);
744 }
745 };
746
747 /*
748 * Now we have to resend all the request in the inflight queue. However,
749 * resend_aioreq() can yield and newly created requests can be added to the
750 * inflight queue before the coroutine is resumed. To avoid mixing them, we
751 * have to move all the inflight requests to the failed queue before
752 * resend_aioreq() is called.
753 */
754 qemu_co_mutex_lock(&s->queue_lock);
755 QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) {
756 QLIST_REMOVE(aio_req, aio_siblings);
757 QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings);
758 }
759
760 /* Resend all the failed aio requests. */
761 while (!QLIST_EMPTY(&s->failed_aio_head)) {
762 aio_req = QLIST_FIRST(&s->failed_aio_head);
763 QLIST_REMOVE(aio_req, aio_siblings);
764 qemu_co_mutex_unlock(&s->queue_lock);
765 resend_aioreq(s, aio_req);
766 qemu_co_mutex_lock(&s->queue_lock);
767 }
768 qemu_co_mutex_unlock(&s->queue_lock);
769 }
770
771 /*
772 * Receive responses of the I/O requests.
773 *
774 * This function is registered as a fd handler, and called from the
775 * main loop when s->fd is ready for reading responses.
776 */
777 static void coroutine_fn aio_read_response(void *opaque)
778 {
779 SheepdogObjRsp rsp;
780 BDRVSheepdogState *s = opaque;
781 int fd = s->fd;
782 int ret;
783 AIOReq *aio_req = NULL;
784 SheepdogAIOCB *acb;
785 uint64_t idx;
786
787 /* read a header */
788 ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
789 if (ret != sizeof(rsp)) {
790 error_report("failed to get the header, %s", strerror(errno));
791 goto err;
792 }
793
794 /* find the right aio_req from the inflight aio list */
795 QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
796 if (aio_req->id == rsp.id) {
797 break;
798 }
799 }
800 if (!aio_req) {
801 error_report("cannot find aio_req %x", rsp.id);
802 goto err;
803 }
804
805 acb = aio_req->aiocb;
806
807 switch (acb->aiocb_type) {
808 case AIOCB_WRITE_UDATA:
809 if (!is_data_obj(aio_req->oid)) {
810 break;
811 }
812 idx = data_oid_to_idx(aio_req->oid);
813
814 if (aio_req->create) {
815 /*
816 * If the object is newly created one, we need to update
817 * the vdi object (metadata object). min_dirty_data_idx
818 * and max_dirty_data_idx are changed to include updated
819 * index between them.
820 */
821 if (rsp.result == SD_RES_SUCCESS) {
822 s->inode.data_vdi_id[idx] = s->inode.vdi_id;
823 acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx);
824 acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx);
825 }
826 }
827 break;
828 case AIOCB_READ_UDATA:
829 ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
830 aio_req->iov_offset, rsp.data_length);
831 if (ret != rsp.data_length) {
832 error_report("failed to get the data, %s", strerror(errno));
833 goto err;
834 }
835 break;
836 case AIOCB_FLUSH_CACHE:
837 if (rsp.result == SD_RES_INVALID_PARMS) {
838 trace_sheepdog_aio_read_response();
839 s->cache_flags = SD_FLAG_CMD_DIRECT;
840 rsp.result = SD_RES_SUCCESS;
841 }
842 break;
843 case AIOCB_DISCARD_OBJ:
844 switch (rsp.result) {
845 case SD_RES_INVALID_PARMS:
846 error_report("server doesn't support discard command");
847 rsp.result = SD_RES_SUCCESS;
848 s->discard_supported = false;
849 break;
850 default:
851 break;
852 }
853 }
854
855 /* No more data for this aio_req (reload_inode below uses its own file
856 * descriptor handler which doesn't use co_recv).
857 */
858 s->co_recv = NULL;
859
860 qemu_co_mutex_lock(&s->queue_lock);
861 QLIST_REMOVE(aio_req, aio_siblings);
862 qemu_co_mutex_unlock(&s->queue_lock);
863
864 switch (rsp.result) {
865 case SD_RES_SUCCESS:
866 break;
867 case SD_RES_READONLY:
868 if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) {
869 ret = reload_inode(s, 0, "");
870 if (ret < 0) {
871 goto err;
872 }
873 }
874 if (is_data_obj(aio_req->oid)) {
875 aio_req->oid = vid_to_data_oid(s->inode.vdi_id,
876 data_oid_to_idx(aio_req->oid));
877 } else {
878 aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id);
879 }
880 resend_aioreq(s, aio_req);
881 return;
882 default:
883 acb->ret = -EIO;
884 error_report("%s", sd_strerror(rsp.result));
885 break;
886 }
887
888 g_free(aio_req);
889
890 if (!--acb->nr_pending) {
891 /*
892 * We've finished all requests which belong to the AIOCB, so
893 * we can switch back to sd_co_readv/writev now.
894 */
895 aio_co_wake(acb->coroutine);
896 }
897
898 return;
899
900 err:
901 reconnect_to_sdog(opaque);
902 }
903
904 static void co_read_response(void *opaque)
905 {
906 BDRVSheepdogState *s = opaque;
907
908 if (!s->co_recv) {
909 s->co_recv = qemu_coroutine_create(aio_read_response, opaque);
910 }
911
912 aio_co_enter(s->aio_context, s->co_recv);
913 }
914
915 static void co_write_request(void *opaque)
916 {
917 BDRVSheepdogState *s = opaque;
918
919 aio_co_wake(s->co_send);
920 }
921
922 /*
923 * Return a socket descriptor to read/write objects.
924 *
925 * We cannot use this descriptor for other operations because
926 * the block driver may be on waiting response from the server.
927 */
928 static int get_sheep_fd(BDRVSheepdogState *s, Error **errp)
929 {
930 int fd;
931
932 fd = connect_to_sdog(s, errp);
933 if (fd < 0) {
934 return fd;
935 }
936
937 aio_set_fd_handler(s->aio_context, fd, false,
938 co_read_response, NULL, NULL, s);
939 return fd;
940 }
941
942 /*
943 * Parse numeric snapshot ID in @str
944 * If @str can't be parsed as number, return false.
945 * Else, if the number is zero or too large, set *@snapid to zero and
946 * return true.
947 * Else, set *@snapid to the number and return true.
948 */
949 static bool sd_parse_snapid(const char *str, uint32_t *snapid)
950 {
951 unsigned long ul;
952 int ret;
953
954 ret = qemu_strtoul(str, NULL, 10, &ul);
955 if (ret == -ERANGE) {
956 ul = ret = 0;
957 }
958 if (ret) {
959 return false;
960 }
961 if (ul > UINT32_MAX) {
962 ul = 0;
963 }
964
965 *snapid = ul;
966 return true;
967 }
968
969 static bool sd_parse_snapid_or_tag(const char *str,
970 uint32_t *snapid, char tag[])
971 {
972 if (!sd_parse_snapid(str, snapid)) {
973 *snapid = 0;
974 if (g_strlcpy(tag, str, SD_MAX_VDI_TAG_LEN) >= SD_MAX_VDI_TAG_LEN) {
975 return false;
976 }
977 } else if (!*snapid) {
978 return false;
979 } else {
980 tag[0] = 0;
981 }
982 return true;
983 }
984
985 typedef struct {
986 const char *path; /* non-null iff transport is tcp */
987 const char *host; /* valid when transport is tcp */
988 int port; /* valid when transport is tcp */
989 char vdi[SD_MAX_VDI_LEN];
990 char tag[SD_MAX_VDI_TAG_LEN];
991 uint32_t snap_id;
992 /* Remainder is only for sd_config_done() */
993 URI *uri;
994 QueryParams *qp;
995 } SheepdogConfig;
996
997 static void sd_config_done(SheepdogConfig *cfg)
998 {
999 if (cfg->qp) {
1000 query_params_free(cfg->qp);
1001 }
1002 uri_free(cfg->uri);
1003 }
1004
1005 static void sd_parse_uri(SheepdogConfig *cfg, const char *filename,
1006 Error **errp)
1007 {
1008 Error *err = NULL;
1009 QueryParams *qp = NULL;
1010 bool is_unix;
1011 URI *uri;
1012
1013 memset(cfg, 0, sizeof(*cfg));
1014
1015 cfg->uri = uri = uri_parse(filename);
1016 if (!uri) {
1017 error_setg(&err, "invalid URI '%s'", filename);
1018 goto out;
1019 }
1020
1021 /* transport */
1022 if (!g_strcmp0(uri->scheme, "sheepdog")) {
1023 is_unix = false;
1024 } else if (!g_strcmp0(uri->scheme, "sheepdog+tcp")) {
1025 is_unix = false;
1026 } else if (!g_strcmp0(uri->scheme, "sheepdog+unix")) {
1027 is_unix = true;
1028 } else {
1029 error_setg(&err, "URI scheme must be 'sheepdog', 'sheepdog+tcp',"
1030 " or 'sheepdog+unix'");
1031 goto out;
1032 }
1033
1034 if (uri->path == NULL || !strcmp(uri->path, "/")) {
1035 error_setg(&err, "missing file path in URI");
1036 goto out;
1037 }
1038 if (g_strlcpy(cfg->vdi, uri->path + 1, SD_MAX_VDI_LEN)
1039 >= SD_MAX_VDI_LEN) {
1040 error_setg(&err, "VDI name is too long");
1041 goto out;
1042 }
1043
1044 cfg->qp = qp = query_params_parse(uri->query);
1045
1046 if (is_unix) {
1047 /* sheepdog+unix:///vdiname?socket=path */
1048 if (uri->server || uri->port) {
1049 error_setg(&err, "URI scheme %s doesn't accept a server address",
1050 uri->scheme);
1051 goto out;
1052 }
1053 if (!qp->n) {
1054 error_setg(&err,
1055 "URI scheme %s requires query parameter 'socket'",
1056 uri->scheme);
1057 goto out;
1058 }
1059 if (qp->n != 1 || strcmp(qp->p[0].name, "socket")) {
1060 error_setg(&err, "unexpected query parameters");
1061 goto out;
1062 }
1063 cfg->path = qp->p[0].value;
1064 } else {
1065 /* sheepdog[+tcp]://[host:port]/vdiname */
1066 if (qp->n) {
1067 error_setg(&err, "unexpected query parameters");
1068 goto out;
1069 }
1070 cfg->host = uri->server;
1071 cfg->port = uri->port;
1072 }
1073
1074 /* snapshot tag */
1075 if (uri->fragment) {
1076 if (!sd_parse_snapid_or_tag(uri->fragment,
1077 &cfg->snap_id, cfg->tag)) {
1078 error_setg(&err, "'%s' is not a valid snapshot ID",
1079 uri->fragment);
1080 goto out;
1081 }
1082 } else {
1083 cfg->snap_id = CURRENT_VDI_ID; /* search current vdi */
1084 }
1085
1086 out:
1087 if (err) {
1088 error_propagate(errp, err);
1089 sd_config_done(cfg);
1090 }
1091 }
1092
1093 /*
1094 * Parse a filename (old syntax)
1095 *
1096 * filename must be one of the following formats:
1097 * 1. [vdiname]
1098 * 2. [vdiname]:[snapid]
1099 * 3. [vdiname]:[tag]
1100 * 4. [hostname]:[port]:[vdiname]
1101 * 5. [hostname]:[port]:[vdiname]:[snapid]
1102 * 6. [hostname]:[port]:[vdiname]:[tag]
1103 *
1104 * You can boot from the snapshot images by specifying `snapid` or
1105 * `tag'.
1106 *
1107 * You can run VMs outside the Sheepdog cluster by specifying
1108 * `hostname' and `port' (experimental).
1109 */
1110 static void parse_vdiname(SheepdogConfig *cfg, const char *filename,
1111 Error **errp)
1112 {
1113 Error *err = NULL;
1114 char *p, *q, *uri;
1115 const char *host_spec, *vdi_spec;
1116 int nr_sep;
1117
1118 strstart(filename, "sheepdog:", &filename);
1119 p = q = g_strdup(filename);
1120
1121 /* count the number of separators */
1122 nr_sep = 0;
1123 while (*p) {
1124 if (*p == ':') {
1125 nr_sep++;
1126 }
1127 p++;
1128 }
1129 p = q;
1130
1131 /* use the first two tokens as host_spec. */
1132 if (nr_sep >= 2) {
1133 host_spec = p;
1134 p = strchr(p, ':');
1135 p++;
1136 p = strchr(p, ':');
1137 *p++ = '\0';
1138 } else {
1139 host_spec = "";
1140 }
1141
1142 vdi_spec = p;
1143
1144 p = strchr(vdi_spec, ':');
1145 if (p) {
1146 *p++ = '#';
1147 }
1148
1149 uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec);
1150
1151 /*
1152 * FIXME We to escape URI meta-characters, e.g. "x?y=z"
1153 * produces "sheepdog://x?y=z". Because of that ...
1154 */
1155 sd_parse_uri(cfg, uri, &err);
1156 if (err) {
1157 /*
1158 * ... this can fail, but the error message is misleading.
1159 * Replace it by the traditional useless one until the
1160 * escaping is fixed.
1161 */
1162 error_free(err);
1163 error_setg(errp, "Can't parse filename");
1164 }
1165
1166 g_free(q);
1167 g_free(uri);
1168 }
1169
1170 static void sd_parse_filename(const char *filename, QDict *options,
1171 Error **errp)
1172 {
1173 Error *err = NULL;
1174 SheepdogConfig cfg;
1175 char buf[32];
1176
1177 if (strstr(filename, "://")) {
1178 sd_parse_uri(&cfg, filename, &err);
1179 } else {
1180 parse_vdiname(&cfg, filename, &err);
1181 }
1182 if (err) {
1183 error_propagate(errp, err);
1184 return;
1185 }
1186
1187 if (cfg.path) {
1188 qdict_set_default_str(options, "server.path", cfg.path);
1189 qdict_set_default_str(options, "server.type", "unix");
1190 } else {
1191 qdict_set_default_str(options, "server.type", "inet");
1192 qdict_set_default_str(options, "server.host",
1193 cfg.host ?: SD_DEFAULT_ADDR);
1194 snprintf(buf, sizeof(buf), "%d", cfg.port ?: SD_DEFAULT_PORT);
1195 qdict_set_default_str(options, "server.port", buf);
1196 }
1197 qdict_set_default_str(options, "vdi", cfg.vdi);
1198 qdict_set_default_str(options, "tag", cfg.tag);
1199 if (cfg.snap_id) {
1200 snprintf(buf, sizeof(buf), "%d", cfg.snap_id);
1201 qdict_set_default_str(options, "snap-id", buf);
1202 }
1203
1204 sd_config_done(&cfg);
1205 }
1206
1207 static int find_vdi_name(BDRVSheepdogState *s, const char *filename,
1208 uint32_t snapid, const char *tag, uint32_t *vid,
1209 bool lock, Error **errp)
1210 {
1211 int ret, fd;
1212 SheepdogVdiReq hdr;
1213 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1214 unsigned int wlen, rlen = 0;
1215 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN] QEMU_NONSTRING;
1216
1217 fd = connect_to_sdog(s, errp);
1218 if (fd < 0) {
1219 return fd;
1220 }
1221
1222 /* This pair of strncpy calls ensures that the buffer is zero-filled,
1223 * which is desirable since we'll soon be sending those bytes, and
1224 * don't want the send_req to read uninitialized data.
1225 */
1226 strncpy(buf, filename, SD_MAX_VDI_LEN);
1227 strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
1228
1229 memset(&hdr, 0, sizeof(hdr));
1230 if (lock) {
1231 hdr.opcode = SD_OP_LOCK_VDI;
1232 hdr.type = LOCK_TYPE_NORMAL;
1233 } else {
1234 hdr.opcode = SD_OP_GET_VDI_INFO;
1235 }
1236 wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
1237 hdr.proto_ver = SD_PROTO_VER;
1238 hdr.data_length = wlen;
1239 hdr.snapid = snapid;
1240 hdr.flags = SD_FLAG_CMD_WRITE;
1241
1242 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1243 if (ret) {
1244 error_setg_errno(errp, -ret, "cannot get vdi info");
1245 goto out;
1246 }
1247
1248 if (rsp->result != SD_RES_SUCCESS) {
1249 error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s",
1250 sd_strerror(rsp->result), filename, snapid, tag);
1251 if (rsp->result == SD_RES_NO_VDI) {
1252 ret = -ENOENT;
1253 } else if (rsp->result == SD_RES_VDI_LOCKED) {
1254 ret = -EBUSY;
1255 } else {
1256 ret = -EIO;
1257 }
1258 goto out;
1259 }
1260 *vid = rsp->vdi_id;
1261
1262 ret = 0;
1263 out:
1264 closesocket(fd);
1265 return ret;
1266 }
1267
1268 static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
1269 struct iovec *iov, int niov,
1270 enum AIOCBState aiocb_type)
1271 {
1272 int nr_copies = s->inode.nr_copies;
1273 SheepdogObjReq hdr;
1274 unsigned int wlen = 0;
1275 int ret;
1276 uint64_t oid = aio_req->oid;
1277 unsigned int datalen = aio_req->data_len;
1278 uint64_t offset = aio_req->offset;
1279 uint8_t flags = aio_req->flags;
1280 uint64_t old_oid = aio_req->base_oid;
1281 bool create = aio_req->create;
1282
1283 qemu_co_mutex_lock(&s->queue_lock);
1284 QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1285 qemu_co_mutex_unlock(&s->queue_lock);
1286
1287 if (!nr_copies) {
1288 error_report("bug");
1289 }
1290
1291 memset(&hdr, 0, sizeof(hdr));
1292
1293 switch (aiocb_type) {
1294 case AIOCB_FLUSH_CACHE:
1295 hdr.opcode = SD_OP_FLUSH_VDI;
1296 break;
1297 case AIOCB_READ_UDATA:
1298 hdr.opcode = SD_OP_READ_OBJ;
1299 hdr.flags = flags;
1300 break;
1301 case AIOCB_WRITE_UDATA:
1302 if (create) {
1303 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1304 } else {
1305 hdr.opcode = SD_OP_WRITE_OBJ;
1306 }
1307 wlen = datalen;
1308 hdr.flags = SD_FLAG_CMD_WRITE | flags;
1309 break;
1310 case AIOCB_DISCARD_OBJ:
1311 hdr.opcode = SD_OP_WRITE_OBJ;
1312 hdr.flags = SD_FLAG_CMD_WRITE | flags;
1313 s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0;
1314 offset = offsetof(SheepdogInode,
1315 data_vdi_id[data_oid_to_idx(oid)]);
1316 oid = vid_to_vdi_oid(s->inode.vdi_id);
1317 wlen = datalen = sizeof(uint32_t);
1318 break;
1319 }
1320
1321 if (s->cache_flags) {
1322 hdr.flags |= s->cache_flags;
1323 }
1324
1325 hdr.oid = oid;
1326 hdr.cow_oid = old_oid;
1327 hdr.copies = s->inode.nr_copies;
1328
1329 hdr.data_length = datalen;
1330 hdr.offset = offset;
1331
1332 hdr.id = aio_req->id;
1333
1334 qemu_co_mutex_lock(&s->lock);
1335 s->co_send = qemu_coroutine_self();
1336 aio_set_fd_handler(s->aio_context, s->fd, false,
1337 co_read_response, co_write_request, NULL, s);
1338 socket_set_cork(s->fd, 1);
1339
1340 /* send a header */
1341 ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
1342 if (ret != sizeof(hdr)) {
1343 error_report("failed to send a req, %s", strerror(errno));
1344 goto out;
1345 }
1346
1347 if (wlen) {
1348 ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1349 if (ret != wlen) {
1350 error_report("failed to send a data, %s", strerror(errno));
1351 }
1352 }
1353 out:
1354 socket_set_cork(s->fd, 0);
1355 aio_set_fd_handler(s->aio_context, s->fd, false,
1356 co_read_response, NULL, NULL, s);
1357 s->co_send = NULL;
1358 qemu_co_mutex_unlock(&s->lock);
1359 }
1360
1361 static int read_write_object(int fd, BlockDriverState *bs, char *buf,
1362 uint64_t oid, uint8_t copies,
1363 unsigned int datalen, uint64_t offset,
1364 bool write, bool create, uint32_t cache_flags)
1365 {
1366 SheepdogObjReq hdr;
1367 SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1368 unsigned int wlen, rlen;
1369 int ret;
1370
1371 memset(&hdr, 0, sizeof(hdr));
1372
1373 if (write) {
1374 wlen = datalen;
1375 rlen = 0;
1376 hdr.flags = SD_FLAG_CMD_WRITE;
1377 if (create) {
1378 hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1379 } else {
1380 hdr.opcode = SD_OP_WRITE_OBJ;
1381 }
1382 } else {
1383 wlen = 0;
1384 rlen = datalen;
1385 hdr.opcode = SD_OP_READ_OBJ;
1386 }
1387
1388 hdr.flags |= cache_flags;
1389
1390 hdr.oid = oid;
1391 hdr.data_length = datalen;
1392 hdr.offset = offset;
1393 hdr.copies = copies;
1394
1395 ret = do_req(fd, bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1396 if (ret) {
1397 error_report("failed to send a request to the sheep");
1398 return ret;
1399 }
1400
1401 switch (rsp->result) {
1402 case SD_RES_SUCCESS:
1403 return 0;
1404 default:
1405 error_report("%s", sd_strerror(rsp->result));
1406 return -EIO;
1407 }
1408 }
1409
1410 static int read_object(int fd, BlockDriverState *bs, char *buf,
1411 uint64_t oid, uint8_t copies,
1412 unsigned int datalen, uint64_t offset,
1413 uint32_t cache_flags)
1414 {
1415 return read_write_object(fd, bs, buf, oid, copies,
1416 datalen, offset, false,
1417 false, cache_flags);
1418 }
1419
1420 static int write_object(int fd, BlockDriverState *bs, char *buf,
1421 uint64_t oid, uint8_t copies,
1422 unsigned int datalen, uint64_t offset, bool create,
1423 uint32_t cache_flags)
1424 {
1425 return read_write_object(fd, bs, buf, oid, copies,
1426 datalen, offset, true,
1427 create, cache_flags);
1428 }
1429
1430 /* update inode with the latest state */
1431 static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
1432 {
1433 Error *local_err = NULL;
1434 SheepdogInode *inode;
1435 int ret = 0, fd;
1436 uint32_t vid = 0;
1437
1438 fd = connect_to_sdog(s, &local_err);
1439 if (fd < 0) {
1440 error_report_err(local_err);
1441 return -EIO;
1442 }
1443
1444 inode = g_malloc(SD_INODE_HEADER_SIZE);
1445
1446 ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err);
1447 if (ret) {
1448 error_report_err(local_err);
1449 goto out;
1450 }
1451
1452 ret = read_object(fd, s->bs, (char *)inode, vid_to_vdi_oid(vid),
1453 s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0,
1454 s->cache_flags);
1455 if (ret < 0) {
1456 goto out;
1457 }
1458
1459 if (inode->vdi_id != s->inode.vdi_id) {
1460 memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE);
1461 }
1462
1463 out:
1464 g_free(inode);
1465 closesocket(fd);
1466
1467 return ret;
1468 }
1469
1470 static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
1471 {
1472 SheepdogAIOCB *acb = aio_req->aiocb;
1473
1474 aio_req->create = false;
1475
1476 /* check whether this request becomes a CoW one */
1477 if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) {
1478 int idx = data_oid_to_idx(aio_req->oid);
1479
1480 if (is_data_obj_writable(&s->inode, idx)) {
1481 goto out;
1482 }
1483
1484 if (s->inode.data_vdi_id[idx]) {
1485 aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx);
1486 aio_req->flags |= SD_FLAG_CMD_COW;
1487 }
1488 aio_req->create = true;
1489 }
1490 out:
1491 if (is_data_obj(aio_req->oid)) {
1492 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
1493 acb->aiocb_type);
1494 } else {
1495 struct iovec iov;
1496 iov.iov_base = &s->inode;
1497 iov.iov_len = sizeof(s->inode);
1498 add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
1499 }
1500 }
1501
1502 static void sd_detach_aio_context(BlockDriverState *bs)
1503 {
1504 BDRVSheepdogState *s = bs->opaque;
1505
1506 aio_set_fd_handler(s->aio_context, s->fd, false, NULL,
1507 NULL, NULL, NULL);
1508 }
1509
1510 static void sd_attach_aio_context(BlockDriverState *bs,
1511 AioContext *new_context)
1512 {
1513 BDRVSheepdogState *s = bs->opaque;
1514
1515 s->aio_context = new_context;
1516 aio_set_fd_handler(new_context, s->fd, false,
1517 co_read_response, NULL, NULL, s);
1518 }
1519
1520 static QemuOptsList runtime_opts = {
1521 .name = "sheepdog",
1522 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1523 .desc = {
1524 {
1525 .name = "vdi",
1526 .type = QEMU_OPT_STRING,
1527 },
1528 {
1529 .name = "snap-id",
1530 .type = QEMU_OPT_NUMBER,
1531 },
1532 {
1533 .name = "tag",
1534 .type = QEMU_OPT_STRING,
1535 },
1536 { /* end of list */ }
1537 },
1538 };
1539
1540 static int sd_open(BlockDriverState *bs, QDict *options, int flags,
1541 Error **errp)
1542 {
1543 int ret, fd;
1544 uint32_t vid = 0;
1545 BDRVSheepdogState *s = bs->opaque;
1546 const char *vdi, *snap_id_str, *tag;
1547 uint64_t snap_id;
1548 char *buf = NULL;
1549 QemuOpts *opts;
1550
1551 s->bs = bs;
1552 s->aio_context = bdrv_get_aio_context(bs);
1553
1554 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1555 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1556 ret = -EINVAL;
1557 goto err_no_fd;
1558 }
1559
1560 s->addr = sd_server_config(options, errp);
1561 if (!s->addr) {
1562 ret = -EINVAL;
1563 goto err_no_fd;
1564 }
1565
1566 vdi = qemu_opt_get(opts, "vdi");
1567 snap_id_str = qemu_opt_get(opts, "snap-id");
1568 snap_id = qemu_opt_get_number(opts, "snap-id", CURRENT_VDI_ID);
1569 tag = qemu_opt_get(opts, "tag");
1570
1571 if (!vdi) {
1572 error_setg(errp, "parameter 'vdi' is missing");
1573 ret = -EINVAL;
1574 goto err_no_fd;
1575 }
1576 if (strlen(vdi) >= SD_MAX_VDI_LEN) {
1577 error_setg(errp, "value of parameter 'vdi' is too long");
1578 ret = -EINVAL;
1579 goto err_no_fd;
1580 }
1581
1582 if (snap_id > UINT32_MAX) {
1583 snap_id = 0;
1584 }
1585 if (snap_id_str && !snap_id) {
1586 error_setg(errp, "'snap-id=%s' is not a valid snapshot ID",
1587 snap_id_str);
1588 ret = -EINVAL;
1589 goto err_no_fd;
1590 }
1591
1592 if (!tag) {
1593 tag = "";
1594 }
1595 if (strlen(tag) >= SD_MAX_VDI_TAG_LEN) {
1596 error_setg(errp, "value of parameter 'tag' is too long");
1597 ret = -EINVAL;
1598 goto err_no_fd;
1599 }
1600
1601 QLIST_INIT(&s->inflight_aio_head);
1602 QLIST_INIT(&s->failed_aio_head);
1603 QLIST_INIT(&s->inflight_aiocb_head);
1604
1605 s->fd = get_sheep_fd(s, errp);
1606 if (s->fd < 0) {
1607 ret = s->fd;
1608 goto err_no_fd;
1609 }
1610
1611 ret = find_vdi_name(s, vdi, (uint32_t)snap_id, tag, &vid, true, errp);
1612 if (ret) {
1613 goto err;
1614 }
1615
1616 /*
1617 * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1618 * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1619 */
1620 s->cache_flags = SD_FLAG_CMD_CACHE;
1621 if (flags & BDRV_O_NOCACHE) {
1622 s->cache_flags = SD_FLAG_CMD_DIRECT;
1623 }
1624 s->discard_supported = true;
1625
1626 if (snap_id || tag[0]) {
1627 trace_sheepdog_open(vid);
1628 s->is_snapshot = true;
1629 }
1630
1631 fd = connect_to_sdog(s, errp);
1632 if (fd < 0) {
1633 ret = fd;
1634 goto err;
1635 }
1636
1637 buf = g_malloc(SD_INODE_SIZE);
1638 ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid),
1639 0, SD_INODE_SIZE, 0, s->cache_flags);
1640
1641 closesocket(fd);
1642
1643 if (ret) {
1644 error_setg(errp, "Can't read snapshot inode");
1645 goto err;
1646 }
1647
1648 memcpy(&s->inode, buf, sizeof(s->inode));
1649
1650 bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE;
1651 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1652 pstrcpy(s->name, sizeof(s->name), vdi);
1653 qemu_co_mutex_init(&s->lock);
1654 qemu_co_mutex_init(&s->queue_lock);
1655 qemu_co_queue_init(&s->overlapping_queue);
1656 qemu_opts_del(opts);
1657 g_free(buf);
1658 return 0;
1659
1660 err:
1661 aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd,
1662 false, NULL, NULL, NULL, NULL);
1663 closesocket(s->fd);
1664 err_no_fd:
1665 qemu_opts_del(opts);
1666 g_free(buf);
1667 return ret;
1668 }
1669
1670 static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue,
1671 Error **errp)
1672 {
1673 BDRVSheepdogState *s = state->bs->opaque;
1674 BDRVSheepdogReopenState *re_s;
1675 int ret = 0;
1676
1677 re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1);
1678
1679 re_s->cache_flags = SD_FLAG_CMD_CACHE;
1680 if (state->flags & BDRV_O_NOCACHE) {
1681 re_s->cache_flags = SD_FLAG_CMD_DIRECT;
1682 }
1683
1684 re_s->fd = get_sheep_fd(s, errp);
1685 if (re_s->fd < 0) {
1686 ret = re_s->fd;
1687 return ret;
1688 }
1689
1690 return ret;
1691 }
1692
1693 static void sd_reopen_commit(BDRVReopenState *state)
1694 {
1695 BDRVSheepdogReopenState *re_s = state->opaque;
1696 BDRVSheepdogState *s = state->bs->opaque;
1697
1698 if (s->fd) {
1699 aio_set_fd_handler(s->aio_context, s->fd, false,
1700 NULL, NULL, NULL, NULL);
1701 closesocket(s->fd);
1702 }
1703
1704 s->fd = re_s->fd;
1705 s->cache_flags = re_s->cache_flags;
1706
1707 g_free(state->opaque);
1708 state->opaque = NULL;
1709
1710 return;
1711 }
1712
1713 static void sd_reopen_abort(BDRVReopenState *state)
1714 {
1715 BDRVSheepdogReopenState *re_s = state->opaque;
1716 BDRVSheepdogState *s = state->bs->opaque;
1717
1718 if (re_s == NULL) {
1719 return;
1720 }
1721
1722 if (re_s->fd) {
1723 aio_set_fd_handler(s->aio_context, re_s->fd, false,
1724 NULL, NULL, NULL, NULL);
1725 closesocket(re_s->fd);
1726 }
1727
1728 g_free(state->opaque);
1729 state->opaque = NULL;
1730
1731 return;
1732 }
1733
1734 static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot,
1735 Error **errp)
1736 {
1737 SheepdogVdiReq hdr;
1738 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1739 int fd, ret;
1740 unsigned int wlen, rlen = 0;
1741 char buf[SD_MAX_VDI_LEN];
1742
1743 fd = connect_to_sdog(s, errp);
1744 if (fd < 0) {
1745 return fd;
1746 }
1747
1748 /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1749 * does not fit in buf? For now, just truncate and avoid buffer overrun.
1750 */
1751 memset(buf, 0, sizeof(buf));
1752 pstrcpy(buf, sizeof(buf), s->name);
1753
1754 memset(&hdr, 0, sizeof(hdr));
1755 hdr.opcode = SD_OP_NEW_VDI;
1756 hdr.base_vdi_id = s->inode.vdi_id;
1757
1758 wlen = SD_MAX_VDI_LEN;
1759
1760 hdr.flags = SD_FLAG_CMD_WRITE;
1761 hdr.snapid = snapshot;
1762
1763 hdr.data_length = wlen;
1764 hdr.vdi_size = s->inode.vdi_size;
1765 hdr.copy_policy = s->inode.copy_policy;
1766 hdr.copies = s->inode.nr_copies;
1767 hdr.block_size_shift = s->inode.block_size_shift;
1768
1769 ret = do_req(fd, NULL, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1770
1771 closesocket(fd);
1772
1773 if (ret) {
1774 error_setg_errno(errp, -ret, "create failed");
1775 return ret;
1776 }
1777
1778 if (rsp->result != SD_RES_SUCCESS) {
1779 error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name);
1780 return -EIO;
1781 }
1782
1783 if (vdi_id) {
1784 *vdi_id = rsp->vdi_id;
1785 }
1786
1787 return 0;
1788 }
1789
1790 static int sd_prealloc(BlockDriverState *bs, int64_t old_size, int64_t new_size,
1791 Error **errp)
1792 {
1793 BlockBackend *blk = NULL;
1794 BDRVSheepdogState *base = bs->opaque;
1795 unsigned long buf_size;
1796 uint32_t idx, max_idx;
1797 uint32_t object_size;
1798 void *buf = NULL;
1799 int ret;
1800
1801 blk = blk_new_with_bs(bs,
1802 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
1803 BLK_PERM_ALL, errp);
1804
1805 if (!blk) {
1806 ret = -EPERM;
1807 goto out_with_err_set;
1808 }
1809
1810 blk_set_allow_write_beyond_eof(blk, true);
1811
1812 object_size = (UINT32_C(1) << base->inode.block_size_shift);
1813 buf_size = MIN(object_size, SD_DATA_OBJ_SIZE);
1814 buf = g_malloc0(buf_size);
1815
1816 max_idx = DIV_ROUND_UP(new_size, buf_size);
1817
1818 for (idx = old_size / buf_size; idx < max_idx; idx++) {
1819 /*
1820 * The created image can be a cloned image, so we need to read
1821 * a data from the source image.
1822 */
1823 ret = blk_pread(blk, idx * buf_size, buf, buf_size);
1824 if (ret < 0) {
1825 goto out;
1826 }
1827 ret = blk_pwrite(blk, idx * buf_size, buf, buf_size, 0);
1828 if (ret < 0) {
1829 goto out;
1830 }
1831 }
1832
1833 ret = 0;
1834 out:
1835 if (ret < 0) {
1836 error_setg_errno(errp, -ret, "Can't pre-allocate");
1837 }
1838 out_with_err_set:
1839 blk_unref(blk);
1840 g_free(buf);
1841
1842 return ret;
1843 }
1844
1845 static int sd_create_prealloc(BlockdevOptionsSheepdog *location, int64_t size,
1846 Error **errp)
1847 {
1848 BlockDriverState *bs;
1849 Visitor *v;
1850 QObject *obj = NULL;
1851 QDict *qdict;
1852 int ret;
1853
1854 v = qobject_output_visitor_new(&obj);
1855 visit_type_BlockdevOptionsSheepdog(v, NULL, &location, &error_abort);
1856 visit_free(v);
1857
1858 qdict = qobject_to(QDict, obj);
1859 qdict_flatten(qdict);
1860
1861 qdict_put_str(qdict, "driver", "sheepdog");
1862
1863 bs = bdrv_open(NULL, NULL, qdict, BDRV_O_PROTOCOL | BDRV_O_RDWR, errp);
1864 if (bs == NULL) {
1865 ret = -EIO;
1866 goto fail;
1867 }
1868
1869 ret = sd_prealloc(bs, 0, size, errp);
1870 fail:
1871 bdrv_unref(bs);
1872 qobject_unref(qdict);
1873 return ret;
1874 }
1875
1876 static int parse_redundancy(BDRVSheepdogState *s, SheepdogRedundancy *opt)
1877 {
1878 struct SheepdogInode *inode = &s->inode;
1879
1880 switch (opt->type) {
1881 case SHEEPDOG_REDUNDANCY_TYPE_FULL:
1882 if (opt->u.full.copies > SD_MAX_COPIES || opt->u.full.copies < 1) {
1883 return -EINVAL;
1884 }
1885 inode->copy_policy = 0;
1886 inode->nr_copies = opt->u.full.copies;
1887 return 0;
1888
1889 case SHEEPDOG_REDUNDANCY_TYPE_ERASURE_CODED:
1890 {
1891 int64_t copy = opt->u.erasure_coded.data_strips;
1892 int64_t parity = opt->u.erasure_coded.parity_strips;
1893
1894 if (copy != 2 && copy != 4 && copy != 8 && copy != 16) {
1895 return -EINVAL;
1896 }
1897
1898 if (parity >= SD_EC_MAX_STRIP || parity < 1) {
1899 return -EINVAL;
1900 }
1901
1902 /*
1903 * 4 bits for parity and 4 bits for data.
1904 * We have to compress upper data bits because it can't represent 16
1905 */
1906 inode->copy_policy = ((copy / 2) << 4) + parity;
1907 inode->nr_copies = copy + parity;
1908 return 0;
1909 }
1910
1911 default:
1912 g_assert_not_reached();
1913 }
1914
1915 return -EINVAL;
1916 }
1917
1918 /*
1919 * Sheepdog support two kinds of redundancy, full replication and erasure
1920 * coding.
1921 *
1922 * # create a fully replicated vdi with x copies
1923 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1924 *
1925 * # create a erasure coded vdi with x data strips and y parity strips
1926 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1927 */
1928 static SheepdogRedundancy *parse_redundancy_str(const char *opt)
1929 {
1930 SheepdogRedundancy *redundancy;
1931 const char *n1, *n2;
1932 long copy, parity;
1933 char p[10];
1934 int ret;
1935
1936 pstrcpy(p, sizeof(p), opt);
1937 n1 = strtok(p, ":");
1938 n2 = strtok(NULL, ":");
1939
1940 if (!n1) {
1941 return NULL;
1942 }
1943
1944 ret = qemu_strtol(n1, NULL, 10, &copy);
1945 if (ret < 0) {
1946 return NULL;
1947 }
1948
1949 redundancy = g_new0(SheepdogRedundancy, 1);
1950 if (!n2) {
1951 *redundancy = (SheepdogRedundancy) {
1952 .type = SHEEPDOG_REDUNDANCY_TYPE_FULL,
1953 .u.full.copies = copy,
1954 };
1955 } else {
1956 ret = qemu_strtol(n2, NULL, 10, &parity);
1957 if (ret < 0) {
1958 g_free(redundancy);
1959 return NULL;
1960 }
1961
1962 *redundancy = (SheepdogRedundancy) {
1963 .type = SHEEPDOG_REDUNDANCY_TYPE_ERASURE_CODED,
1964 .u.erasure_coded = {
1965 .data_strips = copy,
1966 .parity_strips = parity,
1967 },
1968 };
1969 }
1970
1971 return redundancy;
1972 }
1973
1974 static int parse_block_size_shift(BDRVSheepdogState *s,
1975 BlockdevCreateOptionsSheepdog *opts)
1976 {
1977 struct SheepdogInode *inode = &s->inode;
1978 uint64_t object_size;
1979 int obj_order;
1980
1981 if (opts->has_object_size) {
1982 object_size = opts->object_size;
1983
1984 if ((object_size - 1) & object_size) { /* not a power of 2? */
1985 return -EINVAL;
1986 }
1987 obj_order = ctz32(object_size);
1988 if (obj_order < 20 || obj_order > 31) {
1989 return -EINVAL;
1990 }
1991 inode->block_size_shift = (uint8_t)obj_order;
1992 }
1993
1994 return 0;
1995 }
1996
1997 static int sd_co_create(BlockdevCreateOptions *options, Error **errp)
1998 {
1999 BlockdevCreateOptionsSheepdog *opts = &options->u.sheepdog;
2000 int ret = 0;
2001 uint32_t vid = 0;
2002 char *backing_file = NULL;
2003 char *buf = NULL;
2004 BDRVSheepdogState *s;
2005 uint64_t max_vdi_size;
2006 bool prealloc = false;
2007
2008 assert(options->driver == BLOCKDEV_DRIVER_SHEEPDOG);
2009
2010 s = g_new0(BDRVSheepdogState, 1);
2011
2012 /* Steal SocketAddress from QAPI, set NULL to prevent double free */
2013 s->addr = opts->location->server;
2014 opts->location->server = NULL;
2015
2016 if (strlen(opts->location->vdi) >= sizeof(s->name)) {
2017 error_setg(errp, "'vdi' string too long");
2018 ret = -EINVAL;
2019 goto out;
2020 }
2021 pstrcpy(s->name, sizeof(s->name), opts->location->vdi);
2022
2023 s->inode.vdi_size = opts->size;
2024 backing_file = opts->backing_file;
2025
2026 if (!opts->has_preallocation) {
2027 opts->preallocation = PREALLOC_MODE_OFF;
2028 }
2029 switch (opts->preallocation) {
2030 case PREALLOC_MODE_OFF:
2031 prealloc = false;
2032 break;
2033 case PREALLOC_MODE_FULL:
2034 prealloc = true;
2035 break;
2036 default:
2037 error_setg(errp, "Preallocation mode not supported for Sheepdog");
2038 ret = -EINVAL;
2039 goto out;
2040 }
2041
2042 if (opts->has_redundancy) {
2043 ret = parse_redundancy(s, opts->redundancy);
2044 if (ret < 0) {
2045 error_setg(errp, "Invalid redundancy mode");
2046 goto out;
2047 }
2048 }
2049 ret = parse_block_size_shift(s, opts);
2050 if (ret < 0) {
2051 error_setg(errp, "Invalid object_size."
2052 " obect_size needs to be power of 2"
2053 " and be limited from 2^20 to 2^31");
2054 goto out;
2055 }
2056
2057 if (opts->has_backing_file) {
2058 BlockBackend *blk;
2059 BDRVSheepdogState *base;
2060 BlockDriver *drv;
2061
2062 /* Currently, only Sheepdog backing image is supported. */
2063 drv = bdrv_find_protocol(opts->backing_file, true, NULL);
2064 if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
2065 error_setg(errp, "backing_file must be a sheepdog image");
2066 ret = -EINVAL;
2067 goto out;
2068 }
2069
2070 blk = blk_new_open(opts->backing_file, NULL, NULL,
2071 BDRV_O_PROTOCOL, errp);
2072 if (blk == NULL) {
2073 ret = -EIO;
2074 goto out;
2075 }
2076
2077 base = blk_bs(blk)->opaque;
2078
2079 if (!is_snapshot(&base->inode)) {
2080 error_setg(errp, "cannot clone from a non snapshot vdi");
2081 blk_unref(blk);
2082 ret = -EINVAL;
2083 goto out;
2084 }
2085 s->inode.vdi_id = base->inode.vdi_id;
2086 blk_unref(blk);
2087 }
2088
2089 s->aio_context = qemu_get_aio_context();
2090
2091 /* if block_size_shift is not specified, get cluster default value */
2092 if (s->inode.block_size_shift == 0) {
2093 SheepdogVdiReq hdr;
2094 SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr;
2095 int fd;
2096 unsigned int wlen = 0, rlen = 0;
2097
2098 fd = connect_to_sdog(s, errp);
2099 if (fd < 0) {
2100 ret = fd;
2101 goto out;
2102 }
2103
2104 memset(&hdr, 0, sizeof(hdr));
2105 hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT;
2106 hdr.proto_ver = SD_PROTO_VER;
2107
2108 ret = do_req(fd, NULL, (SheepdogReq *)&hdr,
2109 NULL, &wlen, &rlen);
2110 closesocket(fd);
2111 if (ret) {
2112 error_setg_errno(errp, -ret, "failed to get cluster default");
2113 goto out;
2114 }
2115 if (rsp->result == SD_RES_SUCCESS) {
2116 s->inode.block_size_shift = rsp->block_size_shift;
2117 } else {
2118 s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT;
2119 }
2120 }
2121
2122 max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS;
2123
2124 if (s->inode.vdi_size > max_vdi_size) {
2125 error_setg(errp, "An image is too large."
2126 " The maximum image size is %"PRIu64 "GB",
2127 max_vdi_size / 1024 / 1024 / 1024);
2128 ret = -EINVAL;
2129 goto out;
2130 }
2131
2132 ret = do_sd_create(s, &vid, 0, errp);
2133 if (ret) {
2134 goto out;
2135 }
2136
2137 if (prealloc) {
2138 ret = sd_create_prealloc(opts->location, opts->size, errp);
2139 }
2140 out:
2141 g_free(backing_file);
2142 g_free(buf);
2143 g_free(s->addr);
2144 g_free(s);
2145 return ret;
2146 }
2147
2148 static int coroutine_fn sd_co_create_opts(BlockDriver *drv,
2149 const char *filename,
2150 QemuOpts *opts,
2151 Error **errp)
2152 {
2153 BlockdevCreateOptions *create_options = NULL;
2154 QDict *qdict = NULL, *location_qdict;
2155 Visitor *v;
2156 char *redundancy = NULL;
2157 Error *local_err = NULL;
2158 int ret;
2159 char *backing_fmt = NULL;
2160
2161 redundancy = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY);
2162 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2163
2164 if (backing_fmt && strcmp(backing_fmt, "sheepdog") != 0) {
2165 error_setg(errp, "backing_file must be a sheepdog image");
2166 ret = -EINVAL;
2167 goto fail;
2168 }
2169
2170 qdict = qemu_opts_to_qdict(opts, NULL);
2171 qdict_put_str(qdict, "driver", "sheepdog");
2172
2173 location_qdict = qdict_new();
2174 qdict_put(qdict, "location", location_qdict);
2175
2176 sd_parse_filename(filename, location_qdict, &local_err);
2177 if (local_err) {
2178 error_propagate(errp, local_err);
2179 ret = -EINVAL;
2180 goto fail;
2181 }
2182
2183 qdict_flatten(qdict);
2184
2185 /* Change legacy command line options into QMP ones */
2186 static const QDictRenames opt_renames[] = {
2187 { BLOCK_OPT_BACKING_FILE, "backing-file" },
2188 { BLOCK_OPT_OBJECT_SIZE, "object-size" },
2189 { NULL, NULL },
2190 };
2191
2192 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
2193 ret = -EINVAL;
2194 goto fail;
2195 }
2196
2197 /* Get the QAPI object */
2198 v = qobject_input_visitor_new_flat_confused(qdict, errp);
2199 if (!v) {
2200 ret = -EINVAL;
2201 goto fail;
2202 }
2203
2204 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
2205 visit_free(v);
2206 if (!create_options) {
2207 ret = -EINVAL;
2208 goto fail;
2209 }
2210
2211 assert(create_options->driver == BLOCKDEV_DRIVER_SHEEPDOG);
2212 create_options->u.sheepdog.size =
2213 ROUND_UP(create_options->u.sheepdog.size, BDRV_SECTOR_SIZE);
2214
2215 if (redundancy) {
2216 create_options->u.sheepdog.has_redundancy = true;
2217 create_options->u.sheepdog.redundancy =
2218 parse_redundancy_str(redundancy);
2219 if (create_options->u.sheepdog.redundancy == NULL) {
2220 error_setg(errp, "Invalid redundancy mode");
2221 ret = -EINVAL;
2222 goto fail;
2223 }
2224 }
2225
2226 ret = sd_co_create(create_options, errp);
2227 fail:
2228 qapi_free_BlockdevCreateOptions(create_options);
2229 qobject_unref(qdict);
2230 g_free(redundancy);
2231 g_free(backing_fmt);
2232 return ret;
2233 }
2234
2235 static void sd_close(BlockDriverState *bs)
2236 {
2237 Error *local_err = NULL;
2238 BDRVSheepdogState *s = bs->opaque;
2239 SheepdogVdiReq hdr;
2240 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2241 unsigned int wlen, rlen = 0;
2242 int fd, ret;
2243
2244 trace_sheepdog_close(s->name);
2245
2246 fd = connect_to_sdog(s, &local_err);
2247 if (fd < 0) {
2248 error_report_err(local_err);
2249 return;
2250 }
2251
2252 memset(&hdr, 0, sizeof(hdr));
2253
2254 hdr.opcode = SD_OP_RELEASE_VDI;
2255 hdr.type = LOCK_TYPE_NORMAL;
2256 hdr.base_vdi_id = s->inode.vdi_id;
2257 wlen = strlen(s->name) + 1;
2258 hdr.data_length = wlen;
2259 hdr.flags = SD_FLAG_CMD_WRITE;
2260
2261 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2262 s->name, &wlen, &rlen);
2263
2264 closesocket(fd);
2265
2266 if (!ret && rsp->result != SD_RES_SUCCESS &&
2267 rsp->result != SD_RES_VDI_NOT_LOCKED) {
2268 error_report("%s, %s", sd_strerror(rsp->result), s->name);
2269 }
2270
2271 aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd,
2272 false, NULL, NULL, NULL, NULL);
2273 closesocket(s->fd);
2274 qapi_free_SocketAddress(s->addr);
2275 }
2276
2277 static int64_t sd_getlength(BlockDriverState *bs)
2278 {
2279 BDRVSheepdogState *s = bs->opaque;
2280
2281 return s->inode.vdi_size;
2282 }
2283
2284 static int coroutine_fn sd_co_truncate(BlockDriverState *bs, int64_t offset,
2285 bool exact, PreallocMode prealloc,
2286 BdrvRequestFlags flags, Error **errp)
2287 {
2288 BDRVSheepdogState *s = bs->opaque;
2289 int ret, fd;
2290 unsigned int datalen;
2291 uint64_t max_vdi_size;
2292 int64_t old_size = s->inode.vdi_size;
2293
2294 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_FULL) {
2295 error_setg(errp, "Unsupported preallocation mode '%s'",
2296 PreallocMode_str(prealloc));
2297 return -ENOTSUP;
2298 }
2299
2300 max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS;
2301 if (offset < old_size) {
2302 error_setg(errp, "shrinking is not supported");
2303 return -EINVAL;
2304 } else if (offset > max_vdi_size) {
2305 error_setg(errp, "too big image size");
2306 return -EINVAL;
2307 }
2308
2309 fd = connect_to_sdog(s, errp);
2310 if (fd < 0) {
2311 return fd;
2312 }
2313
2314 /* we don't need to update entire object */
2315 datalen = SD_INODE_HEADER_SIZE;
2316 s->inode.vdi_size = offset;
2317 ret = write_object(fd, s->bs, (char *)&s->inode,
2318 vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
2319 datalen, 0, false, s->cache_flags);
2320 close(fd);
2321
2322 if (ret < 0) {
2323 error_setg_errno(errp, -ret, "failed to update an inode");
2324 return ret;
2325 }
2326
2327 if (prealloc == PREALLOC_MODE_FULL) {
2328 ret = sd_prealloc(bs, old_size, offset, errp);
2329 if (ret < 0) {
2330 return ret;
2331 }
2332 }
2333
2334 return 0;
2335 }
2336
2337 /*
2338 * This function is called after writing data objects. If we need to
2339 * update metadata, this sends a write request to the vdi object.
2340 */
2341 static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
2342 {
2343 BDRVSheepdogState *s = acb->s;
2344 struct iovec iov;
2345 AIOReq *aio_req;
2346 uint32_t offset, data_len, mn, mx;
2347
2348 mn = acb->min_dirty_data_idx;
2349 mx = acb->max_dirty_data_idx;
2350 if (mn <= mx) {
2351 /* we need to update the vdi object. */
2352 ++acb->nr_pending;
2353 offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
2354 mn * sizeof(s->inode.data_vdi_id[0]);
2355 data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
2356
2357 acb->min_dirty_data_idx = UINT32_MAX;
2358 acb->max_dirty_data_idx = 0;
2359
2360 iov.iov_base = &s->inode;
2361 iov.iov_len = sizeof(s->inode);
2362 aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
2363 data_len, offset, 0, false, 0, offset);
2364 add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
2365 if (--acb->nr_pending) {
2366 qemu_coroutine_yield();
2367 }
2368 }
2369 }
2370
2371 /* Delete current working VDI on the snapshot chain */
2372 static bool sd_delete(BDRVSheepdogState *s)
2373 {
2374 Error *local_err = NULL;
2375 unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0;
2376 SheepdogVdiReq hdr = {
2377 .opcode = SD_OP_DEL_VDI,
2378 .base_vdi_id = s->inode.vdi_id,
2379 .data_length = wlen,
2380 .flags = SD_FLAG_CMD_WRITE,
2381 };
2382 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2383 int fd, ret;
2384
2385 fd = connect_to_sdog(s, &local_err);
2386 if (fd < 0) {
2387 error_report_err(local_err);
2388 return false;
2389 }
2390
2391 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2392 s->name, &wlen, &rlen);
2393 closesocket(fd);
2394 if (ret) {
2395 return false;
2396 }
2397 switch (rsp->result) {
2398 case SD_RES_NO_VDI:
2399 error_report("%s was already deleted", s->name);
2400 /* fall through */
2401 case SD_RES_SUCCESS:
2402 break;
2403 default:
2404 error_report("%s, %s", sd_strerror(rsp->result), s->name);
2405 return false;
2406 }
2407
2408 return true;
2409 }
2410
2411 /*
2412 * Create a writable VDI from a snapshot
2413 */
2414 static int sd_create_branch(BDRVSheepdogState *s)
2415 {
2416 Error *local_err = NULL;
2417 int ret, fd;
2418 uint32_t vid;
2419 char *buf;
2420 bool deleted;
2421
2422 trace_sheepdog_create_branch_snapshot(s->inode.vdi_id);
2423
2424 buf = g_malloc(SD_INODE_SIZE);
2425
2426 /*
2427 * Even If deletion fails, we will just create extra snapshot based on
2428 * the working VDI which was supposed to be deleted. So no need to
2429 * false bail out.
2430 */
2431 deleted = sd_delete(s);
2432 ret = do_sd_create(s, &vid, !deleted, &local_err);
2433 if (ret) {
2434 error_report_err(local_err);
2435 goto out;
2436 }
2437
2438 trace_sheepdog_create_branch_created(vid);
2439
2440 fd = connect_to_sdog(s, &local_err);
2441 if (fd < 0) {
2442 error_report_err(local_err);
2443 ret = fd;
2444 goto out;
2445 }
2446
2447 ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid),
2448 s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags);
2449
2450 closesocket(fd);
2451
2452 if (ret < 0) {
2453 goto out;
2454 }
2455
2456 memcpy(&s->inode, buf, sizeof(s->inode));
2457
2458 s->is_snapshot = false;
2459 ret = 0;
2460 trace_sheepdog_create_branch_new(s->inode.vdi_id);
2461
2462 out:
2463 g_free(buf);
2464
2465 return ret;
2466 }
2467
2468 /*
2469 * Send I/O requests to the server.
2470 *
2471 * This function sends requests to the server, links the requests to
2472 * the inflight_list in BDRVSheepdogState, and exits without
2473 * waiting the response. The responses are received in the
2474 * `aio_read_response' function which is called from the main loop as
2475 * a fd handler.
2476 *
2477 * Returns 1 when we need to wait a response, 0 when there is no sent
2478 * request and -errno in error cases.
2479 */
2480 static void coroutine_fn sd_co_rw_vector(SheepdogAIOCB *acb)
2481 {
2482 int ret = 0;
2483 unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE;
2484 unsigned long idx;
2485 uint32_t object_size;
2486 uint64_t oid;
2487 uint64_t offset;
2488 BDRVSheepdogState *s = acb->s;
2489 SheepdogInode *inode = &s->inode;
2490 AIOReq *aio_req;
2491
2492 if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
2493 /*
2494 * In the case we open the snapshot VDI, Sheepdog creates the
2495 * writable VDI when we do a write operation first.
2496 */
2497 ret = sd_create_branch(s);
2498 if (ret) {
2499 acb->ret = -EIO;
2500 return;
2501 }
2502 }
2503
2504 object_size = (UINT32_C(1) << inode->block_size_shift);
2505 idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size;
2506 offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size;
2507
2508 /*
2509 * Make sure we don't free the aiocb before we are done with all requests.
2510 * This additional reference is dropped at the end of this function.
2511 */
2512 acb->nr_pending++;
2513
2514 while (done != total) {
2515 uint8_t flags = 0;
2516 uint64_t old_oid = 0;
2517 bool create = false;
2518
2519 oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
2520
2521 len = MIN(total - done, object_size - offset);
2522
2523 switch (acb->aiocb_type) {
2524 case AIOCB_READ_UDATA:
2525 if (!inode->data_vdi_id[idx]) {
2526 qemu_iovec_memset(acb->qiov, done, 0, len);
2527 goto done;
2528 }
2529 break;
2530 case AIOCB_WRITE_UDATA:
2531 if (!inode->data_vdi_id[idx]) {
2532 create = true;
2533 } else if (!is_data_obj_writable(inode, idx)) {
2534 /* Copy-On-Write */
2535 create = true;
2536 old_oid = oid;
2537 flags = SD_FLAG_CMD_COW;
2538 }
2539 break;
2540 case AIOCB_DISCARD_OBJ:
2541 /*
2542 * We discard the object only when the whole object is
2543 * 1) allocated 2) trimmed. Otherwise, simply skip it.
2544 */
2545 if (len != object_size || inode->data_vdi_id[idx] == 0) {
2546 goto done;
2547 }
2548 break;
2549 default:
2550 break;
2551 }
2552
2553 if (create) {
2554 trace_sheepdog_co_rw_vector_update(inode->vdi_id, oid,
2555 vid_to_data_oid(inode->data_vdi_id[idx], idx),
2556 idx);
2557 oid = vid_to_data_oid(inode->vdi_id, idx);
2558 trace_sheepdog_co_rw_vector_new(oid);
2559 }
2560
2561 aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create,
2562 old_oid,
2563 acb->aiocb_type == AIOCB_DISCARD_OBJ ?
2564 0 : done);
2565 add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
2566 acb->aiocb_type);
2567 done:
2568 offset = 0;
2569 idx++;
2570 done += len;
2571 }
2572 if (--acb->nr_pending) {
2573 qemu_coroutine_yield();
2574 }
2575 }
2576
2577 static void sd_aio_complete(SheepdogAIOCB *acb)
2578 {
2579 BDRVSheepdogState *s;
2580 if (acb->aiocb_type == AIOCB_FLUSH_CACHE) {
2581 return;
2582 }
2583
2584 s = acb->s;
2585 qemu_co_mutex_lock(&s->queue_lock);
2586 QLIST_REMOVE(acb, aiocb_siblings);
2587 qemu_co_queue_restart_all(&s->overlapping_queue);
2588 qemu_co_mutex_unlock(&s->queue_lock);
2589 }
2590
2591 static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
2592 int nb_sectors, QEMUIOVector *qiov,
2593 int flags)
2594 {
2595 SheepdogAIOCB acb;
2596 int ret;
2597 int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE;
2598 BDRVSheepdogState *s = bs->opaque;
2599
2600 assert(!flags);
2601 if (offset > s->inode.vdi_size) {
2602 ret = sd_co_truncate(bs, offset, false, PREALLOC_MODE_OFF, 0, NULL);
2603 if (ret < 0) {
2604 return ret;
2605 }
2606 }
2607
2608 sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_WRITE_UDATA);
2609 sd_co_rw_vector(&acb);
2610 sd_write_done(&acb);
2611 sd_aio_complete(&acb);
2612
2613 return acb.ret;
2614 }
2615
2616 static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
2617 int nb_sectors, QEMUIOVector *qiov)
2618 {
2619 SheepdogAIOCB acb;
2620 BDRVSheepdogState *s = bs->opaque;
2621
2622 sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_READ_UDATA);
2623 sd_co_rw_vector(&acb);
2624 sd_aio_complete(&acb);
2625
2626 return acb.ret;
2627 }
2628
2629 static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
2630 {
2631 BDRVSheepdogState *s = bs->opaque;
2632 SheepdogAIOCB acb;
2633 AIOReq *aio_req;
2634
2635 if (s->cache_flags != SD_FLAG_CMD_CACHE) {
2636 return 0;
2637 }
2638
2639 sd_aio_setup(&acb, s, NULL, 0, 0, AIOCB_FLUSH_CACHE);
2640
2641 acb.nr_pending++;
2642 aio_req = alloc_aio_req(s, &acb, vid_to_vdi_oid(s->inode.vdi_id),
2643 0, 0, 0, false, 0, 0);
2644 add_aio_request(s, aio_req, NULL, 0, acb.aiocb_type);
2645
2646 if (--acb.nr_pending) {
2647 qemu_coroutine_yield();
2648 }
2649
2650 sd_aio_complete(&acb);
2651 return acb.ret;
2652 }
2653
2654 static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
2655 {
2656 Error *local_err = NULL;
2657 BDRVSheepdogState *s = bs->opaque;
2658 int ret, fd;
2659 uint32_t new_vid;
2660 SheepdogInode *inode;
2661 unsigned int datalen;
2662
2663 trace_sheepdog_snapshot_create_info(sn_info->name, sn_info->id_str, s->name,
2664 sn_info->vm_state_size, s->is_snapshot);
2665
2666 if (s->is_snapshot) {
2667 error_report("You can't create a snapshot of a snapshot VDI, "
2668 "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
2669
2670 return -EINVAL;
2671 }
2672
2673 trace_sheepdog_snapshot_create(sn_info->name, sn_info->id_str);
2674
2675 s->inode.vm_state_size = sn_info->vm_state_size;
2676 s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
2677 /* It appears that inode.tag does not require a NUL terminator,
2678 * which means this use of strncpy is ok.
2679 */
2680 strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
2681 /* we don't need to update entire object */
2682 datalen = SD_INODE_HEADER_SIZE;
2683 inode = g_malloc(datalen);
2684
2685 /* refresh inode. */
2686 fd = connect_to_sdog(s, &local_err);
2687 if (fd < 0) {
2688 error_report_err(local_err);
2689 ret = fd;
2690 goto cleanup;
2691 }
2692
2693 ret = write_object(fd, s->bs, (char *)&s->inode,
2694 vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
2695 datalen, 0, false, s->cache_flags);
2696 if (ret < 0) {
2697 error_report("failed to write snapshot's inode.");
2698 goto cleanup;
2699 }
2700
2701 ret = do_sd_create(s, &new_vid, 1, &local_err);
2702 if (ret < 0) {
2703 error_reportf_err(local_err,
2704 "failed to create inode for snapshot: ");
2705 goto cleanup;
2706 }
2707
2708 ret = read_object(fd, s->bs, (char *)inode,
2709 vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0,
2710 s->cache_flags);
2711
2712 if (ret < 0) {
2713 error_report("failed to read new inode info. %s", strerror(errno));
2714 goto cleanup;
2715 }
2716
2717 memcpy(&s->inode, inode, datalen);
2718 trace_sheepdog_snapshot_create_inode(s->inode.name, s->inode.snap_id,
2719 s->inode.vdi_id);
2720
2721 cleanup:
2722 g_free(inode);
2723 closesocket(fd);
2724 return ret;
2725 }
2726
2727 /*
2728 * We implement rollback(loadvm) operation to the specified snapshot by
2729 * 1) switch to the snapshot
2730 * 2) rely on sd_create_branch to delete working VDI and
2731 * 3) create a new working VDI based on the specified snapshot
2732 */
2733 static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
2734 {
2735 BDRVSheepdogState *s = bs->opaque;
2736 BDRVSheepdogState *old_s;
2737 char tag[SD_MAX_VDI_TAG_LEN];
2738 uint32_t snapid = 0;
2739 int ret;
2740
2741 if (!sd_parse_snapid_or_tag(snapshot_id, &snapid, tag)) {
2742 return -EINVAL;
2743 }
2744
2745 old_s = g_new(BDRVSheepdogState, 1);
2746
2747 memcpy(old_s, s, sizeof(BDRVSheepdogState));
2748
2749 ret = reload_inode(s, snapid, tag);
2750 if (ret) {
2751 goto out;
2752 }
2753
2754 ret = sd_create_branch(s);
2755 if (ret) {
2756 goto out;
2757 }
2758
2759 g_free(old_s);
2760
2761 return 0;
2762 out:
2763 /* recover bdrv_sd_state */
2764 memcpy(s, old_s, sizeof(BDRVSheepdogState));
2765 g_free(old_s);
2766
2767 error_report("failed to open. recover old bdrv_sd_state.");
2768
2769 return ret;
2770 }
2771
2772 #define NR_BATCHED_DISCARD 128
2773
2774 static int remove_objects(BDRVSheepdogState *s, Error **errp)
2775 {
2776 int fd, i = 0, nr_objs = 0;
2777 int ret;
2778 SheepdogInode *inode = &s->inode;
2779
2780 fd = connect_to_sdog(s, errp);
2781 if (fd < 0) {
2782 return fd;
2783 }
2784
2785 nr_objs = count_data_objs(inode);
2786 while (i < nr_objs) {
2787 int start_idx, nr_filled_idx;
2788
2789 while (i < nr_objs && !inode->data_vdi_id[i]) {
2790 i++;
2791 }
2792 start_idx = i;
2793
2794 nr_filled_idx = 0;
2795 while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) {
2796 if (inode->data_vdi_id[i]) {
2797 inode->data_vdi_id[i] = 0;
2798 nr_filled_idx++;
2799 }
2800
2801 i++;
2802 }
2803
2804 ret = write_object(fd, s->bs,
2805 (char *)&inode->data_vdi_id[start_idx],
2806 vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies,
2807 (i - start_idx) * sizeof(uint32_t),
2808 offsetof(struct SheepdogInode,
2809 data_vdi_id[start_idx]),
2810 false, s->cache_flags);
2811 if (ret < 0) {
2812 error_setg(errp, "Failed to discard snapshot inode");
2813 goto out;
2814 }
2815 }
2816
2817 ret = 0;
2818 out:
2819 closesocket(fd);
2820 return ret;
2821 }
2822
2823 static int sd_snapshot_delete(BlockDriverState *bs,
2824 const char *snapshot_id,
2825 const char *name,
2826 Error **errp)
2827 {
2828 /*
2829 * FIXME should delete the snapshot matching both @snapshot_id and
2830 * @name, but @name not used here
2831 */
2832 unsigned long snap_id = 0;
2833 char snap_tag[SD_MAX_VDI_TAG_LEN];
2834 int fd, ret;
2835 char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
2836 BDRVSheepdogState *s = bs->opaque;
2837 unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0;
2838 uint32_t vid;
2839 SheepdogVdiReq hdr = {
2840 .opcode = SD_OP_DEL_VDI,
2841 .data_length = wlen,
2842 .flags = SD_FLAG_CMD_WRITE,
2843 };
2844 SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2845
2846 ret = remove_objects(s, errp);
2847 if (ret) {
2848 return ret;
2849 }
2850
2851 memset(buf, 0, sizeof(buf));
2852 memset(snap_tag, 0, sizeof(snap_tag));
2853 pstrcpy(buf, SD_MAX_VDI_LEN, s->name);
2854 /* TODO Use sd_parse_snapid() once this mess is cleaned up */
2855 ret = qemu_strtoul(snapshot_id, NULL, 10, &snap_id);
2856 if (ret || snap_id > UINT32_MAX) {
2857 /*
2858 * FIXME Since qemu_strtoul() returns -EINVAL when
2859 * @snapshot_id is null, @snapshot_id is mandatory. Correct
2860 * would be to require at least one of @snapshot_id and @name.
2861 */
2862 error_setg(errp, "Invalid snapshot ID: %s",
2863 snapshot_id ? snapshot_id : "<null>");
2864 return -EINVAL;
2865 }
2866
2867 if (snap_id) {
2868 hdr.snapid = (uint32_t) snap_id;
2869 } else {
2870 /* FIXME I suspect we should use @name here */
2871 /* FIXME don't truncate silently */
2872 pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id);
2873 pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag);
2874 }
2875
2876 ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, errp);
2877 if (ret) {
2878 return ret;
2879 }
2880
2881 fd = connect_to_sdog(s, errp);
2882 if (fd < 0) {
2883 return fd;
2884 }
2885
2886 ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2887 buf, &wlen, &rlen);
2888 closesocket(fd);
2889 if (ret) {
2890 error_setg_errno(errp, -ret, "Couldn't send request to server");
2891 return ret;
2892 }
2893
2894 switch (rsp->result) {
2895 case SD_RES_NO_VDI:
2896 error_setg(errp, "Can't find the snapshot");
2897 return -ENOENT;
2898 case SD_RES_SUCCESS:
2899 break;
2900 default:
2901 error_setg(errp, "%s", sd_strerror(rsp->result));
2902 return -EIO;
2903 }
2904
2905 return 0;
2906 }
2907
2908 static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
2909 {
2910 Error *local_err = NULL;
2911 BDRVSheepdogState *s = bs->opaque;
2912 SheepdogReq req;
2913 int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
2914 QEMUSnapshotInfo *sn_tab = NULL;
2915 unsigned wlen, rlen;
2916 int found = 0;
2917 SheepdogInode *inode;
2918 unsigned long *vdi_inuse;
2919 unsigned int start_nr;
2920 uint64_t hval;
2921 uint32_t vid;
2922
2923 vdi_inuse = g_malloc(max);
2924 inode = g_malloc(SD_INODE_HEADER_SIZE);
2925
2926 fd = connect_to_sdog(s, &local_err);
2927 if (fd < 0) {
2928 error_report_err(local_err);
2929 ret = fd;
2930 goto out;
2931 }
2932
2933 rlen = max;
2934 wlen = 0;
2935
2936 memset(&req, 0, sizeof(req));
2937
2938 req.opcode = SD_OP_READ_VDIS;
2939 req.data_length = max;
2940
2941 ret = do_req(fd, s->bs, &req, vdi_inuse, &wlen, &rlen);
2942
2943 closesocket(fd);
2944 if (ret) {
2945 goto out;
2946 }
2947
2948 sn_tab = g_new0(QEMUSnapshotInfo, nr);
2949
2950 /* calculate a vdi id with hash function */
2951 hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
2952 start_nr = hval & (SD_NR_VDIS - 1);
2953
2954 fd = connect_to_sdog(s, &local_err);
2955 if (fd < 0) {
2956 error_report_err(local_err);
2957 ret = fd;
2958 goto out;
2959 }
2960
2961 for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
2962 if (!test_bit(vid, vdi_inuse)) {
2963 break;
2964 }
2965
2966 /* we don't need to read entire object */
2967 ret = read_object(fd, s->bs, (char *)inode,
2968 vid_to_vdi_oid(vid),
2969 0, SD_INODE_HEADER_SIZE, 0,
2970 s->cache_flags);
2971
2972 if (ret) {
2973 continue;
2974 }
2975
2976 if (!strcmp(inode->name, s->name) && is_snapshot(inode)) {
2977 sn_tab[found].date_sec = inode->snap_ctime >> 32;
2978 sn_tab[found].date_nsec = inode->snap_ctime & 0xffffffff;
2979 sn_tab[found].vm_state_size = inode->vm_state_size;
2980 sn_tab[found].vm_clock_nsec = inode->vm_clock_nsec;
2981
2982 snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str),
2983 "%" PRIu32, inode->snap_id);
2984 pstrcpy(sn_tab[found].name,
2985 MIN(sizeof(sn_tab[found].name), sizeof(inode->tag)),
2986 inode->tag);
2987 found++;
2988 }
2989 }
2990
2991 closesocket(fd);
2992 out:
2993 *psn_tab = sn_tab;
2994
2995 g_free(vdi_inuse);
2996 g_free(inode);
2997
2998 if (ret < 0) {
2999 return ret;
3000 }
3001
3002 return found;
3003 }
3004
3005 static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
3006 int64_t pos, int size, int load)
3007 {
3008 Error *local_err = NULL;
3009 bool create;
3010 int fd, ret = 0, remaining = size;
3011 unsigned int data_len;
3012 uint64_t vmstate_oid;
3013 uint64_t offset;
3014 uint32_t vdi_index;
3015 uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id;
3016 uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift);
3017
3018 fd = connect_to_sdog(s, &local_err);
3019 if (fd < 0) {
3020 error_report_err(local_err);
3021 return fd;
3022 }
3023
3024 while (remaining) {
3025 vdi_index = pos / object_size;
3026 offset = pos % object_size;
3027
3028 data_len = MIN(remaining, object_size - offset);
3029
3030 vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index);
3031
3032 create = (offset == 0);
3033 if (load) {
3034 ret = read_object(fd, s->bs, (char *)data, vmstate_oid,
3035 s->inode.nr_copies, data_len, offset,
3036 s->cache_flags);
3037 } else {
3038 ret = write_object(fd, s->bs, (char *)data, vmstate_oid,
3039 s->inode.nr_copies, data_len, offset, create,
3040 s->cache_flags);
3041 }
3042
3043 if (ret < 0) {
3044 error_report("failed to save vmstate %s", strerror(errno));
3045 goto cleanup;
3046 }
3047
3048 pos += data_len;
3049 data += data_len;
3050 remaining -= data_len;
3051 }
3052 ret = size;
3053 cleanup:
3054 closesocket(fd);
3055 return ret;
3056 }
3057
3058 static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3059 int64_t pos)
3060 {
3061 BDRVSheepdogState *s = bs->opaque;
3062 void *buf;
3063 int ret;
3064
3065 buf = qemu_blockalign(bs, qiov->size);
3066 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
3067 ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0);
3068 qemu_vfree(buf);
3069
3070 return ret;
3071 }
3072
3073 static int sd_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3074 int64_t pos)
3075 {
3076 BDRVSheepdogState *s = bs->opaque;
3077 void *buf;
3078 int ret;
3079
3080 buf = qemu_blockalign(bs, qiov->size);
3081 ret = do_load_save_vmstate(s, buf, pos, qiov->size, 1);
3082 qemu_iovec_from_buf(qiov, 0, buf, qiov->size);
3083 qemu_vfree(buf);
3084
3085 return ret;
3086 }
3087
3088
3089 static coroutine_fn int sd_co_pdiscard(BlockDriverState *bs, int64_t offset,
3090 int bytes)
3091 {
3092 SheepdogAIOCB acb;
3093 BDRVSheepdogState *s = bs->opaque;
3094 QEMUIOVector discard_iov;
3095 struct iovec iov;
3096 uint32_t zero = 0;
3097
3098 if (!s->discard_supported) {
3099 return 0;
3100 }
3101
3102 memset(&discard_iov, 0, sizeof(discard_iov));
3103 memset(&iov, 0, sizeof(iov));
3104 iov.iov_base = &zero;
3105 iov.iov_len = sizeof(zero);
3106 discard_iov.iov = &iov;
3107 discard_iov.niov = 1;
3108 if (!QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE)) {
3109 return -ENOTSUP;
3110 }
3111 sd_aio_setup(&acb, s, &discard_iov, offset >> BDRV_SECTOR_BITS,
3112 bytes >> BDRV_SECTOR_BITS, AIOCB_DISCARD_OBJ);
3113 sd_co_rw_vector(&acb);
3114 sd_aio_complete(&acb);
3115
3116 return acb.ret;
3117 }
3118
3119 static coroutine_fn int
3120 sd_co_block_status(BlockDriverState *bs, bool want_zero, int64_t offset,
3121 int64_t bytes, int64_t *pnum, int64_t *map,
3122 BlockDriverState **file)
3123 {
3124 BDRVSheepdogState *s = bs->opaque;
3125 SheepdogInode *inode = &s->inode;
3126 uint32_t object_size = (UINT32_C(1) << inode->block_size_shift);
3127 unsigned long start = offset / object_size,
3128 end = DIV_ROUND_UP(offset + bytes, object_size);
3129 unsigned long idx;
3130 *map = offset;
3131 int ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
3132
3133 for (idx = start; idx < end; idx++) {
3134 if (inode->data_vdi_id[idx] == 0) {
3135 break;
3136 }
3137 }
3138 if (idx == start) {
3139 /* Get the longest length of unallocated sectors */
3140 ret = 0;
3141 for (idx = start + 1; idx < end; idx++) {
3142 if (inode->data_vdi_id[idx] != 0) {
3143 break;
3144 }
3145 }
3146 }
3147
3148 *pnum = (idx - start) * object_size;
3149 if (*pnum > bytes) {
3150 *pnum = bytes;
3151 }
3152 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
3153 *file = bs;
3154 }
3155 return ret;
3156 }
3157
3158 static int64_t sd_get_allocated_file_size(BlockDriverState *bs)
3159 {
3160 BDRVSheepdogState *s = bs->opaque;
3161 SheepdogInode *inode = &s->inode;
3162 uint32_t object_size = (UINT32_C(1) << inode->block_size_shift);
3163 unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size);
3164 uint64_t size = 0;
3165
3166 for (i = 0; i < last; i++) {
3167 if (inode->data_vdi_id[i] == 0) {
3168 continue;
3169 }
3170 size += object_size;
3171 }
3172 return size;
3173 }
3174
3175 static QemuOptsList sd_create_opts = {
3176 .name = "sheepdog-create-opts",
3177 .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head),
3178 .desc = {
3179 {
3180 .name = BLOCK_OPT_SIZE,
3181 .type = QEMU_OPT_SIZE,
3182 .help = "Virtual disk size"
3183 },
3184 {
3185 .name = BLOCK_OPT_BACKING_FILE,
3186 .type = QEMU_OPT_STRING,
3187 .help = "File name of a base image"
3188 },
3189 {
3190 .name = BLOCK_OPT_BACKING_FMT,
3191 .type = QEMU_OPT_STRING,
3192 .help = "Must be 'sheepdog' if present",
3193 },
3194 {
3195 .name = BLOCK_OPT_PREALLOC,
3196 .type = QEMU_OPT_STRING,
3197 .help = "Preallocation mode (allowed values: off, full)"
3198 },
3199 {
3200 .name = BLOCK_OPT_REDUNDANCY,
3201 .type = QEMU_OPT_STRING,
3202 .help = "Redundancy of the image"
3203 },
3204 {
3205 .name = BLOCK_OPT_OBJECT_SIZE,
3206 .type = QEMU_OPT_SIZE,
3207 .help = "Object size of the image"
3208 },
3209 { /* end of list */ }
3210 }
3211 };
3212
3213 static const char *const sd_strong_runtime_opts[] = {
3214 "vdi",
3215 "snap-id",
3216 "tag",
3217 "server.",
3218
3219 NULL
3220 };
3221
3222 static BlockDriver bdrv_sheepdog = {
3223 .format_name = "sheepdog",
3224 .protocol_name = "sheepdog",
3225 .instance_size = sizeof(BDRVSheepdogState),
3226 .bdrv_parse_filename = sd_parse_filename,
3227 .bdrv_file_open = sd_open,
3228 .bdrv_reopen_prepare = sd_reopen_prepare,
3229 .bdrv_reopen_commit = sd_reopen_commit,
3230 .bdrv_reopen_abort = sd_reopen_abort,
3231 .bdrv_close = sd_close,
3232 .bdrv_co_create = sd_co_create,
3233 .bdrv_co_create_opts = sd_co_create_opts,
3234 .bdrv_has_zero_init = bdrv_has_zero_init_1,
3235 .bdrv_getlength = sd_getlength,
3236 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
3237 .bdrv_co_truncate = sd_co_truncate,
3238
3239 .bdrv_co_readv = sd_co_readv,
3240 .bdrv_co_writev = sd_co_writev,
3241 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
3242 .bdrv_co_pdiscard = sd_co_pdiscard,
3243 .bdrv_co_block_status = sd_co_block_status,
3244
3245 .bdrv_snapshot_create = sd_snapshot_create,
3246 .bdrv_snapshot_goto = sd_snapshot_goto,
3247 .bdrv_snapshot_delete = sd_snapshot_delete,
3248 .bdrv_snapshot_list = sd_snapshot_list,
3249
3250 .bdrv_save_vmstate = sd_save_vmstate,
3251 .bdrv_load_vmstate = sd_load_vmstate,
3252
3253 .bdrv_detach_aio_context = sd_detach_aio_context,
3254 .bdrv_attach_aio_context = sd_attach_aio_context,
3255
3256 .create_opts = &sd_create_opts,
3257 .strong_runtime_opts = sd_strong_runtime_opts,
3258 };
3259
3260 static BlockDriver bdrv_sheepdog_tcp = {
3261 .format_name = "sheepdog",
3262 .protocol_name = "sheepdog+tcp",
3263 .instance_size = sizeof(BDRVSheepdogState),
3264 .bdrv_parse_filename = sd_parse_filename,
3265 .bdrv_file_open = sd_open,
3266 .bdrv_reopen_prepare = sd_reopen_prepare,
3267 .bdrv_reopen_commit = sd_reopen_commit,
3268 .bdrv_reopen_abort = sd_reopen_abort,
3269 .bdrv_close = sd_close,
3270 .bdrv_co_create = sd_co_create,
3271 .bdrv_co_create_opts = sd_co_create_opts,
3272 .bdrv_has_zero_init = bdrv_has_zero_init_1,
3273 .bdrv_getlength = sd_getlength,
3274 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
3275 .bdrv_co_truncate = sd_co_truncate,
3276
3277 .bdrv_co_readv = sd_co_readv,
3278 .bdrv_co_writev = sd_co_writev,
3279 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
3280 .bdrv_co_pdiscard = sd_co_pdiscard,
3281 .bdrv_co_block_status = sd_co_block_status,
3282
3283 .bdrv_snapshot_create = sd_snapshot_create,
3284 .bdrv_snapshot_goto = sd_snapshot_goto,
3285 .bdrv_snapshot_delete = sd_snapshot_delete,
3286 .bdrv_snapshot_list = sd_snapshot_list,
3287
3288 .bdrv_save_vmstate = sd_save_vmstate,
3289 .bdrv_load_vmstate = sd_load_vmstate,
3290
3291 .bdrv_detach_aio_context = sd_detach_aio_context,
3292 .bdrv_attach_aio_context = sd_attach_aio_context,
3293
3294 .create_opts = &sd_create_opts,
3295 .strong_runtime_opts = sd_strong_runtime_opts,
3296 };
3297
3298 static BlockDriver bdrv_sheepdog_unix = {
3299 .format_name = "sheepdog",
3300 .protocol_name = "sheepdog+unix",
3301 .instance_size = sizeof(BDRVSheepdogState),
3302 .bdrv_parse_filename = sd_parse_filename,
3303 .bdrv_file_open = sd_open,
3304 .bdrv_reopen_prepare = sd_reopen_prepare,
3305 .bdrv_reopen_commit = sd_reopen_commit,
3306 .bdrv_reopen_abort = sd_reopen_abort,
3307 .bdrv_close = sd_close,
3308 .bdrv_co_create = sd_co_create,
3309 .bdrv_co_create_opts = sd_co_create_opts,
3310 .bdrv_has_zero_init = bdrv_has_zero_init_1,
3311 .bdrv_getlength = sd_getlength,
3312 .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
3313 .bdrv_co_truncate = sd_co_truncate,
3314
3315 .bdrv_co_readv = sd_co_readv,
3316 .bdrv_co_writev = sd_co_writev,
3317 .bdrv_co_flush_to_disk = sd_co_flush_to_disk,
3318 .bdrv_co_pdiscard = sd_co_pdiscard,
3319 .bdrv_co_block_status = sd_co_block_status,
3320
3321 .bdrv_snapshot_create = sd_snapshot_create,
3322 .bdrv_snapshot_goto = sd_snapshot_goto,
3323 .bdrv_snapshot_delete = sd_snapshot_delete,
3324 .bdrv_snapshot_list = sd_snapshot_list,
3325
3326 .bdrv_save_vmstate = sd_save_vmstate,
3327 .bdrv_load_vmstate = sd_load_vmstate,
3328
3329 .bdrv_detach_aio_context = sd_detach_aio_context,
3330 .bdrv_attach_aio_context = sd_attach_aio_context,
3331
3332 .create_opts = &sd_create_opts,
3333 .strong_runtime_opts = sd_strong_runtime_opts,
3334 };
3335
3336 static void bdrv_sheepdog_init(void)
3337 {
3338 bdrv_register(&bdrv_sheepdog);
3339 bdrv_register(&bdrv_sheepdog_tcp);
3340 bdrv_register(&bdrv_sheepdog_unix);
3341 }
3342 block_init(bdrv_sheepdog_init);