]>
Commit | Line | Data |
---|---|---|
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-common.h" | |
16 | #include "qemu/error-report.h" | |
17 | #include "qemu/sockets.h" | |
18 | #include "block/block_int.h" | |
19 | #include "qemu/bitops.h" | |
20 | ||
21 | #define SD_PROTO_VER 0x01 | |
22 | ||
23 | #define SD_DEFAULT_ADDR "localhost" | |
24 | #define SD_DEFAULT_PORT "7000" | |
25 | ||
26 | #define SD_OP_CREATE_AND_WRITE_OBJ 0x01 | |
27 | #define SD_OP_READ_OBJ 0x02 | |
28 | #define SD_OP_WRITE_OBJ 0x03 | |
29 | ||
30 | #define SD_OP_NEW_VDI 0x11 | |
31 | #define SD_OP_LOCK_VDI 0x12 | |
32 | #define SD_OP_RELEASE_VDI 0x13 | |
33 | #define SD_OP_GET_VDI_INFO 0x14 | |
34 | #define SD_OP_READ_VDIS 0x15 | |
35 | #define SD_OP_FLUSH_VDI 0x16 | |
36 | ||
37 | #define SD_FLAG_CMD_WRITE 0x01 | |
38 | #define SD_FLAG_CMD_COW 0x02 | |
39 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ | |
40 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
41 | ||
42 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
43 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
44 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
45 | #define SD_RES_EIO 0x03 /* I/O error */ | |
46 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
47 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
48 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
49 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
50 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
51 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
52 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
53 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
54 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
55 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
56 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
57 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
58 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
59 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
60 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
61 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
62 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
63 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
64 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
65 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
66 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
67 | ||
68 | /* | |
69 | * Object ID rules | |
70 | * | |
71 | * 0 - 19 (20 bits): data object space | |
72 | * 20 - 31 (12 bits): reserved data object space | |
73 | * 32 - 55 (24 bits): vdi object space | |
74 | * 56 - 59 ( 4 bits): reserved vdi object space | |
75 | * 60 - 63 ( 4 bits): object type identifier space | |
76 | */ | |
77 | ||
78 | #define VDI_SPACE_SHIFT 32 | |
79 | #define VDI_BIT (UINT64_C(1) << 63) | |
80 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
81 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
82 | #define MAX_CHILDREN 1024 | |
83 | #define SD_MAX_VDI_LEN 256 | |
84 | #define SD_MAX_VDI_TAG_LEN 256 | |
85 | #define SD_NR_VDIS (1U << 24) | |
86 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
87 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
88 | #define SECTOR_SIZE 512 | |
89 | ||
90 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
91 | #define CURRENT_VDI_ID 0 | |
92 | ||
93 | typedef struct SheepdogReq { | |
94 | uint8_t proto_ver; | |
95 | uint8_t opcode; | |
96 | uint16_t flags; | |
97 | uint32_t epoch; | |
98 | uint32_t id; | |
99 | uint32_t data_length; | |
100 | uint32_t opcode_specific[8]; | |
101 | } SheepdogReq; | |
102 | ||
103 | typedef struct SheepdogRsp { | |
104 | uint8_t proto_ver; | |
105 | uint8_t opcode; | |
106 | uint16_t flags; | |
107 | uint32_t epoch; | |
108 | uint32_t id; | |
109 | uint32_t data_length; | |
110 | uint32_t result; | |
111 | uint32_t opcode_specific[7]; | |
112 | } SheepdogRsp; | |
113 | ||
114 | typedef struct SheepdogObjReq { | |
115 | uint8_t proto_ver; | |
116 | uint8_t opcode; | |
117 | uint16_t flags; | |
118 | uint32_t epoch; | |
119 | uint32_t id; | |
120 | uint32_t data_length; | |
121 | uint64_t oid; | |
122 | uint64_t cow_oid; | |
123 | uint32_t copies; | |
124 | uint32_t rsvd; | |
125 | uint64_t offset; | |
126 | } SheepdogObjReq; | |
127 | ||
128 | typedef struct SheepdogObjRsp { | |
129 | uint8_t proto_ver; | |
130 | uint8_t opcode; | |
131 | uint16_t flags; | |
132 | uint32_t epoch; | |
133 | uint32_t id; | |
134 | uint32_t data_length; | |
135 | uint32_t result; | |
136 | uint32_t copies; | |
137 | uint32_t pad[6]; | |
138 | } SheepdogObjRsp; | |
139 | ||
140 | typedef struct SheepdogVdiReq { | |
141 | uint8_t proto_ver; | |
142 | uint8_t opcode; | |
143 | uint16_t flags; | |
144 | uint32_t epoch; | |
145 | uint32_t id; | |
146 | uint32_t data_length; | |
147 | uint64_t vdi_size; | |
148 | uint32_t base_vdi_id; | |
149 | uint32_t copies; | |
150 | uint32_t snapid; | |
151 | uint32_t pad[3]; | |
152 | } SheepdogVdiReq; | |
153 | ||
154 | typedef struct SheepdogVdiRsp { | |
155 | uint8_t proto_ver; | |
156 | uint8_t opcode; | |
157 | uint16_t flags; | |
158 | uint32_t epoch; | |
159 | uint32_t id; | |
160 | uint32_t data_length; | |
161 | uint32_t result; | |
162 | uint32_t rsvd; | |
163 | uint32_t vdi_id; | |
164 | uint32_t pad[5]; | |
165 | } SheepdogVdiRsp; | |
166 | ||
167 | typedef struct SheepdogInode { | |
168 | char name[SD_MAX_VDI_LEN]; | |
169 | char tag[SD_MAX_VDI_TAG_LEN]; | |
170 | uint64_t ctime; | |
171 | uint64_t snap_ctime; | |
172 | uint64_t vm_clock_nsec; | |
173 | uint64_t vdi_size; | |
174 | uint64_t vm_state_size; | |
175 | uint16_t copy_policy; | |
176 | uint8_t nr_copies; | |
177 | uint8_t block_size_shift; | |
178 | uint32_t snap_id; | |
179 | uint32_t vdi_id; | |
180 | uint32_t parent_vdi_id; | |
181 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
182 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
183 | } SheepdogInode; | |
184 | ||
185 | /* | |
186 | * 64 bit FNV-1a non-zero initial basis | |
187 | */ | |
188 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
189 | ||
190 | /* | |
191 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
192 | */ | |
193 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
194 | { | |
195 | unsigned char *bp = buf; | |
196 | unsigned char *be = bp + len; | |
197 | while (bp < be) { | |
198 | hval ^= (uint64_t) *bp++; | |
199 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
200 | (hval << 7) + (hval << 8) + (hval << 40); | |
201 | } | |
202 | return hval; | |
203 | } | |
204 | ||
205 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) | |
206 | { | |
207 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
208 | } | |
209 | ||
210 | static inline bool is_data_obj(uint64_t oid) | |
211 | { | |
212 | return !(VDI_BIT & oid); | |
213 | } | |
214 | ||
215 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
216 | { | |
217 | return oid & (MAX_DATA_OBJS - 1); | |
218 | } | |
219 | ||
220 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) | |
221 | { | |
222 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
223 | } | |
224 | ||
225 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
226 | { | |
227 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
228 | } | |
229 | ||
230 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
231 | { | |
232 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
233 | } | |
234 | ||
235 | static inline bool is_snapshot(struct SheepdogInode *inode) | |
236 | { | |
237 | return !!inode->snap_ctime; | |
238 | } | |
239 | ||
240 | #undef dprintf | |
241 | #ifdef DEBUG_SDOG | |
242 | #define dprintf(fmt, args...) \ | |
243 | do { \ | |
244 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
245 | } while (0) | |
246 | #else | |
247 | #define dprintf(fmt, args...) | |
248 | #endif | |
249 | ||
250 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
251 | ||
252 | typedef struct AIOReq { | |
253 | SheepdogAIOCB *aiocb; | |
254 | unsigned int iov_offset; | |
255 | ||
256 | uint64_t oid; | |
257 | uint64_t base_oid; | |
258 | uint64_t offset; | |
259 | unsigned int data_len; | |
260 | uint8_t flags; | |
261 | uint32_t id; | |
262 | ||
263 | QLIST_ENTRY(AIOReq) aio_siblings; | |
264 | } AIOReq; | |
265 | ||
266 | enum AIOCBState { | |
267 | AIOCB_WRITE_UDATA, | |
268 | AIOCB_READ_UDATA, | |
269 | AIOCB_FLUSH_CACHE, | |
270 | }; | |
271 | ||
272 | struct SheepdogAIOCB { | |
273 | BlockDriverAIOCB common; | |
274 | ||
275 | QEMUIOVector *qiov; | |
276 | ||
277 | int64_t sector_num; | |
278 | int nb_sectors; | |
279 | ||
280 | int ret; | |
281 | enum AIOCBState aiocb_type; | |
282 | ||
283 | Coroutine *coroutine; | |
284 | void (*aio_done_func)(SheepdogAIOCB *); | |
285 | ||
286 | bool canceled; | |
287 | int nr_pending; | |
288 | }; | |
289 | ||
290 | typedef struct BDRVSheepdogState { | |
291 | SheepdogInode inode; | |
292 | ||
293 | uint32_t min_dirty_data_idx; | |
294 | uint32_t max_dirty_data_idx; | |
295 | ||
296 | char name[SD_MAX_VDI_LEN]; | |
297 | bool is_snapshot; | |
298 | uint32_t cache_flags; | |
299 | ||
300 | char *addr; | |
301 | char *port; | |
302 | int fd; | |
303 | ||
304 | CoMutex lock; | |
305 | Coroutine *co_send; | |
306 | Coroutine *co_recv; | |
307 | ||
308 | uint32_t aioreq_seq_num; | |
309 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; | |
310 | QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head; | |
311 | } BDRVSheepdogState; | |
312 | ||
313 | static const char * sd_strerror(int err) | |
314 | { | |
315 | int i; | |
316 | ||
317 | static const struct { | |
318 | int err; | |
319 | const char *desc; | |
320 | } errors[] = { | |
321 | {SD_RES_SUCCESS, "Success"}, | |
322 | {SD_RES_UNKNOWN, "Unknown error"}, | |
323 | {SD_RES_NO_OBJ, "No object found"}, | |
324 | {SD_RES_EIO, "I/O error"}, | |
325 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
326 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
327 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
328 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
329 | {SD_RES_NO_VDI, "No vdi found"}, | |
330 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
331 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
332 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
333 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
334 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
335 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
336 | {SD_RES_STARTUP, "The system is still booting"}, | |
337 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
338 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
339 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
340 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
341 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
342 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
343 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
344 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
345 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
346 | }; | |
347 | ||
348 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
349 | if (errors[i].err == err) { | |
350 | return errors[i].desc; | |
351 | } | |
352 | } | |
353 | ||
354 | return "Invalid error code"; | |
355 | } | |
356 | ||
357 | /* | |
358 | * Sheepdog I/O handling: | |
359 | * | |
360 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and | |
361 | * link the requests to the inflight_list in the | |
362 | * BDRVSheepdogState. The function exits without waiting for | |
363 | * receiving the response. | |
364 | * | |
365 | * 2. We receive the response in aio_read_response, the fd handler to | |
366 | * the sheepdog connection. If metadata update is needed, we send | |
367 | * the write request to the vdi object in sd_write_done, the write | |
368 | * completion function. We switch back to sd_co_readv/writev after | |
369 | * all the requests belonging to the AIOCB are finished. | |
370 | */ | |
371 | ||
372 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
373 | uint64_t oid, unsigned int data_len, | |
374 | uint64_t offset, uint8_t flags, | |
375 | uint64_t base_oid, unsigned int iov_offset) | |
376 | { | |
377 | AIOReq *aio_req; | |
378 | ||
379 | aio_req = g_malloc(sizeof(*aio_req)); | |
380 | aio_req->aiocb = acb; | |
381 | aio_req->iov_offset = iov_offset; | |
382 | aio_req->oid = oid; | |
383 | aio_req->base_oid = base_oid; | |
384 | aio_req->offset = offset; | |
385 | aio_req->data_len = data_len; | |
386 | aio_req->flags = flags; | |
387 | aio_req->id = s->aioreq_seq_num++; | |
388 | ||
389 | acb->nr_pending++; | |
390 | return aio_req; | |
391 | } | |
392 | ||
393 | static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) | |
394 | { | |
395 | SheepdogAIOCB *acb = aio_req->aiocb; | |
396 | ||
397 | QLIST_REMOVE(aio_req, aio_siblings); | |
398 | g_free(aio_req); | |
399 | ||
400 | acb->nr_pending--; | |
401 | } | |
402 | ||
403 | static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) | |
404 | { | |
405 | if (!acb->canceled) { | |
406 | qemu_coroutine_enter(acb->coroutine, NULL); | |
407 | } | |
408 | qemu_aio_release(acb); | |
409 | } | |
410 | ||
411 | static void sd_aio_cancel(BlockDriverAIOCB *blockacb) | |
412 | { | |
413 | SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; | |
414 | ||
415 | /* | |
416 | * Sheepdog cannot cancel the requests which are already sent to | |
417 | * the servers, so we just complete the request with -EIO here. | |
418 | */ | |
419 | acb->ret = -EIO; | |
420 | qemu_coroutine_enter(acb->coroutine, NULL); | |
421 | acb->canceled = true; | |
422 | } | |
423 | ||
424 | static const AIOCBInfo sd_aiocb_info = { | |
425 | .aiocb_size = sizeof(SheepdogAIOCB), | |
426 | .cancel = sd_aio_cancel, | |
427 | }; | |
428 | ||
429 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
430 | int64_t sector_num, int nb_sectors) | |
431 | { | |
432 | SheepdogAIOCB *acb; | |
433 | ||
434 | acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); | |
435 | ||
436 | acb->qiov = qiov; | |
437 | ||
438 | acb->sector_num = sector_num; | |
439 | acb->nb_sectors = nb_sectors; | |
440 | ||
441 | acb->aio_done_func = NULL; | |
442 | acb->canceled = false; | |
443 | acb->coroutine = qemu_coroutine_self(); | |
444 | acb->ret = 0; | |
445 | acb->nr_pending = 0; | |
446 | return acb; | |
447 | } | |
448 | ||
449 | static int connect_to_sdog(const char *addr, const char *port) | |
450 | { | |
451 | char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV]; | |
452 | int fd, ret; | |
453 | struct addrinfo hints, *res, *res0; | |
454 | ||
455 | if (!addr) { | |
456 | addr = SD_DEFAULT_ADDR; | |
457 | port = SD_DEFAULT_PORT; | |
458 | } | |
459 | ||
460 | memset(&hints, 0, sizeof(hints)); | |
461 | hints.ai_socktype = SOCK_STREAM; | |
462 | ||
463 | ret = getaddrinfo(addr, port, &hints, &res0); | |
464 | if (ret) { | |
465 | error_report("unable to get address info %s, %s", | |
466 | addr, strerror(errno)); | |
467 | return -errno; | |
468 | } | |
469 | ||
470 | for (res = res0; res; res = res->ai_next) { | |
471 | ret = getnameinfo(res->ai_addr, res->ai_addrlen, hbuf, sizeof(hbuf), | |
472 | sbuf, sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV); | |
473 | if (ret) { | |
474 | continue; | |
475 | } | |
476 | ||
477 | fd = socket(res->ai_family, res->ai_socktype, res->ai_protocol); | |
478 | if (fd < 0) { | |
479 | continue; | |
480 | } | |
481 | ||
482 | reconnect: | |
483 | ret = connect(fd, res->ai_addr, res->ai_addrlen); | |
484 | if (ret < 0) { | |
485 | if (errno == EINTR) { | |
486 | goto reconnect; | |
487 | } | |
488 | close(fd); | |
489 | break; | |
490 | } | |
491 | ||
492 | dprintf("connected to %s:%s\n", addr, port); | |
493 | goto success; | |
494 | } | |
495 | fd = -errno; | |
496 | error_report("failed connect to %s:%s", addr, port); | |
497 | success: | |
498 | freeaddrinfo(res0); | |
499 | return fd; | |
500 | } | |
501 | ||
502 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, | |
503 | unsigned int *wlen) | |
504 | { | |
505 | int ret; | |
506 | ||
507 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
508 | if (ret < sizeof(*hdr)) { | |
509 | error_report("failed to send a req, %s", strerror(errno)); | |
510 | return ret; | |
511 | } | |
512 | ||
513 | ret = qemu_co_send(sockfd, data, *wlen); | |
514 | if (ret < *wlen) { | |
515 | error_report("failed to send a req, %s", strerror(errno)); | |
516 | } | |
517 | ||
518 | return ret; | |
519 | } | |
520 | ||
521 | static void restart_co_req(void *opaque) | |
522 | { | |
523 | Coroutine *co = opaque; | |
524 | ||
525 | qemu_coroutine_enter(co, NULL); | |
526 | } | |
527 | ||
528 | typedef struct SheepdogReqCo { | |
529 | int sockfd; | |
530 | SheepdogReq *hdr; | |
531 | void *data; | |
532 | unsigned int *wlen; | |
533 | unsigned int *rlen; | |
534 | int ret; | |
535 | bool finished; | |
536 | } SheepdogReqCo; | |
537 | ||
538 | static coroutine_fn void do_co_req(void *opaque) | |
539 | { | |
540 | int ret; | |
541 | Coroutine *co; | |
542 | SheepdogReqCo *srco = opaque; | |
543 | int sockfd = srco->sockfd; | |
544 | SheepdogReq *hdr = srco->hdr; | |
545 | void *data = srco->data; | |
546 | unsigned int *wlen = srco->wlen; | |
547 | unsigned int *rlen = srco->rlen; | |
548 | ||
549 | co = qemu_coroutine_self(); | |
550 | qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, NULL, co); | |
551 | ||
552 | socket_set_block(sockfd); | |
553 | ret = send_co_req(sockfd, hdr, data, wlen); | |
554 | if (ret < 0) { | |
555 | goto out; | |
556 | } | |
557 | ||
558 | qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, NULL, co); | |
559 | ||
560 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); | |
561 | if (ret < sizeof(*hdr)) { | |
562 | error_report("failed to get a rsp, %s", strerror(errno)); | |
563 | ret = -errno; | |
564 | goto out; | |
565 | } | |
566 | ||
567 | if (*rlen > hdr->data_length) { | |
568 | *rlen = hdr->data_length; | |
569 | } | |
570 | ||
571 | if (*rlen) { | |
572 | ret = qemu_co_recv(sockfd, data, *rlen); | |
573 | if (ret < *rlen) { | |
574 | error_report("failed to get the data, %s", strerror(errno)); | |
575 | ret = -errno; | |
576 | goto out; | |
577 | } | |
578 | } | |
579 | ret = 0; | |
580 | out: | |
581 | qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL); | |
582 | socket_set_nonblock(sockfd); | |
583 | ||
584 | srco->ret = ret; | |
585 | srco->finished = true; | |
586 | } | |
587 | ||
588 | static int do_req(int sockfd, SheepdogReq *hdr, void *data, | |
589 | unsigned int *wlen, unsigned int *rlen) | |
590 | { | |
591 | Coroutine *co; | |
592 | SheepdogReqCo srco = { | |
593 | .sockfd = sockfd, | |
594 | .hdr = hdr, | |
595 | .data = data, | |
596 | .wlen = wlen, | |
597 | .rlen = rlen, | |
598 | .ret = 0, | |
599 | .finished = false, | |
600 | }; | |
601 | ||
602 | if (qemu_in_coroutine()) { | |
603 | do_co_req(&srco); | |
604 | } else { | |
605 | co = qemu_coroutine_create(do_co_req); | |
606 | qemu_coroutine_enter(co, &srco); | |
607 | while (!srco.finished) { | |
608 | qemu_aio_wait(); | |
609 | } | |
610 | } | |
611 | ||
612 | return srco.ret; | |
613 | } | |
614 | ||
615 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, | |
616 | struct iovec *iov, int niov, bool create, | |
617 | enum AIOCBState aiocb_type); | |
618 | ||
619 | ||
620 | static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid) | |
621 | { | |
622 | AIOReq *aio_req; | |
623 | ||
624 | QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) { | |
625 | if (aio_req->oid == oid) { | |
626 | return aio_req; | |
627 | } | |
628 | } | |
629 | ||
630 | return NULL; | |
631 | } | |
632 | ||
633 | /* | |
634 | * This function searchs pending requests to the object `oid', and | |
635 | * sends them. | |
636 | */ | |
637 | static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid) | |
638 | { | |
639 | AIOReq *aio_req; | |
640 | SheepdogAIOCB *acb; | |
641 | int ret; | |
642 | ||
643 | while ((aio_req = find_pending_req(s, oid)) != NULL) { | |
644 | acb = aio_req->aiocb; | |
645 | /* move aio_req from pending list to inflight one */ | |
646 | QLIST_REMOVE(aio_req, aio_siblings); | |
647 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
648 | ret = add_aio_request(s, aio_req, acb->qiov->iov, | |
649 | acb->qiov->niov, false, acb->aiocb_type); | |
650 | if (ret < 0) { | |
651 | error_report("add_aio_request is failed"); | |
652 | free_aio_req(s, aio_req); | |
653 | if (!acb->nr_pending) { | |
654 | sd_finish_aiocb(acb); | |
655 | } | |
656 | } | |
657 | } | |
658 | } | |
659 | ||
660 | /* | |
661 | * Receive responses of the I/O requests. | |
662 | * | |
663 | * This function is registered as a fd handler, and called from the | |
664 | * main loop when s->fd is ready for reading responses. | |
665 | */ | |
666 | static void coroutine_fn aio_read_response(void *opaque) | |
667 | { | |
668 | SheepdogObjRsp rsp; | |
669 | BDRVSheepdogState *s = opaque; | |
670 | int fd = s->fd; | |
671 | int ret; | |
672 | AIOReq *aio_req = NULL; | |
673 | SheepdogAIOCB *acb; | |
674 | unsigned long idx; | |
675 | ||
676 | if (QLIST_EMPTY(&s->inflight_aio_head)) { | |
677 | goto out; | |
678 | } | |
679 | ||
680 | /* read a header */ | |
681 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); | |
682 | if (ret < 0) { | |
683 | error_report("failed to get the header, %s", strerror(errno)); | |
684 | goto out; | |
685 | } | |
686 | ||
687 | /* find the right aio_req from the inflight aio list */ | |
688 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
689 | if (aio_req->id == rsp.id) { | |
690 | break; | |
691 | } | |
692 | } | |
693 | if (!aio_req) { | |
694 | error_report("cannot find aio_req %x", rsp.id); | |
695 | goto out; | |
696 | } | |
697 | ||
698 | acb = aio_req->aiocb; | |
699 | ||
700 | switch (acb->aiocb_type) { | |
701 | case AIOCB_WRITE_UDATA: | |
702 | /* this coroutine context is no longer suitable for co_recv | |
703 | * because we may send data to update vdi objects */ | |
704 | s->co_recv = NULL; | |
705 | if (!is_data_obj(aio_req->oid)) { | |
706 | break; | |
707 | } | |
708 | idx = data_oid_to_idx(aio_req->oid); | |
709 | ||
710 | if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) { | |
711 | /* | |
712 | * If the object is newly created one, we need to update | |
713 | * the vdi object (metadata object). min_dirty_data_idx | |
714 | * and max_dirty_data_idx are changed to include updated | |
715 | * index between them. | |
716 | */ | |
717 | if (rsp.result == SD_RES_SUCCESS) { | |
718 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
719 | s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); | |
720 | s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); | |
721 | } | |
722 | /* | |
723 | * Some requests may be blocked because simultaneous | |
724 | * create requests are not allowed, so we search the | |
725 | * pending requests here. | |
726 | */ | |
727 | send_pending_req(s, aio_req->oid); | |
728 | } | |
729 | break; | |
730 | case AIOCB_READ_UDATA: | |
731 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, | |
732 | aio_req->iov_offset, rsp.data_length); | |
733 | if (ret < 0) { | |
734 | error_report("failed to get the data, %s", strerror(errno)); | |
735 | goto out; | |
736 | } | |
737 | break; | |
738 | case AIOCB_FLUSH_CACHE: | |
739 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
740 | dprintf("disable cache since the server doesn't support it\n"); | |
741 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
742 | rsp.result = SD_RES_SUCCESS; | |
743 | } | |
744 | break; | |
745 | } | |
746 | ||
747 | if (rsp.result != SD_RES_SUCCESS) { | |
748 | acb->ret = -EIO; | |
749 | error_report("%s", sd_strerror(rsp.result)); | |
750 | } | |
751 | ||
752 | free_aio_req(s, aio_req); | |
753 | if (!acb->nr_pending) { | |
754 | /* | |
755 | * We've finished all requests which belong to the AIOCB, so | |
756 | * we can switch back to sd_co_readv/writev now. | |
757 | */ | |
758 | acb->aio_done_func(acb); | |
759 | } | |
760 | out: | |
761 | s->co_recv = NULL; | |
762 | } | |
763 | ||
764 | static void co_read_response(void *opaque) | |
765 | { | |
766 | BDRVSheepdogState *s = opaque; | |
767 | ||
768 | if (!s->co_recv) { | |
769 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
770 | } | |
771 | ||
772 | qemu_coroutine_enter(s->co_recv, opaque); | |
773 | } | |
774 | ||
775 | static void co_write_request(void *opaque) | |
776 | { | |
777 | BDRVSheepdogState *s = opaque; | |
778 | ||
779 | qemu_coroutine_enter(s->co_send, NULL); | |
780 | } | |
781 | ||
782 | static int aio_flush_request(void *opaque) | |
783 | { | |
784 | BDRVSheepdogState *s = opaque; | |
785 | ||
786 | return !QLIST_EMPTY(&s->inflight_aio_head) || | |
787 | !QLIST_EMPTY(&s->pending_aio_head); | |
788 | } | |
789 | ||
790 | static int set_nodelay(int fd) | |
791 | { | |
792 | int ret, opt; | |
793 | ||
794 | opt = 1; | |
795 | ret = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *)&opt, sizeof(opt)); | |
796 | return ret; | |
797 | } | |
798 | ||
799 | /* | |
800 | * Return a socket discriptor to read/write objects. | |
801 | * | |
802 | * We cannot use this discriptor for other operations because | |
803 | * the block driver may be on waiting response from the server. | |
804 | */ | |
805 | static int get_sheep_fd(BDRVSheepdogState *s) | |
806 | { | |
807 | int ret, fd; | |
808 | ||
809 | fd = connect_to_sdog(s->addr, s->port); | |
810 | if (fd < 0) { | |
811 | error_report("%s", strerror(errno)); | |
812 | return fd; | |
813 | } | |
814 | ||
815 | socket_set_nonblock(fd); | |
816 | ||
817 | ret = set_nodelay(fd); | |
818 | if (ret) { | |
819 | error_report("%s", strerror(errno)); | |
820 | closesocket(fd); | |
821 | return -errno; | |
822 | } | |
823 | ||
824 | qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s); | |
825 | return fd; | |
826 | } | |
827 | ||
828 | /* | |
829 | * Parse a filename | |
830 | * | |
831 | * filename must be one of the following formats: | |
832 | * 1. [vdiname] | |
833 | * 2. [vdiname]:[snapid] | |
834 | * 3. [vdiname]:[tag] | |
835 | * 4. [hostname]:[port]:[vdiname] | |
836 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
837 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
838 | * | |
839 | * You can boot from the snapshot images by specifying `snapid` or | |
840 | * `tag'. | |
841 | * | |
842 | * You can run VMs outside the Sheepdog cluster by specifying | |
843 | * `hostname' and `port' (experimental). | |
844 | */ | |
845 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
846 | char *vdi, uint32_t *snapid, char *tag) | |
847 | { | |
848 | char *p, *q; | |
849 | int nr_sep; | |
850 | ||
851 | p = q = g_strdup(filename); | |
852 | ||
853 | /* count the number of separators */ | |
854 | nr_sep = 0; | |
855 | while (*p) { | |
856 | if (*p == ':') { | |
857 | nr_sep++; | |
858 | } | |
859 | p++; | |
860 | } | |
861 | p = q; | |
862 | ||
863 | /* use the first two tokens as hostname and port number. */ | |
864 | if (nr_sep >= 2) { | |
865 | s->addr = p; | |
866 | p = strchr(p, ':'); | |
867 | *p++ = '\0'; | |
868 | ||
869 | s->port = p; | |
870 | p = strchr(p, ':'); | |
871 | *p++ = '\0'; | |
872 | } else { | |
873 | s->addr = NULL; | |
874 | s->port = 0; | |
875 | } | |
876 | ||
877 | pstrcpy(vdi, SD_MAX_VDI_LEN, p); | |
878 | ||
879 | p = strchr(vdi, ':'); | |
880 | if (p) { | |
881 | *p++ = '\0'; | |
882 | *snapid = strtoul(p, NULL, 10); | |
883 | if (*snapid == 0) { | |
884 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, p); | |
885 | } | |
886 | } else { | |
887 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
888 | } | |
889 | ||
890 | if (s->addr == NULL) { | |
891 | g_free(q); | |
892 | } | |
893 | ||
894 | return 0; | |
895 | } | |
896 | ||
897 | static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid, | |
898 | char *tag, uint32_t *vid, int for_snapshot) | |
899 | { | |
900 | int ret, fd; | |
901 | SheepdogVdiReq hdr; | |
902 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
903 | unsigned int wlen, rlen = 0; | |
904 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
905 | ||
906 | fd = connect_to_sdog(s->addr, s->port); | |
907 | if (fd < 0) { | |
908 | return fd; | |
909 | } | |
910 | ||
911 | /* This pair of strncpy calls ensures that the buffer is zero-filled, | |
912 | * which is desirable since we'll soon be sending those bytes, and | |
913 | * don't want the send_req to read uninitialized data. | |
914 | */ | |
915 | strncpy(buf, filename, SD_MAX_VDI_LEN); | |
916 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
917 | ||
918 | memset(&hdr, 0, sizeof(hdr)); | |
919 | if (for_snapshot) { | |
920 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
921 | } else { | |
922 | hdr.opcode = SD_OP_LOCK_VDI; | |
923 | } | |
924 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
925 | hdr.proto_ver = SD_PROTO_VER; | |
926 | hdr.data_length = wlen; | |
927 | hdr.snapid = snapid; | |
928 | hdr.flags = SD_FLAG_CMD_WRITE; | |
929 | ||
930 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
931 | if (ret) { | |
932 | goto out; | |
933 | } | |
934 | ||
935 | if (rsp->result != SD_RES_SUCCESS) { | |
936 | error_report("cannot get vdi info, %s, %s %d %s", | |
937 | sd_strerror(rsp->result), filename, snapid, tag); | |
938 | if (rsp->result == SD_RES_NO_VDI) { | |
939 | ret = -ENOENT; | |
940 | } else { | |
941 | ret = -EIO; | |
942 | } | |
943 | goto out; | |
944 | } | |
945 | *vid = rsp->vdi_id; | |
946 | ||
947 | ret = 0; | |
948 | out: | |
949 | closesocket(fd); | |
950 | return ret; | |
951 | } | |
952 | ||
953 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, | |
954 | struct iovec *iov, int niov, bool create, | |
955 | enum AIOCBState aiocb_type) | |
956 | { | |
957 | int nr_copies = s->inode.nr_copies; | |
958 | SheepdogObjReq hdr; | |
959 | unsigned int wlen = 0; | |
960 | int ret; | |
961 | uint64_t oid = aio_req->oid; | |
962 | unsigned int datalen = aio_req->data_len; | |
963 | uint64_t offset = aio_req->offset; | |
964 | uint8_t flags = aio_req->flags; | |
965 | uint64_t old_oid = aio_req->base_oid; | |
966 | ||
967 | if (!nr_copies) { | |
968 | error_report("bug"); | |
969 | } | |
970 | ||
971 | memset(&hdr, 0, sizeof(hdr)); | |
972 | ||
973 | switch (aiocb_type) { | |
974 | case AIOCB_FLUSH_CACHE: | |
975 | hdr.opcode = SD_OP_FLUSH_VDI; | |
976 | break; | |
977 | case AIOCB_READ_UDATA: | |
978 | hdr.opcode = SD_OP_READ_OBJ; | |
979 | hdr.flags = flags; | |
980 | break; | |
981 | case AIOCB_WRITE_UDATA: | |
982 | if (create) { | |
983 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
984 | } else { | |
985 | hdr.opcode = SD_OP_WRITE_OBJ; | |
986 | } | |
987 | wlen = datalen; | |
988 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
989 | break; | |
990 | } | |
991 | ||
992 | if (s->cache_flags) { | |
993 | hdr.flags |= s->cache_flags; | |
994 | } | |
995 | ||
996 | hdr.oid = oid; | |
997 | hdr.cow_oid = old_oid; | |
998 | hdr.copies = s->inode.nr_copies; | |
999 | ||
1000 | hdr.data_length = datalen; | |
1001 | hdr.offset = offset; | |
1002 | ||
1003 | hdr.id = aio_req->id; | |
1004 | ||
1005 | qemu_co_mutex_lock(&s->lock); | |
1006 | s->co_send = qemu_coroutine_self(); | |
1007 | qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, | |
1008 | aio_flush_request, s); | |
1009 | socket_set_cork(s->fd, 1); | |
1010 | ||
1011 | /* send a header */ | |
1012 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); | |
1013 | if (ret < 0) { | |
1014 | qemu_co_mutex_unlock(&s->lock); | |
1015 | error_report("failed to send a req, %s", strerror(errno)); | |
1016 | return -errno; | |
1017 | } | |
1018 | ||
1019 | if (wlen) { | |
1020 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); | |
1021 | if (ret < 0) { | |
1022 | qemu_co_mutex_unlock(&s->lock); | |
1023 | error_report("failed to send a data, %s", strerror(errno)); | |
1024 | return -errno; | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | socket_set_cork(s->fd, 0); | |
1029 | qemu_aio_set_fd_handler(s->fd, co_read_response, NULL, | |
1030 | aio_flush_request, s); | |
1031 | qemu_co_mutex_unlock(&s->lock); | |
1032 | ||
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | static int read_write_object(int fd, char *buf, uint64_t oid, int copies, | |
1037 | unsigned int datalen, uint64_t offset, | |
1038 | bool write, bool create, uint32_t cache_flags) | |
1039 | { | |
1040 | SheepdogObjReq hdr; | |
1041 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1042 | unsigned int wlen, rlen; | |
1043 | int ret; | |
1044 | ||
1045 | memset(&hdr, 0, sizeof(hdr)); | |
1046 | ||
1047 | if (write) { | |
1048 | wlen = datalen; | |
1049 | rlen = 0; | |
1050 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1051 | if (create) { | |
1052 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1053 | } else { | |
1054 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1055 | } | |
1056 | } else { | |
1057 | wlen = 0; | |
1058 | rlen = datalen; | |
1059 | hdr.opcode = SD_OP_READ_OBJ; | |
1060 | } | |
1061 | ||
1062 | hdr.flags |= cache_flags; | |
1063 | ||
1064 | hdr.oid = oid; | |
1065 | hdr.data_length = datalen; | |
1066 | hdr.offset = offset; | |
1067 | hdr.copies = copies; | |
1068 | ||
1069 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1070 | if (ret) { | |
1071 | error_report("failed to send a request to the sheep"); | |
1072 | return ret; | |
1073 | } | |
1074 | ||
1075 | switch (rsp->result) { | |
1076 | case SD_RES_SUCCESS: | |
1077 | return 0; | |
1078 | default: | |
1079 | error_report("%s", sd_strerror(rsp->result)); | |
1080 | return -EIO; | |
1081 | } | |
1082 | } | |
1083 | ||
1084 | static int read_object(int fd, char *buf, uint64_t oid, int copies, | |
1085 | unsigned int datalen, uint64_t offset, | |
1086 | uint32_t cache_flags) | |
1087 | { | |
1088 | return read_write_object(fd, buf, oid, copies, datalen, offset, false, | |
1089 | false, cache_flags); | |
1090 | } | |
1091 | ||
1092 | static int write_object(int fd, char *buf, uint64_t oid, int copies, | |
1093 | unsigned int datalen, uint64_t offset, bool create, | |
1094 | uint32_t cache_flags) | |
1095 | { | |
1096 | return read_write_object(fd, buf, oid, copies, datalen, offset, true, | |
1097 | create, cache_flags); | |
1098 | } | |
1099 | ||
1100 | static int sd_open(BlockDriverState *bs, const char *filename, int flags) | |
1101 | { | |
1102 | int ret, fd; | |
1103 | uint32_t vid = 0; | |
1104 | BDRVSheepdogState *s = bs->opaque; | |
1105 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1106 | uint32_t snapid; | |
1107 | char *buf = NULL; | |
1108 | ||
1109 | strstart(filename, "sheepdog:", (const char **)&filename); | |
1110 | ||
1111 | QLIST_INIT(&s->inflight_aio_head); | |
1112 | QLIST_INIT(&s->pending_aio_head); | |
1113 | s->fd = -1; | |
1114 | ||
1115 | memset(vdi, 0, sizeof(vdi)); | |
1116 | memset(tag, 0, sizeof(tag)); | |
1117 | if (parse_vdiname(s, filename, vdi, &snapid, tag) < 0) { | |
1118 | ret = -EINVAL; | |
1119 | goto out; | |
1120 | } | |
1121 | s->fd = get_sheep_fd(s); | |
1122 | if (s->fd < 0) { | |
1123 | ret = s->fd; | |
1124 | goto out; | |
1125 | } | |
1126 | ||
1127 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0); | |
1128 | if (ret) { | |
1129 | goto out; | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1134 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1135 | */ | |
1136 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1137 | if (flags & BDRV_O_NOCACHE) { | |
1138 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1139 | } | |
1140 | ||
1141 | if (snapid || tag[0] != '\0') { | |
1142 | dprintf("%" PRIx32 " snapshot inode was open.\n", vid); | |
1143 | s->is_snapshot = true; | |
1144 | } | |
1145 | ||
1146 | fd = connect_to_sdog(s->addr, s->port); | |
1147 | if (fd < 0) { | |
1148 | error_report("failed to connect"); | |
1149 | ret = fd; | |
1150 | goto out; | |
1151 | } | |
1152 | ||
1153 | buf = g_malloc(SD_INODE_SIZE); | |
1154 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, | |
1155 | s->cache_flags); | |
1156 | ||
1157 | closesocket(fd); | |
1158 | ||
1159 | if (ret) { | |
1160 | goto out; | |
1161 | } | |
1162 | ||
1163 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1164 | s->min_dirty_data_idx = UINT32_MAX; | |
1165 | s->max_dirty_data_idx = 0; | |
1166 | ||
1167 | bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE; | |
1168 | pstrcpy(s->name, sizeof(s->name), vdi); | |
1169 | qemu_co_mutex_init(&s->lock); | |
1170 | g_free(buf); | |
1171 | return 0; | |
1172 | out: | |
1173 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); | |
1174 | if (s->fd >= 0) { | |
1175 | closesocket(s->fd); | |
1176 | } | |
1177 | g_free(buf); | |
1178 | return ret; | |
1179 | } | |
1180 | ||
1181 | static int do_sd_create(char *filename, int64_t vdi_size, | |
1182 | uint32_t base_vid, uint32_t *vdi_id, int snapshot, | |
1183 | const char *addr, const char *port) | |
1184 | { | |
1185 | SheepdogVdiReq hdr; | |
1186 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1187 | int fd, ret; | |
1188 | unsigned int wlen, rlen = 0; | |
1189 | char buf[SD_MAX_VDI_LEN]; | |
1190 | ||
1191 | fd = connect_to_sdog(addr, port); | |
1192 | if (fd < 0) { | |
1193 | return fd; | |
1194 | } | |
1195 | ||
1196 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename | |
1197 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1198 | */ | |
1199 | memset(buf, 0, sizeof(buf)); | |
1200 | pstrcpy(buf, sizeof(buf), filename); | |
1201 | ||
1202 | memset(&hdr, 0, sizeof(hdr)); | |
1203 | hdr.opcode = SD_OP_NEW_VDI; | |
1204 | hdr.base_vdi_id = base_vid; | |
1205 | ||
1206 | wlen = SD_MAX_VDI_LEN; | |
1207 | ||
1208 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1209 | hdr.snapid = snapshot; | |
1210 | ||
1211 | hdr.data_length = wlen; | |
1212 | hdr.vdi_size = vdi_size; | |
1213 | ||
1214 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1215 | ||
1216 | closesocket(fd); | |
1217 | ||
1218 | if (ret) { | |
1219 | return ret; | |
1220 | } | |
1221 | ||
1222 | if (rsp->result != SD_RES_SUCCESS) { | |
1223 | error_report("%s, %s", sd_strerror(rsp->result), filename); | |
1224 | return -EIO; | |
1225 | } | |
1226 | ||
1227 | if (vdi_id) { | |
1228 | *vdi_id = rsp->vdi_id; | |
1229 | } | |
1230 | ||
1231 | return 0; | |
1232 | } | |
1233 | ||
1234 | static int sd_prealloc(const char *filename) | |
1235 | { | |
1236 | BlockDriverState *bs = NULL; | |
1237 | uint32_t idx, max_idx; | |
1238 | int64_t vdi_size; | |
1239 | void *buf = g_malloc0(SD_DATA_OBJ_SIZE); | |
1240 | int ret; | |
1241 | ||
1242 | ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); | |
1243 | if (ret < 0) { | |
1244 | goto out; | |
1245 | } | |
1246 | ||
1247 | vdi_size = bdrv_getlength(bs); | |
1248 | if (vdi_size < 0) { | |
1249 | ret = vdi_size; | |
1250 | goto out; | |
1251 | } | |
1252 | max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); | |
1253 | ||
1254 | for (idx = 0; idx < max_idx; idx++) { | |
1255 | /* | |
1256 | * The created image can be a cloned image, so we need to read | |
1257 | * a data from the source image. | |
1258 | */ | |
1259 | ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1260 | if (ret < 0) { | |
1261 | goto out; | |
1262 | } | |
1263 | ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1264 | if (ret < 0) { | |
1265 | goto out; | |
1266 | } | |
1267 | } | |
1268 | out: | |
1269 | if (bs) { | |
1270 | bdrv_delete(bs); | |
1271 | } | |
1272 | g_free(buf); | |
1273 | ||
1274 | return ret; | |
1275 | } | |
1276 | ||
1277 | static int sd_create(const char *filename, QEMUOptionParameter *options) | |
1278 | { | |
1279 | int ret = 0; | |
1280 | uint32_t vid = 0, base_vid = 0; | |
1281 | int64_t vdi_size = 0; | |
1282 | char *backing_file = NULL; | |
1283 | BDRVSheepdogState *s; | |
1284 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1285 | uint32_t snapid; | |
1286 | bool prealloc = false; | |
1287 | const char *vdiname; | |
1288 | ||
1289 | s = g_malloc0(sizeof(BDRVSheepdogState)); | |
1290 | ||
1291 | strstart(filename, "sheepdog:", &vdiname); | |
1292 | ||
1293 | memset(vdi, 0, sizeof(vdi)); | |
1294 | memset(tag, 0, sizeof(tag)); | |
1295 | if (parse_vdiname(s, vdiname, vdi, &snapid, tag) < 0) { | |
1296 | error_report("invalid filename"); | |
1297 | ret = -EINVAL; | |
1298 | goto out; | |
1299 | } | |
1300 | ||
1301 | while (options && options->name) { | |
1302 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1303 | vdi_size = options->value.n; | |
1304 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1305 | backing_file = options->value.s; | |
1306 | } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { | |
1307 | if (!options->value.s || !strcmp(options->value.s, "off")) { | |
1308 | prealloc = false; | |
1309 | } else if (!strcmp(options->value.s, "full")) { | |
1310 | prealloc = true; | |
1311 | } else { | |
1312 | error_report("Invalid preallocation mode: '%s'", | |
1313 | options->value.s); | |
1314 | ret = -EINVAL; | |
1315 | goto out; | |
1316 | } | |
1317 | } | |
1318 | options++; | |
1319 | } | |
1320 | ||
1321 | if (vdi_size > SD_MAX_VDI_SIZE) { | |
1322 | error_report("too big image size"); | |
1323 | ret = -EINVAL; | |
1324 | goto out; | |
1325 | } | |
1326 | ||
1327 | if (backing_file) { | |
1328 | BlockDriverState *bs; | |
1329 | BDRVSheepdogState *s; | |
1330 | BlockDriver *drv; | |
1331 | ||
1332 | /* Currently, only Sheepdog backing image is supported. */ | |
1333 | drv = bdrv_find_protocol(backing_file); | |
1334 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { | |
1335 | error_report("backing_file must be a sheepdog image"); | |
1336 | ret = -EINVAL; | |
1337 | goto out; | |
1338 | } | |
1339 | ||
1340 | ret = bdrv_file_open(&bs, backing_file, 0); | |
1341 | if (ret < 0) { | |
1342 | goto out; | |
1343 | } | |
1344 | ||
1345 | s = bs->opaque; | |
1346 | ||
1347 | if (!is_snapshot(&s->inode)) { | |
1348 | error_report("cannot clone from a non snapshot vdi"); | |
1349 | bdrv_delete(bs); | |
1350 | ret = -EINVAL; | |
1351 | goto out; | |
1352 | } | |
1353 | ||
1354 | base_vid = s->inode.vdi_id; | |
1355 | bdrv_delete(bs); | |
1356 | } | |
1357 | ||
1358 | ret = do_sd_create(vdi, vdi_size, base_vid, &vid, 0, s->addr, s->port); | |
1359 | if (!prealloc || ret) { | |
1360 | goto out; | |
1361 | } | |
1362 | ||
1363 | ret = sd_prealloc(filename); | |
1364 | out: | |
1365 | g_free(s); | |
1366 | return ret; | |
1367 | } | |
1368 | ||
1369 | static void sd_close(BlockDriverState *bs) | |
1370 | { | |
1371 | BDRVSheepdogState *s = bs->opaque; | |
1372 | SheepdogVdiReq hdr; | |
1373 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1374 | unsigned int wlen, rlen = 0; | |
1375 | int fd, ret; | |
1376 | ||
1377 | dprintf("%s\n", s->name); | |
1378 | ||
1379 | fd = connect_to_sdog(s->addr, s->port); | |
1380 | if (fd < 0) { | |
1381 | return; | |
1382 | } | |
1383 | ||
1384 | memset(&hdr, 0, sizeof(hdr)); | |
1385 | ||
1386 | hdr.opcode = SD_OP_RELEASE_VDI; | |
1387 | wlen = strlen(s->name) + 1; | |
1388 | hdr.data_length = wlen; | |
1389 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1390 | ||
1391 | ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); | |
1392 | ||
1393 | closesocket(fd); | |
1394 | ||
1395 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1396 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
1397 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
1398 | } | |
1399 | ||
1400 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); | |
1401 | closesocket(s->fd); | |
1402 | g_free(s->addr); | |
1403 | } | |
1404 | ||
1405 | static int64_t sd_getlength(BlockDriverState *bs) | |
1406 | { | |
1407 | BDRVSheepdogState *s = bs->opaque; | |
1408 | ||
1409 | return s->inode.vdi_size; | |
1410 | } | |
1411 | ||
1412 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1413 | { | |
1414 | BDRVSheepdogState *s = bs->opaque; | |
1415 | int ret, fd; | |
1416 | unsigned int datalen; | |
1417 | ||
1418 | if (offset < s->inode.vdi_size) { | |
1419 | error_report("shrinking is not supported"); | |
1420 | return -EINVAL; | |
1421 | } else if (offset > SD_MAX_VDI_SIZE) { | |
1422 | error_report("too big image size"); | |
1423 | return -EINVAL; | |
1424 | } | |
1425 | ||
1426 | fd = connect_to_sdog(s->addr, s->port); | |
1427 | if (fd < 0) { | |
1428 | return fd; | |
1429 | } | |
1430 | ||
1431 | /* we don't need to update entire object */ | |
1432 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1433 | s->inode.vdi_size = offset; | |
1434 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
1435 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); | |
1436 | close(fd); | |
1437 | ||
1438 | if (ret < 0) { | |
1439 | error_report("failed to update an inode."); | |
1440 | } | |
1441 | ||
1442 | return ret; | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * This function is called after writing data objects. If we need to | |
1447 | * update metadata, this sends a write request to the vdi object. | |
1448 | * Otherwise, this switches back to sd_co_readv/writev. | |
1449 | */ | |
1450 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) | |
1451 | { | |
1452 | int ret; | |
1453 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1454 | struct iovec iov; | |
1455 | AIOReq *aio_req; | |
1456 | uint32_t offset, data_len, mn, mx; | |
1457 | ||
1458 | mn = s->min_dirty_data_idx; | |
1459 | mx = s->max_dirty_data_idx; | |
1460 | if (mn <= mx) { | |
1461 | /* we need to update the vdi object. */ | |
1462 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
1463 | mn * sizeof(s->inode.data_vdi_id[0]); | |
1464 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
1465 | ||
1466 | s->min_dirty_data_idx = UINT32_MAX; | |
1467 | s->max_dirty_data_idx = 0; | |
1468 | ||
1469 | iov.iov_base = &s->inode; | |
1470 | iov.iov_len = sizeof(s->inode); | |
1471 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
1472 | data_len, offset, 0, 0, offset); | |
1473 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
1474 | ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); | |
1475 | if (ret) { | |
1476 | free_aio_req(s, aio_req); | |
1477 | acb->ret = -EIO; | |
1478 | goto out; | |
1479 | } | |
1480 | ||
1481 | acb->aio_done_func = sd_finish_aiocb; | |
1482 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1483 | return; | |
1484 | } | |
1485 | out: | |
1486 | sd_finish_aiocb(acb); | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * Create a writable VDI from a snapshot | |
1491 | */ | |
1492 | static int sd_create_branch(BDRVSheepdogState *s) | |
1493 | { | |
1494 | int ret, fd; | |
1495 | uint32_t vid; | |
1496 | char *buf; | |
1497 | ||
1498 | dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); | |
1499 | ||
1500 | buf = g_malloc(SD_INODE_SIZE); | |
1501 | ||
1502 | ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1, | |
1503 | s->addr, s->port); | |
1504 | if (ret) { | |
1505 | goto out; | |
1506 | } | |
1507 | ||
1508 | dprintf("%" PRIx32 " is created.\n", vid); | |
1509 | ||
1510 | fd = connect_to_sdog(s->addr, s->port); | |
1511 | if (fd < 0) { | |
1512 | error_report("failed to connect"); | |
1513 | ret = fd; | |
1514 | goto out; | |
1515 | } | |
1516 | ||
1517 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, | |
1518 | SD_INODE_SIZE, 0, s->cache_flags); | |
1519 | ||
1520 | closesocket(fd); | |
1521 | ||
1522 | if (ret < 0) { | |
1523 | goto out; | |
1524 | } | |
1525 | ||
1526 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1527 | ||
1528 | s->is_snapshot = false; | |
1529 | ret = 0; | |
1530 | dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id); | |
1531 | ||
1532 | out: | |
1533 | g_free(buf); | |
1534 | ||
1535 | return ret; | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * Send I/O requests to the server. | |
1540 | * | |
1541 | * This function sends requests to the server, links the requests to | |
1542 | * the inflight_list in BDRVSheepdogState, and exits without | |
1543 | * waiting the response. The responses are received in the | |
1544 | * `aio_read_response' function which is called from the main loop as | |
1545 | * a fd handler. | |
1546 | * | |
1547 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
1548 | * request and -errno in error cases. | |
1549 | */ | |
1550 | static int coroutine_fn sd_co_rw_vector(void *p) | |
1551 | { | |
1552 | SheepdogAIOCB *acb = p; | |
1553 | int ret = 0; | |
1554 | unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE; | |
1555 | unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE; | |
1556 | uint64_t oid; | |
1557 | uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE; | |
1558 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1559 | SheepdogInode *inode = &s->inode; | |
1560 | AIOReq *aio_req; | |
1561 | ||
1562 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { | |
1563 | /* | |
1564 | * In the case we open the snapshot VDI, Sheepdog creates the | |
1565 | * writable VDI when we do a write operation first. | |
1566 | */ | |
1567 | ret = sd_create_branch(s); | |
1568 | if (ret) { | |
1569 | acb->ret = -EIO; | |
1570 | goto out; | |
1571 | } | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * Make sure we don't free the aiocb before we are done with all requests. | |
1576 | * This additional reference is dropped at the end of this function. | |
1577 | */ | |
1578 | acb->nr_pending++; | |
1579 | ||
1580 | while (done != total) { | |
1581 | uint8_t flags = 0; | |
1582 | uint64_t old_oid = 0; | |
1583 | bool create = false; | |
1584 | ||
1585 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
1586 | ||
1587 | len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); | |
1588 | ||
1589 | switch (acb->aiocb_type) { | |
1590 | case AIOCB_READ_UDATA: | |
1591 | if (!inode->data_vdi_id[idx]) { | |
1592 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
1593 | goto done; | |
1594 | } | |
1595 | break; | |
1596 | case AIOCB_WRITE_UDATA: | |
1597 | if (!inode->data_vdi_id[idx]) { | |
1598 | create = true; | |
1599 | } else if (!is_data_obj_writable(inode, idx)) { | |
1600 | /* Copy-On-Write */ | |
1601 | create = true; | |
1602 | old_oid = oid; | |
1603 | flags = SD_FLAG_CMD_COW; | |
1604 | } | |
1605 | break; | |
1606 | default: | |
1607 | break; | |
1608 | } | |
1609 | ||
1610 | if (create) { | |
1611 | dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", | |
1612 | inode->vdi_id, oid, | |
1613 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); | |
1614 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
1615 | dprintf("new oid %" PRIx64 "\n", oid); | |
1616 | } | |
1617 | ||
1618 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); | |
1619 | ||
1620 | if (create) { | |
1621 | AIOReq *areq; | |
1622 | QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { | |
1623 | if (areq->oid == oid) { | |
1624 | /* | |
1625 | * Sheepdog cannot handle simultaneous create | |
1626 | * requests to the same object. So we cannot send | |
1627 | * the request until the previous request | |
1628 | * finishes. | |
1629 | */ | |
1630 | aio_req->flags = 0; | |
1631 | aio_req->base_oid = 0; | |
1632 | QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, | |
1633 | aio_siblings); | |
1634 | goto done; | |
1635 | } | |
1636 | } | |
1637 | } | |
1638 | ||
1639 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
1640 | ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, | |
1641 | create, acb->aiocb_type); | |
1642 | if (ret < 0) { | |
1643 | error_report("add_aio_request is failed"); | |
1644 | free_aio_req(s, aio_req); | |
1645 | acb->ret = -EIO; | |
1646 | goto out; | |
1647 | } | |
1648 | done: | |
1649 | offset = 0; | |
1650 | idx++; | |
1651 | done += len; | |
1652 | } | |
1653 | out: | |
1654 | if (!--acb->nr_pending) { | |
1655 | return acb->ret; | |
1656 | } | |
1657 | return 1; | |
1658 | } | |
1659 | ||
1660 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, | |
1661 | int nb_sectors, QEMUIOVector *qiov) | |
1662 | { | |
1663 | SheepdogAIOCB *acb; | |
1664 | int ret; | |
1665 | ||
1666 | if (bs->growable && sector_num + nb_sectors > bs->total_sectors) { | |
1667 | ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE); | |
1668 | if (ret < 0) { | |
1669 | return ret; | |
1670 | } | |
1671 | bs->total_sectors = sector_num + nb_sectors; | |
1672 | } | |
1673 | ||
1674 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); | |
1675 | acb->aio_done_func = sd_write_done; | |
1676 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1677 | ||
1678 | ret = sd_co_rw_vector(acb); | |
1679 | if (ret <= 0) { | |
1680 | qemu_aio_release(acb); | |
1681 | return ret; | |
1682 | } | |
1683 | ||
1684 | qemu_coroutine_yield(); | |
1685 | ||
1686 | return acb->ret; | |
1687 | } | |
1688 | ||
1689 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, | |
1690 | int nb_sectors, QEMUIOVector *qiov) | |
1691 | { | |
1692 | SheepdogAIOCB *acb; | |
1693 | int ret; | |
1694 | ||
1695 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); | |
1696 | acb->aiocb_type = AIOCB_READ_UDATA; | |
1697 | acb->aio_done_func = sd_finish_aiocb; | |
1698 | ||
1699 | ret = sd_co_rw_vector(acb); | |
1700 | if (ret <= 0) { | |
1701 | qemu_aio_release(acb); | |
1702 | return ret; | |
1703 | } | |
1704 | ||
1705 | qemu_coroutine_yield(); | |
1706 | ||
1707 | return acb->ret; | |
1708 | } | |
1709 | ||
1710 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) | |
1711 | { | |
1712 | BDRVSheepdogState *s = bs->opaque; | |
1713 | SheepdogAIOCB *acb; | |
1714 | AIOReq *aio_req; | |
1715 | int ret; | |
1716 | ||
1717 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { | |
1718 | return 0; | |
1719 | } | |
1720 | ||
1721 | acb = sd_aio_setup(bs, NULL, 0, 0); | |
1722 | acb->aiocb_type = AIOCB_FLUSH_CACHE; | |
1723 | acb->aio_done_func = sd_finish_aiocb; | |
1724 | ||
1725 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
1726 | 0, 0, 0, 0, 0); | |
1727 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
1728 | ret = add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type); | |
1729 | if (ret < 0) { | |
1730 | error_report("add_aio_request is failed"); | |
1731 | free_aio_req(s, aio_req); | |
1732 | qemu_aio_release(acb); | |
1733 | return ret; | |
1734 | } | |
1735 | ||
1736 | qemu_coroutine_yield(); | |
1737 | return acb->ret; | |
1738 | } | |
1739 | ||
1740 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) | |
1741 | { | |
1742 | BDRVSheepdogState *s = bs->opaque; | |
1743 | int ret, fd; | |
1744 | uint32_t new_vid; | |
1745 | SheepdogInode *inode; | |
1746 | unsigned int datalen; | |
1747 | ||
1748 | dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " | |
1749 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, | |
1750 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
1751 | ||
1752 | if (s->is_snapshot) { | |
1753 | error_report("You can't create a snapshot of a snapshot VDI, " | |
1754 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); | |
1755 | ||
1756 | return -EINVAL; | |
1757 | } | |
1758 | ||
1759 | dprintf("%s %s\n", sn_info->name, sn_info->id_str); | |
1760 | ||
1761 | s->inode.vm_state_size = sn_info->vm_state_size; | |
1762 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
1763 | /* It appears that inode.tag does not require a NUL terminator, | |
1764 | * which means this use of strncpy is ok. | |
1765 | */ | |
1766 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); | |
1767 | /* we don't need to update entire object */ | |
1768 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1769 | ||
1770 | /* refresh inode. */ | |
1771 | fd = connect_to_sdog(s->addr, s->port); | |
1772 | if (fd < 0) { | |
1773 | ret = fd; | |
1774 | goto cleanup; | |
1775 | } | |
1776 | ||
1777 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
1778 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); | |
1779 | if (ret < 0) { | |
1780 | error_report("failed to write snapshot's inode."); | |
1781 | goto cleanup; | |
1782 | } | |
1783 | ||
1784 | ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, 1, | |
1785 | s->addr, s->port); | |
1786 | if (ret < 0) { | |
1787 | error_report("failed to create inode for snapshot. %s", | |
1788 | strerror(errno)); | |
1789 | goto cleanup; | |
1790 | } | |
1791 | ||
1792 | inode = (SheepdogInode *)g_malloc(datalen); | |
1793 | ||
1794 | ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid), | |
1795 | s->inode.nr_copies, datalen, 0, s->cache_flags); | |
1796 | ||
1797 | if (ret < 0) { | |
1798 | error_report("failed to read new inode info. %s", strerror(errno)); | |
1799 | goto cleanup; | |
1800 | } | |
1801 | ||
1802 | memcpy(&s->inode, inode, datalen); | |
1803 | dprintf("s->inode: name %s snap_id %x oid %x\n", | |
1804 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); | |
1805 | ||
1806 | cleanup: | |
1807 | closesocket(fd); | |
1808 | return ret; | |
1809 | } | |
1810 | ||
1811 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) | |
1812 | { | |
1813 | BDRVSheepdogState *s = bs->opaque; | |
1814 | BDRVSheepdogState *old_s; | |
1815 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1816 | char *buf = NULL; | |
1817 | uint32_t vid; | |
1818 | uint32_t snapid = 0; | |
1819 | int ret = 0, fd; | |
1820 | ||
1821 | old_s = g_malloc(sizeof(BDRVSheepdogState)); | |
1822 | ||
1823 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
1824 | ||
1825 | pstrcpy(vdi, sizeof(vdi), s->name); | |
1826 | ||
1827 | snapid = strtoul(snapshot_id, NULL, 10); | |
1828 | if (snapid) { | |
1829 | tag[0] = 0; | |
1830 | } else { | |
1831 | pstrcpy(tag, sizeof(tag), s->name); | |
1832 | } | |
1833 | ||
1834 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1); | |
1835 | if (ret) { | |
1836 | error_report("Failed to find_vdi_name"); | |
1837 | goto out; | |
1838 | } | |
1839 | ||
1840 | fd = connect_to_sdog(s->addr, s->port); | |
1841 | if (fd < 0) { | |
1842 | error_report("failed to connect"); | |
1843 | ret = fd; | |
1844 | goto out; | |
1845 | } | |
1846 | ||
1847 | buf = g_malloc(SD_INODE_SIZE); | |
1848 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, | |
1849 | SD_INODE_SIZE, 0, s->cache_flags); | |
1850 | ||
1851 | closesocket(fd); | |
1852 | ||
1853 | if (ret) { | |
1854 | goto out; | |
1855 | } | |
1856 | ||
1857 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1858 | ||
1859 | if (!s->inode.vm_state_size) { | |
1860 | error_report("Invalid snapshot"); | |
1861 | ret = -ENOENT; | |
1862 | goto out; | |
1863 | } | |
1864 | ||
1865 | s->is_snapshot = true; | |
1866 | ||
1867 | g_free(buf); | |
1868 | g_free(old_s); | |
1869 | ||
1870 | return 0; | |
1871 | out: | |
1872 | /* recover bdrv_sd_state */ | |
1873 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
1874 | g_free(buf); | |
1875 | g_free(old_s); | |
1876 | ||
1877 | error_report("failed to open. recover old bdrv_sd_state."); | |
1878 | ||
1879 | return ret; | |
1880 | } | |
1881 | ||
1882 | static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) | |
1883 | { | |
1884 | /* FIXME: Delete specified snapshot id. */ | |
1885 | return 0; | |
1886 | } | |
1887 | ||
1888 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) | |
1889 | { | |
1890 | BDRVSheepdogState *s = bs->opaque; | |
1891 | SheepdogReq req; | |
1892 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
1893 | QEMUSnapshotInfo *sn_tab = NULL; | |
1894 | unsigned wlen, rlen; | |
1895 | int found = 0; | |
1896 | static SheepdogInode inode; | |
1897 | unsigned long *vdi_inuse; | |
1898 | unsigned int start_nr; | |
1899 | uint64_t hval; | |
1900 | uint32_t vid; | |
1901 | ||
1902 | vdi_inuse = g_malloc(max); | |
1903 | ||
1904 | fd = connect_to_sdog(s->addr, s->port); | |
1905 | if (fd < 0) { | |
1906 | ret = fd; | |
1907 | goto out; | |
1908 | } | |
1909 | ||
1910 | rlen = max; | |
1911 | wlen = 0; | |
1912 | ||
1913 | memset(&req, 0, sizeof(req)); | |
1914 | ||
1915 | req.opcode = SD_OP_READ_VDIS; | |
1916 | req.data_length = max; | |
1917 | ||
1918 | ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); | |
1919 | ||
1920 | closesocket(fd); | |
1921 | if (ret) { | |
1922 | goto out; | |
1923 | } | |
1924 | ||
1925 | sn_tab = g_malloc0(nr * sizeof(*sn_tab)); | |
1926 | ||
1927 | /* calculate a vdi id with hash function */ | |
1928 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
1929 | start_nr = hval & (SD_NR_VDIS - 1); | |
1930 | ||
1931 | fd = connect_to_sdog(s->addr, s->port); | |
1932 | if (fd < 0) { | |
1933 | error_report("failed to connect"); | |
1934 | ret = fd; | |
1935 | goto out; | |
1936 | } | |
1937 | ||
1938 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
1939 | if (!test_bit(vid, vdi_inuse)) { | |
1940 | break; | |
1941 | } | |
1942 | ||
1943 | /* we don't need to read entire object */ | |
1944 | ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid), | |
1945 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, | |
1946 | s->cache_flags); | |
1947 | ||
1948 | if (ret) { | |
1949 | continue; | |
1950 | } | |
1951 | ||
1952 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
1953 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
1954 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
1955 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
1956 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
1957 | ||
1958 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", | |
1959 | inode.snap_id); | |
1960 | pstrcpy(sn_tab[found].name, | |
1961 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
1962 | inode.tag); | |
1963 | found++; | |
1964 | } | |
1965 | } | |
1966 | ||
1967 | closesocket(fd); | |
1968 | out: | |
1969 | *psn_tab = sn_tab; | |
1970 | ||
1971 | g_free(vdi_inuse); | |
1972 | ||
1973 | if (ret < 0) { | |
1974 | return ret; | |
1975 | } | |
1976 | ||
1977 | return found; | |
1978 | } | |
1979 | ||
1980 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
1981 | int64_t pos, int size, int load) | |
1982 | { | |
1983 | bool create; | |
1984 | int fd, ret = 0, remaining = size; | |
1985 | unsigned int data_len; | |
1986 | uint64_t vmstate_oid; | |
1987 | uint32_t vdi_index; | |
1988 | uint64_t offset; | |
1989 | ||
1990 | fd = connect_to_sdog(s->addr, s->port); | |
1991 | if (fd < 0) { | |
1992 | return fd; | |
1993 | } | |
1994 | ||
1995 | while (remaining) { | |
1996 | vdi_index = pos / SD_DATA_OBJ_SIZE; | |
1997 | offset = pos % SD_DATA_OBJ_SIZE; | |
1998 | ||
1999 | data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset); | |
2000 | ||
2001 | vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index); | |
2002 | ||
2003 | create = (offset == 0); | |
2004 | if (load) { | |
2005 | ret = read_object(fd, (char *)data, vmstate_oid, | |
2006 | s->inode.nr_copies, data_len, offset, | |
2007 | s->cache_flags); | |
2008 | } else { | |
2009 | ret = write_object(fd, (char *)data, vmstate_oid, | |
2010 | s->inode.nr_copies, data_len, offset, create, | |
2011 | s->cache_flags); | |
2012 | } | |
2013 | ||
2014 | if (ret < 0) { | |
2015 | error_report("failed to save vmstate %s", strerror(errno)); | |
2016 | goto cleanup; | |
2017 | } | |
2018 | ||
2019 | pos += data_len; | |
2020 | data += data_len; | |
2021 | remaining -= data_len; | |
2022 | } | |
2023 | ret = size; | |
2024 | cleanup: | |
2025 | closesocket(fd); | |
2026 | return ret; | |
2027 | } | |
2028 | ||
2029 | static int sd_save_vmstate(BlockDriverState *bs, const uint8_t *data, | |
2030 | int64_t pos, int size) | |
2031 | { | |
2032 | BDRVSheepdogState *s = bs->opaque; | |
2033 | ||
2034 | return do_load_save_vmstate(s, (uint8_t *)data, pos, size, 0); | |
2035 | } | |
2036 | ||
2037 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
2038 | int64_t pos, int size) | |
2039 | { | |
2040 | BDRVSheepdogState *s = bs->opaque; | |
2041 | ||
2042 | return do_load_save_vmstate(s, data, pos, size, 1); | |
2043 | } | |
2044 | ||
2045 | ||
2046 | static QEMUOptionParameter sd_create_options[] = { | |
2047 | { | |
2048 | .name = BLOCK_OPT_SIZE, | |
2049 | .type = OPT_SIZE, | |
2050 | .help = "Virtual disk size" | |
2051 | }, | |
2052 | { | |
2053 | .name = BLOCK_OPT_BACKING_FILE, | |
2054 | .type = OPT_STRING, | |
2055 | .help = "File name of a base image" | |
2056 | }, | |
2057 | { | |
2058 | .name = BLOCK_OPT_PREALLOC, | |
2059 | .type = OPT_STRING, | |
2060 | .help = "Preallocation mode (allowed values: off, full)" | |
2061 | }, | |
2062 | { NULL } | |
2063 | }; | |
2064 | ||
2065 | BlockDriver bdrv_sheepdog = { | |
2066 | .format_name = "sheepdog", | |
2067 | .protocol_name = "sheepdog", | |
2068 | .instance_size = sizeof(BDRVSheepdogState), | |
2069 | .bdrv_file_open = sd_open, | |
2070 | .bdrv_close = sd_close, | |
2071 | .bdrv_create = sd_create, | |
2072 | .bdrv_getlength = sd_getlength, | |
2073 | .bdrv_truncate = sd_truncate, | |
2074 | ||
2075 | .bdrv_co_readv = sd_co_readv, | |
2076 | .bdrv_co_writev = sd_co_writev, | |
2077 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
2078 | ||
2079 | .bdrv_snapshot_create = sd_snapshot_create, | |
2080 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2081 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2082 | .bdrv_snapshot_list = sd_snapshot_list, | |
2083 | ||
2084 | .bdrv_save_vmstate = sd_save_vmstate, | |
2085 | .bdrv_load_vmstate = sd_load_vmstate, | |
2086 | ||
2087 | .create_options = sd_create_options, | |
2088 | }; | |
2089 | ||
2090 | static void bdrv_sheepdog_init(void) | |
2091 | { | |
2092 | bdrv_register(&bdrv_sheepdog); | |
2093 | } | |
2094 | block_init(bdrv_sheepdog_init); |