]>
Commit | Line | Data |
---|---|---|
33b1db1c MK |
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/>. | |
6b620ca3 PB |
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. | |
33b1db1c | 13 | */ |
33b1db1c MK |
14 | |
15 | #include "qemu-common.h" | |
16 | #include "qemu-error.h" | |
17 | #include "qemu_socket.h" | |
18 | #include "block_int.h" | |
8cffde73 | 19 | #include "bitops.h" |
33b1db1c MK |
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 | |
47622c44 | 35 | #define SD_OP_FLUSH_VDI 0x16 |
33b1db1c MK |
36 | |
37 | #define SD_FLAG_CMD_WRITE 0x01 | |
38 | #define SD_FLAG_CMD_COW 0x02 | |
47622c44 | 39 | #define SD_FLAG_CMD_CACHE 0x04 |
33b1db1c MK |
40 | |
41 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
42 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
43 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
44 | #define SD_RES_EIO 0x03 /* I/O error */ | |
45 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
46 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
47 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
48 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
49 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
50 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
51 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
52 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
53 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
54 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
55 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
56 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
57 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
58 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
59 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
60 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
61 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
62 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
63 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
64 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
65 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
66 | ||
67 | /* | |
68 | * Object ID rules | |
69 | * | |
70 | * 0 - 19 (20 bits): data object space | |
71 | * 20 - 31 (12 bits): reserved data object space | |
72 | * 32 - 55 (24 bits): vdi object space | |
73 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 74 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
75 | */ |
76 | ||
77 | #define VDI_SPACE_SHIFT 32 | |
78 | #define VDI_BIT (UINT64_C(1) << 63) | |
79 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
80 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
81 | #define MAX_CHILDREN 1024 | |
82 | #define SD_MAX_VDI_LEN 256 | |
83 | #define SD_MAX_VDI_TAG_LEN 256 | |
84 | #define SD_NR_VDIS (1U << 24) | |
85 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
86 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
87 | #define SECTOR_SIZE 512 | |
88 | ||
89 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
90 | #define CURRENT_VDI_ID 0 | |
91 | ||
92 | typedef struct SheepdogReq { | |
93 | uint8_t proto_ver; | |
94 | uint8_t opcode; | |
95 | uint16_t flags; | |
96 | uint32_t epoch; | |
97 | uint32_t id; | |
98 | uint32_t data_length; | |
99 | uint32_t opcode_specific[8]; | |
100 | } SheepdogReq; | |
101 | ||
102 | typedef struct SheepdogRsp { | |
103 | uint8_t proto_ver; | |
104 | uint8_t opcode; | |
105 | uint16_t flags; | |
106 | uint32_t epoch; | |
107 | uint32_t id; | |
108 | uint32_t data_length; | |
109 | uint32_t result; | |
110 | uint32_t opcode_specific[7]; | |
111 | } SheepdogRsp; | |
112 | ||
113 | typedef struct SheepdogObjReq { | |
114 | uint8_t proto_ver; | |
115 | uint8_t opcode; | |
116 | uint16_t flags; | |
117 | uint32_t epoch; | |
118 | uint32_t id; | |
119 | uint32_t data_length; | |
120 | uint64_t oid; | |
121 | uint64_t cow_oid; | |
122 | uint32_t copies; | |
123 | uint32_t rsvd; | |
124 | uint64_t offset; | |
125 | } SheepdogObjReq; | |
126 | ||
127 | typedef struct SheepdogObjRsp { | |
128 | uint8_t proto_ver; | |
129 | uint8_t opcode; | |
130 | uint16_t flags; | |
131 | uint32_t epoch; | |
132 | uint32_t id; | |
133 | uint32_t data_length; | |
134 | uint32_t result; | |
135 | uint32_t copies; | |
136 | uint32_t pad[6]; | |
137 | } SheepdogObjRsp; | |
138 | ||
139 | typedef struct SheepdogVdiReq { | |
140 | uint8_t proto_ver; | |
141 | uint8_t opcode; | |
142 | uint16_t flags; | |
143 | uint32_t epoch; | |
144 | uint32_t id; | |
145 | uint32_t data_length; | |
146 | uint64_t vdi_size; | |
147 | uint32_t base_vdi_id; | |
148 | uint32_t copies; | |
149 | uint32_t snapid; | |
150 | uint32_t pad[3]; | |
151 | } SheepdogVdiReq; | |
152 | ||
153 | typedef struct SheepdogVdiRsp { | |
154 | uint8_t proto_ver; | |
155 | uint8_t opcode; | |
156 | uint16_t flags; | |
157 | uint32_t epoch; | |
158 | uint32_t id; | |
159 | uint32_t data_length; | |
160 | uint32_t result; | |
161 | uint32_t rsvd; | |
162 | uint32_t vdi_id; | |
163 | uint32_t pad[5]; | |
164 | } SheepdogVdiRsp; | |
165 | ||
166 | typedef struct SheepdogInode { | |
167 | char name[SD_MAX_VDI_LEN]; | |
168 | char tag[SD_MAX_VDI_TAG_LEN]; | |
169 | uint64_t ctime; | |
170 | uint64_t snap_ctime; | |
171 | uint64_t vm_clock_nsec; | |
172 | uint64_t vdi_size; | |
173 | uint64_t vm_state_size; | |
174 | uint16_t copy_policy; | |
175 | uint8_t nr_copies; | |
176 | uint8_t block_size_shift; | |
177 | uint32_t snap_id; | |
178 | uint32_t vdi_id; | |
179 | uint32_t parent_vdi_id; | |
180 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
181 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
182 | } SheepdogInode; | |
183 | ||
184 | /* | |
185 | * 64 bit FNV-1a non-zero initial basis | |
186 | */ | |
187 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
188 | ||
189 | /* | |
190 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
191 | */ | |
192 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
193 | { | |
194 | unsigned char *bp = buf; | |
195 | unsigned char *be = bp + len; | |
196 | while (bp < be) { | |
197 | hval ^= (uint64_t) *bp++; | |
198 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
199 | (hval << 7) + (hval << 8) + (hval << 40); | |
200 | } | |
201 | return hval; | |
202 | } | |
203 | ||
ebabb67a | 204 | static inline int is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
205 | { |
206 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
207 | } | |
208 | ||
209 | static inline int is_data_obj(uint64_t oid) | |
210 | { | |
211 | return !(VDI_BIT & oid); | |
212 | } | |
213 | ||
214 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
215 | { | |
216 | return oid & (MAX_DATA_OBJS - 1); | |
217 | } | |
218 | ||
219 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) | |
220 | { | |
221 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
222 | } | |
223 | ||
224 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
225 | { | |
226 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
227 | } | |
228 | ||
229 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
230 | { | |
231 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
232 | } | |
233 | ||
234 | static inline int is_snapshot(struct SheepdogInode *inode) | |
235 | { | |
236 | return !!inode->snap_ctime; | |
237 | } | |
238 | ||
239 | #undef dprintf | |
240 | #ifdef DEBUG_SDOG | |
241 | #define dprintf(fmt, args...) \ | |
242 | do { \ | |
243 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
244 | } while (0) | |
245 | #else | |
246 | #define dprintf(fmt, args...) | |
247 | #endif | |
248 | ||
249 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
250 | ||
251 | typedef struct AIOReq { | |
252 | SheepdogAIOCB *aiocb; | |
253 | unsigned int iov_offset; | |
254 | ||
255 | uint64_t oid; | |
256 | uint64_t base_oid; | |
257 | uint64_t offset; | |
258 | unsigned int data_len; | |
259 | uint8_t flags; | |
260 | uint32_t id; | |
261 | ||
262 | QLIST_ENTRY(AIOReq) outstanding_aio_siblings; | |
263 | QLIST_ENTRY(AIOReq) aioreq_siblings; | |
264 | } AIOReq; | |
265 | ||
266 | enum AIOCBState { | |
267 | AIOCB_WRITE_UDATA, | |
268 | AIOCB_READ_UDATA, | |
269 | }; | |
270 | ||
271 | struct SheepdogAIOCB { | |
272 | BlockDriverAIOCB common; | |
273 | ||
274 | QEMUIOVector *qiov; | |
275 | ||
276 | int64_t sector_num; | |
277 | int nb_sectors; | |
278 | ||
279 | int ret; | |
280 | enum AIOCBState aiocb_type; | |
281 | ||
2df46246 | 282 | Coroutine *coroutine; |
33b1db1c MK |
283 | void (*aio_done_func)(SheepdogAIOCB *); |
284 | ||
285 | int canceled; | |
286 | ||
287 | QLIST_HEAD(aioreq_head, AIOReq) aioreq_head; | |
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 | int is_snapshot; | |
47622c44 | 298 | uint8_t cache_enabled; |
33b1db1c MK |
299 | |
300 | char *addr; | |
301 | char *port; | |
302 | int fd; | |
47622c44 | 303 | int flush_fd; |
33b1db1c | 304 | |
2df46246 MK |
305 | CoMutex lock; |
306 | Coroutine *co_send; | |
307 | Coroutine *co_recv; | |
308 | ||
33b1db1c MK |
309 | uint32_t aioreq_seq_num; |
310 | QLIST_HEAD(outstanding_aio_head, AIOReq) outstanding_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 | * | |
2df46246 MK |
360 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
361 | * link the requests to the outstanding_list in the | |
362 | * BDRVSheepdogState. The function exits without waiting for | |
363 | * receiving the response. | |
33b1db1c | 364 | * |
2df46246 | 365 | * 2. We receive the response in aio_read_response, the fd handler to |
33b1db1c MK |
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 | |
2df46246 MK |
368 | * completion function. We switch back to sd_co_readv/writev after |
369 | * all the requests belonging to the AIOCB are finished. | |
33b1db1c MK |
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 | ||
7267c094 | 379 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
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 | QLIST_INSERT_HEAD(&s->outstanding_aio_head, aio_req, | |
390 | outstanding_aio_siblings); | |
391 | QLIST_INSERT_HEAD(&acb->aioreq_head, aio_req, aioreq_siblings); | |
392 | ||
393 | return aio_req; | |
394 | } | |
395 | ||
396 | static inline int free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) | |
397 | { | |
398 | SheepdogAIOCB *acb = aio_req->aiocb; | |
399 | QLIST_REMOVE(aio_req, outstanding_aio_siblings); | |
400 | QLIST_REMOVE(aio_req, aioreq_siblings); | |
7267c094 | 401 | g_free(aio_req); |
33b1db1c MK |
402 | |
403 | return !QLIST_EMPTY(&acb->aioreq_head); | |
404 | } | |
405 | ||
d8716b41 | 406 | static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
33b1db1c MK |
407 | { |
408 | if (!acb->canceled) { | |
2df46246 | 409 | qemu_coroutine_enter(acb->coroutine, NULL); |
33b1db1c MK |
410 | } |
411 | qemu_aio_release(acb); | |
412 | } | |
413 | ||
414 | static void sd_aio_cancel(BlockDriverAIOCB *blockacb) | |
415 | { | |
416 | SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; | |
417 | ||
418 | /* | |
419 | * Sheepdog cannot cancel the requests which are already sent to | |
420 | * the servers, so we just complete the request with -EIO here. | |
421 | */ | |
2df46246 MK |
422 | acb->ret = -EIO; |
423 | qemu_coroutine_enter(acb->coroutine, NULL); | |
33b1db1c MK |
424 | acb->canceled = 1; |
425 | } | |
426 | ||
427 | static AIOPool sd_aio_pool = { | |
428 | .aiocb_size = sizeof(SheepdogAIOCB), | |
429 | .cancel = sd_aio_cancel, | |
430 | }; | |
431 | ||
432 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
433 | int64_t sector_num, int nb_sectors, | |
434 | BlockDriverCompletionFunc *cb, void *opaque) | |
435 | { | |
436 | SheepdogAIOCB *acb; | |
437 | ||
438 | acb = qemu_aio_get(&sd_aio_pool, bs, cb, opaque); | |
439 | ||
440 | acb->qiov = qiov; | |
441 | ||
442 | acb->sector_num = sector_num; | |
443 | acb->nb_sectors = nb_sectors; | |
444 | ||
445 | acb->aio_done_func = NULL; | |
446 | acb->canceled = 0; | |
2df46246 | 447 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c MK |
448 | acb->ret = 0; |
449 | QLIST_INIT(&acb->aioreq_head); | |
450 | return acb; | |
451 | } | |
452 | ||
33b1db1c MK |
453 | static int connect_to_sdog(const char *addr, const char *port) |
454 | { | |
455 | char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV]; | |
456 | int fd, ret; | |
457 | struct addrinfo hints, *res, *res0; | |
458 | ||
459 | if (!addr) { | |
460 | addr = SD_DEFAULT_ADDR; | |
461 | port = SD_DEFAULT_PORT; | |
462 | } | |
463 | ||
464 | memset(&hints, 0, sizeof(hints)); | |
465 | hints.ai_socktype = SOCK_STREAM; | |
466 | ||
467 | ret = getaddrinfo(addr, port, &hints, &res0); | |
468 | if (ret) { | |
6daf194d | 469 | error_report("unable to get address info %s, %s", |
33b1db1c MK |
470 | addr, strerror(errno)); |
471 | return -1; | |
472 | } | |
473 | ||
474 | for (res = res0; res; res = res->ai_next) { | |
475 | ret = getnameinfo(res->ai_addr, res->ai_addrlen, hbuf, sizeof(hbuf), | |
476 | sbuf, sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV); | |
477 | if (ret) { | |
478 | continue; | |
479 | } | |
480 | ||
481 | fd = socket(res->ai_family, res->ai_socktype, res->ai_protocol); | |
482 | if (fd < 0) { | |
483 | continue; | |
484 | } | |
485 | ||
486 | reconnect: | |
487 | ret = connect(fd, res->ai_addr, res->ai_addrlen); | |
488 | if (ret < 0) { | |
489 | if (errno == EINTR) { | |
490 | goto reconnect; | |
491 | } | |
492 | break; | |
493 | } | |
494 | ||
495 | dprintf("connected to %s:%s\n", addr, port); | |
496 | goto success; | |
497 | } | |
498 | fd = -1; | |
6daf194d | 499 | error_report("failed connect to %s:%s", addr, port); |
33b1db1c MK |
500 | success: |
501 | freeaddrinfo(res0); | |
502 | return fd; | |
503 | } | |
504 | ||
33b1db1c MK |
505 | static int send_req(int sockfd, SheepdogReq *hdr, void *data, |
506 | unsigned int *wlen) | |
507 | { | |
508 | int ret; | |
33b1db1c | 509 | |
8c5135f9 PB |
510 | ret = qemu_send_full(sockfd, hdr, sizeof(*hdr), 0); |
511 | if (ret < sizeof(*hdr)) { | |
512 | error_report("failed to send a req, %s", strerror(errno)); | |
eb092180 | 513 | return ret; |
33b1db1c MK |
514 | } |
515 | ||
8c5135f9 PB |
516 | ret = qemu_send_full(sockfd, data, *wlen, 0); |
517 | if (ret < *wlen) { | |
6daf194d | 518 | error_report("failed to send a req, %s", strerror(errno)); |
33b1db1c MK |
519 | } |
520 | ||
521 | return ret; | |
522 | } | |
523 | ||
47622c44 LY |
524 | static int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
525 | unsigned int *wlen) | |
526 | { | |
527 | int ret; | |
528 | ||
529 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
530 | if (ret < sizeof(*hdr)) { | |
531 | error_report("failed to send a req, %s", strerror(errno)); | |
eb092180 | 532 | return ret; |
47622c44 LY |
533 | } |
534 | ||
535 | ret = qemu_co_send(sockfd, data, *wlen); | |
536 | if (ret < *wlen) { | |
537 | error_report("failed to send a req, %s", strerror(errno)); | |
538 | } | |
539 | ||
540 | return ret; | |
541 | } | |
33b1db1c MK |
542 | static int do_req(int sockfd, SheepdogReq *hdr, void *data, |
543 | unsigned int *wlen, unsigned int *rlen) | |
544 | { | |
545 | int ret; | |
546 | ||
8c5135f9 | 547 | socket_set_block(sockfd); |
33b1db1c | 548 | ret = send_req(sockfd, hdr, data, wlen); |
8c5135f9 | 549 | if (ret < 0) { |
33b1db1c MK |
550 | goto out; |
551 | } | |
552 | ||
8c5135f9 PB |
553 | ret = qemu_recv_full(sockfd, hdr, sizeof(*hdr), 0); |
554 | if (ret < sizeof(*hdr)) { | |
6daf194d | 555 | error_report("failed to get a rsp, %s", strerror(errno)); |
33b1db1c MK |
556 | goto out; |
557 | } | |
558 | ||
559 | if (*rlen > hdr->data_length) { | |
560 | *rlen = hdr->data_length; | |
561 | } | |
562 | ||
563 | if (*rlen) { | |
8c5135f9 PB |
564 | ret = qemu_recv_full(sockfd, data, *rlen, 0); |
565 | if (ret < *rlen) { | |
6daf194d | 566 | error_report("failed to get the data, %s", strerror(errno)); |
33b1db1c MK |
567 | goto out; |
568 | } | |
569 | } | |
570 | ret = 0; | |
571 | out: | |
8c5135f9 | 572 | socket_set_nonblock(sockfd); |
33b1db1c MK |
573 | return ret; |
574 | } | |
575 | ||
47622c44 LY |
576 | static int do_co_req(int sockfd, SheepdogReq *hdr, void *data, |
577 | unsigned int *wlen, unsigned int *rlen) | |
578 | { | |
579 | int ret; | |
580 | ||
581 | socket_set_block(sockfd); | |
582 | ret = send_co_req(sockfd, hdr, data, wlen); | |
583 | if (ret < 0) { | |
584 | goto out; | |
585 | } | |
586 | ||
587 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); | |
588 | if (ret < sizeof(*hdr)) { | |
589 | error_report("failed to get a rsp, %s", strerror(errno)); | |
590 | goto out; | |
591 | } | |
592 | ||
593 | if (*rlen > hdr->data_length) { | |
594 | *rlen = hdr->data_length; | |
595 | } | |
596 | ||
597 | if (*rlen) { | |
598 | ret = qemu_co_recv(sockfd, data, *rlen); | |
599 | if (ret < *rlen) { | |
600 | error_report("failed to get the data, %s", strerror(errno)); | |
601 | goto out; | |
602 | } | |
603 | } | |
604 | ret = 0; | |
605 | out: | |
606 | socket_set_nonblock(sockfd); | |
607 | return ret; | |
608 | } | |
609 | ||
d8716b41 | 610 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
33b1db1c MK |
611 | struct iovec *iov, int niov, int create, |
612 | enum AIOCBState aiocb_type); | |
613 | ||
614 | /* | |
615 | * This function searchs pending requests to the object `oid', and | |
616 | * sends them. | |
617 | */ | |
d8716b41 | 618 | static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid, uint32_t id) |
33b1db1c MK |
619 | { |
620 | AIOReq *aio_req, *next; | |
621 | SheepdogAIOCB *acb; | |
622 | int ret; | |
623 | ||
624 | QLIST_FOREACH_SAFE(aio_req, &s->outstanding_aio_head, | |
625 | outstanding_aio_siblings, next) { | |
626 | if (id == aio_req->id) { | |
627 | continue; | |
628 | } | |
629 | if (aio_req->oid != oid) { | |
630 | continue; | |
631 | } | |
632 | ||
633 | acb = aio_req->aiocb; | |
634 | ret = add_aio_request(s, aio_req, acb->qiov->iov, | |
635 | acb->qiov->niov, 0, acb->aiocb_type); | |
636 | if (ret < 0) { | |
6daf194d | 637 | error_report("add_aio_request is failed"); |
33b1db1c MK |
638 | free_aio_req(s, aio_req); |
639 | if (QLIST_EMPTY(&acb->aioreq_head)) { | |
640 | sd_finish_aiocb(acb); | |
641 | } | |
642 | } | |
643 | } | |
644 | } | |
645 | ||
646 | /* | |
647 | * Receive responses of the I/O requests. | |
648 | * | |
649 | * This function is registered as a fd handler, and called from the | |
650 | * main loop when s->fd is ready for reading responses. | |
651 | */ | |
d8716b41 | 652 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
653 | { |
654 | SheepdogObjRsp rsp; | |
655 | BDRVSheepdogState *s = opaque; | |
656 | int fd = s->fd; | |
657 | int ret; | |
658 | AIOReq *aio_req = NULL; | |
659 | SheepdogAIOCB *acb; | |
660 | int rest; | |
661 | unsigned long idx; | |
662 | ||
663 | if (QLIST_EMPTY(&s->outstanding_aio_head)) { | |
2df46246 | 664 | goto out; |
33b1db1c MK |
665 | } |
666 | ||
667 | /* read a header */ | |
8c5135f9 PB |
668 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
669 | if (ret < 0) { | |
6daf194d | 670 | error_report("failed to get the header, %s", strerror(errno)); |
2df46246 | 671 | goto out; |
33b1db1c MK |
672 | } |
673 | ||
674 | /* find the right aio_req from the outstanding_aio list */ | |
675 | QLIST_FOREACH(aio_req, &s->outstanding_aio_head, outstanding_aio_siblings) { | |
676 | if (aio_req->id == rsp.id) { | |
677 | break; | |
678 | } | |
679 | } | |
680 | if (!aio_req) { | |
6daf194d | 681 | error_report("cannot find aio_req %x", rsp.id); |
2df46246 | 682 | goto out; |
33b1db1c MK |
683 | } |
684 | ||
685 | acb = aio_req->aiocb; | |
686 | ||
687 | switch (acb->aiocb_type) { | |
688 | case AIOCB_WRITE_UDATA: | |
6d1acda8 MK |
689 | /* this coroutine context is no longer suitable for co_recv |
690 | * because we may send data to update vdi objects */ | |
691 | s->co_recv = NULL; | |
33b1db1c MK |
692 | if (!is_data_obj(aio_req->oid)) { |
693 | break; | |
694 | } | |
695 | idx = data_oid_to_idx(aio_req->oid); | |
696 | ||
697 | if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) { | |
698 | /* | |
699 | * If the object is newly created one, we need to update | |
700 | * the vdi object (metadata object). min_dirty_data_idx | |
701 | * and max_dirty_data_idx are changed to include updated | |
702 | * index between them. | |
703 | */ | |
704 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
705 | s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); | |
706 | s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); | |
707 | ||
708 | /* | |
709 | * Some requests may be blocked because simultaneous | |
710 | * create requests are not allowed, so we search the | |
711 | * pending requests here. | |
712 | */ | |
713 | send_pending_req(s, vid_to_data_oid(s->inode.vdi_id, idx), rsp.id); | |
714 | } | |
715 | break; | |
716 | case AIOCB_READ_UDATA: | |
8c5135f9 PB |
717 | ret = qemu_co_recvv(fd, acb->qiov->iov, rsp.data_length, |
718 | aio_req->iov_offset); | |
719 | if (ret < 0) { | |
6daf194d | 720 | error_report("failed to get the data, %s", strerror(errno)); |
2df46246 | 721 | goto out; |
33b1db1c MK |
722 | } |
723 | break; | |
724 | } | |
725 | ||
726 | if (rsp.result != SD_RES_SUCCESS) { | |
727 | acb->ret = -EIO; | |
6daf194d | 728 | error_report("%s", sd_strerror(rsp.result)); |
33b1db1c MK |
729 | } |
730 | ||
731 | rest = free_aio_req(s, aio_req); | |
732 | if (!rest) { | |
733 | /* | |
734 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 735 | * we can switch back to sd_co_readv/writev now. |
33b1db1c MK |
736 | */ |
737 | acb->aio_done_func(acb); | |
738 | } | |
2df46246 MK |
739 | out: |
740 | s->co_recv = NULL; | |
741 | } | |
742 | ||
743 | static void co_read_response(void *opaque) | |
744 | { | |
745 | BDRVSheepdogState *s = opaque; | |
746 | ||
747 | if (!s->co_recv) { | |
748 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
749 | } | |
750 | ||
751 | qemu_coroutine_enter(s->co_recv, opaque); | |
752 | } | |
753 | ||
754 | static void co_write_request(void *opaque) | |
755 | { | |
756 | BDRVSheepdogState *s = opaque; | |
757 | ||
758 | qemu_coroutine_enter(s->co_send, NULL); | |
33b1db1c MK |
759 | } |
760 | ||
761 | static int aio_flush_request(void *opaque) | |
762 | { | |
763 | BDRVSheepdogState *s = opaque; | |
764 | ||
765 | return !QLIST_EMPTY(&s->outstanding_aio_head); | |
766 | } | |
767 | ||
33b1db1c MK |
768 | static int set_nodelay(int fd) |
769 | { | |
770 | int ret, opt; | |
771 | ||
772 | opt = 1; | |
773 | ret = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *)&opt, sizeof(opt)); | |
774 | return ret; | |
775 | } | |
776 | ||
777 | /* | |
778 | * Return a socket discriptor to read/write objects. | |
779 | * | |
780 | * We cannot use this discriptor for other operations because | |
781 | * the block driver may be on waiting response from the server. | |
782 | */ | |
783 | static int get_sheep_fd(BDRVSheepdogState *s) | |
784 | { | |
785 | int ret, fd; | |
786 | ||
787 | fd = connect_to_sdog(s->addr, s->port); | |
788 | if (fd < 0) { | |
6daf194d | 789 | error_report("%s", strerror(errno)); |
33b1db1c MK |
790 | return -1; |
791 | } | |
792 | ||
793 | socket_set_nonblock(fd); | |
794 | ||
795 | ret = set_nodelay(fd); | |
796 | if (ret) { | |
6daf194d | 797 | error_report("%s", strerror(errno)); |
33b1db1c MK |
798 | closesocket(fd); |
799 | return -1; | |
800 | } | |
801 | ||
2df46246 | 802 | qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, |
33b1db1c MK |
803 | NULL, s); |
804 | return fd; | |
805 | } | |
806 | ||
807 | /* | |
808 | * Parse a filename | |
809 | * | |
810 | * filename must be one of the following formats: | |
811 | * 1. [vdiname] | |
812 | * 2. [vdiname]:[snapid] | |
813 | * 3. [vdiname]:[tag] | |
814 | * 4. [hostname]:[port]:[vdiname] | |
815 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
816 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
817 | * | |
818 | * You can boot from the snapshot images by specifying `snapid` or | |
819 | * `tag'. | |
820 | * | |
821 | * You can run VMs outside the Sheepdog cluster by specifying | |
822 | * `hostname' and `port' (experimental). | |
823 | */ | |
824 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
825 | char *vdi, uint32_t *snapid, char *tag) | |
826 | { | |
827 | char *p, *q; | |
828 | int nr_sep; | |
829 | ||
7267c094 | 830 | p = q = g_strdup(filename); |
33b1db1c MK |
831 | |
832 | /* count the number of separators */ | |
833 | nr_sep = 0; | |
834 | while (*p) { | |
835 | if (*p == ':') { | |
836 | nr_sep++; | |
837 | } | |
838 | p++; | |
839 | } | |
840 | p = q; | |
841 | ||
842 | /* use the first two tokens as hostname and port number. */ | |
843 | if (nr_sep >= 2) { | |
844 | s->addr = p; | |
845 | p = strchr(p, ':'); | |
846 | *p++ = '\0'; | |
847 | ||
848 | s->port = p; | |
849 | p = strchr(p, ':'); | |
850 | *p++ = '\0'; | |
851 | } else { | |
852 | s->addr = NULL; | |
853 | s->port = 0; | |
854 | } | |
855 | ||
856 | strncpy(vdi, p, SD_MAX_VDI_LEN); | |
857 | ||
858 | p = strchr(vdi, ':'); | |
859 | if (p) { | |
860 | *p++ = '\0'; | |
861 | *snapid = strtoul(p, NULL, 10); | |
862 | if (*snapid == 0) { | |
863 | strncpy(tag, p, SD_MAX_VDI_TAG_LEN); | |
864 | } | |
865 | } else { | |
866 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
867 | } | |
868 | ||
869 | if (s->addr == NULL) { | |
7267c094 | 870 | g_free(q); |
33b1db1c MK |
871 | } |
872 | ||
873 | return 0; | |
874 | } | |
875 | ||
876 | static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid, | |
877 | char *tag, uint32_t *vid, int for_snapshot) | |
878 | { | |
879 | int ret, fd; | |
880 | SheepdogVdiReq hdr; | |
881 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
882 | unsigned int wlen, rlen = 0; | |
883 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
884 | ||
885 | fd = connect_to_sdog(s->addr, s->port); | |
886 | if (fd < 0) { | |
887 | return -1; | |
888 | } | |
889 | ||
890 | memset(buf, 0, sizeof(buf)); | |
891 | strncpy(buf, filename, SD_MAX_VDI_LEN); | |
892 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
893 | ||
894 | memset(&hdr, 0, sizeof(hdr)); | |
895 | if (for_snapshot) { | |
896 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
897 | } else { | |
898 | hdr.opcode = SD_OP_LOCK_VDI; | |
899 | } | |
900 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
901 | hdr.proto_ver = SD_PROTO_VER; | |
902 | hdr.data_length = wlen; | |
903 | hdr.snapid = snapid; | |
904 | hdr.flags = SD_FLAG_CMD_WRITE; | |
905 | ||
906 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
907 | if (ret) { | |
908 | ret = -1; | |
909 | goto out; | |
910 | } | |
911 | ||
912 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 913 | error_report("cannot get vdi info, %s, %s %d %s", |
33b1db1c MK |
914 | sd_strerror(rsp->result), filename, snapid, tag); |
915 | ret = -1; | |
916 | goto out; | |
917 | } | |
918 | *vid = rsp->vdi_id; | |
919 | ||
920 | ret = 0; | |
921 | out: | |
922 | closesocket(fd); | |
923 | return ret; | |
924 | } | |
925 | ||
d8716b41 | 926 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
33b1db1c MK |
927 | struct iovec *iov, int niov, int create, |
928 | enum AIOCBState aiocb_type) | |
929 | { | |
930 | int nr_copies = s->inode.nr_copies; | |
931 | SheepdogObjReq hdr; | |
932 | unsigned int wlen; | |
933 | int ret; | |
934 | uint64_t oid = aio_req->oid; | |
935 | unsigned int datalen = aio_req->data_len; | |
936 | uint64_t offset = aio_req->offset; | |
937 | uint8_t flags = aio_req->flags; | |
938 | uint64_t old_oid = aio_req->base_oid; | |
939 | ||
940 | if (!nr_copies) { | |
6daf194d | 941 | error_report("bug"); |
33b1db1c MK |
942 | } |
943 | ||
944 | memset(&hdr, 0, sizeof(hdr)); | |
945 | ||
946 | if (aiocb_type == AIOCB_READ_UDATA) { | |
947 | wlen = 0; | |
948 | hdr.opcode = SD_OP_READ_OBJ; | |
949 | hdr.flags = flags; | |
950 | } else if (create) { | |
951 | wlen = datalen; | |
952 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
953 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
954 | } else { | |
955 | wlen = datalen; | |
956 | hdr.opcode = SD_OP_WRITE_OBJ; | |
957 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
958 | } | |
959 | ||
47622c44 LY |
960 | if (s->cache_enabled) { |
961 | hdr.flags |= SD_FLAG_CMD_CACHE; | |
962 | } | |
963 | ||
33b1db1c MK |
964 | hdr.oid = oid; |
965 | hdr.cow_oid = old_oid; | |
966 | hdr.copies = s->inode.nr_copies; | |
967 | ||
968 | hdr.data_length = datalen; | |
969 | hdr.offset = offset; | |
970 | ||
971 | hdr.id = aio_req->id; | |
972 | ||
2df46246 MK |
973 | qemu_co_mutex_lock(&s->lock); |
974 | s->co_send = qemu_coroutine_self(); | |
975 | qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, | |
976 | aio_flush_request, NULL, s); | |
128aa589 | 977 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
978 | |
979 | /* send a header */ | |
8c5135f9 PB |
980 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
981 | if (ret < 0) { | |
c3fecea5 | 982 | qemu_co_mutex_unlock(&s->lock); |
6daf194d | 983 | error_report("failed to send a req, %s", strerror(errno)); |
33b1db1c MK |
984 | return -EIO; |
985 | } | |
986 | ||
987 | if (wlen) { | |
8c5135f9 PB |
988 | ret = qemu_co_sendv(s->fd, iov, wlen, aio_req->iov_offset); |
989 | if (ret < 0) { | |
c3fecea5 | 990 | qemu_co_mutex_unlock(&s->lock); |
6daf194d | 991 | error_report("failed to send a data, %s", strerror(errno)); |
33b1db1c MK |
992 | return -EIO; |
993 | } | |
994 | } | |
995 | ||
128aa589 | 996 | socket_set_cork(s->fd, 0); |
2df46246 MK |
997 | qemu_aio_set_fd_handler(s->fd, co_read_response, NULL, |
998 | aio_flush_request, NULL, s); | |
999 | qemu_co_mutex_unlock(&s->lock); | |
33b1db1c MK |
1000 | |
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | static int read_write_object(int fd, char *buf, uint64_t oid, int copies, | |
1005 | unsigned int datalen, uint64_t offset, | |
47622c44 | 1006 | int write, int create, uint8_t cache) |
33b1db1c MK |
1007 | { |
1008 | SheepdogObjReq hdr; | |
1009 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1010 | unsigned int wlen, rlen; | |
1011 | int ret; | |
1012 | ||
1013 | memset(&hdr, 0, sizeof(hdr)); | |
1014 | ||
1015 | if (write) { | |
1016 | wlen = datalen; | |
1017 | rlen = 0; | |
1018 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1019 | if (create) { | |
1020 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1021 | } else { | |
1022 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1023 | } | |
1024 | } else { | |
1025 | wlen = 0; | |
1026 | rlen = datalen; | |
1027 | hdr.opcode = SD_OP_READ_OBJ; | |
1028 | } | |
47622c44 LY |
1029 | |
1030 | if (cache) { | |
1031 | hdr.flags |= SD_FLAG_CMD_CACHE; | |
1032 | } | |
1033 | ||
33b1db1c MK |
1034 | hdr.oid = oid; |
1035 | hdr.data_length = datalen; | |
1036 | hdr.offset = offset; | |
1037 | hdr.copies = copies; | |
1038 | ||
1039 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1040 | if (ret) { | |
6daf194d | 1041 | error_report("failed to send a request to the sheep"); |
33b1db1c MK |
1042 | return -1; |
1043 | } | |
1044 | ||
1045 | switch (rsp->result) { | |
1046 | case SD_RES_SUCCESS: | |
1047 | return 0; | |
1048 | default: | |
6daf194d | 1049 | error_report("%s", sd_strerror(rsp->result)); |
33b1db1c MK |
1050 | return -1; |
1051 | } | |
1052 | } | |
1053 | ||
1054 | static int read_object(int fd, char *buf, uint64_t oid, int copies, | |
47622c44 | 1055 | unsigned int datalen, uint64_t offset, uint8_t cache) |
33b1db1c | 1056 | { |
47622c44 LY |
1057 | return read_write_object(fd, buf, oid, copies, datalen, offset, 0, 0, |
1058 | cache); | |
33b1db1c MK |
1059 | } |
1060 | ||
1061 | static int write_object(int fd, char *buf, uint64_t oid, int copies, | |
47622c44 LY |
1062 | unsigned int datalen, uint64_t offset, int create, |
1063 | uint8_t cache) | |
33b1db1c | 1064 | { |
47622c44 LY |
1065 | return read_write_object(fd, buf, oid, copies, datalen, offset, 1, create, |
1066 | cache); | |
33b1db1c MK |
1067 | } |
1068 | ||
1069 | static int sd_open(BlockDriverState *bs, const char *filename, int flags) | |
1070 | { | |
1071 | int ret, fd; | |
1072 | uint32_t vid = 0; | |
1073 | BDRVSheepdogState *s = bs->opaque; | |
1074 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1075 | uint32_t snapid; | |
1076 | char *buf = NULL; | |
1077 | ||
1078 | strstart(filename, "sheepdog:", (const char **)&filename); | |
1079 | ||
1080 | QLIST_INIT(&s->outstanding_aio_head); | |
1081 | s->fd = -1; | |
1082 | ||
1083 | memset(vdi, 0, sizeof(vdi)); | |
1084 | memset(tag, 0, sizeof(tag)); | |
1085 | if (parse_vdiname(s, filename, vdi, &snapid, tag) < 0) { | |
1086 | goto out; | |
1087 | } | |
1088 | s->fd = get_sheep_fd(s); | |
1089 | if (s->fd < 0) { | |
1090 | goto out; | |
1091 | } | |
1092 | ||
1093 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0); | |
1094 | if (ret) { | |
1095 | goto out; | |
1096 | } | |
1097 | ||
47622c44 LY |
1098 | if (flags & BDRV_O_CACHE_WB) { |
1099 | s->cache_enabled = 1; | |
1100 | s->flush_fd = connect_to_sdog(s->addr, s->port); | |
1101 | if (s->flush_fd < 0) { | |
1102 | error_report("failed to connect"); | |
1103 | goto out; | |
1104 | } | |
1105 | } | |
1106 | ||
33b1db1c MK |
1107 | if (snapid) { |
1108 | dprintf("%" PRIx32 " snapshot inode was open.\n", vid); | |
1109 | s->is_snapshot = 1; | |
1110 | } | |
1111 | ||
1112 | fd = connect_to_sdog(s->addr, s->port); | |
1113 | if (fd < 0) { | |
6daf194d | 1114 | error_report("failed to connect"); |
33b1db1c MK |
1115 | goto out; |
1116 | } | |
1117 | ||
7267c094 | 1118 | buf = g_malloc(SD_INODE_SIZE); |
47622c44 LY |
1119 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, |
1120 | s->cache_enabled); | |
33b1db1c MK |
1121 | |
1122 | closesocket(fd); | |
1123 | ||
1124 | if (ret) { | |
1125 | goto out; | |
1126 | } | |
1127 | ||
1128 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1129 | s->min_dirty_data_idx = UINT32_MAX; | |
1130 | s->max_dirty_data_idx = 0; | |
1131 | ||
1132 | bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE; | |
1133 | strncpy(s->name, vdi, sizeof(s->name)); | |
2df46246 | 1134 | qemu_co_mutex_init(&s->lock); |
7267c094 | 1135 | g_free(buf); |
33b1db1c MK |
1136 | return 0; |
1137 | out: | |
1138 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL, NULL); | |
1139 | if (s->fd >= 0) { | |
1140 | closesocket(s->fd); | |
1141 | } | |
7267c094 | 1142 | g_free(buf); |
33b1db1c MK |
1143 | return -1; |
1144 | } | |
1145 | ||
1146 | static int do_sd_create(char *filename, int64_t vdi_size, | |
1147 | uint32_t base_vid, uint32_t *vdi_id, int snapshot, | |
1148 | const char *addr, const char *port) | |
1149 | { | |
1150 | SheepdogVdiReq hdr; | |
1151 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1152 | int fd, ret; | |
1153 | unsigned int wlen, rlen = 0; | |
1154 | char buf[SD_MAX_VDI_LEN]; | |
1155 | ||
1156 | fd = connect_to_sdog(addr, port); | |
1157 | if (fd < 0) { | |
1158 | return -EIO; | |
1159 | } | |
1160 | ||
1161 | memset(buf, 0, sizeof(buf)); | |
1162 | strncpy(buf, filename, SD_MAX_VDI_LEN); | |
1163 | ||
1164 | memset(&hdr, 0, sizeof(hdr)); | |
1165 | hdr.opcode = SD_OP_NEW_VDI; | |
1166 | hdr.base_vdi_id = base_vid; | |
1167 | ||
1168 | wlen = SD_MAX_VDI_LEN; | |
1169 | ||
1170 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1171 | hdr.snapid = snapshot; | |
1172 | ||
1173 | hdr.data_length = wlen; | |
1174 | hdr.vdi_size = vdi_size; | |
1175 | ||
1176 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1177 | ||
1178 | closesocket(fd); | |
1179 | ||
1180 | if (ret) { | |
1181 | return -EIO; | |
1182 | } | |
1183 | ||
1184 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 1185 | error_report("%s, %s", sd_strerror(rsp->result), filename); |
33b1db1c MK |
1186 | return -EIO; |
1187 | } | |
1188 | ||
1189 | if (vdi_id) { | |
1190 | *vdi_id = rsp->vdi_id; | |
1191 | } | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
a8e0fdd7 MK |
1196 | static int sd_prealloc(const char *filename) |
1197 | { | |
1198 | BlockDriverState *bs = NULL; | |
1199 | uint32_t idx, max_idx; | |
1200 | int64_t vdi_size; | |
7267c094 | 1201 | void *buf = g_malloc0(SD_DATA_OBJ_SIZE); |
a8e0fdd7 MK |
1202 | int ret; |
1203 | ||
1204 | ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); | |
1205 | if (ret < 0) { | |
1206 | goto out; | |
1207 | } | |
1208 | ||
1209 | vdi_size = bdrv_getlength(bs); | |
1210 | if (vdi_size < 0) { | |
1211 | ret = vdi_size; | |
1212 | goto out; | |
1213 | } | |
1214 | max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); | |
1215 | ||
1216 | for (idx = 0; idx < max_idx; idx++) { | |
1217 | /* | |
1218 | * The created image can be a cloned image, so we need to read | |
1219 | * a data from the source image. | |
1220 | */ | |
1221 | ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1222 | if (ret < 0) { | |
1223 | goto out; | |
1224 | } | |
1225 | ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1226 | if (ret < 0) { | |
1227 | goto out; | |
1228 | } | |
1229 | } | |
1230 | out: | |
1231 | if (bs) { | |
1232 | bdrv_delete(bs); | |
1233 | } | |
7267c094 | 1234 | g_free(buf); |
a8e0fdd7 MK |
1235 | |
1236 | return ret; | |
1237 | } | |
1238 | ||
33b1db1c MK |
1239 | static int sd_create(const char *filename, QEMUOptionParameter *options) |
1240 | { | |
1241 | int ret; | |
b4447363 | 1242 | uint32_t vid = 0, base_vid = 0; |
33b1db1c MK |
1243 | int64_t vdi_size = 0; |
1244 | char *backing_file = NULL; | |
b4447363 MK |
1245 | BDRVSheepdogState s; |
1246 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1247 | uint32_t snapid; | |
a8e0fdd7 MK |
1248 | int prealloc = 0; |
1249 | const char *vdiname; | |
33b1db1c | 1250 | |
a8e0fdd7 | 1251 | strstart(filename, "sheepdog:", &vdiname); |
33b1db1c | 1252 | |
b4447363 MK |
1253 | memset(&s, 0, sizeof(s)); |
1254 | memset(vdi, 0, sizeof(vdi)); | |
1255 | memset(tag, 0, sizeof(tag)); | |
a8e0fdd7 | 1256 | if (parse_vdiname(&s, vdiname, vdi, &snapid, tag) < 0) { |
6daf194d | 1257 | error_report("invalid filename"); |
b4447363 MK |
1258 | return -EINVAL; |
1259 | } | |
1260 | ||
33b1db1c MK |
1261 | while (options && options->name) { |
1262 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1263 | vdi_size = options->value.n; | |
1264 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1265 | backing_file = options->value.s; | |
a8e0fdd7 MK |
1266 | } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1267 | if (!options->value.s || !strcmp(options->value.s, "off")) { | |
1268 | prealloc = 0; | |
1269 | } else if (!strcmp(options->value.s, "full")) { | |
1270 | prealloc = 1; | |
1271 | } else { | |
1272 | error_report("Invalid preallocation mode: '%s'", | |
1273 | options->value.s); | |
1274 | return -EINVAL; | |
1275 | } | |
33b1db1c MK |
1276 | } |
1277 | options++; | |
1278 | } | |
1279 | ||
1280 | if (vdi_size > SD_MAX_VDI_SIZE) { | |
6daf194d | 1281 | error_report("too big image size"); |
33b1db1c MK |
1282 | return -EINVAL; |
1283 | } | |
1284 | ||
1285 | if (backing_file) { | |
1286 | BlockDriverState *bs; | |
1287 | BDRVSheepdogState *s; | |
1288 | BlockDriver *drv; | |
1289 | ||
1290 | /* Currently, only Sheepdog backing image is supported. */ | |
1291 | drv = bdrv_find_protocol(backing_file); | |
1292 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { | |
6daf194d | 1293 | error_report("backing_file must be a sheepdog image"); |
33b1db1c MK |
1294 | return -EINVAL; |
1295 | } | |
1296 | ||
1297 | ret = bdrv_file_open(&bs, backing_file, 0); | |
1298 | if (ret < 0) | |
1299 | return -EIO; | |
1300 | ||
1301 | s = bs->opaque; | |
1302 | ||
1303 | if (!is_snapshot(&s->inode)) { | |
6daf194d | 1304 | error_report("cannot clone from a non snapshot vdi"); |
33b1db1c MK |
1305 | bdrv_delete(bs); |
1306 | return -EINVAL; | |
1307 | } | |
1308 | ||
b4447363 | 1309 | base_vid = s->inode.vdi_id; |
33b1db1c MK |
1310 | bdrv_delete(bs); |
1311 | } | |
1312 | ||
a8e0fdd7 MK |
1313 | ret = do_sd_create(vdi, vdi_size, base_vid, &vid, 0, s.addr, s.port); |
1314 | if (!prealloc || ret) { | |
1315 | return ret; | |
1316 | } | |
1317 | ||
1318 | return sd_prealloc(filename); | |
33b1db1c MK |
1319 | } |
1320 | ||
1321 | static void sd_close(BlockDriverState *bs) | |
1322 | { | |
1323 | BDRVSheepdogState *s = bs->opaque; | |
1324 | SheepdogVdiReq hdr; | |
1325 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1326 | unsigned int wlen, rlen = 0; | |
1327 | int fd, ret; | |
1328 | ||
1329 | dprintf("%s\n", s->name); | |
1330 | ||
1331 | fd = connect_to_sdog(s->addr, s->port); | |
1332 | if (fd < 0) { | |
1333 | return; | |
1334 | } | |
1335 | ||
1336 | memset(&hdr, 0, sizeof(hdr)); | |
1337 | ||
1338 | hdr.opcode = SD_OP_RELEASE_VDI; | |
1339 | wlen = strlen(s->name) + 1; | |
1340 | hdr.data_length = wlen; | |
1341 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1342 | ||
1343 | ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); | |
1344 | ||
1345 | closesocket(fd); | |
1346 | ||
1347 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1348 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1349 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1350 | } |
1351 | ||
1352 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL, NULL); | |
1353 | closesocket(s->fd); | |
47622c44 LY |
1354 | if (s->cache_enabled) { |
1355 | closesocket(s->flush_fd); | |
1356 | } | |
7267c094 | 1357 | g_free(s->addr); |
33b1db1c MK |
1358 | } |
1359 | ||
1360 | static int64_t sd_getlength(BlockDriverState *bs) | |
1361 | { | |
1362 | BDRVSheepdogState *s = bs->opaque; | |
1363 | ||
1364 | return s->inode.vdi_size; | |
1365 | } | |
1366 | ||
1367 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1368 | { | |
1369 | BDRVSheepdogState *s = bs->opaque; | |
1370 | int ret, fd; | |
1371 | unsigned int datalen; | |
1372 | ||
1373 | if (offset < s->inode.vdi_size) { | |
6daf194d | 1374 | error_report("shrinking is not supported"); |
33b1db1c MK |
1375 | return -EINVAL; |
1376 | } else if (offset > SD_MAX_VDI_SIZE) { | |
6daf194d | 1377 | error_report("too big image size"); |
33b1db1c MK |
1378 | return -EINVAL; |
1379 | } | |
1380 | ||
1381 | fd = connect_to_sdog(s->addr, s->port); | |
1382 | if (fd < 0) { | |
1383 | return -EIO; | |
1384 | } | |
1385 | ||
1386 | /* we don't need to update entire object */ | |
1387 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1388 | s->inode.vdi_size = offset; | |
1389 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
47622c44 | 1390 | s->inode.nr_copies, datalen, 0, 0, s->cache_enabled); |
33b1db1c MK |
1391 | close(fd); |
1392 | ||
1393 | if (ret < 0) { | |
6daf194d | 1394 | error_report("failed to update an inode."); |
33b1db1c MK |
1395 | return -EIO; |
1396 | } | |
1397 | ||
1398 | return 0; | |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * This function is called after writing data objects. If we need to | |
1403 | * update metadata, this sends a write request to the vdi object. | |
2df46246 | 1404 | * Otherwise, this switches back to sd_co_readv/writev. |
33b1db1c | 1405 | */ |
d8716b41 | 1406 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c MK |
1407 | { |
1408 | int ret; | |
1409 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1410 | struct iovec iov; | |
1411 | AIOReq *aio_req; | |
1412 | uint32_t offset, data_len, mn, mx; | |
1413 | ||
1414 | mn = s->min_dirty_data_idx; | |
1415 | mx = s->max_dirty_data_idx; | |
1416 | if (mn <= mx) { | |
1417 | /* we need to update the vdi object. */ | |
1418 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
1419 | mn * sizeof(s->inode.data_vdi_id[0]); | |
1420 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
1421 | ||
1422 | s->min_dirty_data_idx = UINT32_MAX; | |
1423 | s->max_dirty_data_idx = 0; | |
1424 | ||
1425 | iov.iov_base = &s->inode; | |
1426 | iov.iov_len = sizeof(s->inode); | |
1427 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
1428 | data_len, offset, 0, 0, offset); | |
1429 | ret = add_aio_request(s, aio_req, &iov, 1, 0, AIOCB_WRITE_UDATA); | |
1430 | if (ret) { | |
1431 | free_aio_req(s, aio_req); | |
1432 | acb->ret = -EIO; | |
1433 | goto out; | |
1434 | } | |
1435 | ||
1436 | acb->aio_done_func = sd_finish_aiocb; | |
1437 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1438 | return; | |
1439 | } | |
1440 | out: | |
1441 | sd_finish_aiocb(acb); | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * Create a writable VDI from a snapshot | |
1446 | */ | |
1447 | static int sd_create_branch(BDRVSheepdogState *s) | |
1448 | { | |
1449 | int ret, fd; | |
1450 | uint32_t vid; | |
1451 | char *buf; | |
1452 | ||
1453 | dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); | |
1454 | ||
7267c094 | 1455 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c MK |
1456 | |
1457 | ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1, | |
1458 | s->addr, s->port); | |
1459 | if (ret) { | |
1460 | goto out; | |
1461 | } | |
1462 | ||
1463 | dprintf("%" PRIx32 " is created.\n", vid); | |
1464 | ||
1465 | fd = connect_to_sdog(s->addr, s->port); | |
1466 | if (fd < 0) { | |
6daf194d | 1467 | error_report("failed to connect"); |
33b1db1c MK |
1468 | goto out; |
1469 | } | |
1470 | ||
1471 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, | |
47622c44 | 1472 | SD_INODE_SIZE, 0, s->cache_enabled); |
33b1db1c MK |
1473 | |
1474 | closesocket(fd); | |
1475 | ||
1476 | if (ret < 0) { | |
1477 | goto out; | |
1478 | } | |
1479 | ||
1480 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1481 | ||
1482 | s->is_snapshot = 0; | |
1483 | ret = 0; | |
1484 | dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id); | |
1485 | ||
1486 | out: | |
7267c094 | 1487 | g_free(buf); |
33b1db1c MK |
1488 | |
1489 | return ret; | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Send I/O requests to the server. | |
1494 | * | |
1495 | * This function sends requests to the server, links the requests to | |
1496 | * the outstanding_list in BDRVSheepdogState, and exits without | |
1497 | * waiting the response. The responses are received in the | |
1498 | * `aio_read_response' function which is called from the main loop as | |
1499 | * a fd handler. | |
2df46246 MK |
1500 | * |
1501 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
1502 | * request and -errno in error cases. | |
33b1db1c | 1503 | */ |
d8716b41 | 1504 | static int coroutine_fn sd_co_rw_vector(void *p) |
33b1db1c MK |
1505 | { |
1506 | SheepdogAIOCB *acb = p; | |
1507 | int ret = 0; | |
1508 | unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE; | |
1509 | unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE; | |
1510 | uint64_t oid; | |
1511 | uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE; | |
1512 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1513 | SheepdogInode *inode = &s->inode; | |
1514 | AIOReq *aio_req; | |
1515 | ||
33b1db1c MK |
1516 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
1517 | /* | |
1518 | * In the case we open the snapshot VDI, Sheepdog creates the | |
1519 | * writable VDI when we do a write operation first. | |
1520 | */ | |
1521 | ret = sd_create_branch(s); | |
1522 | if (ret) { | |
1523 | acb->ret = -EIO; | |
1524 | goto out; | |
1525 | } | |
1526 | } | |
1527 | ||
1528 | while (done != total) { | |
1529 | uint8_t flags = 0; | |
1530 | uint64_t old_oid = 0; | |
1531 | int create = 0; | |
1532 | ||
1533 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
1534 | ||
1535 | len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); | |
1536 | ||
1537 | if (!inode->data_vdi_id[idx]) { | |
1538 | if (acb->aiocb_type == AIOCB_READ_UDATA) { | |
1539 | goto done; | |
1540 | } | |
1541 | ||
1542 | create = 1; | |
1543 | } else if (acb->aiocb_type == AIOCB_WRITE_UDATA | |
ebabb67a | 1544 | && !is_data_obj_writable(inode, idx)) { |
33b1db1c MK |
1545 | /* Copy-On-Write */ |
1546 | create = 1; | |
1547 | old_oid = oid; | |
1548 | flags = SD_FLAG_CMD_COW; | |
1549 | } | |
1550 | ||
1551 | if (create) { | |
1552 | dprintf("update ino (%" PRIu32") %" PRIu64 " %" PRIu64 | |
1553 | " %" PRIu64 "\n", inode->vdi_id, oid, | |
1554 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); | |
1555 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
1556 | dprintf("new oid %lx\n", oid); | |
1557 | } | |
1558 | ||
1559 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); | |
1560 | ||
1561 | if (create) { | |
1562 | AIOReq *areq; | |
1563 | QLIST_FOREACH(areq, &s->outstanding_aio_head, | |
1564 | outstanding_aio_siblings) { | |
1565 | if (areq == aio_req) { | |
1566 | continue; | |
1567 | } | |
1568 | if (areq->oid == oid) { | |
1569 | /* | |
1570 | * Sheepdog cannot handle simultaneous create | |
1571 | * requests to the same object. So we cannot send | |
1572 | * the request until the previous request | |
1573 | * finishes. | |
1574 | */ | |
1575 | aio_req->flags = 0; | |
1576 | aio_req->base_oid = 0; | |
1577 | goto done; | |
1578 | } | |
1579 | } | |
1580 | } | |
1581 | ||
1582 | ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, | |
1583 | create, acb->aiocb_type); | |
1584 | if (ret < 0) { | |
6daf194d | 1585 | error_report("add_aio_request is failed"); |
33b1db1c MK |
1586 | free_aio_req(s, aio_req); |
1587 | acb->ret = -EIO; | |
1588 | goto out; | |
1589 | } | |
1590 | done: | |
1591 | offset = 0; | |
1592 | idx++; | |
1593 | done += len; | |
1594 | } | |
1595 | out: | |
1596 | if (QLIST_EMPTY(&acb->aioreq_head)) { | |
2df46246 | 1597 | return acb->ret; |
33b1db1c | 1598 | } |
2df46246 | 1599 | return 1; |
33b1db1c MK |
1600 | } |
1601 | ||
a968168c | 1602 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1603 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1604 | { |
1605 | SheepdogAIOCB *acb; | |
2df46246 | 1606 | int ret; |
33b1db1c MK |
1607 | |
1608 | if (bs->growable && sector_num + nb_sectors > bs->total_sectors) { | |
1609 | /* TODO: shouldn't block here */ | |
1610 | if (sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE) < 0) { | |
2df46246 | 1611 | return -EIO; |
33b1db1c MK |
1612 | } |
1613 | bs->total_sectors = sector_num + nb_sectors; | |
1614 | } | |
1615 | ||
2df46246 | 1616 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL); |
33b1db1c MK |
1617 | acb->aio_done_func = sd_write_done; |
1618 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1619 | ||
2df46246 MK |
1620 | ret = sd_co_rw_vector(acb); |
1621 | if (ret <= 0) { | |
1622 | qemu_aio_release(acb); | |
1623 | return ret; | |
1624 | } | |
1625 | ||
1626 | qemu_coroutine_yield(); | |
1627 | ||
1628 | return acb->ret; | |
33b1db1c MK |
1629 | } |
1630 | ||
a968168c | 1631 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1632 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1633 | { |
1634 | SheepdogAIOCB *acb; | |
2df46246 | 1635 | int i, ret; |
33b1db1c | 1636 | |
2df46246 | 1637 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL); |
33b1db1c MK |
1638 | acb->aiocb_type = AIOCB_READ_UDATA; |
1639 | acb->aio_done_func = sd_finish_aiocb; | |
1640 | ||
1641 | /* | |
1642 | * TODO: we can do better; we don't need to initialize | |
1643 | * blindly. | |
1644 | */ | |
1645 | for (i = 0; i < qiov->niov; i++) { | |
1646 | memset(qiov->iov[i].iov_base, 0, qiov->iov[i].iov_len); | |
1647 | } | |
1648 | ||
2df46246 MK |
1649 | ret = sd_co_rw_vector(acb); |
1650 | if (ret <= 0) { | |
1651 | qemu_aio_release(acb); | |
1652 | return ret; | |
1653 | } | |
1654 | ||
1655 | qemu_coroutine_yield(); | |
1656 | ||
1657 | return acb->ret; | |
33b1db1c MK |
1658 | } |
1659 | ||
47622c44 LY |
1660 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
1661 | { | |
1662 | BDRVSheepdogState *s = bs->opaque; | |
1663 | SheepdogObjReq hdr = { 0 }; | |
1664 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1665 | SheepdogInode *inode = &s->inode; | |
1666 | int ret; | |
1667 | unsigned int wlen = 0, rlen = 0; | |
1668 | ||
1669 | if (!s->cache_enabled) { | |
1670 | return 0; | |
1671 | } | |
1672 | ||
1673 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1674 | hdr.oid = vid_to_vdi_oid(inode->vdi_id); | |
1675 | ||
1676 | ret = do_co_req(s->flush_fd, (SheepdogReq *)&hdr, NULL, &wlen, &rlen); | |
1677 | if (ret) { | |
1678 | error_report("failed to send a request to the sheep"); | |
1679 | return ret; | |
1680 | } | |
1681 | ||
1682 | if (rsp->result != SD_RES_SUCCESS) { | |
1683 | error_report("%s", sd_strerror(rsp->result)); | |
1684 | return -EIO; | |
1685 | } | |
1686 | ||
1687 | return 0; | |
1688 | } | |
1689 | ||
33b1db1c MK |
1690 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
1691 | { | |
1692 | BDRVSheepdogState *s = bs->opaque; | |
1693 | int ret, fd; | |
1694 | uint32_t new_vid; | |
1695 | SheepdogInode *inode; | |
1696 | unsigned int datalen; | |
1697 | ||
1698 | dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %d " | |
1699 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, | |
1700 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
1701 | ||
1702 | if (s->is_snapshot) { | |
1703 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 1704 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
1705 | |
1706 | return -EINVAL; | |
1707 | } | |
1708 | ||
1709 | dprintf("%s %s\n", sn_info->name, sn_info->id_str); | |
1710 | ||
1711 | s->inode.vm_state_size = sn_info->vm_state_size; | |
1712 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
1713 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); | |
1714 | /* we don't need to update entire object */ | |
1715 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1716 | ||
1717 | /* refresh inode. */ | |
1718 | fd = connect_to_sdog(s->addr, s->port); | |
1719 | if (fd < 0) { | |
1720 | ret = -EIO; | |
1721 | goto cleanup; | |
1722 | } | |
1723 | ||
1724 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
47622c44 | 1725 | s->inode.nr_copies, datalen, 0, 0, s->cache_enabled); |
33b1db1c | 1726 | if (ret < 0) { |
6daf194d | 1727 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
1728 | ret = -EIO; |
1729 | goto cleanup; | |
1730 | } | |
1731 | ||
1732 | ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, 1, | |
1733 | s->addr, s->port); | |
1734 | if (ret < 0) { | |
6daf194d | 1735 | error_report("failed to create inode for snapshot. %s", |
33b1db1c MK |
1736 | strerror(errno)); |
1737 | ret = -EIO; | |
1738 | goto cleanup; | |
1739 | } | |
1740 | ||
7267c094 | 1741 | inode = (SheepdogInode *)g_malloc(datalen); |
33b1db1c MK |
1742 | |
1743 | ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid), | |
47622c44 | 1744 | s->inode.nr_copies, datalen, 0, s->cache_enabled); |
33b1db1c MK |
1745 | |
1746 | if (ret < 0) { | |
6daf194d | 1747 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
1748 | ret = -EIO; |
1749 | goto cleanup; | |
1750 | } | |
1751 | ||
1752 | memcpy(&s->inode, inode, datalen); | |
1753 | dprintf("s->inode: name %s snap_id %x oid %x\n", | |
1754 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); | |
1755 | ||
1756 | cleanup: | |
1757 | closesocket(fd); | |
1758 | return ret; | |
1759 | } | |
1760 | ||
1761 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) | |
1762 | { | |
1763 | BDRVSheepdogState *s = bs->opaque; | |
1764 | BDRVSheepdogState *old_s; | |
1765 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1766 | char *buf = NULL; | |
1767 | uint32_t vid; | |
1768 | uint32_t snapid = 0; | |
1769 | int ret = -ENOENT, fd; | |
1770 | ||
7267c094 | 1771 | old_s = g_malloc(sizeof(BDRVSheepdogState)); |
33b1db1c MK |
1772 | |
1773 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
1774 | ||
1775 | memset(vdi, 0, sizeof(vdi)); | |
1776 | strncpy(vdi, s->name, sizeof(vdi)); | |
1777 | ||
1778 | memset(tag, 0, sizeof(tag)); | |
1779 | snapid = strtoul(snapshot_id, NULL, 10); | |
1780 | if (!snapid) { | |
1781 | strncpy(tag, s->name, sizeof(tag)); | |
1782 | } | |
1783 | ||
1784 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1); | |
1785 | if (ret) { | |
6daf194d | 1786 | error_report("Failed to find_vdi_name"); |
33b1db1c MK |
1787 | ret = -ENOENT; |
1788 | goto out; | |
1789 | } | |
1790 | ||
1791 | fd = connect_to_sdog(s->addr, s->port); | |
1792 | if (fd < 0) { | |
6daf194d | 1793 | error_report("failed to connect"); |
33b1db1c MK |
1794 | goto out; |
1795 | } | |
1796 | ||
7267c094 | 1797 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 1798 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
47622c44 | 1799 | SD_INODE_SIZE, 0, s->cache_enabled); |
33b1db1c MK |
1800 | |
1801 | closesocket(fd); | |
1802 | ||
1803 | if (ret) { | |
1804 | ret = -ENOENT; | |
1805 | goto out; | |
1806 | } | |
1807 | ||
1808 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1809 | ||
1810 | if (!s->inode.vm_state_size) { | |
6daf194d | 1811 | error_report("Invalid snapshot"); |
33b1db1c MK |
1812 | ret = -ENOENT; |
1813 | goto out; | |
1814 | } | |
1815 | ||
1816 | s->is_snapshot = 1; | |
1817 | ||
7267c094 AL |
1818 | g_free(buf); |
1819 | g_free(old_s); | |
33b1db1c MK |
1820 | |
1821 | return 0; | |
1822 | out: | |
1823 | /* recover bdrv_sd_state */ | |
1824 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 AL |
1825 | g_free(buf); |
1826 | g_free(old_s); | |
33b1db1c | 1827 | |
6daf194d | 1828 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
1829 | |
1830 | return ret; | |
1831 | } | |
1832 | ||
1833 | static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) | |
1834 | { | |
1835 | /* FIXME: Delete specified snapshot id. */ | |
1836 | return 0; | |
1837 | } | |
1838 | ||
33b1db1c MK |
1839 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
1840 | { | |
1841 | BDRVSheepdogState *s = bs->opaque; | |
1842 | SheepdogReq req; | |
1843 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
1844 | QEMUSnapshotInfo *sn_tab = NULL; | |
1845 | unsigned wlen, rlen; | |
1846 | int found = 0; | |
1847 | static SheepdogInode inode; | |
1848 | unsigned long *vdi_inuse; | |
1849 | unsigned int start_nr; | |
1850 | uint64_t hval; | |
1851 | uint32_t vid; | |
1852 | ||
7267c094 | 1853 | vdi_inuse = g_malloc(max); |
33b1db1c MK |
1854 | |
1855 | fd = connect_to_sdog(s->addr, s->port); | |
1856 | if (fd < 0) { | |
1857 | goto out; | |
1858 | } | |
1859 | ||
1860 | rlen = max; | |
1861 | wlen = 0; | |
1862 | ||
1863 | memset(&req, 0, sizeof(req)); | |
1864 | ||
1865 | req.opcode = SD_OP_READ_VDIS; | |
1866 | req.data_length = max; | |
1867 | ||
1868 | ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); | |
1869 | ||
1870 | closesocket(fd); | |
1871 | if (ret) { | |
1872 | goto out; | |
1873 | } | |
1874 | ||
7267c094 | 1875 | sn_tab = g_malloc0(nr * sizeof(*sn_tab)); |
33b1db1c MK |
1876 | |
1877 | /* calculate a vdi id with hash function */ | |
1878 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
1879 | start_nr = hval & (SD_NR_VDIS - 1); | |
1880 | ||
1881 | fd = connect_to_sdog(s->addr, s->port); | |
1882 | if (fd < 0) { | |
6daf194d | 1883 | error_report("failed to connect"); |
33b1db1c MK |
1884 | goto out; |
1885 | } | |
1886 | ||
1887 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
1888 | if (!test_bit(vid, vdi_inuse)) { | |
1889 | break; | |
1890 | } | |
1891 | ||
1892 | /* we don't need to read entire object */ | |
1893 | ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid), | |
47622c44 LY |
1894 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
1895 | s->cache_enabled); | |
33b1db1c MK |
1896 | |
1897 | if (ret) { | |
1898 | continue; | |
1899 | } | |
1900 | ||
1901 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
1902 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
1903 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
1904 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
1905 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
1906 | ||
1907 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", | |
1908 | inode.snap_id); | |
1909 | strncpy(sn_tab[found].name, inode.tag, | |
1910 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag))); | |
1911 | found++; | |
1912 | } | |
1913 | } | |
1914 | ||
1915 | closesocket(fd); | |
1916 | out: | |
1917 | *psn_tab = sn_tab; | |
1918 | ||
7267c094 | 1919 | g_free(vdi_inuse); |
33b1db1c MK |
1920 | |
1921 | return found; | |
1922 | } | |
1923 | ||
1924 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
1925 | int64_t pos, int size, int load) | |
1926 | { | |
1927 | int fd, create; | |
1928 | int ret = 0; | |
1929 | unsigned int data_len; | |
1930 | uint64_t vmstate_oid; | |
1931 | uint32_t vdi_index; | |
1932 | uint64_t offset; | |
1933 | ||
1934 | fd = connect_to_sdog(s->addr, s->port); | |
1935 | if (fd < 0) { | |
1936 | ret = -EIO; | |
1937 | goto cleanup; | |
1938 | } | |
1939 | ||
1940 | while (size) { | |
1941 | vdi_index = pos / SD_DATA_OBJ_SIZE; | |
1942 | offset = pos % SD_DATA_OBJ_SIZE; | |
1943 | ||
1944 | data_len = MIN(size, SD_DATA_OBJ_SIZE); | |
1945 | ||
1946 | vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index); | |
1947 | ||
1948 | create = (offset == 0); | |
1949 | if (load) { | |
1950 | ret = read_object(fd, (char *)data, vmstate_oid, | |
47622c44 LY |
1951 | s->inode.nr_copies, data_len, offset, |
1952 | s->cache_enabled); | |
33b1db1c MK |
1953 | } else { |
1954 | ret = write_object(fd, (char *)data, vmstate_oid, | |
47622c44 LY |
1955 | s->inode.nr_copies, data_len, offset, create, |
1956 | s->cache_enabled); | |
33b1db1c MK |
1957 | } |
1958 | ||
1959 | if (ret < 0) { | |
6daf194d | 1960 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
1961 | ret = -EIO; |
1962 | goto cleanup; | |
1963 | } | |
1964 | ||
1965 | pos += data_len; | |
1966 | size -= data_len; | |
1967 | ret += data_len; | |
1968 | } | |
1969 | cleanup: | |
1970 | closesocket(fd); | |
1971 | return ret; | |
1972 | } | |
1973 | ||
1974 | static int sd_save_vmstate(BlockDriverState *bs, const uint8_t *data, | |
1975 | int64_t pos, int size) | |
1976 | { | |
1977 | BDRVSheepdogState *s = bs->opaque; | |
1978 | ||
1979 | return do_load_save_vmstate(s, (uint8_t *)data, pos, size, 0); | |
1980 | } | |
1981 | ||
1982 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
1983 | int64_t pos, int size) | |
1984 | { | |
1985 | BDRVSheepdogState *s = bs->opaque; | |
1986 | ||
1987 | return do_load_save_vmstate(s, data, pos, size, 1); | |
1988 | } | |
1989 | ||
1990 | ||
1991 | static QEMUOptionParameter sd_create_options[] = { | |
1992 | { | |
1993 | .name = BLOCK_OPT_SIZE, | |
1994 | .type = OPT_SIZE, | |
1995 | .help = "Virtual disk size" | |
1996 | }, | |
1997 | { | |
1998 | .name = BLOCK_OPT_BACKING_FILE, | |
1999 | .type = OPT_STRING, | |
2000 | .help = "File name of a base image" | |
2001 | }, | |
a8e0fdd7 MK |
2002 | { |
2003 | .name = BLOCK_OPT_PREALLOC, | |
2004 | .type = OPT_STRING, | |
2005 | .help = "Preallocation mode (allowed values: off, full)" | |
2006 | }, | |
33b1db1c MK |
2007 | { NULL } |
2008 | }; | |
2009 | ||
2010 | BlockDriver bdrv_sheepdog = { | |
2011 | .format_name = "sheepdog", | |
2012 | .protocol_name = "sheepdog", | |
2013 | .instance_size = sizeof(BDRVSheepdogState), | |
2014 | .bdrv_file_open = sd_open, | |
2015 | .bdrv_close = sd_close, | |
2016 | .bdrv_create = sd_create, | |
2017 | .bdrv_getlength = sd_getlength, | |
2018 | .bdrv_truncate = sd_truncate, | |
2019 | ||
2df46246 MK |
2020 | .bdrv_co_readv = sd_co_readv, |
2021 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2022 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
33b1db1c MK |
2023 | |
2024 | .bdrv_snapshot_create = sd_snapshot_create, | |
2025 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2026 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2027 | .bdrv_snapshot_list = sd_snapshot_list, | |
2028 | ||
2029 | .bdrv_save_vmstate = sd_save_vmstate, | |
2030 | .bdrv_load_vmstate = sd_load_vmstate, | |
2031 | ||
2032 | .create_options = sd_create_options, | |
2033 | }; | |
2034 | ||
2035 | static void bdrv_sheepdog_init(void) | |
2036 | { | |
2037 | bdrv_register(&bdrv_sheepdog); | |
2038 | } | |
2039 | block_init(bdrv_sheepdog_init); |