]>
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" | |
5d6768e3 | 16 | #include "qemu/uri.h" |
1de7afc9 PB |
17 | #include "qemu/error-report.h" |
18 | #include "qemu/sockets.h" | |
737e150e | 19 | #include "block/block_int.h" |
1de7afc9 | 20 | #include "qemu/bitops.h" |
33b1db1c MK |
21 | |
22 | #define SD_PROTO_VER 0x01 | |
23 | ||
24 | #define SD_DEFAULT_ADDR "localhost" | |
25af257d | 25 | #define SD_DEFAULT_PORT 7000 |
33b1db1c MK |
26 | |
27 | #define SD_OP_CREATE_AND_WRITE_OBJ 0x01 | |
28 | #define SD_OP_READ_OBJ 0x02 | |
29 | #define SD_OP_WRITE_OBJ 0x03 | |
30 | ||
31 | #define SD_OP_NEW_VDI 0x11 | |
32 | #define SD_OP_LOCK_VDI 0x12 | |
33 | #define SD_OP_RELEASE_VDI 0x13 | |
34 | #define SD_OP_GET_VDI_INFO 0x14 | |
35 | #define SD_OP_READ_VDIS 0x15 | |
47622c44 | 36 | #define SD_OP_FLUSH_VDI 0x16 |
33b1db1c MK |
37 | |
38 | #define SD_FLAG_CMD_WRITE 0x01 | |
39 | #define SD_FLAG_CMD_COW 0x02 | |
0e7106d8 LY |
40 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
41 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
33b1db1c MK |
42 | |
43 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
44 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
45 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
46 | #define SD_RES_EIO 0x03 /* I/O error */ | |
47 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
48 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
49 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
50 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
51 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
52 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
53 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
54 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
55 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
56 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
57 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
58 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
59 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
60 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
61 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
62 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
63 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
64 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
65 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
66 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
67 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
fca23f0a | 68 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ |
33b1db1c MK |
69 | |
70 | /* | |
71 | * Object ID rules | |
72 | * | |
73 | * 0 - 19 (20 bits): data object space | |
74 | * 20 - 31 (12 bits): reserved data object space | |
75 | * 32 - 55 (24 bits): vdi object space | |
76 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 77 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
78 | */ |
79 | ||
80 | #define VDI_SPACE_SHIFT 32 | |
81 | #define VDI_BIT (UINT64_C(1) << 63) | |
82 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
83 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
84 | #define MAX_CHILDREN 1024 | |
85 | #define SD_MAX_VDI_LEN 256 | |
86 | #define SD_MAX_VDI_TAG_LEN 256 | |
87 | #define SD_NR_VDIS (1U << 24) | |
88 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
89 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
90 | #define SECTOR_SIZE 512 | |
91 | ||
92 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
93 | #define CURRENT_VDI_ID 0 | |
94 | ||
95 | typedef struct SheepdogReq { | |
96 | uint8_t proto_ver; | |
97 | uint8_t opcode; | |
98 | uint16_t flags; | |
99 | uint32_t epoch; | |
100 | uint32_t id; | |
101 | uint32_t data_length; | |
102 | uint32_t opcode_specific[8]; | |
103 | } SheepdogReq; | |
104 | ||
105 | typedef struct SheepdogRsp { | |
106 | uint8_t proto_ver; | |
107 | uint8_t opcode; | |
108 | uint16_t flags; | |
109 | uint32_t epoch; | |
110 | uint32_t id; | |
111 | uint32_t data_length; | |
112 | uint32_t result; | |
113 | uint32_t opcode_specific[7]; | |
114 | } SheepdogRsp; | |
115 | ||
116 | typedef struct SheepdogObjReq { | |
117 | uint8_t proto_ver; | |
118 | uint8_t opcode; | |
119 | uint16_t flags; | |
120 | uint32_t epoch; | |
121 | uint32_t id; | |
122 | uint32_t data_length; | |
123 | uint64_t oid; | |
124 | uint64_t cow_oid; | |
125 | uint32_t copies; | |
126 | uint32_t rsvd; | |
127 | uint64_t offset; | |
128 | } SheepdogObjReq; | |
129 | ||
130 | typedef struct SheepdogObjRsp { | |
131 | uint8_t proto_ver; | |
132 | uint8_t opcode; | |
133 | uint16_t flags; | |
134 | uint32_t epoch; | |
135 | uint32_t id; | |
136 | uint32_t data_length; | |
137 | uint32_t result; | |
138 | uint32_t copies; | |
139 | uint32_t pad[6]; | |
140 | } SheepdogObjRsp; | |
141 | ||
142 | typedef struct SheepdogVdiReq { | |
143 | uint8_t proto_ver; | |
144 | uint8_t opcode; | |
145 | uint16_t flags; | |
146 | uint32_t epoch; | |
147 | uint32_t id; | |
148 | uint32_t data_length; | |
149 | uint64_t vdi_size; | |
6f74c260 | 150 | uint32_t vdi_id; |
33b1db1c MK |
151 | uint32_t copies; |
152 | uint32_t snapid; | |
153 | uint32_t pad[3]; | |
154 | } SheepdogVdiReq; | |
155 | ||
156 | typedef struct SheepdogVdiRsp { | |
157 | uint8_t proto_ver; | |
158 | uint8_t opcode; | |
159 | uint16_t flags; | |
160 | uint32_t epoch; | |
161 | uint32_t id; | |
162 | uint32_t data_length; | |
163 | uint32_t result; | |
164 | uint32_t rsvd; | |
165 | uint32_t vdi_id; | |
166 | uint32_t pad[5]; | |
167 | } SheepdogVdiRsp; | |
168 | ||
169 | typedef struct SheepdogInode { | |
170 | char name[SD_MAX_VDI_LEN]; | |
171 | char tag[SD_MAX_VDI_TAG_LEN]; | |
172 | uint64_t ctime; | |
173 | uint64_t snap_ctime; | |
174 | uint64_t vm_clock_nsec; | |
175 | uint64_t vdi_size; | |
176 | uint64_t vm_state_size; | |
177 | uint16_t copy_policy; | |
178 | uint8_t nr_copies; | |
179 | uint8_t block_size_shift; | |
180 | uint32_t snap_id; | |
181 | uint32_t vdi_id; | |
182 | uint32_t parent_vdi_id; | |
183 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
184 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
185 | } SheepdogInode; | |
186 | ||
187 | /* | |
188 | * 64 bit FNV-1a non-zero initial basis | |
189 | */ | |
190 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
191 | ||
192 | /* | |
193 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
194 | */ | |
195 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
196 | { | |
197 | unsigned char *bp = buf; | |
198 | unsigned char *be = bp + len; | |
199 | while (bp < be) { | |
200 | hval ^= (uint64_t) *bp++; | |
201 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
202 | (hval << 7) + (hval << 8) + (hval << 40); | |
203 | } | |
204 | return hval; | |
205 | } | |
206 | ||
2f536801 | 207 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
208 | { |
209 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
210 | } | |
211 | ||
2f536801 | 212 | static inline bool is_data_obj(uint64_t oid) |
33b1db1c MK |
213 | { |
214 | return !(VDI_BIT & oid); | |
215 | } | |
216 | ||
217 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
218 | { | |
219 | return oid & (MAX_DATA_OBJS - 1); | |
220 | } | |
221 | ||
222 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) | |
223 | { | |
224 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
225 | } | |
226 | ||
227 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
228 | { | |
229 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
230 | } | |
231 | ||
232 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
233 | { | |
234 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
235 | } | |
236 | ||
2f536801 | 237 | static inline bool is_snapshot(struct SheepdogInode *inode) |
33b1db1c MK |
238 | { |
239 | return !!inode->snap_ctime; | |
240 | } | |
241 | ||
242 | #undef dprintf | |
243 | #ifdef DEBUG_SDOG | |
244 | #define dprintf(fmt, args...) \ | |
245 | do { \ | |
246 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
247 | } while (0) | |
248 | #else | |
249 | #define dprintf(fmt, args...) | |
250 | #endif | |
251 | ||
252 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
253 | ||
254 | typedef struct AIOReq { | |
255 | SheepdogAIOCB *aiocb; | |
256 | unsigned int iov_offset; | |
257 | ||
258 | uint64_t oid; | |
259 | uint64_t base_oid; | |
260 | uint64_t offset; | |
261 | unsigned int data_len; | |
262 | uint8_t flags; | |
263 | uint32_t id; | |
264 | ||
c292ee6a | 265 | QLIST_ENTRY(AIOReq) aio_siblings; |
33b1db1c MK |
266 | } AIOReq; |
267 | ||
268 | enum AIOCBState { | |
269 | AIOCB_WRITE_UDATA, | |
270 | AIOCB_READ_UDATA, | |
47783072 | 271 | AIOCB_FLUSH_CACHE, |
33b1db1c MK |
272 | }; |
273 | ||
274 | struct SheepdogAIOCB { | |
275 | BlockDriverAIOCB common; | |
276 | ||
277 | QEMUIOVector *qiov; | |
278 | ||
279 | int64_t sector_num; | |
280 | int nb_sectors; | |
281 | ||
282 | int ret; | |
283 | enum AIOCBState aiocb_type; | |
284 | ||
2df46246 | 285 | Coroutine *coroutine; |
33b1db1c MK |
286 | void (*aio_done_func)(SheepdogAIOCB *); |
287 | ||
2f536801 | 288 | bool canceled; |
1d732d7d | 289 | int nr_pending; |
33b1db1c MK |
290 | }; |
291 | ||
292 | typedef struct BDRVSheepdogState { | |
293 | SheepdogInode inode; | |
294 | ||
295 | uint32_t min_dirty_data_idx; | |
296 | uint32_t max_dirty_data_idx; | |
297 | ||
298 | char name[SD_MAX_VDI_LEN]; | |
2f536801 | 299 | bool is_snapshot; |
0e7106d8 | 300 | uint32_t cache_flags; |
33b1db1c | 301 | |
25af257d | 302 | char *host_spec; |
1b8bbb46 | 303 | bool is_unix; |
33b1db1c MK |
304 | int fd; |
305 | ||
2df46246 MK |
306 | CoMutex lock; |
307 | Coroutine *co_send; | |
308 | Coroutine *co_recv; | |
309 | ||
33b1db1c | 310 | uint32_t aioreq_seq_num; |
c292ee6a MK |
311 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
312 | QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head; | |
33b1db1c MK |
313 | } BDRVSheepdogState; |
314 | ||
315 | static const char * sd_strerror(int err) | |
316 | { | |
317 | int i; | |
318 | ||
319 | static const struct { | |
320 | int err; | |
321 | const char *desc; | |
322 | } errors[] = { | |
323 | {SD_RES_SUCCESS, "Success"}, | |
324 | {SD_RES_UNKNOWN, "Unknown error"}, | |
325 | {SD_RES_NO_OBJ, "No object found"}, | |
326 | {SD_RES_EIO, "I/O error"}, | |
327 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
328 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
329 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
330 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
331 | {SD_RES_NO_VDI, "No vdi found"}, | |
332 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
333 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
334 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
335 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
336 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
337 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
338 | {SD_RES_STARTUP, "The system is still booting"}, | |
339 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
340 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
341 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
342 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
343 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
344 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
345 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
346 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
347 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
fca23f0a | 348 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, |
33b1db1c MK |
349 | }; |
350 | ||
351 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
352 | if (errors[i].err == err) { | |
353 | return errors[i].desc; | |
354 | } | |
355 | } | |
356 | ||
357 | return "Invalid error code"; | |
358 | } | |
359 | ||
360 | /* | |
361 | * Sheepdog I/O handling: | |
362 | * | |
2df46246 | 363 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
c292ee6a | 364 | * link the requests to the inflight_list in the |
2df46246 MK |
365 | * BDRVSheepdogState. The function exits without waiting for |
366 | * receiving the response. | |
33b1db1c | 367 | * |
2df46246 | 368 | * 2. We receive the response in aio_read_response, the fd handler to |
33b1db1c MK |
369 | * the sheepdog connection. If metadata update is needed, we send |
370 | * the write request to the vdi object in sd_write_done, the write | |
2df46246 MK |
371 | * completion function. We switch back to sd_co_readv/writev after |
372 | * all the requests belonging to the AIOCB are finished. | |
33b1db1c MK |
373 | */ |
374 | ||
375 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
376 | uint64_t oid, unsigned int data_len, | |
377 | uint64_t offset, uint8_t flags, | |
378 | uint64_t base_oid, unsigned int iov_offset) | |
379 | { | |
380 | AIOReq *aio_req; | |
381 | ||
7267c094 | 382 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
383 | aio_req->aiocb = acb; |
384 | aio_req->iov_offset = iov_offset; | |
385 | aio_req->oid = oid; | |
386 | aio_req->base_oid = base_oid; | |
387 | aio_req->offset = offset; | |
388 | aio_req->data_len = data_len; | |
389 | aio_req->flags = flags; | |
390 | aio_req->id = s->aioreq_seq_num++; | |
391 | ||
1d732d7d | 392 | acb->nr_pending++; |
33b1db1c MK |
393 | return aio_req; |
394 | } | |
395 | ||
1d732d7d | 396 | static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
33b1db1c MK |
397 | { |
398 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1d732d7d | 399 | |
c292ee6a | 400 | QLIST_REMOVE(aio_req, aio_siblings); |
7267c094 | 401 | g_free(aio_req); |
33b1db1c | 402 | |
1d732d7d | 403 | acb->nr_pending--; |
33b1db1c MK |
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); | |
2f536801 | 424 | acb->canceled = true; |
33b1db1c MK |
425 | } |
426 | ||
d7331bed | 427 | static const AIOCBInfo sd_aiocb_info = { |
33b1db1c MK |
428 | .aiocb_size = sizeof(SheepdogAIOCB), |
429 | .cancel = sd_aio_cancel, | |
430 | }; | |
431 | ||
432 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
f700f8e3 | 433 | int64_t sector_num, int nb_sectors) |
33b1db1c MK |
434 | { |
435 | SheepdogAIOCB *acb; | |
436 | ||
f700f8e3 | 437 | acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); |
33b1db1c MK |
438 | |
439 | acb->qiov = qiov; | |
440 | ||
441 | acb->sector_num = sector_num; | |
442 | acb->nb_sectors = nb_sectors; | |
443 | ||
444 | acb->aio_done_func = NULL; | |
2f536801 | 445 | acb->canceled = false; |
2df46246 | 446 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c | 447 | acb->ret = 0; |
1d732d7d | 448 | acb->nr_pending = 0; |
33b1db1c MK |
449 | return acb; |
450 | } | |
451 | ||
25af257d | 452 | static int connect_to_sdog(BDRVSheepdogState *s) |
33b1db1c | 453 | { |
25af257d MK |
454 | int fd; |
455 | Error *err = NULL; | |
33b1db1c | 456 | |
1b8bbb46 MK |
457 | if (s->is_unix) { |
458 | fd = unix_connect(s->host_spec, &err); | |
459 | } else { | |
460 | fd = inet_connect(s->host_spec, &err); | |
461 | ||
462 | if (err == NULL) { | |
463 | int ret = socket_set_nodelay(fd); | |
464 | if (ret < 0) { | |
465 | error_report("%s", strerror(errno)); | |
466 | } | |
467 | } | |
468 | } | |
33b1db1c | 469 | |
25af257d MK |
470 | if (err != NULL) { |
471 | qerror_report_err(err); | |
472 | error_free(err); | |
0d6db300 | 473 | } else { |
f9e8cacc | 474 | qemu_set_nonblock(fd); |
33b1db1c MK |
475 | } |
476 | ||
33b1db1c MK |
477 | return fd; |
478 | } | |
479 | ||
e0d93a89 MK |
480 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
481 | unsigned int *wlen) | |
47622c44 LY |
482 | { |
483 | int ret; | |
484 | ||
485 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
486 | if (ret < sizeof(*hdr)) { | |
487 | error_report("failed to send a req, %s", strerror(errno)); | |
eb092180 | 488 | return ret; |
47622c44 LY |
489 | } |
490 | ||
491 | ret = qemu_co_send(sockfd, data, *wlen); | |
492 | if (ret < *wlen) { | |
493 | error_report("failed to send a req, %s", strerror(errno)); | |
494 | } | |
495 | ||
496 | return ret; | |
497 | } | |
e0d93a89 | 498 | |
2dfcca3b MK |
499 | static void restart_co_req(void *opaque) |
500 | { | |
501 | Coroutine *co = opaque; | |
502 | ||
503 | qemu_coroutine_enter(co, NULL); | |
504 | } | |
505 | ||
ed9ba724 MK |
506 | static int have_co_req(void *opaque) |
507 | { | |
508 | /* this handler is set only when there is a pending request, so | |
509 | * always returns 1. */ | |
510 | return 1; | |
511 | } | |
512 | ||
cddd4ac7 MK |
513 | typedef struct SheepdogReqCo { |
514 | int sockfd; | |
515 | SheepdogReq *hdr; | |
516 | void *data; | |
517 | unsigned int *wlen; | |
518 | unsigned int *rlen; | |
519 | int ret; | |
520 | bool finished; | |
521 | } SheepdogReqCo; | |
522 | ||
523 | static coroutine_fn void do_co_req(void *opaque) | |
47622c44 LY |
524 | { |
525 | int ret; | |
2dfcca3b | 526 | Coroutine *co; |
cddd4ac7 MK |
527 | SheepdogReqCo *srco = opaque; |
528 | int sockfd = srco->sockfd; | |
529 | SheepdogReq *hdr = srco->hdr; | |
530 | void *data = srco->data; | |
531 | unsigned int *wlen = srco->wlen; | |
532 | unsigned int *rlen = srco->rlen; | |
2dfcca3b MK |
533 | |
534 | co = qemu_coroutine_self(); | |
ed9ba724 | 535 | qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, have_co_req, co); |
47622c44 | 536 | |
47622c44 LY |
537 | ret = send_co_req(sockfd, hdr, data, wlen); |
538 | if (ret < 0) { | |
539 | goto out; | |
540 | } | |
541 | ||
ed9ba724 | 542 | qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, have_co_req, co); |
2dfcca3b | 543 | |
47622c44 LY |
544 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); |
545 | if (ret < sizeof(*hdr)) { | |
546 | error_report("failed to get a rsp, %s", strerror(errno)); | |
cb595887 | 547 | ret = -errno; |
47622c44 LY |
548 | goto out; |
549 | } | |
550 | ||
551 | if (*rlen > hdr->data_length) { | |
552 | *rlen = hdr->data_length; | |
553 | } | |
554 | ||
555 | if (*rlen) { | |
556 | ret = qemu_co_recv(sockfd, data, *rlen); | |
557 | if (ret < *rlen) { | |
558 | error_report("failed to get the data, %s", strerror(errno)); | |
cb595887 | 559 | ret = -errno; |
47622c44 LY |
560 | goto out; |
561 | } | |
562 | } | |
563 | ret = 0; | |
564 | out: | |
ed9ba724 MK |
565 | /* there is at most one request for this sockfd, so it is safe to |
566 | * set each handler to NULL. */ | |
2dfcca3b | 567 | qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL); |
cddd4ac7 MK |
568 | |
569 | srco->ret = ret; | |
570 | srco->finished = true; | |
571 | } | |
572 | ||
573 | static int do_req(int sockfd, SheepdogReq *hdr, void *data, | |
574 | unsigned int *wlen, unsigned int *rlen) | |
575 | { | |
576 | Coroutine *co; | |
577 | SheepdogReqCo srco = { | |
578 | .sockfd = sockfd, | |
579 | .hdr = hdr, | |
580 | .data = data, | |
581 | .wlen = wlen, | |
582 | .rlen = rlen, | |
583 | .ret = 0, | |
584 | .finished = false, | |
585 | }; | |
586 | ||
587 | if (qemu_in_coroutine()) { | |
588 | do_co_req(&srco); | |
589 | } else { | |
590 | co = qemu_coroutine_create(do_co_req); | |
591 | qemu_coroutine_enter(co, &srco); | |
592 | while (!srco.finished) { | |
593 | qemu_aio_wait(); | |
594 | } | |
595 | } | |
596 | ||
597 | return srco.ret; | |
47622c44 LY |
598 | } |
599 | ||
d8716b41 | 600 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
2f536801 | 601 | struct iovec *iov, int niov, bool create, |
33b1db1c MK |
602 | enum AIOCBState aiocb_type); |
603 | ||
7dc1cde0 MK |
604 | |
605 | static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid) | |
606 | { | |
607 | AIOReq *aio_req; | |
608 | ||
609 | QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) { | |
610 | if (aio_req->oid == oid) { | |
611 | return aio_req; | |
612 | } | |
613 | } | |
614 | ||
615 | return NULL; | |
616 | } | |
617 | ||
33b1db1c MK |
618 | /* |
619 | * This function searchs pending requests to the object `oid', and | |
620 | * sends them. | |
621 | */ | |
c292ee6a | 622 | static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid) |
33b1db1c | 623 | { |
7dc1cde0 | 624 | AIOReq *aio_req; |
33b1db1c MK |
625 | SheepdogAIOCB *acb; |
626 | int ret; | |
627 | ||
7dc1cde0 | 628 | while ((aio_req = find_pending_req(s, oid)) != NULL) { |
33b1db1c | 629 | acb = aio_req->aiocb; |
c292ee6a MK |
630 | /* move aio_req from pending list to inflight one */ |
631 | QLIST_REMOVE(aio_req, aio_siblings); | |
632 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
33b1db1c | 633 | ret = add_aio_request(s, aio_req, acb->qiov->iov, |
2f536801 | 634 | acb->qiov->niov, false, acb->aiocb_type); |
33b1db1c | 635 | if (ret < 0) { |
6daf194d | 636 | error_report("add_aio_request is failed"); |
33b1db1c | 637 | free_aio_req(s, aio_req); |
1d732d7d | 638 | if (!acb->nr_pending) { |
33b1db1c MK |
639 | sd_finish_aiocb(acb); |
640 | } | |
641 | } | |
642 | } | |
643 | } | |
644 | ||
645 | /* | |
646 | * Receive responses of the I/O requests. | |
647 | * | |
648 | * This function is registered as a fd handler, and called from the | |
649 | * main loop when s->fd is ready for reading responses. | |
650 | */ | |
d8716b41 | 651 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
652 | { |
653 | SheepdogObjRsp rsp; | |
654 | BDRVSheepdogState *s = opaque; | |
655 | int fd = s->fd; | |
656 | int ret; | |
657 | AIOReq *aio_req = NULL; | |
658 | SheepdogAIOCB *acb; | |
33b1db1c MK |
659 | unsigned long idx; |
660 | ||
c292ee6a | 661 | if (QLIST_EMPTY(&s->inflight_aio_head)) { |
2df46246 | 662 | goto out; |
33b1db1c MK |
663 | } |
664 | ||
665 | /* read a header */ | |
8c5135f9 PB |
666 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
667 | if (ret < 0) { | |
6daf194d | 668 | error_report("failed to get the header, %s", strerror(errno)); |
2df46246 | 669 | goto out; |
33b1db1c MK |
670 | } |
671 | ||
c292ee6a MK |
672 | /* find the right aio_req from the inflight aio list */ |
673 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
33b1db1c MK |
674 | if (aio_req->id == rsp.id) { |
675 | break; | |
676 | } | |
677 | } | |
678 | if (!aio_req) { | |
6daf194d | 679 | error_report("cannot find aio_req %x", rsp.id); |
2df46246 | 680 | goto out; |
33b1db1c MK |
681 | } |
682 | ||
683 | acb = aio_req->aiocb; | |
684 | ||
685 | switch (acb->aiocb_type) { | |
686 | case AIOCB_WRITE_UDATA: | |
6d1acda8 MK |
687 | /* this coroutine context is no longer suitable for co_recv |
688 | * because we may send data to update vdi objects */ | |
689 | s->co_recv = NULL; | |
33b1db1c MK |
690 | if (!is_data_obj(aio_req->oid)) { |
691 | break; | |
692 | } | |
693 | idx = data_oid_to_idx(aio_req->oid); | |
694 | ||
695 | if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) { | |
696 | /* | |
697 | * If the object is newly created one, we need to update | |
698 | * the vdi object (metadata object). min_dirty_data_idx | |
699 | * and max_dirty_data_idx are changed to include updated | |
700 | * index between them. | |
701 | */ | |
bd751f22 LY |
702 | if (rsp.result == SD_RES_SUCCESS) { |
703 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
704 | s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); | |
705 | s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); | |
706 | } | |
33b1db1c MK |
707 | /* |
708 | * Some requests may be blocked because simultaneous | |
709 | * create requests are not allowed, so we search the | |
710 | * pending requests here. | |
711 | */ | |
d6b1ef89 | 712 | send_pending_req(s, aio_req->oid); |
33b1db1c MK |
713 | } |
714 | break; | |
715 | case AIOCB_READ_UDATA: | |
2fc8ae1d MT |
716 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
717 | aio_req->iov_offset, rsp.data_length); | |
8c5135f9 | 718 | if (ret < 0) { |
6daf194d | 719 | error_report("failed to get the data, %s", strerror(errno)); |
2df46246 | 720 | goto out; |
33b1db1c MK |
721 | } |
722 | break; | |
47783072 LY |
723 | case AIOCB_FLUSH_CACHE: |
724 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
725 | dprintf("disable cache since the server doesn't support it\n"); | |
726 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
727 | rsp.result = SD_RES_SUCCESS; | |
728 | } | |
729 | break; | |
33b1db1c MK |
730 | } |
731 | ||
732 | if (rsp.result != SD_RES_SUCCESS) { | |
733 | acb->ret = -EIO; | |
6daf194d | 734 | error_report("%s", sd_strerror(rsp.result)); |
33b1db1c MK |
735 | } |
736 | ||
1d732d7d MK |
737 | free_aio_req(s, aio_req); |
738 | if (!acb->nr_pending) { | |
33b1db1c MK |
739 | /* |
740 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 741 | * we can switch back to sd_co_readv/writev now. |
33b1db1c MK |
742 | */ |
743 | acb->aio_done_func(acb); | |
744 | } | |
2df46246 MK |
745 | out: |
746 | s->co_recv = NULL; | |
747 | } | |
748 | ||
749 | static void co_read_response(void *opaque) | |
750 | { | |
751 | BDRVSheepdogState *s = opaque; | |
752 | ||
753 | if (!s->co_recv) { | |
754 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
755 | } | |
756 | ||
757 | qemu_coroutine_enter(s->co_recv, opaque); | |
758 | } | |
759 | ||
760 | static void co_write_request(void *opaque) | |
761 | { | |
762 | BDRVSheepdogState *s = opaque; | |
763 | ||
764 | qemu_coroutine_enter(s->co_send, NULL); | |
33b1db1c MK |
765 | } |
766 | ||
767 | static int aio_flush_request(void *opaque) | |
768 | { | |
769 | BDRVSheepdogState *s = opaque; | |
770 | ||
c292ee6a MK |
771 | return !QLIST_EMPTY(&s->inflight_aio_head) || |
772 | !QLIST_EMPTY(&s->pending_aio_head); | |
33b1db1c MK |
773 | } |
774 | ||
33b1db1c MK |
775 | /* |
776 | * Return a socket discriptor to read/write objects. | |
777 | * | |
778 | * We cannot use this discriptor for other operations because | |
779 | * the block driver may be on waiting response from the server. | |
780 | */ | |
781 | static int get_sheep_fd(BDRVSheepdogState *s) | |
782 | { | |
1b8bbb46 | 783 | int fd; |
33b1db1c | 784 | |
25af257d | 785 | fd = connect_to_sdog(s); |
33b1db1c | 786 | if (fd < 0) { |
cb595887 | 787 | return fd; |
33b1db1c MK |
788 | } |
789 | ||
bafbd6a1 | 790 | qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s); |
33b1db1c MK |
791 | return fd; |
792 | } | |
793 | ||
5d6768e3 MK |
794 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
795 | char *vdi, uint32_t *snapid, char *tag) | |
796 | { | |
797 | URI *uri; | |
798 | QueryParams *qp = NULL; | |
799 | int ret = 0; | |
800 | ||
801 | uri = uri_parse(filename); | |
802 | if (!uri) { | |
803 | return -EINVAL; | |
804 | } | |
805 | ||
1b8bbb46 MK |
806 | /* transport */ |
807 | if (!strcmp(uri->scheme, "sheepdog")) { | |
808 | s->is_unix = false; | |
809 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
810 | s->is_unix = false; | |
811 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
812 | s->is_unix = true; | |
813 | } else { | |
814 | ret = -EINVAL; | |
815 | goto out; | |
816 | } | |
817 | ||
5d6768e3 MK |
818 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
819 | ret = -EINVAL; | |
820 | goto out; | |
821 | } | |
822 | pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1); | |
823 | ||
1b8bbb46 MK |
824 | qp = query_params_parse(uri->query); |
825 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
826 | ret = -EINVAL; | |
827 | goto out; | |
828 | } | |
829 | ||
830 | if (s->is_unix) { | |
831 | /* sheepdog+unix:///vdiname?socket=path */ | |
832 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
833 | ret = -EINVAL; | |
834 | goto out; | |
835 | } | |
836 | s->host_spec = g_strdup(qp->p[0].value); | |
837 | } else { | |
838 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
839 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
840 | uri->port ?: SD_DEFAULT_PORT); | |
841 | } | |
5d6768e3 MK |
842 | |
843 | /* snapshot tag */ | |
844 | if (uri->fragment) { | |
845 | *snapid = strtoul(uri->fragment, NULL, 10); | |
846 | if (*snapid == 0) { | |
847 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); | |
848 | } | |
849 | } else { | |
850 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
851 | } | |
852 | ||
853 | out: | |
854 | if (qp) { | |
855 | query_params_free(qp); | |
856 | } | |
857 | uri_free(uri); | |
858 | return ret; | |
859 | } | |
860 | ||
33b1db1c | 861 | /* |
5d6768e3 | 862 | * Parse a filename (old syntax) |
33b1db1c MK |
863 | * |
864 | * filename must be one of the following formats: | |
865 | * 1. [vdiname] | |
866 | * 2. [vdiname]:[snapid] | |
867 | * 3. [vdiname]:[tag] | |
868 | * 4. [hostname]:[port]:[vdiname] | |
869 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
870 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
871 | * | |
872 | * You can boot from the snapshot images by specifying `snapid` or | |
873 | * `tag'. | |
874 | * | |
875 | * You can run VMs outside the Sheepdog cluster by specifying | |
876 | * `hostname' and `port' (experimental). | |
877 | */ | |
878 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
879 | char *vdi, uint32_t *snapid, char *tag) | |
880 | { | |
5d6768e3 MK |
881 | char *p, *q, *uri; |
882 | const char *host_spec, *vdi_spec; | |
883 | int nr_sep, ret; | |
33b1db1c | 884 | |
5d6768e3 | 885 | strstart(filename, "sheepdog:", (const char **)&filename); |
7267c094 | 886 | p = q = g_strdup(filename); |
33b1db1c MK |
887 | |
888 | /* count the number of separators */ | |
889 | nr_sep = 0; | |
890 | while (*p) { | |
891 | if (*p == ':') { | |
892 | nr_sep++; | |
893 | } | |
894 | p++; | |
895 | } | |
896 | p = q; | |
897 | ||
5d6768e3 | 898 | /* use the first two tokens as host_spec. */ |
33b1db1c | 899 | if (nr_sep >= 2) { |
5d6768e3 | 900 | host_spec = p; |
33b1db1c | 901 | p = strchr(p, ':'); |
5d6768e3 | 902 | p++; |
33b1db1c MK |
903 | p = strchr(p, ':'); |
904 | *p++ = '\0'; | |
905 | } else { | |
5d6768e3 | 906 | host_spec = ""; |
33b1db1c MK |
907 | } |
908 | ||
5d6768e3 | 909 | vdi_spec = p; |
33b1db1c | 910 | |
5d6768e3 | 911 | p = strchr(vdi_spec, ':'); |
33b1db1c | 912 | if (p) { |
5d6768e3 | 913 | *p++ = '#'; |
33b1db1c MK |
914 | } |
915 | ||
5d6768e3 | 916 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 917 | |
5d6768e3 MK |
918 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
919 | ||
920 | g_free(q); | |
921 | g_free(uri); | |
922 | ||
923 | return ret; | |
33b1db1c MK |
924 | } |
925 | ||
926 | static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid, | |
927 | char *tag, uint32_t *vid, int for_snapshot) | |
928 | { | |
929 | int ret, fd; | |
930 | SheepdogVdiReq hdr; | |
931 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
932 | unsigned int wlen, rlen = 0; | |
933 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
934 | ||
25af257d | 935 | fd = connect_to_sdog(s); |
33b1db1c | 936 | if (fd < 0) { |
cb595887 | 937 | return fd; |
33b1db1c MK |
938 | } |
939 | ||
3178e275 JM |
940 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
941 | * which is desirable since we'll soon be sending those bytes, and | |
942 | * don't want the send_req to read uninitialized data. | |
943 | */ | |
33b1db1c MK |
944 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
945 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
946 | ||
947 | memset(&hdr, 0, sizeof(hdr)); | |
948 | if (for_snapshot) { | |
949 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
950 | } else { | |
951 | hdr.opcode = SD_OP_LOCK_VDI; | |
952 | } | |
953 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
954 | hdr.proto_ver = SD_PROTO_VER; | |
955 | hdr.data_length = wlen; | |
956 | hdr.snapid = snapid; | |
957 | hdr.flags = SD_FLAG_CMD_WRITE; | |
958 | ||
959 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
960 | if (ret) { | |
33b1db1c MK |
961 | goto out; |
962 | } | |
963 | ||
964 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 965 | error_report("cannot get vdi info, %s, %s %d %s", |
33b1db1c | 966 | sd_strerror(rsp->result), filename, snapid, tag); |
cb595887 MK |
967 | if (rsp->result == SD_RES_NO_VDI) { |
968 | ret = -ENOENT; | |
969 | } else { | |
970 | ret = -EIO; | |
971 | } | |
33b1db1c MK |
972 | goto out; |
973 | } | |
974 | *vid = rsp->vdi_id; | |
975 | ||
976 | ret = 0; | |
977 | out: | |
978 | closesocket(fd); | |
979 | return ret; | |
980 | } | |
981 | ||
d8716b41 | 982 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
2f536801 | 983 | struct iovec *iov, int niov, bool create, |
33b1db1c MK |
984 | enum AIOCBState aiocb_type) |
985 | { | |
986 | int nr_copies = s->inode.nr_copies; | |
987 | SheepdogObjReq hdr; | |
47783072 | 988 | unsigned int wlen = 0; |
33b1db1c MK |
989 | int ret; |
990 | uint64_t oid = aio_req->oid; | |
991 | unsigned int datalen = aio_req->data_len; | |
992 | uint64_t offset = aio_req->offset; | |
993 | uint8_t flags = aio_req->flags; | |
994 | uint64_t old_oid = aio_req->base_oid; | |
995 | ||
996 | if (!nr_copies) { | |
6daf194d | 997 | error_report("bug"); |
33b1db1c MK |
998 | } |
999 | ||
1000 | memset(&hdr, 0, sizeof(hdr)); | |
1001 | ||
47783072 LY |
1002 | switch (aiocb_type) { |
1003 | case AIOCB_FLUSH_CACHE: | |
1004 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1005 | break; | |
1006 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1007 | hdr.opcode = SD_OP_READ_OBJ; |
1008 | hdr.flags = flags; | |
47783072 LY |
1009 | break; |
1010 | case AIOCB_WRITE_UDATA: | |
1011 | if (create) { | |
1012 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1013 | } else { | |
1014 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1015 | } | |
33b1db1c | 1016 | wlen = datalen; |
33b1db1c | 1017 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1018 | break; |
33b1db1c MK |
1019 | } |
1020 | ||
0e7106d8 LY |
1021 | if (s->cache_flags) { |
1022 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1023 | } |
1024 | ||
33b1db1c MK |
1025 | hdr.oid = oid; |
1026 | hdr.cow_oid = old_oid; | |
1027 | hdr.copies = s->inode.nr_copies; | |
1028 | ||
1029 | hdr.data_length = datalen; | |
1030 | hdr.offset = offset; | |
1031 | ||
1032 | hdr.id = aio_req->id; | |
1033 | ||
2df46246 MK |
1034 | qemu_co_mutex_lock(&s->lock); |
1035 | s->co_send = qemu_coroutine_self(); | |
1036 | qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, | |
bafbd6a1 | 1037 | aio_flush_request, s); |
128aa589 | 1038 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1039 | |
1040 | /* send a header */ | |
8c5135f9 PB |
1041 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
1042 | if (ret < 0) { | |
c3fecea5 | 1043 | qemu_co_mutex_unlock(&s->lock); |
6daf194d | 1044 | error_report("failed to send a req, %s", strerror(errno)); |
cb595887 | 1045 | return -errno; |
33b1db1c MK |
1046 | } |
1047 | ||
1048 | if (wlen) { | |
2fc8ae1d | 1049 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
8c5135f9 | 1050 | if (ret < 0) { |
c3fecea5 | 1051 | qemu_co_mutex_unlock(&s->lock); |
6daf194d | 1052 | error_report("failed to send a data, %s", strerror(errno)); |
cb595887 | 1053 | return -errno; |
33b1db1c MK |
1054 | } |
1055 | } | |
1056 | ||
128aa589 | 1057 | socket_set_cork(s->fd, 0); |
2df46246 | 1058 | qemu_aio_set_fd_handler(s->fd, co_read_response, NULL, |
bafbd6a1 | 1059 | aio_flush_request, s); |
2df46246 | 1060 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1061 | |
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | static int read_write_object(int fd, char *buf, uint64_t oid, int copies, | |
1066 | unsigned int datalen, uint64_t offset, | |
0e7106d8 | 1067 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1068 | { |
1069 | SheepdogObjReq hdr; | |
1070 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1071 | unsigned int wlen, rlen; | |
1072 | int ret; | |
1073 | ||
1074 | memset(&hdr, 0, sizeof(hdr)); | |
1075 | ||
1076 | if (write) { | |
1077 | wlen = datalen; | |
1078 | rlen = 0; | |
1079 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1080 | if (create) { | |
1081 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1082 | } else { | |
1083 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1084 | } | |
1085 | } else { | |
1086 | wlen = 0; | |
1087 | rlen = datalen; | |
1088 | hdr.opcode = SD_OP_READ_OBJ; | |
1089 | } | |
47622c44 | 1090 | |
0e7106d8 | 1091 | hdr.flags |= cache_flags; |
47622c44 | 1092 | |
33b1db1c MK |
1093 | hdr.oid = oid; |
1094 | hdr.data_length = datalen; | |
1095 | hdr.offset = offset; | |
1096 | hdr.copies = copies; | |
1097 | ||
1098 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1099 | if (ret) { | |
6daf194d | 1100 | error_report("failed to send a request to the sheep"); |
cb595887 | 1101 | return ret; |
33b1db1c MK |
1102 | } |
1103 | ||
1104 | switch (rsp->result) { | |
1105 | case SD_RES_SUCCESS: | |
1106 | return 0; | |
1107 | default: | |
6daf194d | 1108 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1109 | return -EIO; |
33b1db1c MK |
1110 | } |
1111 | } | |
1112 | ||
1113 | static int read_object(int fd, char *buf, uint64_t oid, int copies, | |
0e7106d8 LY |
1114 | unsigned int datalen, uint64_t offset, |
1115 | uint32_t cache_flags) | |
33b1db1c | 1116 | { |
2f536801 | 1117 | return read_write_object(fd, buf, oid, copies, datalen, offset, false, |
0e7106d8 | 1118 | false, cache_flags); |
33b1db1c MK |
1119 | } |
1120 | ||
1121 | static int write_object(int fd, char *buf, uint64_t oid, int copies, | |
2f536801 | 1122 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1123 | uint32_t cache_flags) |
33b1db1c | 1124 | { |
2f536801 | 1125 | return read_write_object(fd, buf, oid, copies, datalen, offset, true, |
0e7106d8 | 1126 | create, cache_flags); |
33b1db1c MK |
1127 | } |
1128 | ||
c8c96350 KW |
1129 | /* TODO Convert to fine grained options */ |
1130 | static QemuOptsList runtime_opts = { | |
1131 | .name = "sheepdog", | |
1132 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1133 | .desc = { | |
1134 | { | |
1135 | .name = "filename", | |
1136 | .type = QEMU_OPT_STRING, | |
1137 | .help = "URL to the sheepdog image", | |
1138 | }, | |
1139 | { /* end of list */ } | |
1140 | }, | |
1141 | }; | |
1142 | ||
56d1b4d2 | 1143 | static int sd_open(BlockDriverState *bs, QDict *options, int flags) |
33b1db1c MK |
1144 | { |
1145 | int ret, fd; | |
1146 | uint32_t vid = 0; | |
1147 | BDRVSheepdogState *s = bs->opaque; | |
1148 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1149 | uint32_t snapid; | |
1150 | char *buf = NULL; | |
c8c96350 KW |
1151 | QemuOpts *opts; |
1152 | Error *local_err = NULL; | |
1153 | const char *filename; | |
1154 | ||
1155 | opts = qemu_opts_create_nofail(&runtime_opts); | |
1156 | qemu_opts_absorb_qdict(opts, options, &local_err); | |
1157 | if (error_is_set(&local_err)) { | |
1158 | qerror_report_err(local_err); | |
1159 | error_free(local_err); | |
1160 | ret = -EINVAL; | |
1161 | goto out; | |
1162 | } | |
1163 | ||
1164 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1165 | |
c292ee6a MK |
1166 | QLIST_INIT(&s->inflight_aio_head); |
1167 | QLIST_INIT(&s->pending_aio_head); | |
33b1db1c MK |
1168 | s->fd = -1; |
1169 | ||
1170 | memset(vdi, 0, sizeof(vdi)); | |
1171 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1172 | |
1173 | if (strstr(filename, "://")) { | |
1174 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1175 | } else { | |
1176 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1177 | } | |
1178 | if (ret < 0) { | |
33b1db1c MK |
1179 | goto out; |
1180 | } | |
1181 | s->fd = get_sheep_fd(s); | |
1182 | if (s->fd < 0) { | |
cb595887 | 1183 | ret = s->fd; |
33b1db1c MK |
1184 | goto out; |
1185 | } | |
1186 | ||
1187 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0); | |
1188 | if (ret) { | |
1189 | goto out; | |
1190 | } | |
1191 | ||
0e7106d8 LY |
1192 | /* |
1193 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1194 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1195 | */ | |
1196 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1197 | if (flags & BDRV_O_NOCACHE) { | |
1198 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1199 | } | |
1200 | ||
622b6057 | 1201 | if (snapid || tag[0] != '\0') { |
33b1db1c | 1202 | dprintf("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1203 | s->is_snapshot = true; |
33b1db1c MK |
1204 | } |
1205 | ||
25af257d | 1206 | fd = connect_to_sdog(s); |
33b1db1c | 1207 | if (fd < 0) { |
cb595887 | 1208 | ret = fd; |
33b1db1c MK |
1209 | goto out; |
1210 | } | |
1211 | ||
7267c094 | 1212 | buf = g_malloc(SD_INODE_SIZE); |
47622c44 | 1213 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, |
0e7106d8 | 1214 | s->cache_flags); |
33b1db1c MK |
1215 | |
1216 | closesocket(fd); | |
1217 | ||
1218 | if (ret) { | |
1219 | goto out; | |
1220 | } | |
1221 | ||
1222 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1223 | s->min_dirty_data_idx = UINT32_MAX; | |
1224 | s->max_dirty_data_idx = 0; | |
1225 | ||
1226 | bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE; | |
3178e275 | 1227 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1228 | qemu_co_mutex_init(&s->lock); |
c8c96350 | 1229 | qemu_opts_del(opts); |
7267c094 | 1230 | g_free(buf); |
33b1db1c MK |
1231 | return 0; |
1232 | out: | |
bafbd6a1 | 1233 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
33b1db1c MK |
1234 | if (s->fd >= 0) { |
1235 | closesocket(s->fd); | |
1236 | } | |
c8c96350 | 1237 | qemu_opts_del(opts); |
7267c094 | 1238 | g_free(buf); |
cb595887 | 1239 | return ret; |
33b1db1c MK |
1240 | } |
1241 | ||
25af257d MK |
1242 | static int do_sd_create(BDRVSheepdogState *s, char *filename, int64_t vdi_size, |
1243 | uint32_t base_vid, uint32_t *vdi_id, int snapshot) | |
33b1db1c MK |
1244 | { |
1245 | SheepdogVdiReq hdr; | |
1246 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1247 | int fd, ret; | |
1248 | unsigned int wlen, rlen = 0; | |
1249 | char buf[SD_MAX_VDI_LEN]; | |
1250 | ||
25af257d | 1251 | fd = connect_to_sdog(s); |
33b1db1c | 1252 | if (fd < 0) { |
cb595887 | 1253 | return fd; |
33b1db1c MK |
1254 | } |
1255 | ||
3178e275 JM |
1256 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1257 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1258 | */ | |
33b1db1c | 1259 | memset(buf, 0, sizeof(buf)); |
3178e275 | 1260 | pstrcpy(buf, sizeof(buf), filename); |
33b1db1c MK |
1261 | |
1262 | memset(&hdr, 0, sizeof(hdr)); | |
1263 | hdr.opcode = SD_OP_NEW_VDI; | |
6f74c260 | 1264 | hdr.vdi_id = base_vid; |
33b1db1c MK |
1265 | |
1266 | wlen = SD_MAX_VDI_LEN; | |
1267 | ||
1268 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1269 | hdr.snapid = snapshot; | |
1270 | ||
1271 | hdr.data_length = wlen; | |
1272 | hdr.vdi_size = vdi_size; | |
1273 | ||
1274 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1275 | ||
1276 | closesocket(fd); | |
1277 | ||
1278 | if (ret) { | |
cb595887 | 1279 | return ret; |
33b1db1c MK |
1280 | } |
1281 | ||
1282 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 1283 | error_report("%s, %s", sd_strerror(rsp->result), filename); |
33b1db1c MK |
1284 | return -EIO; |
1285 | } | |
1286 | ||
1287 | if (vdi_id) { | |
1288 | *vdi_id = rsp->vdi_id; | |
1289 | } | |
1290 | ||
1291 | return 0; | |
1292 | } | |
1293 | ||
a8e0fdd7 MK |
1294 | static int sd_prealloc(const char *filename) |
1295 | { | |
1296 | BlockDriverState *bs = NULL; | |
1297 | uint32_t idx, max_idx; | |
1298 | int64_t vdi_size; | |
7267c094 | 1299 | void *buf = g_malloc0(SD_DATA_OBJ_SIZE); |
a8e0fdd7 MK |
1300 | int ret; |
1301 | ||
787e4a85 | 1302 | ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR); |
a8e0fdd7 MK |
1303 | if (ret < 0) { |
1304 | goto out; | |
1305 | } | |
1306 | ||
1307 | vdi_size = bdrv_getlength(bs); | |
1308 | if (vdi_size < 0) { | |
1309 | ret = vdi_size; | |
1310 | goto out; | |
1311 | } | |
1312 | max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); | |
1313 | ||
1314 | for (idx = 0; idx < max_idx; idx++) { | |
1315 | /* | |
1316 | * The created image can be a cloned image, so we need to read | |
1317 | * a data from the source image. | |
1318 | */ | |
1319 | ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1320 | if (ret < 0) { | |
1321 | goto out; | |
1322 | } | |
1323 | ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1324 | if (ret < 0) { | |
1325 | goto out; | |
1326 | } | |
1327 | } | |
1328 | out: | |
1329 | if (bs) { | |
1330 | bdrv_delete(bs); | |
1331 | } | |
7267c094 | 1332 | g_free(buf); |
a8e0fdd7 MK |
1333 | |
1334 | return ret; | |
1335 | } | |
1336 | ||
33b1db1c MK |
1337 | static int sd_create(const char *filename, QEMUOptionParameter *options) |
1338 | { | |
b6fc8245 | 1339 | int ret = 0; |
b4447363 | 1340 | uint32_t vid = 0, base_vid = 0; |
33b1db1c MK |
1341 | int64_t vdi_size = 0; |
1342 | char *backing_file = NULL; | |
b6fc8245 | 1343 | BDRVSheepdogState *s; |
b4447363 MK |
1344 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; |
1345 | uint32_t snapid; | |
2f536801 | 1346 | bool prealloc = false; |
33b1db1c | 1347 | |
b6fc8245 MK |
1348 | s = g_malloc0(sizeof(BDRVSheepdogState)); |
1349 | ||
b4447363 MK |
1350 | memset(vdi, 0, sizeof(vdi)); |
1351 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1352 | if (strstr(filename, "://")) { |
1353 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1354 | } else { | |
1355 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1356 | } | |
1357 | if (ret < 0) { | |
b6fc8245 | 1358 | goto out; |
b4447363 MK |
1359 | } |
1360 | ||
33b1db1c MK |
1361 | while (options && options->name) { |
1362 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1363 | vdi_size = options->value.n; | |
1364 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1365 | backing_file = options->value.s; | |
a8e0fdd7 MK |
1366 | } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1367 | if (!options->value.s || !strcmp(options->value.s, "off")) { | |
2f536801 | 1368 | prealloc = false; |
a8e0fdd7 | 1369 | } else if (!strcmp(options->value.s, "full")) { |
2f536801 | 1370 | prealloc = true; |
a8e0fdd7 MK |
1371 | } else { |
1372 | error_report("Invalid preallocation mode: '%s'", | |
1373 | options->value.s); | |
b6fc8245 MK |
1374 | ret = -EINVAL; |
1375 | goto out; | |
a8e0fdd7 | 1376 | } |
33b1db1c MK |
1377 | } |
1378 | options++; | |
1379 | } | |
1380 | ||
1381 | if (vdi_size > SD_MAX_VDI_SIZE) { | |
6daf194d | 1382 | error_report("too big image size"); |
b6fc8245 MK |
1383 | ret = -EINVAL; |
1384 | goto out; | |
33b1db1c MK |
1385 | } |
1386 | ||
1387 | if (backing_file) { | |
1388 | BlockDriverState *bs; | |
1389 | BDRVSheepdogState *s; | |
1390 | BlockDriver *drv; | |
1391 | ||
1392 | /* Currently, only Sheepdog backing image is supported. */ | |
1393 | drv = bdrv_find_protocol(backing_file); | |
1394 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { | |
6daf194d | 1395 | error_report("backing_file must be a sheepdog image"); |
b6fc8245 MK |
1396 | ret = -EINVAL; |
1397 | goto out; | |
33b1db1c MK |
1398 | } |
1399 | ||
787e4a85 | 1400 | ret = bdrv_file_open(&bs, backing_file, NULL, 0); |
cb595887 | 1401 | if (ret < 0) { |
b6fc8245 | 1402 | goto out; |
cb595887 | 1403 | } |
33b1db1c MK |
1404 | |
1405 | s = bs->opaque; | |
1406 | ||
1407 | if (!is_snapshot(&s->inode)) { | |
6daf194d | 1408 | error_report("cannot clone from a non snapshot vdi"); |
33b1db1c | 1409 | bdrv_delete(bs); |
b6fc8245 MK |
1410 | ret = -EINVAL; |
1411 | goto out; | |
33b1db1c MK |
1412 | } |
1413 | ||
b4447363 | 1414 | base_vid = s->inode.vdi_id; |
33b1db1c MK |
1415 | bdrv_delete(bs); |
1416 | } | |
1417 | ||
25af257d | 1418 | ret = do_sd_create(s, vdi, vdi_size, base_vid, &vid, 0); |
a8e0fdd7 | 1419 | if (!prealloc || ret) { |
b6fc8245 | 1420 | goto out; |
a8e0fdd7 MK |
1421 | } |
1422 | ||
b6fc8245 MK |
1423 | ret = sd_prealloc(filename); |
1424 | out: | |
1425 | g_free(s); | |
1426 | return ret; | |
33b1db1c MK |
1427 | } |
1428 | ||
1429 | static void sd_close(BlockDriverState *bs) | |
1430 | { | |
1431 | BDRVSheepdogState *s = bs->opaque; | |
1432 | SheepdogVdiReq hdr; | |
1433 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1434 | unsigned int wlen, rlen = 0; | |
1435 | int fd, ret; | |
1436 | ||
1437 | dprintf("%s\n", s->name); | |
1438 | ||
25af257d | 1439 | fd = connect_to_sdog(s); |
33b1db1c MK |
1440 | if (fd < 0) { |
1441 | return; | |
1442 | } | |
1443 | ||
1444 | memset(&hdr, 0, sizeof(hdr)); | |
1445 | ||
1446 | hdr.opcode = SD_OP_RELEASE_VDI; | |
6f74c260 | 1447 | hdr.vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1448 | wlen = strlen(s->name) + 1; |
1449 | hdr.data_length = wlen; | |
1450 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1451 | ||
1452 | ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); | |
1453 | ||
1454 | closesocket(fd); | |
1455 | ||
1456 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1457 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1458 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1459 | } |
1460 | ||
bafbd6a1 | 1461 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
33b1db1c | 1462 | closesocket(s->fd); |
25af257d | 1463 | g_free(s->host_spec); |
33b1db1c MK |
1464 | } |
1465 | ||
1466 | static int64_t sd_getlength(BlockDriverState *bs) | |
1467 | { | |
1468 | BDRVSheepdogState *s = bs->opaque; | |
1469 | ||
1470 | return s->inode.vdi_size; | |
1471 | } | |
1472 | ||
1473 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1474 | { | |
1475 | BDRVSheepdogState *s = bs->opaque; | |
1476 | int ret, fd; | |
1477 | unsigned int datalen; | |
1478 | ||
1479 | if (offset < s->inode.vdi_size) { | |
6daf194d | 1480 | error_report("shrinking is not supported"); |
33b1db1c MK |
1481 | return -EINVAL; |
1482 | } else if (offset > SD_MAX_VDI_SIZE) { | |
6daf194d | 1483 | error_report("too big image size"); |
33b1db1c MK |
1484 | return -EINVAL; |
1485 | } | |
1486 | ||
25af257d | 1487 | fd = connect_to_sdog(s); |
33b1db1c | 1488 | if (fd < 0) { |
cb595887 | 1489 | return fd; |
33b1db1c MK |
1490 | } |
1491 | ||
1492 | /* we don't need to update entire object */ | |
1493 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1494 | s->inode.vdi_size = offset; | |
1495 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
0e7106d8 | 1496 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
33b1db1c MK |
1497 | close(fd); |
1498 | ||
1499 | if (ret < 0) { | |
6daf194d | 1500 | error_report("failed to update an inode."); |
33b1db1c MK |
1501 | } |
1502 | ||
cb595887 | 1503 | return ret; |
33b1db1c MK |
1504 | } |
1505 | ||
1506 | /* | |
1507 | * This function is called after writing data objects. If we need to | |
1508 | * update metadata, this sends a write request to the vdi object. | |
2df46246 | 1509 | * Otherwise, this switches back to sd_co_readv/writev. |
33b1db1c | 1510 | */ |
d8716b41 | 1511 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c MK |
1512 | { |
1513 | int ret; | |
1514 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1515 | struct iovec iov; | |
1516 | AIOReq *aio_req; | |
1517 | uint32_t offset, data_len, mn, mx; | |
1518 | ||
1519 | mn = s->min_dirty_data_idx; | |
1520 | mx = s->max_dirty_data_idx; | |
1521 | if (mn <= mx) { | |
1522 | /* we need to update the vdi object. */ | |
1523 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
1524 | mn * sizeof(s->inode.data_vdi_id[0]); | |
1525 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
1526 | ||
1527 | s->min_dirty_data_idx = UINT32_MAX; | |
1528 | s->max_dirty_data_idx = 0; | |
1529 | ||
1530 | iov.iov_base = &s->inode; | |
1531 | iov.iov_len = sizeof(s->inode); | |
1532 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
1533 | data_len, offset, 0, 0, offset); | |
c292ee6a | 1534 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
2f536801 | 1535 | ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); |
33b1db1c MK |
1536 | if (ret) { |
1537 | free_aio_req(s, aio_req); | |
1538 | acb->ret = -EIO; | |
1539 | goto out; | |
1540 | } | |
1541 | ||
1542 | acb->aio_done_func = sd_finish_aiocb; | |
1543 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1544 | return; | |
1545 | } | |
1546 | out: | |
1547 | sd_finish_aiocb(acb); | |
1548 | } | |
1549 | ||
1550 | /* | |
1551 | * Create a writable VDI from a snapshot | |
1552 | */ | |
1553 | static int sd_create_branch(BDRVSheepdogState *s) | |
1554 | { | |
1555 | int ret, fd; | |
1556 | uint32_t vid; | |
1557 | char *buf; | |
1558 | ||
1559 | dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); | |
1560 | ||
7267c094 | 1561 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 1562 | |
25af257d | 1563 | ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1); |
33b1db1c MK |
1564 | if (ret) { |
1565 | goto out; | |
1566 | } | |
1567 | ||
1568 | dprintf("%" PRIx32 " is created.\n", vid); | |
1569 | ||
25af257d | 1570 | fd = connect_to_sdog(s); |
33b1db1c | 1571 | if (fd < 0) { |
cb595887 | 1572 | ret = fd; |
33b1db1c MK |
1573 | goto out; |
1574 | } | |
1575 | ||
1576 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, | |
0e7106d8 | 1577 | SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1578 | |
1579 | closesocket(fd); | |
1580 | ||
1581 | if (ret < 0) { | |
1582 | goto out; | |
1583 | } | |
1584 | ||
1585 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1586 | ||
2f536801 | 1587 | s->is_snapshot = false; |
33b1db1c MK |
1588 | ret = 0; |
1589 | dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id); | |
1590 | ||
1591 | out: | |
7267c094 | 1592 | g_free(buf); |
33b1db1c MK |
1593 | |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Send I/O requests to the server. | |
1599 | * | |
1600 | * This function sends requests to the server, links the requests to | |
c292ee6a | 1601 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
1602 | * waiting the response. The responses are received in the |
1603 | * `aio_read_response' function which is called from the main loop as | |
1604 | * a fd handler. | |
2df46246 MK |
1605 | * |
1606 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
1607 | * request and -errno in error cases. | |
33b1db1c | 1608 | */ |
d8716b41 | 1609 | static int coroutine_fn sd_co_rw_vector(void *p) |
33b1db1c MK |
1610 | { |
1611 | SheepdogAIOCB *acb = p; | |
1612 | int ret = 0; | |
1613 | unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE; | |
1614 | unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE; | |
1615 | uint64_t oid; | |
1616 | uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE; | |
1617 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1618 | SheepdogInode *inode = &s->inode; | |
1619 | AIOReq *aio_req; | |
1620 | ||
33b1db1c MK |
1621 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
1622 | /* | |
1623 | * In the case we open the snapshot VDI, Sheepdog creates the | |
1624 | * writable VDI when we do a write operation first. | |
1625 | */ | |
1626 | ret = sd_create_branch(s); | |
1627 | if (ret) { | |
1628 | acb->ret = -EIO; | |
1629 | goto out; | |
1630 | } | |
1631 | } | |
1632 | ||
1d732d7d MK |
1633 | /* |
1634 | * Make sure we don't free the aiocb before we are done with all requests. | |
1635 | * This additional reference is dropped at the end of this function. | |
1636 | */ | |
1637 | acb->nr_pending++; | |
1638 | ||
33b1db1c MK |
1639 | while (done != total) { |
1640 | uint8_t flags = 0; | |
1641 | uint64_t old_oid = 0; | |
2f536801 | 1642 | bool create = false; |
33b1db1c MK |
1643 | |
1644 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
1645 | ||
1646 | len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); | |
1647 | ||
19db9b90 CH |
1648 | switch (acb->aiocb_type) { |
1649 | case AIOCB_READ_UDATA: | |
1650 | if (!inode->data_vdi_id[idx]) { | |
1651 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
1652 | goto done; |
1653 | } | |
19db9b90 CH |
1654 | break; |
1655 | case AIOCB_WRITE_UDATA: | |
1656 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 1657 | create = true; |
19db9b90 CH |
1658 | } else if (!is_data_obj_writable(inode, idx)) { |
1659 | /* Copy-On-Write */ | |
2f536801 | 1660 | create = true; |
19db9b90 CH |
1661 | old_oid = oid; |
1662 | flags = SD_FLAG_CMD_COW; | |
1663 | } | |
1664 | break; | |
1665 | default: | |
1666 | break; | |
33b1db1c MK |
1667 | } |
1668 | ||
1669 | if (create) { | |
1b6ac998 MK |
1670 | dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1671 | inode->vdi_id, oid, | |
33b1db1c MK |
1672 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
1673 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
1b6ac998 | 1674 | dprintf("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
1675 | } |
1676 | ||
1677 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); | |
1678 | ||
1679 | if (create) { | |
1680 | AIOReq *areq; | |
c292ee6a | 1681 | QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { |
33b1db1c MK |
1682 | if (areq->oid == oid) { |
1683 | /* | |
1684 | * Sheepdog cannot handle simultaneous create | |
1685 | * requests to the same object. So we cannot send | |
1686 | * the request until the previous request | |
1687 | * finishes. | |
1688 | */ | |
1689 | aio_req->flags = 0; | |
1690 | aio_req->base_oid = 0; | |
c292ee6a MK |
1691 | QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, |
1692 | aio_siblings); | |
33b1db1c MK |
1693 | goto done; |
1694 | } | |
1695 | } | |
1696 | } | |
1697 | ||
c292ee6a | 1698 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
33b1db1c MK |
1699 | ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
1700 | create, acb->aiocb_type); | |
1701 | if (ret < 0) { | |
6daf194d | 1702 | error_report("add_aio_request is failed"); |
33b1db1c MK |
1703 | free_aio_req(s, aio_req); |
1704 | acb->ret = -EIO; | |
1705 | goto out; | |
1706 | } | |
1707 | done: | |
1708 | offset = 0; | |
1709 | idx++; | |
1710 | done += len; | |
1711 | } | |
1712 | out: | |
1d732d7d | 1713 | if (!--acb->nr_pending) { |
2df46246 | 1714 | return acb->ret; |
33b1db1c | 1715 | } |
2df46246 | 1716 | return 1; |
33b1db1c MK |
1717 | } |
1718 | ||
a968168c | 1719 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1720 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1721 | { |
1722 | SheepdogAIOCB *acb; | |
2df46246 | 1723 | int ret; |
33b1db1c MK |
1724 | |
1725 | if (bs->growable && sector_num + nb_sectors > bs->total_sectors) { | |
cb595887 MK |
1726 | ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE); |
1727 | if (ret < 0) { | |
1728 | return ret; | |
33b1db1c MK |
1729 | } |
1730 | bs->total_sectors = sector_num + nb_sectors; | |
1731 | } | |
1732 | ||
f700f8e3 | 1733 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
1734 | acb->aio_done_func = sd_write_done; |
1735 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1736 | ||
2df46246 MK |
1737 | ret = sd_co_rw_vector(acb); |
1738 | if (ret <= 0) { | |
1739 | qemu_aio_release(acb); | |
1740 | return ret; | |
1741 | } | |
1742 | ||
1743 | qemu_coroutine_yield(); | |
1744 | ||
1745 | return acb->ret; | |
33b1db1c MK |
1746 | } |
1747 | ||
a968168c | 1748 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1749 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1750 | { |
1751 | SheepdogAIOCB *acb; | |
19db9b90 | 1752 | int ret; |
33b1db1c | 1753 | |
f700f8e3 | 1754 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
1755 | acb->aiocb_type = AIOCB_READ_UDATA; |
1756 | acb->aio_done_func = sd_finish_aiocb; | |
1757 | ||
2df46246 MK |
1758 | ret = sd_co_rw_vector(acb); |
1759 | if (ret <= 0) { | |
1760 | qemu_aio_release(acb); | |
1761 | return ret; | |
1762 | } | |
1763 | ||
1764 | qemu_coroutine_yield(); | |
1765 | ||
1766 | return acb->ret; | |
33b1db1c MK |
1767 | } |
1768 | ||
47622c44 LY |
1769 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
1770 | { | |
1771 | BDRVSheepdogState *s = bs->opaque; | |
47783072 LY |
1772 | SheepdogAIOCB *acb; |
1773 | AIOReq *aio_req; | |
47622c44 | 1774 | int ret; |
47622c44 | 1775 | |
0e7106d8 | 1776 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
1777 | return 0; |
1778 | } | |
1779 | ||
f700f8e3 | 1780 | acb = sd_aio_setup(bs, NULL, 0, 0); |
47783072 LY |
1781 | acb->aiocb_type = AIOCB_FLUSH_CACHE; |
1782 | acb->aio_done_func = sd_finish_aiocb; | |
47622c44 | 1783 | |
47783072 LY |
1784 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
1785 | 0, 0, 0, 0, 0); | |
1786 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
1787 | ret = add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type); | |
1788 | if (ret < 0) { | |
1789 | error_report("add_aio_request is failed"); | |
1790 | free_aio_req(s, aio_req); | |
1791 | qemu_aio_release(acb); | |
47622c44 LY |
1792 | return ret; |
1793 | } | |
1794 | ||
47783072 LY |
1795 | qemu_coroutine_yield(); |
1796 | return acb->ret; | |
47622c44 LY |
1797 | } |
1798 | ||
33b1db1c MK |
1799 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
1800 | { | |
1801 | BDRVSheepdogState *s = bs->opaque; | |
1802 | int ret, fd; | |
1803 | uint32_t new_vid; | |
1804 | SheepdogInode *inode; | |
1805 | unsigned int datalen; | |
1806 | ||
1b6ac998 | 1807 | dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
1808 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
1809 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
1810 | ||
1811 | if (s->is_snapshot) { | |
1812 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 1813 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
1814 | |
1815 | return -EINVAL; | |
1816 | } | |
1817 | ||
1818 | dprintf("%s %s\n", sn_info->name, sn_info->id_str); | |
1819 | ||
1820 | s->inode.vm_state_size = sn_info->vm_state_size; | |
1821 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
1822 | /* It appears that inode.tag does not require a NUL terminator, |
1823 | * which means this use of strncpy is ok. | |
1824 | */ | |
33b1db1c MK |
1825 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
1826 | /* we don't need to update entire object */ | |
1827 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1828 | ||
1829 | /* refresh inode. */ | |
25af257d | 1830 | fd = connect_to_sdog(s); |
33b1db1c | 1831 | if (fd < 0) { |
cb595887 | 1832 | ret = fd; |
33b1db1c MK |
1833 | goto cleanup; |
1834 | } | |
1835 | ||
1836 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
0e7106d8 | 1837 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
33b1db1c | 1838 | if (ret < 0) { |
6daf194d | 1839 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
1840 | goto cleanup; |
1841 | } | |
1842 | ||
25af257d MK |
1843 | ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, |
1844 | 1); | |
33b1db1c | 1845 | if (ret < 0) { |
6daf194d | 1846 | error_report("failed to create inode for snapshot. %s", |
33b1db1c | 1847 | strerror(errno)); |
33b1db1c MK |
1848 | goto cleanup; |
1849 | } | |
1850 | ||
7267c094 | 1851 | inode = (SheepdogInode *)g_malloc(datalen); |
33b1db1c MK |
1852 | |
1853 | ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid), | |
0e7106d8 | 1854 | s->inode.nr_copies, datalen, 0, s->cache_flags); |
33b1db1c MK |
1855 | |
1856 | if (ret < 0) { | |
6daf194d | 1857 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
1858 | goto cleanup; |
1859 | } | |
1860 | ||
1861 | memcpy(&s->inode, inode, datalen); | |
1862 | dprintf("s->inode: name %s snap_id %x oid %x\n", | |
1863 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); | |
1864 | ||
1865 | cleanup: | |
1866 | closesocket(fd); | |
1867 | return ret; | |
1868 | } | |
1869 | ||
1870 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) | |
1871 | { | |
1872 | BDRVSheepdogState *s = bs->opaque; | |
1873 | BDRVSheepdogState *old_s; | |
1874 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1875 | char *buf = NULL; | |
1876 | uint32_t vid; | |
1877 | uint32_t snapid = 0; | |
cb595887 | 1878 | int ret = 0, fd; |
33b1db1c | 1879 | |
7267c094 | 1880 | old_s = g_malloc(sizeof(BDRVSheepdogState)); |
33b1db1c MK |
1881 | |
1882 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
1883 | ||
3178e275 | 1884 | pstrcpy(vdi, sizeof(vdi), s->name); |
33b1db1c | 1885 | |
33b1db1c | 1886 | snapid = strtoul(snapshot_id, NULL, 10); |
3178e275 JM |
1887 | if (snapid) { |
1888 | tag[0] = 0; | |
1889 | } else { | |
1890 | pstrcpy(tag, sizeof(tag), s->name); | |
33b1db1c MK |
1891 | } |
1892 | ||
1893 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1); | |
1894 | if (ret) { | |
6daf194d | 1895 | error_report("Failed to find_vdi_name"); |
33b1db1c MK |
1896 | goto out; |
1897 | } | |
1898 | ||
25af257d | 1899 | fd = connect_to_sdog(s); |
33b1db1c | 1900 | if (fd < 0) { |
cb595887 | 1901 | ret = fd; |
33b1db1c MK |
1902 | goto out; |
1903 | } | |
1904 | ||
7267c094 | 1905 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 1906 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
0e7106d8 | 1907 | SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1908 | |
1909 | closesocket(fd); | |
1910 | ||
1911 | if (ret) { | |
33b1db1c MK |
1912 | goto out; |
1913 | } | |
1914 | ||
1915 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1916 | ||
1917 | if (!s->inode.vm_state_size) { | |
6daf194d | 1918 | error_report("Invalid snapshot"); |
33b1db1c MK |
1919 | ret = -ENOENT; |
1920 | goto out; | |
1921 | } | |
1922 | ||
2f536801 | 1923 | s->is_snapshot = true; |
33b1db1c | 1924 | |
7267c094 AL |
1925 | g_free(buf); |
1926 | g_free(old_s); | |
33b1db1c MK |
1927 | |
1928 | return 0; | |
1929 | out: | |
1930 | /* recover bdrv_sd_state */ | |
1931 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 AL |
1932 | g_free(buf); |
1933 | g_free(old_s); | |
33b1db1c | 1934 | |
6daf194d | 1935 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
1936 | |
1937 | return ret; | |
1938 | } | |
1939 | ||
1940 | static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) | |
1941 | { | |
1942 | /* FIXME: Delete specified snapshot id. */ | |
1943 | return 0; | |
1944 | } | |
1945 | ||
33b1db1c MK |
1946 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
1947 | { | |
1948 | BDRVSheepdogState *s = bs->opaque; | |
1949 | SheepdogReq req; | |
1950 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
1951 | QEMUSnapshotInfo *sn_tab = NULL; | |
1952 | unsigned wlen, rlen; | |
1953 | int found = 0; | |
1954 | static SheepdogInode inode; | |
1955 | unsigned long *vdi_inuse; | |
1956 | unsigned int start_nr; | |
1957 | uint64_t hval; | |
1958 | uint32_t vid; | |
1959 | ||
7267c094 | 1960 | vdi_inuse = g_malloc(max); |
33b1db1c | 1961 | |
25af257d | 1962 | fd = connect_to_sdog(s); |
33b1db1c | 1963 | if (fd < 0) { |
cb595887 | 1964 | ret = fd; |
33b1db1c MK |
1965 | goto out; |
1966 | } | |
1967 | ||
1968 | rlen = max; | |
1969 | wlen = 0; | |
1970 | ||
1971 | memset(&req, 0, sizeof(req)); | |
1972 | ||
1973 | req.opcode = SD_OP_READ_VDIS; | |
1974 | req.data_length = max; | |
1975 | ||
1976 | ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); | |
1977 | ||
1978 | closesocket(fd); | |
1979 | if (ret) { | |
1980 | goto out; | |
1981 | } | |
1982 | ||
7267c094 | 1983 | sn_tab = g_malloc0(nr * sizeof(*sn_tab)); |
33b1db1c MK |
1984 | |
1985 | /* calculate a vdi id with hash function */ | |
1986 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
1987 | start_nr = hval & (SD_NR_VDIS - 1); | |
1988 | ||
25af257d | 1989 | fd = connect_to_sdog(s); |
33b1db1c | 1990 | if (fd < 0) { |
cb595887 | 1991 | ret = fd; |
33b1db1c MK |
1992 | goto out; |
1993 | } | |
1994 | ||
1995 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
1996 | if (!test_bit(vid, vdi_inuse)) { | |
1997 | break; | |
1998 | } | |
1999 | ||
2000 | /* we don't need to read entire object */ | |
2001 | ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid), | |
47622c44 | 2002 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2003 | s->cache_flags); |
33b1db1c MK |
2004 | |
2005 | if (ret) { | |
2006 | continue; | |
2007 | } | |
2008 | ||
2009 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2010 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2011 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2012 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2013 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2014 | ||
2015 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", | |
2016 | inode.snap_id); | |
3178e275 JM |
2017 | pstrcpy(sn_tab[found].name, |
2018 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2019 | inode.tag); | |
33b1db1c MK |
2020 | found++; |
2021 | } | |
2022 | } | |
2023 | ||
2024 | closesocket(fd); | |
2025 | out: | |
2026 | *psn_tab = sn_tab; | |
2027 | ||
7267c094 | 2028 | g_free(vdi_inuse); |
33b1db1c | 2029 | |
cb595887 MK |
2030 | if (ret < 0) { |
2031 | return ret; | |
2032 | } | |
2033 | ||
33b1db1c MK |
2034 | return found; |
2035 | } | |
2036 | ||
2037 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2038 | int64_t pos, int size, int load) | |
2039 | { | |
2f536801 MK |
2040 | bool create; |
2041 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2042 | unsigned int data_len; |
2043 | uint64_t vmstate_oid; | |
2044 | uint32_t vdi_index; | |
2045 | uint64_t offset; | |
2046 | ||
25af257d | 2047 | fd = connect_to_sdog(s); |
33b1db1c | 2048 | if (fd < 0) { |
cb595887 | 2049 | return fd; |
33b1db1c MK |
2050 | } |
2051 | ||
6f3c714e | 2052 | while (remaining) { |
33b1db1c MK |
2053 | vdi_index = pos / SD_DATA_OBJ_SIZE; |
2054 | offset = pos % SD_DATA_OBJ_SIZE; | |
2055 | ||
1f7a48de | 2056 | data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset); |
33b1db1c MK |
2057 | |
2058 | vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index); | |
2059 | ||
2060 | create = (offset == 0); | |
2061 | if (load) { | |
2062 | ret = read_object(fd, (char *)data, vmstate_oid, | |
47622c44 | 2063 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2064 | s->cache_flags); |
33b1db1c MK |
2065 | } else { |
2066 | ret = write_object(fd, (char *)data, vmstate_oid, | |
47622c44 | 2067 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2068 | s->cache_flags); |
33b1db1c MK |
2069 | } |
2070 | ||
2071 | if (ret < 0) { | |
6daf194d | 2072 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2073 | goto cleanup; |
2074 | } | |
2075 | ||
2076 | pos += data_len; | |
1f7a48de | 2077 | data += data_len; |
6f3c714e | 2078 | remaining -= data_len; |
33b1db1c | 2079 | } |
6f3c714e | 2080 | ret = size; |
33b1db1c MK |
2081 | cleanup: |
2082 | closesocket(fd); | |
2083 | return ret; | |
2084 | } | |
2085 | ||
cf8074b3 KW |
2086 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2087 | int64_t pos) | |
33b1db1c MK |
2088 | { |
2089 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2090 | void *buf; |
2091 | int ret; | |
33b1db1c | 2092 | |
cf8074b3 KW |
2093 | buf = qemu_blockalign(bs, qiov->size); |
2094 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2095 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2096 | qemu_vfree(buf); | |
2097 | ||
2098 | return ret; | |
33b1db1c MK |
2099 | } |
2100 | ||
2101 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
2102 | int64_t pos, int size) | |
2103 | { | |
2104 | BDRVSheepdogState *s = bs->opaque; | |
2105 | ||
2106 | return do_load_save_vmstate(s, data, pos, size, 1); | |
2107 | } | |
2108 | ||
2109 | ||
2110 | static QEMUOptionParameter sd_create_options[] = { | |
2111 | { | |
2112 | .name = BLOCK_OPT_SIZE, | |
2113 | .type = OPT_SIZE, | |
2114 | .help = "Virtual disk size" | |
2115 | }, | |
2116 | { | |
2117 | .name = BLOCK_OPT_BACKING_FILE, | |
2118 | .type = OPT_STRING, | |
2119 | .help = "File name of a base image" | |
2120 | }, | |
a8e0fdd7 MK |
2121 | { |
2122 | .name = BLOCK_OPT_PREALLOC, | |
2123 | .type = OPT_STRING, | |
2124 | .help = "Preallocation mode (allowed values: off, full)" | |
2125 | }, | |
33b1db1c MK |
2126 | { NULL } |
2127 | }; | |
2128 | ||
5d6768e3 | 2129 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2130 | .format_name = "sheepdog", |
2131 | .protocol_name = "sheepdog", | |
2132 | .instance_size = sizeof(BDRVSheepdogState), | |
2133 | .bdrv_file_open = sd_open, | |
2134 | .bdrv_close = sd_close, | |
2135 | .bdrv_create = sd_create, | |
2136 | .bdrv_getlength = sd_getlength, | |
2137 | .bdrv_truncate = sd_truncate, | |
2138 | ||
2df46246 MK |
2139 | .bdrv_co_readv = sd_co_readv, |
2140 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2141 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
33b1db1c MK |
2142 | |
2143 | .bdrv_snapshot_create = sd_snapshot_create, | |
2144 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2145 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2146 | .bdrv_snapshot_list = sd_snapshot_list, | |
2147 | ||
2148 | .bdrv_save_vmstate = sd_save_vmstate, | |
2149 | .bdrv_load_vmstate = sd_load_vmstate, | |
2150 | ||
2151 | .create_options = sd_create_options, | |
2152 | }; | |
2153 | ||
5d6768e3 MK |
2154 | static BlockDriver bdrv_sheepdog_tcp = { |
2155 | .format_name = "sheepdog", | |
2156 | .protocol_name = "sheepdog+tcp", | |
2157 | .instance_size = sizeof(BDRVSheepdogState), | |
2158 | .bdrv_file_open = sd_open, | |
2159 | .bdrv_close = sd_close, | |
2160 | .bdrv_create = sd_create, | |
2161 | .bdrv_getlength = sd_getlength, | |
2162 | .bdrv_truncate = sd_truncate, | |
2163 | ||
2164 | .bdrv_co_readv = sd_co_readv, | |
2165 | .bdrv_co_writev = sd_co_writev, | |
2166 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
2167 | ||
2168 | .bdrv_snapshot_create = sd_snapshot_create, | |
2169 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2170 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2171 | .bdrv_snapshot_list = sd_snapshot_list, | |
2172 | ||
2173 | .bdrv_save_vmstate = sd_save_vmstate, | |
2174 | .bdrv_load_vmstate = sd_load_vmstate, | |
2175 | ||
2176 | .create_options = sd_create_options, | |
2177 | }; | |
2178 | ||
1b8bbb46 MK |
2179 | static BlockDriver bdrv_sheepdog_unix = { |
2180 | .format_name = "sheepdog", | |
2181 | .protocol_name = "sheepdog+unix", | |
2182 | .instance_size = sizeof(BDRVSheepdogState), | |
2183 | .bdrv_file_open = sd_open, | |
2184 | .bdrv_close = sd_close, | |
2185 | .bdrv_create = sd_create, | |
2186 | .bdrv_getlength = sd_getlength, | |
2187 | .bdrv_truncate = sd_truncate, | |
2188 | ||
2189 | .bdrv_co_readv = sd_co_readv, | |
2190 | .bdrv_co_writev = sd_co_writev, | |
2191 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
2192 | ||
2193 | .bdrv_snapshot_create = sd_snapshot_create, | |
2194 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2195 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2196 | .bdrv_snapshot_list = sd_snapshot_list, | |
2197 | ||
2198 | .bdrv_save_vmstate = sd_save_vmstate, | |
2199 | .bdrv_load_vmstate = sd_load_vmstate, | |
2200 | ||
2201 | .create_options = sd_create_options, | |
2202 | }; | |
2203 | ||
33b1db1c MK |
2204 | static void bdrv_sheepdog_init(void) |
2205 | { | |
2206 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 2207 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 2208 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
2209 | } |
2210 | block_init(bdrv_sheepdog_init); |