]>
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 | |
cac8f4a6 LY |
30 | /* 0x04 is used internally by Sheepdog */ |
31 | #define SD_OP_DISCARD_OBJ 0x05 | |
33b1db1c MK |
32 | |
33 | #define SD_OP_NEW_VDI 0x11 | |
34 | #define SD_OP_LOCK_VDI 0x12 | |
35 | #define SD_OP_RELEASE_VDI 0x13 | |
36 | #define SD_OP_GET_VDI_INFO 0x14 | |
37 | #define SD_OP_READ_VDIS 0x15 | |
47622c44 | 38 | #define SD_OP_FLUSH_VDI 0x16 |
859e5553 | 39 | #define SD_OP_DEL_VDI 0x17 |
33b1db1c MK |
40 | |
41 | #define SD_FLAG_CMD_WRITE 0x01 | |
42 | #define SD_FLAG_CMD_COW 0x02 | |
0e7106d8 LY |
43 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
44 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
33b1db1c MK |
45 | |
46 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
47 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
48 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
49 | #define SD_RES_EIO 0x03 /* I/O error */ | |
50 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
51 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
52 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
53 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
54 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
55 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
56 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
57 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
58 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
59 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
60 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
61 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
62 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
63 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
64 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
65 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
66 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
67 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
68 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
69 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
70 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
fca23f0a | 71 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ |
6a0b5490 | 72 | #define SD_RES_READONLY 0x1A /* Object is read-only */ |
33b1db1c MK |
73 | |
74 | /* | |
75 | * Object ID rules | |
76 | * | |
77 | * 0 - 19 (20 bits): data object space | |
78 | * 20 - 31 (12 bits): reserved data object space | |
79 | * 32 - 55 (24 bits): vdi object space | |
80 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 81 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
82 | */ |
83 | ||
84 | #define VDI_SPACE_SHIFT 32 | |
85 | #define VDI_BIT (UINT64_C(1) << 63) | |
86 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
87 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
88 | #define MAX_CHILDREN 1024 | |
89 | #define SD_MAX_VDI_LEN 256 | |
90 | #define SD_MAX_VDI_TAG_LEN 256 | |
91 | #define SD_NR_VDIS (1U << 24) | |
92 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
93 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
33b1db1c MK |
94 | |
95 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
96 | #define CURRENT_VDI_ID 0 | |
97 | ||
98 | typedef struct SheepdogReq { | |
99 | uint8_t proto_ver; | |
100 | uint8_t opcode; | |
101 | uint16_t flags; | |
102 | uint32_t epoch; | |
103 | uint32_t id; | |
104 | uint32_t data_length; | |
105 | uint32_t opcode_specific[8]; | |
106 | } SheepdogReq; | |
107 | ||
108 | typedef struct SheepdogRsp { | |
109 | uint8_t proto_ver; | |
110 | uint8_t opcode; | |
111 | uint16_t flags; | |
112 | uint32_t epoch; | |
113 | uint32_t id; | |
114 | uint32_t data_length; | |
115 | uint32_t result; | |
116 | uint32_t opcode_specific[7]; | |
117 | } SheepdogRsp; | |
118 | ||
119 | typedef struct SheepdogObjReq { | |
120 | uint8_t proto_ver; | |
121 | uint8_t opcode; | |
122 | uint16_t flags; | |
123 | uint32_t epoch; | |
124 | uint32_t id; | |
125 | uint32_t data_length; | |
126 | uint64_t oid; | |
127 | uint64_t cow_oid; | |
29a67f7e | 128 | uint8_t copies; |
1841f880 LY |
129 | uint8_t copy_policy; |
130 | uint8_t reserved[6]; | |
33b1db1c MK |
131 | uint64_t offset; |
132 | } SheepdogObjReq; | |
133 | ||
134 | typedef struct SheepdogObjRsp { | |
135 | uint8_t proto_ver; | |
136 | uint8_t opcode; | |
137 | uint16_t flags; | |
138 | uint32_t epoch; | |
139 | uint32_t id; | |
140 | uint32_t data_length; | |
141 | uint32_t result; | |
29a67f7e | 142 | uint8_t copies; |
1841f880 LY |
143 | uint8_t copy_policy; |
144 | uint8_t reserved[2]; | |
33b1db1c MK |
145 | uint32_t pad[6]; |
146 | } SheepdogObjRsp; | |
147 | ||
148 | typedef struct SheepdogVdiReq { | |
149 | uint8_t proto_ver; | |
150 | uint8_t opcode; | |
151 | uint16_t flags; | |
152 | uint32_t epoch; | |
153 | uint32_t id; | |
154 | uint32_t data_length; | |
155 | uint64_t vdi_size; | |
6f74c260 | 156 | uint32_t vdi_id; |
29a67f7e | 157 | uint8_t copies; |
1841f880 LY |
158 | uint8_t copy_policy; |
159 | uint8_t reserved[2]; | |
33b1db1c MK |
160 | uint32_t snapid; |
161 | uint32_t pad[3]; | |
162 | } SheepdogVdiReq; | |
163 | ||
164 | typedef struct SheepdogVdiRsp { | |
165 | uint8_t proto_ver; | |
166 | uint8_t opcode; | |
167 | uint16_t flags; | |
168 | uint32_t epoch; | |
169 | uint32_t id; | |
170 | uint32_t data_length; | |
171 | uint32_t result; | |
172 | uint32_t rsvd; | |
173 | uint32_t vdi_id; | |
174 | uint32_t pad[5]; | |
175 | } SheepdogVdiRsp; | |
176 | ||
177 | typedef struct SheepdogInode { | |
178 | char name[SD_MAX_VDI_LEN]; | |
179 | char tag[SD_MAX_VDI_TAG_LEN]; | |
180 | uint64_t ctime; | |
181 | uint64_t snap_ctime; | |
182 | uint64_t vm_clock_nsec; | |
183 | uint64_t vdi_size; | |
184 | uint64_t vm_state_size; | |
185 | uint16_t copy_policy; | |
186 | uint8_t nr_copies; | |
187 | uint8_t block_size_shift; | |
188 | uint32_t snap_id; | |
189 | uint32_t vdi_id; | |
190 | uint32_t parent_vdi_id; | |
191 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
192 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
193 | } SheepdogInode; | |
194 | ||
195 | /* | |
196 | * 64 bit FNV-1a non-zero initial basis | |
197 | */ | |
198 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
199 | ||
200 | /* | |
201 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
202 | */ | |
203 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
204 | { | |
205 | unsigned char *bp = buf; | |
206 | unsigned char *be = bp + len; | |
207 | while (bp < be) { | |
208 | hval ^= (uint64_t) *bp++; | |
209 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
210 | (hval << 7) + (hval << 8) + (hval << 40); | |
211 | } | |
212 | return hval; | |
213 | } | |
214 | ||
2f536801 | 215 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
216 | { |
217 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
218 | } | |
219 | ||
2f536801 | 220 | static inline bool is_data_obj(uint64_t oid) |
33b1db1c MK |
221 | { |
222 | return !(VDI_BIT & oid); | |
223 | } | |
224 | ||
225 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
226 | { | |
227 | return oid & (MAX_DATA_OBJS - 1); | |
228 | } | |
229 | ||
72e0996c MK |
230 | static inline uint32_t oid_to_vid(uint64_t oid) |
231 | { | |
232 | return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT; | |
233 | } | |
234 | ||
33b1db1c MK |
235 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) |
236 | { | |
237 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
238 | } | |
239 | ||
240 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
241 | { | |
242 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
243 | } | |
244 | ||
245 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
246 | { | |
247 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
248 | } | |
249 | ||
2f536801 | 250 | static inline bool is_snapshot(struct SheepdogInode *inode) |
33b1db1c MK |
251 | { |
252 | return !!inode->snap_ctime; | |
253 | } | |
254 | ||
2440a2c3 | 255 | #undef DPRINTF |
33b1db1c | 256 | #ifdef DEBUG_SDOG |
2440a2c3 | 257 | #define DPRINTF(fmt, args...) \ |
33b1db1c MK |
258 | do { \ |
259 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
260 | } while (0) | |
261 | #else | |
2440a2c3 | 262 | #define DPRINTF(fmt, args...) |
33b1db1c MK |
263 | #endif |
264 | ||
265 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
266 | ||
267 | typedef struct AIOReq { | |
268 | SheepdogAIOCB *aiocb; | |
269 | unsigned int iov_offset; | |
270 | ||
271 | uint64_t oid; | |
272 | uint64_t base_oid; | |
273 | uint64_t offset; | |
274 | unsigned int data_len; | |
275 | uint8_t flags; | |
276 | uint32_t id; | |
277 | ||
c292ee6a | 278 | QLIST_ENTRY(AIOReq) aio_siblings; |
33b1db1c MK |
279 | } AIOReq; |
280 | ||
281 | enum AIOCBState { | |
282 | AIOCB_WRITE_UDATA, | |
283 | AIOCB_READ_UDATA, | |
47783072 | 284 | AIOCB_FLUSH_CACHE, |
cac8f4a6 | 285 | AIOCB_DISCARD_OBJ, |
33b1db1c MK |
286 | }; |
287 | ||
288 | struct SheepdogAIOCB { | |
289 | BlockDriverAIOCB common; | |
290 | ||
291 | QEMUIOVector *qiov; | |
292 | ||
293 | int64_t sector_num; | |
294 | int nb_sectors; | |
295 | ||
296 | int ret; | |
297 | enum AIOCBState aiocb_type; | |
298 | ||
2df46246 | 299 | Coroutine *coroutine; |
33b1db1c MK |
300 | void (*aio_done_func)(SheepdogAIOCB *); |
301 | ||
2f536801 | 302 | bool canceled; |
1d732d7d | 303 | int nr_pending; |
33b1db1c MK |
304 | }; |
305 | ||
306 | typedef struct BDRVSheepdogState { | |
011603ca MK |
307 | BlockDriverState *bs; |
308 | ||
33b1db1c MK |
309 | SheepdogInode inode; |
310 | ||
311 | uint32_t min_dirty_data_idx; | |
312 | uint32_t max_dirty_data_idx; | |
313 | ||
314 | char name[SD_MAX_VDI_LEN]; | |
2f536801 | 315 | bool is_snapshot; |
0e7106d8 | 316 | uint32_t cache_flags; |
cac8f4a6 | 317 | bool discard_supported; |
33b1db1c | 318 | |
25af257d | 319 | char *host_spec; |
1b8bbb46 | 320 | bool is_unix; |
33b1db1c MK |
321 | int fd; |
322 | ||
2df46246 MK |
323 | CoMutex lock; |
324 | Coroutine *co_send; | |
325 | Coroutine *co_recv; | |
326 | ||
33b1db1c | 327 | uint32_t aioreq_seq_num; |
011603ca MK |
328 | |
329 | /* Every aio request must be linked to either of these queues. */ | |
c292ee6a MK |
330 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
331 | QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head; | |
011603ca | 332 | QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head; |
33b1db1c MK |
333 | } BDRVSheepdogState; |
334 | ||
335 | static const char * sd_strerror(int err) | |
336 | { | |
337 | int i; | |
338 | ||
339 | static const struct { | |
340 | int err; | |
341 | const char *desc; | |
342 | } errors[] = { | |
343 | {SD_RES_SUCCESS, "Success"}, | |
344 | {SD_RES_UNKNOWN, "Unknown error"}, | |
345 | {SD_RES_NO_OBJ, "No object found"}, | |
346 | {SD_RES_EIO, "I/O error"}, | |
347 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
348 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
349 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
350 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
351 | {SD_RES_NO_VDI, "No vdi found"}, | |
352 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
353 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
354 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
355 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
356 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
357 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
358 | {SD_RES_STARTUP, "The system is still booting"}, | |
359 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
360 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
361 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
362 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
363 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
364 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
365 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
366 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
367 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
fca23f0a | 368 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, |
6a0b5490 | 369 | {SD_RES_READONLY, "Object is read-only"}, |
33b1db1c MK |
370 | }; |
371 | ||
372 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
373 | if (errors[i].err == err) { | |
374 | return errors[i].desc; | |
375 | } | |
376 | } | |
377 | ||
378 | return "Invalid error code"; | |
379 | } | |
380 | ||
381 | /* | |
382 | * Sheepdog I/O handling: | |
383 | * | |
2df46246 | 384 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
c292ee6a | 385 | * link the requests to the inflight_list in the |
2df46246 MK |
386 | * BDRVSheepdogState. The function exits without waiting for |
387 | * receiving the response. | |
33b1db1c | 388 | * |
2df46246 | 389 | * 2. We receive the response in aio_read_response, the fd handler to |
33b1db1c MK |
390 | * the sheepdog connection. If metadata update is needed, we send |
391 | * the write request to the vdi object in sd_write_done, the write | |
2df46246 MK |
392 | * completion function. We switch back to sd_co_readv/writev after |
393 | * all the requests belonging to the AIOCB are finished. | |
33b1db1c MK |
394 | */ |
395 | ||
396 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
397 | uint64_t oid, unsigned int data_len, | |
398 | uint64_t offset, uint8_t flags, | |
399 | uint64_t base_oid, unsigned int iov_offset) | |
400 | { | |
401 | AIOReq *aio_req; | |
402 | ||
7267c094 | 403 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
404 | aio_req->aiocb = acb; |
405 | aio_req->iov_offset = iov_offset; | |
406 | aio_req->oid = oid; | |
407 | aio_req->base_oid = base_oid; | |
408 | aio_req->offset = offset; | |
409 | aio_req->data_len = data_len; | |
410 | aio_req->flags = flags; | |
411 | aio_req->id = s->aioreq_seq_num++; | |
412 | ||
1d732d7d | 413 | acb->nr_pending++; |
33b1db1c MK |
414 | return aio_req; |
415 | } | |
416 | ||
1d732d7d | 417 | static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
33b1db1c MK |
418 | { |
419 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1d732d7d | 420 | |
c292ee6a | 421 | QLIST_REMOVE(aio_req, aio_siblings); |
7267c094 | 422 | g_free(aio_req); |
33b1db1c | 423 | |
1d732d7d | 424 | acb->nr_pending--; |
33b1db1c MK |
425 | } |
426 | ||
d8716b41 | 427 | static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
33b1db1c MK |
428 | { |
429 | if (!acb->canceled) { | |
2df46246 | 430 | qemu_coroutine_enter(acb->coroutine, NULL); |
33b1db1c MK |
431 | } |
432 | qemu_aio_release(acb); | |
433 | } | |
434 | ||
435 | static void sd_aio_cancel(BlockDriverAIOCB *blockacb) | |
436 | { | |
437 | SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; | |
438 | ||
439 | /* | |
440 | * Sheepdog cannot cancel the requests which are already sent to | |
441 | * the servers, so we just complete the request with -EIO here. | |
442 | */ | |
2df46246 MK |
443 | acb->ret = -EIO; |
444 | qemu_coroutine_enter(acb->coroutine, NULL); | |
2f536801 | 445 | acb->canceled = true; |
33b1db1c MK |
446 | } |
447 | ||
d7331bed | 448 | static const AIOCBInfo sd_aiocb_info = { |
33b1db1c MK |
449 | .aiocb_size = sizeof(SheepdogAIOCB), |
450 | .cancel = sd_aio_cancel, | |
451 | }; | |
452 | ||
453 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
f700f8e3 | 454 | int64_t sector_num, int nb_sectors) |
33b1db1c MK |
455 | { |
456 | SheepdogAIOCB *acb; | |
457 | ||
f700f8e3 | 458 | acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); |
33b1db1c MK |
459 | |
460 | acb->qiov = qiov; | |
461 | ||
462 | acb->sector_num = sector_num; | |
463 | acb->nb_sectors = nb_sectors; | |
464 | ||
465 | acb->aio_done_func = NULL; | |
2f536801 | 466 | acb->canceled = false; |
2df46246 | 467 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c | 468 | acb->ret = 0; |
1d732d7d | 469 | acb->nr_pending = 0; |
33b1db1c MK |
470 | return acb; |
471 | } | |
472 | ||
25af257d | 473 | static int connect_to_sdog(BDRVSheepdogState *s) |
33b1db1c | 474 | { |
25af257d MK |
475 | int fd; |
476 | Error *err = NULL; | |
33b1db1c | 477 | |
1b8bbb46 MK |
478 | if (s->is_unix) { |
479 | fd = unix_connect(s->host_spec, &err); | |
480 | } else { | |
481 | fd = inet_connect(s->host_spec, &err); | |
482 | ||
483 | if (err == NULL) { | |
484 | int ret = socket_set_nodelay(fd); | |
485 | if (ret < 0) { | |
486 | error_report("%s", strerror(errno)); | |
487 | } | |
488 | } | |
489 | } | |
33b1db1c | 490 | |
25af257d MK |
491 | if (err != NULL) { |
492 | qerror_report_err(err); | |
493 | error_free(err); | |
0d6db300 | 494 | } else { |
f9e8cacc | 495 | qemu_set_nonblock(fd); |
33b1db1c MK |
496 | } |
497 | ||
33b1db1c MK |
498 | return fd; |
499 | } | |
500 | ||
e0d93a89 MK |
501 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
502 | unsigned int *wlen) | |
47622c44 LY |
503 | { |
504 | int ret; | |
505 | ||
506 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
80731d9d | 507 | if (ret != sizeof(*hdr)) { |
47622c44 | 508 | error_report("failed to send a req, %s", strerror(errno)); |
eb092180 | 509 | return ret; |
47622c44 LY |
510 | } |
511 | ||
512 | ret = qemu_co_send(sockfd, data, *wlen); | |
80731d9d | 513 | if (ret != *wlen) { |
47622c44 LY |
514 | error_report("failed to send a req, %s", strerror(errno)); |
515 | } | |
516 | ||
517 | return ret; | |
518 | } | |
e0d93a89 | 519 | |
2dfcca3b MK |
520 | static void restart_co_req(void *opaque) |
521 | { | |
522 | Coroutine *co = opaque; | |
523 | ||
524 | qemu_coroutine_enter(co, NULL); | |
525 | } | |
526 | ||
cddd4ac7 MK |
527 | typedef struct SheepdogReqCo { |
528 | int sockfd; | |
529 | SheepdogReq *hdr; | |
530 | void *data; | |
531 | unsigned int *wlen; | |
532 | unsigned int *rlen; | |
533 | int ret; | |
534 | bool finished; | |
535 | } SheepdogReqCo; | |
536 | ||
537 | static coroutine_fn void do_co_req(void *opaque) | |
47622c44 LY |
538 | { |
539 | int ret; | |
2dfcca3b | 540 | Coroutine *co; |
cddd4ac7 MK |
541 | SheepdogReqCo *srco = opaque; |
542 | int sockfd = srco->sockfd; | |
543 | SheepdogReq *hdr = srco->hdr; | |
544 | void *data = srco->data; | |
545 | unsigned int *wlen = srco->wlen; | |
546 | unsigned int *rlen = srco->rlen; | |
2dfcca3b MK |
547 | |
548 | co = qemu_coroutine_self(); | |
f2e5dca4 | 549 | qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, co); |
47622c44 | 550 | |
47622c44 LY |
551 | ret = send_co_req(sockfd, hdr, data, wlen); |
552 | if (ret < 0) { | |
553 | goto out; | |
554 | } | |
555 | ||
f2e5dca4 | 556 | qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, co); |
2dfcca3b | 557 | |
47622c44 | 558 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); |
80731d9d | 559 | if (ret != sizeof(*hdr)) { |
47622c44 | 560 | error_report("failed to get a rsp, %s", strerror(errno)); |
cb595887 | 561 | ret = -errno; |
47622c44 LY |
562 | goto out; |
563 | } | |
564 | ||
565 | if (*rlen > hdr->data_length) { | |
566 | *rlen = hdr->data_length; | |
567 | } | |
568 | ||
569 | if (*rlen) { | |
570 | ret = qemu_co_recv(sockfd, data, *rlen); | |
80731d9d | 571 | if (ret != *rlen) { |
47622c44 | 572 | error_report("failed to get the data, %s", strerror(errno)); |
cb595887 | 573 | ret = -errno; |
47622c44 LY |
574 | goto out; |
575 | } | |
576 | } | |
577 | ret = 0; | |
578 | out: | |
ed9ba724 MK |
579 | /* there is at most one request for this sockfd, so it is safe to |
580 | * set each handler to NULL. */ | |
f2e5dca4 | 581 | qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL); |
cddd4ac7 MK |
582 | |
583 | srco->ret = ret; | |
584 | srco->finished = true; | |
585 | } | |
586 | ||
587 | static int do_req(int sockfd, SheepdogReq *hdr, void *data, | |
588 | unsigned int *wlen, unsigned int *rlen) | |
589 | { | |
590 | Coroutine *co; | |
591 | SheepdogReqCo srco = { | |
592 | .sockfd = sockfd, | |
593 | .hdr = hdr, | |
594 | .data = data, | |
595 | .wlen = wlen, | |
596 | .rlen = rlen, | |
597 | .ret = 0, | |
598 | .finished = false, | |
599 | }; | |
600 | ||
601 | if (qemu_in_coroutine()) { | |
602 | do_co_req(&srco); | |
603 | } else { | |
604 | co = qemu_coroutine_create(do_co_req); | |
605 | qemu_coroutine_enter(co, &srco); | |
606 | while (!srco.finished) { | |
607 | qemu_aio_wait(); | |
608 | } | |
609 | } | |
610 | ||
611 | return srco.ret; | |
47622c44 LY |
612 | } |
613 | ||
d8716b41 | 614 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
2f536801 | 615 | struct iovec *iov, int niov, bool create, |
33b1db1c | 616 | enum AIOCBState aiocb_type); |
13c31de2 | 617 | static int coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req); |
72e0996c | 618 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag); |
011603ca MK |
619 | static int get_sheep_fd(BDRVSheepdogState *s); |
620 | static void co_write_request(void *opaque); | |
7dc1cde0 MK |
621 | |
622 | static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid) | |
623 | { | |
624 | AIOReq *aio_req; | |
625 | ||
626 | QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) { | |
627 | if (aio_req->oid == oid) { | |
628 | return aio_req; | |
629 | } | |
630 | } | |
631 | ||
632 | return NULL; | |
633 | } | |
634 | ||
33b1db1c MK |
635 | /* |
636 | * This function searchs pending requests to the object `oid', and | |
637 | * sends them. | |
638 | */ | |
c292ee6a | 639 | static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid) |
33b1db1c | 640 | { |
7dc1cde0 | 641 | AIOReq *aio_req; |
33b1db1c MK |
642 | SheepdogAIOCB *acb; |
643 | int ret; | |
644 | ||
7dc1cde0 | 645 | while ((aio_req = find_pending_req(s, oid)) != NULL) { |
33b1db1c | 646 | acb = aio_req->aiocb; |
c292ee6a MK |
647 | /* move aio_req from pending list to inflight one */ |
648 | QLIST_REMOVE(aio_req, aio_siblings); | |
649 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
33b1db1c | 650 | ret = add_aio_request(s, aio_req, acb->qiov->iov, |
2f536801 | 651 | acb->qiov->niov, false, acb->aiocb_type); |
33b1db1c | 652 | if (ret < 0) { |
6daf194d | 653 | error_report("add_aio_request is failed"); |
33b1db1c | 654 | free_aio_req(s, aio_req); |
1d732d7d | 655 | if (!acb->nr_pending) { |
33b1db1c MK |
656 | sd_finish_aiocb(acb); |
657 | } | |
658 | } | |
659 | } | |
660 | } | |
661 | ||
011603ca MK |
662 | static coroutine_fn void reconnect_to_sdog(void *opaque) |
663 | { | |
664 | BDRVSheepdogState *s = opaque; | |
665 | AIOReq *aio_req, *next; | |
666 | ||
667 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL); | |
668 | close(s->fd); | |
669 | s->fd = -1; | |
670 | ||
671 | /* Wait for outstanding write requests to be completed. */ | |
672 | while (s->co_send != NULL) { | |
673 | co_write_request(opaque); | |
674 | } | |
675 | ||
676 | /* Try to reconnect the sheepdog server every one second. */ | |
677 | while (s->fd < 0) { | |
678 | s->fd = get_sheep_fd(s); | |
679 | if (s->fd < 0) { | |
680 | DPRINTF("Wait for connection to be established\n"); | |
681 | co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME, | |
682 | 1000000000ULL); | |
683 | } | |
684 | }; | |
685 | ||
686 | /* | |
687 | * Now we have to resend all the request in the inflight queue. However, | |
688 | * resend_aioreq() can yield and newly created requests can be added to the | |
689 | * inflight queue before the coroutine is resumed. To avoid mixing them, we | |
690 | * have to move all the inflight requests to the failed queue before | |
691 | * resend_aioreq() is called. | |
692 | */ | |
693 | QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) { | |
694 | QLIST_REMOVE(aio_req, aio_siblings); | |
695 | QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings); | |
696 | } | |
697 | ||
698 | /* Resend all the failed aio requests. */ | |
699 | while (!QLIST_EMPTY(&s->failed_aio_head)) { | |
700 | aio_req = QLIST_FIRST(&s->failed_aio_head); | |
701 | QLIST_REMOVE(aio_req, aio_siblings); | |
702 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
703 | resend_aioreq(s, aio_req); | |
704 | } | |
705 | } | |
706 | ||
33b1db1c MK |
707 | /* |
708 | * Receive responses of the I/O requests. | |
709 | * | |
710 | * This function is registered as a fd handler, and called from the | |
711 | * main loop when s->fd is ready for reading responses. | |
712 | */ | |
d8716b41 | 713 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
714 | { |
715 | SheepdogObjRsp rsp; | |
716 | BDRVSheepdogState *s = opaque; | |
717 | int fd = s->fd; | |
718 | int ret; | |
719 | AIOReq *aio_req = NULL; | |
720 | SheepdogAIOCB *acb; | |
cac8f4a6 | 721 | uint64_t idx; |
33b1db1c | 722 | |
33b1db1c | 723 | /* read a header */ |
8c5135f9 | 724 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
80731d9d | 725 | if (ret != sizeof(rsp)) { |
6daf194d | 726 | error_report("failed to get the header, %s", strerror(errno)); |
011603ca | 727 | goto err; |
33b1db1c MK |
728 | } |
729 | ||
c292ee6a MK |
730 | /* find the right aio_req from the inflight aio list */ |
731 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
33b1db1c MK |
732 | if (aio_req->id == rsp.id) { |
733 | break; | |
734 | } | |
735 | } | |
736 | if (!aio_req) { | |
6daf194d | 737 | error_report("cannot find aio_req %x", rsp.id); |
011603ca | 738 | goto err; |
33b1db1c MK |
739 | } |
740 | ||
741 | acb = aio_req->aiocb; | |
742 | ||
743 | switch (acb->aiocb_type) { | |
744 | case AIOCB_WRITE_UDATA: | |
6d1acda8 MK |
745 | /* this coroutine context is no longer suitable for co_recv |
746 | * because we may send data to update vdi objects */ | |
747 | s->co_recv = NULL; | |
33b1db1c MK |
748 | if (!is_data_obj(aio_req->oid)) { |
749 | break; | |
750 | } | |
751 | idx = data_oid_to_idx(aio_req->oid); | |
752 | ||
753 | if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) { | |
754 | /* | |
755 | * If the object is newly created one, we need to update | |
756 | * the vdi object (metadata object). min_dirty_data_idx | |
757 | * and max_dirty_data_idx are changed to include updated | |
758 | * index between them. | |
759 | */ | |
bd751f22 LY |
760 | if (rsp.result == SD_RES_SUCCESS) { |
761 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
762 | s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); | |
763 | s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); | |
764 | } | |
33b1db1c MK |
765 | /* |
766 | * Some requests may be blocked because simultaneous | |
767 | * create requests are not allowed, so we search the | |
768 | * pending requests here. | |
769 | */ | |
d6b1ef89 | 770 | send_pending_req(s, aio_req->oid); |
33b1db1c MK |
771 | } |
772 | break; | |
773 | case AIOCB_READ_UDATA: | |
2fc8ae1d MT |
774 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
775 | aio_req->iov_offset, rsp.data_length); | |
80731d9d | 776 | if (ret != rsp.data_length) { |
6daf194d | 777 | error_report("failed to get the data, %s", strerror(errno)); |
011603ca | 778 | goto err; |
33b1db1c MK |
779 | } |
780 | break; | |
47783072 LY |
781 | case AIOCB_FLUSH_CACHE: |
782 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
2440a2c3 | 783 | DPRINTF("disable cache since the server doesn't support it\n"); |
47783072 LY |
784 | s->cache_flags = SD_FLAG_CMD_DIRECT; |
785 | rsp.result = SD_RES_SUCCESS; | |
786 | } | |
787 | break; | |
cac8f4a6 LY |
788 | case AIOCB_DISCARD_OBJ: |
789 | switch (rsp.result) { | |
790 | case SD_RES_INVALID_PARMS: | |
791 | error_report("sheep(%s) doesn't support discard command", | |
792 | s->host_spec); | |
793 | rsp.result = SD_RES_SUCCESS; | |
794 | s->discard_supported = false; | |
795 | break; | |
796 | case SD_RES_SUCCESS: | |
797 | idx = data_oid_to_idx(aio_req->oid); | |
798 | s->inode.data_vdi_id[idx] = 0; | |
799 | break; | |
800 | default: | |
801 | break; | |
802 | } | |
33b1db1c MK |
803 | } |
804 | ||
13c31de2 MK |
805 | switch (rsp.result) { |
806 | case SD_RES_SUCCESS: | |
807 | break; | |
808 | case SD_RES_READONLY: | |
72e0996c MK |
809 | if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) { |
810 | ret = reload_inode(s, 0, ""); | |
811 | if (ret < 0) { | |
011603ca | 812 | goto err; |
72e0996c MK |
813 | } |
814 | } | |
72e0996c MK |
815 | if (is_data_obj(aio_req->oid)) { |
816 | aio_req->oid = vid_to_data_oid(s->inode.vdi_id, | |
817 | data_oid_to_idx(aio_req->oid)); | |
818 | } else { | |
819 | aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id); | |
820 | } | |
13c31de2 MK |
821 | ret = resend_aioreq(s, aio_req); |
822 | if (ret == SD_RES_SUCCESS) { | |
823 | goto out; | |
824 | } | |
825 | /* fall through */ | |
826 | default: | |
33b1db1c | 827 | acb->ret = -EIO; |
6daf194d | 828 | error_report("%s", sd_strerror(rsp.result)); |
13c31de2 | 829 | break; |
33b1db1c MK |
830 | } |
831 | ||
1d732d7d MK |
832 | free_aio_req(s, aio_req); |
833 | if (!acb->nr_pending) { | |
33b1db1c MK |
834 | /* |
835 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 836 | * we can switch back to sd_co_readv/writev now. |
33b1db1c MK |
837 | */ |
838 | acb->aio_done_func(acb); | |
839 | } | |
2df46246 MK |
840 | out: |
841 | s->co_recv = NULL; | |
011603ca MK |
842 | return; |
843 | err: | |
844 | s->co_recv = NULL; | |
845 | reconnect_to_sdog(opaque); | |
2df46246 MK |
846 | } |
847 | ||
848 | static void co_read_response(void *opaque) | |
849 | { | |
850 | BDRVSheepdogState *s = opaque; | |
851 | ||
852 | if (!s->co_recv) { | |
853 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
854 | } | |
855 | ||
856 | qemu_coroutine_enter(s->co_recv, opaque); | |
857 | } | |
858 | ||
859 | static void co_write_request(void *opaque) | |
860 | { | |
861 | BDRVSheepdogState *s = opaque; | |
862 | ||
863 | qemu_coroutine_enter(s->co_send, NULL); | |
33b1db1c MK |
864 | } |
865 | ||
33b1db1c MK |
866 | /* |
867 | * Return a socket discriptor to read/write objects. | |
868 | * | |
869 | * We cannot use this discriptor for other operations because | |
870 | * the block driver may be on waiting response from the server. | |
871 | */ | |
872 | static int get_sheep_fd(BDRVSheepdogState *s) | |
873 | { | |
1b8bbb46 | 874 | int fd; |
33b1db1c | 875 | |
25af257d | 876 | fd = connect_to_sdog(s); |
33b1db1c | 877 | if (fd < 0) { |
cb595887 | 878 | return fd; |
33b1db1c MK |
879 | } |
880 | ||
f2e5dca4 | 881 | qemu_aio_set_fd_handler(fd, co_read_response, NULL, s); |
33b1db1c MK |
882 | return fd; |
883 | } | |
884 | ||
5d6768e3 MK |
885 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
886 | char *vdi, uint32_t *snapid, char *tag) | |
887 | { | |
888 | URI *uri; | |
889 | QueryParams *qp = NULL; | |
890 | int ret = 0; | |
891 | ||
892 | uri = uri_parse(filename); | |
893 | if (!uri) { | |
894 | return -EINVAL; | |
895 | } | |
896 | ||
1b8bbb46 MK |
897 | /* transport */ |
898 | if (!strcmp(uri->scheme, "sheepdog")) { | |
899 | s->is_unix = false; | |
900 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
901 | s->is_unix = false; | |
902 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
903 | s->is_unix = true; | |
904 | } else { | |
905 | ret = -EINVAL; | |
906 | goto out; | |
907 | } | |
908 | ||
5d6768e3 MK |
909 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
910 | ret = -EINVAL; | |
911 | goto out; | |
912 | } | |
913 | pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1); | |
914 | ||
1b8bbb46 MK |
915 | qp = query_params_parse(uri->query); |
916 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
917 | ret = -EINVAL; | |
918 | goto out; | |
919 | } | |
920 | ||
921 | if (s->is_unix) { | |
922 | /* sheepdog+unix:///vdiname?socket=path */ | |
923 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
924 | ret = -EINVAL; | |
925 | goto out; | |
926 | } | |
927 | s->host_spec = g_strdup(qp->p[0].value); | |
928 | } else { | |
929 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
930 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
931 | uri->port ?: SD_DEFAULT_PORT); | |
932 | } | |
5d6768e3 MK |
933 | |
934 | /* snapshot tag */ | |
935 | if (uri->fragment) { | |
936 | *snapid = strtoul(uri->fragment, NULL, 10); | |
937 | if (*snapid == 0) { | |
938 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); | |
939 | } | |
940 | } else { | |
941 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
942 | } | |
943 | ||
944 | out: | |
945 | if (qp) { | |
946 | query_params_free(qp); | |
947 | } | |
948 | uri_free(uri); | |
949 | return ret; | |
950 | } | |
951 | ||
33b1db1c | 952 | /* |
5d6768e3 | 953 | * Parse a filename (old syntax) |
33b1db1c MK |
954 | * |
955 | * filename must be one of the following formats: | |
956 | * 1. [vdiname] | |
957 | * 2. [vdiname]:[snapid] | |
958 | * 3. [vdiname]:[tag] | |
959 | * 4. [hostname]:[port]:[vdiname] | |
960 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
961 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
962 | * | |
963 | * You can boot from the snapshot images by specifying `snapid` or | |
964 | * `tag'. | |
965 | * | |
966 | * You can run VMs outside the Sheepdog cluster by specifying | |
967 | * `hostname' and `port' (experimental). | |
968 | */ | |
969 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
970 | char *vdi, uint32_t *snapid, char *tag) | |
971 | { | |
5d6768e3 MK |
972 | char *p, *q, *uri; |
973 | const char *host_spec, *vdi_spec; | |
974 | int nr_sep, ret; | |
33b1db1c | 975 | |
5d6768e3 | 976 | strstart(filename, "sheepdog:", (const char **)&filename); |
7267c094 | 977 | p = q = g_strdup(filename); |
33b1db1c MK |
978 | |
979 | /* count the number of separators */ | |
980 | nr_sep = 0; | |
981 | while (*p) { | |
982 | if (*p == ':') { | |
983 | nr_sep++; | |
984 | } | |
985 | p++; | |
986 | } | |
987 | p = q; | |
988 | ||
5d6768e3 | 989 | /* use the first two tokens as host_spec. */ |
33b1db1c | 990 | if (nr_sep >= 2) { |
5d6768e3 | 991 | host_spec = p; |
33b1db1c | 992 | p = strchr(p, ':'); |
5d6768e3 | 993 | p++; |
33b1db1c MK |
994 | p = strchr(p, ':'); |
995 | *p++ = '\0'; | |
996 | } else { | |
5d6768e3 | 997 | host_spec = ""; |
33b1db1c MK |
998 | } |
999 | ||
5d6768e3 | 1000 | vdi_spec = p; |
33b1db1c | 1001 | |
5d6768e3 | 1002 | p = strchr(vdi_spec, ':'); |
33b1db1c | 1003 | if (p) { |
5d6768e3 | 1004 | *p++ = '#'; |
33b1db1c MK |
1005 | } |
1006 | ||
5d6768e3 | 1007 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 1008 | |
5d6768e3 MK |
1009 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
1010 | ||
1011 | g_free(q); | |
1012 | g_free(uri); | |
1013 | ||
1014 | return ret; | |
33b1db1c MK |
1015 | } |
1016 | ||
982dcbf4 MK |
1017 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, |
1018 | uint32_t snapid, const char *tag, uint32_t *vid, | |
1019 | bool lock) | |
33b1db1c MK |
1020 | { |
1021 | int ret, fd; | |
1022 | SheepdogVdiReq hdr; | |
1023 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1024 | unsigned int wlen, rlen = 0; | |
1025 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
1026 | ||
25af257d | 1027 | fd = connect_to_sdog(s); |
33b1db1c | 1028 | if (fd < 0) { |
cb595887 | 1029 | return fd; |
33b1db1c MK |
1030 | } |
1031 | ||
3178e275 JM |
1032 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
1033 | * which is desirable since we'll soon be sending those bytes, and | |
1034 | * don't want the send_req to read uninitialized data. | |
1035 | */ | |
33b1db1c MK |
1036 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
1037 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
1038 | ||
1039 | memset(&hdr, 0, sizeof(hdr)); | |
982dcbf4 | 1040 | if (lock) { |
33b1db1c | 1041 | hdr.opcode = SD_OP_LOCK_VDI; |
982dcbf4 MK |
1042 | } else { |
1043 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
33b1db1c MK |
1044 | } |
1045 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
1046 | hdr.proto_ver = SD_PROTO_VER; | |
1047 | hdr.data_length = wlen; | |
1048 | hdr.snapid = snapid; | |
1049 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1050 | ||
1051 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1052 | if (ret) { | |
33b1db1c MK |
1053 | goto out; |
1054 | } | |
1055 | ||
1056 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 1057 | error_report("cannot get vdi info, %s, %s %d %s", |
33b1db1c | 1058 | sd_strerror(rsp->result), filename, snapid, tag); |
cb595887 MK |
1059 | if (rsp->result == SD_RES_NO_VDI) { |
1060 | ret = -ENOENT; | |
1061 | } else { | |
1062 | ret = -EIO; | |
1063 | } | |
33b1db1c MK |
1064 | goto out; |
1065 | } | |
1066 | *vid = rsp->vdi_id; | |
1067 | ||
1068 | ret = 0; | |
1069 | out: | |
1070 | closesocket(fd); | |
1071 | return ret; | |
1072 | } | |
1073 | ||
d8716b41 | 1074 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
2f536801 | 1075 | struct iovec *iov, int niov, bool create, |
33b1db1c MK |
1076 | enum AIOCBState aiocb_type) |
1077 | { | |
1078 | int nr_copies = s->inode.nr_copies; | |
1079 | SheepdogObjReq hdr; | |
47783072 | 1080 | unsigned int wlen = 0; |
33b1db1c MK |
1081 | int ret; |
1082 | uint64_t oid = aio_req->oid; | |
1083 | unsigned int datalen = aio_req->data_len; | |
1084 | uint64_t offset = aio_req->offset; | |
1085 | uint8_t flags = aio_req->flags; | |
1086 | uint64_t old_oid = aio_req->base_oid; | |
1087 | ||
1088 | if (!nr_copies) { | |
6daf194d | 1089 | error_report("bug"); |
33b1db1c MK |
1090 | } |
1091 | ||
1092 | memset(&hdr, 0, sizeof(hdr)); | |
1093 | ||
47783072 LY |
1094 | switch (aiocb_type) { |
1095 | case AIOCB_FLUSH_CACHE: | |
1096 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1097 | break; | |
1098 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1099 | hdr.opcode = SD_OP_READ_OBJ; |
1100 | hdr.flags = flags; | |
47783072 LY |
1101 | break; |
1102 | case AIOCB_WRITE_UDATA: | |
1103 | if (create) { | |
1104 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1105 | } else { | |
1106 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1107 | } | |
33b1db1c | 1108 | wlen = datalen; |
33b1db1c | 1109 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1110 | break; |
cac8f4a6 LY |
1111 | case AIOCB_DISCARD_OBJ: |
1112 | hdr.opcode = SD_OP_DISCARD_OBJ; | |
1113 | break; | |
33b1db1c MK |
1114 | } |
1115 | ||
0e7106d8 LY |
1116 | if (s->cache_flags) { |
1117 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1118 | } |
1119 | ||
33b1db1c MK |
1120 | hdr.oid = oid; |
1121 | hdr.cow_oid = old_oid; | |
1122 | hdr.copies = s->inode.nr_copies; | |
1123 | ||
1124 | hdr.data_length = datalen; | |
1125 | hdr.offset = offset; | |
1126 | ||
1127 | hdr.id = aio_req->id; | |
1128 | ||
2df46246 MK |
1129 | qemu_co_mutex_lock(&s->lock); |
1130 | s->co_send = qemu_coroutine_self(); | |
f2e5dca4 | 1131 | qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, s); |
128aa589 | 1132 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1133 | |
1134 | /* send a header */ | |
8c5135f9 | 1135 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
80731d9d | 1136 | if (ret != sizeof(hdr)) { |
6daf194d | 1137 | error_report("failed to send a req, %s", strerror(errno)); |
011603ca | 1138 | goto out; |
33b1db1c MK |
1139 | } |
1140 | ||
1141 | if (wlen) { | |
2fc8ae1d | 1142 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
80731d9d | 1143 | if (ret != wlen) { |
6daf194d | 1144 | error_report("failed to send a data, %s", strerror(errno)); |
33b1db1c MK |
1145 | } |
1146 | } | |
011603ca | 1147 | out: |
128aa589 | 1148 | socket_set_cork(s->fd, 0); |
f2e5dca4 | 1149 | qemu_aio_set_fd_handler(s->fd, co_read_response, NULL, s); |
011603ca | 1150 | s->co_send = NULL; |
2df46246 | 1151 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1152 | |
1153 | return 0; | |
1154 | } | |
1155 | ||
29a67f7e | 1156 | static int read_write_object(int fd, char *buf, uint64_t oid, uint8_t copies, |
33b1db1c | 1157 | unsigned int datalen, uint64_t offset, |
0e7106d8 | 1158 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1159 | { |
1160 | SheepdogObjReq hdr; | |
1161 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1162 | unsigned int wlen, rlen; | |
1163 | int ret; | |
1164 | ||
1165 | memset(&hdr, 0, sizeof(hdr)); | |
1166 | ||
1167 | if (write) { | |
1168 | wlen = datalen; | |
1169 | rlen = 0; | |
1170 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1171 | if (create) { | |
1172 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1173 | } else { | |
1174 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1175 | } | |
1176 | } else { | |
1177 | wlen = 0; | |
1178 | rlen = datalen; | |
1179 | hdr.opcode = SD_OP_READ_OBJ; | |
1180 | } | |
47622c44 | 1181 | |
0e7106d8 | 1182 | hdr.flags |= cache_flags; |
47622c44 | 1183 | |
33b1db1c MK |
1184 | hdr.oid = oid; |
1185 | hdr.data_length = datalen; | |
1186 | hdr.offset = offset; | |
1187 | hdr.copies = copies; | |
1188 | ||
1189 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1190 | if (ret) { | |
6daf194d | 1191 | error_report("failed to send a request to the sheep"); |
cb595887 | 1192 | return ret; |
33b1db1c MK |
1193 | } |
1194 | ||
1195 | switch (rsp->result) { | |
1196 | case SD_RES_SUCCESS: | |
1197 | return 0; | |
1198 | default: | |
6daf194d | 1199 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1200 | return -EIO; |
33b1db1c MK |
1201 | } |
1202 | } | |
1203 | ||
29a67f7e | 1204 | static int read_object(int fd, char *buf, uint64_t oid, uint8_t copies, |
0e7106d8 LY |
1205 | unsigned int datalen, uint64_t offset, |
1206 | uint32_t cache_flags) | |
33b1db1c | 1207 | { |
2f536801 | 1208 | return read_write_object(fd, buf, oid, copies, datalen, offset, false, |
0e7106d8 | 1209 | false, cache_flags); |
33b1db1c MK |
1210 | } |
1211 | ||
29a67f7e | 1212 | static int write_object(int fd, char *buf, uint64_t oid, uint8_t copies, |
2f536801 | 1213 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1214 | uint32_t cache_flags) |
33b1db1c | 1215 | { |
2f536801 | 1216 | return read_write_object(fd, buf, oid, copies, datalen, offset, true, |
0e7106d8 | 1217 | create, cache_flags); |
33b1db1c MK |
1218 | } |
1219 | ||
9ff53a0e MK |
1220 | /* update inode with the latest state */ |
1221 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag) | |
1222 | { | |
1223 | SheepdogInode *inode; | |
1224 | int ret = 0, fd; | |
1225 | uint32_t vid = 0; | |
1226 | ||
1227 | fd = connect_to_sdog(s); | |
1228 | if (fd < 0) { | |
1229 | return -EIO; | |
1230 | } | |
1231 | ||
1232 | inode = g_malloc(sizeof(s->inode)); | |
1233 | ||
1234 | ret = find_vdi_name(s, s->name, snapid, tag, &vid, false); | |
1235 | if (ret) { | |
1236 | goto out; | |
1237 | } | |
1238 | ||
1239 | ret = read_object(fd, (char *)inode, vid_to_vdi_oid(vid), | |
1240 | s->inode.nr_copies, sizeof(*inode), 0, s->cache_flags); | |
1241 | if (ret < 0) { | |
1242 | goto out; | |
1243 | } | |
1244 | ||
1245 | if (inode->vdi_id != s->inode.vdi_id) { | |
1246 | memcpy(&s->inode, inode, sizeof(s->inode)); | |
1247 | } | |
1248 | ||
1249 | out: | |
1250 | g_free(inode); | |
1251 | closesocket(fd); | |
1252 | ||
1253 | return ret; | |
1254 | } | |
1255 | ||
13c31de2 MK |
1256 | static int coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req) |
1257 | { | |
1258 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1259 | bool create = false; | |
13c31de2 MK |
1260 | |
1261 | /* check whether this request becomes a CoW one */ | |
2412aec7 | 1262 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) { |
13c31de2 MK |
1263 | int idx = data_oid_to_idx(aio_req->oid); |
1264 | AIOReq *areq; | |
1265 | ||
1266 | if (s->inode.data_vdi_id[idx] == 0) { | |
1267 | create = true; | |
1268 | goto out; | |
1269 | } | |
1270 | if (is_data_obj_writable(&s->inode, idx)) { | |
1271 | goto out; | |
1272 | } | |
1273 | ||
1274 | /* link to the pending list if there is another CoW request to | |
1275 | * the same object */ | |
1276 | QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { | |
1277 | if (areq != aio_req && areq->oid == aio_req->oid) { | |
2440a2c3 | 1278 | DPRINTF("simultaneous CoW to %" PRIx64 "\n", aio_req->oid); |
13c31de2 MK |
1279 | QLIST_REMOVE(aio_req, aio_siblings); |
1280 | QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, aio_siblings); | |
1281 | return SD_RES_SUCCESS; | |
1282 | } | |
1283 | } | |
1284 | ||
1285 | aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx); | |
1286 | aio_req->flags |= SD_FLAG_CMD_COW; | |
1287 | create = true; | |
1288 | } | |
1289 | out: | |
2412aec7 MK |
1290 | if (is_data_obj(aio_req->oid)) { |
1291 | return add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, | |
1292 | create, acb->aiocb_type); | |
1293 | } else { | |
1294 | struct iovec iov; | |
1295 | iov.iov_base = &s->inode; | |
1296 | iov.iov_len = sizeof(s->inode); | |
1297 | return add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); | |
1298 | } | |
13c31de2 MK |
1299 | } |
1300 | ||
c8c96350 KW |
1301 | /* TODO Convert to fine grained options */ |
1302 | static QemuOptsList runtime_opts = { | |
1303 | .name = "sheepdog", | |
1304 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1305 | .desc = { | |
1306 | { | |
1307 | .name = "filename", | |
1308 | .type = QEMU_OPT_STRING, | |
1309 | .help = "URL to the sheepdog image", | |
1310 | }, | |
1311 | { /* end of list */ } | |
1312 | }, | |
1313 | }; | |
1314 | ||
015a1036 MR |
1315 | static int sd_open(BlockDriverState *bs, QDict *options, int flags, |
1316 | Error **errp) | |
33b1db1c MK |
1317 | { |
1318 | int ret, fd; | |
1319 | uint32_t vid = 0; | |
1320 | BDRVSheepdogState *s = bs->opaque; | |
1321 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1322 | uint32_t snapid; | |
1323 | char *buf = NULL; | |
c8c96350 KW |
1324 | QemuOpts *opts; |
1325 | Error *local_err = NULL; | |
1326 | const char *filename; | |
1327 | ||
011603ca MK |
1328 | s->bs = bs; |
1329 | ||
c8c96350 KW |
1330 | opts = qemu_opts_create_nofail(&runtime_opts); |
1331 | qemu_opts_absorb_qdict(opts, options, &local_err); | |
1332 | if (error_is_set(&local_err)) { | |
1333 | qerror_report_err(local_err); | |
1334 | error_free(local_err); | |
1335 | ret = -EINVAL; | |
1336 | goto out; | |
1337 | } | |
1338 | ||
1339 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1340 | |
c292ee6a MK |
1341 | QLIST_INIT(&s->inflight_aio_head); |
1342 | QLIST_INIT(&s->pending_aio_head); | |
011603ca | 1343 | QLIST_INIT(&s->failed_aio_head); |
33b1db1c MK |
1344 | s->fd = -1; |
1345 | ||
1346 | memset(vdi, 0, sizeof(vdi)); | |
1347 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1348 | |
1349 | if (strstr(filename, "://")) { | |
1350 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1351 | } else { | |
1352 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1353 | } | |
1354 | if (ret < 0) { | |
33b1db1c MK |
1355 | goto out; |
1356 | } | |
1357 | s->fd = get_sheep_fd(s); | |
1358 | if (s->fd < 0) { | |
cb595887 | 1359 | ret = s->fd; |
33b1db1c MK |
1360 | goto out; |
1361 | } | |
1362 | ||
982dcbf4 | 1363 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, true); |
33b1db1c MK |
1364 | if (ret) { |
1365 | goto out; | |
1366 | } | |
1367 | ||
0e7106d8 LY |
1368 | /* |
1369 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1370 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1371 | */ | |
1372 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1373 | if (flags & BDRV_O_NOCACHE) { | |
1374 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1375 | } | |
cac8f4a6 | 1376 | s->discard_supported = true; |
0e7106d8 | 1377 | |
622b6057 | 1378 | if (snapid || tag[0] != '\0') { |
2440a2c3 | 1379 | DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1380 | s->is_snapshot = true; |
33b1db1c MK |
1381 | } |
1382 | ||
25af257d | 1383 | fd = connect_to_sdog(s); |
33b1db1c | 1384 | if (fd < 0) { |
cb595887 | 1385 | ret = fd; |
33b1db1c MK |
1386 | goto out; |
1387 | } | |
1388 | ||
7267c094 | 1389 | buf = g_malloc(SD_INODE_SIZE); |
47622c44 | 1390 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, |
0e7106d8 | 1391 | s->cache_flags); |
33b1db1c MK |
1392 | |
1393 | closesocket(fd); | |
1394 | ||
1395 | if (ret) { | |
1396 | goto out; | |
1397 | } | |
1398 | ||
1399 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1400 | s->min_dirty_data_idx = UINT32_MAX; | |
1401 | s->max_dirty_data_idx = 0; | |
1402 | ||
e8bfaa2f | 1403 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; |
3178e275 | 1404 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1405 | qemu_co_mutex_init(&s->lock); |
c8c96350 | 1406 | qemu_opts_del(opts); |
7267c094 | 1407 | g_free(buf); |
33b1db1c MK |
1408 | return 0; |
1409 | out: | |
f2e5dca4 | 1410 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL); |
33b1db1c MK |
1411 | if (s->fd >= 0) { |
1412 | closesocket(s->fd); | |
1413 | } | |
c8c96350 | 1414 | qemu_opts_del(opts); |
7267c094 | 1415 | g_free(buf); |
cb595887 | 1416 | return ret; |
33b1db1c MK |
1417 | } |
1418 | ||
25af257d | 1419 | static int do_sd_create(BDRVSheepdogState *s, char *filename, int64_t vdi_size, |
1841f880 LY |
1420 | uint32_t base_vid, uint32_t *vdi_id, int snapshot, |
1421 | uint8_t copy_policy) | |
33b1db1c MK |
1422 | { |
1423 | SheepdogVdiReq hdr; | |
1424 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1425 | int fd, ret; | |
1426 | unsigned int wlen, rlen = 0; | |
1427 | char buf[SD_MAX_VDI_LEN]; | |
1428 | ||
25af257d | 1429 | fd = connect_to_sdog(s); |
33b1db1c | 1430 | if (fd < 0) { |
cb595887 | 1431 | return fd; |
33b1db1c MK |
1432 | } |
1433 | ||
3178e275 JM |
1434 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1435 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1436 | */ | |
33b1db1c | 1437 | memset(buf, 0, sizeof(buf)); |
3178e275 | 1438 | pstrcpy(buf, sizeof(buf), filename); |
33b1db1c MK |
1439 | |
1440 | memset(&hdr, 0, sizeof(hdr)); | |
1441 | hdr.opcode = SD_OP_NEW_VDI; | |
6f74c260 | 1442 | hdr.vdi_id = base_vid; |
33b1db1c MK |
1443 | |
1444 | wlen = SD_MAX_VDI_LEN; | |
1445 | ||
1446 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1447 | hdr.snapid = snapshot; | |
1448 | ||
1449 | hdr.data_length = wlen; | |
1450 | hdr.vdi_size = vdi_size; | |
1841f880 | 1451 | hdr.copy_policy = copy_policy; |
33b1db1c MK |
1452 | |
1453 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1454 | ||
1455 | closesocket(fd); | |
1456 | ||
1457 | if (ret) { | |
cb595887 | 1458 | return ret; |
33b1db1c MK |
1459 | } |
1460 | ||
1461 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 1462 | error_report("%s, %s", sd_strerror(rsp->result), filename); |
33b1db1c MK |
1463 | return -EIO; |
1464 | } | |
1465 | ||
1466 | if (vdi_id) { | |
1467 | *vdi_id = rsp->vdi_id; | |
1468 | } | |
1469 | ||
1470 | return 0; | |
1471 | } | |
1472 | ||
a8e0fdd7 MK |
1473 | static int sd_prealloc(const char *filename) |
1474 | { | |
1475 | BlockDriverState *bs = NULL; | |
1476 | uint32_t idx, max_idx; | |
1477 | int64_t vdi_size; | |
7267c094 | 1478 | void *buf = g_malloc0(SD_DATA_OBJ_SIZE); |
34b5d2c6 | 1479 | Error *local_err = NULL; |
a8e0fdd7 MK |
1480 | int ret; |
1481 | ||
34b5d2c6 | 1482 | ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR, &local_err); |
a8e0fdd7 | 1483 | if (ret < 0) { |
34b5d2c6 MR |
1484 | qerror_report_err(local_err); |
1485 | error_free(local_err); | |
a8e0fdd7 MK |
1486 | goto out; |
1487 | } | |
1488 | ||
1489 | vdi_size = bdrv_getlength(bs); | |
1490 | if (vdi_size < 0) { | |
1491 | ret = vdi_size; | |
1492 | goto out; | |
1493 | } | |
1494 | max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); | |
1495 | ||
1496 | for (idx = 0; idx < max_idx; idx++) { | |
1497 | /* | |
1498 | * The created image can be a cloned image, so we need to read | |
1499 | * a data from the source image. | |
1500 | */ | |
1501 | ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1502 | if (ret < 0) { | |
1503 | goto out; | |
1504 | } | |
1505 | ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1506 | if (ret < 0) { | |
1507 | goto out; | |
1508 | } | |
1509 | } | |
1510 | out: | |
1511 | if (bs) { | |
4f6fd349 | 1512 | bdrv_unref(bs); |
a8e0fdd7 | 1513 | } |
7267c094 | 1514 | g_free(buf); |
a8e0fdd7 MK |
1515 | |
1516 | return ret; | |
1517 | } | |
1518 | ||
d5124c00 MR |
1519 | static int sd_create(const char *filename, QEMUOptionParameter *options, |
1520 | Error **errp) | |
33b1db1c | 1521 | { |
b6fc8245 | 1522 | int ret = 0; |
b4447363 | 1523 | uint32_t vid = 0, base_vid = 0; |
33b1db1c MK |
1524 | int64_t vdi_size = 0; |
1525 | char *backing_file = NULL; | |
b6fc8245 | 1526 | BDRVSheepdogState *s; |
b4447363 MK |
1527 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; |
1528 | uint32_t snapid; | |
2f536801 | 1529 | bool prealloc = false; |
34b5d2c6 | 1530 | Error *local_err = NULL; |
33b1db1c | 1531 | |
b6fc8245 MK |
1532 | s = g_malloc0(sizeof(BDRVSheepdogState)); |
1533 | ||
b4447363 MK |
1534 | memset(vdi, 0, sizeof(vdi)); |
1535 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1536 | if (strstr(filename, "://")) { |
1537 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1538 | } else { | |
1539 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1540 | } | |
1541 | if (ret < 0) { | |
b6fc8245 | 1542 | goto out; |
b4447363 MK |
1543 | } |
1544 | ||
33b1db1c MK |
1545 | while (options && options->name) { |
1546 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1547 | vdi_size = options->value.n; | |
1548 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1549 | backing_file = options->value.s; | |
a8e0fdd7 MK |
1550 | } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1551 | if (!options->value.s || !strcmp(options->value.s, "off")) { | |
2f536801 | 1552 | prealloc = false; |
a8e0fdd7 | 1553 | } else if (!strcmp(options->value.s, "full")) { |
2f536801 | 1554 | prealloc = true; |
a8e0fdd7 MK |
1555 | } else { |
1556 | error_report("Invalid preallocation mode: '%s'", | |
1557 | options->value.s); | |
b6fc8245 MK |
1558 | ret = -EINVAL; |
1559 | goto out; | |
a8e0fdd7 | 1560 | } |
33b1db1c MK |
1561 | } |
1562 | options++; | |
1563 | } | |
1564 | ||
1565 | if (vdi_size > SD_MAX_VDI_SIZE) { | |
6daf194d | 1566 | error_report("too big image size"); |
b6fc8245 MK |
1567 | ret = -EINVAL; |
1568 | goto out; | |
33b1db1c MK |
1569 | } |
1570 | ||
1571 | if (backing_file) { | |
1572 | BlockDriverState *bs; | |
1573 | BDRVSheepdogState *s; | |
1574 | BlockDriver *drv; | |
1575 | ||
1576 | /* Currently, only Sheepdog backing image is supported. */ | |
98289620 | 1577 | drv = bdrv_find_protocol(backing_file, true); |
33b1db1c | 1578 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
6daf194d | 1579 | error_report("backing_file must be a sheepdog image"); |
b6fc8245 MK |
1580 | ret = -EINVAL; |
1581 | goto out; | |
33b1db1c MK |
1582 | } |
1583 | ||
34b5d2c6 | 1584 | ret = bdrv_file_open(&bs, backing_file, NULL, 0, &local_err); |
cb595887 | 1585 | if (ret < 0) { |
34b5d2c6 MR |
1586 | qerror_report_err(local_err); |
1587 | error_free(local_err); | |
b6fc8245 | 1588 | goto out; |
cb595887 | 1589 | } |
33b1db1c MK |
1590 | |
1591 | s = bs->opaque; | |
1592 | ||
1593 | if (!is_snapshot(&s->inode)) { | |
6daf194d | 1594 | error_report("cannot clone from a non snapshot vdi"); |
4f6fd349 | 1595 | bdrv_unref(bs); |
b6fc8245 MK |
1596 | ret = -EINVAL; |
1597 | goto out; | |
33b1db1c MK |
1598 | } |
1599 | ||
b4447363 | 1600 | base_vid = s->inode.vdi_id; |
4f6fd349 | 1601 | bdrv_unref(bs); |
33b1db1c MK |
1602 | } |
1603 | ||
1841f880 LY |
1604 | /* TODO: allow users to specify copy number */ |
1605 | ret = do_sd_create(s, vdi, vdi_size, base_vid, &vid, 0, 0); | |
a8e0fdd7 | 1606 | if (!prealloc || ret) { |
b6fc8245 | 1607 | goto out; |
a8e0fdd7 MK |
1608 | } |
1609 | ||
b6fc8245 MK |
1610 | ret = sd_prealloc(filename); |
1611 | out: | |
1612 | g_free(s); | |
1613 | return ret; | |
33b1db1c MK |
1614 | } |
1615 | ||
1616 | static void sd_close(BlockDriverState *bs) | |
1617 | { | |
1618 | BDRVSheepdogState *s = bs->opaque; | |
1619 | SheepdogVdiReq hdr; | |
1620 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1621 | unsigned int wlen, rlen = 0; | |
1622 | int fd, ret; | |
1623 | ||
2440a2c3 | 1624 | DPRINTF("%s\n", s->name); |
33b1db1c | 1625 | |
25af257d | 1626 | fd = connect_to_sdog(s); |
33b1db1c MK |
1627 | if (fd < 0) { |
1628 | return; | |
1629 | } | |
1630 | ||
1631 | memset(&hdr, 0, sizeof(hdr)); | |
1632 | ||
1633 | hdr.opcode = SD_OP_RELEASE_VDI; | |
6f74c260 | 1634 | hdr.vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1635 | wlen = strlen(s->name) + 1; |
1636 | hdr.data_length = wlen; | |
1637 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1638 | ||
1639 | ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); | |
1640 | ||
1641 | closesocket(fd); | |
1642 | ||
1643 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1644 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1645 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1646 | } |
1647 | ||
f2e5dca4 | 1648 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL); |
33b1db1c | 1649 | closesocket(s->fd); |
25af257d | 1650 | g_free(s->host_spec); |
33b1db1c MK |
1651 | } |
1652 | ||
1653 | static int64_t sd_getlength(BlockDriverState *bs) | |
1654 | { | |
1655 | BDRVSheepdogState *s = bs->opaque; | |
1656 | ||
1657 | return s->inode.vdi_size; | |
1658 | } | |
1659 | ||
1660 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1661 | { | |
1662 | BDRVSheepdogState *s = bs->opaque; | |
1663 | int ret, fd; | |
1664 | unsigned int datalen; | |
1665 | ||
1666 | if (offset < s->inode.vdi_size) { | |
6daf194d | 1667 | error_report("shrinking is not supported"); |
33b1db1c MK |
1668 | return -EINVAL; |
1669 | } else if (offset > SD_MAX_VDI_SIZE) { | |
6daf194d | 1670 | error_report("too big image size"); |
33b1db1c MK |
1671 | return -EINVAL; |
1672 | } | |
1673 | ||
25af257d | 1674 | fd = connect_to_sdog(s); |
33b1db1c | 1675 | if (fd < 0) { |
cb595887 | 1676 | return fd; |
33b1db1c MK |
1677 | } |
1678 | ||
1679 | /* we don't need to update entire object */ | |
1680 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1681 | s->inode.vdi_size = offset; | |
1682 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
0e7106d8 | 1683 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
33b1db1c MK |
1684 | close(fd); |
1685 | ||
1686 | if (ret < 0) { | |
6daf194d | 1687 | error_report("failed to update an inode."); |
33b1db1c MK |
1688 | } |
1689 | ||
cb595887 | 1690 | return ret; |
33b1db1c MK |
1691 | } |
1692 | ||
1693 | /* | |
1694 | * This function is called after writing data objects. If we need to | |
1695 | * update metadata, this sends a write request to the vdi object. | |
2df46246 | 1696 | * Otherwise, this switches back to sd_co_readv/writev. |
33b1db1c | 1697 | */ |
d8716b41 | 1698 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c MK |
1699 | { |
1700 | int ret; | |
1701 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1702 | struct iovec iov; | |
1703 | AIOReq *aio_req; | |
1704 | uint32_t offset, data_len, mn, mx; | |
1705 | ||
1706 | mn = s->min_dirty_data_idx; | |
1707 | mx = s->max_dirty_data_idx; | |
1708 | if (mn <= mx) { | |
1709 | /* we need to update the vdi object. */ | |
1710 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
1711 | mn * sizeof(s->inode.data_vdi_id[0]); | |
1712 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
1713 | ||
1714 | s->min_dirty_data_idx = UINT32_MAX; | |
1715 | s->max_dirty_data_idx = 0; | |
1716 | ||
1717 | iov.iov_base = &s->inode; | |
1718 | iov.iov_len = sizeof(s->inode); | |
1719 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
1720 | data_len, offset, 0, 0, offset); | |
c292ee6a | 1721 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
2f536801 | 1722 | ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); |
33b1db1c MK |
1723 | if (ret) { |
1724 | free_aio_req(s, aio_req); | |
1725 | acb->ret = -EIO; | |
1726 | goto out; | |
1727 | } | |
1728 | ||
1729 | acb->aio_done_func = sd_finish_aiocb; | |
1730 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1731 | return; | |
1732 | } | |
1733 | out: | |
1734 | sd_finish_aiocb(acb); | |
1735 | } | |
1736 | ||
859e5553 LY |
1737 | /* Delete current working VDI on the snapshot chain */ |
1738 | static bool sd_delete(BDRVSheepdogState *s) | |
1739 | { | |
1740 | unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0; | |
1741 | SheepdogVdiReq hdr = { | |
1742 | .opcode = SD_OP_DEL_VDI, | |
1743 | .vdi_id = s->inode.vdi_id, | |
1744 | .data_length = wlen, | |
1745 | .flags = SD_FLAG_CMD_WRITE, | |
1746 | }; | |
1747 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1748 | int fd, ret; | |
1749 | ||
1750 | fd = connect_to_sdog(s); | |
1751 | if (fd < 0) { | |
1752 | return false; | |
1753 | } | |
1754 | ||
1755 | ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); | |
1756 | closesocket(fd); | |
1757 | if (ret) { | |
1758 | return false; | |
1759 | } | |
1760 | switch (rsp->result) { | |
1761 | case SD_RES_NO_VDI: | |
1762 | error_report("%s was already deleted", s->name); | |
1763 | /* fall through */ | |
1764 | case SD_RES_SUCCESS: | |
1765 | break; | |
1766 | default: | |
1767 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
1768 | return false; | |
1769 | } | |
1770 | ||
1771 | return true; | |
1772 | } | |
1773 | ||
33b1db1c MK |
1774 | /* |
1775 | * Create a writable VDI from a snapshot | |
1776 | */ | |
1777 | static int sd_create_branch(BDRVSheepdogState *s) | |
1778 | { | |
1779 | int ret, fd; | |
1780 | uint32_t vid; | |
1781 | char *buf; | |
859e5553 | 1782 | bool deleted; |
33b1db1c | 1783 | |
2440a2c3 | 1784 | DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
33b1db1c | 1785 | |
7267c094 | 1786 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 1787 | |
859e5553 LY |
1788 | /* |
1789 | * Even If deletion fails, we will just create extra snapshot based on | |
1790 | * the workding VDI which was supposed to be deleted. So no need to | |
1791 | * false bail out. | |
1792 | */ | |
1793 | deleted = sd_delete(s); | |
1794 | ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, | |
1841f880 | 1795 | !deleted, s->inode.copy_policy); |
33b1db1c MK |
1796 | if (ret) { |
1797 | goto out; | |
1798 | } | |
1799 | ||
2440a2c3 | 1800 | DPRINTF("%" PRIx32 " is created.\n", vid); |
33b1db1c | 1801 | |
25af257d | 1802 | fd = connect_to_sdog(s); |
33b1db1c | 1803 | if (fd < 0) { |
cb595887 | 1804 | ret = fd; |
33b1db1c MK |
1805 | goto out; |
1806 | } | |
1807 | ||
1808 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, | |
0e7106d8 | 1809 | SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1810 | |
1811 | closesocket(fd); | |
1812 | ||
1813 | if (ret < 0) { | |
1814 | goto out; | |
1815 | } | |
1816 | ||
1817 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1818 | ||
2f536801 | 1819 | s->is_snapshot = false; |
33b1db1c | 1820 | ret = 0; |
2440a2c3 | 1821 | DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
33b1db1c MK |
1822 | |
1823 | out: | |
7267c094 | 1824 | g_free(buf); |
33b1db1c MK |
1825 | |
1826 | return ret; | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * Send I/O requests to the server. | |
1831 | * | |
1832 | * This function sends requests to the server, links the requests to | |
c292ee6a | 1833 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
1834 | * waiting the response. The responses are received in the |
1835 | * `aio_read_response' function which is called from the main loop as | |
1836 | * a fd handler. | |
2df46246 MK |
1837 | * |
1838 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
1839 | * request and -errno in error cases. | |
33b1db1c | 1840 | */ |
d8716b41 | 1841 | static int coroutine_fn sd_co_rw_vector(void *p) |
33b1db1c MK |
1842 | { |
1843 | SheepdogAIOCB *acb = p; | |
1844 | int ret = 0; | |
e8bfaa2f LY |
1845 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; |
1846 | unsigned long idx = acb->sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE; | |
33b1db1c | 1847 | uint64_t oid; |
e8bfaa2f | 1848 | uint64_t offset = (acb->sector_num * BDRV_SECTOR_SIZE) % SD_DATA_OBJ_SIZE; |
33b1db1c MK |
1849 | BDRVSheepdogState *s = acb->common.bs->opaque; |
1850 | SheepdogInode *inode = &s->inode; | |
1851 | AIOReq *aio_req; | |
1852 | ||
33b1db1c MK |
1853 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
1854 | /* | |
1855 | * In the case we open the snapshot VDI, Sheepdog creates the | |
1856 | * writable VDI when we do a write operation first. | |
1857 | */ | |
1858 | ret = sd_create_branch(s); | |
1859 | if (ret) { | |
1860 | acb->ret = -EIO; | |
1861 | goto out; | |
1862 | } | |
1863 | } | |
1864 | ||
1d732d7d MK |
1865 | /* |
1866 | * Make sure we don't free the aiocb before we are done with all requests. | |
1867 | * This additional reference is dropped at the end of this function. | |
1868 | */ | |
1869 | acb->nr_pending++; | |
1870 | ||
33b1db1c MK |
1871 | while (done != total) { |
1872 | uint8_t flags = 0; | |
1873 | uint64_t old_oid = 0; | |
2f536801 | 1874 | bool create = false; |
33b1db1c MK |
1875 | |
1876 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
1877 | ||
1878 | len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); | |
1879 | ||
19db9b90 CH |
1880 | switch (acb->aiocb_type) { |
1881 | case AIOCB_READ_UDATA: | |
1882 | if (!inode->data_vdi_id[idx]) { | |
1883 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
1884 | goto done; |
1885 | } | |
19db9b90 CH |
1886 | break; |
1887 | case AIOCB_WRITE_UDATA: | |
1888 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 1889 | create = true; |
19db9b90 CH |
1890 | } else if (!is_data_obj_writable(inode, idx)) { |
1891 | /* Copy-On-Write */ | |
2f536801 | 1892 | create = true; |
19db9b90 CH |
1893 | old_oid = oid; |
1894 | flags = SD_FLAG_CMD_COW; | |
1895 | } | |
1896 | break; | |
cac8f4a6 LY |
1897 | case AIOCB_DISCARD_OBJ: |
1898 | /* | |
1899 | * We discard the object only when the whole object is | |
1900 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
1901 | */ | |
1902 | if (len != SD_DATA_OBJ_SIZE || inode->data_vdi_id[idx] == 0) { | |
1903 | goto done; | |
1904 | } | |
1905 | break; | |
19db9b90 CH |
1906 | default: |
1907 | break; | |
33b1db1c MK |
1908 | } |
1909 | ||
1910 | if (create) { | |
2440a2c3 | 1911 | DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1b6ac998 | 1912 | inode->vdi_id, oid, |
33b1db1c MK |
1913 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
1914 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
2440a2c3 | 1915 | DPRINTF("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
1916 | } |
1917 | ||
1918 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); | |
1919 | ||
1920 | if (create) { | |
1921 | AIOReq *areq; | |
c292ee6a | 1922 | QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { |
33b1db1c MK |
1923 | if (areq->oid == oid) { |
1924 | /* | |
1925 | * Sheepdog cannot handle simultaneous create | |
1926 | * requests to the same object. So we cannot send | |
1927 | * the request until the previous request | |
1928 | * finishes. | |
1929 | */ | |
1930 | aio_req->flags = 0; | |
1931 | aio_req->base_oid = 0; | |
c292ee6a MK |
1932 | QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, |
1933 | aio_siblings); | |
33b1db1c MK |
1934 | goto done; |
1935 | } | |
1936 | } | |
1937 | } | |
1938 | ||
c292ee6a | 1939 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
33b1db1c MK |
1940 | ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
1941 | create, acb->aiocb_type); | |
1942 | if (ret < 0) { | |
6daf194d | 1943 | error_report("add_aio_request is failed"); |
33b1db1c MK |
1944 | free_aio_req(s, aio_req); |
1945 | acb->ret = -EIO; | |
1946 | goto out; | |
1947 | } | |
1948 | done: | |
1949 | offset = 0; | |
1950 | idx++; | |
1951 | done += len; | |
1952 | } | |
1953 | out: | |
1d732d7d | 1954 | if (!--acb->nr_pending) { |
2df46246 | 1955 | return acb->ret; |
33b1db1c | 1956 | } |
2df46246 | 1957 | return 1; |
33b1db1c MK |
1958 | } |
1959 | ||
a968168c | 1960 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1961 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1962 | { |
1963 | SheepdogAIOCB *acb; | |
2df46246 | 1964 | int ret; |
33b1db1c MK |
1965 | |
1966 | if (bs->growable && sector_num + nb_sectors > bs->total_sectors) { | |
e8bfaa2f | 1967 | ret = sd_truncate(bs, (sector_num + nb_sectors) * BDRV_SECTOR_SIZE); |
cb595887 MK |
1968 | if (ret < 0) { |
1969 | return ret; | |
33b1db1c MK |
1970 | } |
1971 | bs->total_sectors = sector_num + nb_sectors; | |
1972 | } | |
1973 | ||
f700f8e3 | 1974 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
1975 | acb->aio_done_func = sd_write_done; |
1976 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1977 | ||
2df46246 MK |
1978 | ret = sd_co_rw_vector(acb); |
1979 | if (ret <= 0) { | |
1980 | qemu_aio_release(acb); | |
1981 | return ret; | |
1982 | } | |
1983 | ||
1984 | qemu_coroutine_yield(); | |
1985 | ||
1986 | return acb->ret; | |
33b1db1c MK |
1987 | } |
1988 | ||
a968168c | 1989 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1990 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1991 | { |
1992 | SheepdogAIOCB *acb; | |
19db9b90 | 1993 | int ret; |
33b1db1c | 1994 | |
f700f8e3 | 1995 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
1996 | acb->aiocb_type = AIOCB_READ_UDATA; |
1997 | acb->aio_done_func = sd_finish_aiocb; | |
1998 | ||
2df46246 MK |
1999 | ret = sd_co_rw_vector(acb); |
2000 | if (ret <= 0) { | |
2001 | qemu_aio_release(acb); | |
2002 | return ret; | |
2003 | } | |
2004 | ||
2005 | qemu_coroutine_yield(); | |
2006 | ||
2007 | return acb->ret; | |
33b1db1c MK |
2008 | } |
2009 | ||
47622c44 LY |
2010 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
2011 | { | |
2012 | BDRVSheepdogState *s = bs->opaque; | |
47783072 LY |
2013 | SheepdogAIOCB *acb; |
2014 | AIOReq *aio_req; | |
47622c44 | 2015 | int ret; |
47622c44 | 2016 | |
0e7106d8 | 2017 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
2018 | return 0; |
2019 | } | |
2020 | ||
f700f8e3 | 2021 | acb = sd_aio_setup(bs, NULL, 0, 0); |
47783072 LY |
2022 | acb->aiocb_type = AIOCB_FLUSH_CACHE; |
2023 | acb->aio_done_func = sd_finish_aiocb; | |
47622c44 | 2024 | |
47783072 LY |
2025 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
2026 | 0, 0, 0, 0, 0); | |
2027 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
2028 | ret = add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type); | |
2029 | if (ret < 0) { | |
2030 | error_report("add_aio_request is failed"); | |
2031 | free_aio_req(s, aio_req); | |
2032 | qemu_aio_release(acb); | |
47622c44 LY |
2033 | return ret; |
2034 | } | |
2035 | ||
47783072 LY |
2036 | qemu_coroutine_yield(); |
2037 | return acb->ret; | |
47622c44 LY |
2038 | } |
2039 | ||
33b1db1c MK |
2040 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
2041 | { | |
2042 | BDRVSheepdogState *s = bs->opaque; | |
2043 | int ret, fd; | |
2044 | uint32_t new_vid; | |
2045 | SheepdogInode *inode; | |
2046 | unsigned int datalen; | |
2047 | ||
2440a2c3 | 2048 | DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
2049 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
2050 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
2051 | ||
2052 | if (s->is_snapshot) { | |
2053 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 2054 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
2055 | |
2056 | return -EINVAL; | |
2057 | } | |
2058 | ||
2440a2c3 | 2059 | DPRINTF("%s %s\n", sn_info->name, sn_info->id_str); |
33b1db1c MK |
2060 | |
2061 | s->inode.vm_state_size = sn_info->vm_state_size; | |
2062 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
2063 | /* It appears that inode.tag does not require a NUL terminator, |
2064 | * which means this use of strncpy is ok. | |
2065 | */ | |
33b1db1c MK |
2066 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
2067 | /* we don't need to update entire object */ | |
2068 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
2069 | ||
2070 | /* refresh inode. */ | |
25af257d | 2071 | fd = connect_to_sdog(s); |
33b1db1c | 2072 | if (fd < 0) { |
cb595887 | 2073 | ret = fd; |
33b1db1c MK |
2074 | goto cleanup; |
2075 | } | |
2076 | ||
2077 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
0e7106d8 | 2078 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
33b1db1c | 2079 | if (ret < 0) { |
6daf194d | 2080 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
2081 | goto cleanup; |
2082 | } | |
2083 | ||
25af257d | 2084 | ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, |
1841f880 | 2085 | 1, s->inode.copy_policy); |
33b1db1c | 2086 | if (ret < 0) { |
6daf194d | 2087 | error_report("failed to create inode for snapshot. %s", |
33b1db1c | 2088 | strerror(errno)); |
33b1db1c MK |
2089 | goto cleanup; |
2090 | } | |
2091 | ||
7267c094 | 2092 | inode = (SheepdogInode *)g_malloc(datalen); |
33b1db1c MK |
2093 | |
2094 | ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid), | |
0e7106d8 | 2095 | s->inode.nr_copies, datalen, 0, s->cache_flags); |
33b1db1c MK |
2096 | |
2097 | if (ret < 0) { | |
6daf194d | 2098 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
2099 | goto cleanup; |
2100 | } | |
2101 | ||
2102 | memcpy(&s->inode, inode, datalen); | |
2440a2c3 | 2103 | DPRINTF("s->inode: name %s snap_id %x oid %x\n", |
33b1db1c MK |
2104 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
2105 | ||
2106 | cleanup: | |
2107 | closesocket(fd); | |
2108 | return ret; | |
2109 | } | |
2110 | ||
859e5553 LY |
2111 | /* |
2112 | * We implement rollback(loadvm) operation to the specified snapshot by | |
2113 | * 1) switch to the snapshot | |
2114 | * 2) rely on sd_create_branch to delete working VDI and | |
2115 | * 3) create a new working VDI based on the speicified snapshot | |
2116 | */ | |
33b1db1c MK |
2117 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
2118 | { | |
2119 | BDRVSheepdogState *s = bs->opaque; | |
2120 | BDRVSheepdogState *old_s; | |
9ff53a0e | 2121 | char tag[SD_MAX_VDI_TAG_LEN]; |
33b1db1c | 2122 | uint32_t snapid = 0; |
9ff53a0e | 2123 | int ret = 0; |
33b1db1c | 2124 | |
7267c094 | 2125 | old_s = g_malloc(sizeof(BDRVSheepdogState)); |
33b1db1c MK |
2126 | |
2127 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
2128 | ||
33b1db1c | 2129 | snapid = strtoul(snapshot_id, NULL, 10); |
3178e275 JM |
2130 | if (snapid) { |
2131 | tag[0] = 0; | |
2132 | } else { | |
b579ffb3 | 2133 | pstrcpy(tag, sizeof(tag), snapshot_id); |
33b1db1c MK |
2134 | } |
2135 | ||
9ff53a0e | 2136 | ret = reload_inode(s, snapid, tag); |
33b1db1c | 2137 | if (ret) { |
33b1db1c MK |
2138 | goto out; |
2139 | } | |
2140 | ||
cede621f LY |
2141 | ret = sd_create_branch(s); |
2142 | if (ret) { | |
33b1db1c MK |
2143 | goto out; |
2144 | } | |
2145 | ||
7267c094 | 2146 | g_free(old_s); |
33b1db1c MK |
2147 | |
2148 | return 0; | |
2149 | out: | |
2150 | /* recover bdrv_sd_state */ | |
2151 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 | 2152 | g_free(old_s); |
33b1db1c | 2153 | |
6daf194d | 2154 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
2155 | |
2156 | return ret; | |
2157 | } | |
2158 | ||
a89d89d3 WX |
2159 | static int sd_snapshot_delete(BlockDriverState *bs, |
2160 | const char *snapshot_id, | |
2161 | const char *name, | |
2162 | Error **errp) | |
33b1db1c MK |
2163 | { |
2164 | /* FIXME: Delete specified snapshot id. */ | |
2165 | return 0; | |
2166 | } | |
2167 | ||
33b1db1c MK |
2168 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
2169 | { | |
2170 | BDRVSheepdogState *s = bs->opaque; | |
2171 | SheepdogReq req; | |
2172 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
2173 | QEMUSnapshotInfo *sn_tab = NULL; | |
2174 | unsigned wlen, rlen; | |
2175 | int found = 0; | |
2176 | static SheepdogInode inode; | |
2177 | unsigned long *vdi_inuse; | |
2178 | unsigned int start_nr; | |
2179 | uint64_t hval; | |
2180 | uint32_t vid; | |
2181 | ||
7267c094 | 2182 | vdi_inuse = g_malloc(max); |
33b1db1c | 2183 | |
25af257d | 2184 | fd = connect_to_sdog(s); |
33b1db1c | 2185 | if (fd < 0) { |
cb595887 | 2186 | ret = fd; |
33b1db1c MK |
2187 | goto out; |
2188 | } | |
2189 | ||
2190 | rlen = max; | |
2191 | wlen = 0; | |
2192 | ||
2193 | memset(&req, 0, sizeof(req)); | |
2194 | ||
2195 | req.opcode = SD_OP_READ_VDIS; | |
2196 | req.data_length = max; | |
2197 | ||
2198 | ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); | |
2199 | ||
2200 | closesocket(fd); | |
2201 | if (ret) { | |
2202 | goto out; | |
2203 | } | |
2204 | ||
7267c094 | 2205 | sn_tab = g_malloc0(nr * sizeof(*sn_tab)); |
33b1db1c MK |
2206 | |
2207 | /* calculate a vdi id with hash function */ | |
2208 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2209 | start_nr = hval & (SD_NR_VDIS - 1); | |
2210 | ||
25af257d | 2211 | fd = connect_to_sdog(s); |
33b1db1c | 2212 | if (fd < 0) { |
cb595887 | 2213 | ret = fd; |
33b1db1c MK |
2214 | goto out; |
2215 | } | |
2216 | ||
2217 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2218 | if (!test_bit(vid, vdi_inuse)) { | |
2219 | break; | |
2220 | } | |
2221 | ||
2222 | /* we don't need to read entire object */ | |
2223 | ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid), | |
47622c44 | 2224 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2225 | s->cache_flags); |
33b1db1c MK |
2226 | |
2227 | if (ret) { | |
2228 | continue; | |
2229 | } | |
2230 | ||
2231 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2232 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2233 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2234 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2235 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2236 | ||
2237 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", | |
2238 | inode.snap_id); | |
3178e275 JM |
2239 | pstrcpy(sn_tab[found].name, |
2240 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2241 | inode.tag); | |
33b1db1c MK |
2242 | found++; |
2243 | } | |
2244 | } | |
2245 | ||
2246 | closesocket(fd); | |
2247 | out: | |
2248 | *psn_tab = sn_tab; | |
2249 | ||
7267c094 | 2250 | g_free(vdi_inuse); |
33b1db1c | 2251 | |
cb595887 MK |
2252 | if (ret < 0) { |
2253 | return ret; | |
2254 | } | |
2255 | ||
33b1db1c MK |
2256 | return found; |
2257 | } | |
2258 | ||
2259 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2260 | int64_t pos, int size, int load) | |
2261 | { | |
2f536801 MK |
2262 | bool create; |
2263 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2264 | unsigned int data_len; |
2265 | uint64_t vmstate_oid; | |
33b1db1c | 2266 | uint64_t offset; |
cede621f LY |
2267 | uint32_t vdi_index; |
2268 | uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id; | |
33b1db1c | 2269 | |
25af257d | 2270 | fd = connect_to_sdog(s); |
33b1db1c | 2271 | if (fd < 0) { |
cb595887 | 2272 | return fd; |
33b1db1c MK |
2273 | } |
2274 | ||
6f3c714e | 2275 | while (remaining) { |
33b1db1c MK |
2276 | vdi_index = pos / SD_DATA_OBJ_SIZE; |
2277 | offset = pos % SD_DATA_OBJ_SIZE; | |
2278 | ||
1f7a48de | 2279 | data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset); |
33b1db1c | 2280 | |
cede621f | 2281 | vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index); |
33b1db1c MK |
2282 | |
2283 | create = (offset == 0); | |
2284 | if (load) { | |
2285 | ret = read_object(fd, (char *)data, vmstate_oid, | |
47622c44 | 2286 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2287 | s->cache_flags); |
33b1db1c MK |
2288 | } else { |
2289 | ret = write_object(fd, (char *)data, vmstate_oid, | |
47622c44 | 2290 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2291 | s->cache_flags); |
33b1db1c MK |
2292 | } |
2293 | ||
2294 | if (ret < 0) { | |
6daf194d | 2295 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2296 | goto cleanup; |
2297 | } | |
2298 | ||
2299 | pos += data_len; | |
1f7a48de | 2300 | data += data_len; |
6f3c714e | 2301 | remaining -= data_len; |
33b1db1c | 2302 | } |
6f3c714e | 2303 | ret = size; |
33b1db1c MK |
2304 | cleanup: |
2305 | closesocket(fd); | |
2306 | return ret; | |
2307 | } | |
2308 | ||
cf8074b3 KW |
2309 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2310 | int64_t pos) | |
33b1db1c MK |
2311 | { |
2312 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2313 | void *buf; |
2314 | int ret; | |
33b1db1c | 2315 | |
cf8074b3 KW |
2316 | buf = qemu_blockalign(bs, qiov->size); |
2317 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2318 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2319 | qemu_vfree(buf); | |
2320 | ||
2321 | return ret; | |
33b1db1c MK |
2322 | } |
2323 | ||
2324 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
2325 | int64_t pos, int size) | |
2326 | { | |
2327 | BDRVSheepdogState *s = bs->opaque; | |
2328 | ||
2329 | return do_load_save_vmstate(s, data, pos, size, 1); | |
2330 | } | |
2331 | ||
2332 | ||
cac8f4a6 LY |
2333 | static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num, |
2334 | int nb_sectors) | |
2335 | { | |
2336 | SheepdogAIOCB *acb; | |
2337 | QEMUIOVector dummy; | |
2338 | BDRVSheepdogState *s = bs->opaque; | |
2339 | int ret; | |
2340 | ||
2341 | if (!s->discard_supported) { | |
2342 | return 0; | |
2343 | } | |
2344 | ||
2345 | acb = sd_aio_setup(bs, &dummy, sector_num, nb_sectors); | |
2346 | acb->aiocb_type = AIOCB_DISCARD_OBJ; | |
2347 | acb->aio_done_func = sd_finish_aiocb; | |
2348 | ||
2349 | ret = sd_co_rw_vector(acb); | |
2350 | if (ret <= 0) { | |
2351 | qemu_aio_release(acb); | |
2352 | return ret; | |
2353 | } | |
2354 | ||
2355 | qemu_coroutine_yield(); | |
2356 | ||
2357 | return acb->ret; | |
2358 | } | |
2359 | ||
b6b8a333 PB |
2360 | static coroutine_fn int64_t |
2361 | sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
2362 | int *pnum) | |
8d71c631 LY |
2363 | { |
2364 | BDRVSheepdogState *s = bs->opaque; | |
2365 | SheepdogInode *inode = &s->inode; | |
2366 | unsigned long start = sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE, | |
2367 | end = DIV_ROUND_UP((sector_num + nb_sectors) * | |
2368 | BDRV_SECTOR_SIZE, SD_DATA_OBJ_SIZE); | |
2369 | unsigned long idx; | |
4bc74be9 | 2370 | int64_t ret = BDRV_BLOCK_DATA; |
8d71c631 LY |
2371 | |
2372 | for (idx = start; idx < end; idx++) { | |
2373 | if (inode->data_vdi_id[idx] == 0) { | |
2374 | break; | |
2375 | } | |
2376 | } | |
2377 | if (idx == start) { | |
2378 | /* Get the longest length of unallocated sectors */ | |
2379 | ret = 0; | |
2380 | for (idx = start + 1; idx < end; idx++) { | |
2381 | if (inode->data_vdi_id[idx] != 0) { | |
2382 | break; | |
2383 | } | |
2384 | } | |
2385 | } | |
2386 | ||
2387 | *pnum = (idx - start) * SD_DATA_OBJ_SIZE / BDRV_SECTOR_SIZE; | |
2388 | if (*pnum > nb_sectors) { | |
2389 | *pnum = nb_sectors; | |
2390 | } | |
2391 | return ret; | |
2392 | } | |
2393 | ||
33b1db1c MK |
2394 | static QEMUOptionParameter sd_create_options[] = { |
2395 | { | |
2396 | .name = BLOCK_OPT_SIZE, | |
2397 | .type = OPT_SIZE, | |
2398 | .help = "Virtual disk size" | |
2399 | }, | |
2400 | { | |
2401 | .name = BLOCK_OPT_BACKING_FILE, | |
2402 | .type = OPT_STRING, | |
2403 | .help = "File name of a base image" | |
2404 | }, | |
a8e0fdd7 MK |
2405 | { |
2406 | .name = BLOCK_OPT_PREALLOC, | |
2407 | .type = OPT_STRING, | |
2408 | .help = "Preallocation mode (allowed values: off, full)" | |
2409 | }, | |
33b1db1c MK |
2410 | { NULL } |
2411 | }; | |
2412 | ||
5d6768e3 | 2413 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2414 | .format_name = "sheepdog", |
2415 | .protocol_name = "sheepdog", | |
2416 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2417 | .bdrv_needs_filename = true, |
33b1db1c MK |
2418 | .bdrv_file_open = sd_open, |
2419 | .bdrv_close = sd_close, | |
2420 | .bdrv_create = sd_create, | |
e4f5c1bf | 2421 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
33b1db1c MK |
2422 | .bdrv_getlength = sd_getlength, |
2423 | .bdrv_truncate = sd_truncate, | |
2424 | ||
2df46246 MK |
2425 | .bdrv_co_readv = sd_co_readv, |
2426 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2427 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
cac8f4a6 | 2428 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2429 | .bdrv_co_get_block_status = sd_co_get_block_status, |
33b1db1c MK |
2430 | |
2431 | .bdrv_snapshot_create = sd_snapshot_create, | |
2432 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2433 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2434 | .bdrv_snapshot_list = sd_snapshot_list, | |
2435 | ||
2436 | .bdrv_save_vmstate = sd_save_vmstate, | |
2437 | .bdrv_load_vmstate = sd_load_vmstate, | |
2438 | ||
2439 | .create_options = sd_create_options, | |
2440 | }; | |
2441 | ||
5d6768e3 MK |
2442 | static BlockDriver bdrv_sheepdog_tcp = { |
2443 | .format_name = "sheepdog", | |
2444 | .protocol_name = "sheepdog+tcp", | |
2445 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2446 | .bdrv_needs_filename = true, |
5d6768e3 MK |
2447 | .bdrv_file_open = sd_open, |
2448 | .bdrv_close = sd_close, | |
2449 | .bdrv_create = sd_create, | |
e4f5c1bf | 2450 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
5d6768e3 MK |
2451 | .bdrv_getlength = sd_getlength, |
2452 | .bdrv_truncate = sd_truncate, | |
2453 | ||
2454 | .bdrv_co_readv = sd_co_readv, | |
2455 | .bdrv_co_writev = sd_co_writev, | |
2456 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2457 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2458 | .bdrv_co_get_block_status = sd_co_get_block_status, |
5d6768e3 MK |
2459 | |
2460 | .bdrv_snapshot_create = sd_snapshot_create, | |
2461 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2462 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2463 | .bdrv_snapshot_list = sd_snapshot_list, | |
2464 | ||
2465 | .bdrv_save_vmstate = sd_save_vmstate, | |
2466 | .bdrv_load_vmstate = sd_load_vmstate, | |
2467 | ||
2468 | .create_options = sd_create_options, | |
2469 | }; | |
2470 | ||
1b8bbb46 MK |
2471 | static BlockDriver bdrv_sheepdog_unix = { |
2472 | .format_name = "sheepdog", | |
2473 | .protocol_name = "sheepdog+unix", | |
2474 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2475 | .bdrv_needs_filename = true, |
1b8bbb46 MK |
2476 | .bdrv_file_open = sd_open, |
2477 | .bdrv_close = sd_close, | |
2478 | .bdrv_create = sd_create, | |
3ac21627 | 2479 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
1b8bbb46 MK |
2480 | .bdrv_getlength = sd_getlength, |
2481 | .bdrv_truncate = sd_truncate, | |
2482 | ||
2483 | .bdrv_co_readv = sd_co_readv, | |
2484 | .bdrv_co_writev = sd_co_writev, | |
2485 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2486 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2487 | .bdrv_co_get_block_status = sd_co_get_block_status, |
1b8bbb46 MK |
2488 | |
2489 | .bdrv_snapshot_create = sd_snapshot_create, | |
2490 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2491 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2492 | .bdrv_snapshot_list = sd_snapshot_list, | |
2493 | ||
2494 | .bdrv_save_vmstate = sd_save_vmstate, | |
2495 | .bdrv_load_vmstate = sd_load_vmstate, | |
2496 | ||
2497 | .create_options = sd_create_options, | |
2498 | }; | |
2499 | ||
33b1db1c MK |
2500 | static void bdrv_sheepdog_init(void) |
2501 | { | |
2502 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 2503 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 2504 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
2505 | } |
2506 | block_init(bdrv_sheepdog_init); |