]>
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 |
876eb1b0 | 40 | #define SD_OP_GET_CLUSTER_DEFAULT 0x18 |
33b1db1c MK |
41 | |
42 | #define SD_FLAG_CMD_WRITE 0x01 | |
43 | #define SD_FLAG_CMD_COW 0x02 | |
0e7106d8 LY |
44 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
45 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
33b1db1c MK |
46 | |
47 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
48 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
49 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
50 | #define SD_RES_EIO 0x03 /* I/O error */ | |
51 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
52 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
53 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
54 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
55 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
56 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
57 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
58 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
59 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
60 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
61 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
62 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
63 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
64 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
65 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
66 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
67 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
68 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
69 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
70 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
71 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
fca23f0a | 72 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ |
6a0b5490 | 73 | #define SD_RES_READONLY 0x1A /* Object is read-only */ |
33b1db1c MK |
74 | |
75 | /* | |
76 | * Object ID rules | |
77 | * | |
78 | * 0 - 19 (20 bits): data object space | |
79 | * 20 - 31 (12 bits): reserved data object space | |
80 | * 32 - 55 (24 bits): vdi object space | |
81 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 82 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
83 | */ |
84 | ||
85 | #define VDI_SPACE_SHIFT 32 | |
86 | #define VDI_BIT (UINT64_C(1) << 63) | |
87 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
88 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
89 | #define MAX_CHILDREN 1024 | |
90 | #define SD_MAX_VDI_LEN 256 | |
91 | #define SD_MAX_VDI_TAG_LEN 256 | |
92 | #define SD_NR_VDIS (1U << 24) | |
93 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
94 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
876eb1b0 | 95 | #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22 |
b3af018f LY |
96 | /* |
97 | * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and | |
98 | * (SD_EC_MAX_STRIP - 1) for parity strips | |
99 | * | |
100 | * SD_MAX_COPIES is sum of number of data strips and parity strips. | |
101 | */ | |
102 | #define SD_EC_MAX_STRIP 16 | |
103 | #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1) | |
33b1db1c MK |
104 | |
105 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
106 | #define CURRENT_VDI_ID 0 | |
107 | ||
1dbfafed HM |
108 | #define LOCK_TYPE_NORMAL 0 |
109 | #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */ | |
110 | ||
33b1db1c MK |
111 | typedef struct SheepdogReq { |
112 | uint8_t proto_ver; | |
113 | uint8_t opcode; | |
114 | uint16_t flags; | |
115 | uint32_t epoch; | |
116 | uint32_t id; | |
117 | uint32_t data_length; | |
118 | uint32_t opcode_specific[8]; | |
119 | } SheepdogReq; | |
120 | ||
121 | typedef struct SheepdogRsp { | |
122 | uint8_t proto_ver; | |
123 | uint8_t opcode; | |
124 | uint16_t flags; | |
125 | uint32_t epoch; | |
126 | uint32_t id; | |
127 | uint32_t data_length; | |
128 | uint32_t result; | |
129 | uint32_t opcode_specific[7]; | |
130 | } SheepdogRsp; | |
131 | ||
132 | typedef struct SheepdogObjReq { | |
133 | uint8_t proto_ver; | |
134 | uint8_t opcode; | |
135 | uint16_t flags; | |
136 | uint32_t epoch; | |
137 | uint32_t id; | |
138 | uint32_t data_length; | |
139 | uint64_t oid; | |
140 | uint64_t cow_oid; | |
29a67f7e | 141 | uint8_t copies; |
1841f880 LY |
142 | uint8_t copy_policy; |
143 | uint8_t reserved[6]; | |
33b1db1c MK |
144 | uint64_t offset; |
145 | } SheepdogObjReq; | |
146 | ||
147 | typedef struct SheepdogObjRsp { | |
148 | uint8_t proto_ver; | |
149 | uint8_t opcode; | |
150 | uint16_t flags; | |
151 | uint32_t epoch; | |
152 | uint32_t id; | |
153 | uint32_t data_length; | |
154 | uint32_t result; | |
29a67f7e | 155 | uint8_t copies; |
1841f880 LY |
156 | uint8_t copy_policy; |
157 | uint8_t reserved[2]; | |
33b1db1c MK |
158 | uint32_t pad[6]; |
159 | } SheepdogObjRsp; | |
160 | ||
161 | typedef struct SheepdogVdiReq { | |
162 | uint8_t proto_ver; | |
163 | uint8_t opcode; | |
164 | uint16_t flags; | |
165 | uint32_t epoch; | |
166 | uint32_t id; | |
167 | uint32_t data_length; | |
168 | uint64_t vdi_size; | |
9f23fce7 | 169 | uint32_t base_vdi_id; |
29a67f7e | 170 | uint8_t copies; |
1841f880 | 171 | uint8_t copy_policy; |
876eb1b0 TI |
172 | uint8_t store_policy; |
173 | uint8_t block_size_shift; | |
33b1db1c | 174 | uint32_t snapid; |
1dbfafed HM |
175 | uint32_t type; |
176 | uint32_t pad[2]; | |
33b1db1c MK |
177 | } SheepdogVdiReq; |
178 | ||
179 | typedef struct SheepdogVdiRsp { | |
180 | uint8_t proto_ver; | |
181 | uint8_t opcode; | |
182 | uint16_t flags; | |
183 | uint32_t epoch; | |
184 | uint32_t id; | |
185 | uint32_t data_length; | |
186 | uint32_t result; | |
187 | uint32_t rsvd; | |
188 | uint32_t vdi_id; | |
189 | uint32_t pad[5]; | |
190 | } SheepdogVdiRsp; | |
191 | ||
876eb1b0 TI |
192 | typedef struct SheepdogClusterRsp { |
193 | uint8_t proto_ver; | |
194 | uint8_t opcode; | |
195 | uint16_t flags; | |
196 | uint32_t epoch; | |
197 | uint32_t id; | |
198 | uint32_t data_length; | |
199 | uint32_t result; | |
200 | uint8_t nr_copies; | |
201 | uint8_t copy_policy; | |
202 | uint8_t block_size_shift; | |
203 | uint8_t __pad1; | |
204 | uint32_t __pad2[6]; | |
205 | } SheepdogClusterRsp; | |
206 | ||
33b1db1c MK |
207 | typedef struct SheepdogInode { |
208 | char name[SD_MAX_VDI_LEN]; | |
209 | char tag[SD_MAX_VDI_TAG_LEN]; | |
210 | uint64_t ctime; | |
211 | uint64_t snap_ctime; | |
212 | uint64_t vm_clock_nsec; | |
213 | uint64_t vdi_size; | |
214 | uint64_t vm_state_size; | |
215 | uint16_t copy_policy; | |
216 | uint8_t nr_copies; | |
217 | uint8_t block_size_shift; | |
218 | uint32_t snap_id; | |
219 | uint32_t vdi_id; | |
220 | uint32_t parent_vdi_id; | |
221 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
222 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
223 | } SheepdogInode; | |
224 | ||
5d039bab HM |
225 | #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id) |
226 | ||
33b1db1c MK |
227 | /* |
228 | * 64 bit FNV-1a non-zero initial basis | |
229 | */ | |
230 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
231 | ||
232 | /* | |
233 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
234 | */ | |
235 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
236 | { | |
237 | unsigned char *bp = buf; | |
238 | unsigned char *be = bp + len; | |
239 | while (bp < be) { | |
240 | hval ^= (uint64_t) *bp++; | |
241 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
242 | (hval << 7) + (hval << 8) + (hval << 40); | |
243 | } | |
244 | return hval; | |
245 | } | |
246 | ||
2f536801 | 247 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
248 | { |
249 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
250 | } | |
251 | ||
2f536801 | 252 | static inline bool is_data_obj(uint64_t oid) |
33b1db1c MK |
253 | { |
254 | return !(VDI_BIT & oid); | |
255 | } | |
256 | ||
257 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
258 | { | |
259 | return oid & (MAX_DATA_OBJS - 1); | |
260 | } | |
261 | ||
72e0996c MK |
262 | static inline uint32_t oid_to_vid(uint64_t oid) |
263 | { | |
264 | return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT; | |
265 | } | |
266 | ||
33b1db1c MK |
267 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) |
268 | { | |
269 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
270 | } | |
271 | ||
272 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
273 | { | |
274 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
275 | } | |
276 | ||
277 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
278 | { | |
279 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
280 | } | |
281 | ||
2f536801 | 282 | static inline bool is_snapshot(struct SheepdogInode *inode) |
33b1db1c MK |
283 | { |
284 | return !!inode->snap_ctime; | |
285 | } | |
286 | ||
2440a2c3 | 287 | #undef DPRINTF |
33b1db1c | 288 | #ifdef DEBUG_SDOG |
2440a2c3 | 289 | #define DPRINTF(fmt, args...) \ |
33b1db1c MK |
290 | do { \ |
291 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
292 | } while (0) | |
293 | #else | |
2440a2c3 | 294 | #define DPRINTF(fmt, args...) |
33b1db1c MK |
295 | #endif |
296 | ||
297 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
298 | ||
299 | typedef struct AIOReq { | |
300 | SheepdogAIOCB *aiocb; | |
301 | unsigned int iov_offset; | |
302 | ||
303 | uint64_t oid; | |
304 | uint64_t base_oid; | |
305 | uint64_t offset; | |
306 | unsigned int data_len; | |
307 | uint8_t flags; | |
308 | uint32_t id; | |
b544c1ab | 309 | bool create; |
33b1db1c | 310 | |
c292ee6a | 311 | QLIST_ENTRY(AIOReq) aio_siblings; |
33b1db1c MK |
312 | } AIOReq; |
313 | ||
314 | enum AIOCBState { | |
315 | AIOCB_WRITE_UDATA, | |
316 | AIOCB_READ_UDATA, | |
47783072 | 317 | AIOCB_FLUSH_CACHE, |
cac8f4a6 | 318 | AIOCB_DISCARD_OBJ, |
33b1db1c MK |
319 | }; |
320 | ||
498f2140 | 321 | #define AIOCBOverlapping(x, y) \ |
6a55c82c HM |
322 | (!(x->max_affect_data_idx < y->min_affect_data_idx \ |
323 | || y->max_affect_data_idx < x->min_affect_data_idx)) | |
324 | ||
33b1db1c | 325 | struct SheepdogAIOCB { |
7c84b1b8 | 326 | BlockAIOCB common; |
33b1db1c MK |
327 | |
328 | QEMUIOVector *qiov; | |
329 | ||
330 | int64_t sector_num; | |
331 | int nb_sectors; | |
332 | ||
333 | int ret; | |
334 | enum AIOCBState aiocb_type; | |
335 | ||
2df46246 | 336 | Coroutine *coroutine; |
33b1db1c MK |
337 | void (*aio_done_func)(SheepdogAIOCB *); |
338 | ||
35200687 | 339 | bool cancelable; |
1d732d7d | 340 | int nr_pending; |
6a55c82c HM |
341 | |
342 | uint32_t min_affect_data_idx; | |
343 | uint32_t max_affect_data_idx; | |
344 | ||
498f2140 HM |
345 | /* |
346 | * The difference between affect_data_idx and dirty_data_idx: | |
347 | * affect_data_idx represents range of index of all request types. | |
348 | * dirty_data_idx represents range of index updated by COW requests. | |
349 | * dirty_data_idx is used for updating an inode object. | |
350 | */ | |
351 | uint32_t min_dirty_data_idx; | |
352 | uint32_t max_dirty_data_idx; | |
353 | ||
6a55c82c | 354 | QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings; |
33b1db1c MK |
355 | }; |
356 | ||
357 | typedef struct BDRVSheepdogState { | |
011603ca | 358 | BlockDriverState *bs; |
84390bed | 359 | AioContext *aio_context; |
011603ca | 360 | |
33b1db1c MK |
361 | SheepdogInode inode; |
362 | ||
33b1db1c | 363 | char name[SD_MAX_VDI_LEN]; |
2f536801 | 364 | bool is_snapshot; |
0e7106d8 | 365 | uint32_t cache_flags; |
cac8f4a6 | 366 | bool discard_supported; |
33b1db1c | 367 | |
25af257d | 368 | char *host_spec; |
1b8bbb46 | 369 | bool is_unix; |
33b1db1c MK |
370 | int fd; |
371 | ||
2df46246 MK |
372 | CoMutex lock; |
373 | Coroutine *co_send; | |
374 | Coroutine *co_recv; | |
375 | ||
33b1db1c | 376 | uint32_t aioreq_seq_num; |
011603ca MK |
377 | |
378 | /* Every aio request must be linked to either of these queues. */ | |
c292ee6a | 379 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
011603ca | 380 | QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head; |
6a55c82c | 381 | |
498f2140 | 382 | CoQueue overlapping_queue; |
6a55c82c | 383 | QLIST_HEAD(inflight_aiocb_head, SheepdogAIOCB) inflight_aiocb_head; |
33b1db1c MK |
384 | } BDRVSheepdogState; |
385 | ||
4da65c80 LY |
386 | typedef struct BDRVSheepdogReopenState { |
387 | int fd; | |
388 | int cache_flags; | |
389 | } BDRVSheepdogReopenState; | |
390 | ||
33b1db1c MK |
391 | static const char * sd_strerror(int err) |
392 | { | |
393 | int i; | |
394 | ||
395 | static const struct { | |
396 | int err; | |
397 | const char *desc; | |
398 | } errors[] = { | |
399 | {SD_RES_SUCCESS, "Success"}, | |
400 | {SD_RES_UNKNOWN, "Unknown error"}, | |
401 | {SD_RES_NO_OBJ, "No object found"}, | |
402 | {SD_RES_EIO, "I/O error"}, | |
403 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
404 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
405 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
406 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
407 | {SD_RES_NO_VDI, "No vdi found"}, | |
408 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
409 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
410 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
411 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
412 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
413 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
414 | {SD_RES_STARTUP, "The system is still booting"}, | |
415 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
416 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
417 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
418 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
419 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
420 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
421 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
422 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
423 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
fca23f0a | 424 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, |
6a0b5490 | 425 | {SD_RES_READONLY, "Object is read-only"}, |
33b1db1c MK |
426 | }; |
427 | ||
428 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
429 | if (errors[i].err == err) { | |
430 | return errors[i].desc; | |
431 | } | |
432 | } | |
433 | ||
434 | return "Invalid error code"; | |
435 | } | |
436 | ||
437 | /* | |
438 | * Sheepdog I/O handling: | |
439 | * | |
2df46246 | 440 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
c292ee6a | 441 | * link the requests to the inflight_list in the |
2df46246 MK |
442 | * BDRVSheepdogState. The function exits without waiting for |
443 | * receiving the response. | |
33b1db1c | 444 | * |
2df46246 | 445 | * 2. We receive the response in aio_read_response, the fd handler to |
33b1db1c MK |
446 | * the sheepdog connection. If metadata update is needed, we send |
447 | * the write request to the vdi object in sd_write_done, the write | |
2df46246 MK |
448 | * completion function. We switch back to sd_co_readv/writev after |
449 | * all the requests belonging to the AIOCB are finished. | |
33b1db1c MK |
450 | */ |
451 | ||
452 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
453 | uint64_t oid, unsigned int data_len, | |
b544c1ab | 454 | uint64_t offset, uint8_t flags, bool create, |
33b1db1c MK |
455 | uint64_t base_oid, unsigned int iov_offset) |
456 | { | |
457 | AIOReq *aio_req; | |
458 | ||
7267c094 | 459 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
460 | aio_req->aiocb = acb; |
461 | aio_req->iov_offset = iov_offset; | |
462 | aio_req->oid = oid; | |
463 | aio_req->base_oid = base_oid; | |
464 | aio_req->offset = offset; | |
465 | aio_req->data_len = data_len; | |
466 | aio_req->flags = flags; | |
467 | aio_req->id = s->aioreq_seq_num++; | |
b544c1ab | 468 | aio_req->create = create; |
33b1db1c | 469 | |
1d732d7d | 470 | acb->nr_pending++; |
33b1db1c MK |
471 | return aio_req; |
472 | } | |
473 | ||
1d732d7d | 474 | static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
33b1db1c MK |
475 | { |
476 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1d732d7d | 477 | |
35200687 | 478 | acb->cancelable = false; |
c292ee6a | 479 | QLIST_REMOVE(aio_req, aio_siblings); |
7267c094 | 480 | g_free(aio_req); |
33b1db1c | 481 | |
1d732d7d | 482 | acb->nr_pending--; |
33b1db1c MK |
483 | } |
484 | ||
d8716b41 | 485 | static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
33b1db1c | 486 | { |
35200687 | 487 | qemu_coroutine_enter(acb->coroutine, NULL); |
8007429a | 488 | qemu_aio_unref(acb); |
33b1db1c MK |
489 | } |
490 | ||
35200687 MK |
491 | /* |
492 | * Check whether the specified acb can be canceled | |
493 | * | |
494 | * We can cancel aio when any request belonging to the acb is: | |
495 | * - Not processed by the sheepdog server. | |
496 | * - Not linked to the inflight queue. | |
497 | */ | |
498 | static bool sd_acb_cancelable(const SheepdogAIOCB *acb) | |
499 | { | |
500 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
501 | AIOReq *aioreq; | |
502 | ||
503 | if (!acb->cancelable) { | |
504 | return false; | |
505 | } | |
506 | ||
507 | QLIST_FOREACH(aioreq, &s->inflight_aio_head, aio_siblings) { | |
508 | if (aioreq->aiocb == acb) { | |
509 | return false; | |
510 | } | |
511 | } | |
512 | ||
513 | return true; | |
514 | } | |
515 | ||
7c84b1b8 | 516 | static void sd_aio_cancel(BlockAIOCB *blockacb) |
33b1db1c MK |
517 | { |
518 | SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; | |
35200687 MK |
519 | BDRVSheepdogState *s = acb->common.bs->opaque; |
520 | AIOReq *aioreq, *next; | |
6d24b4df FZ |
521 | |
522 | if (sd_acb_cancelable(acb)) { | |
6a55c82c | 523 | /* Remove outstanding requests from failed queue. */ |
6d24b4df FZ |
524 | QLIST_FOREACH_SAFE(aioreq, &s->failed_aio_head, aio_siblings, |
525 | next) { | |
526 | if (aioreq->aiocb == acb) { | |
527 | free_aio_req(s, aioreq); | |
35200687 | 528 | } |
6d24b4df | 529 | } |
35200687 | 530 | |
6d24b4df FZ |
531 | assert(acb->nr_pending == 0); |
532 | if (acb->common.cb) { | |
533 | acb->common.cb(acb->common.opaque, -ECANCELED); | |
35200687 | 534 | } |
6d24b4df | 535 | sd_finish_aiocb(acb); |
35200687 | 536 | } |
33b1db1c MK |
537 | } |
538 | ||
d7331bed | 539 | static const AIOCBInfo sd_aiocb_info = { |
6d24b4df FZ |
540 | .aiocb_size = sizeof(SheepdogAIOCB), |
541 | .cancel_async = sd_aio_cancel, | |
33b1db1c MK |
542 | }; |
543 | ||
544 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
f700f8e3 | 545 | int64_t sector_num, int nb_sectors) |
33b1db1c MK |
546 | { |
547 | SheepdogAIOCB *acb; | |
6a55c82c HM |
548 | uint32_t object_size; |
549 | BDRVSheepdogState *s = bs->opaque; | |
550 | ||
551 | object_size = (UINT32_C(1) << s->inode.block_size_shift); | |
33b1db1c | 552 | |
f700f8e3 | 553 | acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); |
33b1db1c MK |
554 | |
555 | acb->qiov = qiov; | |
556 | ||
557 | acb->sector_num = sector_num; | |
558 | acb->nb_sectors = nb_sectors; | |
559 | ||
560 | acb->aio_done_func = NULL; | |
35200687 | 561 | acb->cancelable = true; |
2df46246 | 562 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c | 563 | acb->ret = 0; |
1d732d7d | 564 | acb->nr_pending = 0; |
6a55c82c HM |
565 | |
566 | acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
567 | acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE + | |
568 | acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size; | |
569 | ||
498f2140 HM |
570 | acb->min_dirty_data_idx = UINT32_MAX; |
571 | acb->max_dirty_data_idx = 0; | |
572 | ||
33b1db1c MK |
573 | return acb; |
574 | } | |
575 | ||
833a7cc3 | 576 | /* Return -EIO in case of error, file descriptor on success */ |
dfb12bf8 | 577 | static int connect_to_sdog(BDRVSheepdogState *s, Error **errp) |
33b1db1c | 578 | { |
25af257d | 579 | int fd; |
33b1db1c | 580 | |
1b8bbb46 | 581 | if (s->is_unix) { |
dfb12bf8 | 582 | fd = unix_connect(s->host_spec, errp); |
1b8bbb46 | 583 | } else { |
dfb12bf8 | 584 | fd = inet_connect(s->host_spec, errp); |
1b8bbb46 | 585 | |
dfb12bf8 | 586 | if (fd >= 0) { |
1b8bbb46 MK |
587 | int ret = socket_set_nodelay(fd); |
588 | if (ret < 0) { | |
589 | error_report("%s", strerror(errno)); | |
590 | } | |
591 | } | |
592 | } | |
33b1db1c | 593 | |
dfb12bf8 | 594 | if (fd >= 0) { |
f9e8cacc | 595 | qemu_set_nonblock(fd); |
833a7cc3 LY |
596 | } else { |
597 | fd = -EIO; | |
33b1db1c MK |
598 | } |
599 | ||
33b1db1c MK |
600 | return fd; |
601 | } | |
602 | ||
833a7cc3 | 603 | /* Return 0 on success and -errno in case of error */ |
e0d93a89 MK |
604 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
605 | unsigned int *wlen) | |
47622c44 LY |
606 | { |
607 | int ret; | |
608 | ||
609 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
80731d9d | 610 | if (ret != sizeof(*hdr)) { |
47622c44 | 611 | error_report("failed to send a req, %s", strerror(errno)); |
833a7cc3 | 612 | ret = -socket_error(); |
eb092180 | 613 | return ret; |
47622c44 LY |
614 | } |
615 | ||
616 | ret = qemu_co_send(sockfd, data, *wlen); | |
80731d9d | 617 | if (ret != *wlen) { |
833a7cc3 | 618 | ret = -socket_error(); |
47622c44 LY |
619 | error_report("failed to send a req, %s", strerror(errno)); |
620 | } | |
621 | ||
622 | return ret; | |
623 | } | |
e0d93a89 | 624 | |
2dfcca3b MK |
625 | static void restart_co_req(void *opaque) |
626 | { | |
627 | Coroutine *co = opaque; | |
628 | ||
629 | qemu_coroutine_enter(co, NULL); | |
630 | } | |
631 | ||
cddd4ac7 MK |
632 | typedef struct SheepdogReqCo { |
633 | int sockfd; | |
84390bed | 634 | AioContext *aio_context; |
cddd4ac7 MK |
635 | SheepdogReq *hdr; |
636 | void *data; | |
637 | unsigned int *wlen; | |
638 | unsigned int *rlen; | |
639 | int ret; | |
640 | bool finished; | |
641 | } SheepdogReqCo; | |
642 | ||
643 | static coroutine_fn void do_co_req(void *opaque) | |
47622c44 LY |
644 | { |
645 | int ret; | |
2dfcca3b | 646 | Coroutine *co; |
cddd4ac7 MK |
647 | SheepdogReqCo *srco = opaque; |
648 | int sockfd = srco->sockfd; | |
649 | SheepdogReq *hdr = srco->hdr; | |
650 | void *data = srco->data; | |
651 | unsigned int *wlen = srco->wlen; | |
652 | unsigned int *rlen = srco->rlen; | |
2dfcca3b MK |
653 | |
654 | co = qemu_coroutine_self(); | |
84390bed | 655 | aio_set_fd_handler(srco->aio_context, sockfd, NULL, restart_co_req, co); |
47622c44 | 656 | |
47622c44 LY |
657 | ret = send_co_req(sockfd, hdr, data, wlen); |
658 | if (ret < 0) { | |
659 | goto out; | |
660 | } | |
661 | ||
84390bed | 662 | aio_set_fd_handler(srco->aio_context, sockfd, restart_co_req, NULL, co); |
2dfcca3b | 663 | |
47622c44 | 664 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); |
80731d9d | 665 | if (ret != sizeof(*hdr)) { |
47622c44 | 666 | error_report("failed to get a rsp, %s", strerror(errno)); |
cb595887 | 667 | ret = -errno; |
47622c44 LY |
668 | goto out; |
669 | } | |
670 | ||
671 | if (*rlen > hdr->data_length) { | |
672 | *rlen = hdr->data_length; | |
673 | } | |
674 | ||
675 | if (*rlen) { | |
676 | ret = qemu_co_recv(sockfd, data, *rlen); | |
80731d9d | 677 | if (ret != *rlen) { |
47622c44 | 678 | error_report("failed to get the data, %s", strerror(errno)); |
cb595887 | 679 | ret = -errno; |
47622c44 LY |
680 | goto out; |
681 | } | |
682 | } | |
683 | ret = 0; | |
684 | out: | |
ed9ba724 MK |
685 | /* there is at most one request for this sockfd, so it is safe to |
686 | * set each handler to NULL. */ | |
84390bed | 687 | aio_set_fd_handler(srco->aio_context, sockfd, NULL, NULL, NULL); |
cddd4ac7 MK |
688 | |
689 | srco->ret = ret; | |
690 | srco->finished = true; | |
691 | } | |
692 | ||
833a7cc3 LY |
693 | /* |
694 | * Send the request to the sheep in a synchronous manner. | |
695 | * | |
696 | * Return 0 on success, -errno in case of error. | |
697 | */ | |
84390bed SH |
698 | static int do_req(int sockfd, AioContext *aio_context, SheepdogReq *hdr, |
699 | void *data, unsigned int *wlen, unsigned int *rlen) | |
cddd4ac7 MK |
700 | { |
701 | Coroutine *co; | |
702 | SheepdogReqCo srco = { | |
703 | .sockfd = sockfd, | |
84390bed | 704 | .aio_context = aio_context, |
cddd4ac7 MK |
705 | .hdr = hdr, |
706 | .data = data, | |
707 | .wlen = wlen, | |
708 | .rlen = rlen, | |
709 | .ret = 0, | |
710 | .finished = false, | |
711 | }; | |
712 | ||
713 | if (qemu_in_coroutine()) { | |
714 | do_co_req(&srco); | |
715 | } else { | |
716 | co = qemu_coroutine_create(do_co_req); | |
717 | qemu_coroutine_enter(co, &srco); | |
718 | while (!srco.finished) { | |
84390bed | 719 | aio_poll(aio_context, true); |
cddd4ac7 MK |
720 | } |
721 | } | |
722 | ||
723 | return srco.ret; | |
47622c44 LY |
724 | } |
725 | ||
a37dcdf9 | 726 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
727 | struct iovec *iov, int niov, |
728 | enum AIOCBState aiocb_type); | |
a37dcdf9 | 729 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req); |
72e0996c | 730 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag); |
356b4ca2 | 731 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp); |
011603ca | 732 | static void co_write_request(void *opaque); |
7dc1cde0 | 733 | |
011603ca MK |
734 | static coroutine_fn void reconnect_to_sdog(void *opaque) |
735 | { | |
736 | BDRVSheepdogState *s = opaque; | |
737 | AIOReq *aio_req, *next; | |
738 | ||
84390bed | 739 | aio_set_fd_handler(s->aio_context, s->fd, NULL, NULL, NULL); |
011603ca MK |
740 | close(s->fd); |
741 | s->fd = -1; | |
742 | ||
743 | /* Wait for outstanding write requests to be completed. */ | |
744 | while (s->co_send != NULL) { | |
745 | co_write_request(opaque); | |
746 | } | |
747 | ||
748 | /* Try to reconnect the sheepdog server every one second. */ | |
749 | while (s->fd < 0) { | |
a780dea0 | 750 | Error *local_err = NULL; |
356b4ca2 | 751 | s->fd = get_sheep_fd(s, &local_err); |
011603ca MK |
752 | if (s->fd < 0) { |
753 | DPRINTF("Wait for connection to be established\n"); | |
565f65d2 | 754 | error_report_err(local_err); |
011603ca MK |
755 | co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME, |
756 | 1000000000ULL); | |
757 | } | |
758 | }; | |
759 | ||
760 | /* | |
761 | * Now we have to resend all the request in the inflight queue. However, | |
762 | * resend_aioreq() can yield and newly created requests can be added to the | |
763 | * inflight queue before the coroutine is resumed. To avoid mixing them, we | |
764 | * have to move all the inflight requests to the failed queue before | |
765 | * resend_aioreq() is called. | |
766 | */ | |
767 | QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) { | |
768 | QLIST_REMOVE(aio_req, aio_siblings); | |
769 | QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings); | |
770 | } | |
771 | ||
772 | /* Resend all the failed aio requests. */ | |
773 | while (!QLIST_EMPTY(&s->failed_aio_head)) { | |
774 | aio_req = QLIST_FIRST(&s->failed_aio_head); | |
775 | QLIST_REMOVE(aio_req, aio_siblings); | |
776 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
777 | resend_aioreq(s, aio_req); | |
778 | } | |
779 | } | |
780 | ||
33b1db1c MK |
781 | /* |
782 | * Receive responses of the I/O requests. | |
783 | * | |
784 | * This function is registered as a fd handler, and called from the | |
785 | * main loop when s->fd is ready for reading responses. | |
786 | */ | |
d8716b41 | 787 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
788 | { |
789 | SheepdogObjRsp rsp; | |
790 | BDRVSheepdogState *s = opaque; | |
791 | int fd = s->fd; | |
792 | int ret; | |
793 | AIOReq *aio_req = NULL; | |
794 | SheepdogAIOCB *acb; | |
cac8f4a6 | 795 | uint64_t idx; |
33b1db1c | 796 | |
33b1db1c | 797 | /* read a header */ |
8c5135f9 | 798 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
80731d9d | 799 | if (ret != sizeof(rsp)) { |
6daf194d | 800 | error_report("failed to get the header, %s", strerror(errno)); |
011603ca | 801 | goto err; |
33b1db1c MK |
802 | } |
803 | ||
c292ee6a MK |
804 | /* find the right aio_req from the inflight aio list */ |
805 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
33b1db1c MK |
806 | if (aio_req->id == rsp.id) { |
807 | break; | |
808 | } | |
809 | } | |
810 | if (!aio_req) { | |
6daf194d | 811 | error_report("cannot find aio_req %x", rsp.id); |
011603ca | 812 | goto err; |
33b1db1c MK |
813 | } |
814 | ||
815 | acb = aio_req->aiocb; | |
816 | ||
817 | switch (acb->aiocb_type) { | |
818 | case AIOCB_WRITE_UDATA: | |
6d1acda8 MK |
819 | /* this coroutine context is no longer suitable for co_recv |
820 | * because we may send data to update vdi objects */ | |
821 | s->co_recv = NULL; | |
33b1db1c MK |
822 | if (!is_data_obj(aio_req->oid)) { |
823 | break; | |
824 | } | |
825 | idx = data_oid_to_idx(aio_req->oid); | |
826 | ||
b544c1ab | 827 | if (aio_req->create) { |
33b1db1c MK |
828 | /* |
829 | * If the object is newly created one, we need to update | |
830 | * the vdi object (metadata object). min_dirty_data_idx | |
831 | * and max_dirty_data_idx are changed to include updated | |
832 | * index between them. | |
833 | */ | |
bd751f22 LY |
834 | if (rsp.result == SD_RES_SUCCESS) { |
835 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
498f2140 HM |
836 | acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx); |
837 | acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx); | |
bd751f22 | 838 | } |
33b1db1c MK |
839 | } |
840 | break; | |
841 | case AIOCB_READ_UDATA: | |
2fc8ae1d MT |
842 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
843 | aio_req->iov_offset, rsp.data_length); | |
80731d9d | 844 | if (ret != rsp.data_length) { |
6daf194d | 845 | error_report("failed to get the data, %s", strerror(errno)); |
011603ca | 846 | goto err; |
33b1db1c MK |
847 | } |
848 | break; | |
47783072 LY |
849 | case AIOCB_FLUSH_CACHE: |
850 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
2440a2c3 | 851 | DPRINTF("disable cache since the server doesn't support it\n"); |
47783072 LY |
852 | s->cache_flags = SD_FLAG_CMD_DIRECT; |
853 | rsp.result = SD_RES_SUCCESS; | |
854 | } | |
855 | break; | |
cac8f4a6 LY |
856 | case AIOCB_DISCARD_OBJ: |
857 | switch (rsp.result) { | |
858 | case SD_RES_INVALID_PARMS: | |
859 | error_report("sheep(%s) doesn't support discard command", | |
860 | s->host_spec); | |
861 | rsp.result = SD_RES_SUCCESS; | |
862 | s->discard_supported = false; | |
863 | break; | |
864 | case SD_RES_SUCCESS: | |
865 | idx = data_oid_to_idx(aio_req->oid); | |
866 | s->inode.data_vdi_id[idx] = 0; | |
867 | break; | |
868 | default: | |
869 | break; | |
870 | } | |
33b1db1c MK |
871 | } |
872 | ||
13c31de2 MK |
873 | switch (rsp.result) { |
874 | case SD_RES_SUCCESS: | |
875 | break; | |
876 | case SD_RES_READONLY: | |
72e0996c MK |
877 | if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) { |
878 | ret = reload_inode(s, 0, ""); | |
879 | if (ret < 0) { | |
011603ca | 880 | goto err; |
72e0996c MK |
881 | } |
882 | } | |
72e0996c MK |
883 | if (is_data_obj(aio_req->oid)) { |
884 | aio_req->oid = vid_to_data_oid(s->inode.vdi_id, | |
885 | data_oid_to_idx(aio_req->oid)); | |
886 | } else { | |
887 | aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id); | |
888 | } | |
a37dcdf9 MK |
889 | resend_aioreq(s, aio_req); |
890 | goto out; | |
13c31de2 | 891 | default: |
33b1db1c | 892 | acb->ret = -EIO; |
6daf194d | 893 | error_report("%s", sd_strerror(rsp.result)); |
13c31de2 | 894 | break; |
33b1db1c MK |
895 | } |
896 | ||
1d732d7d MK |
897 | free_aio_req(s, aio_req); |
898 | if (!acb->nr_pending) { | |
33b1db1c MK |
899 | /* |
900 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 901 | * we can switch back to sd_co_readv/writev now. |
33b1db1c MK |
902 | */ |
903 | acb->aio_done_func(acb); | |
904 | } | |
2df46246 MK |
905 | out: |
906 | s->co_recv = NULL; | |
011603ca MK |
907 | return; |
908 | err: | |
909 | s->co_recv = NULL; | |
910 | reconnect_to_sdog(opaque); | |
2df46246 MK |
911 | } |
912 | ||
913 | static void co_read_response(void *opaque) | |
914 | { | |
915 | BDRVSheepdogState *s = opaque; | |
916 | ||
917 | if (!s->co_recv) { | |
918 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
919 | } | |
920 | ||
921 | qemu_coroutine_enter(s->co_recv, opaque); | |
922 | } | |
923 | ||
924 | static void co_write_request(void *opaque) | |
925 | { | |
926 | BDRVSheepdogState *s = opaque; | |
927 | ||
928 | qemu_coroutine_enter(s->co_send, NULL); | |
33b1db1c MK |
929 | } |
930 | ||
33b1db1c | 931 | /* |
dc6fb73d | 932 | * Return a socket descriptor to read/write objects. |
33b1db1c | 933 | * |
dc6fb73d | 934 | * We cannot use this descriptor for other operations because |
33b1db1c MK |
935 | * the block driver may be on waiting response from the server. |
936 | */ | |
356b4ca2 | 937 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp) |
33b1db1c | 938 | { |
1b8bbb46 | 939 | int fd; |
33b1db1c | 940 | |
356b4ca2 | 941 | fd = connect_to_sdog(s, errp); |
33b1db1c | 942 | if (fd < 0) { |
cb595887 | 943 | return fd; |
33b1db1c MK |
944 | } |
945 | ||
84390bed | 946 | aio_set_fd_handler(s->aio_context, fd, co_read_response, NULL, s); |
33b1db1c MK |
947 | return fd; |
948 | } | |
949 | ||
5d6768e3 MK |
950 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
951 | char *vdi, uint32_t *snapid, char *tag) | |
952 | { | |
953 | URI *uri; | |
954 | QueryParams *qp = NULL; | |
955 | int ret = 0; | |
956 | ||
957 | uri = uri_parse(filename); | |
958 | if (!uri) { | |
959 | return -EINVAL; | |
960 | } | |
961 | ||
1b8bbb46 MK |
962 | /* transport */ |
963 | if (!strcmp(uri->scheme, "sheepdog")) { | |
964 | s->is_unix = false; | |
965 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
966 | s->is_unix = false; | |
967 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
968 | s->is_unix = true; | |
969 | } else { | |
970 | ret = -EINVAL; | |
971 | goto out; | |
972 | } | |
973 | ||
5d6768e3 MK |
974 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
975 | ret = -EINVAL; | |
976 | goto out; | |
977 | } | |
978 | pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1); | |
979 | ||
1b8bbb46 MK |
980 | qp = query_params_parse(uri->query); |
981 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
982 | ret = -EINVAL; | |
983 | goto out; | |
984 | } | |
985 | ||
986 | if (s->is_unix) { | |
987 | /* sheepdog+unix:///vdiname?socket=path */ | |
988 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
989 | ret = -EINVAL; | |
990 | goto out; | |
991 | } | |
992 | s->host_spec = g_strdup(qp->p[0].value); | |
993 | } else { | |
994 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
995 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
996 | uri->port ?: SD_DEFAULT_PORT); | |
997 | } | |
5d6768e3 MK |
998 | |
999 | /* snapshot tag */ | |
1000 | if (uri->fragment) { | |
1001 | *snapid = strtoul(uri->fragment, NULL, 10); | |
1002 | if (*snapid == 0) { | |
1003 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); | |
1004 | } | |
1005 | } else { | |
1006 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
1007 | } | |
1008 | ||
1009 | out: | |
1010 | if (qp) { | |
1011 | query_params_free(qp); | |
1012 | } | |
1013 | uri_free(uri); | |
1014 | return ret; | |
1015 | } | |
1016 | ||
33b1db1c | 1017 | /* |
5d6768e3 | 1018 | * Parse a filename (old syntax) |
33b1db1c MK |
1019 | * |
1020 | * filename must be one of the following formats: | |
1021 | * 1. [vdiname] | |
1022 | * 2. [vdiname]:[snapid] | |
1023 | * 3. [vdiname]:[tag] | |
1024 | * 4. [hostname]:[port]:[vdiname] | |
1025 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
1026 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
1027 | * | |
1028 | * You can boot from the snapshot images by specifying `snapid` or | |
1029 | * `tag'. | |
1030 | * | |
1031 | * You can run VMs outside the Sheepdog cluster by specifying | |
1032 | * `hostname' and `port' (experimental). | |
1033 | */ | |
1034 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
1035 | char *vdi, uint32_t *snapid, char *tag) | |
1036 | { | |
5d6768e3 MK |
1037 | char *p, *q, *uri; |
1038 | const char *host_spec, *vdi_spec; | |
1039 | int nr_sep, ret; | |
33b1db1c | 1040 | |
5d6768e3 | 1041 | strstart(filename, "sheepdog:", (const char **)&filename); |
7267c094 | 1042 | p = q = g_strdup(filename); |
33b1db1c MK |
1043 | |
1044 | /* count the number of separators */ | |
1045 | nr_sep = 0; | |
1046 | while (*p) { | |
1047 | if (*p == ':') { | |
1048 | nr_sep++; | |
1049 | } | |
1050 | p++; | |
1051 | } | |
1052 | p = q; | |
1053 | ||
5d6768e3 | 1054 | /* use the first two tokens as host_spec. */ |
33b1db1c | 1055 | if (nr_sep >= 2) { |
5d6768e3 | 1056 | host_spec = p; |
33b1db1c | 1057 | p = strchr(p, ':'); |
5d6768e3 | 1058 | p++; |
33b1db1c MK |
1059 | p = strchr(p, ':'); |
1060 | *p++ = '\0'; | |
1061 | } else { | |
5d6768e3 | 1062 | host_spec = ""; |
33b1db1c MK |
1063 | } |
1064 | ||
5d6768e3 | 1065 | vdi_spec = p; |
33b1db1c | 1066 | |
5d6768e3 | 1067 | p = strchr(vdi_spec, ':'); |
33b1db1c | 1068 | if (p) { |
5d6768e3 | 1069 | *p++ = '#'; |
33b1db1c MK |
1070 | } |
1071 | ||
5d6768e3 | 1072 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 1073 | |
5d6768e3 MK |
1074 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
1075 | ||
1076 | g_free(q); | |
1077 | g_free(uri); | |
1078 | ||
1079 | return ret; | |
33b1db1c MK |
1080 | } |
1081 | ||
982dcbf4 MK |
1082 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, |
1083 | uint32_t snapid, const char *tag, uint32_t *vid, | |
dc83cd42 | 1084 | bool lock, Error **errp) |
33b1db1c MK |
1085 | { |
1086 | int ret, fd; | |
1087 | SheepdogVdiReq hdr; | |
1088 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1089 | unsigned int wlen, rlen = 0; | |
1090 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
1091 | ||
dc83cd42 | 1092 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1093 | if (fd < 0) { |
cb595887 | 1094 | return fd; |
33b1db1c MK |
1095 | } |
1096 | ||
3178e275 JM |
1097 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
1098 | * which is desirable since we'll soon be sending those bytes, and | |
1099 | * don't want the send_req to read uninitialized data. | |
1100 | */ | |
33b1db1c MK |
1101 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
1102 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
1103 | ||
1104 | memset(&hdr, 0, sizeof(hdr)); | |
982dcbf4 | 1105 | if (lock) { |
33b1db1c | 1106 | hdr.opcode = SD_OP_LOCK_VDI; |
1dbfafed | 1107 | hdr.type = LOCK_TYPE_NORMAL; |
982dcbf4 MK |
1108 | } else { |
1109 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
33b1db1c MK |
1110 | } |
1111 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
1112 | hdr.proto_ver = SD_PROTO_VER; | |
1113 | hdr.data_length = wlen; | |
1114 | hdr.snapid = snapid; | |
1115 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1116 | ||
84390bed | 1117 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1118 | if (ret) { |
dc83cd42 | 1119 | error_setg_errno(errp, -ret, "cannot get vdi info"); |
33b1db1c MK |
1120 | goto out; |
1121 | } | |
1122 | ||
1123 | if (rsp->result != SD_RES_SUCCESS) { | |
dc83cd42 MA |
1124 | error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s", |
1125 | sd_strerror(rsp->result), filename, snapid, tag); | |
cb595887 MK |
1126 | if (rsp->result == SD_RES_NO_VDI) { |
1127 | ret = -ENOENT; | |
38890b24 HM |
1128 | } else if (rsp->result == SD_RES_VDI_LOCKED) { |
1129 | ret = -EBUSY; | |
cb595887 MK |
1130 | } else { |
1131 | ret = -EIO; | |
1132 | } | |
33b1db1c MK |
1133 | goto out; |
1134 | } | |
1135 | *vid = rsp->vdi_id; | |
1136 | ||
1137 | ret = 0; | |
1138 | out: | |
1139 | closesocket(fd); | |
1140 | return ret; | |
1141 | } | |
1142 | ||
a37dcdf9 | 1143 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
1144 | struct iovec *iov, int niov, |
1145 | enum AIOCBState aiocb_type) | |
33b1db1c MK |
1146 | { |
1147 | int nr_copies = s->inode.nr_copies; | |
1148 | SheepdogObjReq hdr; | |
47783072 | 1149 | unsigned int wlen = 0; |
33b1db1c MK |
1150 | int ret; |
1151 | uint64_t oid = aio_req->oid; | |
1152 | unsigned int datalen = aio_req->data_len; | |
1153 | uint64_t offset = aio_req->offset; | |
1154 | uint8_t flags = aio_req->flags; | |
1155 | uint64_t old_oid = aio_req->base_oid; | |
b544c1ab | 1156 | bool create = aio_req->create; |
33b1db1c MK |
1157 | |
1158 | if (!nr_copies) { | |
6daf194d | 1159 | error_report("bug"); |
33b1db1c MK |
1160 | } |
1161 | ||
1162 | memset(&hdr, 0, sizeof(hdr)); | |
1163 | ||
47783072 LY |
1164 | switch (aiocb_type) { |
1165 | case AIOCB_FLUSH_CACHE: | |
1166 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1167 | break; | |
1168 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1169 | hdr.opcode = SD_OP_READ_OBJ; |
1170 | hdr.flags = flags; | |
47783072 LY |
1171 | break; |
1172 | case AIOCB_WRITE_UDATA: | |
1173 | if (create) { | |
1174 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1175 | } else { | |
1176 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1177 | } | |
33b1db1c | 1178 | wlen = datalen; |
33b1db1c | 1179 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1180 | break; |
cac8f4a6 LY |
1181 | case AIOCB_DISCARD_OBJ: |
1182 | hdr.opcode = SD_OP_DISCARD_OBJ; | |
1183 | break; | |
33b1db1c MK |
1184 | } |
1185 | ||
0e7106d8 LY |
1186 | if (s->cache_flags) { |
1187 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1188 | } |
1189 | ||
33b1db1c MK |
1190 | hdr.oid = oid; |
1191 | hdr.cow_oid = old_oid; | |
1192 | hdr.copies = s->inode.nr_copies; | |
1193 | ||
1194 | hdr.data_length = datalen; | |
1195 | hdr.offset = offset; | |
1196 | ||
1197 | hdr.id = aio_req->id; | |
1198 | ||
2df46246 MK |
1199 | qemu_co_mutex_lock(&s->lock); |
1200 | s->co_send = qemu_coroutine_self(); | |
84390bed SH |
1201 | aio_set_fd_handler(s->aio_context, s->fd, |
1202 | co_read_response, co_write_request, s); | |
128aa589 | 1203 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1204 | |
1205 | /* send a header */ | |
8c5135f9 | 1206 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
80731d9d | 1207 | if (ret != sizeof(hdr)) { |
6daf194d | 1208 | error_report("failed to send a req, %s", strerror(errno)); |
011603ca | 1209 | goto out; |
33b1db1c MK |
1210 | } |
1211 | ||
1212 | if (wlen) { | |
2fc8ae1d | 1213 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
80731d9d | 1214 | if (ret != wlen) { |
6daf194d | 1215 | error_report("failed to send a data, %s", strerror(errno)); |
33b1db1c MK |
1216 | } |
1217 | } | |
011603ca | 1218 | out: |
128aa589 | 1219 | socket_set_cork(s->fd, 0); |
84390bed | 1220 | aio_set_fd_handler(s->aio_context, s->fd, co_read_response, NULL, s); |
011603ca | 1221 | s->co_send = NULL; |
2df46246 | 1222 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1223 | } |
1224 | ||
84390bed SH |
1225 | static int read_write_object(int fd, AioContext *aio_context, char *buf, |
1226 | uint64_t oid, uint8_t copies, | |
33b1db1c | 1227 | unsigned int datalen, uint64_t offset, |
0e7106d8 | 1228 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1229 | { |
1230 | SheepdogObjReq hdr; | |
1231 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1232 | unsigned int wlen, rlen; | |
1233 | int ret; | |
1234 | ||
1235 | memset(&hdr, 0, sizeof(hdr)); | |
1236 | ||
1237 | if (write) { | |
1238 | wlen = datalen; | |
1239 | rlen = 0; | |
1240 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1241 | if (create) { | |
1242 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1243 | } else { | |
1244 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1245 | } | |
1246 | } else { | |
1247 | wlen = 0; | |
1248 | rlen = datalen; | |
1249 | hdr.opcode = SD_OP_READ_OBJ; | |
1250 | } | |
47622c44 | 1251 | |
0e7106d8 | 1252 | hdr.flags |= cache_flags; |
47622c44 | 1253 | |
33b1db1c MK |
1254 | hdr.oid = oid; |
1255 | hdr.data_length = datalen; | |
1256 | hdr.offset = offset; | |
1257 | hdr.copies = copies; | |
1258 | ||
84390bed | 1259 | ret = do_req(fd, aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1260 | if (ret) { |
6daf194d | 1261 | error_report("failed to send a request to the sheep"); |
cb595887 | 1262 | return ret; |
33b1db1c MK |
1263 | } |
1264 | ||
1265 | switch (rsp->result) { | |
1266 | case SD_RES_SUCCESS: | |
1267 | return 0; | |
1268 | default: | |
6daf194d | 1269 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1270 | return -EIO; |
33b1db1c MK |
1271 | } |
1272 | } | |
1273 | ||
84390bed SH |
1274 | static int read_object(int fd, AioContext *aio_context, char *buf, |
1275 | uint64_t oid, uint8_t copies, | |
0e7106d8 LY |
1276 | unsigned int datalen, uint64_t offset, |
1277 | uint32_t cache_flags) | |
33b1db1c | 1278 | { |
84390bed SH |
1279 | return read_write_object(fd, aio_context, buf, oid, copies, |
1280 | datalen, offset, false, | |
0e7106d8 | 1281 | false, cache_flags); |
33b1db1c MK |
1282 | } |
1283 | ||
84390bed SH |
1284 | static int write_object(int fd, AioContext *aio_context, char *buf, |
1285 | uint64_t oid, uint8_t copies, | |
2f536801 | 1286 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1287 | uint32_t cache_flags) |
33b1db1c | 1288 | { |
84390bed SH |
1289 | return read_write_object(fd, aio_context, buf, oid, copies, |
1290 | datalen, offset, true, | |
0e7106d8 | 1291 | create, cache_flags); |
33b1db1c MK |
1292 | } |
1293 | ||
9ff53a0e MK |
1294 | /* update inode with the latest state */ |
1295 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag) | |
1296 | { | |
dfb12bf8 | 1297 | Error *local_err = NULL; |
9ff53a0e MK |
1298 | SheepdogInode *inode; |
1299 | int ret = 0, fd; | |
1300 | uint32_t vid = 0; | |
1301 | ||
dfb12bf8 | 1302 | fd = connect_to_sdog(s, &local_err); |
9ff53a0e | 1303 | if (fd < 0) { |
565f65d2 | 1304 | error_report_err(local_err); |
9ff53a0e MK |
1305 | return -EIO; |
1306 | } | |
1307 | ||
5d039bab | 1308 | inode = g_malloc(SD_INODE_HEADER_SIZE); |
9ff53a0e | 1309 | |
dc83cd42 | 1310 | ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err); |
9ff53a0e | 1311 | if (ret) { |
565f65d2 | 1312 | error_report_err(local_err); |
9ff53a0e MK |
1313 | goto out; |
1314 | } | |
1315 | ||
84390bed | 1316 | ret = read_object(fd, s->aio_context, (char *)inode, vid_to_vdi_oid(vid), |
5d039bab HM |
1317 | s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0, |
1318 | s->cache_flags); | |
9ff53a0e MK |
1319 | if (ret < 0) { |
1320 | goto out; | |
1321 | } | |
1322 | ||
1323 | if (inode->vdi_id != s->inode.vdi_id) { | |
5d039bab | 1324 | memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE); |
9ff53a0e MK |
1325 | } |
1326 | ||
1327 | out: | |
1328 | g_free(inode); | |
1329 | closesocket(fd); | |
1330 | ||
1331 | return ret; | |
1332 | } | |
1333 | ||
a37dcdf9 | 1334 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req) |
13c31de2 MK |
1335 | { |
1336 | SheepdogAIOCB *acb = aio_req->aiocb; | |
b544c1ab HM |
1337 | |
1338 | aio_req->create = false; | |
13c31de2 MK |
1339 | |
1340 | /* check whether this request becomes a CoW one */ | |
2412aec7 | 1341 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) { |
13c31de2 | 1342 | int idx = data_oid_to_idx(aio_req->oid); |
13c31de2 | 1343 | |
13c31de2 MK |
1344 | if (is_data_obj_writable(&s->inode, idx)) { |
1345 | goto out; | |
1346 | } | |
1347 | ||
80308d33 MK |
1348 | if (s->inode.data_vdi_id[idx]) { |
1349 | aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx); | |
1350 | aio_req->flags |= SD_FLAG_CMD_COW; | |
1351 | } | |
b544c1ab | 1352 | aio_req->create = true; |
13c31de2 MK |
1353 | } |
1354 | out: | |
2412aec7 | 1355 | if (is_data_obj(aio_req->oid)) { |
b544c1ab | 1356 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 1357 | acb->aiocb_type); |
2412aec7 MK |
1358 | } else { |
1359 | struct iovec iov; | |
1360 | iov.iov_base = &s->inode; | |
1361 | iov.iov_len = sizeof(s->inode); | |
b544c1ab | 1362 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
2412aec7 | 1363 | } |
13c31de2 MK |
1364 | } |
1365 | ||
84390bed SH |
1366 | static void sd_detach_aio_context(BlockDriverState *bs) |
1367 | { | |
1368 | BDRVSheepdogState *s = bs->opaque; | |
1369 | ||
1370 | aio_set_fd_handler(s->aio_context, s->fd, NULL, NULL, NULL); | |
1371 | } | |
1372 | ||
1373 | static void sd_attach_aio_context(BlockDriverState *bs, | |
1374 | AioContext *new_context) | |
1375 | { | |
1376 | BDRVSheepdogState *s = bs->opaque; | |
1377 | ||
1378 | s->aio_context = new_context; | |
1379 | aio_set_fd_handler(new_context, s->fd, co_read_response, NULL, s); | |
1380 | } | |
1381 | ||
c8c96350 KW |
1382 | /* TODO Convert to fine grained options */ |
1383 | static QemuOptsList runtime_opts = { | |
1384 | .name = "sheepdog", | |
1385 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1386 | .desc = { | |
1387 | { | |
1388 | .name = "filename", | |
1389 | .type = QEMU_OPT_STRING, | |
1390 | .help = "URL to the sheepdog image", | |
1391 | }, | |
1392 | { /* end of list */ } | |
1393 | }, | |
1394 | }; | |
1395 | ||
015a1036 HR |
1396 | static int sd_open(BlockDriverState *bs, QDict *options, int flags, |
1397 | Error **errp) | |
33b1db1c MK |
1398 | { |
1399 | int ret, fd; | |
1400 | uint32_t vid = 0; | |
1401 | BDRVSheepdogState *s = bs->opaque; | |
1402 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1403 | uint32_t snapid; | |
1404 | char *buf = NULL; | |
c8c96350 KW |
1405 | QemuOpts *opts; |
1406 | Error *local_err = NULL; | |
1407 | const char *filename; | |
1408 | ||
011603ca | 1409 | s->bs = bs; |
84390bed | 1410 | s->aio_context = bdrv_get_aio_context(bs); |
011603ca | 1411 | |
87ea75d5 | 1412 | opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); |
c8c96350 | 1413 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 1414 | if (local_err) { |
e67c3993 | 1415 | error_propagate(errp, local_err); |
c8c96350 KW |
1416 | ret = -EINVAL; |
1417 | goto out; | |
1418 | } | |
1419 | ||
1420 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1421 | |
c292ee6a | 1422 | QLIST_INIT(&s->inflight_aio_head); |
011603ca | 1423 | QLIST_INIT(&s->failed_aio_head); |
6a55c82c | 1424 | QLIST_INIT(&s->inflight_aiocb_head); |
33b1db1c MK |
1425 | s->fd = -1; |
1426 | ||
1427 | memset(vdi, 0, sizeof(vdi)); | |
1428 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1429 | |
1430 | if (strstr(filename, "://")) { | |
1431 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1432 | } else { | |
1433 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1434 | } | |
1435 | if (ret < 0) { | |
efde4b62 | 1436 | error_setg(errp, "Can't parse filename"); |
33b1db1c MK |
1437 | goto out; |
1438 | } | |
e67c3993 | 1439 | s->fd = get_sheep_fd(s, errp); |
33b1db1c | 1440 | if (s->fd < 0) { |
cb595887 | 1441 | ret = s->fd; |
33b1db1c MK |
1442 | goto out; |
1443 | } | |
1444 | ||
e67c3993 | 1445 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, true, errp); |
33b1db1c MK |
1446 | if (ret) { |
1447 | goto out; | |
1448 | } | |
1449 | ||
0e7106d8 LY |
1450 | /* |
1451 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1452 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1453 | */ | |
1454 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1455 | if (flags & BDRV_O_NOCACHE) { | |
1456 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1457 | } | |
cac8f4a6 | 1458 | s->discard_supported = true; |
0e7106d8 | 1459 | |
622b6057 | 1460 | if (snapid || tag[0] != '\0') { |
2440a2c3 | 1461 | DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1462 | s->is_snapshot = true; |
33b1db1c MK |
1463 | } |
1464 | ||
e67c3993 | 1465 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1466 | if (fd < 0) { |
cb595887 | 1467 | ret = fd; |
33b1db1c MK |
1468 | goto out; |
1469 | } | |
1470 | ||
7267c094 | 1471 | buf = g_malloc(SD_INODE_SIZE); |
84390bed SH |
1472 | ret = read_object(fd, s->aio_context, buf, vid_to_vdi_oid(vid), |
1473 | 0, SD_INODE_SIZE, 0, s->cache_flags); | |
33b1db1c MK |
1474 | |
1475 | closesocket(fd); | |
1476 | ||
1477 | if (ret) { | |
efde4b62 | 1478 | error_setg(errp, "Can't read snapshot inode"); |
33b1db1c MK |
1479 | goto out; |
1480 | } | |
1481 | ||
1482 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
33b1db1c | 1483 | |
e8bfaa2f | 1484 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; |
3178e275 | 1485 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1486 | qemu_co_mutex_init(&s->lock); |
498f2140 | 1487 | qemu_co_queue_init(&s->overlapping_queue); |
c8c96350 | 1488 | qemu_opts_del(opts); |
7267c094 | 1489 | g_free(buf); |
33b1db1c MK |
1490 | return 0; |
1491 | out: | |
84390bed | 1492 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, NULL, NULL, NULL); |
33b1db1c MK |
1493 | if (s->fd >= 0) { |
1494 | closesocket(s->fd); | |
1495 | } | |
c8c96350 | 1496 | qemu_opts_del(opts); |
7267c094 | 1497 | g_free(buf); |
cb595887 | 1498 | return ret; |
33b1db1c MK |
1499 | } |
1500 | ||
4da65c80 LY |
1501 | static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue, |
1502 | Error **errp) | |
1503 | { | |
1504 | BDRVSheepdogState *s = state->bs->opaque; | |
1505 | BDRVSheepdogReopenState *re_s; | |
1506 | int ret = 0; | |
1507 | ||
1508 | re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1); | |
1509 | ||
1510 | re_s->cache_flags = SD_FLAG_CMD_CACHE; | |
1511 | if (state->flags & BDRV_O_NOCACHE) { | |
1512 | re_s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1513 | } | |
1514 | ||
1515 | re_s->fd = get_sheep_fd(s, errp); | |
1516 | if (re_s->fd < 0) { | |
1517 | ret = re_s->fd; | |
1518 | return ret; | |
1519 | } | |
1520 | ||
1521 | return ret; | |
1522 | } | |
1523 | ||
1524 | static void sd_reopen_commit(BDRVReopenState *state) | |
1525 | { | |
1526 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1527 | BDRVSheepdogState *s = state->bs->opaque; | |
1528 | ||
1529 | if (s->fd) { | |
1530 | aio_set_fd_handler(s->aio_context, s->fd, NULL, NULL, NULL); | |
1531 | closesocket(s->fd); | |
1532 | } | |
1533 | ||
1534 | s->fd = re_s->fd; | |
1535 | s->cache_flags = re_s->cache_flags; | |
1536 | ||
1537 | g_free(state->opaque); | |
1538 | state->opaque = NULL; | |
1539 | ||
1540 | return; | |
1541 | } | |
1542 | ||
1543 | static void sd_reopen_abort(BDRVReopenState *state) | |
1544 | { | |
1545 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1546 | BDRVSheepdogState *s = state->bs->opaque; | |
1547 | ||
1548 | if (re_s == NULL) { | |
1549 | return; | |
1550 | } | |
1551 | ||
1552 | if (re_s->fd) { | |
1553 | aio_set_fd_handler(s->aio_context, re_s->fd, NULL, NULL, NULL); | |
1554 | closesocket(re_s->fd); | |
1555 | } | |
1556 | ||
1557 | g_free(state->opaque); | |
1558 | state->opaque = NULL; | |
1559 | ||
1560 | return; | |
1561 | } | |
1562 | ||
7d2d3e74 MA |
1563 | static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot, |
1564 | Error **errp) | |
33b1db1c MK |
1565 | { |
1566 | SheepdogVdiReq hdr; | |
1567 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1568 | int fd, ret; | |
1569 | unsigned int wlen, rlen = 0; | |
1570 | char buf[SD_MAX_VDI_LEN]; | |
1571 | ||
7d2d3e74 | 1572 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1573 | if (fd < 0) { |
cb595887 | 1574 | return fd; |
33b1db1c MK |
1575 | } |
1576 | ||
3178e275 JM |
1577 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1578 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1579 | */ | |
33b1db1c | 1580 | memset(buf, 0, sizeof(buf)); |
c31d482f | 1581 | pstrcpy(buf, sizeof(buf), s->name); |
33b1db1c MK |
1582 | |
1583 | memset(&hdr, 0, sizeof(hdr)); | |
1584 | hdr.opcode = SD_OP_NEW_VDI; | |
9f23fce7 | 1585 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1586 | |
1587 | wlen = SD_MAX_VDI_LEN; | |
1588 | ||
1589 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1590 | hdr.snapid = snapshot; | |
1591 | ||
1592 | hdr.data_length = wlen; | |
c31d482f LY |
1593 | hdr.vdi_size = s->inode.vdi_size; |
1594 | hdr.copy_policy = s->inode.copy_policy; | |
b3af018f | 1595 | hdr.copies = s->inode.nr_copies; |
876eb1b0 | 1596 | hdr.block_size_shift = s->inode.block_size_shift; |
33b1db1c | 1597 | |
84390bed | 1598 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c MK |
1599 | |
1600 | closesocket(fd); | |
1601 | ||
1602 | if (ret) { | |
7d2d3e74 | 1603 | error_setg_errno(errp, -ret, "create failed"); |
cb595887 | 1604 | return ret; |
33b1db1c MK |
1605 | } |
1606 | ||
1607 | if (rsp->result != SD_RES_SUCCESS) { | |
7d2d3e74 | 1608 | error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name); |
33b1db1c MK |
1609 | return -EIO; |
1610 | } | |
1611 | ||
1612 | if (vdi_id) { | |
1613 | *vdi_id = rsp->vdi_id; | |
1614 | } | |
1615 | ||
1616 | return 0; | |
1617 | } | |
1618 | ||
318df29e | 1619 | static int sd_prealloc(const char *filename, Error **errp) |
a8e0fdd7 MK |
1620 | { |
1621 | BlockDriverState *bs = NULL; | |
876eb1b0 TI |
1622 | BDRVSheepdogState *base = NULL; |
1623 | unsigned long buf_size; | |
a8e0fdd7 | 1624 | uint32_t idx, max_idx; |
876eb1b0 | 1625 | uint32_t object_size; |
a8e0fdd7 | 1626 | int64_t vdi_size; |
876eb1b0 | 1627 | void *buf = NULL; |
a8e0fdd7 MK |
1628 | int ret; |
1629 | ||
2e40134b | 1630 | ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, |
6ebf9aa2 | 1631 | errp); |
a8e0fdd7 | 1632 | if (ret < 0) { |
318df29e | 1633 | goto out_with_err_set; |
a8e0fdd7 MK |
1634 | } |
1635 | ||
1636 | vdi_size = bdrv_getlength(bs); | |
1637 | if (vdi_size < 0) { | |
1638 | ret = vdi_size; | |
1639 | goto out; | |
1640 | } | |
876eb1b0 TI |
1641 | |
1642 | base = bs->opaque; | |
1643 | object_size = (UINT32_C(1) << base->inode.block_size_shift); | |
1644 | buf_size = MIN(object_size, SD_DATA_OBJ_SIZE); | |
1645 | buf = g_malloc0(buf_size); | |
1646 | ||
1647 | max_idx = DIV_ROUND_UP(vdi_size, buf_size); | |
a8e0fdd7 MK |
1648 | |
1649 | for (idx = 0; idx < max_idx; idx++) { | |
1650 | /* | |
1651 | * The created image can be a cloned image, so we need to read | |
1652 | * a data from the source image. | |
1653 | */ | |
876eb1b0 | 1654 | ret = bdrv_pread(bs, idx * buf_size, buf, buf_size); |
a8e0fdd7 MK |
1655 | if (ret < 0) { |
1656 | goto out; | |
1657 | } | |
876eb1b0 | 1658 | ret = bdrv_pwrite(bs, idx * buf_size, buf, buf_size); |
a8e0fdd7 MK |
1659 | if (ret < 0) { |
1660 | goto out; | |
1661 | } | |
1662 | } | |
318df29e | 1663 | |
a8e0fdd7 | 1664 | out: |
318df29e MA |
1665 | if (ret < 0) { |
1666 | error_setg_errno(errp, -ret, "Can't pre-allocate"); | |
1667 | } | |
1668 | out_with_err_set: | |
a8e0fdd7 | 1669 | if (bs) { |
4f6fd349 | 1670 | bdrv_unref(bs); |
a8e0fdd7 | 1671 | } |
7267c094 | 1672 | g_free(buf); |
a8e0fdd7 MK |
1673 | |
1674 | return ret; | |
1675 | } | |
1676 | ||
b3af018f LY |
1677 | /* |
1678 | * Sheepdog support two kinds of redundancy, full replication and erasure | |
1679 | * coding. | |
1680 | * | |
1681 | * # create a fully replicated vdi with x copies | |
1682 | * -o redundancy=x (1 <= x <= SD_MAX_COPIES) | |
1683 | * | |
1684 | * # create a erasure coded vdi with x data strips and y parity strips | |
1685 | * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP) | |
1686 | */ | |
1687 | static int parse_redundancy(BDRVSheepdogState *s, const char *opt) | |
1688 | { | |
1689 | struct SheepdogInode *inode = &s->inode; | |
1690 | const char *n1, *n2; | |
1691 | long copy, parity; | |
1692 | char p[10]; | |
1693 | ||
1694 | pstrcpy(p, sizeof(p), opt); | |
1695 | n1 = strtok(p, ":"); | |
1696 | n2 = strtok(NULL, ":"); | |
1697 | ||
1698 | if (!n1) { | |
1699 | return -EINVAL; | |
1700 | } | |
1701 | ||
1702 | copy = strtol(n1, NULL, 10); | |
1703 | if (copy > SD_MAX_COPIES || copy < 1) { | |
1704 | return -EINVAL; | |
1705 | } | |
1706 | if (!n2) { | |
1707 | inode->copy_policy = 0; | |
1708 | inode->nr_copies = copy; | |
1709 | return 0; | |
1710 | } | |
1711 | ||
1712 | if (copy != 2 && copy != 4 && copy != 8 && copy != 16) { | |
1713 | return -EINVAL; | |
1714 | } | |
1715 | ||
1716 | parity = strtol(n2, NULL, 10); | |
1717 | if (parity >= SD_EC_MAX_STRIP || parity < 1) { | |
1718 | return -EINVAL; | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * 4 bits for parity and 4 bits for data. | |
1723 | * We have to compress upper data bits because it can't represent 16 | |
1724 | */ | |
1725 | inode->copy_policy = ((copy / 2) << 4) + parity; | |
1726 | inode->nr_copies = copy + parity; | |
1727 | ||
1728 | return 0; | |
1729 | } | |
1730 | ||
876eb1b0 TI |
1731 | static int parse_block_size_shift(BDRVSheepdogState *s, QemuOpts *opt) |
1732 | { | |
1733 | struct SheepdogInode *inode = &s->inode; | |
1734 | uint64_t object_size; | |
1735 | int obj_order; | |
1736 | ||
1737 | object_size = qemu_opt_get_size_del(opt, BLOCK_OPT_OBJECT_SIZE, 0); | |
1738 | if (object_size) { | |
1739 | if ((object_size - 1) & object_size) { /* not a power of 2? */ | |
1740 | return -EINVAL; | |
1741 | } | |
786a4ea8 | 1742 | obj_order = ctz32(object_size); |
876eb1b0 TI |
1743 | if (obj_order < 20 || obj_order > 31) { |
1744 | return -EINVAL; | |
1745 | } | |
1746 | inode->block_size_shift = (uint8_t)obj_order; | |
1747 | } | |
1748 | ||
1749 | return 0; | |
1750 | } | |
1751 | ||
b222237b | 1752 | static int sd_create(const char *filename, QemuOpts *opts, |
d5124c00 | 1753 | Error **errp) |
33b1db1c | 1754 | { |
b6fc8245 | 1755 | int ret = 0; |
c31d482f | 1756 | uint32_t vid = 0; |
33b1db1c | 1757 | char *backing_file = NULL; |
b222237b | 1758 | char *buf = NULL; |
b6fc8245 | 1759 | BDRVSheepdogState *s; |
c31d482f | 1760 | char tag[SD_MAX_VDI_TAG_LEN]; |
b4447363 | 1761 | uint32_t snapid; |
876eb1b0 | 1762 | uint64_t max_vdi_size; |
2f536801 | 1763 | bool prealloc = false; |
33b1db1c | 1764 | |
5839e53b | 1765 | s = g_new0(BDRVSheepdogState, 1); |
b6fc8245 | 1766 | |
b4447363 | 1767 | memset(tag, 0, sizeof(tag)); |
5d6768e3 | 1768 | if (strstr(filename, "://")) { |
c31d482f | 1769 | ret = sd_parse_uri(s, filename, s->name, &snapid, tag); |
5d6768e3 | 1770 | } else { |
c31d482f | 1771 | ret = parse_vdiname(s, filename, s->name, &snapid, tag); |
5d6768e3 MK |
1772 | } |
1773 | if (ret < 0) { | |
efde4b62 | 1774 | error_setg(errp, "Can't parse filename"); |
b6fc8245 | 1775 | goto out; |
b4447363 MK |
1776 | } |
1777 | ||
c2eb918e HT |
1778 | s->inode.vdi_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
1779 | BDRV_SECTOR_SIZE); | |
b222237b CL |
1780 | backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); |
1781 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
1782 | if (!buf || !strcmp(buf, "off")) { | |
1783 | prealloc = false; | |
1784 | } else if (!strcmp(buf, "full")) { | |
1785 | prealloc = true; | |
1786 | } else { | |
1787 | error_setg(errp, "Invalid preallocation mode: '%s'", buf); | |
1788 | ret = -EINVAL; | |
1789 | goto out; | |
1790 | } | |
1791 | ||
1792 | g_free(buf); | |
1793 | buf = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY); | |
1794 | if (buf) { | |
1795 | ret = parse_redundancy(s, buf); | |
1796 | if (ret < 0) { | |
1797 | error_setg(errp, "Invalid redundancy mode: '%s'", buf); | |
1798 | goto out; | |
33b1db1c | 1799 | } |
33b1db1c | 1800 | } |
876eb1b0 TI |
1801 | ret = parse_block_size_shift(s, opts); |
1802 | if (ret < 0) { | |
1803 | error_setg(errp, "Invalid object_size." | |
1804 | " obect_size needs to be power of 2" | |
1805 | " and be limited from 2^20 to 2^31"); | |
b6fc8245 | 1806 | goto out; |
33b1db1c MK |
1807 | } |
1808 | ||
1809 | if (backing_file) { | |
1810 | BlockDriverState *bs; | |
9f23fce7 | 1811 | BDRVSheepdogState *base; |
33b1db1c MK |
1812 | BlockDriver *drv; |
1813 | ||
1814 | /* Currently, only Sheepdog backing image is supported. */ | |
b65a5e12 | 1815 | drv = bdrv_find_protocol(backing_file, true, NULL); |
33b1db1c | 1816 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
e67c3993 | 1817 | error_setg(errp, "backing_file must be a sheepdog image"); |
b6fc8245 MK |
1818 | ret = -EINVAL; |
1819 | goto out; | |
33b1db1c MK |
1820 | } |
1821 | ||
2e40134b | 1822 | bs = NULL; |
6ebf9aa2 | 1823 | ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_PROTOCOL, errp); |
cb595887 | 1824 | if (ret < 0) { |
b6fc8245 | 1825 | goto out; |
cb595887 | 1826 | } |
33b1db1c | 1827 | |
9f23fce7 | 1828 | base = bs->opaque; |
33b1db1c | 1829 | |
9f23fce7 | 1830 | if (!is_snapshot(&base->inode)) { |
e67c3993 | 1831 | error_setg(errp, "cannot clone from a non snapshot vdi"); |
4f6fd349 | 1832 | bdrv_unref(bs); |
b6fc8245 MK |
1833 | ret = -EINVAL; |
1834 | goto out; | |
33b1db1c | 1835 | } |
9f23fce7 | 1836 | s->inode.vdi_id = base->inode.vdi_id; |
4f6fd349 | 1837 | bdrv_unref(bs); |
33b1db1c MK |
1838 | } |
1839 | ||
5d5da114 | 1840 | s->aio_context = qemu_get_aio_context(); |
876eb1b0 TI |
1841 | |
1842 | /* if block_size_shift is not specified, get cluster default value */ | |
1843 | if (s->inode.block_size_shift == 0) { | |
1844 | SheepdogVdiReq hdr; | |
1845 | SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr; | |
1846 | Error *local_err = NULL; | |
1847 | int fd; | |
1848 | unsigned int wlen = 0, rlen = 0; | |
1849 | ||
1850 | fd = connect_to_sdog(s, &local_err); | |
1851 | if (fd < 0) { | |
1852 | error_report("%s", error_get_pretty(local_err)); | |
1853 | error_free(local_err); | |
1854 | ret = -EIO; | |
1855 | goto out; | |
1856 | } | |
1857 | ||
1858 | memset(&hdr, 0, sizeof(hdr)); | |
1859 | hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT; | |
1860 | hdr.proto_ver = SD_PROTO_VER; | |
1861 | ||
1862 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, | |
1863 | NULL, &wlen, &rlen); | |
1864 | closesocket(fd); | |
1865 | if (ret) { | |
1866 | error_setg_errno(errp, -ret, "failed to get cluster default"); | |
1867 | goto out; | |
1868 | } | |
1869 | if (rsp->result == SD_RES_SUCCESS) { | |
1870 | s->inode.block_size_shift = rsp->block_size_shift; | |
1871 | } else { | |
1872 | s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT; | |
1873 | } | |
1874 | } | |
1875 | ||
1876 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; | |
1877 | ||
1878 | if (s->inode.vdi_size > max_vdi_size) { | |
1879 | error_setg(errp, "An image is too large." | |
1880 | " The maximum image size is %"PRIu64 "GB", | |
1881 | max_vdi_size / 1024 / 1024 / 1024); | |
1882 | ret = -EINVAL; | |
1883 | goto out; | |
1884 | } | |
1885 | ||
e67c3993 | 1886 | ret = do_sd_create(s, &vid, 0, errp); |
7d2d3e74 | 1887 | if (ret) { |
b6fc8245 | 1888 | goto out; |
a8e0fdd7 MK |
1889 | } |
1890 | ||
7d2d3e74 | 1891 | if (prealloc) { |
e67c3993 | 1892 | ret = sd_prealloc(filename, errp); |
318df29e | 1893 | } |
b6fc8245 | 1894 | out: |
b222237b CL |
1895 | g_free(backing_file); |
1896 | g_free(buf); | |
b6fc8245 MK |
1897 | g_free(s); |
1898 | return ret; | |
33b1db1c MK |
1899 | } |
1900 | ||
1901 | static void sd_close(BlockDriverState *bs) | |
1902 | { | |
dfb12bf8 | 1903 | Error *local_err = NULL; |
33b1db1c MK |
1904 | BDRVSheepdogState *s = bs->opaque; |
1905 | SheepdogVdiReq hdr; | |
1906 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1907 | unsigned int wlen, rlen = 0; | |
1908 | int fd, ret; | |
1909 | ||
2440a2c3 | 1910 | DPRINTF("%s\n", s->name); |
33b1db1c | 1911 | |
dfb12bf8 | 1912 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1913 | if (fd < 0) { |
565f65d2 | 1914 | error_report_err(local_err); |
33b1db1c MK |
1915 | return; |
1916 | } | |
1917 | ||
1918 | memset(&hdr, 0, sizeof(hdr)); | |
1919 | ||
1920 | hdr.opcode = SD_OP_RELEASE_VDI; | |
1dbfafed | 1921 | hdr.type = LOCK_TYPE_NORMAL; |
9f23fce7 | 1922 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1923 | wlen = strlen(s->name) + 1; |
1924 | hdr.data_length = wlen; | |
1925 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1926 | ||
84390bed SH |
1927 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, |
1928 | s->name, &wlen, &rlen); | |
33b1db1c MK |
1929 | |
1930 | closesocket(fd); | |
1931 | ||
1932 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1933 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1934 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1935 | } |
1936 | ||
84390bed | 1937 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, NULL, NULL, NULL); |
33b1db1c | 1938 | closesocket(s->fd); |
25af257d | 1939 | g_free(s->host_spec); |
33b1db1c MK |
1940 | } |
1941 | ||
1942 | static int64_t sd_getlength(BlockDriverState *bs) | |
1943 | { | |
1944 | BDRVSheepdogState *s = bs->opaque; | |
1945 | ||
1946 | return s->inode.vdi_size; | |
1947 | } | |
1948 | ||
1949 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1950 | { | |
dfb12bf8 | 1951 | Error *local_err = NULL; |
33b1db1c MK |
1952 | BDRVSheepdogState *s = bs->opaque; |
1953 | int ret, fd; | |
1954 | unsigned int datalen; | |
876eb1b0 | 1955 | uint64_t max_vdi_size; |
33b1db1c | 1956 | |
876eb1b0 | 1957 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; |
33b1db1c | 1958 | if (offset < s->inode.vdi_size) { |
6daf194d | 1959 | error_report("shrinking is not supported"); |
33b1db1c | 1960 | return -EINVAL; |
876eb1b0 | 1961 | } else if (offset > max_vdi_size) { |
6daf194d | 1962 | error_report("too big image size"); |
33b1db1c MK |
1963 | return -EINVAL; |
1964 | } | |
1965 | ||
dfb12bf8 | 1966 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1967 | if (fd < 0) { |
565f65d2 | 1968 | error_report_err(local_err); |
cb595887 | 1969 | return fd; |
33b1db1c MK |
1970 | } |
1971 | ||
1972 | /* we don't need to update entire object */ | |
1973 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1974 | s->inode.vdi_size = offset; | |
84390bed SH |
1975 | ret = write_object(fd, s->aio_context, (char *)&s->inode, |
1976 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, | |
1977 | datalen, 0, false, s->cache_flags); | |
33b1db1c MK |
1978 | close(fd); |
1979 | ||
1980 | if (ret < 0) { | |
6daf194d | 1981 | error_report("failed to update an inode."); |
33b1db1c MK |
1982 | } |
1983 | ||
cb595887 | 1984 | return ret; |
33b1db1c MK |
1985 | } |
1986 | ||
1987 | /* | |
1988 | * This function is called after writing data objects. If we need to | |
1989 | * update metadata, this sends a write request to the vdi object. | |
2df46246 | 1990 | * Otherwise, this switches back to sd_co_readv/writev. |
33b1db1c | 1991 | */ |
d8716b41 | 1992 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c | 1993 | { |
33b1db1c MK |
1994 | BDRVSheepdogState *s = acb->common.bs->opaque; |
1995 | struct iovec iov; | |
1996 | AIOReq *aio_req; | |
1997 | uint32_t offset, data_len, mn, mx; | |
1998 | ||
498f2140 HM |
1999 | mn = acb->min_dirty_data_idx; |
2000 | mx = acb->max_dirty_data_idx; | |
33b1db1c MK |
2001 | if (mn <= mx) { |
2002 | /* we need to update the vdi object. */ | |
2003 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
2004 | mn * sizeof(s->inode.data_vdi_id[0]); | |
2005 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
2006 | ||
498f2140 HM |
2007 | acb->min_dirty_data_idx = UINT32_MAX; |
2008 | acb->max_dirty_data_idx = 0; | |
33b1db1c MK |
2009 | |
2010 | iov.iov_base = &s->inode; | |
2011 | iov.iov_len = sizeof(s->inode); | |
2012 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
b544c1ab | 2013 | data_len, offset, 0, false, 0, offset); |
c292ee6a | 2014 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
b544c1ab | 2015 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
33b1db1c MK |
2016 | |
2017 | acb->aio_done_func = sd_finish_aiocb; | |
2018 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
2019 | return; | |
2020 | } | |
a37dcdf9 | 2021 | |
33b1db1c MK |
2022 | sd_finish_aiocb(acb); |
2023 | } | |
2024 | ||
859e5553 LY |
2025 | /* Delete current working VDI on the snapshot chain */ |
2026 | static bool sd_delete(BDRVSheepdogState *s) | |
2027 | { | |
dfb12bf8 | 2028 | Error *local_err = NULL; |
859e5553 LY |
2029 | unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0; |
2030 | SheepdogVdiReq hdr = { | |
2031 | .opcode = SD_OP_DEL_VDI, | |
9f23fce7 | 2032 | .base_vdi_id = s->inode.vdi_id, |
859e5553 LY |
2033 | .data_length = wlen, |
2034 | .flags = SD_FLAG_CMD_WRITE, | |
2035 | }; | |
2036 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2037 | int fd, ret; | |
2038 | ||
dfb12bf8 | 2039 | fd = connect_to_sdog(s, &local_err); |
859e5553 | 2040 | if (fd < 0) { |
565f65d2 | 2041 | error_report_err(local_err); |
859e5553 LY |
2042 | return false; |
2043 | } | |
2044 | ||
84390bed SH |
2045 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, |
2046 | s->name, &wlen, &rlen); | |
859e5553 LY |
2047 | closesocket(fd); |
2048 | if (ret) { | |
2049 | return false; | |
2050 | } | |
2051 | switch (rsp->result) { | |
2052 | case SD_RES_NO_VDI: | |
2053 | error_report("%s was already deleted", s->name); | |
2054 | /* fall through */ | |
2055 | case SD_RES_SUCCESS: | |
2056 | break; | |
2057 | default: | |
2058 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2059 | return false; | |
2060 | } | |
2061 | ||
2062 | return true; | |
2063 | } | |
2064 | ||
33b1db1c MK |
2065 | /* |
2066 | * Create a writable VDI from a snapshot | |
2067 | */ | |
2068 | static int sd_create_branch(BDRVSheepdogState *s) | |
2069 | { | |
dfb12bf8 | 2070 | Error *local_err = NULL; |
33b1db1c MK |
2071 | int ret, fd; |
2072 | uint32_t vid; | |
2073 | char *buf; | |
859e5553 | 2074 | bool deleted; |
33b1db1c | 2075 | |
2440a2c3 | 2076 | DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
33b1db1c | 2077 | |
7267c094 | 2078 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 2079 | |
859e5553 LY |
2080 | /* |
2081 | * Even If deletion fails, we will just create extra snapshot based on | |
dc6fb73d | 2082 | * the working VDI which was supposed to be deleted. So no need to |
859e5553 LY |
2083 | * false bail out. |
2084 | */ | |
2085 | deleted = sd_delete(s); | |
7d2d3e74 | 2086 | ret = do_sd_create(s, &vid, !deleted, &local_err); |
33b1db1c | 2087 | if (ret) { |
565f65d2 | 2088 | error_report_err(local_err); |
33b1db1c MK |
2089 | goto out; |
2090 | } | |
2091 | ||
2440a2c3 | 2092 | DPRINTF("%" PRIx32 " is created.\n", vid); |
33b1db1c | 2093 | |
dfb12bf8 | 2094 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2095 | if (fd < 0) { |
565f65d2 | 2096 | error_report_err(local_err); |
cb595887 | 2097 | ret = fd; |
33b1db1c MK |
2098 | goto out; |
2099 | } | |
2100 | ||
84390bed SH |
2101 | ret = read_object(fd, s->aio_context, buf, vid_to_vdi_oid(vid), |
2102 | s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags); | |
33b1db1c MK |
2103 | |
2104 | closesocket(fd); | |
2105 | ||
2106 | if (ret < 0) { | |
2107 | goto out; | |
2108 | } | |
2109 | ||
2110 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
2111 | ||
2f536801 | 2112 | s->is_snapshot = false; |
33b1db1c | 2113 | ret = 0; |
2440a2c3 | 2114 | DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
33b1db1c MK |
2115 | |
2116 | out: | |
7267c094 | 2117 | g_free(buf); |
33b1db1c MK |
2118 | |
2119 | return ret; | |
2120 | } | |
2121 | ||
2122 | /* | |
2123 | * Send I/O requests to the server. | |
2124 | * | |
2125 | * This function sends requests to the server, links the requests to | |
c292ee6a | 2126 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
2127 | * waiting the response. The responses are received in the |
2128 | * `aio_read_response' function which is called from the main loop as | |
2129 | * a fd handler. | |
2df46246 MK |
2130 | * |
2131 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
2132 | * request and -errno in error cases. | |
33b1db1c | 2133 | */ |
d8716b41 | 2134 | static int coroutine_fn sd_co_rw_vector(void *p) |
33b1db1c MK |
2135 | { |
2136 | SheepdogAIOCB *acb = p; | |
2137 | int ret = 0; | |
e8bfaa2f | 2138 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; |
876eb1b0 TI |
2139 | unsigned long idx; |
2140 | uint32_t object_size; | |
33b1db1c | 2141 | uint64_t oid; |
876eb1b0 | 2142 | uint64_t offset; |
33b1db1c MK |
2143 | BDRVSheepdogState *s = acb->common.bs->opaque; |
2144 | SheepdogInode *inode = &s->inode; | |
2145 | AIOReq *aio_req; | |
2146 | ||
33b1db1c MK |
2147 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
2148 | /* | |
2149 | * In the case we open the snapshot VDI, Sheepdog creates the | |
2150 | * writable VDI when we do a write operation first. | |
2151 | */ | |
2152 | ret = sd_create_branch(s); | |
2153 | if (ret) { | |
2154 | acb->ret = -EIO; | |
2155 | goto out; | |
2156 | } | |
2157 | } | |
2158 | ||
876eb1b0 TI |
2159 | object_size = (UINT32_C(1) << inode->block_size_shift); |
2160 | idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
2161 | offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size; | |
2162 | ||
1d732d7d MK |
2163 | /* |
2164 | * Make sure we don't free the aiocb before we are done with all requests. | |
2165 | * This additional reference is dropped at the end of this function. | |
2166 | */ | |
2167 | acb->nr_pending++; | |
2168 | ||
33b1db1c MK |
2169 | while (done != total) { |
2170 | uint8_t flags = 0; | |
2171 | uint64_t old_oid = 0; | |
2f536801 | 2172 | bool create = false; |
33b1db1c MK |
2173 | |
2174 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
2175 | ||
876eb1b0 | 2176 | len = MIN(total - done, object_size - offset); |
33b1db1c | 2177 | |
19db9b90 CH |
2178 | switch (acb->aiocb_type) { |
2179 | case AIOCB_READ_UDATA: | |
2180 | if (!inode->data_vdi_id[idx]) { | |
2181 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
2182 | goto done; |
2183 | } | |
19db9b90 CH |
2184 | break; |
2185 | case AIOCB_WRITE_UDATA: | |
2186 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 2187 | create = true; |
19db9b90 CH |
2188 | } else if (!is_data_obj_writable(inode, idx)) { |
2189 | /* Copy-On-Write */ | |
2f536801 | 2190 | create = true; |
19db9b90 CH |
2191 | old_oid = oid; |
2192 | flags = SD_FLAG_CMD_COW; | |
2193 | } | |
2194 | break; | |
cac8f4a6 LY |
2195 | case AIOCB_DISCARD_OBJ: |
2196 | /* | |
2197 | * We discard the object only when the whole object is | |
2198 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
2199 | */ | |
876eb1b0 | 2200 | if (len != object_size || inode->data_vdi_id[idx] == 0) { |
cac8f4a6 LY |
2201 | goto done; |
2202 | } | |
2203 | break; | |
19db9b90 CH |
2204 | default: |
2205 | break; | |
33b1db1c MK |
2206 | } |
2207 | ||
2208 | if (create) { | |
2440a2c3 | 2209 | DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1b6ac998 | 2210 | inode->vdi_id, oid, |
33b1db1c MK |
2211 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
2212 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
2440a2c3 | 2213 | DPRINTF("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
2214 | } |
2215 | ||
b544c1ab HM |
2216 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create, |
2217 | old_oid, done); | |
80308d33 | 2218 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
33b1db1c | 2219 | |
b544c1ab | 2220 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 2221 | acb->aiocb_type); |
33b1db1c MK |
2222 | done: |
2223 | offset = 0; | |
2224 | idx++; | |
2225 | done += len; | |
2226 | } | |
2227 | out: | |
1d732d7d | 2228 | if (!--acb->nr_pending) { |
2df46246 | 2229 | return acb->ret; |
33b1db1c | 2230 | } |
2df46246 | 2231 | return 1; |
33b1db1c MK |
2232 | } |
2233 | ||
498f2140 | 2234 | static bool check_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *aiocb) |
6a55c82c HM |
2235 | { |
2236 | SheepdogAIOCB *cb; | |
2237 | ||
2238 | QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) { | |
498f2140 | 2239 | if (AIOCBOverlapping(aiocb, cb)) { |
6a55c82c HM |
2240 | return true; |
2241 | } | |
2242 | } | |
2243 | ||
2244 | QLIST_INSERT_HEAD(&s->inflight_aiocb_head, aiocb, aiocb_siblings); | |
2245 | return false; | |
2246 | } | |
2247 | ||
a968168c | 2248 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2249 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
2250 | { |
2251 | SheepdogAIOCB *acb; | |
2df46246 | 2252 | int ret; |
e50d7607 LY |
2253 | int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE; |
2254 | BDRVSheepdogState *s = bs->opaque; | |
33b1db1c | 2255 | |
c0191e76 | 2256 | if (offset > s->inode.vdi_size) { |
e50d7607 | 2257 | ret = sd_truncate(bs, offset); |
cb595887 MK |
2258 | if (ret < 0) { |
2259 | return ret; | |
33b1db1c | 2260 | } |
33b1db1c MK |
2261 | } |
2262 | ||
f700f8e3 | 2263 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
2264 | acb->aio_done_func = sd_write_done; |
2265 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
2266 | ||
6a55c82c | 2267 | retry: |
498f2140 HM |
2268 | if (check_overlapping_aiocb(s, acb)) { |
2269 | qemu_co_queue_wait(&s->overlapping_queue); | |
6a55c82c HM |
2270 | goto retry; |
2271 | } | |
2272 | ||
2df46246 MK |
2273 | ret = sd_co_rw_vector(acb); |
2274 | if (ret <= 0) { | |
6a55c82c | 2275 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2276 | qemu_co_queue_restart_all(&s->overlapping_queue); |
8007429a | 2277 | qemu_aio_unref(acb); |
2df46246 MK |
2278 | return ret; |
2279 | } | |
2280 | ||
2281 | qemu_coroutine_yield(); | |
2282 | ||
6a55c82c | 2283 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2284 | qemu_co_queue_restart_all(&s->overlapping_queue); |
6a55c82c | 2285 | |
2df46246 | 2286 | return acb->ret; |
33b1db1c MK |
2287 | } |
2288 | ||
a968168c | 2289 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2290 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
2291 | { |
2292 | SheepdogAIOCB *acb; | |
19db9b90 | 2293 | int ret; |
6a55c82c | 2294 | BDRVSheepdogState *s = bs->opaque; |
33b1db1c | 2295 | |
f700f8e3 | 2296 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
2297 | acb->aiocb_type = AIOCB_READ_UDATA; |
2298 | acb->aio_done_func = sd_finish_aiocb; | |
2299 | ||
6a55c82c | 2300 | retry: |
498f2140 HM |
2301 | if (check_overlapping_aiocb(s, acb)) { |
2302 | qemu_co_queue_wait(&s->overlapping_queue); | |
6a55c82c HM |
2303 | goto retry; |
2304 | } | |
2305 | ||
2df46246 MK |
2306 | ret = sd_co_rw_vector(acb); |
2307 | if (ret <= 0) { | |
6a55c82c | 2308 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2309 | qemu_co_queue_restart_all(&s->overlapping_queue); |
8007429a | 2310 | qemu_aio_unref(acb); |
2df46246 MK |
2311 | return ret; |
2312 | } | |
2313 | ||
2314 | qemu_coroutine_yield(); | |
2315 | ||
6a55c82c | 2316 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2317 | qemu_co_queue_restart_all(&s->overlapping_queue); |
2df46246 | 2318 | return acb->ret; |
33b1db1c MK |
2319 | } |
2320 | ||
47622c44 LY |
2321 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
2322 | { | |
2323 | BDRVSheepdogState *s = bs->opaque; | |
47783072 LY |
2324 | SheepdogAIOCB *acb; |
2325 | AIOReq *aio_req; | |
47622c44 | 2326 | |
0e7106d8 | 2327 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
2328 | return 0; |
2329 | } | |
2330 | ||
f700f8e3 | 2331 | acb = sd_aio_setup(bs, NULL, 0, 0); |
47783072 LY |
2332 | acb->aiocb_type = AIOCB_FLUSH_CACHE; |
2333 | acb->aio_done_func = sd_finish_aiocb; | |
47622c44 | 2334 | |
47783072 | 2335 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
b544c1ab | 2336 | 0, 0, 0, false, 0, 0); |
47783072 | 2337 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
b544c1ab | 2338 | add_aio_request(s, aio_req, NULL, 0, acb->aiocb_type); |
47622c44 | 2339 | |
47783072 LY |
2340 | qemu_coroutine_yield(); |
2341 | return acb->ret; | |
47622c44 LY |
2342 | } |
2343 | ||
33b1db1c MK |
2344 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
2345 | { | |
dfb12bf8 | 2346 | Error *local_err = NULL; |
33b1db1c MK |
2347 | BDRVSheepdogState *s = bs->opaque; |
2348 | int ret, fd; | |
2349 | uint32_t new_vid; | |
2350 | SheepdogInode *inode; | |
2351 | unsigned int datalen; | |
2352 | ||
2440a2c3 | 2353 | DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
2354 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
2355 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
2356 | ||
2357 | if (s->is_snapshot) { | |
2358 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 2359 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
2360 | |
2361 | return -EINVAL; | |
2362 | } | |
2363 | ||
2440a2c3 | 2364 | DPRINTF("%s %s\n", sn_info->name, sn_info->id_str); |
33b1db1c MK |
2365 | |
2366 | s->inode.vm_state_size = sn_info->vm_state_size; | |
2367 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
2368 | /* It appears that inode.tag does not require a NUL terminator, |
2369 | * which means this use of strncpy is ok. | |
2370 | */ | |
33b1db1c MK |
2371 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
2372 | /* we don't need to update entire object */ | |
2373 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
2df5fee2 | 2374 | inode = g_malloc(datalen); |
33b1db1c MK |
2375 | |
2376 | /* refresh inode. */ | |
dfb12bf8 | 2377 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2378 | if (fd < 0) { |
565f65d2 | 2379 | error_report_err(local_err); |
cb595887 | 2380 | ret = fd; |
33b1db1c MK |
2381 | goto cleanup; |
2382 | } | |
2383 | ||
84390bed SH |
2384 | ret = write_object(fd, s->aio_context, (char *)&s->inode, |
2385 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, | |
2386 | datalen, 0, false, s->cache_flags); | |
33b1db1c | 2387 | if (ret < 0) { |
6daf194d | 2388 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
2389 | goto cleanup; |
2390 | } | |
2391 | ||
7d2d3e74 | 2392 | ret = do_sd_create(s, &new_vid, 1, &local_err); |
33b1db1c | 2393 | if (ret < 0) { |
27994d58 MA |
2394 | error_report("failed to create inode for snapshot: %s", |
2395 | error_get_pretty(local_err)); | |
973a8529 | 2396 | error_free(local_err); |
33b1db1c MK |
2397 | goto cleanup; |
2398 | } | |
2399 | ||
84390bed SH |
2400 | ret = read_object(fd, s->aio_context, (char *)inode, |
2401 | vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0, | |
2402 | s->cache_flags); | |
33b1db1c MK |
2403 | |
2404 | if (ret < 0) { | |
6daf194d | 2405 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
2406 | goto cleanup; |
2407 | } | |
2408 | ||
2409 | memcpy(&s->inode, inode, datalen); | |
2440a2c3 | 2410 | DPRINTF("s->inode: name %s snap_id %x oid %x\n", |
33b1db1c MK |
2411 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
2412 | ||
2413 | cleanup: | |
2df5fee2 | 2414 | g_free(inode); |
33b1db1c MK |
2415 | closesocket(fd); |
2416 | return ret; | |
2417 | } | |
2418 | ||
859e5553 LY |
2419 | /* |
2420 | * We implement rollback(loadvm) operation to the specified snapshot by | |
2421 | * 1) switch to the snapshot | |
2422 | * 2) rely on sd_create_branch to delete working VDI and | |
dc6fb73d | 2423 | * 3) create a new working VDI based on the specified snapshot |
859e5553 | 2424 | */ |
33b1db1c MK |
2425 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
2426 | { | |
2427 | BDRVSheepdogState *s = bs->opaque; | |
2428 | BDRVSheepdogState *old_s; | |
9ff53a0e | 2429 | char tag[SD_MAX_VDI_TAG_LEN]; |
33b1db1c | 2430 | uint32_t snapid = 0; |
9ff53a0e | 2431 | int ret = 0; |
33b1db1c | 2432 | |
5839e53b | 2433 | old_s = g_new(BDRVSheepdogState, 1); |
33b1db1c MK |
2434 | |
2435 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
2436 | ||
33b1db1c | 2437 | snapid = strtoul(snapshot_id, NULL, 10); |
3178e275 JM |
2438 | if (snapid) { |
2439 | tag[0] = 0; | |
2440 | } else { | |
b579ffb3 | 2441 | pstrcpy(tag, sizeof(tag), snapshot_id); |
33b1db1c MK |
2442 | } |
2443 | ||
9ff53a0e | 2444 | ret = reload_inode(s, snapid, tag); |
33b1db1c | 2445 | if (ret) { |
33b1db1c MK |
2446 | goto out; |
2447 | } | |
2448 | ||
cede621f LY |
2449 | ret = sd_create_branch(s); |
2450 | if (ret) { | |
33b1db1c MK |
2451 | goto out; |
2452 | } | |
2453 | ||
7267c094 | 2454 | g_free(old_s); |
33b1db1c MK |
2455 | |
2456 | return 0; | |
2457 | out: | |
2458 | /* recover bdrv_sd_state */ | |
2459 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 | 2460 | g_free(old_s); |
33b1db1c | 2461 | |
6daf194d | 2462 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
2463 | |
2464 | return ret; | |
2465 | } | |
2466 | ||
a89d89d3 WX |
2467 | static int sd_snapshot_delete(BlockDriverState *bs, |
2468 | const char *snapshot_id, | |
2469 | const char *name, | |
2470 | Error **errp) | |
33b1db1c MK |
2471 | { |
2472 | /* FIXME: Delete specified snapshot id. */ | |
2473 | return 0; | |
2474 | } | |
2475 | ||
33b1db1c MK |
2476 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
2477 | { | |
dfb12bf8 | 2478 | Error *local_err = NULL; |
33b1db1c MK |
2479 | BDRVSheepdogState *s = bs->opaque; |
2480 | SheepdogReq req; | |
2481 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
2482 | QEMUSnapshotInfo *sn_tab = NULL; | |
2483 | unsigned wlen, rlen; | |
2484 | int found = 0; | |
2485 | static SheepdogInode inode; | |
2486 | unsigned long *vdi_inuse; | |
2487 | unsigned int start_nr; | |
2488 | uint64_t hval; | |
2489 | uint32_t vid; | |
2490 | ||
7267c094 | 2491 | vdi_inuse = g_malloc(max); |
33b1db1c | 2492 | |
dfb12bf8 | 2493 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2494 | if (fd < 0) { |
565f65d2 | 2495 | error_report_err(local_err); |
cb595887 | 2496 | ret = fd; |
33b1db1c MK |
2497 | goto out; |
2498 | } | |
2499 | ||
2500 | rlen = max; | |
2501 | wlen = 0; | |
2502 | ||
2503 | memset(&req, 0, sizeof(req)); | |
2504 | ||
2505 | req.opcode = SD_OP_READ_VDIS; | |
2506 | req.data_length = max; | |
2507 | ||
84390bed SH |
2508 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&req, |
2509 | vdi_inuse, &wlen, &rlen); | |
33b1db1c MK |
2510 | |
2511 | closesocket(fd); | |
2512 | if (ret) { | |
2513 | goto out; | |
2514 | } | |
2515 | ||
02c4f26b | 2516 | sn_tab = g_new0(QEMUSnapshotInfo, nr); |
33b1db1c MK |
2517 | |
2518 | /* calculate a vdi id with hash function */ | |
2519 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2520 | start_nr = hval & (SD_NR_VDIS - 1); | |
2521 | ||
dfb12bf8 | 2522 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2523 | if (fd < 0) { |
565f65d2 | 2524 | error_report_err(local_err); |
cb595887 | 2525 | ret = fd; |
33b1db1c MK |
2526 | goto out; |
2527 | } | |
2528 | ||
2529 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2530 | if (!test_bit(vid, vdi_inuse)) { | |
2531 | break; | |
2532 | } | |
2533 | ||
2534 | /* we don't need to read entire object */ | |
84390bed SH |
2535 | ret = read_object(fd, s->aio_context, (char *)&inode, |
2536 | vid_to_vdi_oid(vid), | |
47622c44 | 2537 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2538 | s->cache_flags); |
33b1db1c MK |
2539 | |
2540 | if (ret) { | |
2541 | continue; | |
2542 | } | |
2543 | ||
2544 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2545 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2546 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2547 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2548 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2549 | ||
521b2b5d HR |
2550 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), |
2551 | "%" PRIu32, inode.snap_id); | |
3178e275 JM |
2552 | pstrcpy(sn_tab[found].name, |
2553 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2554 | inode.tag); | |
33b1db1c MK |
2555 | found++; |
2556 | } | |
2557 | } | |
2558 | ||
2559 | closesocket(fd); | |
2560 | out: | |
2561 | *psn_tab = sn_tab; | |
2562 | ||
7267c094 | 2563 | g_free(vdi_inuse); |
33b1db1c | 2564 | |
cb595887 MK |
2565 | if (ret < 0) { |
2566 | return ret; | |
2567 | } | |
2568 | ||
33b1db1c MK |
2569 | return found; |
2570 | } | |
2571 | ||
2572 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2573 | int64_t pos, int size, int load) | |
2574 | { | |
dfb12bf8 | 2575 | Error *local_err = NULL; |
2f536801 MK |
2576 | bool create; |
2577 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2578 | unsigned int data_len; |
2579 | uint64_t vmstate_oid; | |
33b1db1c | 2580 | uint64_t offset; |
cede621f LY |
2581 | uint32_t vdi_index; |
2582 | uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id; | |
876eb1b0 | 2583 | uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift); |
33b1db1c | 2584 | |
dfb12bf8 | 2585 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2586 | if (fd < 0) { |
565f65d2 | 2587 | error_report_err(local_err); |
cb595887 | 2588 | return fd; |
33b1db1c MK |
2589 | } |
2590 | ||
6f3c714e | 2591 | while (remaining) { |
876eb1b0 TI |
2592 | vdi_index = pos / object_size; |
2593 | offset = pos % object_size; | |
33b1db1c | 2594 | |
876eb1b0 | 2595 | data_len = MIN(remaining, object_size - offset); |
33b1db1c | 2596 | |
cede621f | 2597 | vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index); |
33b1db1c MK |
2598 | |
2599 | create = (offset == 0); | |
2600 | if (load) { | |
84390bed | 2601 | ret = read_object(fd, s->aio_context, (char *)data, vmstate_oid, |
47622c44 | 2602 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2603 | s->cache_flags); |
33b1db1c | 2604 | } else { |
84390bed | 2605 | ret = write_object(fd, s->aio_context, (char *)data, vmstate_oid, |
47622c44 | 2606 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2607 | s->cache_flags); |
33b1db1c MK |
2608 | } |
2609 | ||
2610 | if (ret < 0) { | |
6daf194d | 2611 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2612 | goto cleanup; |
2613 | } | |
2614 | ||
2615 | pos += data_len; | |
1f7a48de | 2616 | data += data_len; |
6f3c714e | 2617 | remaining -= data_len; |
33b1db1c | 2618 | } |
6f3c714e | 2619 | ret = size; |
33b1db1c MK |
2620 | cleanup: |
2621 | closesocket(fd); | |
2622 | return ret; | |
2623 | } | |
2624 | ||
cf8074b3 KW |
2625 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2626 | int64_t pos) | |
33b1db1c MK |
2627 | { |
2628 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2629 | void *buf; |
2630 | int ret; | |
33b1db1c | 2631 | |
cf8074b3 KW |
2632 | buf = qemu_blockalign(bs, qiov->size); |
2633 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2634 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2635 | qemu_vfree(buf); | |
2636 | ||
2637 | return ret; | |
33b1db1c MK |
2638 | } |
2639 | ||
2640 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
2641 | int64_t pos, int size) | |
2642 | { | |
2643 | BDRVSheepdogState *s = bs->opaque; | |
2644 | ||
2645 | return do_load_save_vmstate(s, data, pos, size, 1); | |
2646 | } | |
2647 | ||
2648 | ||
cac8f4a6 LY |
2649 | static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num, |
2650 | int nb_sectors) | |
2651 | { | |
2652 | SheepdogAIOCB *acb; | |
2653 | QEMUIOVector dummy; | |
2654 | BDRVSheepdogState *s = bs->opaque; | |
2655 | int ret; | |
2656 | ||
2657 | if (!s->discard_supported) { | |
2658 | return 0; | |
2659 | } | |
2660 | ||
2661 | acb = sd_aio_setup(bs, &dummy, sector_num, nb_sectors); | |
2662 | acb->aiocb_type = AIOCB_DISCARD_OBJ; | |
2663 | acb->aio_done_func = sd_finish_aiocb; | |
2664 | ||
6a55c82c | 2665 | retry: |
498f2140 HM |
2666 | if (check_overlapping_aiocb(s, acb)) { |
2667 | qemu_co_queue_wait(&s->overlapping_queue); | |
6a55c82c HM |
2668 | goto retry; |
2669 | } | |
2670 | ||
cac8f4a6 LY |
2671 | ret = sd_co_rw_vector(acb); |
2672 | if (ret <= 0) { | |
6a55c82c | 2673 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2674 | qemu_co_queue_restart_all(&s->overlapping_queue); |
8007429a | 2675 | qemu_aio_unref(acb); |
cac8f4a6 LY |
2676 | return ret; |
2677 | } | |
2678 | ||
2679 | qemu_coroutine_yield(); | |
2680 | ||
6a55c82c | 2681 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2682 | qemu_co_queue_restart_all(&s->overlapping_queue); |
6a55c82c | 2683 | |
cac8f4a6 LY |
2684 | return acb->ret; |
2685 | } | |
2686 | ||
b6b8a333 PB |
2687 | static coroutine_fn int64_t |
2688 | sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
2689 | int *pnum) | |
8d71c631 LY |
2690 | { |
2691 | BDRVSheepdogState *s = bs->opaque; | |
2692 | SheepdogInode *inode = &s->inode; | |
876eb1b0 | 2693 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
9cd76737 | 2694 | uint64_t offset = sector_num * BDRV_SECTOR_SIZE; |
876eb1b0 | 2695 | unsigned long start = offset / object_size, |
8d71c631 | 2696 | end = DIV_ROUND_UP((sector_num + nb_sectors) * |
876eb1b0 | 2697 | BDRV_SECTOR_SIZE, object_size); |
8d71c631 | 2698 | unsigned long idx; |
9cd76737 | 2699 | int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset; |
8d71c631 LY |
2700 | |
2701 | for (idx = start; idx < end; idx++) { | |
2702 | if (inode->data_vdi_id[idx] == 0) { | |
2703 | break; | |
2704 | } | |
2705 | } | |
2706 | if (idx == start) { | |
2707 | /* Get the longest length of unallocated sectors */ | |
2708 | ret = 0; | |
2709 | for (idx = start + 1; idx < end; idx++) { | |
2710 | if (inode->data_vdi_id[idx] != 0) { | |
2711 | break; | |
2712 | } | |
2713 | } | |
2714 | } | |
2715 | ||
876eb1b0 | 2716 | *pnum = (idx - start) * object_size / BDRV_SECTOR_SIZE; |
8d71c631 LY |
2717 | if (*pnum > nb_sectors) { |
2718 | *pnum = nb_sectors; | |
2719 | } | |
2720 | return ret; | |
2721 | } | |
2722 | ||
85829722 LY |
2723 | static int64_t sd_get_allocated_file_size(BlockDriverState *bs) |
2724 | { | |
2725 | BDRVSheepdogState *s = bs->opaque; | |
2726 | SheepdogInode *inode = &s->inode; | |
876eb1b0 TI |
2727 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
2728 | unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size); | |
85829722 LY |
2729 | uint64_t size = 0; |
2730 | ||
2731 | for (i = 0; i < last; i++) { | |
2732 | if (inode->data_vdi_id[i] == 0) { | |
2733 | continue; | |
2734 | } | |
876eb1b0 | 2735 | size += object_size; |
85829722 LY |
2736 | } |
2737 | return size; | |
2738 | } | |
2739 | ||
b222237b CL |
2740 | static QemuOptsList sd_create_opts = { |
2741 | .name = "sheepdog-create-opts", | |
2742 | .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head), | |
2743 | .desc = { | |
2744 | { | |
2745 | .name = BLOCK_OPT_SIZE, | |
2746 | .type = QEMU_OPT_SIZE, | |
2747 | .help = "Virtual disk size" | |
2748 | }, | |
2749 | { | |
2750 | .name = BLOCK_OPT_BACKING_FILE, | |
2751 | .type = QEMU_OPT_STRING, | |
2752 | .help = "File name of a base image" | |
2753 | }, | |
2754 | { | |
2755 | .name = BLOCK_OPT_PREALLOC, | |
2756 | .type = QEMU_OPT_STRING, | |
2757 | .help = "Preallocation mode (allowed values: off, full)" | |
2758 | }, | |
2759 | { | |
2760 | .name = BLOCK_OPT_REDUNDANCY, | |
2761 | .type = QEMU_OPT_STRING, | |
2762 | .help = "Redundancy of the image" | |
2763 | }, | |
876eb1b0 TI |
2764 | { |
2765 | .name = BLOCK_OPT_OBJECT_SIZE, | |
2766 | .type = QEMU_OPT_SIZE, | |
2767 | .help = "Object size of the image" | |
2768 | }, | |
b222237b CL |
2769 | { /* end of list */ } |
2770 | } | |
33b1db1c MK |
2771 | }; |
2772 | ||
5d6768e3 | 2773 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2774 | .format_name = "sheepdog", |
2775 | .protocol_name = "sheepdog", | |
2776 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2777 | .bdrv_needs_filename = true, |
33b1db1c | 2778 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2779 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2780 | .bdrv_reopen_commit = sd_reopen_commit, | |
2781 | .bdrv_reopen_abort = sd_reopen_abort, | |
33b1db1c | 2782 | .bdrv_close = sd_close, |
c282e1fd | 2783 | .bdrv_create = sd_create, |
e4f5c1bf | 2784 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
33b1db1c | 2785 | .bdrv_getlength = sd_getlength, |
85829722 | 2786 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
33b1db1c MK |
2787 | .bdrv_truncate = sd_truncate, |
2788 | ||
2df46246 MK |
2789 | .bdrv_co_readv = sd_co_readv, |
2790 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2791 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
cac8f4a6 | 2792 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2793 | .bdrv_co_get_block_status = sd_co_get_block_status, |
33b1db1c MK |
2794 | |
2795 | .bdrv_snapshot_create = sd_snapshot_create, | |
2796 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2797 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2798 | .bdrv_snapshot_list = sd_snapshot_list, | |
2799 | ||
2800 | .bdrv_save_vmstate = sd_save_vmstate, | |
2801 | .bdrv_load_vmstate = sd_load_vmstate, | |
2802 | ||
84390bed SH |
2803 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2804 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2805 | ||
b222237b | 2806 | .create_opts = &sd_create_opts, |
33b1db1c MK |
2807 | }; |
2808 | ||
5d6768e3 MK |
2809 | static BlockDriver bdrv_sheepdog_tcp = { |
2810 | .format_name = "sheepdog", | |
2811 | .protocol_name = "sheepdog+tcp", | |
2812 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2813 | .bdrv_needs_filename = true, |
5d6768e3 | 2814 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2815 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2816 | .bdrv_reopen_commit = sd_reopen_commit, | |
2817 | .bdrv_reopen_abort = sd_reopen_abort, | |
5d6768e3 | 2818 | .bdrv_close = sd_close, |
c282e1fd | 2819 | .bdrv_create = sd_create, |
e4f5c1bf | 2820 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
5d6768e3 | 2821 | .bdrv_getlength = sd_getlength, |
85829722 | 2822 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
5d6768e3 MK |
2823 | .bdrv_truncate = sd_truncate, |
2824 | ||
2825 | .bdrv_co_readv = sd_co_readv, | |
2826 | .bdrv_co_writev = sd_co_writev, | |
2827 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2828 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2829 | .bdrv_co_get_block_status = sd_co_get_block_status, |
5d6768e3 MK |
2830 | |
2831 | .bdrv_snapshot_create = sd_snapshot_create, | |
2832 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2833 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2834 | .bdrv_snapshot_list = sd_snapshot_list, | |
2835 | ||
2836 | .bdrv_save_vmstate = sd_save_vmstate, | |
2837 | .bdrv_load_vmstate = sd_load_vmstate, | |
2838 | ||
84390bed SH |
2839 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2840 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2841 | ||
b222237b | 2842 | .create_opts = &sd_create_opts, |
5d6768e3 MK |
2843 | }; |
2844 | ||
1b8bbb46 MK |
2845 | static BlockDriver bdrv_sheepdog_unix = { |
2846 | .format_name = "sheepdog", | |
2847 | .protocol_name = "sheepdog+unix", | |
2848 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2849 | .bdrv_needs_filename = true, |
1b8bbb46 | 2850 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2851 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2852 | .bdrv_reopen_commit = sd_reopen_commit, | |
2853 | .bdrv_reopen_abort = sd_reopen_abort, | |
1b8bbb46 | 2854 | .bdrv_close = sd_close, |
c282e1fd | 2855 | .bdrv_create = sd_create, |
3ac21627 | 2856 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
1b8bbb46 | 2857 | .bdrv_getlength = sd_getlength, |
85829722 | 2858 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
1b8bbb46 MK |
2859 | .bdrv_truncate = sd_truncate, |
2860 | ||
2861 | .bdrv_co_readv = sd_co_readv, | |
2862 | .bdrv_co_writev = sd_co_writev, | |
2863 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2864 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2865 | .bdrv_co_get_block_status = sd_co_get_block_status, |
1b8bbb46 MK |
2866 | |
2867 | .bdrv_snapshot_create = sd_snapshot_create, | |
2868 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2869 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2870 | .bdrv_snapshot_list = sd_snapshot_list, | |
2871 | ||
2872 | .bdrv_save_vmstate = sd_save_vmstate, | |
2873 | .bdrv_load_vmstate = sd_load_vmstate, | |
2874 | ||
84390bed SH |
2875 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2876 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2877 | ||
b222237b | 2878 | .create_opts = &sd_create_opts, |
1b8bbb46 MK |
2879 | }; |
2880 | ||
33b1db1c MK |
2881 | static void bdrv_sheepdog_init(void) |
2882 | { | |
2883 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 2884 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 2885 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
2886 | } |
2887 | block_init(bdrv_sheepdog_init); |