]>
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 | 14 | |
80c71a24 | 15 | #include "qemu/osdep.h" |
da34e65c | 16 | #include "qapi/error.h" |
5d6768e3 | 17 | #include "qemu/uri.h" |
1de7afc9 PB |
18 | #include "qemu/error-report.h" |
19 | #include "qemu/sockets.h" | |
737e150e | 20 | #include "block/block_int.h" |
fba98d45 | 21 | #include "sysemu/block-backend.h" |
1de7afc9 | 22 | #include "qemu/bitops.h" |
f348b6d1 | 23 | #include "qemu/cutils.h" |
33b1db1c MK |
24 | |
25 | #define SD_PROTO_VER 0x01 | |
26 | ||
27 | #define SD_DEFAULT_ADDR "localhost" | |
25af257d | 28 | #define SD_DEFAULT_PORT 7000 |
33b1db1c MK |
29 | |
30 | #define SD_OP_CREATE_AND_WRITE_OBJ 0x01 | |
31 | #define SD_OP_READ_OBJ 0x02 | |
32 | #define SD_OP_WRITE_OBJ 0x03 | |
cac8f4a6 | 33 | /* 0x04 is used internally by Sheepdog */ |
33b1db1c MK |
34 | |
35 | #define SD_OP_NEW_VDI 0x11 | |
36 | #define SD_OP_LOCK_VDI 0x12 | |
37 | #define SD_OP_RELEASE_VDI 0x13 | |
38 | #define SD_OP_GET_VDI_INFO 0x14 | |
39 | #define SD_OP_READ_VDIS 0x15 | |
47622c44 | 40 | #define SD_OP_FLUSH_VDI 0x16 |
859e5553 | 41 | #define SD_OP_DEL_VDI 0x17 |
876eb1b0 | 42 | #define SD_OP_GET_CLUSTER_DEFAULT 0x18 |
33b1db1c MK |
43 | |
44 | #define SD_FLAG_CMD_WRITE 0x01 | |
45 | #define SD_FLAG_CMD_COW 0x02 | |
0e7106d8 LY |
46 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
47 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
33b1db1c MK |
48 | |
49 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
50 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
51 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
52 | #define SD_RES_EIO 0x03 /* I/O error */ | |
53 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
54 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
55 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
56 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
57 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
58 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
59 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
60 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
61 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
62 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
63 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
64 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
65 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
66 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
67 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
68 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
69 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
70 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
71 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
72 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
73 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
fca23f0a | 74 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ |
6a0b5490 | 75 | #define SD_RES_READONLY 0x1A /* Object is read-only */ |
33b1db1c MK |
76 | |
77 | /* | |
78 | * Object ID rules | |
79 | * | |
80 | * 0 - 19 (20 bits): data object space | |
81 | * 20 - 31 (12 bits): reserved data object space | |
82 | * 32 - 55 (24 bits): vdi object space | |
83 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 84 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
85 | */ |
86 | ||
87 | #define VDI_SPACE_SHIFT 32 | |
88 | #define VDI_BIT (UINT64_C(1) << 63) | |
89 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
90 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
91 | #define MAX_CHILDREN 1024 | |
92 | #define SD_MAX_VDI_LEN 256 | |
93 | #define SD_MAX_VDI_TAG_LEN 256 | |
94 | #define SD_NR_VDIS (1U << 24) | |
95 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
96 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
876eb1b0 | 97 | #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22 |
b3af018f LY |
98 | /* |
99 | * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and | |
100 | * (SD_EC_MAX_STRIP - 1) for parity strips | |
101 | * | |
102 | * SD_MAX_COPIES is sum of number of data strips and parity strips. | |
103 | */ | |
104 | #define SD_EC_MAX_STRIP 16 | |
105 | #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1) | |
33b1db1c MK |
106 | |
107 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
108 | #define CURRENT_VDI_ID 0 | |
109 | ||
1dbfafed HM |
110 | #define LOCK_TYPE_NORMAL 0 |
111 | #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */ | |
112 | ||
33b1db1c MK |
113 | typedef struct SheepdogReq { |
114 | uint8_t proto_ver; | |
115 | uint8_t opcode; | |
116 | uint16_t flags; | |
117 | uint32_t epoch; | |
118 | uint32_t id; | |
119 | uint32_t data_length; | |
120 | uint32_t opcode_specific[8]; | |
121 | } SheepdogReq; | |
122 | ||
123 | typedef struct SheepdogRsp { | |
124 | uint8_t proto_ver; | |
125 | uint8_t opcode; | |
126 | uint16_t flags; | |
127 | uint32_t epoch; | |
128 | uint32_t id; | |
129 | uint32_t data_length; | |
130 | uint32_t result; | |
131 | uint32_t opcode_specific[7]; | |
132 | } SheepdogRsp; | |
133 | ||
134 | typedef struct SheepdogObjReq { | |
135 | uint8_t proto_ver; | |
136 | uint8_t opcode; | |
137 | uint16_t flags; | |
138 | uint32_t epoch; | |
139 | uint32_t id; | |
140 | uint32_t data_length; | |
141 | uint64_t oid; | |
142 | uint64_t cow_oid; | |
29a67f7e | 143 | uint8_t copies; |
1841f880 LY |
144 | uint8_t copy_policy; |
145 | uint8_t reserved[6]; | |
33b1db1c MK |
146 | uint64_t offset; |
147 | } SheepdogObjReq; | |
148 | ||
149 | typedef struct SheepdogObjRsp { | |
150 | uint8_t proto_ver; | |
151 | uint8_t opcode; | |
152 | uint16_t flags; | |
153 | uint32_t epoch; | |
154 | uint32_t id; | |
155 | uint32_t data_length; | |
156 | uint32_t result; | |
29a67f7e | 157 | uint8_t copies; |
1841f880 LY |
158 | uint8_t copy_policy; |
159 | uint8_t reserved[2]; | |
33b1db1c MK |
160 | uint32_t pad[6]; |
161 | } SheepdogObjRsp; | |
162 | ||
163 | typedef struct SheepdogVdiReq { | |
164 | uint8_t proto_ver; | |
165 | uint8_t opcode; | |
166 | uint16_t flags; | |
167 | uint32_t epoch; | |
168 | uint32_t id; | |
169 | uint32_t data_length; | |
170 | uint64_t vdi_size; | |
9f23fce7 | 171 | uint32_t base_vdi_id; |
29a67f7e | 172 | uint8_t copies; |
1841f880 | 173 | uint8_t copy_policy; |
876eb1b0 TI |
174 | uint8_t store_policy; |
175 | uint8_t block_size_shift; | |
33b1db1c | 176 | uint32_t snapid; |
1dbfafed HM |
177 | uint32_t type; |
178 | uint32_t pad[2]; | |
33b1db1c MK |
179 | } SheepdogVdiReq; |
180 | ||
181 | typedef struct SheepdogVdiRsp { | |
182 | uint8_t proto_ver; | |
183 | uint8_t opcode; | |
184 | uint16_t flags; | |
185 | uint32_t epoch; | |
186 | uint32_t id; | |
187 | uint32_t data_length; | |
188 | uint32_t result; | |
189 | uint32_t rsvd; | |
190 | uint32_t vdi_id; | |
191 | uint32_t pad[5]; | |
192 | } SheepdogVdiRsp; | |
193 | ||
876eb1b0 TI |
194 | typedef struct SheepdogClusterRsp { |
195 | uint8_t proto_ver; | |
196 | uint8_t opcode; | |
197 | uint16_t flags; | |
198 | uint32_t epoch; | |
199 | uint32_t id; | |
200 | uint32_t data_length; | |
201 | uint32_t result; | |
202 | uint8_t nr_copies; | |
203 | uint8_t copy_policy; | |
204 | uint8_t block_size_shift; | |
205 | uint8_t __pad1; | |
206 | uint32_t __pad2[6]; | |
207 | } SheepdogClusterRsp; | |
208 | ||
33b1db1c MK |
209 | typedef struct SheepdogInode { |
210 | char name[SD_MAX_VDI_LEN]; | |
211 | char tag[SD_MAX_VDI_TAG_LEN]; | |
212 | uint64_t ctime; | |
213 | uint64_t snap_ctime; | |
214 | uint64_t vm_clock_nsec; | |
215 | uint64_t vdi_size; | |
216 | uint64_t vm_state_size; | |
217 | uint16_t copy_policy; | |
218 | uint8_t nr_copies; | |
219 | uint8_t block_size_shift; | |
220 | uint32_t snap_id; | |
221 | uint32_t vdi_id; | |
222 | uint32_t parent_vdi_id; | |
223 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
224 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
225 | } SheepdogInode; | |
226 | ||
5d039bab HM |
227 | #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id) |
228 | ||
33b1db1c MK |
229 | /* |
230 | * 64 bit FNV-1a non-zero initial basis | |
231 | */ | |
232 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
233 | ||
234 | /* | |
235 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
236 | */ | |
237 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
238 | { | |
239 | unsigned char *bp = buf; | |
240 | unsigned char *be = bp + len; | |
241 | while (bp < be) { | |
242 | hval ^= (uint64_t) *bp++; | |
243 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
244 | (hval << 7) + (hval << 8) + (hval << 40); | |
245 | } | |
246 | return hval; | |
247 | } | |
248 | ||
2f536801 | 249 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
250 | { |
251 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
252 | } | |
253 | ||
2f536801 | 254 | static inline bool is_data_obj(uint64_t oid) |
33b1db1c MK |
255 | { |
256 | return !(VDI_BIT & oid); | |
257 | } | |
258 | ||
259 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
260 | { | |
261 | return oid & (MAX_DATA_OBJS - 1); | |
262 | } | |
263 | ||
72e0996c MK |
264 | static inline uint32_t oid_to_vid(uint64_t oid) |
265 | { | |
266 | return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT; | |
267 | } | |
268 | ||
33b1db1c MK |
269 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) |
270 | { | |
271 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
272 | } | |
273 | ||
274 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
275 | { | |
276 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
277 | } | |
278 | ||
279 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
280 | { | |
281 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
282 | } | |
283 | ||
2f536801 | 284 | static inline bool is_snapshot(struct SheepdogInode *inode) |
33b1db1c MK |
285 | { |
286 | return !!inode->snap_ctime; | |
287 | } | |
288 | ||
eab8eb8d VT |
289 | static inline size_t count_data_objs(const struct SheepdogInode *inode) |
290 | { | |
291 | return DIV_ROUND_UP(inode->vdi_size, | |
292 | (1UL << inode->block_size_shift)); | |
293 | } | |
294 | ||
2440a2c3 | 295 | #undef DPRINTF |
33b1db1c | 296 | #ifdef DEBUG_SDOG |
ed79f37d | 297 | #define DEBUG_SDOG_PRINT 1 |
33b1db1c | 298 | #else |
ed79f37d | 299 | #define DEBUG_SDOG_PRINT 0 |
33b1db1c | 300 | #endif |
ed79f37d ZJ |
301 | #define DPRINTF(fmt, args...) \ |
302 | do { \ | |
303 | if (DEBUG_SDOG_PRINT) { \ | |
304 | fprintf(stderr, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
305 | } \ | |
306 | } while (0) | |
33b1db1c MK |
307 | |
308 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
28ddd08c | 309 | typedef struct BDRVSheepdogState BDRVSheepdogState; |
33b1db1c MK |
310 | |
311 | typedef struct AIOReq { | |
312 | SheepdogAIOCB *aiocb; | |
313 | unsigned int iov_offset; | |
314 | ||
315 | uint64_t oid; | |
316 | uint64_t base_oid; | |
317 | uint64_t offset; | |
318 | unsigned int data_len; | |
319 | uint8_t flags; | |
320 | uint32_t id; | |
b544c1ab | 321 | bool create; |
33b1db1c | 322 | |
c292ee6a | 323 | QLIST_ENTRY(AIOReq) aio_siblings; |
33b1db1c MK |
324 | } AIOReq; |
325 | ||
326 | enum AIOCBState { | |
327 | AIOCB_WRITE_UDATA, | |
328 | AIOCB_READ_UDATA, | |
47783072 | 329 | AIOCB_FLUSH_CACHE, |
cac8f4a6 | 330 | AIOCB_DISCARD_OBJ, |
33b1db1c MK |
331 | }; |
332 | ||
498f2140 | 333 | #define AIOCBOverlapping(x, y) \ |
6a55c82c HM |
334 | (!(x->max_affect_data_idx < y->min_affect_data_idx \ |
335 | || y->max_affect_data_idx < x->min_affect_data_idx)) | |
336 | ||
33b1db1c | 337 | struct SheepdogAIOCB { |
28ddd08c | 338 | BDRVSheepdogState *s; |
33b1db1c MK |
339 | |
340 | QEMUIOVector *qiov; | |
341 | ||
342 | int64_t sector_num; | |
343 | int nb_sectors; | |
344 | ||
345 | int ret; | |
346 | enum AIOCBState aiocb_type; | |
347 | ||
2df46246 | 348 | Coroutine *coroutine; |
1d732d7d | 349 | int nr_pending; |
6a55c82c HM |
350 | |
351 | uint32_t min_affect_data_idx; | |
352 | uint32_t max_affect_data_idx; | |
353 | ||
498f2140 HM |
354 | /* |
355 | * The difference between affect_data_idx and dirty_data_idx: | |
356 | * affect_data_idx represents range of index of all request types. | |
357 | * dirty_data_idx represents range of index updated by COW requests. | |
358 | * dirty_data_idx is used for updating an inode object. | |
359 | */ | |
360 | uint32_t min_dirty_data_idx; | |
361 | uint32_t max_dirty_data_idx; | |
362 | ||
6a55c82c | 363 | QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings; |
33b1db1c MK |
364 | }; |
365 | ||
28ddd08c | 366 | struct BDRVSheepdogState { |
011603ca | 367 | BlockDriverState *bs; |
84390bed | 368 | AioContext *aio_context; |
011603ca | 369 | |
33b1db1c MK |
370 | SheepdogInode inode; |
371 | ||
33b1db1c | 372 | char name[SD_MAX_VDI_LEN]; |
2f536801 | 373 | bool is_snapshot; |
0e7106d8 | 374 | uint32_t cache_flags; |
cac8f4a6 | 375 | bool discard_supported; |
33b1db1c | 376 | |
25af257d | 377 | char *host_spec; |
1b8bbb46 | 378 | bool is_unix; |
33b1db1c MK |
379 | int fd; |
380 | ||
2df46246 MK |
381 | CoMutex lock; |
382 | Coroutine *co_send; | |
383 | Coroutine *co_recv; | |
384 | ||
33b1db1c | 385 | uint32_t aioreq_seq_num; |
011603ca MK |
386 | |
387 | /* Every aio request must be linked to either of these queues. */ | |
c292ee6a | 388 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
011603ca | 389 | QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head; |
6a55c82c | 390 | |
498f2140 | 391 | CoQueue overlapping_queue; |
6a55c82c | 392 | QLIST_HEAD(inflight_aiocb_head, SheepdogAIOCB) inflight_aiocb_head; |
28ddd08c | 393 | }; |
33b1db1c | 394 | |
4da65c80 LY |
395 | typedef struct BDRVSheepdogReopenState { |
396 | int fd; | |
397 | int cache_flags; | |
398 | } BDRVSheepdogReopenState; | |
399 | ||
33b1db1c MK |
400 | static const char * sd_strerror(int err) |
401 | { | |
402 | int i; | |
403 | ||
404 | static const struct { | |
405 | int err; | |
406 | const char *desc; | |
407 | } errors[] = { | |
408 | {SD_RES_SUCCESS, "Success"}, | |
409 | {SD_RES_UNKNOWN, "Unknown error"}, | |
410 | {SD_RES_NO_OBJ, "No object found"}, | |
411 | {SD_RES_EIO, "I/O error"}, | |
412 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
413 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
414 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
415 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
416 | {SD_RES_NO_VDI, "No vdi found"}, | |
417 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
418 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
419 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
420 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
421 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
422 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
423 | {SD_RES_STARTUP, "The system is still booting"}, | |
424 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
425 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
426 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
427 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
428 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
429 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
430 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
431 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
432 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
fca23f0a | 433 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, |
6a0b5490 | 434 | {SD_RES_READONLY, "Object is read-only"}, |
33b1db1c MK |
435 | }; |
436 | ||
437 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
438 | if (errors[i].err == err) { | |
439 | return errors[i].desc; | |
440 | } | |
441 | } | |
442 | ||
443 | return "Invalid error code"; | |
444 | } | |
445 | ||
446 | /* | |
447 | * Sheepdog I/O handling: | |
448 | * | |
2df46246 | 449 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
c292ee6a | 450 | * link the requests to the inflight_list in the |
e80ab33d | 451 | * BDRVSheepdogState. The function yields while waiting for |
2df46246 | 452 | * receiving the response. |
33b1db1c | 453 | * |
2df46246 | 454 | * 2. We receive the response in aio_read_response, the fd handler to |
e80ab33d PB |
455 | * the sheepdog connection. We switch back to sd_co_readv/sd_writev |
456 | * after all the requests belonging to the AIOCB are finished. If | |
457 | * needed, sd_co_writev will send another requests for the vdi object. | |
33b1db1c MK |
458 | */ |
459 | ||
460 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
461 | uint64_t oid, unsigned int data_len, | |
b544c1ab | 462 | uint64_t offset, uint8_t flags, bool create, |
33b1db1c MK |
463 | uint64_t base_oid, unsigned int iov_offset) |
464 | { | |
465 | AIOReq *aio_req; | |
466 | ||
7267c094 | 467 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
468 | aio_req->aiocb = acb; |
469 | aio_req->iov_offset = iov_offset; | |
470 | aio_req->oid = oid; | |
471 | aio_req->base_oid = base_oid; | |
472 | aio_req->offset = offset; | |
473 | aio_req->data_len = data_len; | |
474 | aio_req->flags = flags; | |
475 | aio_req->id = s->aioreq_seq_num++; | |
b544c1ab | 476 | aio_req->create = create; |
33b1db1c | 477 | |
1d732d7d | 478 | acb->nr_pending++; |
33b1db1c MK |
479 | return aio_req; |
480 | } | |
481 | ||
acf6e5f0 PB |
482 | static void wait_for_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *acb) |
483 | { | |
484 | SheepdogAIOCB *cb; | |
485 | ||
486 | retry: | |
487 | QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) { | |
488 | if (AIOCBOverlapping(acb, cb)) { | |
1ace7cea | 489 | qemu_co_queue_wait(&s->overlapping_queue, NULL); |
acf6e5f0 PB |
490 | goto retry; |
491 | } | |
492 | } | |
493 | } | |
494 | ||
28ddd08c PB |
495 | static void sd_aio_setup(SheepdogAIOCB *acb, BDRVSheepdogState *s, |
496 | QEMUIOVector *qiov, int64_t sector_num, int nb_sectors, | |
497 | int type) | |
33b1db1c | 498 | { |
6a55c82c | 499 | uint32_t object_size; |
6a55c82c HM |
500 | |
501 | object_size = (UINT32_C(1) << s->inode.block_size_shift); | |
33b1db1c | 502 | |
28ddd08c | 503 | acb->s = s; |
33b1db1c MK |
504 | |
505 | acb->qiov = qiov; | |
506 | ||
507 | acb->sector_num = sector_num; | |
508 | acb->nb_sectors = nb_sectors; | |
509 | ||
2df46246 | 510 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c | 511 | acb->ret = 0; |
1d732d7d | 512 | acb->nr_pending = 0; |
6a55c82c HM |
513 | |
514 | acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
515 | acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE + | |
516 | acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size; | |
517 | ||
498f2140 HM |
518 | acb->min_dirty_data_idx = UINT32_MAX; |
519 | acb->max_dirty_data_idx = 0; | |
28ddd08c | 520 | acb->aiocb_type = type; |
acf6e5f0 PB |
521 | |
522 | if (type == AIOCB_FLUSH_CACHE) { | |
523 | return; | |
524 | } | |
525 | ||
526 | wait_for_overlapping_aiocb(s, acb); | |
527 | QLIST_INSERT_HEAD(&s->inflight_aiocb_head, acb, aiocb_siblings); | |
33b1db1c MK |
528 | } |
529 | ||
833a7cc3 | 530 | /* Return -EIO in case of error, file descriptor on success */ |
dfb12bf8 | 531 | static int connect_to_sdog(BDRVSheepdogState *s, Error **errp) |
33b1db1c | 532 | { |
25af257d | 533 | int fd; |
33b1db1c | 534 | |
1b8bbb46 | 535 | if (s->is_unix) { |
dfb12bf8 | 536 | fd = unix_connect(s->host_spec, errp); |
1b8bbb46 | 537 | } else { |
dfb12bf8 | 538 | fd = inet_connect(s->host_spec, errp); |
1b8bbb46 | 539 | |
dfb12bf8 | 540 | if (fd >= 0) { |
1b8bbb46 MK |
541 | int ret = socket_set_nodelay(fd); |
542 | if (ret < 0) { | |
543 | error_report("%s", strerror(errno)); | |
544 | } | |
545 | } | |
546 | } | |
33b1db1c | 547 | |
dfb12bf8 | 548 | if (fd >= 0) { |
f9e8cacc | 549 | qemu_set_nonblock(fd); |
833a7cc3 LY |
550 | } else { |
551 | fd = -EIO; | |
33b1db1c MK |
552 | } |
553 | ||
33b1db1c MK |
554 | return fd; |
555 | } | |
556 | ||
833a7cc3 | 557 | /* Return 0 on success and -errno in case of error */ |
e0d93a89 MK |
558 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
559 | unsigned int *wlen) | |
47622c44 LY |
560 | { |
561 | int ret; | |
562 | ||
563 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
80731d9d | 564 | if (ret != sizeof(*hdr)) { |
47622c44 | 565 | error_report("failed to send a req, %s", strerror(errno)); |
b16a44e1 | 566 | return -errno; |
47622c44 LY |
567 | } |
568 | ||
569 | ret = qemu_co_send(sockfd, data, *wlen); | |
80731d9d | 570 | if (ret != *wlen) { |
47622c44 | 571 | error_report("failed to send a req, %s", strerror(errno)); |
b16a44e1 | 572 | return -errno; |
47622c44 LY |
573 | } |
574 | ||
575 | return ret; | |
576 | } | |
e0d93a89 | 577 | |
cddd4ac7 MK |
578 | typedef struct SheepdogReqCo { |
579 | int sockfd; | |
f11672db | 580 | BlockDriverState *bs; |
84390bed | 581 | AioContext *aio_context; |
cddd4ac7 MK |
582 | SheepdogReq *hdr; |
583 | void *data; | |
584 | unsigned int *wlen; | |
585 | unsigned int *rlen; | |
586 | int ret; | |
587 | bool finished; | |
9d456654 | 588 | Coroutine *co; |
cddd4ac7 MK |
589 | } SheepdogReqCo; |
590 | ||
9d456654 PB |
591 | static void restart_co_req(void *opaque) |
592 | { | |
593 | SheepdogReqCo *srco = opaque; | |
594 | ||
595 | aio_co_wake(srco->co); | |
596 | } | |
597 | ||
cddd4ac7 | 598 | static coroutine_fn void do_co_req(void *opaque) |
47622c44 LY |
599 | { |
600 | int ret; | |
cddd4ac7 MK |
601 | SheepdogReqCo *srco = opaque; |
602 | int sockfd = srco->sockfd; | |
603 | SheepdogReq *hdr = srco->hdr; | |
604 | void *data = srco->data; | |
605 | unsigned int *wlen = srco->wlen; | |
606 | unsigned int *rlen = srco->rlen; | |
2dfcca3b | 607 | |
9d456654 | 608 | srco->co = qemu_coroutine_self(); |
dca21ef2 | 609 | aio_set_fd_handler(srco->aio_context, sockfd, false, |
9d456654 | 610 | NULL, restart_co_req, NULL, srco); |
47622c44 | 611 | |
47622c44 LY |
612 | ret = send_co_req(sockfd, hdr, data, wlen); |
613 | if (ret < 0) { | |
614 | goto out; | |
615 | } | |
616 | ||
dca21ef2 | 617 | aio_set_fd_handler(srco->aio_context, sockfd, false, |
9d456654 | 618 | restart_co_req, NULL, NULL, srco); |
2dfcca3b | 619 | |
47622c44 | 620 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); |
80731d9d | 621 | if (ret != sizeof(*hdr)) { |
47622c44 | 622 | error_report("failed to get a rsp, %s", strerror(errno)); |
cb595887 | 623 | ret = -errno; |
47622c44 LY |
624 | goto out; |
625 | } | |
626 | ||
627 | if (*rlen > hdr->data_length) { | |
628 | *rlen = hdr->data_length; | |
629 | } | |
630 | ||
631 | if (*rlen) { | |
632 | ret = qemu_co_recv(sockfd, data, *rlen); | |
80731d9d | 633 | if (ret != *rlen) { |
47622c44 | 634 | error_report("failed to get the data, %s", strerror(errno)); |
cb595887 | 635 | ret = -errno; |
47622c44 LY |
636 | goto out; |
637 | } | |
638 | } | |
639 | ret = 0; | |
640 | out: | |
ed9ba724 MK |
641 | /* there is at most one request for this sockfd, so it is safe to |
642 | * set each handler to NULL. */ | |
dca21ef2 | 643 | aio_set_fd_handler(srco->aio_context, sockfd, false, |
f6a51c84 | 644 | NULL, NULL, NULL, NULL); |
cddd4ac7 | 645 | |
9d456654 | 646 | srco->co = NULL; |
cddd4ac7 MK |
647 | srco->ret = ret; |
648 | srco->finished = true; | |
c9d1a561 PB |
649 | if (srco->bs) { |
650 | bdrv_wakeup(srco->bs); | |
651 | } | |
cddd4ac7 MK |
652 | } |
653 | ||
833a7cc3 LY |
654 | /* |
655 | * Send the request to the sheep in a synchronous manner. | |
656 | * | |
657 | * Return 0 on success, -errno in case of error. | |
658 | */ | |
f11672db | 659 | static int do_req(int sockfd, BlockDriverState *bs, SheepdogReq *hdr, |
84390bed | 660 | void *data, unsigned int *wlen, unsigned int *rlen) |
cddd4ac7 MK |
661 | { |
662 | Coroutine *co; | |
663 | SheepdogReqCo srco = { | |
664 | .sockfd = sockfd, | |
f11672db PB |
665 | .aio_context = bs ? bdrv_get_aio_context(bs) : qemu_get_aio_context(), |
666 | .bs = bs, | |
cddd4ac7 MK |
667 | .hdr = hdr, |
668 | .data = data, | |
669 | .wlen = wlen, | |
670 | .rlen = rlen, | |
671 | .ret = 0, | |
672 | .finished = false, | |
673 | }; | |
674 | ||
675 | if (qemu_in_coroutine()) { | |
676 | do_co_req(&srco); | |
677 | } else { | |
0b8b8753 | 678 | co = qemu_coroutine_create(do_co_req, &srco); |
f11672db PB |
679 | if (bs) { |
680 | qemu_coroutine_enter(co); | |
681 | BDRV_POLL_WHILE(bs, !srco.finished); | |
682 | } else { | |
683 | qemu_coroutine_enter(co); | |
684 | while (!srco.finished) { | |
685 | aio_poll(qemu_get_aio_context(), true); | |
686 | } | |
cddd4ac7 MK |
687 | } |
688 | } | |
689 | ||
690 | return srco.ret; | |
47622c44 LY |
691 | } |
692 | ||
a37dcdf9 | 693 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
694 | struct iovec *iov, int niov, |
695 | enum AIOCBState aiocb_type); | |
a37dcdf9 | 696 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req); |
72e0996c | 697 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag); |
356b4ca2 | 698 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp); |
011603ca | 699 | static void co_write_request(void *opaque); |
7dc1cde0 | 700 | |
011603ca MK |
701 | static coroutine_fn void reconnect_to_sdog(void *opaque) |
702 | { | |
703 | BDRVSheepdogState *s = opaque; | |
704 | AIOReq *aio_req, *next; | |
705 | ||
dca21ef2 | 706 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, |
f6a51c84 | 707 | NULL, NULL, NULL); |
011603ca MK |
708 | close(s->fd); |
709 | s->fd = -1; | |
710 | ||
711 | /* Wait for outstanding write requests to be completed. */ | |
712 | while (s->co_send != NULL) { | |
713 | co_write_request(opaque); | |
714 | } | |
715 | ||
716 | /* Try to reconnect the sheepdog server every one second. */ | |
717 | while (s->fd < 0) { | |
a780dea0 | 718 | Error *local_err = NULL; |
356b4ca2 | 719 | s->fd = get_sheep_fd(s, &local_err); |
011603ca MK |
720 | if (s->fd < 0) { |
721 | DPRINTF("Wait for connection to be established\n"); | |
565f65d2 | 722 | error_report_err(local_err); |
011603ca MK |
723 | co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME, |
724 | 1000000000ULL); | |
725 | } | |
726 | }; | |
727 | ||
728 | /* | |
729 | * Now we have to resend all the request in the inflight queue. However, | |
730 | * resend_aioreq() can yield and newly created requests can be added to the | |
731 | * inflight queue before the coroutine is resumed. To avoid mixing them, we | |
732 | * have to move all the inflight requests to the failed queue before | |
733 | * resend_aioreq() is called. | |
734 | */ | |
735 | QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) { | |
736 | QLIST_REMOVE(aio_req, aio_siblings); | |
737 | QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings); | |
738 | } | |
739 | ||
740 | /* Resend all the failed aio requests. */ | |
741 | while (!QLIST_EMPTY(&s->failed_aio_head)) { | |
742 | aio_req = QLIST_FIRST(&s->failed_aio_head); | |
743 | QLIST_REMOVE(aio_req, aio_siblings); | |
011603ca MK |
744 | resend_aioreq(s, aio_req); |
745 | } | |
746 | } | |
747 | ||
33b1db1c MK |
748 | /* |
749 | * Receive responses of the I/O requests. | |
750 | * | |
751 | * This function is registered as a fd handler, and called from the | |
752 | * main loop when s->fd is ready for reading responses. | |
753 | */ | |
d8716b41 | 754 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
755 | { |
756 | SheepdogObjRsp rsp; | |
757 | BDRVSheepdogState *s = opaque; | |
758 | int fd = s->fd; | |
759 | int ret; | |
760 | AIOReq *aio_req = NULL; | |
761 | SheepdogAIOCB *acb; | |
cac8f4a6 | 762 | uint64_t idx; |
33b1db1c | 763 | |
33b1db1c | 764 | /* read a header */ |
8c5135f9 | 765 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
80731d9d | 766 | if (ret != sizeof(rsp)) { |
6daf194d | 767 | error_report("failed to get the header, %s", strerror(errno)); |
011603ca | 768 | goto err; |
33b1db1c MK |
769 | } |
770 | ||
c292ee6a MK |
771 | /* find the right aio_req from the inflight aio list */ |
772 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
33b1db1c MK |
773 | if (aio_req->id == rsp.id) { |
774 | break; | |
775 | } | |
776 | } | |
777 | if (!aio_req) { | |
6daf194d | 778 | error_report("cannot find aio_req %x", rsp.id); |
011603ca | 779 | goto err; |
33b1db1c MK |
780 | } |
781 | ||
782 | acb = aio_req->aiocb; | |
783 | ||
784 | switch (acb->aiocb_type) { | |
785 | case AIOCB_WRITE_UDATA: | |
786 | if (!is_data_obj(aio_req->oid)) { | |
787 | break; | |
788 | } | |
789 | idx = data_oid_to_idx(aio_req->oid); | |
790 | ||
b544c1ab | 791 | if (aio_req->create) { |
33b1db1c MK |
792 | /* |
793 | * If the object is newly created one, we need to update | |
794 | * the vdi object (metadata object). min_dirty_data_idx | |
795 | * and max_dirty_data_idx are changed to include updated | |
796 | * index between them. | |
797 | */ | |
bd751f22 LY |
798 | if (rsp.result == SD_RES_SUCCESS) { |
799 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
498f2140 HM |
800 | acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx); |
801 | acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx); | |
bd751f22 | 802 | } |
33b1db1c MK |
803 | } |
804 | break; | |
805 | case AIOCB_READ_UDATA: | |
2fc8ae1d MT |
806 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
807 | aio_req->iov_offset, rsp.data_length); | |
80731d9d | 808 | if (ret != rsp.data_length) { |
6daf194d | 809 | error_report("failed to get the data, %s", strerror(errno)); |
011603ca | 810 | goto err; |
33b1db1c MK |
811 | } |
812 | break; | |
47783072 LY |
813 | case AIOCB_FLUSH_CACHE: |
814 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
2440a2c3 | 815 | DPRINTF("disable cache since the server doesn't support it\n"); |
47783072 LY |
816 | s->cache_flags = SD_FLAG_CMD_DIRECT; |
817 | rsp.result = SD_RES_SUCCESS; | |
818 | } | |
819 | break; | |
cac8f4a6 LY |
820 | case AIOCB_DISCARD_OBJ: |
821 | switch (rsp.result) { | |
822 | case SD_RES_INVALID_PARMS: | |
823 | error_report("sheep(%s) doesn't support discard command", | |
824 | s->host_spec); | |
825 | rsp.result = SD_RES_SUCCESS; | |
826 | s->discard_supported = false; | |
827 | break; | |
cac8f4a6 LY |
828 | default: |
829 | break; | |
830 | } | |
33b1db1c MK |
831 | } |
832 | ||
e80ab33d PB |
833 | /* No more data for this aio_req (reload_inode below uses its own file |
834 | * descriptor handler which doesn't use co_recv). | |
835 | */ | |
836 | s->co_recv = NULL; | |
837 | ||
c4080e93 | 838 | QLIST_REMOVE(aio_req, aio_siblings); |
13c31de2 MK |
839 | switch (rsp.result) { |
840 | case SD_RES_SUCCESS: | |
841 | break; | |
842 | case SD_RES_READONLY: | |
72e0996c MK |
843 | if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) { |
844 | ret = reload_inode(s, 0, ""); | |
845 | if (ret < 0) { | |
011603ca | 846 | goto err; |
72e0996c MK |
847 | } |
848 | } | |
72e0996c MK |
849 | if (is_data_obj(aio_req->oid)) { |
850 | aio_req->oid = vid_to_data_oid(s->inode.vdi_id, | |
851 | data_oid_to_idx(aio_req->oid)); | |
852 | } else { | |
853 | aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id); | |
854 | } | |
a37dcdf9 | 855 | resend_aioreq(s, aio_req); |
e80ab33d | 856 | return; |
13c31de2 | 857 | default: |
33b1db1c | 858 | acb->ret = -EIO; |
6daf194d | 859 | error_report("%s", sd_strerror(rsp.result)); |
13c31de2 | 860 | break; |
33b1db1c MK |
861 | } |
862 | ||
c4080e93 PB |
863 | g_free(aio_req); |
864 | ||
865 | if (!--acb->nr_pending) { | |
33b1db1c MK |
866 | /* |
867 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 868 | * we can switch back to sd_co_readv/writev now. |
33b1db1c | 869 | */ |
9d456654 | 870 | aio_co_wake(acb->coroutine); |
33b1db1c | 871 | } |
e80ab33d | 872 | |
011603ca | 873 | return; |
e80ab33d | 874 | |
011603ca | 875 | err: |
011603ca | 876 | reconnect_to_sdog(opaque); |
2df46246 MK |
877 | } |
878 | ||
879 | static void co_read_response(void *opaque) | |
880 | { | |
881 | BDRVSheepdogState *s = opaque; | |
882 | ||
883 | if (!s->co_recv) { | |
0b8b8753 | 884 | s->co_recv = qemu_coroutine_create(aio_read_response, opaque); |
2df46246 MK |
885 | } |
886 | ||
9d456654 | 887 | aio_co_wake(s->co_recv); |
2df46246 MK |
888 | } |
889 | ||
890 | static void co_write_request(void *opaque) | |
891 | { | |
892 | BDRVSheepdogState *s = opaque; | |
893 | ||
9d456654 | 894 | aio_co_wake(s->co_send); |
33b1db1c MK |
895 | } |
896 | ||
33b1db1c | 897 | /* |
dc6fb73d | 898 | * Return a socket descriptor to read/write objects. |
33b1db1c | 899 | * |
dc6fb73d | 900 | * We cannot use this descriptor for other operations because |
33b1db1c MK |
901 | * the block driver may be on waiting response from the server. |
902 | */ | |
356b4ca2 | 903 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp) |
33b1db1c | 904 | { |
1b8bbb46 | 905 | int fd; |
33b1db1c | 906 | |
356b4ca2 | 907 | fd = connect_to_sdog(s, errp); |
33b1db1c | 908 | if (fd < 0) { |
cb595887 | 909 | return fd; |
33b1db1c MK |
910 | } |
911 | ||
dca21ef2 | 912 | aio_set_fd_handler(s->aio_context, fd, false, |
f6a51c84 | 913 | co_read_response, NULL, NULL, s); |
33b1db1c MK |
914 | return fd; |
915 | } | |
916 | ||
5d6768e3 MK |
917 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
918 | char *vdi, uint32_t *snapid, char *tag) | |
919 | { | |
920 | URI *uri; | |
921 | QueryParams *qp = NULL; | |
922 | int ret = 0; | |
923 | ||
924 | uri = uri_parse(filename); | |
925 | if (!uri) { | |
926 | return -EINVAL; | |
927 | } | |
928 | ||
1b8bbb46 MK |
929 | /* transport */ |
930 | if (!strcmp(uri->scheme, "sheepdog")) { | |
931 | s->is_unix = false; | |
932 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
933 | s->is_unix = false; | |
934 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
935 | s->is_unix = true; | |
936 | } else { | |
937 | ret = -EINVAL; | |
938 | goto out; | |
939 | } | |
940 | ||
5d6768e3 MK |
941 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
942 | ret = -EINVAL; | |
943 | goto out; | |
944 | } | |
945 | pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1); | |
946 | ||
1b8bbb46 MK |
947 | qp = query_params_parse(uri->query); |
948 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
949 | ret = -EINVAL; | |
950 | goto out; | |
951 | } | |
952 | ||
953 | if (s->is_unix) { | |
954 | /* sheepdog+unix:///vdiname?socket=path */ | |
955 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
956 | ret = -EINVAL; | |
957 | goto out; | |
958 | } | |
959 | s->host_spec = g_strdup(qp->p[0].value); | |
960 | } else { | |
961 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
962 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
963 | uri->port ?: SD_DEFAULT_PORT); | |
964 | } | |
5d6768e3 MK |
965 | |
966 | /* snapshot tag */ | |
967 | if (uri->fragment) { | |
968 | *snapid = strtoul(uri->fragment, NULL, 10); | |
969 | if (*snapid == 0) { | |
970 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); | |
971 | } | |
972 | } else { | |
973 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
974 | } | |
975 | ||
976 | out: | |
977 | if (qp) { | |
978 | query_params_free(qp); | |
979 | } | |
980 | uri_free(uri); | |
981 | return ret; | |
982 | } | |
983 | ||
33b1db1c | 984 | /* |
5d6768e3 | 985 | * Parse a filename (old syntax) |
33b1db1c MK |
986 | * |
987 | * filename must be one of the following formats: | |
988 | * 1. [vdiname] | |
989 | * 2. [vdiname]:[snapid] | |
990 | * 3. [vdiname]:[tag] | |
991 | * 4. [hostname]:[port]:[vdiname] | |
992 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
993 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
994 | * | |
995 | * You can boot from the snapshot images by specifying `snapid` or | |
996 | * `tag'. | |
997 | * | |
998 | * You can run VMs outside the Sheepdog cluster by specifying | |
999 | * `hostname' and `port' (experimental). | |
1000 | */ | |
1001 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
1002 | char *vdi, uint32_t *snapid, char *tag) | |
1003 | { | |
5d6768e3 MK |
1004 | char *p, *q, *uri; |
1005 | const char *host_spec, *vdi_spec; | |
1006 | int nr_sep, ret; | |
33b1db1c | 1007 | |
11d816a5 | 1008 | strstart(filename, "sheepdog:", &filename); |
7267c094 | 1009 | p = q = g_strdup(filename); |
33b1db1c MK |
1010 | |
1011 | /* count the number of separators */ | |
1012 | nr_sep = 0; | |
1013 | while (*p) { | |
1014 | if (*p == ':') { | |
1015 | nr_sep++; | |
1016 | } | |
1017 | p++; | |
1018 | } | |
1019 | p = q; | |
1020 | ||
5d6768e3 | 1021 | /* use the first two tokens as host_spec. */ |
33b1db1c | 1022 | if (nr_sep >= 2) { |
5d6768e3 | 1023 | host_spec = p; |
33b1db1c | 1024 | p = strchr(p, ':'); |
5d6768e3 | 1025 | p++; |
33b1db1c MK |
1026 | p = strchr(p, ':'); |
1027 | *p++ = '\0'; | |
1028 | } else { | |
5d6768e3 | 1029 | host_spec = ""; |
33b1db1c MK |
1030 | } |
1031 | ||
5d6768e3 | 1032 | vdi_spec = p; |
33b1db1c | 1033 | |
5d6768e3 | 1034 | p = strchr(vdi_spec, ':'); |
33b1db1c | 1035 | if (p) { |
5d6768e3 | 1036 | *p++ = '#'; |
33b1db1c MK |
1037 | } |
1038 | ||
5d6768e3 | 1039 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 1040 | |
5d6768e3 MK |
1041 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
1042 | ||
1043 | g_free(q); | |
1044 | g_free(uri); | |
1045 | ||
1046 | return ret; | |
33b1db1c MK |
1047 | } |
1048 | ||
982dcbf4 MK |
1049 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, |
1050 | uint32_t snapid, const char *tag, uint32_t *vid, | |
dc83cd42 | 1051 | bool lock, Error **errp) |
33b1db1c MK |
1052 | { |
1053 | int ret, fd; | |
1054 | SheepdogVdiReq hdr; | |
1055 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1056 | unsigned int wlen, rlen = 0; | |
1057 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
1058 | ||
dc83cd42 | 1059 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1060 | if (fd < 0) { |
cb595887 | 1061 | return fd; |
33b1db1c MK |
1062 | } |
1063 | ||
3178e275 JM |
1064 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
1065 | * which is desirable since we'll soon be sending those bytes, and | |
1066 | * don't want the send_req to read uninitialized data. | |
1067 | */ | |
33b1db1c MK |
1068 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
1069 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
1070 | ||
1071 | memset(&hdr, 0, sizeof(hdr)); | |
982dcbf4 | 1072 | if (lock) { |
33b1db1c | 1073 | hdr.opcode = SD_OP_LOCK_VDI; |
1dbfafed | 1074 | hdr.type = LOCK_TYPE_NORMAL; |
982dcbf4 MK |
1075 | } else { |
1076 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
33b1db1c MK |
1077 | } |
1078 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
1079 | hdr.proto_ver = SD_PROTO_VER; | |
1080 | hdr.data_length = wlen; | |
1081 | hdr.snapid = snapid; | |
1082 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1083 | ||
f11672db | 1084 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1085 | if (ret) { |
dc83cd42 | 1086 | error_setg_errno(errp, -ret, "cannot get vdi info"); |
33b1db1c MK |
1087 | goto out; |
1088 | } | |
1089 | ||
1090 | if (rsp->result != SD_RES_SUCCESS) { | |
dc83cd42 MA |
1091 | error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s", |
1092 | sd_strerror(rsp->result), filename, snapid, tag); | |
cb595887 MK |
1093 | if (rsp->result == SD_RES_NO_VDI) { |
1094 | ret = -ENOENT; | |
38890b24 HM |
1095 | } else if (rsp->result == SD_RES_VDI_LOCKED) { |
1096 | ret = -EBUSY; | |
cb595887 MK |
1097 | } else { |
1098 | ret = -EIO; | |
1099 | } | |
33b1db1c MK |
1100 | goto out; |
1101 | } | |
1102 | *vid = rsp->vdi_id; | |
1103 | ||
1104 | ret = 0; | |
1105 | out: | |
1106 | closesocket(fd); | |
1107 | return ret; | |
1108 | } | |
1109 | ||
a37dcdf9 | 1110 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
1111 | struct iovec *iov, int niov, |
1112 | enum AIOCBState aiocb_type) | |
33b1db1c MK |
1113 | { |
1114 | int nr_copies = s->inode.nr_copies; | |
1115 | SheepdogObjReq hdr; | |
47783072 | 1116 | unsigned int wlen = 0; |
33b1db1c MK |
1117 | int ret; |
1118 | uint64_t oid = aio_req->oid; | |
1119 | unsigned int datalen = aio_req->data_len; | |
1120 | uint64_t offset = aio_req->offset; | |
1121 | uint8_t flags = aio_req->flags; | |
1122 | uint64_t old_oid = aio_req->base_oid; | |
b544c1ab | 1123 | bool create = aio_req->create; |
33b1db1c | 1124 | |
c4080e93 PB |
1125 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1126 | ||
33b1db1c | 1127 | if (!nr_copies) { |
6daf194d | 1128 | error_report("bug"); |
33b1db1c MK |
1129 | } |
1130 | ||
1131 | memset(&hdr, 0, sizeof(hdr)); | |
1132 | ||
47783072 LY |
1133 | switch (aiocb_type) { |
1134 | case AIOCB_FLUSH_CACHE: | |
1135 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1136 | break; | |
1137 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1138 | hdr.opcode = SD_OP_READ_OBJ; |
1139 | hdr.flags = flags; | |
47783072 LY |
1140 | break; |
1141 | case AIOCB_WRITE_UDATA: | |
1142 | if (create) { | |
1143 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1144 | } else { | |
1145 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1146 | } | |
33b1db1c | 1147 | wlen = datalen; |
33b1db1c | 1148 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1149 | break; |
cac8f4a6 | 1150 | case AIOCB_DISCARD_OBJ: |
e6fd57ea HM |
1151 | hdr.opcode = SD_OP_WRITE_OBJ; |
1152 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
1153 | s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0; | |
1154 | offset = offsetof(SheepdogInode, | |
1155 | data_vdi_id[data_oid_to_idx(oid)]); | |
1156 | oid = vid_to_vdi_oid(s->inode.vdi_id); | |
1157 | wlen = datalen = sizeof(uint32_t); | |
cac8f4a6 | 1158 | break; |
33b1db1c MK |
1159 | } |
1160 | ||
0e7106d8 LY |
1161 | if (s->cache_flags) { |
1162 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1163 | } |
1164 | ||
33b1db1c MK |
1165 | hdr.oid = oid; |
1166 | hdr.cow_oid = old_oid; | |
1167 | hdr.copies = s->inode.nr_copies; | |
1168 | ||
1169 | hdr.data_length = datalen; | |
1170 | hdr.offset = offset; | |
1171 | ||
1172 | hdr.id = aio_req->id; | |
1173 | ||
2df46246 MK |
1174 | qemu_co_mutex_lock(&s->lock); |
1175 | s->co_send = qemu_coroutine_self(); | |
dca21ef2 | 1176 | aio_set_fd_handler(s->aio_context, s->fd, false, |
f6a51c84 | 1177 | co_read_response, co_write_request, NULL, s); |
128aa589 | 1178 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1179 | |
1180 | /* send a header */ | |
8c5135f9 | 1181 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
80731d9d | 1182 | if (ret != sizeof(hdr)) { |
6daf194d | 1183 | error_report("failed to send a req, %s", strerror(errno)); |
011603ca | 1184 | goto out; |
33b1db1c MK |
1185 | } |
1186 | ||
1187 | if (wlen) { | |
2fc8ae1d | 1188 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
80731d9d | 1189 | if (ret != wlen) { |
6daf194d | 1190 | error_report("failed to send a data, %s", strerror(errno)); |
33b1db1c MK |
1191 | } |
1192 | } | |
011603ca | 1193 | out: |
128aa589 | 1194 | socket_set_cork(s->fd, 0); |
dca21ef2 | 1195 | aio_set_fd_handler(s->aio_context, s->fd, false, |
f6a51c84 | 1196 | co_read_response, NULL, NULL, s); |
011603ca | 1197 | s->co_send = NULL; |
2df46246 | 1198 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1199 | } |
1200 | ||
f11672db | 1201 | static int read_write_object(int fd, BlockDriverState *bs, char *buf, |
84390bed | 1202 | uint64_t oid, uint8_t copies, |
33b1db1c | 1203 | unsigned int datalen, uint64_t offset, |
0e7106d8 | 1204 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1205 | { |
1206 | SheepdogObjReq hdr; | |
1207 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1208 | unsigned int wlen, rlen; | |
1209 | int ret; | |
1210 | ||
1211 | memset(&hdr, 0, sizeof(hdr)); | |
1212 | ||
1213 | if (write) { | |
1214 | wlen = datalen; | |
1215 | rlen = 0; | |
1216 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1217 | if (create) { | |
1218 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1219 | } else { | |
1220 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1221 | } | |
1222 | } else { | |
1223 | wlen = 0; | |
1224 | rlen = datalen; | |
1225 | hdr.opcode = SD_OP_READ_OBJ; | |
1226 | } | |
47622c44 | 1227 | |
0e7106d8 | 1228 | hdr.flags |= cache_flags; |
47622c44 | 1229 | |
33b1db1c MK |
1230 | hdr.oid = oid; |
1231 | hdr.data_length = datalen; | |
1232 | hdr.offset = offset; | |
1233 | hdr.copies = copies; | |
1234 | ||
f11672db | 1235 | ret = do_req(fd, bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1236 | if (ret) { |
6daf194d | 1237 | error_report("failed to send a request to the sheep"); |
cb595887 | 1238 | return ret; |
33b1db1c MK |
1239 | } |
1240 | ||
1241 | switch (rsp->result) { | |
1242 | case SD_RES_SUCCESS: | |
1243 | return 0; | |
1244 | default: | |
6daf194d | 1245 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1246 | return -EIO; |
33b1db1c MK |
1247 | } |
1248 | } | |
1249 | ||
f11672db | 1250 | static int read_object(int fd, BlockDriverState *bs, char *buf, |
84390bed | 1251 | uint64_t oid, uint8_t copies, |
0e7106d8 LY |
1252 | unsigned int datalen, uint64_t offset, |
1253 | uint32_t cache_flags) | |
33b1db1c | 1254 | { |
f11672db | 1255 | return read_write_object(fd, bs, buf, oid, copies, |
84390bed | 1256 | datalen, offset, false, |
0e7106d8 | 1257 | false, cache_flags); |
33b1db1c MK |
1258 | } |
1259 | ||
f11672db | 1260 | static int write_object(int fd, BlockDriverState *bs, char *buf, |
84390bed | 1261 | uint64_t oid, uint8_t copies, |
2f536801 | 1262 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1263 | uint32_t cache_flags) |
33b1db1c | 1264 | { |
f11672db | 1265 | return read_write_object(fd, bs, buf, oid, copies, |
84390bed | 1266 | datalen, offset, true, |
0e7106d8 | 1267 | create, cache_flags); |
33b1db1c MK |
1268 | } |
1269 | ||
9ff53a0e MK |
1270 | /* update inode with the latest state */ |
1271 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag) | |
1272 | { | |
dfb12bf8 | 1273 | Error *local_err = NULL; |
9ff53a0e MK |
1274 | SheepdogInode *inode; |
1275 | int ret = 0, fd; | |
1276 | uint32_t vid = 0; | |
1277 | ||
dfb12bf8 | 1278 | fd = connect_to_sdog(s, &local_err); |
9ff53a0e | 1279 | if (fd < 0) { |
565f65d2 | 1280 | error_report_err(local_err); |
9ff53a0e MK |
1281 | return -EIO; |
1282 | } | |
1283 | ||
5d039bab | 1284 | inode = g_malloc(SD_INODE_HEADER_SIZE); |
9ff53a0e | 1285 | |
dc83cd42 | 1286 | ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err); |
9ff53a0e | 1287 | if (ret) { |
565f65d2 | 1288 | error_report_err(local_err); |
9ff53a0e MK |
1289 | goto out; |
1290 | } | |
1291 | ||
f11672db | 1292 | ret = read_object(fd, s->bs, (char *)inode, vid_to_vdi_oid(vid), |
5d039bab HM |
1293 | s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0, |
1294 | s->cache_flags); | |
9ff53a0e MK |
1295 | if (ret < 0) { |
1296 | goto out; | |
1297 | } | |
1298 | ||
1299 | if (inode->vdi_id != s->inode.vdi_id) { | |
5d039bab | 1300 | memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE); |
9ff53a0e MK |
1301 | } |
1302 | ||
1303 | out: | |
1304 | g_free(inode); | |
1305 | closesocket(fd); | |
1306 | ||
1307 | return ret; | |
1308 | } | |
1309 | ||
a37dcdf9 | 1310 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req) |
13c31de2 MK |
1311 | { |
1312 | SheepdogAIOCB *acb = aio_req->aiocb; | |
b544c1ab HM |
1313 | |
1314 | aio_req->create = false; | |
13c31de2 MK |
1315 | |
1316 | /* check whether this request becomes a CoW one */ | |
2412aec7 | 1317 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) { |
13c31de2 | 1318 | int idx = data_oid_to_idx(aio_req->oid); |
13c31de2 | 1319 | |
13c31de2 MK |
1320 | if (is_data_obj_writable(&s->inode, idx)) { |
1321 | goto out; | |
1322 | } | |
1323 | ||
80308d33 MK |
1324 | if (s->inode.data_vdi_id[idx]) { |
1325 | aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx); | |
1326 | aio_req->flags |= SD_FLAG_CMD_COW; | |
1327 | } | |
b544c1ab | 1328 | aio_req->create = true; |
13c31de2 MK |
1329 | } |
1330 | out: | |
2412aec7 | 1331 | if (is_data_obj(aio_req->oid)) { |
b544c1ab | 1332 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 1333 | acb->aiocb_type); |
2412aec7 MK |
1334 | } else { |
1335 | struct iovec iov; | |
1336 | iov.iov_base = &s->inode; | |
1337 | iov.iov_len = sizeof(s->inode); | |
b544c1ab | 1338 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
2412aec7 | 1339 | } |
13c31de2 MK |
1340 | } |
1341 | ||
84390bed SH |
1342 | static void sd_detach_aio_context(BlockDriverState *bs) |
1343 | { | |
1344 | BDRVSheepdogState *s = bs->opaque; | |
1345 | ||
dca21ef2 | 1346 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, |
f6a51c84 | 1347 | NULL, NULL, NULL); |
84390bed SH |
1348 | } |
1349 | ||
1350 | static void sd_attach_aio_context(BlockDriverState *bs, | |
1351 | AioContext *new_context) | |
1352 | { | |
1353 | BDRVSheepdogState *s = bs->opaque; | |
1354 | ||
1355 | s->aio_context = new_context; | |
dca21ef2 | 1356 | aio_set_fd_handler(new_context, s->fd, false, |
f6a51c84 | 1357 | co_read_response, NULL, NULL, s); |
84390bed SH |
1358 | } |
1359 | ||
c8c96350 KW |
1360 | /* TODO Convert to fine grained options */ |
1361 | static QemuOptsList runtime_opts = { | |
1362 | .name = "sheepdog", | |
1363 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1364 | .desc = { | |
1365 | { | |
1366 | .name = "filename", | |
1367 | .type = QEMU_OPT_STRING, | |
1368 | .help = "URL to the sheepdog image", | |
1369 | }, | |
1370 | { /* end of list */ } | |
1371 | }, | |
1372 | }; | |
1373 | ||
015a1036 HR |
1374 | static int sd_open(BlockDriverState *bs, QDict *options, int flags, |
1375 | Error **errp) | |
33b1db1c MK |
1376 | { |
1377 | int ret, fd; | |
1378 | uint32_t vid = 0; | |
1379 | BDRVSheepdogState *s = bs->opaque; | |
1380 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1381 | uint32_t snapid; | |
1382 | char *buf = NULL; | |
c8c96350 KW |
1383 | QemuOpts *opts; |
1384 | Error *local_err = NULL; | |
1385 | const char *filename; | |
1386 | ||
011603ca | 1387 | s->bs = bs; |
84390bed | 1388 | s->aio_context = bdrv_get_aio_context(bs); |
011603ca | 1389 | |
87ea75d5 | 1390 | opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); |
c8c96350 | 1391 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 1392 | if (local_err) { |
e67c3993 | 1393 | error_propagate(errp, local_err); |
c8c96350 KW |
1394 | ret = -EINVAL; |
1395 | goto out; | |
1396 | } | |
1397 | ||
1398 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1399 | |
c292ee6a | 1400 | QLIST_INIT(&s->inflight_aio_head); |
011603ca | 1401 | QLIST_INIT(&s->failed_aio_head); |
6a55c82c | 1402 | QLIST_INIT(&s->inflight_aiocb_head); |
33b1db1c MK |
1403 | s->fd = -1; |
1404 | ||
1405 | memset(vdi, 0, sizeof(vdi)); | |
1406 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1407 | |
1408 | if (strstr(filename, "://")) { | |
1409 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1410 | } else { | |
1411 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1412 | } | |
1413 | if (ret < 0) { | |
efde4b62 | 1414 | error_setg(errp, "Can't parse filename"); |
33b1db1c MK |
1415 | goto out; |
1416 | } | |
e67c3993 | 1417 | s->fd = get_sheep_fd(s, errp); |
33b1db1c | 1418 | if (s->fd < 0) { |
cb595887 | 1419 | ret = s->fd; |
33b1db1c MK |
1420 | goto out; |
1421 | } | |
1422 | ||
e67c3993 | 1423 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, true, errp); |
33b1db1c MK |
1424 | if (ret) { |
1425 | goto out; | |
1426 | } | |
1427 | ||
0e7106d8 LY |
1428 | /* |
1429 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1430 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1431 | */ | |
1432 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1433 | if (flags & BDRV_O_NOCACHE) { | |
1434 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1435 | } | |
cac8f4a6 | 1436 | s->discard_supported = true; |
0e7106d8 | 1437 | |
622b6057 | 1438 | if (snapid || tag[0] != '\0') { |
2440a2c3 | 1439 | DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1440 | s->is_snapshot = true; |
33b1db1c MK |
1441 | } |
1442 | ||
e67c3993 | 1443 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1444 | if (fd < 0) { |
cb595887 | 1445 | ret = fd; |
33b1db1c MK |
1446 | goto out; |
1447 | } | |
1448 | ||
7267c094 | 1449 | buf = g_malloc(SD_INODE_SIZE); |
f11672db | 1450 | ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid), |
84390bed | 1451 | 0, SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1452 | |
1453 | closesocket(fd); | |
1454 | ||
1455 | if (ret) { | |
efde4b62 | 1456 | error_setg(errp, "Can't read snapshot inode"); |
33b1db1c MK |
1457 | goto out; |
1458 | } | |
1459 | ||
1460 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
33b1db1c | 1461 | |
e8bfaa2f | 1462 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; |
3178e275 | 1463 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1464 | qemu_co_mutex_init(&s->lock); |
498f2140 | 1465 | qemu_co_queue_init(&s->overlapping_queue); |
c8c96350 | 1466 | qemu_opts_del(opts); |
7267c094 | 1467 | g_free(buf); |
33b1db1c MK |
1468 | return 0; |
1469 | out: | |
dca21ef2 | 1470 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, |
f6a51c84 | 1471 | false, NULL, NULL, NULL, NULL); |
33b1db1c MK |
1472 | if (s->fd >= 0) { |
1473 | closesocket(s->fd); | |
1474 | } | |
c8c96350 | 1475 | qemu_opts_del(opts); |
7267c094 | 1476 | g_free(buf); |
cb595887 | 1477 | return ret; |
33b1db1c MK |
1478 | } |
1479 | ||
4da65c80 LY |
1480 | static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue, |
1481 | Error **errp) | |
1482 | { | |
1483 | BDRVSheepdogState *s = state->bs->opaque; | |
1484 | BDRVSheepdogReopenState *re_s; | |
1485 | int ret = 0; | |
1486 | ||
1487 | re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1); | |
1488 | ||
1489 | re_s->cache_flags = SD_FLAG_CMD_CACHE; | |
1490 | if (state->flags & BDRV_O_NOCACHE) { | |
1491 | re_s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1492 | } | |
1493 | ||
1494 | re_s->fd = get_sheep_fd(s, errp); | |
1495 | if (re_s->fd < 0) { | |
1496 | ret = re_s->fd; | |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | return ret; | |
1501 | } | |
1502 | ||
1503 | static void sd_reopen_commit(BDRVReopenState *state) | |
1504 | { | |
1505 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1506 | BDRVSheepdogState *s = state->bs->opaque; | |
1507 | ||
1508 | if (s->fd) { | |
dca21ef2 | 1509 | aio_set_fd_handler(s->aio_context, s->fd, false, |
f6a51c84 | 1510 | NULL, NULL, NULL, NULL); |
4da65c80 LY |
1511 | closesocket(s->fd); |
1512 | } | |
1513 | ||
1514 | s->fd = re_s->fd; | |
1515 | s->cache_flags = re_s->cache_flags; | |
1516 | ||
1517 | g_free(state->opaque); | |
1518 | state->opaque = NULL; | |
1519 | ||
1520 | return; | |
1521 | } | |
1522 | ||
1523 | static void sd_reopen_abort(BDRVReopenState *state) | |
1524 | { | |
1525 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1526 | BDRVSheepdogState *s = state->bs->opaque; | |
1527 | ||
1528 | if (re_s == NULL) { | |
1529 | return; | |
1530 | } | |
1531 | ||
1532 | if (re_s->fd) { | |
dca21ef2 | 1533 | aio_set_fd_handler(s->aio_context, re_s->fd, false, |
f6a51c84 | 1534 | NULL, NULL, NULL, NULL); |
4da65c80 LY |
1535 | closesocket(re_s->fd); |
1536 | } | |
1537 | ||
1538 | g_free(state->opaque); | |
1539 | state->opaque = NULL; | |
1540 | ||
1541 | return; | |
1542 | } | |
1543 | ||
7d2d3e74 MA |
1544 | static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot, |
1545 | Error **errp) | |
33b1db1c MK |
1546 | { |
1547 | SheepdogVdiReq hdr; | |
1548 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1549 | int fd, ret; | |
1550 | unsigned int wlen, rlen = 0; | |
1551 | char buf[SD_MAX_VDI_LEN]; | |
1552 | ||
7d2d3e74 | 1553 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1554 | if (fd < 0) { |
cb595887 | 1555 | return fd; |
33b1db1c MK |
1556 | } |
1557 | ||
3178e275 JM |
1558 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1559 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1560 | */ | |
33b1db1c | 1561 | memset(buf, 0, sizeof(buf)); |
c31d482f | 1562 | pstrcpy(buf, sizeof(buf), s->name); |
33b1db1c MK |
1563 | |
1564 | memset(&hdr, 0, sizeof(hdr)); | |
1565 | hdr.opcode = SD_OP_NEW_VDI; | |
9f23fce7 | 1566 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1567 | |
1568 | wlen = SD_MAX_VDI_LEN; | |
1569 | ||
1570 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1571 | hdr.snapid = snapshot; | |
1572 | ||
1573 | hdr.data_length = wlen; | |
c31d482f LY |
1574 | hdr.vdi_size = s->inode.vdi_size; |
1575 | hdr.copy_policy = s->inode.copy_policy; | |
b3af018f | 1576 | hdr.copies = s->inode.nr_copies; |
876eb1b0 | 1577 | hdr.block_size_shift = s->inode.block_size_shift; |
33b1db1c | 1578 | |
f11672db | 1579 | ret = do_req(fd, NULL, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c MK |
1580 | |
1581 | closesocket(fd); | |
1582 | ||
1583 | if (ret) { | |
7d2d3e74 | 1584 | error_setg_errno(errp, -ret, "create failed"); |
cb595887 | 1585 | return ret; |
33b1db1c MK |
1586 | } |
1587 | ||
1588 | if (rsp->result != SD_RES_SUCCESS) { | |
7d2d3e74 | 1589 | error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name); |
33b1db1c MK |
1590 | return -EIO; |
1591 | } | |
1592 | ||
1593 | if (vdi_id) { | |
1594 | *vdi_id = rsp->vdi_id; | |
1595 | } | |
1596 | ||
1597 | return 0; | |
1598 | } | |
1599 | ||
318df29e | 1600 | static int sd_prealloc(const char *filename, Error **errp) |
a8e0fdd7 | 1601 | { |
fba98d45 | 1602 | BlockBackend *blk = NULL; |
876eb1b0 TI |
1603 | BDRVSheepdogState *base = NULL; |
1604 | unsigned long buf_size; | |
a8e0fdd7 | 1605 | uint32_t idx, max_idx; |
876eb1b0 | 1606 | uint32_t object_size; |
a8e0fdd7 | 1607 | int64_t vdi_size; |
876eb1b0 | 1608 | void *buf = NULL; |
a8e0fdd7 MK |
1609 | int ret; |
1610 | ||
efaa7c4e | 1611 | blk = blk_new_open(filename, NULL, NULL, |
72e775c7 | 1612 | BDRV_O_RDWR | BDRV_O_PROTOCOL, errp); |
fba98d45 KW |
1613 | if (blk == NULL) { |
1614 | ret = -EIO; | |
318df29e | 1615 | goto out_with_err_set; |
a8e0fdd7 MK |
1616 | } |
1617 | ||
fba98d45 KW |
1618 | blk_set_allow_write_beyond_eof(blk, true); |
1619 | ||
1620 | vdi_size = blk_getlength(blk); | |
a8e0fdd7 MK |
1621 | if (vdi_size < 0) { |
1622 | ret = vdi_size; | |
1623 | goto out; | |
1624 | } | |
876eb1b0 | 1625 | |
fba98d45 | 1626 | base = blk_bs(blk)->opaque; |
876eb1b0 TI |
1627 | object_size = (UINT32_C(1) << base->inode.block_size_shift); |
1628 | buf_size = MIN(object_size, SD_DATA_OBJ_SIZE); | |
1629 | buf = g_malloc0(buf_size); | |
1630 | ||
1631 | max_idx = DIV_ROUND_UP(vdi_size, buf_size); | |
a8e0fdd7 MK |
1632 | |
1633 | for (idx = 0; idx < max_idx; idx++) { | |
1634 | /* | |
1635 | * The created image can be a cloned image, so we need to read | |
1636 | * a data from the source image. | |
1637 | */ | |
fba98d45 | 1638 | ret = blk_pread(blk, idx * buf_size, buf, buf_size); |
a8e0fdd7 MK |
1639 | if (ret < 0) { |
1640 | goto out; | |
1641 | } | |
8341f00d | 1642 | ret = blk_pwrite(blk, idx * buf_size, buf, buf_size, 0); |
a8e0fdd7 MK |
1643 | if (ret < 0) { |
1644 | goto out; | |
1645 | } | |
1646 | } | |
318df29e | 1647 | |
fba98d45 | 1648 | ret = 0; |
a8e0fdd7 | 1649 | out: |
318df29e MA |
1650 | if (ret < 0) { |
1651 | error_setg_errno(errp, -ret, "Can't pre-allocate"); | |
1652 | } | |
1653 | out_with_err_set: | |
fba98d45 KW |
1654 | if (blk) { |
1655 | blk_unref(blk); | |
a8e0fdd7 | 1656 | } |
7267c094 | 1657 | g_free(buf); |
a8e0fdd7 MK |
1658 | |
1659 | return ret; | |
1660 | } | |
1661 | ||
b3af018f LY |
1662 | /* |
1663 | * Sheepdog support two kinds of redundancy, full replication and erasure | |
1664 | * coding. | |
1665 | * | |
1666 | * # create a fully replicated vdi with x copies | |
1667 | * -o redundancy=x (1 <= x <= SD_MAX_COPIES) | |
1668 | * | |
1669 | * # create a erasure coded vdi with x data strips and y parity strips | |
1670 | * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP) | |
1671 | */ | |
1672 | static int parse_redundancy(BDRVSheepdogState *s, const char *opt) | |
1673 | { | |
1674 | struct SheepdogInode *inode = &s->inode; | |
1675 | const char *n1, *n2; | |
1676 | long copy, parity; | |
1677 | char p[10]; | |
1678 | ||
1679 | pstrcpy(p, sizeof(p), opt); | |
1680 | n1 = strtok(p, ":"); | |
1681 | n2 = strtok(NULL, ":"); | |
1682 | ||
1683 | if (!n1) { | |
1684 | return -EINVAL; | |
1685 | } | |
1686 | ||
1687 | copy = strtol(n1, NULL, 10); | |
1688 | if (copy > SD_MAX_COPIES || copy < 1) { | |
1689 | return -EINVAL; | |
1690 | } | |
1691 | if (!n2) { | |
1692 | inode->copy_policy = 0; | |
1693 | inode->nr_copies = copy; | |
1694 | return 0; | |
1695 | } | |
1696 | ||
1697 | if (copy != 2 && copy != 4 && copy != 8 && copy != 16) { | |
1698 | return -EINVAL; | |
1699 | } | |
1700 | ||
1701 | parity = strtol(n2, NULL, 10); | |
1702 | if (parity >= SD_EC_MAX_STRIP || parity < 1) { | |
1703 | return -EINVAL; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * 4 bits for parity and 4 bits for data. | |
1708 | * We have to compress upper data bits because it can't represent 16 | |
1709 | */ | |
1710 | inode->copy_policy = ((copy / 2) << 4) + parity; | |
1711 | inode->nr_copies = copy + parity; | |
1712 | ||
1713 | return 0; | |
1714 | } | |
1715 | ||
876eb1b0 TI |
1716 | static int parse_block_size_shift(BDRVSheepdogState *s, QemuOpts *opt) |
1717 | { | |
1718 | struct SheepdogInode *inode = &s->inode; | |
1719 | uint64_t object_size; | |
1720 | int obj_order; | |
1721 | ||
1722 | object_size = qemu_opt_get_size_del(opt, BLOCK_OPT_OBJECT_SIZE, 0); | |
1723 | if (object_size) { | |
1724 | if ((object_size - 1) & object_size) { /* not a power of 2? */ | |
1725 | return -EINVAL; | |
1726 | } | |
786a4ea8 | 1727 | obj_order = ctz32(object_size); |
876eb1b0 TI |
1728 | if (obj_order < 20 || obj_order > 31) { |
1729 | return -EINVAL; | |
1730 | } | |
1731 | inode->block_size_shift = (uint8_t)obj_order; | |
1732 | } | |
1733 | ||
1734 | return 0; | |
1735 | } | |
1736 | ||
b222237b | 1737 | static int sd_create(const char *filename, QemuOpts *opts, |
d5124c00 | 1738 | Error **errp) |
33b1db1c | 1739 | { |
b6fc8245 | 1740 | int ret = 0; |
c31d482f | 1741 | uint32_t vid = 0; |
33b1db1c | 1742 | char *backing_file = NULL; |
b222237b | 1743 | char *buf = NULL; |
b6fc8245 | 1744 | BDRVSheepdogState *s; |
c31d482f | 1745 | char tag[SD_MAX_VDI_TAG_LEN]; |
b4447363 | 1746 | uint32_t snapid; |
876eb1b0 | 1747 | uint64_t max_vdi_size; |
2f536801 | 1748 | bool prealloc = false; |
33b1db1c | 1749 | |
5839e53b | 1750 | s = g_new0(BDRVSheepdogState, 1); |
b6fc8245 | 1751 | |
b4447363 | 1752 | memset(tag, 0, sizeof(tag)); |
5d6768e3 | 1753 | if (strstr(filename, "://")) { |
c31d482f | 1754 | ret = sd_parse_uri(s, filename, s->name, &snapid, tag); |
5d6768e3 | 1755 | } else { |
c31d482f | 1756 | ret = parse_vdiname(s, filename, s->name, &snapid, tag); |
5d6768e3 MK |
1757 | } |
1758 | if (ret < 0) { | |
efde4b62 | 1759 | error_setg(errp, "Can't parse filename"); |
b6fc8245 | 1760 | goto out; |
b4447363 MK |
1761 | } |
1762 | ||
c2eb918e HT |
1763 | s->inode.vdi_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
1764 | BDRV_SECTOR_SIZE); | |
b222237b CL |
1765 | backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); |
1766 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
1767 | if (!buf || !strcmp(buf, "off")) { | |
1768 | prealloc = false; | |
1769 | } else if (!strcmp(buf, "full")) { | |
1770 | prealloc = true; | |
1771 | } else { | |
1772 | error_setg(errp, "Invalid preallocation mode: '%s'", buf); | |
1773 | ret = -EINVAL; | |
1774 | goto out; | |
1775 | } | |
1776 | ||
1777 | g_free(buf); | |
1778 | buf = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY); | |
1779 | if (buf) { | |
1780 | ret = parse_redundancy(s, buf); | |
1781 | if (ret < 0) { | |
1782 | error_setg(errp, "Invalid redundancy mode: '%s'", buf); | |
1783 | goto out; | |
33b1db1c | 1784 | } |
33b1db1c | 1785 | } |
876eb1b0 TI |
1786 | ret = parse_block_size_shift(s, opts); |
1787 | if (ret < 0) { | |
1788 | error_setg(errp, "Invalid object_size." | |
1789 | " obect_size needs to be power of 2" | |
1790 | " and be limited from 2^20 to 2^31"); | |
b6fc8245 | 1791 | goto out; |
33b1db1c MK |
1792 | } |
1793 | ||
1794 | if (backing_file) { | |
fba98d45 | 1795 | BlockBackend *blk; |
9f23fce7 | 1796 | BDRVSheepdogState *base; |
33b1db1c MK |
1797 | BlockDriver *drv; |
1798 | ||
1799 | /* Currently, only Sheepdog backing image is supported. */ | |
b65a5e12 | 1800 | drv = bdrv_find_protocol(backing_file, true, NULL); |
33b1db1c | 1801 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
e67c3993 | 1802 | error_setg(errp, "backing_file must be a sheepdog image"); |
b6fc8245 MK |
1803 | ret = -EINVAL; |
1804 | goto out; | |
33b1db1c MK |
1805 | } |
1806 | ||
efaa7c4e | 1807 | blk = blk_new_open(backing_file, NULL, NULL, |
72e775c7 | 1808 | BDRV_O_PROTOCOL, errp); |
fba98d45 KW |
1809 | if (blk == NULL) { |
1810 | ret = -EIO; | |
b6fc8245 | 1811 | goto out; |
cb595887 | 1812 | } |
33b1db1c | 1813 | |
fba98d45 | 1814 | base = blk_bs(blk)->opaque; |
33b1db1c | 1815 | |
9f23fce7 | 1816 | if (!is_snapshot(&base->inode)) { |
e67c3993 | 1817 | error_setg(errp, "cannot clone from a non snapshot vdi"); |
fba98d45 | 1818 | blk_unref(blk); |
b6fc8245 MK |
1819 | ret = -EINVAL; |
1820 | goto out; | |
33b1db1c | 1821 | } |
9f23fce7 | 1822 | s->inode.vdi_id = base->inode.vdi_id; |
fba98d45 | 1823 | blk_unref(blk); |
33b1db1c MK |
1824 | } |
1825 | ||
5d5da114 | 1826 | s->aio_context = qemu_get_aio_context(); |
876eb1b0 TI |
1827 | |
1828 | /* if block_size_shift is not specified, get cluster default value */ | |
1829 | if (s->inode.block_size_shift == 0) { | |
1830 | SheepdogVdiReq hdr; | |
1831 | SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr; | |
1832 | Error *local_err = NULL; | |
1833 | int fd; | |
1834 | unsigned int wlen = 0, rlen = 0; | |
1835 | ||
1836 | fd = connect_to_sdog(s, &local_err); | |
1837 | if (fd < 0) { | |
4fffeb5e | 1838 | error_report_err(local_err); |
876eb1b0 TI |
1839 | ret = -EIO; |
1840 | goto out; | |
1841 | } | |
1842 | ||
1843 | memset(&hdr, 0, sizeof(hdr)); | |
1844 | hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT; | |
1845 | hdr.proto_ver = SD_PROTO_VER; | |
1846 | ||
f11672db | 1847 | ret = do_req(fd, NULL, (SheepdogReq *)&hdr, |
876eb1b0 TI |
1848 | NULL, &wlen, &rlen); |
1849 | closesocket(fd); | |
1850 | if (ret) { | |
1851 | error_setg_errno(errp, -ret, "failed to get cluster default"); | |
1852 | goto out; | |
1853 | } | |
1854 | if (rsp->result == SD_RES_SUCCESS) { | |
1855 | s->inode.block_size_shift = rsp->block_size_shift; | |
1856 | } else { | |
1857 | s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT; | |
1858 | } | |
1859 | } | |
1860 | ||
1861 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; | |
1862 | ||
1863 | if (s->inode.vdi_size > max_vdi_size) { | |
1864 | error_setg(errp, "An image is too large." | |
1865 | " The maximum image size is %"PRIu64 "GB", | |
1866 | max_vdi_size / 1024 / 1024 / 1024); | |
1867 | ret = -EINVAL; | |
1868 | goto out; | |
1869 | } | |
1870 | ||
e67c3993 | 1871 | ret = do_sd_create(s, &vid, 0, errp); |
7d2d3e74 | 1872 | if (ret) { |
b6fc8245 | 1873 | goto out; |
a8e0fdd7 MK |
1874 | } |
1875 | ||
7d2d3e74 | 1876 | if (prealloc) { |
e67c3993 | 1877 | ret = sd_prealloc(filename, errp); |
318df29e | 1878 | } |
b6fc8245 | 1879 | out: |
b222237b CL |
1880 | g_free(backing_file); |
1881 | g_free(buf); | |
b6fc8245 MK |
1882 | g_free(s); |
1883 | return ret; | |
33b1db1c MK |
1884 | } |
1885 | ||
1886 | static void sd_close(BlockDriverState *bs) | |
1887 | { | |
dfb12bf8 | 1888 | Error *local_err = NULL; |
33b1db1c MK |
1889 | BDRVSheepdogState *s = bs->opaque; |
1890 | SheepdogVdiReq hdr; | |
1891 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1892 | unsigned int wlen, rlen = 0; | |
1893 | int fd, ret; | |
1894 | ||
2440a2c3 | 1895 | DPRINTF("%s\n", s->name); |
33b1db1c | 1896 | |
dfb12bf8 | 1897 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1898 | if (fd < 0) { |
565f65d2 | 1899 | error_report_err(local_err); |
33b1db1c MK |
1900 | return; |
1901 | } | |
1902 | ||
1903 | memset(&hdr, 0, sizeof(hdr)); | |
1904 | ||
1905 | hdr.opcode = SD_OP_RELEASE_VDI; | |
1dbfafed | 1906 | hdr.type = LOCK_TYPE_NORMAL; |
9f23fce7 | 1907 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1908 | wlen = strlen(s->name) + 1; |
1909 | hdr.data_length = wlen; | |
1910 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1911 | ||
f11672db | 1912 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, |
84390bed | 1913 | s->name, &wlen, &rlen); |
33b1db1c MK |
1914 | |
1915 | closesocket(fd); | |
1916 | ||
1917 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1918 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1919 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1920 | } |
1921 | ||
dca21ef2 | 1922 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, |
f6a51c84 | 1923 | false, NULL, NULL, NULL, NULL); |
33b1db1c | 1924 | closesocket(s->fd); |
25af257d | 1925 | g_free(s->host_spec); |
33b1db1c MK |
1926 | } |
1927 | ||
1928 | static int64_t sd_getlength(BlockDriverState *bs) | |
1929 | { | |
1930 | BDRVSheepdogState *s = bs->opaque; | |
1931 | ||
1932 | return s->inode.vdi_size; | |
1933 | } | |
1934 | ||
1935 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1936 | { | |
dfb12bf8 | 1937 | Error *local_err = NULL; |
33b1db1c MK |
1938 | BDRVSheepdogState *s = bs->opaque; |
1939 | int ret, fd; | |
1940 | unsigned int datalen; | |
876eb1b0 | 1941 | uint64_t max_vdi_size; |
33b1db1c | 1942 | |
876eb1b0 | 1943 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; |
33b1db1c | 1944 | if (offset < s->inode.vdi_size) { |
6daf194d | 1945 | error_report("shrinking is not supported"); |
33b1db1c | 1946 | return -EINVAL; |
876eb1b0 | 1947 | } else if (offset > max_vdi_size) { |
6daf194d | 1948 | error_report("too big image size"); |
33b1db1c MK |
1949 | return -EINVAL; |
1950 | } | |
1951 | ||
dfb12bf8 | 1952 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1953 | if (fd < 0) { |
565f65d2 | 1954 | error_report_err(local_err); |
cb595887 | 1955 | return fd; |
33b1db1c MK |
1956 | } |
1957 | ||
1958 | /* we don't need to update entire object */ | |
1959 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1960 | s->inode.vdi_size = offset; | |
f11672db | 1961 | ret = write_object(fd, s->bs, (char *)&s->inode, |
84390bed SH |
1962 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, |
1963 | datalen, 0, false, s->cache_flags); | |
33b1db1c MK |
1964 | close(fd); |
1965 | ||
1966 | if (ret < 0) { | |
6daf194d | 1967 | error_report("failed to update an inode."); |
33b1db1c MK |
1968 | } |
1969 | ||
cb595887 | 1970 | return ret; |
33b1db1c MK |
1971 | } |
1972 | ||
1973 | /* | |
1974 | * This function is called after writing data objects. If we need to | |
1975 | * update metadata, this sends a write request to the vdi object. | |
33b1db1c | 1976 | */ |
d8716b41 | 1977 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c | 1978 | { |
28ddd08c | 1979 | BDRVSheepdogState *s = acb->s; |
33b1db1c MK |
1980 | struct iovec iov; |
1981 | AIOReq *aio_req; | |
1982 | uint32_t offset, data_len, mn, mx; | |
1983 | ||
498f2140 HM |
1984 | mn = acb->min_dirty_data_idx; |
1985 | mx = acb->max_dirty_data_idx; | |
33b1db1c MK |
1986 | if (mn <= mx) { |
1987 | /* we need to update the vdi object. */ | |
e80ab33d | 1988 | ++acb->nr_pending; |
33b1db1c MK |
1989 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + |
1990 | mn * sizeof(s->inode.data_vdi_id[0]); | |
1991 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
1992 | ||
498f2140 HM |
1993 | acb->min_dirty_data_idx = UINT32_MAX; |
1994 | acb->max_dirty_data_idx = 0; | |
33b1db1c MK |
1995 | |
1996 | iov.iov_base = &s->inode; | |
1997 | iov.iov_len = sizeof(s->inode); | |
1998 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
b544c1ab | 1999 | data_len, offset, 0, false, 0, offset); |
b544c1ab | 2000 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
e80ab33d PB |
2001 | if (--acb->nr_pending) { |
2002 | qemu_coroutine_yield(); | |
2003 | } | |
33b1db1c | 2004 | } |
33b1db1c MK |
2005 | } |
2006 | ||
859e5553 LY |
2007 | /* Delete current working VDI on the snapshot chain */ |
2008 | static bool sd_delete(BDRVSheepdogState *s) | |
2009 | { | |
dfb12bf8 | 2010 | Error *local_err = NULL; |
859e5553 LY |
2011 | unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0; |
2012 | SheepdogVdiReq hdr = { | |
2013 | .opcode = SD_OP_DEL_VDI, | |
9f23fce7 | 2014 | .base_vdi_id = s->inode.vdi_id, |
859e5553 LY |
2015 | .data_length = wlen, |
2016 | .flags = SD_FLAG_CMD_WRITE, | |
2017 | }; | |
2018 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2019 | int fd, ret; | |
2020 | ||
dfb12bf8 | 2021 | fd = connect_to_sdog(s, &local_err); |
859e5553 | 2022 | if (fd < 0) { |
565f65d2 | 2023 | error_report_err(local_err); |
859e5553 LY |
2024 | return false; |
2025 | } | |
2026 | ||
f11672db | 2027 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, |
84390bed | 2028 | s->name, &wlen, &rlen); |
859e5553 LY |
2029 | closesocket(fd); |
2030 | if (ret) { | |
2031 | return false; | |
2032 | } | |
2033 | switch (rsp->result) { | |
2034 | case SD_RES_NO_VDI: | |
2035 | error_report("%s was already deleted", s->name); | |
2036 | /* fall through */ | |
2037 | case SD_RES_SUCCESS: | |
2038 | break; | |
2039 | default: | |
2040 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2041 | return false; | |
2042 | } | |
2043 | ||
2044 | return true; | |
2045 | } | |
2046 | ||
33b1db1c MK |
2047 | /* |
2048 | * Create a writable VDI from a snapshot | |
2049 | */ | |
2050 | static int sd_create_branch(BDRVSheepdogState *s) | |
2051 | { | |
dfb12bf8 | 2052 | Error *local_err = NULL; |
33b1db1c MK |
2053 | int ret, fd; |
2054 | uint32_t vid; | |
2055 | char *buf; | |
859e5553 | 2056 | bool deleted; |
33b1db1c | 2057 | |
2440a2c3 | 2058 | DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
33b1db1c | 2059 | |
7267c094 | 2060 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 2061 | |
859e5553 LY |
2062 | /* |
2063 | * Even If deletion fails, we will just create extra snapshot based on | |
dc6fb73d | 2064 | * the working VDI which was supposed to be deleted. So no need to |
859e5553 LY |
2065 | * false bail out. |
2066 | */ | |
2067 | deleted = sd_delete(s); | |
7d2d3e74 | 2068 | ret = do_sd_create(s, &vid, !deleted, &local_err); |
33b1db1c | 2069 | if (ret) { |
565f65d2 | 2070 | error_report_err(local_err); |
33b1db1c MK |
2071 | goto out; |
2072 | } | |
2073 | ||
2440a2c3 | 2074 | DPRINTF("%" PRIx32 " is created.\n", vid); |
33b1db1c | 2075 | |
dfb12bf8 | 2076 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2077 | if (fd < 0) { |
565f65d2 | 2078 | error_report_err(local_err); |
cb595887 | 2079 | ret = fd; |
33b1db1c MK |
2080 | goto out; |
2081 | } | |
2082 | ||
f11672db | 2083 | ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid), |
84390bed | 2084 | s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
2085 | |
2086 | closesocket(fd); | |
2087 | ||
2088 | if (ret < 0) { | |
2089 | goto out; | |
2090 | } | |
2091 | ||
2092 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
2093 | ||
2f536801 | 2094 | s->is_snapshot = false; |
33b1db1c | 2095 | ret = 0; |
2440a2c3 | 2096 | DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
33b1db1c MK |
2097 | |
2098 | out: | |
7267c094 | 2099 | g_free(buf); |
33b1db1c MK |
2100 | |
2101 | return ret; | |
2102 | } | |
2103 | ||
2104 | /* | |
2105 | * Send I/O requests to the server. | |
2106 | * | |
2107 | * This function sends requests to the server, links the requests to | |
c292ee6a | 2108 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
2109 | * waiting the response. The responses are received in the |
2110 | * `aio_read_response' function which is called from the main loop as | |
2111 | * a fd handler. | |
2df46246 MK |
2112 | * |
2113 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
2114 | * request and -errno in error cases. | |
33b1db1c | 2115 | */ |
28ddd08c | 2116 | static void coroutine_fn sd_co_rw_vector(SheepdogAIOCB *acb) |
33b1db1c | 2117 | { |
33b1db1c | 2118 | int ret = 0; |
e8bfaa2f | 2119 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; |
876eb1b0 TI |
2120 | unsigned long idx; |
2121 | uint32_t object_size; | |
33b1db1c | 2122 | uint64_t oid; |
876eb1b0 | 2123 | uint64_t offset; |
28ddd08c | 2124 | BDRVSheepdogState *s = acb->s; |
33b1db1c MK |
2125 | SheepdogInode *inode = &s->inode; |
2126 | AIOReq *aio_req; | |
2127 | ||
33b1db1c MK |
2128 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
2129 | /* | |
2130 | * In the case we open the snapshot VDI, Sheepdog creates the | |
2131 | * writable VDI when we do a write operation first. | |
2132 | */ | |
2133 | ret = sd_create_branch(s); | |
2134 | if (ret) { | |
2135 | acb->ret = -EIO; | |
e80ab33d | 2136 | return; |
33b1db1c MK |
2137 | } |
2138 | } | |
2139 | ||
876eb1b0 TI |
2140 | object_size = (UINT32_C(1) << inode->block_size_shift); |
2141 | idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
2142 | offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size; | |
2143 | ||
1d732d7d MK |
2144 | /* |
2145 | * Make sure we don't free the aiocb before we are done with all requests. | |
2146 | * This additional reference is dropped at the end of this function. | |
2147 | */ | |
2148 | acb->nr_pending++; | |
2149 | ||
33b1db1c MK |
2150 | while (done != total) { |
2151 | uint8_t flags = 0; | |
2152 | uint64_t old_oid = 0; | |
2f536801 | 2153 | bool create = false; |
33b1db1c MK |
2154 | |
2155 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
2156 | ||
876eb1b0 | 2157 | len = MIN(total - done, object_size - offset); |
33b1db1c | 2158 | |
19db9b90 CH |
2159 | switch (acb->aiocb_type) { |
2160 | case AIOCB_READ_UDATA: | |
2161 | if (!inode->data_vdi_id[idx]) { | |
2162 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
2163 | goto done; |
2164 | } | |
19db9b90 CH |
2165 | break; |
2166 | case AIOCB_WRITE_UDATA: | |
2167 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 2168 | create = true; |
19db9b90 CH |
2169 | } else if (!is_data_obj_writable(inode, idx)) { |
2170 | /* Copy-On-Write */ | |
2f536801 | 2171 | create = true; |
19db9b90 CH |
2172 | old_oid = oid; |
2173 | flags = SD_FLAG_CMD_COW; | |
2174 | } | |
2175 | break; | |
cac8f4a6 LY |
2176 | case AIOCB_DISCARD_OBJ: |
2177 | /* | |
2178 | * We discard the object only when the whole object is | |
2179 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
2180 | */ | |
876eb1b0 | 2181 | if (len != object_size || inode->data_vdi_id[idx] == 0) { |
cac8f4a6 LY |
2182 | goto done; |
2183 | } | |
2184 | break; | |
19db9b90 CH |
2185 | default: |
2186 | break; | |
33b1db1c MK |
2187 | } |
2188 | ||
2189 | if (create) { | |
2440a2c3 | 2190 | DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1b6ac998 | 2191 | inode->vdi_id, oid, |
33b1db1c MK |
2192 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
2193 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
2440a2c3 | 2194 | DPRINTF("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
2195 | } |
2196 | ||
b544c1ab | 2197 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create, |
e6fd57ea HM |
2198 | old_oid, |
2199 | acb->aiocb_type == AIOCB_DISCARD_OBJ ? | |
2200 | 0 : done); | |
b544c1ab | 2201 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 2202 | acb->aiocb_type); |
33b1db1c MK |
2203 | done: |
2204 | offset = 0; | |
2205 | idx++; | |
2206 | done += len; | |
2207 | } | |
e80ab33d PB |
2208 | if (--acb->nr_pending) { |
2209 | qemu_coroutine_yield(); | |
33b1db1c MK |
2210 | } |
2211 | } | |
2212 | ||
acf6e5f0 | 2213 | static void sd_aio_complete(SheepdogAIOCB *acb) |
6a55c82c | 2214 | { |
acf6e5f0 PB |
2215 | if (acb->aiocb_type == AIOCB_FLUSH_CACHE) { |
2216 | return; | |
6a55c82c HM |
2217 | } |
2218 | ||
acf6e5f0 PB |
2219 | QLIST_REMOVE(acb, aiocb_siblings); |
2220 | qemu_co_queue_restart_all(&acb->s->overlapping_queue); | |
6a55c82c HM |
2221 | } |
2222 | ||
a968168c | 2223 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2224 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c | 2225 | { |
28ddd08c | 2226 | SheepdogAIOCB acb; |
2df46246 | 2227 | int ret; |
e50d7607 LY |
2228 | int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE; |
2229 | BDRVSheepdogState *s = bs->opaque; | |
33b1db1c | 2230 | |
c0191e76 | 2231 | if (offset > s->inode.vdi_size) { |
e50d7607 | 2232 | ret = sd_truncate(bs, offset); |
cb595887 MK |
2233 | if (ret < 0) { |
2234 | return ret; | |
33b1db1c | 2235 | } |
33b1db1c MK |
2236 | } |
2237 | ||
28ddd08c | 2238 | sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_WRITE_UDATA); |
28ddd08c PB |
2239 | sd_co_rw_vector(&acb); |
2240 | sd_write_done(&acb); | |
acf6e5f0 | 2241 | sd_aio_complete(&acb); |
2df46246 | 2242 | |
28ddd08c | 2243 | return acb.ret; |
33b1db1c MK |
2244 | } |
2245 | ||
a968168c | 2246 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2247 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c | 2248 | { |
28ddd08c | 2249 | SheepdogAIOCB acb; |
6a55c82c | 2250 | BDRVSheepdogState *s = bs->opaque; |
33b1db1c | 2251 | |
28ddd08c | 2252 | sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_READ_UDATA); |
28ddd08c | 2253 | sd_co_rw_vector(&acb); |
acf6e5f0 | 2254 | sd_aio_complete(&acb); |
2df46246 | 2255 | |
28ddd08c | 2256 | return acb.ret; |
33b1db1c MK |
2257 | } |
2258 | ||
47622c44 LY |
2259 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
2260 | { | |
2261 | BDRVSheepdogState *s = bs->opaque; | |
28ddd08c | 2262 | SheepdogAIOCB acb; |
47783072 | 2263 | AIOReq *aio_req; |
47622c44 | 2264 | |
0e7106d8 | 2265 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
2266 | return 0; |
2267 | } | |
2268 | ||
28ddd08c | 2269 | sd_aio_setup(&acb, s, NULL, 0, 0, AIOCB_FLUSH_CACHE); |
47622c44 | 2270 | |
28ddd08c PB |
2271 | acb.nr_pending++; |
2272 | aio_req = alloc_aio_req(s, &acb, vid_to_vdi_oid(s->inode.vdi_id), | |
b544c1ab | 2273 | 0, 0, 0, false, 0, 0); |
28ddd08c | 2274 | add_aio_request(s, aio_req, NULL, 0, acb.aiocb_type); |
47622c44 | 2275 | |
28ddd08c | 2276 | if (--acb.nr_pending) { |
e80ab33d PB |
2277 | qemu_coroutine_yield(); |
2278 | } | |
acf6e5f0 PB |
2279 | |
2280 | sd_aio_complete(&acb); | |
28ddd08c | 2281 | return acb.ret; |
47622c44 LY |
2282 | } |
2283 | ||
33b1db1c MK |
2284 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
2285 | { | |
dfb12bf8 | 2286 | Error *local_err = NULL; |
33b1db1c MK |
2287 | BDRVSheepdogState *s = bs->opaque; |
2288 | int ret, fd; | |
2289 | uint32_t new_vid; | |
2290 | SheepdogInode *inode; | |
2291 | unsigned int datalen; | |
2292 | ||
2440a2c3 | 2293 | DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
2294 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
2295 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
2296 | ||
2297 | if (s->is_snapshot) { | |
2298 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 2299 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
2300 | |
2301 | return -EINVAL; | |
2302 | } | |
2303 | ||
2440a2c3 | 2304 | DPRINTF("%s %s\n", sn_info->name, sn_info->id_str); |
33b1db1c MK |
2305 | |
2306 | s->inode.vm_state_size = sn_info->vm_state_size; | |
2307 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
2308 | /* It appears that inode.tag does not require a NUL terminator, |
2309 | * which means this use of strncpy is ok. | |
2310 | */ | |
33b1db1c MK |
2311 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
2312 | /* we don't need to update entire object */ | |
2313 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
2df5fee2 | 2314 | inode = g_malloc(datalen); |
33b1db1c MK |
2315 | |
2316 | /* refresh inode. */ | |
dfb12bf8 | 2317 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2318 | if (fd < 0) { |
565f65d2 | 2319 | error_report_err(local_err); |
cb595887 | 2320 | ret = fd; |
33b1db1c MK |
2321 | goto cleanup; |
2322 | } | |
2323 | ||
f11672db | 2324 | ret = write_object(fd, s->bs, (char *)&s->inode, |
84390bed SH |
2325 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, |
2326 | datalen, 0, false, s->cache_flags); | |
33b1db1c | 2327 | if (ret < 0) { |
6daf194d | 2328 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
2329 | goto cleanup; |
2330 | } | |
2331 | ||
7d2d3e74 | 2332 | ret = do_sd_create(s, &new_vid, 1, &local_err); |
33b1db1c | 2333 | if (ret < 0) { |
c29b77f9 MA |
2334 | error_reportf_err(local_err, |
2335 | "failed to create inode for snapshot: "); | |
33b1db1c MK |
2336 | goto cleanup; |
2337 | } | |
2338 | ||
f11672db | 2339 | ret = read_object(fd, s->bs, (char *)inode, |
84390bed SH |
2340 | vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0, |
2341 | s->cache_flags); | |
33b1db1c MK |
2342 | |
2343 | if (ret < 0) { | |
6daf194d | 2344 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
2345 | goto cleanup; |
2346 | } | |
2347 | ||
2348 | memcpy(&s->inode, inode, datalen); | |
2440a2c3 | 2349 | DPRINTF("s->inode: name %s snap_id %x oid %x\n", |
33b1db1c MK |
2350 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
2351 | ||
2352 | cleanup: | |
2df5fee2 | 2353 | g_free(inode); |
33b1db1c MK |
2354 | closesocket(fd); |
2355 | return ret; | |
2356 | } | |
2357 | ||
859e5553 LY |
2358 | /* |
2359 | * We implement rollback(loadvm) operation to the specified snapshot by | |
2360 | * 1) switch to the snapshot | |
2361 | * 2) rely on sd_create_branch to delete working VDI and | |
dc6fb73d | 2362 | * 3) create a new working VDI based on the specified snapshot |
859e5553 | 2363 | */ |
33b1db1c MK |
2364 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
2365 | { | |
2366 | BDRVSheepdogState *s = bs->opaque; | |
2367 | BDRVSheepdogState *old_s; | |
9ff53a0e | 2368 | char tag[SD_MAX_VDI_TAG_LEN]; |
33b1db1c | 2369 | uint32_t snapid = 0; |
9ff53a0e | 2370 | int ret = 0; |
33b1db1c | 2371 | |
5839e53b | 2372 | old_s = g_new(BDRVSheepdogState, 1); |
33b1db1c MK |
2373 | |
2374 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
2375 | ||
33b1db1c | 2376 | snapid = strtoul(snapshot_id, NULL, 10); |
3178e275 JM |
2377 | if (snapid) { |
2378 | tag[0] = 0; | |
2379 | } else { | |
b579ffb3 | 2380 | pstrcpy(tag, sizeof(tag), snapshot_id); |
33b1db1c MK |
2381 | } |
2382 | ||
9ff53a0e | 2383 | ret = reload_inode(s, snapid, tag); |
33b1db1c | 2384 | if (ret) { |
33b1db1c MK |
2385 | goto out; |
2386 | } | |
2387 | ||
cede621f LY |
2388 | ret = sd_create_branch(s); |
2389 | if (ret) { | |
33b1db1c MK |
2390 | goto out; |
2391 | } | |
2392 | ||
7267c094 | 2393 | g_free(old_s); |
33b1db1c MK |
2394 | |
2395 | return 0; | |
2396 | out: | |
2397 | /* recover bdrv_sd_state */ | |
2398 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 | 2399 | g_free(old_s); |
33b1db1c | 2400 | |
6daf194d | 2401 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
2402 | |
2403 | return ret; | |
2404 | } | |
2405 | ||
eab8eb8d VT |
2406 | #define NR_BATCHED_DISCARD 128 |
2407 | ||
2408 | static bool remove_objects(BDRVSheepdogState *s) | |
2409 | { | |
2410 | int fd, i = 0, nr_objs = 0; | |
2411 | Error *local_err = NULL; | |
2412 | int ret = 0; | |
2413 | bool result = true; | |
2414 | SheepdogInode *inode = &s->inode; | |
2415 | ||
2416 | fd = connect_to_sdog(s, &local_err); | |
2417 | if (fd < 0) { | |
2418 | error_report_err(local_err); | |
2419 | return false; | |
2420 | } | |
2421 | ||
2422 | nr_objs = count_data_objs(inode); | |
2423 | while (i < nr_objs) { | |
2424 | int start_idx, nr_filled_idx; | |
2425 | ||
2426 | while (i < nr_objs && !inode->data_vdi_id[i]) { | |
2427 | i++; | |
2428 | } | |
2429 | start_idx = i; | |
2430 | ||
2431 | nr_filled_idx = 0; | |
2432 | while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) { | |
2433 | if (inode->data_vdi_id[i]) { | |
2434 | inode->data_vdi_id[i] = 0; | |
2435 | nr_filled_idx++; | |
2436 | } | |
2437 | ||
2438 | i++; | |
2439 | } | |
2440 | ||
f11672db | 2441 | ret = write_object(fd, s->bs, |
eab8eb8d VT |
2442 | (char *)&inode->data_vdi_id[start_idx], |
2443 | vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies, | |
2444 | (i - start_idx) * sizeof(uint32_t), | |
2445 | offsetof(struct SheepdogInode, | |
2446 | data_vdi_id[start_idx]), | |
2447 | false, s->cache_flags); | |
2448 | if (ret < 0) { | |
2449 | error_report("failed to discard snapshot inode."); | |
2450 | result = false; | |
2451 | goto out; | |
2452 | } | |
2453 | } | |
2454 | ||
2455 | out: | |
2456 | closesocket(fd); | |
2457 | return result; | |
2458 | } | |
2459 | ||
a89d89d3 WX |
2460 | static int sd_snapshot_delete(BlockDriverState *bs, |
2461 | const char *snapshot_id, | |
2462 | const char *name, | |
2463 | Error **errp) | |
33b1db1c | 2464 | { |
03c698f0 | 2465 | unsigned long snap_id = 0; |
eab8eb8d VT |
2466 | char snap_tag[SD_MAX_VDI_TAG_LEN]; |
2467 | Error *local_err = NULL; | |
2468 | int fd, ret; | |
2469 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
2470 | BDRVSheepdogState *s = bs->opaque; | |
2471 | unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0; | |
2472 | uint32_t vid; | |
2473 | SheepdogVdiReq hdr = { | |
2474 | .opcode = SD_OP_DEL_VDI, | |
2475 | .data_length = wlen, | |
2476 | .flags = SD_FLAG_CMD_WRITE, | |
2477 | }; | |
2478 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2479 | ||
2480 | if (!remove_objects(s)) { | |
2481 | return -1; | |
2482 | } | |
2483 | ||
2484 | memset(buf, 0, sizeof(buf)); | |
2485 | memset(snap_tag, 0, sizeof(snap_tag)); | |
2486 | pstrcpy(buf, SD_MAX_VDI_LEN, s->name); | |
03c698f0 JC |
2487 | ret = qemu_strtoul(snapshot_id, NULL, 10, &snap_id); |
2488 | if (ret || snap_id > UINT32_MAX) { | |
2489 | error_setg(errp, "Invalid snapshot ID: %s", | |
2490 | snapshot_id ? snapshot_id : "<null>"); | |
2491 | return -EINVAL; | |
eab8eb8d VT |
2492 | } |
2493 | ||
2494 | if (snap_id) { | |
03c698f0 | 2495 | hdr.snapid = (uint32_t) snap_id; |
eab8eb8d VT |
2496 | } else { |
2497 | pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id); | |
2498 | pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag); | |
2499 | } | |
2500 | ||
2501 | ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, | |
2502 | &local_err); | |
2503 | if (ret) { | |
2504 | return ret; | |
2505 | } | |
2506 | ||
2507 | fd = connect_to_sdog(s, &local_err); | |
2508 | if (fd < 0) { | |
2509 | error_report_err(local_err); | |
2510 | return -1; | |
2511 | } | |
2512 | ||
f11672db | 2513 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, |
eab8eb8d VT |
2514 | buf, &wlen, &rlen); |
2515 | closesocket(fd); | |
2516 | if (ret) { | |
2517 | return ret; | |
2518 | } | |
2519 | ||
2520 | switch (rsp->result) { | |
2521 | case SD_RES_NO_VDI: | |
2522 | error_report("%s was already deleted", s->name); | |
2523 | case SD_RES_SUCCESS: | |
2524 | break; | |
2525 | default: | |
2526 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2527 | return -1; | |
2528 | } | |
2529 | ||
2530 | return ret; | |
33b1db1c MK |
2531 | } |
2532 | ||
33b1db1c MK |
2533 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
2534 | { | |
dfb12bf8 | 2535 | Error *local_err = NULL; |
33b1db1c MK |
2536 | BDRVSheepdogState *s = bs->opaque; |
2537 | SheepdogReq req; | |
2538 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
2539 | QEMUSnapshotInfo *sn_tab = NULL; | |
2540 | unsigned wlen, rlen; | |
2541 | int found = 0; | |
2542 | static SheepdogInode inode; | |
2543 | unsigned long *vdi_inuse; | |
2544 | unsigned int start_nr; | |
2545 | uint64_t hval; | |
2546 | uint32_t vid; | |
2547 | ||
7267c094 | 2548 | vdi_inuse = g_malloc(max); |
33b1db1c | 2549 | |
dfb12bf8 | 2550 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2551 | if (fd < 0) { |
565f65d2 | 2552 | error_report_err(local_err); |
cb595887 | 2553 | ret = fd; |
33b1db1c MK |
2554 | goto out; |
2555 | } | |
2556 | ||
2557 | rlen = max; | |
2558 | wlen = 0; | |
2559 | ||
2560 | memset(&req, 0, sizeof(req)); | |
2561 | ||
2562 | req.opcode = SD_OP_READ_VDIS; | |
2563 | req.data_length = max; | |
2564 | ||
f11672db | 2565 | ret = do_req(fd, s->bs, &req, vdi_inuse, &wlen, &rlen); |
33b1db1c MK |
2566 | |
2567 | closesocket(fd); | |
2568 | if (ret) { | |
2569 | goto out; | |
2570 | } | |
2571 | ||
02c4f26b | 2572 | sn_tab = g_new0(QEMUSnapshotInfo, nr); |
33b1db1c MK |
2573 | |
2574 | /* calculate a vdi id with hash function */ | |
2575 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2576 | start_nr = hval & (SD_NR_VDIS - 1); | |
2577 | ||
dfb12bf8 | 2578 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2579 | if (fd < 0) { |
565f65d2 | 2580 | error_report_err(local_err); |
cb595887 | 2581 | ret = fd; |
33b1db1c MK |
2582 | goto out; |
2583 | } | |
2584 | ||
2585 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2586 | if (!test_bit(vid, vdi_inuse)) { | |
2587 | break; | |
2588 | } | |
2589 | ||
2590 | /* we don't need to read entire object */ | |
f11672db | 2591 | ret = read_object(fd, s->bs, (char *)&inode, |
84390bed | 2592 | vid_to_vdi_oid(vid), |
47622c44 | 2593 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2594 | s->cache_flags); |
33b1db1c MK |
2595 | |
2596 | if (ret) { | |
2597 | continue; | |
2598 | } | |
2599 | ||
2600 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2601 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2602 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2603 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2604 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2605 | ||
521b2b5d HR |
2606 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), |
2607 | "%" PRIu32, inode.snap_id); | |
3178e275 JM |
2608 | pstrcpy(sn_tab[found].name, |
2609 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2610 | inode.tag); | |
33b1db1c MK |
2611 | found++; |
2612 | } | |
2613 | } | |
2614 | ||
2615 | closesocket(fd); | |
2616 | out: | |
2617 | *psn_tab = sn_tab; | |
2618 | ||
7267c094 | 2619 | g_free(vdi_inuse); |
33b1db1c | 2620 | |
cb595887 MK |
2621 | if (ret < 0) { |
2622 | return ret; | |
2623 | } | |
2624 | ||
33b1db1c MK |
2625 | return found; |
2626 | } | |
2627 | ||
2628 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2629 | int64_t pos, int size, int load) | |
2630 | { | |
dfb12bf8 | 2631 | Error *local_err = NULL; |
2f536801 MK |
2632 | bool create; |
2633 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2634 | unsigned int data_len; |
2635 | uint64_t vmstate_oid; | |
33b1db1c | 2636 | uint64_t offset; |
cede621f LY |
2637 | uint32_t vdi_index; |
2638 | uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id; | |
876eb1b0 | 2639 | uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift); |
33b1db1c | 2640 | |
dfb12bf8 | 2641 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2642 | if (fd < 0) { |
565f65d2 | 2643 | error_report_err(local_err); |
cb595887 | 2644 | return fd; |
33b1db1c MK |
2645 | } |
2646 | ||
6f3c714e | 2647 | while (remaining) { |
876eb1b0 TI |
2648 | vdi_index = pos / object_size; |
2649 | offset = pos % object_size; | |
33b1db1c | 2650 | |
876eb1b0 | 2651 | data_len = MIN(remaining, object_size - offset); |
33b1db1c | 2652 | |
cede621f | 2653 | vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index); |
33b1db1c MK |
2654 | |
2655 | create = (offset == 0); | |
2656 | if (load) { | |
f11672db | 2657 | ret = read_object(fd, s->bs, (char *)data, vmstate_oid, |
47622c44 | 2658 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2659 | s->cache_flags); |
33b1db1c | 2660 | } else { |
f11672db | 2661 | ret = write_object(fd, s->bs, (char *)data, vmstate_oid, |
47622c44 | 2662 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2663 | s->cache_flags); |
33b1db1c MK |
2664 | } |
2665 | ||
2666 | if (ret < 0) { | |
6daf194d | 2667 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2668 | goto cleanup; |
2669 | } | |
2670 | ||
2671 | pos += data_len; | |
1f7a48de | 2672 | data += data_len; |
6f3c714e | 2673 | remaining -= data_len; |
33b1db1c | 2674 | } |
6f3c714e | 2675 | ret = size; |
33b1db1c MK |
2676 | cleanup: |
2677 | closesocket(fd); | |
2678 | return ret; | |
2679 | } | |
2680 | ||
cf8074b3 KW |
2681 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2682 | int64_t pos) | |
33b1db1c MK |
2683 | { |
2684 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2685 | void *buf; |
2686 | int ret; | |
33b1db1c | 2687 | |
cf8074b3 KW |
2688 | buf = qemu_blockalign(bs, qiov->size); |
2689 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2690 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2691 | qemu_vfree(buf); | |
2692 | ||
2693 | return ret; | |
33b1db1c MK |
2694 | } |
2695 | ||
5ddda0b8 KW |
2696 | static int sd_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2697 | int64_t pos) | |
33b1db1c MK |
2698 | { |
2699 | BDRVSheepdogState *s = bs->opaque; | |
5ddda0b8 KW |
2700 | void *buf; |
2701 | int ret; | |
33b1db1c | 2702 | |
5ddda0b8 KW |
2703 | buf = qemu_blockalign(bs, qiov->size); |
2704 | ret = do_load_save_vmstate(s, buf, pos, qiov->size, 1); | |
2705 | qemu_iovec_from_buf(qiov, 0, buf, qiov->size); | |
2706 | qemu_vfree(buf); | |
2707 | ||
2708 | return ret; | |
33b1db1c MK |
2709 | } |
2710 | ||
2711 | ||
dde47537 EB |
2712 | static coroutine_fn int sd_co_pdiscard(BlockDriverState *bs, int64_t offset, |
2713 | int count) | |
cac8f4a6 | 2714 | { |
28ddd08c | 2715 | SheepdogAIOCB acb; |
cac8f4a6 | 2716 | BDRVSheepdogState *s = bs->opaque; |
e6fd57ea HM |
2717 | QEMUIOVector discard_iov; |
2718 | struct iovec iov; | |
2719 | uint32_t zero = 0; | |
cac8f4a6 LY |
2720 | |
2721 | if (!s->discard_supported) { | |
dde47537 | 2722 | return 0; |
cac8f4a6 LY |
2723 | } |
2724 | ||
e6fd57ea HM |
2725 | memset(&discard_iov, 0, sizeof(discard_iov)); |
2726 | memset(&iov, 0, sizeof(iov)); | |
2727 | iov.iov_base = &zero; | |
2728 | iov.iov_len = sizeof(zero); | |
2729 | discard_iov.iov = &iov; | |
2730 | discard_iov.niov = 1; | |
49228d1e EB |
2731 | if (!QEMU_IS_ALIGNED(offset | count, BDRV_SECTOR_SIZE)) { |
2732 | return -ENOTSUP; | |
2733 | } | |
28ddd08c PB |
2734 | sd_aio_setup(&acb, s, &discard_iov, offset >> BDRV_SECTOR_BITS, |
2735 | count >> BDRV_SECTOR_BITS, AIOCB_DISCARD_OBJ); | |
28ddd08c | 2736 | sd_co_rw_vector(&acb); |
acf6e5f0 | 2737 | sd_aio_complete(&acb); |
cac8f4a6 | 2738 | |
28ddd08c | 2739 | return acb.ret; |
cac8f4a6 LY |
2740 | } |
2741 | ||
b6b8a333 PB |
2742 | static coroutine_fn int64_t |
2743 | sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
67a0fd2a | 2744 | int *pnum, BlockDriverState **file) |
8d71c631 LY |
2745 | { |
2746 | BDRVSheepdogState *s = bs->opaque; | |
2747 | SheepdogInode *inode = &s->inode; | |
876eb1b0 | 2748 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
9cd76737 | 2749 | uint64_t offset = sector_num * BDRV_SECTOR_SIZE; |
876eb1b0 | 2750 | unsigned long start = offset / object_size, |
8d71c631 | 2751 | end = DIV_ROUND_UP((sector_num + nb_sectors) * |
876eb1b0 | 2752 | BDRV_SECTOR_SIZE, object_size); |
8d71c631 | 2753 | unsigned long idx; |
9cd76737 | 2754 | int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset; |
8d71c631 LY |
2755 | |
2756 | for (idx = start; idx < end; idx++) { | |
2757 | if (inode->data_vdi_id[idx] == 0) { | |
2758 | break; | |
2759 | } | |
2760 | } | |
2761 | if (idx == start) { | |
2762 | /* Get the longest length of unallocated sectors */ | |
2763 | ret = 0; | |
2764 | for (idx = start + 1; idx < end; idx++) { | |
2765 | if (inode->data_vdi_id[idx] != 0) { | |
2766 | break; | |
2767 | } | |
2768 | } | |
2769 | } | |
2770 | ||
876eb1b0 | 2771 | *pnum = (idx - start) * object_size / BDRV_SECTOR_SIZE; |
8d71c631 LY |
2772 | if (*pnum > nb_sectors) { |
2773 | *pnum = nb_sectors; | |
2774 | } | |
d234c929 FZ |
2775 | if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) { |
2776 | *file = bs; | |
2777 | } | |
8d71c631 LY |
2778 | return ret; |
2779 | } | |
2780 | ||
85829722 LY |
2781 | static int64_t sd_get_allocated_file_size(BlockDriverState *bs) |
2782 | { | |
2783 | BDRVSheepdogState *s = bs->opaque; | |
2784 | SheepdogInode *inode = &s->inode; | |
876eb1b0 TI |
2785 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
2786 | unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size); | |
85829722 LY |
2787 | uint64_t size = 0; |
2788 | ||
2789 | for (i = 0; i < last; i++) { | |
2790 | if (inode->data_vdi_id[i] == 0) { | |
2791 | continue; | |
2792 | } | |
876eb1b0 | 2793 | size += object_size; |
85829722 LY |
2794 | } |
2795 | return size; | |
2796 | } | |
2797 | ||
b222237b CL |
2798 | static QemuOptsList sd_create_opts = { |
2799 | .name = "sheepdog-create-opts", | |
2800 | .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head), | |
2801 | .desc = { | |
2802 | { | |
2803 | .name = BLOCK_OPT_SIZE, | |
2804 | .type = QEMU_OPT_SIZE, | |
2805 | .help = "Virtual disk size" | |
2806 | }, | |
2807 | { | |
2808 | .name = BLOCK_OPT_BACKING_FILE, | |
2809 | .type = QEMU_OPT_STRING, | |
2810 | .help = "File name of a base image" | |
2811 | }, | |
2812 | { | |
2813 | .name = BLOCK_OPT_PREALLOC, | |
2814 | .type = QEMU_OPT_STRING, | |
2815 | .help = "Preallocation mode (allowed values: off, full)" | |
2816 | }, | |
2817 | { | |
2818 | .name = BLOCK_OPT_REDUNDANCY, | |
2819 | .type = QEMU_OPT_STRING, | |
2820 | .help = "Redundancy of the image" | |
2821 | }, | |
876eb1b0 TI |
2822 | { |
2823 | .name = BLOCK_OPT_OBJECT_SIZE, | |
2824 | .type = QEMU_OPT_SIZE, | |
2825 | .help = "Object size of the image" | |
2826 | }, | |
b222237b CL |
2827 | { /* end of list */ } |
2828 | } | |
33b1db1c MK |
2829 | }; |
2830 | ||
5d6768e3 | 2831 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2832 | .format_name = "sheepdog", |
2833 | .protocol_name = "sheepdog", | |
2834 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2835 | .bdrv_needs_filename = true, |
33b1db1c | 2836 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2837 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2838 | .bdrv_reopen_commit = sd_reopen_commit, | |
2839 | .bdrv_reopen_abort = sd_reopen_abort, | |
33b1db1c | 2840 | .bdrv_close = sd_close, |
c282e1fd | 2841 | .bdrv_create = sd_create, |
e4f5c1bf | 2842 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
33b1db1c | 2843 | .bdrv_getlength = sd_getlength, |
85829722 | 2844 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
33b1db1c MK |
2845 | .bdrv_truncate = sd_truncate, |
2846 | ||
2df46246 MK |
2847 | .bdrv_co_readv = sd_co_readv, |
2848 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2849 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
dde47537 | 2850 | .bdrv_co_pdiscard = sd_co_pdiscard, |
b6b8a333 | 2851 | .bdrv_co_get_block_status = sd_co_get_block_status, |
33b1db1c MK |
2852 | |
2853 | .bdrv_snapshot_create = sd_snapshot_create, | |
2854 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2855 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2856 | .bdrv_snapshot_list = sd_snapshot_list, | |
2857 | ||
2858 | .bdrv_save_vmstate = sd_save_vmstate, | |
2859 | .bdrv_load_vmstate = sd_load_vmstate, | |
2860 | ||
84390bed SH |
2861 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2862 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2863 | ||
b222237b | 2864 | .create_opts = &sd_create_opts, |
33b1db1c MK |
2865 | }; |
2866 | ||
5d6768e3 MK |
2867 | static BlockDriver bdrv_sheepdog_tcp = { |
2868 | .format_name = "sheepdog", | |
2869 | .protocol_name = "sheepdog+tcp", | |
2870 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2871 | .bdrv_needs_filename = true, |
5d6768e3 | 2872 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2873 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2874 | .bdrv_reopen_commit = sd_reopen_commit, | |
2875 | .bdrv_reopen_abort = sd_reopen_abort, | |
5d6768e3 | 2876 | .bdrv_close = sd_close, |
c282e1fd | 2877 | .bdrv_create = sd_create, |
e4f5c1bf | 2878 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
5d6768e3 | 2879 | .bdrv_getlength = sd_getlength, |
85829722 | 2880 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
5d6768e3 MK |
2881 | .bdrv_truncate = sd_truncate, |
2882 | ||
2883 | .bdrv_co_readv = sd_co_readv, | |
2884 | .bdrv_co_writev = sd_co_writev, | |
2885 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
dde47537 | 2886 | .bdrv_co_pdiscard = sd_co_pdiscard, |
b6b8a333 | 2887 | .bdrv_co_get_block_status = sd_co_get_block_status, |
5d6768e3 MK |
2888 | |
2889 | .bdrv_snapshot_create = sd_snapshot_create, | |
2890 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2891 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2892 | .bdrv_snapshot_list = sd_snapshot_list, | |
2893 | ||
2894 | .bdrv_save_vmstate = sd_save_vmstate, | |
2895 | .bdrv_load_vmstate = sd_load_vmstate, | |
2896 | ||
84390bed SH |
2897 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2898 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2899 | ||
b222237b | 2900 | .create_opts = &sd_create_opts, |
5d6768e3 MK |
2901 | }; |
2902 | ||
1b8bbb46 MK |
2903 | static BlockDriver bdrv_sheepdog_unix = { |
2904 | .format_name = "sheepdog", | |
2905 | .protocol_name = "sheepdog+unix", | |
2906 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2907 | .bdrv_needs_filename = true, |
1b8bbb46 | 2908 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2909 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2910 | .bdrv_reopen_commit = sd_reopen_commit, | |
2911 | .bdrv_reopen_abort = sd_reopen_abort, | |
1b8bbb46 | 2912 | .bdrv_close = sd_close, |
c282e1fd | 2913 | .bdrv_create = sd_create, |
3ac21627 | 2914 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
1b8bbb46 | 2915 | .bdrv_getlength = sd_getlength, |
85829722 | 2916 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
1b8bbb46 MK |
2917 | .bdrv_truncate = sd_truncate, |
2918 | ||
2919 | .bdrv_co_readv = sd_co_readv, | |
2920 | .bdrv_co_writev = sd_co_writev, | |
2921 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
dde47537 | 2922 | .bdrv_co_pdiscard = sd_co_pdiscard, |
b6b8a333 | 2923 | .bdrv_co_get_block_status = sd_co_get_block_status, |
1b8bbb46 MK |
2924 | |
2925 | .bdrv_snapshot_create = sd_snapshot_create, | |
2926 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2927 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2928 | .bdrv_snapshot_list = sd_snapshot_list, | |
2929 | ||
2930 | .bdrv_save_vmstate = sd_save_vmstate, | |
2931 | .bdrv_load_vmstate = sd_load_vmstate, | |
2932 | ||
84390bed SH |
2933 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2934 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2935 | ||
b222237b | 2936 | .create_opts = &sd_create_opts, |
1b8bbb46 MK |
2937 | }; |
2938 | ||
33b1db1c MK |
2939 | static void bdrv_sheepdog_init(void) |
2940 | { | |
2941 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 2942 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 2943 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
2944 | } |
2945 | block_init(bdrv_sheepdog_init); |