]>
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 | ||
89e2a31d MA |
917 | /* |
918 | * Parse numeric snapshot ID in @str | |
919 | * If @str can't be parsed as number, return false. | |
920 | * Else, if the number is zero or too large, set *@snapid to zero and | |
921 | * return true. | |
922 | * Else, set *@snapid to the number and return true. | |
923 | */ | |
924 | static bool sd_parse_snapid(const char *str, uint32_t *snapid) | |
925 | { | |
926 | unsigned long ul; | |
927 | int ret; | |
928 | ||
929 | ret = qemu_strtoul(str, NULL, 10, &ul); | |
930 | if (ret == -ERANGE) { | |
931 | ul = ret = 0; | |
932 | } | |
933 | if (ret) { | |
934 | return false; | |
935 | } | |
936 | if (ul > UINT32_MAX) { | |
937 | ul = 0; | |
938 | } | |
939 | ||
940 | *snapid = ul; | |
941 | return true; | |
942 | } | |
943 | ||
944 | static bool sd_parse_snapid_or_tag(const char *str, | |
945 | uint32_t *snapid, char tag[]) | |
946 | { | |
947 | if (!sd_parse_snapid(str, snapid)) { | |
948 | *snapid = 0; | |
949 | if (g_strlcpy(tag, str, SD_MAX_VDI_TAG_LEN) >= SD_MAX_VDI_TAG_LEN) { | |
950 | return false; | |
951 | } | |
952 | } else if (!*snapid) { | |
953 | return false; | |
954 | } else { | |
955 | tag[0] = 0; | |
956 | } | |
957 | return true; | |
958 | } | |
959 | ||
5d6768e3 MK |
960 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
961 | char *vdi, uint32_t *snapid, char *tag) | |
962 | { | |
963 | URI *uri; | |
964 | QueryParams *qp = NULL; | |
965 | int ret = 0; | |
966 | ||
967 | uri = uri_parse(filename); | |
968 | if (!uri) { | |
969 | return -EINVAL; | |
970 | } | |
971 | ||
1b8bbb46 MK |
972 | /* transport */ |
973 | if (!strcmp(uri->scheme, "sheepdog")) { | |
974 | s->is_unix = false; | |
975 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
976 | s->is_unix = false; | |
977 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
978 | s->is_unix = true; | |
979 | } else { | |
980 | ret = -EINVAL; | |
981 | goto out; | |
982 | } | |
983 | ||
5d6768e3 MK |
984 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
985 | ret = -EINVAL; | |
986 | goto out; | |
987 | } | |
daa0b0d4 MA |
988 | if (g_strlcpy(vdi, uri->path + 1, SD_MAX_VDI_LEN) >= SD_MAX_VDI_LEN) { |
989 | ret = -EINVAL; | |
990 | goto out; | |
991 | } | |
5d6768e3 | 992 | |
1b8bbb46 MK |
993 | qp = query_params_parse(uri->query); |
994 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
995 | ret = -EINVAL; | |
996 | goto out; | |
997 | } | |
998 | ||
999 | if (s->is_unix) { | |
1000 | /* sheepdog+unix:///vdiname?socket=path */ | |
1001 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
1002 | ret = -EINVAL; | |
1003 | goto out; | |
1004 | } | |
1005 | s->host_spec = g_strdup(qp->p[0].value); | |
1006 | } else { | |
1007 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
1008 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
1009 | uri->port ?: SD_DEFAULT_PORT); | |
1010 | } | |
5d6768e3 MK |
1011 | |
1012 | /* snapshot tag */ | |
1013 | if (uri->fragment) { | |
89e2a31d MA |
1014 | if (!sd_parse_snapid_or_tag(uri->fragment, snapid, tag)) { |
1015 | ret = -EINVAL; | |
1016 | goto out; | |
5d6768e3 MK |
1017 | } |
1018 | } else { | |
1019 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
1020 | } | |
1021 | ||
1022 | out: | |
1023 | if (qp) { | |
1024 | query_params_free(qp); | |
1025 | } | |
1026 | uri_free(uri); | |
1027 | return ret; | |
1028 | } | |
1029 | ||
33b1db1c | 1030 | /* |
5d6768e3 | 1031 | * Parse a filename (old syntax) |
33b1db1c MK |
1032 | * |
1033 | * filename must be one of the following formats: | |
1034 | * 1. [vdiname] | |
1035 | * 2. [vdiname]:[snapid] | |
1036 | * 3. [vdiname]:[tag] | |
1037 | * 4. [hostname]:[port]:[vdiname] | |
1038 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
1039 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
1040 | * | |
1041 | * You can boot from the snapshot images by specifying `snapid` or | |
1042 | * `tag'. | |
1043 | * | |
1044 | * You can run VMs outside the Sheepdog cluster by specifying | |
1045 | * `hostname' and `port' (experimental). | |
1046 | */ | |
1047 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
1048 | char *vdi, uint32_t *snapid, char *tag) | |
1049 | { | |
5d6768e3 MK |
1050 | char *p, *q, *uri; |
1051 | const char *host_spec, *vdi_spec; | |
1052 | int nr_sep, ret; | |
33b1db1c | 1053 | |
11d816a5 | 1054 | strstart(filename, "sheepdog:", &filename); |
7267c094 | 1055 | p = q = g_strdup(filename); |
33b1db1c MK |
1056 | |
1057 | /* count the number of separators */ | |
1058 | nr_sep = 0; | |
1059 | while (*p) { | |
1060 | if (*p == ':') { | |
1061 | nr_sep++; | |
1062 | } | |
1063 | p++; | |
1064 | } | |
1065 | p = q; | |
1066 | ||
5d6768e3 | 1067 | /* use the first two tokens as host_spec. */ |
33b1db1c | 1068 | if (nr_sep >= 2) { |
5d6768e3 | 1069 | host_spec = p; |
33b1db1c | 1070 | p = strchr(p, ':'); |
5d6768e3 | 1071 | p++; |
33b1db1c MK |
1072 | p = strchr(p, ':'); |
1073 | *p++ = '\0'; | |
1074 | } else { | |
5d6768e3 | 1075 | host_spec = ""; |
33b1db1c MK |
1076 | } |
1077 | ||
5d6768e3 | 1078 | vdi_spec = p; |
33b1db1c | 1079 | |
5d6768e3 | 1080 | p = strchr(vdi_spec, ':'); |
33b1db1c | 1081 | if (p) { |
5d6768e3 | 1082 | *p++ = '#'; |
33b1db1c MK |
1083 | } |
1084 | ||
5d6768e3 | 1085 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 1086 | |
5d6768e3 MK |
1087 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
1088 | ||
1089 | g_free(q); | |
1090 | g_free(uri); | |
1091 | ||
1092 | return ret; | |
33b1db1c MK |
1093 | } |
1094 | ||
982dcbf4 MK |
1095 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, |
1096 | uint32_t snapid, const char *tag, uint32_t *vid, | |
dc83cd42 | 1097 | bool lock, Error **errp) |
33b1db1c MK |
1098 | { |
1099 | int ret, fd; | |
1100 | SheepdogVdiReq hdr; | |
1101 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1102 | unsigned int wlen, rlen = 0; | |
1103 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
1104 | ||
dc83cd42 | 1105 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1106 | if (fd < 0) { |
cb595887 | 1107 | return fd; |
33b1db1c MK |
1108 | } |
1109 | ||
3178e275 JM |
1110 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
1111 | * which is desirable since we'll soon be sending those bytes, and | |
1112 | * don't want the send_req to read uninitialized data. | |
1113 | */ | |
33b1db1c MK |
1114 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
1115 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
1116 | ||
1117 | memset(&hdr, 0, sizeof(hdr)); | |
982dcbf4 | 1118 | if (lock) { |
33b1db1c | 1119 | hdr.opcode = SD_OP_LOCK_VDI; |
1dbfafed | 1120 | hdr.type = LOCK_TYPE_NORMAL; |
982dcbf4 MK |
1121 | } else { |
1122 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
33b1db1c MK |
1123 | } |
1124 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
1125 | hdr.proto_ver = SD_PROTO_VER; | |
1126 | hdr.data_length = wlen; | |
1127 | hdr.snapid = snapid; | |
1128 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1129 | ||
f11672db | 1130 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1131 | if (ret) { |
dc83cd42 | 1132 | error_setg_errno(errp, -ret, "cannot get vdi info"); |
33b1db1c MK |
1133 | goto out; |
1134 | } | |
1135 | ||
1136 | if (rsp->result != SD_RES_SUCCESS) { | |
dc83cd42 MA |
1137 | error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s", |
1138 | sd_strerror(rsp->result), filename, snapid, tag); | |
cb595887 MK |
1139 | if (rsp->result == SD_RES_NO_VDI) { |
1140 | ret = -ENOENT; | |
38890b24 HM |
1141 | } else if (rsp->result == SD_RES_VDI_LOCKED) { |
1142 | ret = -EBUSY; | |
cb595887 MK |
1143 | } else { |
1144 | ret = -EIO; | |
1145 | } | |
33b1db1c MK |
1146 | goto out; |
1147 | } | |
1148 | *vid = rsp->vdi_id; | |
1149 | ||
1150 | ret = 0; | |
1151 | out: | |
1152 | closesocket(fd); | |
1153 | return ret; | |
1154 | } | |
1155 | ||
a37dcdf9 | 1156 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
1157 | struct iovec *iov, int niov, |
1158 | enum AIOCBState aiocb_type) | |
33b1db1c MK |
1159 | { |
1160 | int nr_copies = s->inode.nr_copies; | |
1161 | SheepdogObjReq hdr; | |
47783072 | 1162 | unsigned int wlen = 0; |
33b1db1c MK |
1163 | int ret; |
1164 | uint64_t oid = aio_req->oid; | |
1165 | unsigned int datalen = aio_req->data_len; | |
1166 | uint64_t offset = aio_req->offset; | |
1167 | uint8_t flags = aio_req->flags; | |
1168 | uint64_t old_oid = aio_req->base_oid; | |
b544c1ab | 1169 | bool create = aio_req->create; |
33b1db1c | 1170 | |
c4080e93 PB |
1171 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1172 | ||
33b1db1c | 1173 | if (!nr_copies) { |
6daf194d | 1174 | error_report("bug"); |
33b1db1c MK |
1175 | } |
1176 | ||
1177 | memset(&hdr, 0, sizeof(hdr)); | |
1178 | ||
47783072 LY |
1179 | switch (aiocb_type) { |
1180 | case AIOCB_FLUSH_CACHE: | |
1181 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1182 | break; | |
1183 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1184 | hdr.opcode = SD_OP_READ_OBJ; |
1185 | hdr.flags = flags; | |
47783072 LY |
1186 | break; |
1187 | case AIOCB_WRITE_UDATA: | |
1188 | if (create) { | |
1189 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1190 | } else { | |
1191 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1192 | } | |
33b1db1c | 1193 | wlen = datalen; |
33b1db1c | 1194 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1195 | break; |
cac8f4a6 | 1196 | case AIOCB_DISCARD_OBJ: |
e6fd57ea HM |
1197 | hdr.opcode = SD_OP_WRITE_OBJ; |
1198 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
1199 | s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0; | |
1200 | offset = offsetof(SheepdogInode, | |
1201 | data_vdi_id[data_oid_to_idx(oid)]); | |
1202 | oid = vid_to_vdi_oid(s->inode.vdi_id); | |
1203 | wlen = datalen = sizeof(uint32_t); | |
cac8f4a6 | 1204 | break; |
33b1db1c MK |
1205 | } |
1206 | ||
0e7106d8 LY |
1207 | if (s->cache_flags) { |
1208 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1209 | } |
1210 | ||
33b1db1c MK |
1211 | hdr.oid = oid; |
1212 | hdr.cow_oid = old_oid; | |
1213 | hdr.copies = s->inode.nr_copies; | |
1214 | ||
1215 | hdr.data_length = datalen; | |
1216 | hdr.offset = offset; | |
1217 | ||
1218 | hdr.id = aio_req->id; | |
1219 | ||
2df46246 MK |
1220 | qemu_co_mutex_lock(&s->lock); |
1221 | s->co_send = qemu_coroutine_self(); | |
dca21ef2 | 1222 | aio_set_fd_handler(s->aio_context, s->fd, false, |
f6a51c84 | 1223 | co_read_response, co_write_request, NULL, s); |
128aa589 | 1224 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1225 | |
1226 | /* send a header */ | |
8c5135f9 | 1227 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
80731d9d | 1228 | if (ret != sizeof(hdr)) { |
6daf194d | 1229 | error_report("failed to send a req, %s", strerror(errno)); |
011603ca | 1230 | goto out; |
33b1db1c MK |
1231 | } |
1232 | ||
1233 | if (wlen) { | |
2fc8ae1d | 1234 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
80731d9d | 1235 | if (ret != wlen) { |
6daf194d | 1236 | error_report("failed to send a data, %s", strerror(errno)); |
33b1db1c MK |
1237 | } |
1238 | } | |
011603ca | 1239 | out: |
128aa589 | 1240 | socket_set_cork(s->fd, 0); |
dca21ef2 | 1241 | aio_set_fd_handler(s->aio_context, s->fd, false, |
f6a51c84 | 1242 | co_read_response, NULL, NULL, s); |
011603ca | 1243 | s->co_send = NULL; |
2df46246 | 1244 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1245 | } |
1246 | ||
f11672db | 1247 | static int read_write_object(int fd, BlockDriverState *bs, char *buf, |
84390bed | 1248 | uint64_t oid, uint8_t copies, |
33b1db1c | 1249 | unsigned int datalen, uint64_t offset, |
0e7106d8 | 1250 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1251 | { |
1252 | SheepdogObjReq hdr; | |
1253 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1254 | unsigned int wlen, rlen; | |
1255 | int ret; | |
1256 | ||
1257 | memset(&hdr, 0, sizeof(hdr)); | |
1258 | ||
1259 | if (write) { | |
1260 | wlen = datalen; | |
1261 | rlen = 0; | |
1262 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1263 | if (create) { | |
1264 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1265 | } else { | |
1266 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1267 | } | |
1268 | } else { | |
1269 | wlen = 0; | |
1270 | rlen = datalen; | |
1271 | hdr.opcode = SD_OP_READ_OBJ; | |
1272 | } | |
47622c44 | 1273 | |
0e7106d8 | 1274 | hdr.flags |= cache_flags; |
47622c44 | 1275 | |
33b1db1c MK |
1276 | hdr.oid = oid; |
1277 | hdr.data_length = datalen; | |
1278 | hdr.offset = offset; | |
1279 | hdr.copies = copies; | |
1280 | ||
f11672db | 1281 | ret = do_req(fd, bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1282 | if (ret) { |
6daf194d | 1283 | error_report("failed to send a request to the sheep"); |
cb595887 | 1284 | return ret; |
33b1db1c MK |
1285 | } |
1286 | ||
1287 | switch (rsp->result) { | |
1288 | case SD_RES_SUCCESS: | |
1289 | return 0; | |
1290 | default: | |
6daf194d | 1291 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1292 | return -EIO; |
33b1db1c MK |
1293 | } |
1294 | } | |
1295 | ||
f11672db | 1296 | static int read_object(int fd, BlockDriverState *bs, char *buf, |
84390bed | 1297 | uint64_t oid, uint8_t copies, |
0e7106d8 LY |
1298 | unsigned int datalen, uint64_t offset, |
1299 | uint32_t cache_flags) | |
33b1db1c | 1300 | { |
f11672db | 1301 | return read_write_object(fd, bs, buf, oid, copies, |
84390bed | 1302 | datalen, offset, false, |
0e7106d8 | 1303 | false, cache_flags); |
33b1db1c MK |
1304 | } |
1305 | ||
f11672db | 1306 | static int write_object(int fd, BlockDriverState *bs, char *buf, |
84390bed | 1307 | uint64_t oid, uint8_t copies, |
2f536801 | 1308 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1309 | uint32_t cache_flags) |
33b1db1c | 1310 | { |
f11672db | 1311 | return read_write_object(fd, bs, buf, oid, copies, |
84390bed | 1312 | datalen, offset, true, |
0e7106d8 | 1313 | create, cache_flags); |
33b1db1c MK |
1314 | } |
1315 | ||
9ff53a0e MK |
1316 | /* update inode with the latest state */ |
1317 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag) | |
1318 | { | |
dfb12bf8 | 1319 | Error *local_err = NULL; |
9ff53a0e MK |
1320 | SheepdogInode *inode; |
1321 | int ret = 0, fd; | |
1322 | uint32_t vid = 0; | |
1323 | ||
dfb12bf8 | 1324 | fd = connect_to_sdog(s, &local_err); |
9ff53a0e | 1325 | if (fd < 0) { |
565f65d2 | 1326 | error_report_err(local_err); |
9ff53a0e MK |
1327 | return -EIO; |
1328 | } | |
1329 | ||
5d039bab | 1330 | inode = g_malloc(SD_INODE_HEADER_SIZE); |
9ff53a0e | 1331 | |
dc83cd42 | 1332 | ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err); |
9ff53a0e | 1333 | if (ret) { |
565f65d2 | 1334 | error_report_err(local_err); |
9ff53a0e MK |
1335 | goto out; |
1336 | } | |
1337 | ||
f11672db | 1338 | ret = read_object(fd, s->bs, (char *)inode, vid_to_vdi_oid(vid), |
5d039bab HM |
1339 | s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0, |
1340 | s->cache_flags); | |
9ff53a0e MK |
1341 | if (ret < 0) { |
1342 | goto out; | |
1343 | } | |
1344 | ||
1345 | if (inode->vdi_id != s->inode.vdi_id) { | |
5d039bab | 1346 | memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE); |
9ff53a0e MK |
1347 | } |
1348 | ||
1349 | out: | |
1350 | g_free(inode); | |
1351 | closesocket(fd); | |
1352 | ||
1353 | return ret; | |
1354 | } | |
1355 | ||
a37dcdf9 | 1356 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req) |
13c31de2 MK |
1357 | { |
1358 | SheepdogAIOCB *acb = aio_req->aiocb; | |
b544c1ab HM |
1359 | |
1360 | aio_req->create = false; | |
13c31de2 MK |
1361 | |
1362 | /* check whether this request becomes a CoW one */ | |
2412aec7 | 1363 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) { |
13c31de2 | 1364 | int idx = data_oid_to_idx(aio_req->oid); |
13c31de2 | 1365 | |
13c31de2 MK |
1366 | if (is_data_obj_writable(&s->inode, idx)) { |
1367 | goto out; | |
1368 | } | |
1369 | ||
80308d33 MK |
1370 | if (s->inode.data_vdi_id[idx]) { |
1371 | aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx); | |
1372 | aio_req->flags |= SD_FLAG_CMD_COW; | |
1373 | } | |
b544c1ab | 1374 | aio_req->create = true; |
13c31de2 MK |
1375 | } |
1376 | out: | |
2412aec7 | 1377 | if (is_data_obj(aio_req->oid)) { |
b544c1ab | 1378 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 1379 | acb->aiocb_type); |
2412aec7 MK |
1380 | } else { |
1381 | struct iovec iov; | |
1382 | iov.iov_base = &s->inode; | |
1383 | iov.iov_len = sizeof(s->inode); | |
b544c1ab | 1384 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
2412aec7 | 1385 | } |
13c31de2 MK |
1386 | } |
1387 | ||
84390bed SH |
1388 | static void sd_detach_aio_context(BlockDriverState *bs) |
1389 | { | |
1390 | BDRVSheepdogState *s = bs->opaque; | |
1391 | ||
dca21ef2 | 1392 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, |
f6a51c84 | 1393 | NULL, NULL, NULL); |
84390bed SH |
1394 | } |
1395 | ||
1396 | static void sd_attach_aio_context(BlockDriverState *bs, | |
1397 | AioContext *new_context) | |
1398 | { | |
1399 | BDRVSheepdogState *s = bs->opaque; | |
1400 | ||
1401 | s->aio_context = new_context; | |
dca21ef2 | 1402 | aio_set_fd_handler(new_context, s->fd, false, |
f6a51c84 | 1403 | co_read_response, NULL, NULL, s); |
84390bed SH |
1404 | } |
1405 | ||
c8c96350 KW |
1406 | /* TODO Convert to fine grained options */ |
1407 | static QemuOptsList runtime_opts = { | |
1408 | .name = "sheepdog", | |
1409 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1410 | .desc = { | |
1411 | { | |
1412 | .name = "filename", | |
1413 | .type = QEMU_OPT_STRING, | |
1414 | .help = "URL to the sheepdog image", | |
1415 | }, | |
1416 | { /* end of list */ } | |
1417 | }, | |
1418 | }; | |
1419 | ||
015a1036 HR |
1420 | static int sd_open(BlockDriverState *bs, QDict *options, int flags, |
1421 | Error **errp) | |
33b1db1c MK |
1422 | { |
1423 | int ret, fd; | |
1424 | uint32_t vid = 0; | |
1425 | BDRVSheepdogState *s = bs->opaque; | |
1426 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1427 | uint32_t snapid; | |
1428 | char *buf = NULL; | |
c8c96350 KW |
1429 | QemuOpts *opts; |
1430 | Error *local_err = NULL; | |
1431 | const char *filename; | |
1432 | ||
011603ca | 1433 | s->bs = bs; |
84390bed | 1434 | s->aio_context = bdrv_get_aio_context(bs); |
011603ca | 1435 | |
87ea75d5 | 1436 | opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); |
c8c96350 | 1437 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 1438 | if (local_err) { |
e67c3993 | 1439 | error_propagate(errp, local_err); |
c8c96350 | 1440 | ret = -EINVAL; |
cbc488ee | 1441 | goto err_no_fd; |
c8c96350 KW |
1442 | } |
1443 | ||
1444 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1445 | |
c292ee6a | 1446 | QLIST_INIT(&s->inflight_aio_head); |
011603ca | 1447 | QLIST_INIT(&s->failed_aio_head); |
6a55c82c | 1448 | QLIST_INIT(&s->inflight_aiocb_head); |
33b1db1c MK |
1449 | s->fd = -1; |
1450 | ||
1451 | memset(vdi, 0, sizeof(vdi)); | |
1452 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1453 | |
1454 | if (strstr(filename, "://")) { | |
1455 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1456 | } else { | |
1457 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1458 | } | |
1459 | if (ret < 0) { | |
efde4b62 | 1460 | error_setg(errp, "Can't parse filename"); |
cbc488ee | 1461 | goto err_no_fd; |
33b1db1c | 1462 | } |
e67c3993 | 1463 | s->fd = get_sheep_fd(s, errp); |
33b1db1c | 1464 | if (s->fd < 0) { |
cb595887 | 1465 | ret = s->fd; |
cbc488ee | 1466 | goto err_no_fd; |
33b1db1c MK |
1467 | } |
1468 | ||
e67c3993 | 1469 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, true, errp); |
33b1db1c | 1470 | if (ret) { |
cbc488ee | 1471 | goto err; |
33b1db1c MK |
1472 | } |
1473 | ||
0e7106d8 LY |
1474 | /* |
1475 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1476 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1477 | */ | |
1478 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1479 | if (flags & BDRV_O_NOCACHE) { | |
1480 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1481 | } | |
cac8f4a6 | 1482 | s->discard_supported = true; |
0e7106d8 | 1483 | |
622b6057 | 1484 | if (snapid || tag[0] != '\0') { |
2440a2c3 | 1485 | DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1486 | s->is_snapshot = true; |
33b1db1c MK |
1487 | } |
1488 | ||
e67c3993 | 1489 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1490 | if (fd < 0) { |
cb595887 | 1491 | ret = fd; |
cbc488ee | 1492 | goto err; |
33b1db1c MK |
1493 | } |
1494 | ||
7267c094 | 1495 | buf = g_malloc(SD_INODE_SIZE); |
f11672db | 1496 | ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid), |
84390bed | 1497 | 0, SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1498 | |
1499 | closesocket(fd); | |
1500 | ||
1501 | if (ret) { | |
efde4b62 | 1502 | error_setg(errp, "Can't read snapshot inode"); |
cbc488ee | 1503 | goto err; |
33b1db1c MK |
1504 | } |
1505 | ||
1506 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
33b1db1c | 1507 | |
e8bfaa2f | 1508 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; |
3178e275 | 1509 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1510 | qemu_co_mutex_init(&s->lock); |
498f2140 | 1511 | qemu_co_queue_init(&s->overlapping_queue); |
c8c96350 | 1512 | qemu_opts_del(opts); |
7267c094 | 1513 | g_free(buf); |
33b1db1c | 1514 | return 0; |
cbc488ee MA |
1515 | |
1516 | err: | |
dca21ef2 | 1517 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, |
f6a51c84 | 1518 | false, NULL, NULL, NULL, NULL); |
cbc488ee MA |
1519 | closesocket(s->fd); |
1520 | err_no_fd: | |
c8c96350 | 1521 | qemu_opts_del(opts); |
7267c094 | 1522 | g_free(buf); |
cb595887 | 1523 | return ret; |
33b1db1c MK |
1524 | } |
1525 | ||
4da65c80 LY |
1526 | static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue, |
1527 | Error **errp) | |
1528 | { | |
1529 | BDRVSheepdogState *s = state->bs->opaque; | |
1530 | BDRVSheepdogReopenState *re_s; | |
1531 | int ret = 0; | |
1532 | ||
1533 | re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1); | |
1534 | ||
1535 | re_s->cache_flags = SD_FLAG_CMD_CACHE; | |
1536 | if (state->flags & BDRV_O_NOCACHE) { | |
1537 | re_s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1538 | } | |
1539 | ||
1540 | re_s->fd = get_sheep_fd(s, errp); | |
1541 | if (re_s->fd < 0) { | |
1542 | ret = re_s->fd; | |
1543 | return ret; | |
1544 | } | |
1545 | ||
1546 | return ret; | |
1547 | } | |
1548 | ||
1549 | static void sd_reopen_commit(BDRVReopenState *state) | |
1550 | { | |
1551 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1552 | BDRVSheepdogState *s = state->bs->opaque; | |
1553 | ||
1554 | if (s->fd) { | |
dca21ef2 | 1555 | aio_set_fd_handler(s->aio_context, s->fd, false, |
f6a51c84 | 1556 | NULL, NULL, NULL, NULL); |
4da65c80 LY |
1557 | closesocket(s->fd); |
1558 | } | |
1559 | ||
1560 | s->fd = re_s->fd; | |
1561 | s->cache_flags = re_s->cache_flags; | |
1562 | ||
1563 | g_free(state->opaque); | |
1564 | state->opaque = NULL; | |
1565 | ||
1566 | return; | |
1567 | } | |
1568 | ||
1569 | static void sd_reopen_abort(BDRVReopenState *state) | |
1570 | { | |
1571 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1572 | BDRVSheepdogState *s = state->bs->opaque; | |
1573 | ||
1574 | if (re_s == NULL) { | |
1575 | return; | |
1576 | } | |
1577 | ||
1578 | if (re_s->fd) { | |
dca21ef2 | 1579 | aio_set_fd_handler(s->aio_context, re_s->fd, false, |
f6a51c84 | 1580 | NULL, NULL, NULL, NULL); |
4da65c80 LY |
1581 | closesocket(re_s->fd); |
1582 | } | |
1583 | ||
1584 | g_free(state->opaque); | |
1585 | state->opaque = NULL; | |
1586 | ||
1587 | return; | |
1588 | } | |
1589 | ||
7d2d3e74 MA |
1590 | static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot, |
1591 | Error **errp) | |
33b1db1c MK |
1592 | { |
1593 | SheepdogVdiReq hdr; | |
1594 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1595 | int fd, ret; | |
1596 | unsigned int wlen, rlen = 0; | |
1597 | char buf[SD_MAX_VDI_LEN]; | |
1598 | ||
7d2d3e74 | 1599 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1600 | if (fd < 0) { |
cb595887 | 1601 | return fd; |
33b1db1c MK |
1602 | } |
1603 | ||
3178e275 JM |
1604 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1605 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1606 | */ | |
33b1db1c | 1607 | memset(buf, 0, sizeof(buf)); |
c31d482f | 1608 | pstrcpy(buf, sizeof(buf), s->name); |
33b1db1c MK |
1609 | |
1610 | memset(&hdr, 0, sizeof(hdr)); | |
1611 | hdr.opcode = SD_OP_NEW_VDI; | |
9f23fce7 | 1612 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1613 | |
1614 | wlen = SD_MAX_VDI_LEN; | |
1615 | ||
1616 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1617 | hdr.snapid = snapshot; | |
1618 | ||
1619 | hdr.data_length = wlen; | |
c31d482f LY |
1620 | hdr.vdi_size = s->inode.vdi_size; |
1621 | hdr.copy_policy = s->inode.copy_policy; | |
b3af018f | 1622 | hdr.copies = s->inode.nr_copies; |
876eb1b0 | 1623 | hdr.block_size_shift = s->inode.block_size_shift; |
33b1db1c | 1624 | |
f11672db | 1625 | ret = do_req(fd, NULL, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c MK |
1626 | |
1627 | closesocket(fd); | |
1628 | ||
1629 | if (ret) { | |
7d2d3e74 | 1630 | error_setg_errno(errp, -ret, "create failed"); |
cb595887 | 1631 | return ret; |
33b1db1c MK |
1632 | } |
1633 | ||
1634 | if (rsp->result != SD_RES_SUCCESS) { | |
7d2d3e74 | 1635 | error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name); |
33b1db1c MK |
1636 | return -EIO; |
1637 | } | |
1638 | ||
1639 | if (vdi_id) { | |
1640 | *vdi_id = rsp->vdi_id; | |
1641 | } | |
1642 | ||
1643 | return 0; | |
1644 | } | |
1645 | ||
318df29e | 1646 | static int sd_prealloc(const char *filename, Error **errp) |
a8e0fdd7 | 1647 | { |
fba98d45 | 1648 | BlockBackend *blk = NULL; |
876eb1b0 TI |
1649 | BDRVSheepdogState *base = NULL; |
1650 | unsigned long buf_size; | |
a8e0fdd7 | 1651 | uint32_t idx, max_idx; |
876eb1b0 | 1652 | uint32_t object_size; |
a8e0fdd7 | 1653 | int64_t vdi_size; |
876eb1b0 | 1654 | void *buf = NULL; |
a8e0fdd7 MK |
1655 | int ret; |
1656 | ||
efaa7c4e | 1657 | blk = blk_new_open(filename, NULL, NULL, |
55880601 | 1658 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); |
fba98d45 KW |
1659 | if (blk == NULL) { |
1660 | ret = -EIO; | |
318df29e | 1661 | goto out_with_err_set; |
a8e0fdd7 MK |
1662 | } |
1663 | ||
fba98d45 KW |
1664 | blk_set_allow_write_beyond_eof(blk, true); |
1665 | ||
1666 | vdi_size = blk_getlength(blk); | |
a8e0fdd7 MK |
1667 | if (vdi_size < 0) { |
1668 | ret = vdi_size; | |
1669 | goto out; | |
1670 | } | |
876eb1b0 | 1671 | |
fba98d45 | 1672 | base = blk_bs(blk)->opaque; |
876eb1b0 TI |
1673 | object_size = (UINT32_C(1) << base->inode.block_size_shift); |
1674 | buf_size = MIN(object_size, SD_DATA_OBJ_SIZE); | |
1675 | buf = g_malloc0(buf_size); | |
1676 | ||
1677 | max_idx = DIV_ROUND_UP(vdi_size, buf_size); | |
a8e0fdd7 MK |
1678 | |
1679 | for (idx = 0; idx < max_idx; idx++) { | |
1680 | /* | |
1681 | * The created image can be a cloned image, so we need to read | |
1682 | * a data from the source image. | |
1683 | */ | |
fba98d45 | 1684 | ret = blk_pread(blk, idx * buf_size, buf, buf_size); |
a8e0fdd7 MK |
1685 | if (ret < 0) { |
1686 | goto out; | |
1687 | } | |
8341f00d | 1688 | ret = blk_pwrite(blk, idx * buf_size, buf, buf_size, 0); |
a8e0fdd7 MK |
1689 | if (ret < 0) { |
1690 | goto out; | |
1691 | } | |
1692 | } | |
318df29e | 1693 | |
fba98d45 | 1694 | ret = 0; |
a8e0fdd7 | 1695 | out: |
318df29e MA |
1696 | if (ret < 0) { |
1697 | error_setg_errno(errp, -ret, "Can't pre-allocate"); | |
1698 | } | |
1699 | out_with_err_set: | |
fba98d45 KW |
1700 | if (blk) { |
1701 | blk_unref(blk); | |
a8e0fdd7 | 1702 | } |
7267c094 | 1703 | g_free(buf); |
a8e0fdd7 MK |
1704 | |
1705 | return ret; | |
1706 | } | |
1707 | ||
b3af018f LY |
1708 | /* |
1709 | * Sheepdog support two kinds of redundancy, full replication and erasure | |
1710 | * coding. | |
1711 | * | |
1712 | * # create a fully replicated vdi with x copies | |
1713 | * -o redundancy=x (1 <= x <= SD_MAX_COPIES) | |
1714 | * | |
1715 | * # create a erasure coded vdi with x data strips and y parity strips | |
1716 | * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP) | |
1717 | */ | |
1718 | static int parse_redundancy(BDRVSheepdogState *s, const char *opt) | |
1719 | { | |
1720 | struct SheepdogInode *inode = &s->inode; | |
1721 | const char *n1, *n2; | |
1722 | long copy, parity; | |
1723 | char p[10]; | |
1724 | ||
1725 | pstrcpy(p, sizeof(p), opt); | |
1726 | n1 = strtok(p, ":"); | |
1727 | n2 = strtok(NULL, ":"); | |
1728 | ||
1729 | if (!n1) { | |
1730 | return -EINVAL; | |
1731 | } | |
1732 | ||
1733 | copy = strtol(n1, NULL, 10); | |
89e2a31d | 1734 | /* FIXME fix error checking by switching to qemu_strtol() */ |
b3af018f LY |
1735 | if (copy > SD_MAX_COPIES || copy < 1) { |
1736 | return -EINVAL; | |
1737 | } | |
1738 | if (!n2) { | |
1739 | inode->copy_policy = 0; | |
1740 | inode->nr_copies = copy; | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | if (copy != 2 && copy != 4 && copy != 8 && copy != 16) { | |
1745 | return -EINVAL; | |
1746 | } | |
1747 | ||
1748 | parity = strtol(n2, NULL, 10); | |
89e2a31d | 1749 | /* FIXME fix error checking by switching to qemu_strtol() */ |
b3af018f LY |
1750 | if (parity >= SD_EC_MAX_STRIP || parity < 1) { |
1751 | return -EINVAL; | |
1752 | } | |
1753 | ||
1754 | /* | |
1755 | * 4 bits for parity and 4 bits for data. | |
1756 | * We have to compress upper data bits because it can't represent 16 | |
1757 | */ | |
1758 | inode->copy_policy = ((copy / 2) << 4) + parity; | |
1759 | inode->nr_copies = copy + parity; | |
1760 | ||
1761 | return 0; | |
1762 | } | |
1763 | ||
876eb1b0 TI |
1764 | static int parse_block_size_shift(BDRVSheepdogState *s, QemuOpts *opt) |
1765 | { | |
1766 | struct SheepdogInode *inode = &s->inode; | |
1767 | uint64_t object_size; | |
1768 | int obj_order; | |
1769 | ||
1770 | object_size = qemu_opt_get_size_del(opt, BLOCK_OPT_OBJECT_SIZE, 0); | |
1771 | if (object_size) { | |
1772 | if ((object_size - 1) & object_size) { /* not a power of 2? */ | |
1773 | return -EINVAL; | |
1774 | } | |
786a4ea8 | 1775 | obj_order = ctz32(object_size); |
876eb1b0 TI |
1776 | if (obj_order < 20 || obj_order > 31) { |
1777 | return -EINVAL; | |
1778 | } | |
1779 | inode->block_size_shift = (uint8_t)obj_order; | |
1780 | } | |
1781 | ||
1782 | return 0; | |
1783 | } | |
1784 | ||
b222237b | 1785 | static int sd_create(const char *filename, QemuOpts *opts, |
d5124c00 | 1786 | Error **errp) |
33b1db1c | 1787 | { |
b6fc8245 | 1788 | int ret = 0; |
c31d482f | 1789 | uint32_t vid = 0; |
33b1db1c | 1790 | char *backing_file = NULL; |
b222237b | 1791 | char *buf = NULL; |
b6fc8245 | 1792 | BDRVSheepdogState *s; |
c31d482f | 1793 | char tag[SD_MAX_VDI_TAG_LEN]; |
b4447363 | 1794 | uint32_t snapid; |
876eb1b0 | 1795 | uint64_t max_vdi_size; |
2f536801 | 1796 | bool prealloc = false; |
33b1db1c | 1797 | |
5839e53b | 1798 | s = g_new0(BDRVSheepdogState, 1); |
b6fc8245 | 1799 | |
b4447363 | 1800 | memset(tag, 0, sizeof(tag)); |
5d6768e3 | 1801 | if (strstr(filename, "://")) { |
c31d482f | 1802 | ret = sd_parse_uri(s, filename, s->name, &snapid, tag); |
5d6768e3 | 1803 | } else { |
c31d482f | 1804 | ret = parse_vdiname(s, filename, s->name, &snapid, tag); |
5d6768e3 MK |
1805 | } |
1806 | if (ret < 0) { | |
efde4b62 | 1807 | error_setg(errp, "Can't parse filename"); |
b6fc8245 | 1808 | goto out; |
b4447363 MK |
1809 | } |
1810 | ||
c2eb918e HT |
1811 | s->inode.vdi_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
1812 | BDRV_SECTOR_SIZE); | |
b222237b CL |
1813 | backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); |
1814 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
1815 | if (!buf || !strcmp(buf, "off")) { | |
1816 | prealloc = false; | |
1817 | } else if (!strcmp(buf, "full")) { | |
1818 | prealloc = true; | |
1819 | } else { | |
1820 | error_setg(errp, "Invalid preallocation mode: '%s'", buf); | |
1821 | ret = -EINVAL; | |
1822 | goto out; | |
1823 | } | |
1824 | ||
1825 | g_free(buf); | |
1826 | buf = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY); | |
1827 | if (buf) { | |
1828 | ret = parse_redundancy(s, buf); | |
1829 | if (ret < 0) { | |
1830 | error_setg(errp, "Invalid redundancy mode: '%s'", buf); | |
1831 | goto out; | |
33b1db1c | 1832 | } |
33b1db1c | 1833 | } |
876eb1b0 TI |
1834 | ret = parse_block_size_shift(s, opts); |
1835 | if (ret < 0) { | |
1836 | error_setg(errp, "Invalid object_size." | |
1837 | " obect_size needs to be power of 2" | |
1838 | " and be limited from 2^20 to 2^31"); | |
b6fc8245 | 1839 | goto out; |
33b1db1c MK |
1840 | } |
1841 | ||
1842 | if (backing_file) { | |
fba98d45 | 1843 | BlockBackend *blk; |
9f23fce7 | 1844 | BDRVSheepdogState *base; |
33b1db1c MK |
1845 | BlockDriver *drv; |
1846 | ||
1847 | /* Currently, only Sheepdog backing image is supported. */ | |
b65a5e12 | 1848 | drv = bdrv_find_protocol(backing_file, true, NULL); |
33b1db1c | 1849 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
e67c3993 | 1850 | error_setg(errp, "backing_file must be a sheepdog image"); |
b6fc8245 MK |
1851 | ret = -EINVAL; |
1852 | goto out; | |
33b1db1c MK |
1853 | } |
1854 | ||
efaa7c4e | 1855 | blk = blk_new_open(backing_file, NULL, NULL, |
72e775c7 | 1856 | BDRV_O_PROTOCOL, errp); |
fba98d45 KW |
1857 | if (blk == NULL) { |
1858 | ret = -EIO; | |
b6fc8245 | 1859 | goto out; |
cb595887 | 1860 | } |
33b1db1c | 1861 | |
fba98d45 | 1862 | base = blk_bs(blk)->opaque; |
33b1db1c | 1863 | |
9f23fce7 | 1864 | if (!is_snapshot(&base->inode)) { |
e67c3993 | 1865 | error_setg(errp, "cannot clone from a non snapshot vdi"); |
fba98d45 | 1866 | blk_unref(blk); |
b6fc8245 MK |
1867 | ret = -EINVAL; |
1868 | goto out; | |
33b1db1c | 1869 | } |
9f23fce7 | 1870 | s->inode.vdi_id = base->inode.vdi_id; |
fba98d45 | 1871 | blk_unref(blk); |
33b1db1c MK |
1872 | } |
1873 | ||
5d5da114 | 1874 | s->aio_context = qemu_get_aio_context(); |
876eb1b0 TI |
1875 | |
1876 | /* if block_size_shift is not specified, get cluster default value */ | |
1877 | if (s->inode.block_size_shift == 0) { | |
1878 | SheepdogVdiReq hdr; | |
1879 | SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr; | |
876eb1b0 TI |
1880 | int fd; |
1881 | unsigned int wlen = 0, rlen = 0; | |
1882 | ||
48d7c4af | 1883 | fd = connect_to_sdog(s, errp); |
876eb1b0 | 1884 | if (fd < 0) { |
48d7c4af | 1885 | ret = fd; |
876eb1b0 TI |
1886 | goto out; |
1887 | } | |
1888 | ||
1889 | memset(&hdr, 0, sizeof(hdr)); | |
1890 | hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT; | |
1891 | hdr.proto_ver = SD_PROTO_VER; | |
1892 | ||
f11672db | 1893 | ret = do_req(fd, NULL, (SheepdogReq *)&hdr, |
876eb1b0 TI |
1894 | NULL, &wlen, &rlen); |
1895 | closesocket(fd); | |
1896 | if (ret) { | |
1897 | error_setg_errno(errp, -ret, "failed to get cluster default"); | |
1898 | goto out; | |
1899 | } | |
1900 | if (rsp->result == SD_RES_SUCCESS) { | |
1901 | s->inode.block_size_shift = rsp->block_size_shift; | |
1902 | } else { | |
1903 | s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT; | |
1904 | } | |
1905 | } | |
1906 | ||
1907 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; | |
1908 | ||
1909 | if (s->inode.vdi_size > max_vdi_size) { | |
1910 | error_setg(errp, "An image is too large." | |
1911 | " The maximum image size is %"PRIu64 "GB", | |
1912 | max_vdi_size / 1024 / 1024 / 1024); | |
1913 | ret = -EINVAL; | |
1914 | goto out; | |
1915 | } | |
1916 | ||
e67c3993 | 1917 | ret = do_sd_create(s, &vid, 0, errp); |
7d2d3e74 | 1918 | if (ret) { |
b6fc8245 | 1919 | goto out; |
a8e0fdd7 MK |
1920 | } |
1921 | ||
7d2d3e74 | 1922 | if (prealloc) { |
e67c3993 | 1923 | ret = sd_prealloc(filename, errp); |
318df29e | 1924 | } |
b6fc8245 | 1925 | out: |
b222237b CL |
1926 | g_free(backing_file); |
1927 | g_free(buf); | |
b6fc8245 MK |
1928 | g_free(s); |
1929 | return ret; | |
33b1db1c MK |
1930 | } |
1931 | ||
1932 | static void sd_close(BlockDriverState *bs) | |
1933 | { | |
dfb12bf8 | 1934 | Error *local_err = NULL; |
33b1db1c MK |
1935 | BDRVSheepdogState *s = bs->opaque; |
1936 | SheepdogVdiReq hdr; | |
1937 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1938 | unsigned int wlen, rlen = 0; | |
1939 | int fd, ret; | |
1940 | ||
2440a2c3 | 1941 | DPRINTF("%s\n", s->name); |
33b1db1c | 1942 | |
dfb12bf8 | 1943 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1944 | if (fd < 0) { |
565f65d2 | 1945 | error_report_err(local_err); |
33b1db1c MK |
1946 | return; |
1947 | } | |
1948 | ||
1949 | memset(&hdr, 0, sizeof(hdr)); | |
1950 | ||
1951 | hdr.opcode = SD_OP_RELEASE_VDI; | |
1dbfafed | 1952 | hdr.type = LOCK_TYPE_NORMAL; |
9f23fce7 | 1953 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1954 | wlen = strlen(s->name) + 1; |
1955 | hdr.data_length = wlen; | |
1956 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1957 | ||
f11672db | 1958 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, |
84390bed | 1959 | s->name, &wlen, &rlen); |
33b1db1c MK |
1960 | |
1961 | closesocket(fd); | |
1962 | ||
1963 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1964 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1965 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1966 | } |
1967 | ||
dca21ef2 | 1968 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, |
f6a51c84 | 1969 | false, NULL, NULL, NULL, NULL); |
33b1db1c | 1970 | closesocket(s->fd); |
25af257d | 1971 | g_free(s->host_spec); |
33b1db1c MK |
1972 | } |
1973 | ||
1974 | static int64_t sd_getlength(BlockDriverState *bs) | |
1975 | { | |
1976 | BDRVSheepdogState *s = bs->opaque; | |
1977 | ||
1978 | return s->inode.vdi_size; | |
1979 | } | |
1980 | ||
1981 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1982 | { | |
dfb12bf8 | 1983 | Error *local_err = NULL; |
33b1db1c MK |
1984 | BDRVSheepdogState *s = bs->opaque; |
1985 | int ret, fd; | |
1986 | unsigned int datalen; | |
876eb1b0 | 1987 | uint64_t max_vdi_size; |
33b1db1c | 1988 | |
876eb1b0 | 1989 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; |
33b1db1c | 1990 | if (offset < s->inode.vdi_size) { |
6daf194d | 1991 | error_report("shrinking is not supported"); |
33b1db1c | 1992 | return -EINVAL; |
876eb1b0 | 1993 | } else if (offset > max_vdi_size) { |
6daf194d | 1994 | error_report("too big image size"); |
33b1db1c MK |
1995 | return -EINVAL; |
1996 | } | |
1997 | ||
dfb12bf8 | 1998 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1999 | if (fd < 0) { |
565f65d2 | 2000 | error_report_err(local_err); |
cb595887 | 2001 | return fd; |
33b1db1c MK |
2002 | } |
2003 | ||
2004 | /* we don't need to update entire object */ | |
2005 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
2006 | s->inode.vdi_size = offset; | |
f11672db | 2007 | ret = write_object(fd, s->bs, (char *)&s->inode, |
84390bed SH |
2008 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, |
2009 | datalen, 0, false, s->cache_flags); | |
33b1db1c MK |
2010 | close(fd); |
2011 | ||
2012 | if (ret < 0) { | |
6daf194d | 2013 | error_report("failed to update an inode."); |
33b1db1c MK |
2014 | } |
2015 | ||
cb595887 | 2016 | return ret; |
33b1db1c MK |
2017 | } |
2018 | ||
2019 | /* | |
2020 | * This function is called after writing data objects. If we need to | |
2021 | * update metadata, this sends a write request to the vdi object. | |
33b1db1c | 2022 | */ |
d8716b41 | 2023 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c | 2024 | { |
28ddd08c | 2025 | BDRVSheepdogState *s = acb->s; |
33b1db1c MK |
2026 | struct iovec iov; |
2027 | AIOReq *aio_req; | |
2028 | uint32_t offset, data_len, mn, mx; | |
2029 | ||
498f2140 HM |
2030 | mn = acb->min_dirty_data_idx; |
2031 | mx = acb->max_dirty_data_idx; | |
33b1db1c MK |
2032 | if (mn <= mx) { |
2033 | /* we need to update the vdi object. */ | |
e80ab33d | 2034 | ++acb->nr_pending; |
33b1db1c MK |
2035 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + |
2036 | mn * sizeof(s->inode.data_vdi_id[0]); | |
2037 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
2038 | ||
498f2140 HM |
2039 | acb->min_dirty_data_idx = UINT32_MAX; |
2040 | acb->max_dirty_data_idx = 0; | |
33b1db1c MK |
2041 | |
2042 | iov.iov_base = &s->inode; | |
2043 | iov.iov_len = sizeof(s->inode); | |
2044 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
b544c1ab | 2045 | data_len, offset, 0, false, 0, offset); |
b544c1ab | 2046 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
e80ab33d PB |
2047 | if (--acb->nr_pending) { |
2048 | qemu_coroutine_yield(); | |
2049 | } | |
33b1db1c | 2050 | } |
33b1db1c MK |
2051 | } |
2052 | ||
859e5553 LY |
2053 | /* Delete current working VDI on the snapshot chain */ |
2054 | static bool sd_delete(BDRVSheepdogState *s) | |
2055 | { | |
dfb12bf8 | 2056 | Error *local_err = NULL; |
859e5553 LY |
2057 | unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0; |
2058 | SheepdogVdiReq hdr = { | |
2059 | .opcode = SD_OP_DEL_VDI, | |
9f23fce7 | 2060 | .base_vdi_id = s->inode.vdi_id, |
859e5553 LY |
2061 | .data_length = wlen, |
2062 | .flags = SD_FLAG_CMD_WRITE, | |
2063 | }; | |
2064 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2065 | int fd, ret; | |
2066 | ||
dfb12bf8 | 2067 | fd = connect_to_sdog(s, &local_err); |
859e5553 | 2068 | if (fd < 0) { |
565f65d2 | 2069 | error_report_err(local_err); |
859e5553 LY |
2070 | return false; |
2071 | } | |
2072 | ||
f11672db | 2073 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, |
84390bed | 2074 | s->name, &wlen, &rlen); |
859e5553 LY |
2075 | closesocket(fd); |
2076 | if (ret) { | |
2077 | return false; | |
2078 | } | |
2079 | switch (rsp->result) { | |
2080 | case SD_RES_NO_VDI: | |
2081 | error_report("%s was already deleted", s->name); | |
2082 | /* fall through */ | |
2083 | case SD_RES_SUCCESS: | |
2084 | break; | |
2085 | default: | |
2086 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2087 | return false; | |
2088 | } | |
2089 | ||
2090 | return true; | |
2091 | } | |
2092 | ||
33b1db1c MK |
2093 | /* |
2094 | * Create a writable VDI from a snapshot | |
2095 | */ | |
2096 | static int sd_create_branch(BDRVSheepdogState *s) | |
2097 | { | |
dfb12bf8 | 2098 | Error *local_err = NULL; |
33b1db1c MK |
2099 | int ret, fd; |
2100 | uint32_t vid; | |
2101 | char *buf; | |
859e5553 | 2102 | bool deleted; |
33b1db1c | 2103 | |
2440a2c3 | 2104 | DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
33b1db1c | 2105 | |
7267c094 | 2106 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 2107 | |
859e5553 LY |
2108 | /* |
2109 | * Even If deletion fails, we will just create extra snapshot based on | |
dc6fb73d | 2110 | * the working VDI which was supposed to be deleted. So no need to |
859e5553 LY |
2111 | * false bail out. |
2112 | */ | |
2113 | deleted = sd_delete(s); | |
7d2d3e74 | 2114 | ret = do_sd_create(s, &vid, !deleted, &local_err); |
33b1db1c | 2115 | if (ret) { |
565f65d2 | 2116 | error_report_err(local_err); |
33b1db1c MK |
2117 | goto out; |
2118 | } | |
2119 | ||
2440a2c3 | 2120 | DPRINTF("%" PRIx32 " is created.\n", vid); |
33b1db1c | 2121 | |
dfb12bf8 | 2122 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2123 | if (fd < 0) { |
565f65d2 | 2124 | error_report_err(local_err); |
cb595887 | 2125 | ret = fd; |
33b1db1c MK |
2126 | goto out; |
2127 | } | |
2128 | ||
f11672db | 2129 | ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid), |
84390bed | 2130 | s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
2131 | |
2132 | closesocket(fd); | |
2133 | ||
2134 | if (ret < 0) { | |
2135 | goto out; | |
2136 | } | |
2137 | ||
2138 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
2139 | ||
2f536801 | 2140 | s->is_snapshot = false; |
33b1db1c | 2141 | ret = 0; |
2440a2c3 | 2142 | DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
33b1db1c MK |
2143 | |
2144 | out: | |
7267c094 | 2145 | g_free(buf); |
33b1db1c MK |
2146 | |
2147 | return ret; | |
2148 | } | |
2149 | ||
2150 | /* | |
2151 | * Send I/O requests to the server. | |
2152 | * | |
2153 | * This function sends requests to the server, links the requests to | |
c292ee6a | 2154 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
2155 | * waiting the response. The responses are received in the |
2156 | * `aio_read_response' function which is called from the main loop as | |
2157 | * a fd handler. | |
2df46246 MK |
2158 | * |
2159 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
2160 | * request and -errno in error cases. | |
33b1db1c | 2161 | */ |
28ddd08c | 2162 | static void coroutine_fn sd_co_rw_vector(SheepdogAIOCB *acb) |
33b1db1c | 2163 | { |
33b1db1c | 2164 | int ret = 0; |
e8bfaa2f | 2165 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; |
876eb1b0 TI |
2166 | unsigned long idx; |
2167 | uint32_t object_size; | |
33b1db1c | 2168 | uint64_t oid; |
876eb1b0 | 2169 | uint64_t offset; |
28ddd08c | 2170 | BDRVSheepdogState *s = acb->s; |
33b1db1c MK |
2171 | SheepdogInode *inode = &s->inode; |
2172 | AIOReq *aio_req; | |
2173 | ||
33b1db1c MK |
2174 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
2175 | /* | |
2176 | * In the case we open the snapshot VDI, Sheepdog creates the | |
2177 | * writable VDI when we do a write operation first. | |
2178 | */ | |
2179 | ret = sd_create_branch(s); | |
2180 | if (ret) { | |
2181 | acb->ret = -EIO; | |
e80ab33d | 2182 | return; |
33b1db1c MK |
2183 | } |
2184 | } | |
2185 | ||
876eb1b0 TI |
2186 | object_size = (UINT32_C(1) << inode->block_size_shift); |
2187 | idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
2188 | offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size; | |
2189 | ||
1d732d7d MK |
2190 | /* |
2191 | * Make sure we don't free the aiocb before we are done with all requests. | |
2192 | * This additional reference is dropped at the end of this function. | |
2193 | */ | |
2194 | acb->nr_pending++; | |
2195 | ||
33b1db1c MK |
2196 | while (done != total) { |
2197 | uint8_t flags = 0; | |
2198 | uint64_t old_oid = 0; | |
2f536801 | 2199 | bool create = false; |
33b1db1c MK |
2200 | |
2201 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
2202 | ||
876eb1b0 | 2203 | len = MIN(total - done, object_size - offset); |
33b1db1c | 2204 | |
19db9b90 CH |
2205 | switch (acb->aiocb_type) { |
2206 | case AIOCB_READ_UDATA: | |
2207 | if (!inode->data_vdi_id[idx]) { | |
2208 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
2209 | goto done; |
2210 | } | |
19db9b90 CH |
2211 | break; |
2212 | case AIOCB_WRITE_UDATA: | |
2213 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 2214 | create = true; |
19db9b90 CH |
2215 | } else if (!is_data_obj_writable(inode, idx)) { |
2216 | /* Copy-On-Write */ | |
2f536801 | 2217 | create = true; |
19db9b90 CH |
2218 | old_oid = oid; |
2219 | flags = SD_FLAG_CMD_COW; | |
2220 | } | |
2221 | break; | |
cac8f4a6 LY |
2222 | case AIOCB_DISCARD_OBJ: |
2223 | /* | |
2224 | * We discard the object only when the whole object is | |
2225 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
2226 | */ | |
876eb1b0 | 2227 | if (len != object_size || inode->data_vdi_id[idx] == 0) { |
cac8f4a6 LY |
2228 | goto done; |
2229 | } | |
2230 | break; | |
19db9b90 CH |
2231 | default: |
2232 | break; | |
33b1db1c MK |
2233 | } |
2234 | ||
2235 | if (create) { | |
2440a2c3 | 2236 | DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1b6ac998 | 2237 | inode->vdi_id, oid, |
33b1db1c MK |
2238 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
2239 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
2440a2c3 | 2240 | DPRINTF("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
2241 | } |
2242 | ||
b544c1ab | 2243 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create, |
e6fd57ea HM |
2244 | old_oid, |
2245 | acb->aiocb_type == AIOCB_DISCARD_OBJ ? | |
2246 | 0 : done); | |
b544c1ab | 2247 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 2248 | acb->aiocb_type); |
33b1db1c MK |
2249 | done: |
2250 | offset = 0; | |
2251 | idx++; | |
2252 | done += len; | |
2253 | } | |
e80ab33d PB |
2254 | if (--acb->nr_pending) { |
2255 | qemu_coroutine_yield(); | |
33b1db1c MK |
2256 | } |
2257 | } | |
2258 | ||
acf6e5f0 | 2259 | static void sd_aio_complete(SheepdogAIOCB *acb) |
6a55c82c | 2260 | { |
acf6e5f0 PB |
2261 | if (acb->aiocb_type == AIOCB_FLUSH_CACHE) { |
2262 | return; | |
6a55c82c HM |
2263 | } |
2264 | ||
acf6e5f0 PB |
2265 | QLIST_REMOVE(acb, aiocb_siblings); |
2266 | qemu_co_queue_restart_all(&acb->s->overlapping_queue); | |
6a55c82c HM |
2267 | } |
2268 | ||
a968168c | 2269 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2270 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c | 2271 | { |
28ddd08c | 2272 | SheepdogAIOCB acb; |
2df46246 | 2273 | int ret; |
e50d7607 LY |
2274 | int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE; |
2275 | BDRVSheepdogState *s = bs->opaque; | |
33b1db1c | 2276 | |
c0191e76 | 2277 | if (offset > s->inode.vdi_size) { |
e50d7607 | 2278 | ret = sd_truncate(bs, offset); |
cb595887 MK |
2279 | if (ret < 0) { |
2280 | return ret; | |
33b1db1c | 2281 | } |
33b1db1c MK |
2282 | } |
2283 | ||
28ddd08c | 2284 | sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_WRITE_UDATA); |
28ddd08c PB |
2285 | sd_co_rw_vector(&acb); |
2286 | sd_write_done(&acb); | |
acf6e5f0 | 2287 | sd_aio_complete(&acb); |
2df46246 | 2288 | |
28ddd08c | 2289 | return acb.ret; |
33b1db1c MK |
2290 | } |
2291 | ||
a968168c | 2292 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2293 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c | 2294 | { |
28ddd08c | 2295 | SheepdogAIOCB acb; |
6a55c82c | 2296 | BDRVSheepdogState *s = bs->opaque; |
33b1db1c | 2297 | |
28ddd08c | 2298 | sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_READ_UDATA); |
28ddd08c | 2299 | sd_co_rw_vector(&acb); |
acf6e5f0 | 2300 | sd_aio_complete(&acb); |
2df46246 | 2301 | |
28ddd08c | 2302 | return acb.ret; |
33b1db1c MK |
2303 | } |
2304 | ||
47622c44 LY |
2305 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
2306 | { | |
2307 | BDRVSheepdogState *s = bs->opaque; | |
28ddd08c | 2308 | SheepdogAIOCB acb; |
47783072 | 2309 | AIOReq *aio_req; |
47622c44 | 2310 | |
0e7106d8 | 2311 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
2312 | return 0; |
2313 | } | |
2314 | ||
28ddd08c | 2315 | sd_aio_setup(&acb, s, NULL, 0, 0, AIOCB_FLUSH_CACHE); |
47622c44 | 2316 | |
28ddd08c PB |
2317 | acb.nr_pending++; |
2318 | aio_req = alloc_aio_req(s, &acb, vid_to_vdi_oid(s->inode.vdi_id), | |
b544c1ab | 2319 | 0, 0, 0, false, 0, 0); |
28ddd08c | 2320 | add_aio_request(s, aio_req, NULL, 0, acb.aiocb_type); |
47622c44 | 2321 | |
28ddd08c | 2322 | if (--acb.nr_pending) { |
e80ab33d PB |
2323 | qemu_coroutine_yield(); |
2324 | } | |
acf6e5f0 PB |
2325 | |
2326 | sd_aio_complete(&acb); | |
28ddd08c | 2327 | return acb.ret; |
47622c44 LY |
2328 | } |
2329 | ||
33b1db1c MK |
2330 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
2331 | { | |
dfb12bf8 | 2332 | Error *local_err = NULL; |
33b1db1c MK |
2333 | BDRVSheepdogState *s = bs->opaque; |
2334 | int ret, fd; | |
2335 | uint32_t new_vid; | |
2336 | SheepdogInode *inode; | |
2337 | unsigned int datalen; | |
2338 | ||
2440a2c3 | 2339 | DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
2340 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
2341 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
2342 | ||
2343 | if (s->is_snapshot) { | |
2344 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 2345 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
2346 | |
2347 | return -EINVAL; | |
2348 | } | |
2349 | ||
2440a2c3 | 2350 | DPRINTF("%s %s\n", sn_info->name, sn_info->id_str); |
33b1db1c MK |
2351 | |
2352 | s->inode.vm_state_size = sn_info->vm_state_size; | |
2353 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
2354 | /* It appears that inode.tag does not require a NUL terminator, |
2355 | * which means this use of strncpy is ok. | |
2356 | */ | |
33b1db1c MK |
2357 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
2358 | /* we don't need to update entire object */ | |
2359 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
2df5fee2 | 2360 | inode = g_malloc(datalen); |
33b1db1c MK |
2361 | |
2362 | /* refresh inode. */ | |
dfb12bf8 | 2363 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2364 | if (fd < 0) { |
565f65d2 | 2365 | error_report_err(local_err); |
cb595887 | 2366 | ret = fd; |
33b1db1c MK |
2367 | goto cleanup; |
2368 | } | |
2369 | ||
f11672db | 2370 | ret = write_object(fd, s->bs, (char *)&s->inode, |
84390bed SH |
2371 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, |
2372 | datalen, 0, false, s->cache_flags); | |
33b1db1c | 2373 | if (ret < 0) { |
6daf194d | 2374 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
2375 | goto cleanup; |
2376 | } | |
2377 | ||
7d2d3e74 | 2378 | ret = do_sd_create(s, &new_vid, 1, &local_err); |
33b1db1c | 2379 | if (ret < 0) { |
c29b77f9 MA |
2380 | error_reportf_err(local_err, |
2381 | "failed to create inode for snapshot: "); | |
33b1db1c MK |
2382 | goto cleanup; |
2383 | } | |
2384 | ||
f11672db | 2385 | ret = read_object(fd, s->bs, (char *)inode, |
84390bed SH |
2386 | vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0, |
2387 | s->cache_flags); | |
33b1db1c MK |
2388 | |
2389 | if (ret < 0) { | |
6daf194d | 2390 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
2391 | goto cleanup; |
2392 | } | |
2393 | ||
2394 | memcpy(&s->inode, inode, datalen); | |
2440a2c3 | 2395 | DPRINTF("s->inode: name %s snap_id %x oid %x\n", |
33b1db1c MK |
2396 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
2397 | ||
2398 | cleanup: | |
2df5fee2 | 2399 | g_free(inode); |
33b1db1c MK |
2400 | closesocket(fd); |
2401 | return ret; | |
2402 | } | |
2403 | ||
859e5553 LY |
2404 | /* |
2405 | * We implement rollback(loadvm) operation to the specified snapshot by | |
2406 | * 1) switch to the snapshot | |
2407 | * 2) rely on sd_create_branch to delete working VDI and | |
dc6fb73d | 2408 | * 3) create a new working VDI based on the specified snapshot |
859e5553 | 2409 | */ |
33b1db1c MK |
2410 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
2411 | { | |
2412 | BDRVSheepdogState *s = bs->opaque; | |
2413 | BDRVSheepdogState *old_s; | |
9ff53a0e | 2414 | char tag[SD_MAX_VDI_TAG_LEN]; |
33b1db1c | 2415 | uint32_t snapid = 0; |
89e2a31d MA |
2416 | int ret; |
2417 | ||
2418 | if (!sd_parse_snapid_or_tag(snapshot_id, &snapid, tag)) { | |
2419 | return -EINVAL; | |
2420 | } | |
33b1db1c | 2421 | |
5839e53b | 2422 | old_s = g_new(BDRVSheepdogState, 1); |
33b1db1c MK |
2423 | |
2424 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
2425 | ||
9ff53a0e | 2426 | ret = reload_inode(s, snapid, tag); |
33b1db1c | 2427 | if (ret) { |
33b1db1c MK |
2428 | goto out; |
2429 | } | |
2430 | ||
cede621f LY |
2431 | ret = sd_create_branch(s); |
2432 | if (ret) { | |
33b1db1c MK |
2433 | goto out; |
2434 | } | |
2435 | ||
7267c094 | 2436 | g_free(old_s); |
33b1db1c MK |
2437 | |
2438 | return 0; | |
2439 | out: | |
2440 | /* recover bdrv_sd_state */ | |
2441 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 | 2442 | g_free(old_s); |
33b1db1c | 2443 | |
6daf194d | 2444 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
2445 | |
2446 | return ret; | |
2447 | } | |
2448 | ||
eab8eb8d VT |
2449 | #define NR_BATCHED_DISCARD 128 |
2450 | ||
e25cad69 | 2451 | static int remove_objects(BDRVSheepdogState *s, Error **errp) |
eab8eb8d VT |
2452 | { |
2453 | int fd, i = 0, nr_objs = 0; | |
e25cad69 | 2454 | int ret; |
eab8eb8d VT |
2455 | SheepdogInode *inode = &s->inode; |
2456 | ||
e25cad69 | 2457 | fd = connect_to_sdog(s, errp); |
eab8eb8d | 2458 | if (fd < 0) { |
e25cad69 | 2459 | return fd; |
eab8eb8d VT |
2460 | } |
2461 | ||
2462 | nr_objs = count_data_objs(inode); | |
2463 | while (i < nr_objs) { | |
2464 | int start_idx, nr_filled_idx; | |
2465 | ||
2466 | while (i < nr_objs && !inode->data_vdi_id[i]) { | |
2467 | i++; | |
2468 | } | |
2469 | start_idx = i; | |
2470 | ||
2471 | nr_filled_idx = 0; | |
2472 | while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) { | |
2473 | if (inode->data_vdi_id[i]) { | |
2474 | inode->data_vdi_id[i] = 0; | |
2475 | nr_filled_idx++; | |
2476 | } | |
2477 | ||
2478 | i++; | |
2479 | } | |
2480 | ||
f11672db | 2481 | ret = write_object(fd, s->bs, |
eab8eb8d VT |
2482 | (char *)&inode->data_vdi_id[start_idx], |
2483 | vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies, | |
2484 | (i - start_idx) * sizeof(uint32_t), | |
2485 | offsetof(struct SheepdogInode, | |
2486 | data_vdi_id[start_idx]), | |
2487 | false, s->cache_flags); | |
2488 | if (ret < 0) { | |
e25cad69 | 2489 | error_setg(errp, "Failed to discard snapshot inode"); |
eab8eb8d VT |
2490 | goto out; |
2491 | } | |
2492 | } | |
2493 | ||
e25cad69 | 2494 | ret = 0; |
eab8eb8d VT |
2495 | out: |
2496 | closesocket(fd); | |
e25cad69 | 2497 | return ret; |
eab8eb8d VT |
2498 | } |
2499 | ||
a89d89d3 WX |
2500 | static int sd_snapshot_delete(BlockDriverState *bs, |
2501 | const char *snapshot_id, | |
2502 | const char *name, | |
2503 | Error **errp) | |
33b1db1c | 2504 | { |
a0dc0e2b MA |
2505 | /* |
2506 | * FIXME should delete the snapshot matching both @snapshot_id and | |
2507 | * @name, but @name not used here | |
2508 | */ | |
03c698f0 | 2509 | unsigned long snap_id = 0; |
eab8eb8d | 2510 | char snap_tag[SD_MAX_VDI_TAG_LEN]; |
eab8eb8d VT |
2511 | int fd, ret; |
2512 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
2513 | BDRVSheepdogState *s = bs->opaque; | |
2514 | unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0; | |
2515 | uint32_t vid; | |
2516 | SheepdogVdiReq hdr = { | |
2517 | .opcode = SD_OP_DEL_VDI, | |
2518 | .data_length = wlen, | |
2519 | .flags = SD_FLAG_CMD_WRITE, | |
2520 | }; | |
2521 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2522 | ||
e25cad69 MA |
2523 | ret = remove_objects(s, errp); |
2524 | if (ret) { | |
2525 | return ret; | |
eab8eb8d VT |
2526 | } |
2527 | ||
2528 | memset(buf, 0, sizeof(buf)); | |
2529 | memset(snap_tag, 0, sizeof(snap_tag)); | |
2530 | pstrcpy(buf, SD_MAX_VDI_LEN, s->name); | |
89e2a31d | 2531 | /* TODO Use sd_parse_snapid() once this mess is cleaned up */ |
03c698f0 JC |
2532 | ret = qemu_strtoul(snapshot_id, NULL, 10, &snap_id); |
2533 | if (ret || snap_id > UINT32_MAX) { | |
a0dc0e2b MA |
2534 | /* |
2535 | * FIXME Since qemu_strtoul() returns -EINVAL when | |
2536 | * @snapshot_id is null, @snapshot_id is mandatory. Correct | |
2537 | * would be to require at least one of @snapshot_id and @name. | |
2538 | */ | |
03c698f0 JC |
2539 | error_setg(errp, "Invalid snapshot ID: %s", |
2540 | snapshot_id ? snapshot_id : "<null>"); | |
2541 | return -EINVAL; | |
eab8eb8d VT |
2542 | } |
2543 | ||
2544 | if (snap_id) { | |
03c698f0 | 2545 | hdr.snapid = (uint32_t) snap_id; |
eab8eb8d | 2546 | } else { |
a0dc0e2b | 2547 | /* FIXME I suspect we should use @name here */ |
89e2a31d | 2548 | /* FIXME don't truncate silently */ |
eab8eb8d VT |
2549 | pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id); |
2550 | pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag); | |
2551 | } | |
2552 | ||
e25cad69 | 2553 | ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, errp); |
eab8eb8d VT |
2554 | if (ret) { |
2555 | return ret; | |
2556 | } | |
2557 | ||
e25cad69 | 2558 | fd = connect_to_sdog(s, errp); |
eab8eb8d | 2559 | if (fd < 0) { |
e25cad69 | 2560 | return fd; |
eab8eb8d VT |
2561 | } |
2562 | ||
f11672db | 2563 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, |
eab8eb8d VT |
2564 | buf, &wlen, &rlen); |
2565 | closesocket(fd); | |
2566 | if (ret) { | |
e25cad69 | 2567 | error_setg_errno(errp, -ret, "Couldn't send request to server"); |
eab8eb8d VT |
2568 | return ret; |
2569 | } | |
2570 | ||
2571 | switch (rsp->result) { | |
2572 | case SD_RES_NO_VDI: | |
e25cad69 MA |
2573 | error_setg(errp, "Can't find the snapshot"); |
2574 | return -ENOENT; | |
eab8eb8d VT |
2575 | case SD_RES_SUCCESS: |
2576 | break; | |
2577 | default: | |
e25cad69 MA |
2578 | error_setg(errp, "%s", sd_strerror(rsp->result)); |
2579 | return -EIO; | |
eab8eb8d VT |
2580 | } |
2581 | ||
e25cad69 | 2582 | return 0; |
33b1db1c MK |
2583 | } |
2584 | ||
33b1db1c MK |
2585 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
2586 | { | |
dfb12bf8 | 2587 | Error *local_err = NULL; |
33b1db1c MK |
2588 | BDRVSheepdogState *s = bs->opaque; |
2589 | SheepdogReq req; | |
2590 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
2591 | QEMUSnapshotInfo *sn_tab = NULL; | |
2592 | unsigned wlen, rlen; | |
2593 | int found = 0; | |
2594 | static SheepdogInode inode; | |
2595 | unsigned long *vdi_inuse; | |
2596 | unsigned int start_nr; | |
2597 | uint64_t hval; | |
2598 | uint32_t vid; | |
2599 | ||
7267c094 | 2600 | vdi_inuse = g_malloc(max); |
33b1db1c | 2601 | |
dfb12bf8 | 2602 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2603 | if (fd < 0) { |
565f65d2 | 2604 | error_report_err(local_err); |
cb595887 | 2605 | ret = fd; |
33b1db1c MK |
2606 | goto out; |
2607 | } | |
2608 | ||
2609 | rlen = max; | |
2610 | wlen = 0; | |
2611 | ||
2612 | memset(&req, 0, sizeof(req)); | |
2613 | ||
2614 | req.opcode = SD_OP_READ_VDIS; | |
2615 | req.data_length = max; | |
2616 | ||
f11672db | 2617 | ret = do_req(fd, s->bs, &req, vdi_inuse, &wlen, &rlen); |
33b1db1c MK |
2618 | |
2619 | closesocket(fd); | |
2620 | if (ret) { | |
2621 | goto out; | |
2622 | } | |
2623 | ||
02c4f26b | 2624 | sn_tab = g_new0(QEMUSnapshotInfo, nr); |
33b1db1c MK |
2625 | |
2626 | /* calculate a vdi id with hash function */ | |
2627 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2628 | start_nr = hval & (SD_NR_VDIS - 1); | |
2629 | ||
dfb12bf8 | 2630 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2631 | if (fd < 0) { |
565f65d2 | 2632 | error_report_err(local_err); |
cb595887 | 2633 | ret = fd; |
33b1db1c MK |
2634 | goto out; |
2635 | } | |
2636 | ||
2637 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2638 | if (!test_bit(vid, vdi_inuse)) { | |
2639 | break; | |
2640 | } | |
2641 | ||
2642 | /* we don't need to read entire object */ | |
f11672db | 2643 | ret = read_object(fd, s->bs, (char *)&inode, |
84390bed | 2644 | vid_to_vdi_oid(vid), |
47622c44 | 2645 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2646 | s->cache_flags); |
33b1db1c MK |
2647 | |
2648 | if (ret) { | |
2649 | continue; | |
2650 | } | |
2651 | ||
2652 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2653 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2654 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2655 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2656 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2657 | ||
521b2b5d HR |
2658 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), |
2659 | "%" PRIu32, inode.snap_id); | |
3178e275 JM |
2660 | pstrcpy(sn_tab[found].name, |
2661 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2662 | inode.tag); | |
33b1db1c MK |
2663 | found++; |
2664 | } | |
2665 | } | |
2666 | ||
2667 | closesocket(fd); | |
2668 | out: | |
2669 | *psn_tab = sn_tab; | |
2670 | ||
7267c094 | 2671 | g_free(vdi_inuse); |
33b1db1c | 2672 | |
cb595887 MK |
2673 | if (ret < 0) { |
2674 | return ret; | |
2675 | } | |
2676 | ||
33b1db1c MK |
2677 | return found; |
2678 | } | |
2679 | ||
2680 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2681 | int64_t pos, int size, int load) | |
2682 | { | |
dfb12bf8 | 2683 | Error *local_err = NULL; |
2f536801 MK |
2684 | bool create; |
2685 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2686 | unsigned int data_len; |
2687 | uint64_t vmstate_oid; | |
33b1db1c | 2688 | uint64_t offset; |
cede621f LY |
2689 | uint32_t vdi_index; |
2690 | uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id; | |
876eb1b0 | 2691 | uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift); |
33b1db1c | 2692 | |
dfb12bf8 | 2693 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2694 | if (fd < 0) { |
565f65d2 | 2695 | error_report_err(local_err); |
cb595887 | 2696 | return fd; |
33b1db1c MK |
2697 | } |
2698 | ||
6f3c714e | 2699 | while (remaining) { |
876eb1b0 TI |
2700 | vdi_index = pos / object_size; |
2701 | offset = pos % object_size; | |
33b1db1c | 2702 | |
876eb1b0 | 2703 | data_len = MIN(remaining, object_size - offset); |
33b1db1c | 2704 | |
cede621f | 2705 | vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index); |
33b1db1c MK |
2706 | |
2707 | create = (offset == 0); | |
2708 | if (load) { | |
f11672db | 2709 | ret = read_object(fd, s->bs, (char *)data, vmstate_oid, |
47622c44 | 2710 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2711 | s->cache_flags); |
33b1db1c | 2712 | } else { |
f11672db | 2713 | ret = write_object(fd, s->bs, (char *)data, vmstate_oid, |
47622c44 | 2714 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2715 | s->cache_flags); |
33b1db1c MK |
2716 | } |
2717 | ||
2718 | if (ret < 0) { | |
6daf194d | 2719 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2720 | goto cleanup; |
2721 | } | |
2722 | ||
2723 | pos += data_len; | |
1f7a48de | 2724 | data += data_len; |
6f3c714e | 2725 | remaining -= data_len; |
33b1db1c | 2726 | } |
6f3c714e | 2727 | ret = size; |
33b1db1c MK |
2728 | cleanup: |
2729 | closesocket(fd); | |
2730 | return ret; | |
2731 | } | |
2732 | ||
cf8074b3 KW |
2733 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2734 | int64_t pos) | |
33b1db1c MK |
2735 | { |
2736 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2737 | void *buf; |
2738 | int ret; | |
33b1db1c | 2739 | |
cf8074b3 KW |
2740 | buf = qemu_blockalign(bs, qiov->size); |
2741 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2742 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2743 | qemu_vfree(buf); | |
2744 | ||
2745 | return ret; | |
33b1db1c MK |
2746 | } |
2747 | ||
5ddda0b8 KW |
2748 | static int sd_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2749 | int64_t pos) | |
33b1db1c MK |
2750 | { |
2751 | BDRVSheepdogState *s = bs->opaque; | |
5ddda0b8 KW |
2752 | void *buf; |
2753 | int ret; | |
33b1db1c | 2754 | |
5ddda0b8 KW |
2755 | buf = qemu_blockalign(bs, qiov->size); |
2756 | ret = do_load_save_vmstate(s, buf, pos, qiov->size, 1); | |
2757 | qemu_iovec_from_buf(qiov, 0, buf, qiov->size); | |
2758 | qemu_vfree(buf); | |
2759 | ||
2760 | return ret; | |
33b1db1c MK |
2761 | } |
2762 | ||
2763 | ||
dde47537 EB |
2764 | static coroutine_fn int sd_co_pdiscard(BlockDriverState *bs, int64_t offset, |
2765 | int count) | |
cac8f4a6 | 2766 | { |
28ddd08c | 2767 | SheepdogAIOCB acb; |
cac8f4a6 | 2768 | BDRVSheepdogState *s = bs->opaque; |
e6fd57ea HM |
2769 | QEMUIOVector discard_iov; |
2770 | struct iovec iov; | |
2771 | uint32_t zero = 0; | |
cac8f4a6 LY |
2772 | |
2773 | if (!s->discard_supported) { | |
dde47537 | 2774 | return 0; |
cac8f4a6 LY |
2775 | } |
2776 | ||
e6fd57ea HM |
2777 | memset(&discard_iov, 0, sizeof(discard_iov)); |
2778 | memset(&iov, 0, sizeof(iov)); | |
2779 | iov.iov_base = &zero; | |
2780 | iov.iov_len = sizeof(zero); | |
2781 | discard_iov.iov = &iov; | |
2782 | discard_iov.niov = 1; | |
49228d1e EB |
2783 | if (!QEMU_IS_ALIGNED(offset | count, BDRV_SECTOR_SIZE)) { |
2784 | return -ENOTSUP; | |
2785 | } | |
28ddd08c PB |
2786 | sd_aio_setup(&acb, s, &discard_iov, offset >> BDRV_SECTOR_BITS, |
2787 | count >> BDRV_SECTOR_BITS, AIOCB_DISCARD_OBJ); | |
28ddd08c | 2788 | sd_co_rw_vector(&acb); |
acf6e5f0 | 2789 | sd_aio_complete(&acb); |
cac8f4a6 | 2790 | |
28ddd08c | 2791 | return acb.ret; |
cac8f4a6 LY |
2792 | } |
2793 | ||
b6b8a333 PB |
2794 | static coroutine_fn int64_t |
2795 | sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
67a0fd2a | 2796 | int *pnum, BlockDriverState **file) |
8d71c631 LY |
2797 | { |
2798 | BDRVSheepdogState *s = bs->opaque; | |
2799 | SheepdogInode *inode = &s->inode; | |
876eb1b0 | 2800 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
9cd76737 | 2801 | uint64_t offset = sector_num * BDRV_SECTOR_SIZE; |
876eb1b0 | 2802 | unsigned long start = offset / object_size, |
8d71c631 | 2803 | end = DIV_ROUND_UP((sector_num + nb_sectors) * |
876eb1b0 | 2804 | BDRV_SECTOR_SIZE, object_size); |
8d71c631 | 2805 | unsigned long idx; |
9cd76737 | 2806 | int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset; |
8d71c631 LY |
2807 | |
2808 | for (idx = start; idx < end; idx++) { | |
2809 | if (inode->data_vdi_id[idx] == 0) { | |
2810 | break; | |
2811 | } | |
2812 | } | |
2813 | if (idx == start) { | |
2814 | /* Get the longest length of unallocated sectors */ | |
2815 | ret = 0; | |
2816 | for (idx = start + 1; idx < end; idx++) { | |
2817 | if (inode->data_vdi_id[idx] != 0) { | |
2818 | break; | |
2819 | } | |
2820 | } | |
2821 | } | |
2822 | ||
876eb1b0 | 2823 | *pnum = (idx - start) * object_size / BDRV_SECTOR_SIZE; |
8d71c631 LY |
2824 | if (*pnum > nb_sectors) { |
2825 | *pnum = nb_sectors; | |
2826 | } | |
d234c929 FZ |
2827 | if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) { |
2828 | *file = bs; | |
2829 | } | |
8d71c631 LY |
2830 | return ret; |
2831 | } | |
2832 | ||
85829722 LY |
2833 | static int64_t sd_get_allocated_file_size(BlockDriverState *bs) |
2834 | { | |
2835 | BDRVSheepdogState *s = bs->opaque; | |
2836 | SheepdogInode *inode = &s->inode; | |
876eb1b0 TI |
2837 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
2838 | unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size); | |
85829722 LY |
2839 | uint64_t size = 0; |
2840 | ||
2841 | for (i = 0; i < last; i++) { | |
2842 | if (inode->data_vdi_id[i] == 0) { | |
2843 | continue; | |
2844 | } | |
876eb1b0 | 2845 | size += object_size; |
85829722 LY |
2846 | } |
2847 | return size; | |
2848 | } | |
2849 | ||
b222237b CL |
2850 | static QemuOptsList sd_create_opts = { |
2851 | .name = "sheepdog-create-opts", | |
2852 | .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head), | |
2853 | .desc = { | |
2854 | { | |
2855 | .name = BLOCK_OPT_SIZE, | |
2856 | .type = QEMU_OPT_SIZE, | |
2857 | .help = "Virtual disk size" | |
2858 | }, | |
2859 | { | |
2860 | .name = BLOCK_OPT_BACKING_FILE, | |
2861 | .type = QEMU_OPT_STRING, | |
2862 | .help = "File name of a base image" | |
2863 | }, | |
2864 | { | |
2865 | .name = BLOCK_OPT_PREALLOC, | |
2866 | .type = QEMU_OPT_STRING, | |
2867 | .help = "Preallocation mode (allowed values: off, full)" | |
2868 | }, | |
2869 | { | |
2870 | .name = BLOCK_OPT_REDUNDANCY, | |
2871 | .type = QEMU_OPT_STRING, | |
2872 | .help = "Redundancy of the image" | |
2873 | }, | |
876eb1b0 TI |
2874 | { |
2875 | .name = BLOCK_OPT_OBJECT_SIZE, | |
2876 | .type = QEMU_OPT_SIZE, | |
2877 | .help = "Object size of the image" | |
2878 | }, | |
b222237b CL |
2879 | { /* end of list */ } |
2880 | } | |
33b1db1c MK |
2881 | }; |
2882 | ||
5d6768e3 | 2883 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2884 | .format_name = "sheepdog", |
2885 | .protocol_name = "sheepdog", | |
2886 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2887 | .bdrv_needs_filename = true, |
33b1db1c | 2888 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2889 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2890 | .bdrv_reopen_commit = sd_reopen_commit, | |
2891 | .bdrv_reopen_abort = sd_reopen_abort, | |
33b1db1c | 2892 | .bdrv_close = sd_close, |
c282e1fd | 2893 | .bdrv_create = sd_create, |
e4f5c1bf | 2894 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
33b1db1c | 2895 | .bdrv_getlength = sd_getlength, |
85829722 | 2896 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
33b1db1c MK |
2897 | .bdrv_truncate = sd_truncate, |
2898 | ||
2df46246 MK |
2899 | .bdrv_co_readv = sd_co_readv, |
2900 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2901 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
dde47537 | 2902 | .bdrv_co_pdiscard = sd_co_pdiscard, |
b6b8a333 | 2903 | .bdrv_co_get_block_status = sd_co_get_block_status, |
33b1db1c MK |
2904 | |
2905 | .bdrv_snapshot_create = sd_snapshot_create, | |
2906 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2907 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2908 | .bdrv_snapshot_list = sd_snapshot_list, | |
2909 | ||
2910 | .bdrv_save_vmstate = sd_save_vmstate, | |
2911 | .bdrv_load_vmstate = sd_load_vmstate, | |
2912 | ||
84390bed SH |
2913 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2914 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2915 | ||
b222237b | 2916 | .create_opts = &sd_create_opts, |
33b1db1c MK |
2917 | }; |
2918 | ||
5d6768e3 MK |
2919 | static BlockDriver bdrv_sheepdog_tcp = { |
2920 | .format_name = "sheepdog", | |
2921 | .protocol_name = "sheepdog+tcp", | |
2922 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2923 | .bdrv_needs_filename = true, |
5d6768e3 | 2924 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2925 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2926 | .bdrv_reopen_commit = sd_reopen_commit, | |
2927 | .bdrv_reopen_abort = sd_reopen_abort, | |
5d6768e3 | 2928 | .bdrv_close = sd_close, |
c282e1fd | 2929 | .bdrv_create = sd_create, |
e4f5c1bf | 2930 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
5d6768e3 | 2931 | .bdrv_getlength = sd_getlength, |
85829722 | 2932 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
5d6768e3 MK |
2933 | .bdrv_truncate = sd_truncate, |
2934 | ||
2935 | .bdrv_co_readv = sd_co_readv, | |
2936 | .bdrv_co_writev = sd_co_writev, | |
2937 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
dde47537 | 2938 | .bdrv_co_pdiscard = sd_co_pdiscard, |
b6b8a333 | 2939 | .bdrv_co_get_block_status = sd_co_get_block_status, |
5d6768e3 MK |
2940 | |
2941 | .bdrv_snapshot_create = sd_snapshot_create, | |
2942 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2943 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2944 | .bdrv_snapshot_list = sd_snapshot_list, | |
2945 | ||
2946 | .bdrv_save_vmstate = sd_save_vmstate, | |
2947 | .bdrv_load_vmstate = sd_load_vmstate, | |
2948 | ||
84390bed SH |
2949 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2950 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2951 | ||
b222237b | 2952 | .create_opts = &sd_create_opts, |
5d6768e3 MK |
2953 | }; |
2954 | ||
1b8bbb46 MK |
2955 | static BlockDriver bdrv_sheepdog_unix = { |
2956 | .format_name = "sheepdog", | |
2957 | .protocol_name = "sheepdog+unix", | |
2958 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2959 | .bdrv_needs_filename = true, |
1b8bbb46 | 2960 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2961 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2962 | .bdrv_reopen_commit = sd_reopen_commit, | |
2963 | .bdrv_reopen_abort = sd_reopen_abort, | |
1b8bbb46 | 2964 | .bdrv_close = sd_close, |
c282e1fd | 2965 | .bdrv_create = sd_create, |
3ac21627 | 2966 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
1b8bbb46 | 2967 | .bdrv_getlength = sd_getlength, |
85829722 | 2968 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
1b8bbb46 MK |
2969 | .bdrv_truncate = sd_truncate, |
2970 | ||
2971 | .bdrv_co_readv = sd_co_readv, | |
2972 | .bdrv_co_writev = sd_co_writev, | |
2973 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
dde47537 | 2974 | .bdrv_co_pdiscard = sd_co_pdiscard, |
b6b8a333 | 2975 | .bdrv_co_get_block_status = sd_co_get_block_status, |
1b8bbb46 MK |
2976 | |
2977 | .bdrv_snapshot_create = sd_snapshot_create, | |
2978 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2979 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2980 | .bdrv_snapshot_list = sd_snapshot_list, | |
2981 | ||
2982 | .bdrv_save_vmstate = sd_save_vmstate, | |
2983 | .bdrv_load_vmstate = sd_load_vmstate, | |
2984 | ||
84390bed SH |
2985 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2986 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2987 | ||
b222237b | 2988 | .create_opts = &sd_create_opts, |
1b8bbb46 MK |
2989 | }; |
2990 | ||
33b1db1c MK |
2991 | static void bdrv_sheepdog_init(void) |
2992 | { | |
2993 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 2994 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 2995 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
2996 | } |
2997 | block_init(bdrv_sheepdog_init); |