]>
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" |
33b1db1c | 16 | #include "qemu-common.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" |
33b1db1c MK |
23 | |
24 | #define SD_PROTO_VER 0x01 | |
25 | ||
26 | #define SD_DEFAULT_ADDR "localhost" | |
25af257d | 27 | #define SD_DEFAULT_PORT 7000 |
33b1db1c MK |
28 | |
29 | #define SD_OP_CREATE_AND_WRITE_OBJ 0x01 | |
30 | #define SD_OP_READ_OBJ 0x02 | |
31 | #define SD_OP_WRITE_OBJ 0x03 | |
cac8f4a6 | 32 | /* 0x04 is used internally by Sheepdog */ |
33b1db1c MK |
33 | |
34 | #define SD_OP_NEW_VDI 0x11 | |
35 | #define SD_OP_LOCK_VDI 0x12 | |
36 | #define SD_OP_RELEASE_VDI 0x13 | |
37 | #define SD_OP_GET_VDI_INFO 0x14 | |
38 | #define SD_OP_READ_VDIS 0x15 | |
47622c44 | 39 | #define SD_OP_FLUSH_VDI 0x16 |
859e5553 | 40 | #define SD_OP_DEL_VDI 0x17 |
876eb1b0 | 41 | #define SD_OP_GET_CLUSTER_DEFAULT 0x18 |
33b1db1c MK |
42 | |
43 | #define SD_FLAG_CMD_WRITE 0x01 | |
44 | #define SD_FLAG_CMD_COW 0x02 | |
0e7106d8 LY |
45 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
46 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
33b1db1c MK |
47 | |
48 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
49 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
50 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
51 | #define SD_RES_EIO 0x03 /* I/O error */ | |
52 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
53 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
54 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
55 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
56 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
57 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
58 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
59 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
60 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
61 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
62 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
63 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
64 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
65 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
66 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
67 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
68 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
69 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
70 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
71 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
72 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
fca23f0a | 73 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ |
6a0b5490 | 74 | #define SD_RES_READONLY 0x1A /* Object is read-only */ |
33b1db1c MK |
75 | |
76 | /* | |
77 | * Object ID rules | |
78 | * | |
79 | * 0 - 19 (20 bits): data object space | |
80 | * 20 - 31 (12 bits): reserved data object space | |
81 | * 32 - 55 (24 bits): vdi object space | |
82 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 83 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
84 | */ |
85 | ||
86 | #define VDI_SPACE_SHIFT 32 | |
87 | #define VDI_BIT (UINT64_C(1) << 63) | |
88 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
89 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
90 | #define MAX_CHILDREN 1024 | |
91 | #define SD_MAX_VDI_LEN 256 | |
92 | #define SD_MAX_VDI_TAG_LEN 256 | |
93 | #define SD_NR_VDIS (1U << 24) | |
94 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
95 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
876eb1b0 | 96 | #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22 |
b3af018f LY |
97 | /* |
98 | * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and | |
99 | * (SD_EC_MAX_STRIP - 1) for parity strips | |
100 | * | |
101 | * SD_MAX_COPIES is sum of number of data strips and parity strips. | |
102 | */ | |
103 | #define SD_EC_MAX_STRIP 16 | |
104 | #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1) | |
33b1db1c MK |
105 | |
106 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
107 | #define CURRENT_VDI_ID 0 | |
108 | ||
1dbfafed HM |
109 | #define LOCK_TYPE_NORMAL 0 |
110 | #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */ | |
111 | ||
33b1db1c MK |
112 | typedef struct SheepdogReq { |
113 | uint8_t proto_ver; | |
114 | uint8_t opcode; | |
115 | uint16_t flags; | |
116 | uint32_t epoch; | |
117 | uint32_t id; | |
118 | uint32_t data_length; | |
119 | uint32_t opcode_specific[8]; | |
120 | } SheepdogReq; | |
121 | ||
122 | typedef struct SheepdogRsp { | |
123 | uint8_t proto_ver; | |
124 | uint8_t opcode; | |
125 | uint16_t flags; | |
126 | uint32_t epoch; | |
127 | uint32_t id; | |
128 | uint32_t data_length; | |
129 | uint32_t result; | |
130 | uint32_t opcode_specific[7]; | |
131 | } SheepdogRsp; | |
132 | ||
133 | typedef struct SheepdogObjReq { | |
134 | uint8_t proto_ver; | |
135 | uint8_t opcode; | |
136 | uint16_t flags; | |
137 | uint32_t epoch; | |
138 | uint32_t id; | |
139 | uint32_t data_length; | |
140 | uint64_t oid; | |
141 | uint64_t cow_oid; | |
29a67f7e | 142 | uint8_t copies; |
1841f880 LY |
143 | uint8_t copy_policy; |
144 | uint8_t reserved[6]; | |
33b1db1c MK |
145 | uint64_t offset; |
146 | } SheepdogObjReq; | |
147 | ||
148 | typedef struct SheepdogObjRsp { | |
149 | uint8_t proto_ver; | |
150 | uint8_t opcode; | |
151 | uint16_t flags; | |
152 | uint32_t epoch; | |
153 | uint32_t id; | |
154 | uint32_t data_length; | |
155 | uint32_t result; | |
29a67f7e | 156 | uint8_t copies; |
1841f880 LY |
157 | uint8_t copy_policy; |
158 | uint8_t reserved[2]; | |
33b1db1c MK |
159 | uint32_t pad[6]; |
160 | } SheepdogObjRsp; | |
161 | ||
162 | typedef struct SheepdogVdiReq { | |
163 | uint8_t proto_ver; | |
164 | uint8_t opcode; | |
165 | uint16_t flags; | |
166 | uint32_t epoch; | |
167 | uint32_t id; | |
168 | uint32_t data_length; | |
169 | uint64_t vdi_size; | |
9f23fce7 | 170 | uint32_t base_vdi_id; |
29a67f7e | 171 | uint8_t copies; |
1841f880 | 172 | uint8_t copy_policy; |
876eb1b0 TI |
173 | uint8_t store_policy; |
174 | uint8_t block_size_shift; | |
33b1db1c | 175 | uint32_t snapid; |
1dbfafed HM |
176 | uint32_t type; |
177 | uint32_t pad[2]; | |
33b1db1c MK |
178 | } SheepdogVdiReq; |
179 | ||
180 | typedef struct SheepdogVdiRsp { | |
181 | uint8_t proto_ver; | |
182 | uint8_t opcode; | |
183 | uint16_t flags; | |
184 | uint32_t epoch; | |
185 | uint32_t id; | |
186 | uint32_t data_length; | |
187 | uint32_t result; | |
188 | uint32_t rsvd; | |
189 | uint32_t vdi_id; | |
190 | uint32_t pad[5]; | |
191 | } SheepdogVdiRsp; | |
192 | ||
876eb1b0 TI |
193 | typedef struct SheepdogClusterRsp { |
194 | uint8_t proto_ver; | |
195 | uint8_t opcode; | |
196 | uint16_t flags; | |
197 | uint32_t epoch; | |
198 | uint32_t id; | |
199 | uint32_t data_length; | |
200 | uint32_t result; | |
201 | uint8_t nr_copies; | |
202 | uint8_t copy_policy; | |
203 | uint8_t block_size_shift; | |
204 | uint8_t __pad1; | |
205 | uint32_t __pad2[6]; | |
206 | } SheepdogClusterRsp; | |
207 | ||
33b1db1c MK |
208 | typedef struct SheepdogInode { |
209 | char name[SD_MAX_VDI_LEN]; | |
210 | char tag[SD_MAX_VDI_TAG_LEN]; | |
211 | uint64_t ctime; | |
212 | uint64_t snap_ctime; | |
213 | uint64_t vm_clock_nsec; | |
214 | uint64_t vdi_size; | |
215 | uint64_t vm_state_size; | |
216 | uint16_t copy_policy; | |
217 | uint8_t nr_copies; | |
218 | uint8_t block_size_shift; | |
219 | uint32_t snap_id; | |
220 | uint32_t vdi_id; | |
221 | uint32_t parent_vdi_id; | |
222 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
223 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
224 | } SheepdogInode; | |
225 | ||
5d039bab HM |
226 | #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id) |
227 | ||
33b1db1c MK |
228 | /* |
229 | * 64 bit FNV-1a non-zero initial basis | |
230 | */ | |
231 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
232 | ||
233 | /* | |
234 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
235 | */ | |
236 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
237 | { | |
238 | unsigned char *bp = buf; | |
239 | unsigned char *be = bp + len; | |
240 | while (bp < be) { | |
241 | hval ^= (uint64_t) *bp++; | |
242 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
243 | (hval << 7) + (hval << 8) + (hval << 40); | |
244 | } | |
245 | return hval; | |
246 | } | |
247 | ||
2f536801 | 248 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
249 | { |
250 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
251 | } | |
252 | ||
2f536801 | 253 | static inline bool is_data_obj(uint64_t oid) |
33b1db1c MK |
254 | { |
255 | return !(VDI_BIT & oid); | |
256 | } | |
257 | ||
258 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
259 | { | |
260 | return oid & (MAX_DATA_OBJS - 1); | |
261 | } | |
262 | ||
72e0996c MK |
263 | static inline uint32_t oid_to_vid(uint64_t oid) |
264 | { | |
265 | return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT; | |
266 | } | |
267 | ||
33b1db1c MK |
268 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) |
269 | { | |
270 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
271 | } | |
272 | ||
273 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
274 | { | |
275 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
276 | } | |
277 | ||
278 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
279 | { | |
280 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
281 | } | |
282 | ||
2f536801 | 283 | static inline bool is_snapshot(struct SheepdogInode *inode) |
33b1db1c MK |
284 | { |
285 | return !!inode->snap_ctime; | |
286 | } | |
287 | ||
eab8eb8d VT |
288 | static inline size_t count_data_objs(const struct SheepdogInode *inode) |
289 | { | |
290 | return DIV_ROUND_UP(inode->vdi_size, | |
291 | (1UL << inode->block_size_shift)); | |
292 | } | |
293 | ||
2440a2c3 | 294 | #undef DPRINTF |
33b1db1c | 295 | #ifdef DEBUG_SDOG |
2440a2c3 | 296 | #define DPRINTF(fmt, args...) \ |
33b1db1c MK |
297 | do { \ |
298 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
299 | } while (0) | |
300 | #else | |
2440a2c3 | 301 | #define DPRINTF(fmt, args...) |
33b1db1c MK |
302 | #endif |
303 | ||
304 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
305 | ||
306 | typedef struct AIOReq { | |
307 | SheepdogAIOCB *aiocb; | |
308 | unsigned int iov_offset; | |
309 | ||
310 | uint64_t oid; | |
311 | uint64_t base_oid; | |
312 | uint64_t offset; | |
313 | unsigned int data_len; | |
314 | uint8_t flags; | |
315 | uint32_t id; | |
b544c1ab | 316 | bool create; |
33b1db1c | 317 | |
c292ee6a | 318 | QLIST_ENTRY(AIOReq) aio_siblings; |
33b1db1c MK |
319 | } AIOReq; |
320 | ||
321 | enum AIOCBState { | |
322 | AIOCB_WRITE_UDATA, | |
323 | AIOCB_READ_UDATA, | |
47783072 | 324 | AIOCB_FLUSH_CACHE, |
cac8f4a6 | 325 | AIOCB_DISCARD_OBJ, |
33b1db1c MK |
326 | }; |
327 | ||
498f2140 | 328 | #define AIOCBOverlapping(x, y) \ |
6a55c82c HM |
329 | (!(x->max_affect_data_idx < y->min_affect_data_idx \ |
330 | || y->max_affect_data_idx < x->min_affect_data_idx)) | |
331 | ||
33b1db1c | 332 | struct SheepdogAIOCB { |
7c84b1b8 | 333 | BlockAIOCB common; |
33b1db1c MK |
334 | |
335 | QEMUIOVector *qiov; | |
336 | ||
337 | int64_t sector_num; | |
338 | int nb_sectors; | |
339 | ||
340 | int ret; | |
341 | enum AIOCBState aiocb_type; | |
342 | ||
2df46246 | 343 | Coroutine *coroutine; |
33b1db1c MK |
344 | void (*aio_done_func)(SheepdogAIOCB *); |
345 | ||
35200687 | 346 | bool cancelable; |
1d732d7d | 347 | int nr_pending; |
6a55c82c HM |
348 | |
349 | uint32_t min_affect_data_idx; | |
350 | uint32_t max_affect_data_idx; | |
351 | ||
498f2140 HM |
352 | /* |
353 | * The difference between affect_data_idx and dirty_data_idx: | |
354 | * affect_data_idx represents range of index of all request types. | |
355 | * dirty_data_idx represents range of index updated by COW requests. | |
356 | * dirty_data_idx is used for updating an inode object. | |
357 | */ | |
358 | uint32_t min_dirty_data_idx; | |
359 | uint32_t max_dirty_data_idx; | |
360 | ||
6a55c82c | 361 | QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings; |
33b1db1c MK |
362 | }; |
363 | ||
364 | typedef struct BDRVSheepdogState { | |
011603ca | 365 | BlockDriverState *bs; |
84390bed | 366 | AioContext *aio_context; |
011603ca | 367 | |
33b1db1c MK |
368 | SheepdogInode inode; |
369 | ||
33b1db1c | 370 | char name[SD_MAX_VDI_LEN]; |
2f536801 | 371 | bool is_snapshot; |
0e7106d8 | 372 | uint32_t cache_flags; |
cac8f4a6 | 373 | bool discard_supported; |
33b1db1c | 374 | |
25af257d | 375 | char *host_spec; |
1b8bbb46 | 376 | bool is_unix; |
33b1db1c MK |
377 | int fd; |
378 | ||
2df46246 MK |
379 | CoMutex lock; |
380 | Coroutine *co_send; | |
381 | Coroutine *co_recv; | |
382 | ||
33b1db1c | 383 | uint32_t aioreq_seq_num; |
011603ca MK |
384 | |
385 | /* Every aio request must be linked to either of these queues. */ | |
c292ee6a | 386 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
011603ca | 387 | QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head; |
6a55c82c | 388 | |
498f2140 | 389 | CoQueue overlapping_queue; |
6a55c82c | 390 | QLIST_HEAD(inflight_aiocb_head, SheepdogAIOCB) inflight_aiocb_head; |
33b1db1c MK |
391 | } BDRVSheepdogState; |
392 | ||
4da65c80 LY |
393 | typedef struct BDRVSheepdogReopenState { |
394 | int fd; | |
395 | int cache_flags; | |
396 | } BDRVSheepdogReopenState; | |
397 | ||
33b1db1c MK |
398 | static const char * sd_strerror(int err) |
399 | { | |
400 | int i; | |
401 | ||
402 | static const struct { | |
403 | int err; | |
404 | const char *desc; | |
405 | } errors[] = { | |
406 | {SD_RES_SUCCESS, "Success"}, | |
407 | {SD_RES_UNKNOWN, "Unknown error"}, | |
408 | {SD_RES_NO_OBJ, "No object found"}, | |
409 | {SD_RES_EIO, "I/O error"}, | |
410 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
411 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
412 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
413 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
414 | {SD_RES_NO_VDI, "No vdi found"}, | |
415 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
416 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
417 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
418 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
419 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
420 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
421 | {SD_RES_STARTUP, "The system is still booting"}, | |
422 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
423 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
424 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
425 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
426 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
427 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
428 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
429 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
430 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
fca23f0a | 431 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, |
6a0b5490 | 432 | {SD_RES_READONLY, "Object is read-only"}, |
33b1db1c MK |
433 | }; |
434 | ||
435 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
436 | if (errors[i].err == err) { | |
437 | return errors[i].desc; | |
438 | } | |
439 | } | |
440 | ||
441 | return "Invalid error code"; | |
442 | } | |
443 | ||
444 | /* | |
445 | * Sheepdog I/O handling: | |
446 | * | |
2df46246 | 447 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
c292ee6a | 448 | * link the requests to the inflight_list in the |
2df46246 MK |
449 | * BDRVSheepdogState. The function exits without waiting for |
450 | * receiving the response. | |
33b1db1c | 451 | * |
2df46246 | 452 | * 2. We receive the response in aio_read_response, the fd handler to |
33b1db1c MK |
453 | * the sheepdog connection. If metadata update is needed, we send |
454 | * the write request to the vdi object in sd_write_done, the write | |
2df46246 MK |
455 | * completion function. We switch back to sd_co_readv/writev after |
456 | * all the requests belonging to the AIOCB are finished. | |
33b1db1c MK |
457 | */ |
458 | ||
459 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
460 | uint64_t oid, unsigned int data_len, | |
b544c1ab | 461 | uint64_t offset, uint8_t flags, bool create, |
33b1db1c MK |
462 | uint64_t base_oid, unsigned int iov_offset) |
463 | { | |
464 | AIOReq *aio_req; | |
465 | ||
7267c094 | 466 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
467 | aio_req->aiocb = acb; |
468 | aio_req->iov_offset = iov_offset; | |
469 | aio_req->oid = oid; | |
470 | aio_req->base_oid = base_oid; | |
471 | aio_req->offset = offset; | |
472 | aio_req->data_len = data_len; | |
473 | aio_req->flags = flags; | |
474 | aio_req->id = s->aioreq_seq_num++; | |
b544c1ab | 475 | aio_req->create = create; |
33b1db1c | 476 | |
1d732d7d | 477 | acb->nr_pending++; |
33b1db1c MK |
478 | return aio_req; |
479 | } | |
480 | ||
1d732d7d | 481 | static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
33b1db1c MK |
482 | { |
483 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1d732d7d | 484 | |
35200687 | 485 | acb->cancelable = false; |
c292ee6a | 486 | QLIST_REMOVE(aio_req, aio_siblings); |
7267c094 | 487 | g_free(aio_req); |
33b1db1c | 488 | |
1d732d7d | 489 | acb->nr_pending--; |
33b1db1c MK |
490 | } |
491 | ||
d8716b41 | 492 | static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
33b1db1c | 493 | { |
35200687 | 494 | qemu_coroutine_enter(acb->coroutine, NULL); |
8007429a | 495 | qemu_aio_unref(acb); |
33b1db1c MK |
496 | } |
497 | ||
35200687 MK |
498 | /* |
499 | * Check whether the specified acb can be canceled | |
500 | * | |
501 | * We can cancel aio when any request belonging to the acb is: | |
502 | * - Not processed by the sheepdog server. | |
503 | * - Not linked to the inflight queue. | |
504 | */ | |
505 | static bool sd_acb_cancelable(const SheepdogAIOCB *acb) | |
506 | { | |
507 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
508 | AIOReq *aioreq; | |
509 | ||
510 | if (!acb->cancelable) { | |
511 | return false; | |
512 | } | |
513 | ||
514 | QLIST_FOREACH(aioreq, &s->inflight_aio_head, aio_siblings) { | |
515 | if (aioreq->aiocb == acb) { | |
516 | return false; | |
517 | } | |
518 | } | |
519 | ||
520 | return true; | |
521 | } | |
522 | ||
7c84b1b8 | 523 | static void sd_aio_cancel(BlockAIOCB *blockacb) |
33b1db1c MK |
524 | { |
525 | SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; | |
35200687 MK |
526 | BDRVSheepdogState *s = acb->common.bs->opaque; |
527 | AIOReq *aioreq, *next; | |
6d24b4df FZ |
528 | |
529 | if (sd_acb_cancelable(acb)) { | |
6a55c82c | 530 | /* Remove outstanding requests from failed queue. */ |
6d24b4df FZ |
531 | QLIST_FOREACH_SAFE(aioreq, &s->failed_aio_head, aio_siblings, |
532 | next) { | |
533 | if (aioreq->aiocb == acb) { | |
534 | free_aio_req(s, aioreq); | |
35200687 | 535 | } |
6d24b4df | 536 | } |
35200687 | 537 | |
6d24b4df FZ |
538 | assert(acb->nr_pending == 0); |
539 | if (acb->common.cb) { | |
540 | acb->common.cb(acb->common.opaque, -ECANCELED); | |
35200687 | 541 | } |
6d24b4df | 542 | sd_finish_aiocb(acb); |
35200687 | 543 | } |
33b1db1c MK |
544 | } |
545 | ||
d7331bed | 546 | static const AIOCBInfo sd_aiocb_info = { |
6d24b4df FZ |
547 | .aiocb_size = sizeof(SheepdogAIOCB), |
548 | .cancel_async = sd_aio_cancel, | |
33b1db1c MK |
549 | }; |
550 | ||
551 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
f700f8e3 | 552 | int64_t sector_num, int nb_sectors) |
33b1db1c MK |
553 | { |
554 | SheepdogAIOCB *acb; | |
6a55c82c HM |
555 | uint32_t object_size; |
556 | BDRVSheepdogState *s = bs->opaque; | |
557 | ||
558 | object_size = (UINT32_C(1) << s->inode.block_size_shift); | |
33b1db1c | 559 | |
f700f8e3 | 560 | acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); |
33b1db1c MK |
561 | |
562 | acb->qiov = qiov; | |
563 | ||
564 | acb->sector_num = sector_num; | |
565 | acb->nb_sectors = nb_sectors; | |
566 | ||
567 | acb->aio_done_func = NULL; | |
35200687 | 568 | acb->cancelable = true; |
2df46246 | 569 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c | 570 | acb->ret = 0; |
1d732d7d | 571 | acb->nr_pending = 0; |
6a55c82c HM |
572 | |
573 | acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
574 | acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE + | |
575 | acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size; | |
576 | ||
498f2140 HM |
577 | acb->min_dirty_data_idx = UINT32_MAX; |
578 | acb->max_dirty_data_idx = 0; | |
579 | ||
33b1db1c MK |
580 | return acb; |
581 | } | |
582 | ||
833a7cc3 | 583 | /* Return -EIO in case of error, file descriptor on success */ |
dfb12bf8 | 584 | static int connect_to_sdog(BDRVSheepdogState *s, Error **errp) |
33b1db1c | 585 | { |
25af257d | 586 | int fd; |
33b1db1c | 587 | |
1b8bbb46 | 588 | if (s->is_unix) { |
dfb12bf8 | 589 | fd = unix_connect(s->host_spec, errp); |
1b8bbb46 | 590 | } else { |
dfb12bf8 | 591 | fd = inet_connect(s->host_spec, errp); |
1b8bbb46 | 592 | |
dfb12bf8 | 593 | if (fd >= 0) { |
1b8bbb46 MK |
594 | int ret = socket_set_nodelay(fd); |
595 | if (ret < 0) { | |
596 | error_report("%s", strerror(errno)); | |
597 | } | |
598 | } | |
599 | } | |
33b1db1c | 600 | |
dfb12bf8 | 601 | if (fd >= 0) { |
f9e8cacc | 602 | qemu_set_nonblock(fd); |
833a7cc3 LY |
603 | } else { |
604 | fd = -EIO; | |
33b1db1c MK |
605 | } |
606 | ||
33b1db1c MK |
607 | return fd; |
608 | } | |
609 | ||
833a7cc3 | 610 | /* Return 0 on success and -errno in case of error */ |
e0d93a89 MK |
611 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
612 | unsigned int *wlen) | |
47622c44 LY |
613 | { |
614 | int ret; | |
615 | ||
616 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
80731d9d | 617 | if (ret != sizeof(*hdr)) { |
47622c44 | 618 | error_report("failed to send a req, %s", strerror(errno)); |
b16a44e1 | 619 | return -errno; |
47622c44 LY |
620 | } |
621 | ||
622 | ret = qemu_co_send(sockfd, data, *wlen); | |
80731d9d | 623 | if (ret != *wlen) { |
47622c44 | 624 | error_report("failed to send a req, %s", strerror(errno)); |
b16a44e1 | 625 | return -errno; |
47622c44 LY |
626 | } |
627 | ||
628 | return ret; | |
629 | } | |
e0d93a89 | 630 | |
2dfcca3b MK |
631 | static void restart_co_req(void *opaque) |
632 | { | |
633 | Coroutine *co = opaque; | |
634 | ||
635 | qemu_coroutine_enter(co, NULL); | |
636 | } | |
637 | ||
cddd4ac7 MK |
638 | typedef struct SheepdogReqCo { |
639 | int sockfd; | |
84390bed | 640 | AioContext *aio_context; |
cddd4ac7 MK |
641 | SheepdogReq *hdr; |
642 | void *data; | |
643 | unsigned int *wlen; | |
644 | unsigned int *rlen; | |
645 | int ret; | |
646 | bool finished; | |
647 | } SheepdogReqCo; | |
648 | ||
649 | static coroutine_fn void do_co_req(void *opaque) | |
47622c44 LY |
650 | { |
651 | int ret; | |
2dfcca3b | 652 | Coroutine *co; |
cddd4ac7 MK |
653 | SheepdogReqCo *srco = opaque; |
654 | int sockfd = srco->sockfd; | |
655 | SheepdogReq *hdr = srco->hdr; | |
656 | void *data = srco->data; | |
657 | unsigned int *wlen = srco->wlen; | |
658 | unsigned int *rlen = srco->rlen; | |
2dfcca3b MK |
659 | |
660 | co = qemu_coroutine_self(); | |
dca21ef2 FZ |
661 | aio_set_fd_handler(srco->aio_context, sockfd, false, |
662 | NULL, restart_co_req, co); | |
47622c44 | 663 | |
47622c44 LY |
664 | ret = send_co_req(sockfd, hdr, data, wlen); |
665 | if (ret < 0) { | |
666 | goto out; | |
667 | } | |
668 | ||
dca21ef2 FZ |
669 | aio_set_fd_handler(srco->aio_context, sockfd, false, |
670 | restart_co_req, NULL, co); | |
2dfcca3b | 671 | |
47622c44 | 672 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); |
80731d9d | 673 | if (ret != sizeof(*hdr)) { |
47622c44 | 674 | error_report("failed to get a rsp, %s", strerror(errno)); |
cb595887 | 675 | ret = -errno; |
47622c44 LY |
676 | goto out; |
677 | } | |
678 | ||
679 | if (*rlen > hdr->data_length) { | |
680 | *rlen = hdr->data_length; | |
681 | } | |
682 | ||
683 | if (*rlen) { | |
684 | ret = qemu_co_recv(sockfd, data, *rlen); | |
80731d9d | 685 | if (ret != *rlen) { |
47622c44 | 686 | error_report("failed to get the data, %s", strerror(errno)); |
cb595887 | 687 | ret = -errno; |
47622c44 LY |
688 | goto out; |
689 | } | |
690 | } | |
691 | ret = 0; | |
692 | out: | |
ed9ba724 MK |
693 | /* there is at most one request for this sockfd, so it is safe to |
694 | * set each handler to NULL. */ | |
dca21ef2 FZ |
695 | aio_set_fd_handler(srco->aio_context, sockfd, false, |
696 | NULL, NULL, NULL); | |
cddd4ac7 MK |
697 | |
698 | srco->ret = ret; | |
699 | srco->finished = true; | |
700 | } | |
701 | ||
833a7cc3 LY |
702 | /* |
703 | * Send the request to the sheep in a synchronous manner. | |
704 | * | |
705 | * Return 0 on success, -errno in case of error. | |
706 | */ | |
84390bed SH |
707 | static int do_req(int sockfd, AioContext *aio_context, SheepdogReq *hdr, |
708 | void *data, unsigned int *wlen, unsigned int *rlen) | |
cddd4ac7 MK |
709 | { |
710 | Coroutine *co; | |
711 | SheepdogReqCo srco = { | |
712 | .sockfd = sockfd, | |
84390bed | 713 | .aio_context = aio_context, |
cddd4ac7 MK |
714 | .hdr = hdr, |
715 | .data = data, | |
716 | .wlen = wlen, | |
717 | .rlen = rlen, | |
718 | .ret = 0, | |
719 | .finished = false, | |
720 | }; | |
721 | ||
722 | if (qemu_in_coroutine()) { | |
723 | do_co_req(&srco); | |
724 | } else { | |
725 | co = qemu_coroutine_create(do_co_req); | |
726 | qemu_coroutine_enter(co, &srco); | |
727 | while (!srco.finished) { | |
84390bed | 728 | aio_poll(aio_context, true); |
cddd4ac7 MK |
729 | } |
730 | } | |
731 | ||
732 | return srco.ret; | |
47622c44 LY |
733 | } |
734 | ||
a37dcdf9 | 735 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
736 | struct iovec *iov, int niov, |
737 | enum AIOCBState aiocb_type); | |
a37dcdf9 | 738 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req); |
72e0996c | 739 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag); |
356b4ca2 | 740 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp); |
011603ca | 741 | static void co_write_request(void *opaque); |
7dc1cde0 | 742 | |
011603ca MK |
743 | static coroutine_fn void reconnect_to_sdog(void *opaque) |
744 | { | |
745 | BDRVSheepdogState *s = opaque; | |
746 | AIOReq *aio_req, *next; | |
747 | ||
dca21ef2 FZ |
748 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, |
749 | NULL, NULL); | |
011603ca MK |
750 | close(s->fd); |
751 | s->fd = -1; | |
752 | ||
753 | /* Wait for outstanding write requests to be completed. */ | |
754 | while (s->co_send != NULL) { | |
755 | co_write_request(opaque); | |
756 | } | |
757 | ||
758 | /* Try to reconnect the sheepdog server every one second. */ | |
759 | while (s->fd < 0) { | |
a780dea0 | 760 | Error *local_err = NULL; |
356b4ca2 | 761 | s->fd = get_sheep_fd(s, &local_err); |
011603ca MK |
762 | if (s->fd < 0) { |
763 | DPRINTF("Wait for connection to be established\n"); | |
565f65d2 | 764 | error_report_err(local_err); |
011603ca MK |
765 | co_aio_sleep_ns(bdrv_get_aio_context(s->bs), QEMU_CLOCK_REALTIME, |
766 | 1000000000ULL); | |
767 | } | |
768 | }; | |
769 | ||
770 | /* | |
771 | * Now we have to resend all the request in the inflight queue. However, | |
772 | * resend_aioreq() can yield and newly created requests can be added to the | |
773 | * inflight queue before the coroutine is resumed. To avoid mixing them, we | |
774 | * have to move all the inflight requests to the failed queue before | |
775 | * resend_aioreq() is called. | |
776 | */ | |
777 | QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) { | |
778 | QLIST_REMOVE(aio_req, aio_siblings); | |
779 | QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings); | |
780 | } | |
781 | ||
782 | /* Resend all the failed aio requests. */ | |
783 | while (!QLIST_EMPTY(&s->failed_aio_head)) { | |
784 | aio_req = QLIST_FIRST(&s->failed_aio_head); | |
785 | QLIST_REMOVE(aio_req, aio_siblings); | |
786 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
787 | resend_aioreq(s, aio_req); | |
788 | } | |
789 | } | |
790 | ||
33b1db1c MK |
791 | /* |
792 | * Receive responses of the I/O requests. | |
793 | * | |
794 | * This function is registered as a fd handler, and called from the | |
795 | * main loop when s->fd is ready for reading responses. | |
796 | */ | |
d8716b41 | 797 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
798 | { |
799 | SheepdogObjRsp rsp; | |
800 | BDRVSheepdogState *s = opaque; | |
801 | int fd = s->fd; | |
802 | int ret; | |
803 | AIOReq *aio_req = NULL; | |
804 | SheepdogAIOCB *acb; | |
cac8f4a6 | 805 | uint64_t idx; |
33b1db1c | 806 | |
33b1db1c | 807 | /* read a header */ |
8c5135f9 | 808 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
80731d9d | 809 | if (ret != sizeof(rsp)) { |
6daf194d | 810 | error_report("failed to get the header, %s", strerror(errno)); |
011603ca | 811 | goto err; |
33b1db1c MK |
812 | } |
813 | ||
c292ee6a MK |
814 | /* find the right aio_req from the inflight aio list */ |
815 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
33b1db1c MK |
816 | if (aio_req->id == rsp.id) { |
817 | break; | |
818 | } | |
819 | } | |
820 | if (!aio_req) { | |
6daf194d | 821 | error_report("cannot find aio_req %x", rsp.id); |
011603ca | 822 | goto err; |
33b1db1c MK |
823 | } |
824 | ||
825 | acb = aio_req->aiocb; | |
826 | ||
827 | switch (acb->aiocb_type) { | |
828 | case AIOCB_WRITE_UDATA: | |
6d1acda8 MK |
829 | /* this coroutine context is no longer suitable for co_recv |
830 | * because we may send data to update vdi objects */ | |
831 | s->co_recv = NULL; | |
33b1db1c MK |
832 | if (!is_data_obj(aio_req->oid)) { |
833 | break; | |
834 | } | |
835 | idx = data_oid_to_idx(aio_req->oid); | |
836 | ||
b544c1ab | 837 | if (aio_req->create) { |
33b1db1c MK |
838 | /* |
839 | * If the object is newly created one, we need to update | |
840 | * the vdi object (metadata object). min_dirty_data_idx | |
841 | * and max_dirty_data_idx are changed to include updated | |
842 | * index between them. | |
843 | */ | |
bd751f22 LY |
844 | if (rsp.result == SD_RES_SUCCESS) { |
845 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
498f2140 HM |
846 | acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx); |
847 | acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx); | |
bd751f22 | 848 | } |
33b1db1c MK |
849 | } |
850 | break; | |
851 | case AIOCB_READ_UDATA: | |
2fc8ae1d MT |
852 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
853 | aio_req->iov_offset, rsp.data_length); | |
80731d9d | 854 | if (ret != rsp.data_length) { |
6daf194d | 855 | error_report("failed to get the data, %s", strerror(errno)); |
011603ca | 856 | goto err; |
33b1db1c MK |
857 | } |
858 | break; | |
47783072 LY |
859 | case AIOCB_FLUSH_CACHE: |
860 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
2440a2c3 | 861 | DPRINTF("disable cache since the server doesn't support it\n"); |
47783072 LY |
862 | s->cache_flags = SD_FLAG_CMD_DIRECT; |
863 | rsp.result = SD_RES_SUCCESS; | |
864 | } | |
865 | break; | |
cac8f4a6 LY |
866 | case AIOCB_DISCARD_OBJ: |
867 | switch (rsp.result) { | |
868 | case SD_RES_INVALID_PARMS: | |
869 | error_report("sheep(%s) doesn't support discard command", | |
870 | s->host_spec); | |
871 | rsp.result = SD_RES_SUCCESS; | |
872 | s->discard_supported = false; | |
873 | break; | |
cac8f4a6 LY |
874 | default: |
875 | break; | |
876 | } | |
33b1db1c MK |
877 | } |
878 | ||
13c31de2 MK |
879 | switch (rsp.result) { |
880 | case SD_RES_SUCCESS: | |
881 | break; | |
882 | case SD_RES_READONLY: | |
72e0996c MK |
883 | if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) { |
884 | ret = reload_inode(s, 0, ""); | |
885 | if (ret < 0) { | |
011603ca | 886 | goto err; |
72e0996c MK |
887 | } |
888 | } | |
72e0996c MK |
889 | if (is_data_obj(aio_req->oid)) { |
890 | aio_req->oid = vid_to_data_oid(s->inode.vdi_id, | |
891 | data_oid_to_idx(aio_req->oid)); | |
892 | } else { | |
893 | aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id); | |
894 | } | |
a37dcdf9 MK |
895 | resend_aioreq(s, aio_req); |
896 | goto out; | |
13c31de2 | 897 | default: |
33b1db1c | 898 | acb->ret = -EIO; |
6daf194d | 899 | error_report("%s", sd_strerror(rsp.result)); |
13c31de2 | 900 | break; |
33b1db1c MK |
901 | } |
902 | ||
1d732d7d MK |
903 | free_aio_req(s, aio_req); |
904 | if (!acb->nr_pending) { | |
33b1db1c MK |
905 | /* |
906 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 907 | * we can switch back to sd_co_readv/writev now. |
33b1db1c MK |
908 | */ |
909 | acb->aio_done_func(acb); | |
910 | } | |
2df46246 MK |
911 | out: |
912 | s->co_recv = NULL; | |
011603ca MK |
913 | return; |
914 | err: | |
915 | s->co_recv = NULL; | |
916 | reconnect_to_sdog(opaque); | |
2df46246 MK |
917 | } |
918 | ||
919 | static void co_read_response(void *opaque) | |
920 | { | |
921 | BDRVSheepdogState *s = opaque; | |
922 | ||
923 | if (!s->co_recv) { | |
924 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
925 | } | |
926 | ||
927 | qemu_coroutine_enter(s->co_recv, opaque); | |
928 | } | |
929 | ||
930 | static void co_write_request(void *opaque) | |
931 | { | |
932 | BDRVSheepdogState *s = opaque; | |
933 | ||
934 | qemu_coroutine_enter(s->co_send, NULL); | |
33b1db1c MK |
935 | } |
936 | ||
33b1db1c | 937 | /* |
dc6fb73d | 938 | * Return a socket descriptor to read/write objects. |
33b1db1c | 939 | * |
dc6fb73d | 940 | * We cannot use this descriptor for other operations because |
33b1db1c MK |
941 | * the block driver may be on waiting response from the server. |
942 | */ | |
356b4ca2 | 943 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp) |
33b1db1c | 944 | { |
1b8bbb46 | 945 | int fd; |
33b1db1c | 946 | |
356b4ca2 | 947 | fd = connect_to_sdog(s, errp); |
33b1db1c | 948 | if (fd < 0) { |
cb595887 | 949 | return fd; |
33b1db1c MK |
950 | } |
951 | ||
dca21ef2 FZ |
952 | aio_set_fd_handler(s->aio_context, fd, false, |
953 | co_read_response, NULL, s); | |
33b1db1c MK |
954 | return fd; |
955 | } | |
956 | ||
5d6768e3 MK |
957 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
958 | char *vdi, uint32_t *snapid, char *tag) | |
959 | { | |
960 | URI *uri; | |
961 | QueryParams *qp = NULL; | |
962 | int ret = 0; | |
963 | ||
964 | uri = uri_parse(filename); | |
965 | if (!uri) { | |
966 | return -EINVAL; | |
967 | } | |
968 | ||
1b8bbb46 MK |
969 | /* transport */ |
970 | if (!strcmp(uri->scheme, "sheepdog")) { | |
971 | s->is_unix = false; | |
972 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
973 | s->is_unix = false; | |
974 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
975 | s->is_unix = true; | |
976 | } else { | |
977 | ret = -EINVAL; | |
978 | goto out; | |
979 | } | |
980 | ||
5d6768e3 MK |
981 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
982 | ret = -EINVAL; | |
983 | goto out; | |
984 | } | |
985 | pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1); | |
986 | ||
1b8bbb46 MK |
987 | qp = query_params_parse(uri->query); |
988 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
989 | ret = -EINVAL; | |
990 | goto out; | |
991 | } | |
992 | ||
993 | if (s->is_unix) { | |
994 | /* sheepdog+unix:///vdiname?socket=path */ | |
995 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
996 | ret = -EINVAL; | |
997 | goto out; | |
998 | } | |
999 | s->host_spec = g_strdup(qp->p[0].value); | |
1000 | } else { | |
1001 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
1002 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
1003 | uri->port ?: SD_DEFAULT_PORT); | |
1004 | } | |
5d6768e3 MK |
1005 | |
1006 | /* snapshot tag */ | |
1007 | if (uri->fragment) { | |
1008 | *snapid = strtoul(uri->fragment, NULL, 10); | |
1009 | if (*snapid == 0) { | |
1010 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); | |
1011 | } | |
1012 | } else { | |
1013 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
1014 | } | |
1015 | ||
1016 | out: | |
1017 | if (qp) { | |
1018 | query_params_free(qp); | |
1019 | } | |
1020 | uri_free(uri); | |
1021 | return ret; | |
1022 | } | |
1023 | ||
33b1db1c | 1024 | /* |
5d6768e3 | 1025 | * Parse a filename (old syntax) |
33b1db1c MK |
1026 | * |
1027 | * filename must be one of the following formats: | |
1028 | * 1. [vdiname] | |
1029 | * 2. [vdiname]:[snapid] | |
1030 | * 3. [vdiname]:[tag] | |
1031 | * 4. [hostname]:[port]:[vdiname] | |
1032 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
1033 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
1034 | * | |
1035 | * You can boot from the snapshot images by specifying `snapid` or | |
1036 | * `tag'. | |
1037 | * | |
1038 | * You can run VMs outside the Sheepdog cluster by specifying | |
1039 | * `hostname' and `port' (experimental). | |
1040 | */ | |
1041 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
1042 | char *vdi, uint32_t *snapid, char *tag) | |
1043 | { | |
5d6768e3 MK |
1044 | char *p, *q, *uri; |
1045 | const char *host_spec, *vdi_spec; | |
1046 | int nr_sep, ret; | |
33b1db1c | 1047 | |
5d6768e3 | 1048 | strstart(filename, "sheepdog:", (const char **)&filename); |
7267c094 | 1049 | p = q = g_strdup(filename); |
33b1db1c MK |
1050 | |
1051 | /* count the number of separators */ | |
1052 | nr_sep = 0; | |
1053 | while (*p) { | |
1054 | if (*p == ':') { | |
1055 | nr_sep++; | |
1056 | } | |
1057 | p++; | |
1058 | } | |
1059 | p = q; | |
1060 | ||
5d6768e3 | 1061 | /* use the first two tokens as host_spec. */ |
33b1db1c | 1062 | if (nr_sep >= 2) { |
5d6768e3 | 1063 | host_spec = p; |
33b1db1c | 1064 | p = strchr(p, ':'); |
5d6768e3 | 1065 | p++; |
33b1db1c MK |
1066 | p = strchr(p, ':'); |
1067 | *p++ = '\0'; | |
1068 | } else { | |
5d6768e3 | 1069 | host_spec = ""; |
33b1db1c MK |
1070 | } |
1071 | ||
5d6768e3 | 1072 | vdi_spec = p; |
33b1db1c | 1073 | |
5d6768e3 | 1074 | p = strchr(vdi_spec, ':'); |
33b1db1c | 1075 | if (p) { |
5d6768e3 | 1076 | *p++ = '#'; |
33b1db1c MK |
1077 | } |
1078 | ||
5d6768e3 | 1079 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 1080 | |
5d6768e3 MK |
1081 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
1082 | ||
1083 | g_free(q); | |
1084 | g_free(uri); | |
1085 | ||
1086 | return ret; | |
33b1db1c MK |
1087 | } |
1088 | ||
982dcbf4 MK |
1089 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, |
1090 | uint32_t snapid, const char *tag, uint32_t *vid, | |
dc83cd42 | 1091 | bool lock, Error **errp) |
33b1db1c MK |
1092 | { |
1093 | int ret, fd; | |
1094 | SheepdogVdiReq hdr; | |
1095 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1096 | unsigned int wlen, rlen = 0; | |
1097 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
1098 | ||
dc83cd42 | 1099 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1100 | if (fd < 0) { |
cb595887 | 1101 | return fd; |
33b1db1c MK |
1102 | } |
1103 | ||
3178e275 JM |
1104 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
1105 | * which is desirable since we'll soon be sending those bytes, and | |
1106 | * don't want the send_req to read uninitialized data. | |
1107 | */ | |
33b1db1c MK |
1108 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
1109 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
1110 | ||
1111 | memset(&hdr, 0, sizeof(hdr)); | |
982dcbf4 | 1112 | if (lock) { |
33b1db1c | 1113 | hdr.opcode = SD_OP_LOCK_VDI; |
1dbfafed | 1114 | hdr.type = LOCK_TYPE_NORMAL; |
982dcbf4 MK |
1115 | } else { |
1116 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
33b1db1c MK |
1117 | } |
1118 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
1119 | hdr.proto_ver = SD_PROTO_VER; | |
1120 | hdr.data_length = wlen; | |
1121 | hdr.snapid = snapid; | |
1122 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1123 | ||
84390bed | 1124 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1125 | if (ret) { |
dc83cd42 | 1126 | error_setg_errno(errp, -ret, "cannot get vdi info"); |
33b1db1c MK |
1127 | goto out; |
1128 | } | |
1129 | ||
1130 | if (rsp->result != SD_RES_SUCCESS) { | |
dc83cd42 MA |
1131 | error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s", |
1132 | sd_strerror(rsp->result), filename, snapid, tag); | |
cb595887 MK |
1133 | if (rsp->result == SD_RES_NO_VDI) { |
1134 | ret = -ENOENT; | |
38890b24 HM |
1135 | } else if (rsp->result == SD_RES_VDI_LOCKED) { |
1136 | ret = -EBUSY; | |
cb595887 MK |
1137 | } else { |
1138 | ret = -EIO; | |
1139 | } | |
33b1db1c MK |
1140 | goto out; |
1141 | } | |
1142 | *vid = rsp->vdi_id; | |
1143 | ||
1144 | ret = 0; | |
1145 | out: | |
1146 | closesocket(fd); | |
1147 | return ret; | |
1148 | } | |
1149 | ||
a37dcdf9 | 1150 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
b544c1ab HM |
1151 | struct iovec *iov, int niov, |
1152 | enum AIOCBState aiocb_type) | |
33b1db1c MK |
1153 | { |
1154 | int nr_copies = s->inode.nr_copies; | |
1155 | SheepdogObjReq hdr; | |
47783072 | 1156 | unsigned int wlen = 0; |
33b1db1c MK |
1157 | int ret; |
1158 | uint64_t oid = aio_req->oid; | |
1159 | unsigned int datalen = aio_req->data_len; | |
1160 | uint64_t offset = aio_req->offset; | |
1161 | uint8_t flags = aio_req->flags; | |
1162 | uint64_t old_oid = aio_req->base_oid; | |
b544c1ab | 1163 | bool create = aio_req->create; |
33b1db1c MK |
1164 | |
1165 | if (!nr_copies) { | |
6daf194d | 1166 | error_report("bug"); |
33b1db1c MK |
1167 | } |
1168 | ||
1169 | memset(&hdr, 0, sizeof(hdr)); | |
1170 | ||
47783072 LY |
1171 | switch (aiocb_type) { |
1172 | case AIOCB_FLUSH_CACHE: | |
1173 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1174 | break; | |
1175 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1176 | hdr.opcode = SD_OP_READ_OBJ; |
1177 | hdr.flags = flags; | |
47783072 LY |
1178 | break; |
1179 | case AIOCB_WRITE_UDATA: | |
1180 | if (create) { | |
1181 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1182 | } else { | |
1183 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1184 | } | |
33b1db1c | 1185 | wlen = datalen; |
33b1db1c | 1186 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1187 | break; |
cac8f4a6 | 1188 | case AIOCB_DISCARD_OBJ: |
e6fd57ea HM |
1189 | hdr.opcode = SD_OP_WRITE_OBJ; |
1190 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
1191 | s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0; | |
1192 | offset = offsetof(SheepdogInode, | |
1193 | data_vdi_id[data_oid_to_idx(oid)]); | |
1194 | oid = vid_to_vdi_oid(s->inode.vdi_id); | |
1195 | wlen = datalen = sizeof(uint32_t); | |
cac8f4a6 | 1196 | break; |
33b1db1c MK |
1197 | } |
1198 | ||
0e7106d8 LY |
1199 | if (s->cache_flags) { |
1200 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1201 | } |
1202 | ||
33b1db1c MK |
1203 | hdr.oid = oid; |
1204 | hdr.cow_oid = old_oid; | |
1205 | hdr.copies = s->inode.nr_copies; | |
1206 | ||
1207 | hdr.data_length = datalen; | |
1208 | hdr.offset = offset; | |
1209 | ||
1210 | hdr.id = aio_req->id; | |
1211 | ||
2df46246 MK |
1212 | qemu_co_mutex_lock(&s->lock); |
1213 | s->co_send = qemu_coroutine_self(); | |
dca21ef2 | 1214 | aio_set_fd_handler(s->aio_context, s->fd, false, |
84390bed | 1215 | co_read_response, co_write_request, s); |
128aa589 | 1216 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1217 | |
1218 | /* send a header */ | |
8c5135f9 | 1219 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
80731d9d | 1220 | if (ret != sizeof(hdr)) { |
6daf194d | 1221 | error_report("failed to send a req, %s", strerror(errno)); |
011603ca | 1222 | goto out; |
33b1db1c MK |
1223 | } |
1224 | ||
1225 | if (wlen) { | |
2fc8ae1d | 1226 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
80731d9d | 1227 | if (ret != wlen) { |
6daf194d | 1228 | error_report("failed to send a data, %s", strerror(errno)); |
33b1db1c MK |
1229 | } |
1230 | } | |
011603ca | 1231 | out: |
128aa589 | 1232 | socket_set_cork(s->fd, 0); |
dca21ef2 FZ |
1233 | aio_set_fd_handler(s->aio_context, s->fd, false, |
1234 | co_read_response, NULL, s); | |
011603ca | 1235 | s->co_send = NULL; |
2df46246 | 1236 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1237 | } |
1238 | ||
84390bed SH |
1239 | static int read_write_object(int fd, AioContext *aio_context, char *buf, |
1240 | uint64_t oid, uint8_t copies, | |
33b1db1c | 1241 | unsigned int datalen, uint64_t offset, |
0e7106d8 | 1242 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1243 | { |
1244 | SheepdogObjReq hdr; | |
1245 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1246 | unsigned int wlen, rlen; | |
1247 | int ret; | |
1248 | ||
1249 | memset(&hdr, 0, sizeof(hdr)); | |
1250 | ||
1251 | if (write) { | |
1252 | wlen = datalen; | |
1253 | rlen = 0; | |
1254 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1255 | if (create) { | |
1256 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1257 | } else { | |
1258 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1259 | } | |
1260 | } else { | |
1261 | wlen = 0; | |
1262 | rlen = datalen; | |
1263 | hdr.opcode = SD_OP_READ_OBJ; | |
1264 | } | |
47622c44 | 1265 | |
0e7106d8 | 1266 | hdr.flags |= cache_flags; |
47622c44 | 1267 | |
33b1db1c MK |
1268 | hdr.oid = oid; |
1269 | hdr.data_length = datalen; | |
1270 | hdr.offset = offset; | |
1271 | hdr.copies = copies; | |
1272 | ||
84390bed | 1273 | ret = do_req(fd, aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c | 1274 | if (ret) { |
6daf194d | 1275 | error_report("failed to send a request to the sheep"); |
cb595887 | 1276 | return ret; |
33b1db1c MK |
1277 | } |
1278 | ||
1279 | switch (rsp->result) { | |
1280 | case SD_RES_SUCCESS: | |
1281 | return 0; | |
1282 | default: | |
6daf194d | 1283 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1284 | return -EIO; |
33b1db1c MK |
1285 | } |
1286 | } | |
1287 | ||
84390bed SH |
1288 | static int read_object(int fd, AioContext *aio_context, char *buf, |
1289 | uint64_t oid, uint8_t copies, | |
0e7106d8 LY |
1290 | unsigned int datalen, uint64_t offset, |
1291 | uint32_t cache_flags) | |
33b1db1c | 1292 | { |
84390bed SH |
1293 | return read_write_object(fd, aio_context, buf, oid, copies, |
1294 | datalen, offset, false, | |
0e7106d8 | 1295 | false, cache_flags); |
33b1db1c MK |
1296 | } |
1297 | ||
84390bed SH |
1298 | static int write_object(int fd, AioContext *aio_context, char *buf, |
1299 | uint64_t oid, uint8_t copies, | |
2f536801 | 1300 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1301 | uint32_t cache_flags) |
33b1db1c | 1302 | { |
84390bed SH |
1303 | return read_write_object(fd, aio_context, buf, oid, copies, |
1304 | datalen, offset, true, | |
0e7106d8 | 1305 | create, cache_flags); |
33b1db1c MK |
1306 | } |
1307 | ||
9ff53a0e MK |
1308 | /* update inode with the latest state */ |
1309 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag) | |
1310 | { | |
dfb12bf8 | 1311 | Error *local_err = NULL; |
9ff53a0e MK |
1312 | SheepdogInode *inode; |
1313 | int ret = 0, fd; | |
1314 | uint32_t vid = 0; | |
1315 | ||
dfb12bf8 | 1316 | fd = connect_to_sdog(s, &local_err); |
9ff53a0e | 1317 | if (fd < 0) { |
565f65d2 | 1318 | error_report_err(local_err); |
9ff53a0e MK |
1319 | return -EIO; |
1320 | } | |
1321 | ||
5d039bab | 1322 | inode = g_malloc(SD_INODE_HEADER_SIZE); |
9ff53a0e | 1323 | |
dc83cd42 | 1324 | ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err); |
9ff53a0e | 1325 | if (ret) { |
565f65d2 | 1326 | error_report_err(local_err); |
9ff53a0e MK |
1327 | goto out; |
1328 | } | |
1329 | ||
84390bed | 1330 | ret = read_object(fd, s->aio_context, (char *)inode, vid_to_vdi_oid(vid), |
5d039bab HM |
1331 | s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0, |
1332 | s->cache_flags); | |
9ff53a0e MK |
1333 | if (ret < 0) { |
1334 | goto out; | |
1335 | } | |
1336 | ||
1337 | if (inode->vdi_id != s->inode.vdi_id) { | |
5d039bab | 1338 | memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE); |
9ff53a0e MK |
1339 | } |
1340 | ||
1341 | out: | |
1342 | g_free(inode); | |
1343 | closesocket(fd); | |
1344 | ||
1345 | return ret; | |
1346 | } | |
1347 | ||
a37dcdf9 | 1348 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req) |
13c31de2 MK |
1349 | { |
1350 | SheepdogAIOCB *acb = aio_req->aiocb; | |
b544c1ab HM |
1351 | |
1352 | aio_req->create = false; | |
13c31de2 MK |
1353 | |
1354 | /* check whether this request becomes a CoW one */ | |
2412aec7 | 1355 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) { |
13c31de2 | 1356 | int idx = data_oid_to_idx(aio_req->oid); |
13c31de2 | 1357 | |
13c31de2 MK |
1358 | if (is_data_obj_writable(&s->inode, idx)) { |
1359 | goto out; | |
1360 | } | |
1361 | ||
80308d33 MK |
1362 | if (s->inode.data_vdi_id[idx]) { |
1363 | aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx); | |
1364 | aio_req->flags |= SD_FLAG_CMD_COW; | |
1365 | } | |
b544c1ab | 1366 | aio_req->create = true; |
13c31de2 MK |
1367 | } |
1368 | out: | |
2412aec7 | 1369 | if (is_data_obj(aio_req->oid)) { |
b544c1ab | 1370 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
a37dcdf9 | 1371 | acb->aiocb_type); |
2412aec7 MK |
1372 | } else { |
1373 | struct iovec iov; | |
1374 | iov.iov_base = &s->inode; | |
1375 | iov.iov_len = sizeof(s->inode); | |
b544c1ab | 1376 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
2412aec7 | 1377 | } |
13c31de2 MK |
1378 | } |
1379 | ||
84390bed SH |
1380 | static void sd_detach_aio_context(BlockDriverState *bs) |
1381 | { | |
1382 | BDRVSheepdogState *s = bs->opaque; | |
1383 | ||
dca21ef2 FZ |
1384 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, |
1385 | NULL, NULL); | |
84390bed SH |
1386 | } |
1387 | ||
1388 | static void sd_attach_aio_context(BlockDriverState *bs, | |
1389 | AioContext *new_context) | |
1390 | { | |
1391 | BDRVSheepdogState *s = bs->opaque; | |
1392 | ||
1393 | s->aio_context = new_context; | |
dca21ef2 FZ |
1394 | aio_set_fd_handler(new_context, s->fd, false, |
1395 | co_read_response, NULL, s); | |
84390bed SH |
1396 | } |
1397 | ||
c8c96350 KW |
1398 | /* TODO Convert to fine grained options */ |
1399 | static QemuOptsList runtime_opts = { | |
1400 | .name = "sheepdog", | |
1401 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1402 | .desc = { | |
1403 | { | |
1404 | .name = "filename", | |
1405 | .type = QEMU_OPT_STRING, | |
1406 | .help = "URL to the sheepdog image", | |
1407 | }, | |
1408 | { /* end of list */ } | |
1409 | }, | |
1410 | }; | |
1411 | ||
015a1036 HR |
1412 | static int sd_open(BlockDriverState *bs, QDict *options, int flags, |
1413 | Error **errp) | |
33b1db1c MK |
1414 | { |
1415 | int ret, fd; | |
1416 | uint32_t vid = 0; | |
1417 | BDRVSheepdogState *s = bs->opaque; | |
1418 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1419 | uint32_t snapid; | |
1420 | char *buf = NULL; | |
c8c96350 KW |
1421 | QemuOpts *opts; |
1422 | Error *local_err = NULL; | |
1423 | const char *filename; | |
1424 | ||
011603ca | 1425 | s->bs = bs; |
84390bed | 1426 | s->aio_context = bdrv_get_aio_context(bs); |
011603ca | 1427 | |
87ea75d5 | 1428 | opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); |
c8c96350 | 1429 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 1430 | if (local_err) { |
e67c3993 | 1431 | error_propagate(errp, local_err); |
c8c96350 KW |
1432 | ret = -EINVAL; |
1433 | goto out; | |
1434 | } | |
1435 | ||
1436 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1437 | |
c292ee6a | 1438 | QLIST_INIT(&s->inflight_aio_head); |
011603ca | 1439 | QLIST_INIT(&s->failed_aio_head); |
6a55c82c | 1440 | QLIST_INIT(&s->inflight_aiocb_head); |
33b1db1c MK |
1441 | s->fd = -1; |
1442 | ||
1443 | memset(vdi, 0, sizeof(vdi)); | |
1444 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1445 | |
1446 | if (strstr(filename, "://")) { | |
1447 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1448 | } else { | |
1449 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1450 | } | |
1451 | if (ret < 0) { | |
efde4b62 | 1452 | error_setg(errp, "Can't parse filename"); |
33b1db1c MK |
1453 | goto out; |
1454 | } | |
e67c3993 | 1455 | s->fd = get_sheep_fd(s, errp); |
33b1db1c | 1456 | if (s->fd < 0) { |
cb595887 | 1457 | ret = s->fd; |
33b1db1c MK |
1458 | goto out; |
1459 | } | |
1460 | ||
e67c3993 | 1461 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, true, errp); |
33b1db1c MK |
1462 | if (ret) { |
1463 | goto out; | |
1464 | } | |
1465 | ||
0e7106d8 LY |
1466 | /* |
1467 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1468 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1469 | */ | |
1470 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1471 | if (flags & BDRV_O_NOCACHE) { | |
1472 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1473 | } | |
cac8f4a6 | 1474 | s->discard_supported = true; |
0e7106d8 | 1475 | |
622b6057 | 1476 | if (snapid || tag[0] != '\0') { |
2440a2c3 | 1477 | DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1478 | s->is_snapshot = true; |
33b1db1c MK |
1479 | } |
1480 | ||
e67c3993 | 1481 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1482 | if (fd < 0) { |
cb595887 | 1483 | ret = fd; |
33b1db1c MK |
1484 | goto out; |
1485 | } | |
1486 | ||
7267c094 | 1487 | buf = g_malloc(SD_INODE_SIZE); |
84390bed SH |
1488 | ret = read_object(fd, s->aio_context, buf, vid_to_vdi_oid(vid), |
1489 | 0, SD_INODE_SIZE, 0, s->cache_flags); | |
33b1db1c MK |
1490 | |
1491 | closesocket(fd); | |
1492 | ||
1493 | if (ret) { | |
efde4b62 | 1494 | error_setg(errp, "Can't read snapshot inode"); |
33b1db1c MK |
1495 | goto out; |
1496 | } | |
1497 | ||
1498 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
33b1db1c | 1499 | |
e8bfaa2f | 1500 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; |
3178e275 | 1501 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1502 | qemu_co_mutex_init(&s->lock); |
498f2140 | 1503 | qemu_co_queue_init(&s->overlapping_queue); |
c8c96350 | 1504 | qemu_opts_del(opts); |
7267c094 | 1505 | g_free(buf); |
33b1db1c MK |
1506 | return 0; |
1507 | out: | |
dca21ef2 FZ |
1508 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, |
1509 | false, NULL, NULL, NULL); | |
33b1db1c MK |
1510 | if (s->fd >= 0) { |
1511 | closesocket(s->fd); | |
1512 | } | |
c8c96350 | 1513 | qemu_opts_del(opts); |
7267c094 | 1514 | g_free(buf); |
cb595887 | 1515 | return ret; |
33b1db1c MK |
1516 | } |
1517 | ||
4da65c80 LY |
1518 | static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue, |
1519 | Error **errp) | |
1520 | { | |
1521 | BDRVSheepdogState *s = state->bs->opaque; | |
1522 | BDRVSheepdogReopenState *re_s; | |
1523 | int ret = 0; | |
1524 | ||
1525 | re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1); | |
1526 | ||
1527 | re_s->cache_flags = SD_FLAG_CMD_CACHE; | |
1528 | if (state->flags & BDRV_O_NOCACHE) { | |
1529 | re_s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1530 | } | |
1531 | ||
1532 | re_s->fd = get_sheep_fd(s, errp); | |
1533 | if (re_s->fd < 0) { | |
1534 | ret = re_s->fd; | |
1535 | return ret; | |
1536 | } | |
1537 | ||
1538 | return ret; | |
1539 | } | |
1540 | ||
1541 | static void sd_reopen_commit(BDRVReopenState *state) | |
1542 | { | |
1543 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1544 | BDRVSheepdogState *s = state->bs->opaque; | |
1545 | ||
1546 | if (s->fd) { | |
dca21ef2 FZ |
1547 | aio_set_fd_handler(s->aio_context, s->fd, false, |
1548 | NULL, NULL, NULL); | |
4da65c80 LY |
1549 | closesocket(s->fd); |
1550 | } | |
1551 | ||
1552 | s->fd = re_s->fd; | |
1553 | s->cache_flags = re_s->cache_flags; | |
1554 | ||
1555 | g_free(state->opaque); | |
1556 | state->opaque = NULL; | |
1557 | ||
1558 | return; | |
1559 | } | |
1560 | ||
1561 | static void sd_reopen_abort(BDRVReopenState *state) | |
1562 | { | |
1563 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1564 | BDRVSheepdogState *s = state->bs->opaque; | |
1565 | ||
1566 | if (re_s == NULL) { | |
1567 | return; | |
1568 | } | |
1569 | ||
1570 | if (re_s->fd) { | |
dca21ef2 FZ |
1571 | aio_set_fd_handler(s->aio_context, re_s->fd, false, |
1572 | NULL, NULL, NULL); | |
4da65c80 LY |
1573 | closesocket(re_s->fd); |
1574 | } | |
1575 | ||
1576 | g_free(state->opaque); | |
1577 | state->opaque = NULL; | |
1578 | ||
1579 | return; | |
1580 | } | |
1581 | ||
7d2d3e74 MA |
1582 | static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot, |
1583 | Error **errp) | |
33b1db1c MK |
1584 | { |
1585 | SheepdogVdiReq hdr; | |
1586 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1587 | int fd, ret; | |
1588 | unsigned int wlen, rlen = 0; | |
1589 | char buf[SD_MAX_VDI_LEN]; | |
1590 | ||
7d2d3e74 | 1591 | fd = connect_to_sdog(s, errp); |
33b1db1c | 1592 | if (fd < 0) { |
cb595887 | 1593 | return fd; |
33b1db1c MK |
1594 | } |
1595 | ||
3178e275 JM |
1596 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1597 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1598 | */ | |
33b1db1c | 1599 | memset(buf, 0, sizeof(buf)); |
c31d482f | 1600 | pstrcpy(buf, sizeof(buf), s->name); |
33b1db1c MK |
1601 | |
1602 | memset(&hdr, 0, sizeof(hdr)); | |
1603 | hdr.opcode = SD_OP_NEW_VDI; | |
9f23fce7 | 1604 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1605 | |
1606 | wlen = SD_MAX_VDI_LEN; | |
1607 | ||
1608 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1609 | hdr.snapid = snapshot; | |
1610 | ||
1611 | hdr.data_length = wlen; | |
c31d482f LY |
1612 | hdr.vdi_size = s->inode.vdi_size; |
1613 | hdr.copy_policy = s->inode.copy_policy; | |
b3af018f | 1614 | hdr.copies = s->inode.nr_copies; |
876eb1b0 | 1615 | hdr.block_size_shift = s->inode.block_size_shift; |
33b1db1c | 1616 | |
84390bed | 1617 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
33b1db1c MK |
1618 | |
1619 | closesocket(fd); | |
1620 | ||
1621 | if (ret) { | |
7d2d3e74 | 1622 | error_setg_errno(errp, -ret, "create failed"); |
cb595887 | 1623 | return ret; |
33b1db1c MK |
1624 | } |
1625 | ||
1626 | if (rsp->result != SD_RES_SUCCESS) { | |
7d2d3e74 | 1627 | error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name); |
33b1db1c MK |
1628 | return -EIO; |
1629 | } | |
1630 | ||
1631 | if (vdi_id) { | |
1632 | *vdi_id = rsp->vdi_id; | |
1633 | } | |
1634 | ||
1635 | return 0; | |
1636 | } | |
1637 | ||
318df29e | 1638 | static int sd_prealloc(const char *filename, Error **errp) |
a8e0fdd7 | 1639 | { |
fba98d45 | 1640 | BlockBackend *blk = NULL; |
876eb1b0 TI |
1641 | BDRVSheepdogState *base = NULL; |
1642 | unsigned long buf_size; | |
a8e0fdd7 | 1643 | uint32_t idx, max_idx; |
876eb1b0 | 1644 | uint32_t object_size; |
a8e0fdd7 | 1645 | int64_t vdi_size; |
876eb1b0 | 1646 | void *buf = NULL; |
a8e0fdd7 MK |
1647 | int ret; |
1648 | ||
fba98d45 KW |
1649 | blk = blk_new_open("image-prealloc", filename, NULL, NULL, |
1650 | BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL, | |
1651 | errp); | |
1652 | if (blk == NULL) { | |
1653 | ret = -EIO; | |
318df29e | 1654 | goto out_with_err_set; |
a8e0fdd7 MK |
1655 | } |
1656 | ||
fba98d45 KW |
1657 | blk_set_allow_write_beyond_eof(blk, true); |
1658 | ||
1659 | vdi_size = blk_getlength(blk); | |
a8e0fdd7 MK |
1660 | if (vdi_size < 0) { |
1661 | ret = vdi_size; | |
1662 | goto out; | |
1663 | } | |
876eb1b0 | 1664 | |
fba98d45 | 1665 | base = blk_bs(blk)->opaque; |
876eb1b0 TI |
1666 | object_size = (UINT32_C(1) << base->inode.block_size_shift); |
1667 | buf_size = MIN(object_size, SD_DATA_OBJ_SIZE); | |
1668 | buf = g_malloc0(buf_size); | |
1669 | ||
1670 | max_idx = DIV_ROUND_UP(vdi_size, buf_size); | |
a8e0fdd7 MK |
1671 | |
1672 | for (idx = 0; idx < max_idx; idx++) { | |
1673 | /* | |
1674 | * The created image can be a cloned image, so we need to read | |
1675 | * a data from the source image. | |
1676 | */ | |
fba98d45 | 1677 | ret = blk_pread(blk, idx * buf_size, buf, buf_size); |
a8e0fdd7 MK |
1678 | if (ret < 0) { |
1679 | goto out; | |
1680 | } | |
fba98d45 | 1681 | ret = blk_pwrite(blk, idx * buf_size, buf, buf_size); |
a8e0fdd7 MK |
1682 | if (ret < 0) { |
1683 | goto out; | |
1684 | } | |
1685 | } | |
318df29e | 1686 | |
fba98d45 | 1687 | ret = 0; |
a8e0fdd7 | 1688 | out: |
318df29e MA |
1689 | if (ret < 0) { |
1690 | error_setg_errno(errp, -ret, "Can't pre-allocate"); | |
1691 | } | |
1692 | out_with_err_set: | |
fba98d45 KW |
1693 | if (blk) { |
1694 | blk_unref(blk); | |
a8e0fdd7 | 1695 | } |
7267c094 | 1696 | g_free(buf); |
a8e0fdd7 MK |
1697 | |
1698 | return ret; | |
1699 | } | |
1700 | ||
b3af018f LY |
1701 | /* |
1702 | * Sheepdog support two kinds of redundancy, full replication and erasure | |
1703 | * coding. | |
1704 | * | |
1705 | * # create a fully replicated vdi with x copies | |
1706 | * -o redundancy=x (1 <= x <= SD_MAX_COPIES) | |
1707 | * | |
1708 | * # create a erasure coded vdi with x data strips and y parity strips | |
1709 | * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP) | |
1710 | */ | |
1711 | static int parse_redundancy(BDRVSheepdogState *s, const char *opt) | |
1712 | { | |
1713 | struct SheepdogInode *inode = &s->inode; | |
1714 | const char *n1, *n2; | |
1715 | long copy, parity; | |
1716 | char p[10]; | |
1717 | ||
1718 | pstrcpy(p, sizeof(p), opt); | |
1719 | n1 = strtok(p, ":"); | |
1720 | n2 = strtok(NULL, ":"); | |
1721 | ||
1722 | if (!n1) { | |
1723 | return -EINVAL; | |
1724 | } | |
1725 | ||
1726 | copy = strtol(n1, NULL, 10); | |
1727 | if (copy > SD_MAX_COPIES || copy < 1) { | |
1728 | return -EINVAL; | |
1729 | } | |
1730 | if (!n2) { | |
1731 | inode->copy_policy = 0; | |
1732 | inode->nr_copies = copy; | |
1733 | return 0; | |
1734 | } | |
1735 | ||
1736 | if (copy != 2 && copy != 4 && copy != 8 && copy != 16) { | |
1737 | return -EINVAL; | |
1738 | } | |
1739 | ||
1740 | parity = strtol(n2, NULL, 10); | |
1741 | if (parity >= SD_EC_MAX_STRIP || parity < 1) { | |
1742 | return -EINVAL; | |
1743 | } | |
1744 | ||
1745 | /* | |
1746 | * 4 bits for parity and 4 bits for data. | |
1747 | * We have to compress upper data bits because it can't represent 16 | |
1748 | */ | |
1749 | inode->copy_policy = ((copy / 2) << 4) + parity; | |
1750 | inode->nr_copies = copy + parity; | |
1751 | ||
1752 | return 0; | |
1753 | } | |
1754 | ||
876eb1b0 TI |
1755 | static int parse_block_size_shift(BDRVSheepdogState *s, QemuOpts *opt) |
1756 | { | |
1757 | struct SheepdogInode *inode = &s->inode; | |
1758 | uint64_t object_size; | |
1759 | int obj_order; | |
1760 | ||
1761 | object_size = qemu_opt_get_size_del(opt, BLOCK_OPT_OBJECT_SIZE, 0); | |
1762 | if (object_size) { | |
1763 | if ((object_size - 1) & object_size) { /* not a power of 2? */ | |
1764 | return -EINVAL; | |
1765 | } | |
786a4ea8 | 1766 | obj_order = ctz32(object_size); |
876eb1b0 TI |
1767 | if (obj_order < 20 || obj_order > 31) { |
1768 | return -EINVAL; | |
1769 | } | |
1770 | inode->block_size_shift = (uint8_t)obj_order; | |
1771 | } | |
1772 | ||
1773 | return 0; | |
1774 | } | |
1775 | ||
b222237b | 1776 | static int sd_create(const char *filename, QemuOpts *opts, |
d5124c00 | 1777 | Error **errp) |
33b1db1c | 1778 | { |
b6fc8245 | 1779 | int ret = 0; |
c31d482f | 1780 | uint32_t vid = 0; |
33b1db1c | 1781 | char *backing_file = NULL; |
b222237b | 1782 | char *buf = NULL; |
b6fc8245 | 1783 | BDRVSheepdogState *s; |
c31d482f | 1784 | char tag[SD_MAX_VDI_TAG_LEN]; |
b4447363 | 1785 | uint32_t snapid; |
876eb1b0 | 1786 | uint64_t max_vdi_size; |
2f536801 | 1787 | bool prealloc = false; |
33b1db1c | 1788 | |
5839e53b | 1789 | s = g_new0(BDRVSheepdogState, 1); |
b6fc8245 | 1790 | |
b4447363 | 1791 | memset(tag, 0, sizeof(tag)); |
5d6768e3 | 1792 | if (strstr(filename, "://")) { |
c31d482f | 1793 | ret = sd_parse_uri(s, filename, s->name, &snapid, tag); |
5d6768e3 | 1794 | } else { |
c31d482f | 1795 | ret = parse_vdiname(s, filename, s->name, &snapid, tag); |
5d6768e3 MK |
1796 | } |
1797 | if (ret < 0) { | |
efde4b62 | 1798 | error_setg(errp, "Can't parse filename"); |
b6fc8245 | 1799 | goto out; |
b4447363 MK |
1800 | } |
1801 | ||
c2eb918e HT |
1802 | s->inode.vdi_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
1803 | BDRV_SECTOR_SIZE); | |
b222237b CL |
1804 | backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); |
1805 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
1806 | if (!buf || !strcmp(buf, "off")) { | |
1807 | prealloc = false; | |
1808 | } else if (!strcmp(buf, "full")) { | |
1809 | prealloc = true; | |
1810 | } else { | |
1811 | error_setg(errp, "Invalid preallocation mode: '%s'", buf); | |
1812 | ret = -EINVAL; | |
1813 | goto out; | |
1814 | } | |
1815 | ||
1816 | g_free(buf); | |
1817 | buf = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY); | |
1818 | if (buf) { | |
1819 | ret = parse_redundancy(s, buf); | |
1820 | if (ret < 0) { | |
1821 | error_setg(errp, "Invalid redundancy mode: '%s'", buf); | |
1822 | goto out; | |
33b1db1c | 1823 | } |
33b1db1c | 1824 | } |
876eb1b0 TI |
1825 | ret = parse_block_size_shift(s, opts); |
1826 | if (ret < 0) { | |
1827 | error_setg(errp, "Invalid object_size." | |
1828 | " obect_size needs to be power of 2" | |
1829 | " and be limited from 2^20 to 2^31"); | |
b6fc8245 | 1830 | goto out; |
33b1db1c MK |
1831 | } |
1832 | ||
1833 | if (backing_file) { | |
fba98d45 | 1834 | BlockBackend *blk; |
9f23fce7 | 1835 | BDRVSheepdogState *base; |
33b1db1c MK |
1836 | BlockDriver *drv; |
1837 | ||
1838 | /* Currently, only Sheepdog backing image is supported. */ | |
b65a5e12 | 1839 | drv = bdrv_find_protocol(backing_file, true, NULL); |
33b1db1c | 1840 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
e67c3993 | 1841 | error_setg(errp, "backing_file must be a sheepdog image"); |
b6fc8245 MK |
1842 | ret = -EINVAL; |
1843 | goto out; | |
33b1db1c MK |
1844 | } |
1845 | ||
fba98d45 KW |
1846 | blk = blk_new_open("backing", backing_file, NULL, NULL, |
1847 | BDRV_O_PROTOCOL | BDRV_O_CACHE_WB, errp); | |
1848 | if (blk == NULL) { | |
1849 | ret = -EIO; | |
b6fc8245 | 1850 | goto out; |
cb595887 | 1851 | } |
33b1db1c | 1852 | |
fba98d45 | 1853 | base = blk_bs(blk)->opaque; |
33b1db1c | 1854 | |
9f23fce7 | 1855 | if (!is_snapshot(&base->inode)) { |
e67c3993 | 1856 | error_setg(errp, "cannot clone from a non snapshot vdi"); |
fba98d45 | 1857 | blk_unref(blk); |
b6fc8245 MK |
1858 | ret = -EINVAL; |
1859 | goto out; | |
33b1db1c | 1860 | } |
9f23fce7 | 1861 | s->inode.vdi_id = base->inode.vdi_id; |
fba98d45 | 1862 | blk_unref(blk); |
33b1db1c MK |
1863 | } |
1864 | ||
5d5da114 | 1865 | s->aio_context = qemu_get_aio_context(); |
876eb1b0 TI |
1866 | |
1867 | /* if block_size_shift is not specified, get cluster default value */ | |
1868 | if (s->inode.block_size_shift == 0) { | |
1869 | SheepdogVdiReq hdr; | |
1870 | SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr; | |
1871 | Error *local_err = NULL; | |
1872 | int fd; | |
1873 | unsigned int wlen = 0, rlen = 0; | |
1874 | ||
1875 | fd = connect_to_sdog(s, &local_err); | |
1876 | if (fd < 0) { | |
4fffeb5e | 1877 | error_report_err(local_err); |
876eb1b0 TI |
1878 | ret = -EIO; |
1879 | goto out; | |
1880 | } | |
1881 | ||
1882 | memset(&hdr, 0, sizeof(hdr)); | |
1883 | hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT; | |
1884 | hdr.proto_ver = SD_PROTO_VER; | |
1885 | ||
1886 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, | |
1887 | NULL, &wlen, &rlen); | |
1888 | closesocket(fd); | |
1889 | if (ret) { | |
1890 | error_setg_errno(errp, -ret, "failed to get cluster default"); | |
1891 | goto out; | |
1892 | } | |
1893 | if (rsp->result == SD_RES_SUCCESS) { | |
1894 | s->inode.block_size_shift = rsp->block_size_shift; | |
1895 | } else { | |
1896 | s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT; | |
1897 | } | |
1898 | } | |
1899 | ||
1900 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; | |
1901 | ||
1902 | if (s->inode.vdi_size > max_vdi_size) { | |
1903 | error_setg(errp, "An image is too large." | |
1904 | " The maximum image size is %"PRIu64 "GB", | |
1905 | max_vdi_size / 1024 / 1024 / 1024); | |
1906 | ret = -EINVAL; | |
1907 | goto out; | |
1908 | } | |
1909 | ||
e67c3993 | 1910 | ret = do_sd_create(s, &vid, 0, errp); |
7d2d3e74 | 1911 | if (ret) { |
b6fc8245 | 1912 | goto out; |
a8e0fdd7 MK |
1913 | } |
1914 | ||
7d2d3e74 | 1915 | if (prealloc) { |
e67c3993 | 1916 | ret = sd_prealloc(filename, errp); |
318df29e | 1917 | } |
b6fc8245 | 1918 | out: |
b222237b CL |
1919 | g_free(backing_file); |
1920 | g_free(buf); | |
b6fc8245 MK |
1921 | g_free(s); |
1922 | return ret; | |
33b1db1c MK |
1923 | } |
1924 | ||
1925 | static void sd_close(BlockDriverState *bs) | |
1926 | { | |
dfb12bf8 | 1927 | Error *local_err = NULL; |
33b1db1c MK |
1928 | BDRVSheepdogState *s = bs->opaque; |
1929 | SheepdogVdiReq hdr; | |
1930 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1931 | unsigned int wlen, rlen = 0; | |
1932 | int fd, ret; | |
1933 | ||
2440a2c3 | 1934 | DPRINTF("%s\n", s->name); |
33b1db1c | 1935 | |
dfb12bf8 | 1936 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1937 | if (fd < 0) { |
565f65d2 | 1938 | error_report_err(local_err); |
33b1db1c MK |
1939 | return; |
1940 | } | |
1941 | ||
1942 | memset(&hdr, 0, sizeof(hdr)); | |
1943 | ||
1944 | hdr.opcode = SD_OP_RELEASE_VDI; | |
1dbfafed | 1945 | hdr.type = LOCK_TYPE_NORMAL; |
9f23fce7 | 1946 | hdr.base_vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1947 | wlen = strlen(s->name) + 1; |
1948 | hdr.data_length = wlen; | |
1949 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1950 | ||
84390bed SH |
1951 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, |
1952 | s->name, &wlen, &rlen); | |
33b1db1c MK |
1953 | |
1954 | closesocket(fd); | |
1955 | ||
1956 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1957 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1958 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1959 | } |
1960 | ||
dca21ef2 FZ |
1961 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, |
1962 | false, NULL, NULL, NULL); | |
33b1db1c | 1963 | closesocket(s->fd); |
25af257d | 1964 | g_free(s->host_spec); |
33b1db1c MK |
1965 | } |
1966 | ||
1967 | static int64_t sd_getlength(BlockDriverState *bs) | |
1968 | { | |
1969 | BDRVSheepdogState *s = bs->opaque; | |
1970 | ||
1971 | return s->inode.vdi_size; | |
1972 | } | |
1973 | ||
1974 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1975 | { | |
dfb12bf8 | 1976 | Error *local_err = NULL; |
33b1db1c MK |
1977 | BDRVSheepdogState *s = bs->opaque; |
1978 | int ret, fd; | |
1979 | unsigned int datalen; | |
876eb1b0 | 1980 | uint64_t max_vdi_size; |
33b1db1c | 1981 | |
876eb1b0 | 1982 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; |
33b1db1c | 1983 | if (offset < s->inode.vdi_size) { |
6daf194d | 1984 | error_report("shrinking is not supported"); |
33b1db1c | 1985 | return -EINVAL; |
876eb1b0 | 1986 | } else if (offset > max_vdi_size) { |
6daf194d | 1987 | error_report("too big image size"); |
33b1db1c MK |
1988 | return -EINVAL; |
1989 | } | |
1990 | ||
dfb12bf8 | 1991 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 1992 | if (fd < 0) { |
565f65d2 | 1993 | error_report_err(local_err); |
cb595887 | 1994 | return fd; |
33b1db1c MK |
1995 | } |
1996 | ||
1997 | /* we don't need to update entire object */ | |
1998 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1999 | s->inode.vdi_size = offset; | |
84390bed SH |
2000 | ret = write_object(fd, s->aio_context, (char *)&s->inode, |
2001 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, | |
2002 | datalen, 0, false, s->cache_flags); | |
33b1db1c MK |
2003 | close(fd); |
2004 | ||
2005 | if (ret < 0) { | |
6daf194d | 2006 | error_report("failed to update an inode."); |
33b1db1c MK |
2007 | } |
2008 | ||
cb595887 | 2009 | return ret; |
33b1db1c MK |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * This function is called after writing data objects. If we need to | |
2014 | * update metadata, this sends a write request to the vdi object. | |
2df46246 | 2015 | * Otherwise, this switches back to sd_co_readv/writev. |
33b1db1c | 2016 | */ |
d8716b41 | 2017 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c | 2018 | { |
33b1db1c MK |
2019 | BDRVSheepdogState *s = acb->common.bs->opaque; |
2020 | struct iovec iov; | |
2021 | AIOReq *aio_req; | |
2022 | uint32_t offset, data_len, mn, mx; | |
2023 | ||
498f2140 HM |
2024 | mn = acb->min_dirty_data_idx; |
2025 | mx = acb->max_dirty_data_idx; | |
33b1db1c MK |
2026 | if (mn <= mx) { |
2027 | /* we need to update the vdi object. */ | |
2028 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
2029 | mn * sizeof(s->inode.data_vdi_id[0]); | |
2030 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
2031 | ||
498f2140 HM |
2032 | acb->min_dirty_data_idx = UINT32_MAX; |
2033 | acb->max_dirty_data_idx = 0; | |
33b1db1c MK |
2034 | |
2035 | iov.iov_base = &s->inode; | |
2036 | iov.iov_len = sizeof(s->inode); | |
2037 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
b544c1ab | 2038 | data_len, offset, 0, false, 0, offset); |
c292ee6a | 2039 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
b544c1ab | 2040 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); |
33b1db1c MK |
2041 | |
2042 | acb->aio_done_func = sd_finish_aiocb; | |
2043 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
2044 | return; | |
2045 | } | |
a37dcdf9 | 2046 | |
33b1db1c MK |
2047 | sd_finish_aiocb(acb); |
2048 | } | |
2049 | ||
859e5553 LY |
2050 | /* Delete current working VDI on the snapshot chain */ |
2051 | static bool sd_delete(BDRVSheepdogState *s) | |
2052 | { | |
dfb12bf8 | 2053 | Error *local_err = NULL; |
859e5553 LY |
2054 | unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0; |
2055 | SheepdogVdiReq hdr = { | |
2056 | .opcode = SD_OP_DEL_VDI, | |
9f23fce7 | 2057 | .base_vdi_id = s->inode.vdi_id, |
859e5553 LY |
2058 | .data_length = wlen, |
2059 | .flags = SD_FLAG_CMD_WRITE, | |
2060 | }; | |
2061 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2062 | int fd, ret; | |
2063 | ||
dfb12bf8 | 2064 | fd = connect_to_sdog(s, &local_err); |
859e5553 | 2065 | if (fd < 0) { |
565f65d2 | 2066 | error_report_err(local_err); |
859e5553 LY |
2067 | return false; |
2068 | } | |
2069 | ||
84390bed SH |
2070 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, |
2071 | s->name, &wlen, &rlen); | |
859e5553 LY |
2072 | closesocket(fd); |
2073 | if (ret) { | |
2074 | return false; | |
2075 | } | |
2076 | switch (rsp->result) { | |
2077 | case SD_RES_NO_VDI: | |
2078 | error_report("%s was already deleted", s->name); | |
2079 | /* fall through */ | |
2080 | case SD_RES_SUCCESS: | |
2081 | break; | |
2082 | default: | |
2083 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2084 | return false; | |
2085 | } | |
2086 | ||
2087 | return true; | |
2088 | } | |
2089 | ||
33b1db1c MK |
2090 | /* |
2091 | * Create a writable VDI from a snapshot | |
2092 | */ | |
2093 | static int sd_create_branch(BDRVSheepdogState *s) | |
2094 | { | |
dfb12bf8 | 2095 | Error *local_err = NULL; |
33b1db1c MK |
2096 | int ret, fd; |
2097 | uint32_t vid; | |
2098 | char *buf; | |
859e5553 | 2099 | bool deleted; |
33b1db1c | 2100 | |
2440a2c3 | 2101 | DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
33b1db1c | 2102 | |
7267c094 | 2103 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 2104 | |
859e5553 LY |
2105 | /* |
2106 | * Even If deletion fails, we will just create extra snapshot based on | |
dc6fb73d | 2107 | * the working VDI which was supposed to be deleted. So no need to |
859e5553 LY |
2108 | * false bail out. |
2109 | */ | |
2110 | deleted = sd_delete(s); | |
7d2d3e74 | 2111 | ret = do_sd_create(s, &vid, !deleted, &local_err); |
33b1db1c | 2112 | if (ret) { |
565f65d2 | 2113 | error_report_err(local_err); |
33b1db1c MK |
2114 | goto out; |
2115 | } | |
2116 | ||
2440a2c3 | 2117 | DPRINTF("%" PRIx32 " is created.\n", vid); |
33b1db1c | 2118 | |
dfb12bf8 | 2119 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2120 | if (fd < 0) { |
565f65d2 | 2121 | error_report_err(local_err); |
cb595887 | 2122 | ret = fd; |
33b1db1c MK |
2123 | goto out; |
2124 | } | |
2125 | ||
84390bed SH |
2126 | ret = read_object(fd, s->aio_context, buf, vid_to_vdi_oid(vid), |
2127 | s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags); | |
33b1db1c MK |
2128 | |
2129 | closesocket(fd); | |
2130 | ||
2131 | if (ret < 0) { | |
2132 | goto out; | |
2133 | } | |
2134 | ||
2135 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
2136 | ||
2f536801 | 2137 | s->is_snapshot = false; |
33b1db1c | 2138 | ret = 0; |
2440a2c3 | 2139 | DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
33b1db1c MK |
2140 | |
2141 | out: | |
7267c094 | 2142 | g_free(buf); |
33b1db1c MK |
2143 | |
2144 | return ret; | |
2145 | } | |
2146 | ||
2147 | /* | |
2148 | * Send I/O requests to the server. | |
2149 | * | |
2150 | * This function sends requests to the server, links the requests to | |
c292ee6a | 2151 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
2152 | * waiting the response. The responses are received in the |
2153 | * `aio_read_response' function which is called from the main loop as | |
2154 | * a fd handler. | |
2df46246 MK |
2155 | * |
2156 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
2157 | * request and -errno in error cases. | |
33b1db1c | 2158 | */ |
d8716b41 | 2159 | static int coroutine_fn sd_co_rw_vector(void *p) |
33b1db1c MK |
2160 | { |
2161 | SheepdogAIOCB *acb = p; | |
2162 | int ret = 0; | |
e8bfaa2f | 2163 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; |
876eb1b0 TI |
2164 | unsigned long idx; |
2165 | uint32_t object_size; | |
33b1db1c | 2166 | uint64_t oid; |
876eb1b0 | 2167 | uint64_t offset; |
33b1db1c MK |
2168 | BDRVSheepdogState *s = acb->common.bs->opaque; |
2169 | SheepdogInode *inode = &s->inode; | |
2170 | AIOReq *aio_req; | |
2171 | ||
33b1db1c MK |
2172 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
2173 | /* | |
2174 | * In the case we open the snapshot VDI, Sheepdog creates the | |
2175 | * writable VDI when we do a write operation first. | |
2176 | */ | |
2177 | ret = sd_create_branch(s); | |
2178 | if (ret) { | |
2179 | acb->ret = -EIO; | |
2180 | goto out; | |
2181 | } | |
2182 | } | |
2183 | ||
876eb1b0 TI |
2184 | object_size = (UINT32_C(1) << inode->block_size_shift); |
2185 | idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
2186 | offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size; | |
2187 | ||
1d732d7d MK |
2188 | /* |
2189 | * Make sure we don't free the aiocb before we are done with all requests. | |
2190 | * This additional reference is dropped at the end of this function. | |
2191 | */ | |
2192 | acb->nr_pending++; | |
2193 | ||
33b1db1c MK |
2194 | while (done != total) { |
2195 | uint8_t flags = 0; | |
2196 | uint64_t old_oid = 0; | |
2f536801 | 2197 | bool create = false; |
33b1db1c MK |
2198 | |
2199 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
2200 | ||
876eb1b0 | 2201 | len = MIN(total - done, object_size - offset); |
33b1db1c | 2202 | |
19db9b90 CH |
2203 | switch (acb->aiocb_type) { |
2204 | case AIOCB_READ_UDATA: | |
2205 | if (!inode->data_vdi_id[idx]) { | |
2206 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
2207 | goto done; |
2208 | } | |
19db9b90 CH |
2209 | break; |
2210 | case AIOCB_WRITE_UDATA: | |
2211 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 2212 | create = true; |
19db9b90 CH |
2213 | } else if (!is_data_obj_writable(inode, idx)) { |
2214 | /* Copy-On-Write */ | |
2f536801 | 2215 | create = true; |
19db9b90 CH |
2216 | old_oid = oid; |
2217 | flags = SD_FLAG_CMD_COW; | |
2218 | } | |
2219 | break; | |
cac8f4a6 LY |
2220 | case AIOCB_DISCARD_OBJ: |
2221 | /* | |
2222 | * We discard the object only when the whole object is | |
2223 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
2224 | */ | |
876eb1b0 | 2225 | if (len != object_size || inode->data_vdi_id[idx] == 0) { |
cac8f4a6 LY |
2226 | goto done; |
2227 | } | |
2228 | break; | |
19db9b90 CH |
2229 | default: |
2230 | break; | |
33b1db1c MK |
2231 | } |
2232 | ||
2233 | if (create) { | |
2440a2c3 | 2234 | DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1b6ac998 | 2235 | inode->vdi_id, oid, |
33b1db1c MK |
2236 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
2237 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
2440a2c3 | 2238 | DPRINTF("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
2239 | } |
2240 | ||
b544c1ab | 2241 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create, |
e6fd57ea HM |
2242 | old_oid, |
2243 | acb->aiocb_type == AIOCB_DISCARD_OBJ ? | |
2244 | 0 : done); | |
80308d33 | 2245 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
33b1db1c | 2246 | |
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 | } | |
2254 | out: | |
1d732d7d | 2255 | if (!--acb->nr_pending) { |
2df46246 | 2256 | return acb->ret; |
33b1db1c | 2257 | } |
2df46246 | 2258 | return 1; |
33b1db1c MK |
2259 | } |
2260 | ||
498f2140 | 2261 | static bool check_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *aiocb) |
6a55c82c HM |
2262 | { |
2263 | SheepdogAIOCB *cb; | |
2264 | ||
2265 | QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) { | |
498f2140 | 2266 | if (AIOCBOverlapping(aiocb, cb)) { |
6a55c82c HM |
2267 | return true; |
2268 | } | |
2269 | } | |
2270 | ||
2271 | QLIST_INSERT_HEAD(&s->inflight_aiocb_head, aiocb, aiocb_siblings); | |
2272 | return false; | |
2273 | } | |
2274 | ||
a968168c | 2275 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2276 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
2277 | { |
2278 | SheepdogAIOCB *acb; | |
2df46246 | 2279 | int ret; |
e50d7607 LY |
2280 | int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE; |
2281 | BDRVSheepdogState *s = bs->opaque; | |
33b1db1c | 2282 | |
c0191e76 | 2283 | if (offset > s->inode.vdi_size) { |
e50d7607 | 2284 | ret = sd_truncate(bs, offset); |
cb595887 MK |
2285 | if (ret < 0) { |
2286 | return ret; | |
33b1db1c | 2287 | } |
33b1db1c MK |
2288 | } |
2289 | ||
f700f8e3 | 2290 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
2291 | acb->aio_done_func = sd_write_done; |
2292 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
2293 | ||
6a55c82c | 2294 | retry: |
498f2140 HM |
2295 | if (check_overlapping_aiocb(s, acb)) { |
2296 | qemu_co_queue_wait(&s->overlapping_queue); | |
6a55c82c HM |
2297 | goto retry; |
2298 | } | |
2299 | ||
2df46246 MK |
2300 | ret = sd_co_rw_vector(acb); |
2301 | if (ret <= 0) { | |
6a55c82c | 2302 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2303 | qemu_co_queue_restart_all(&s->overlapping_queue); |
8007429a | 2304 | qemu_aio_unref(acb); |
2df46246 MK |
2305 | return ret; |
2306 | } | |
2307 | ||
2308 | qemu_coroutine_yield(); | |
2309 | ||
6a55c82c | 2310 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2311 | qemu_co_queue_restart_all(&s->overlapping_queue); |
6a55c82c | 2312 | |
2df46246 | 2313 | return acb->ret; |
33b1db1c MK |
2314 | } |
2315 | ||
a968168c | 2316 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 2317 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
2318 | { |
2319 | SheepdogAIOCB *acb; | |
19db9b90 | 2320 | int ret; |
6a55c82c | 2321 | BDRVSheepdogState *s = bs->opaque; |
33b1db1c | 2322 | |
f700f8e3 | 2323 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
2324 | acb->aiocb_type = AIOCB_READ_UDATA; |
2325 | acb->aio_done_func = sd_finish_aiocb; | |
2326 | ||
6a55c82c | 2327 | retry: |
498f2140 HM |
2328 | if (check_overlapping_aiocb(s, acb)) { |
2329 | qemu_co_queue_wait(&s->overlapping_queue); | |
6a55c82c HM |
2330 | goto retry; |
2331 | } | |
2332 | ||
2df46246 MK |
2333 | ret = sd_co_rw_vector(acb); |
2334 | if (ret <= 0) { | |
6a55c82c | 2335 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2336 | qemu_co_queue_restart_all(&s->overlapping_queue); |
8007429a | 2337 | qemu_aio_unref(acb); |
2df46246 MK |
2338 | return ret; |
2339 | } | |
2340 | ||
2341 | qemu_coroutine_yield(); | |
2342 | ||
6a55c82c | 2343 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2344 | qemu_co_queue_restart_all(&s->overlapping_queue); |
2df46246 | 2345 | return acb->ret; |
33b1db1c MK |
2346 | } |
2347 | ||
47622c44 LY |
2348 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
2349 | { | |
2350 | BDRVSheepdogState *s = bs->opaque; | |
47783072 LY |
2351 | SheepdogAIOCB *acb; |
2352 | AIOReq *aio_req; | |
47622c44 | 2353 | |
0e7106d8 | 2354 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
2355 | return 0; |
2356 | } | |
2357 | ||
f700f8e3 | 2358 | acb = sd_aio_setup(bs, NULL, 0, 0); |
47783072 LY |
2359 | acb->aiocb_type = AIOCB_FLUSH_CACHE; |
2360 | acb->aio_done_func = sd_finish_aiocb; | |
47622c44 | 2361 | |
47783072 | 2362 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
b544c1ab | 2363 | 0, 0, 0, false, 0, 0); |
47783072 | 2364 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
b544c1ab | 2365 | add_aio_request(s, aio_req, NULL, 0, acb->aiocb_type); |
47622c44 | 2366 | |
47783072 LY |
2367 | qemu_coroutine_yield(); |
2368 | return acb->ret; | |
47622c44 LY |
2369 | } |
2370 | ||
33b1db1c MK |
2371 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
2372 | { | |
dfb12bf8 | 2373 | Error *local_err = NULL; |
33b1db1c MK |
2374 | BDRVSheepdogState *s = bs->opaque; |
2375 | int ret, fd; | |
2376 | uint32_t new_vid; | |
2377 | SheepdogInode *inode; | |
2378 | unsigned int datalen; | |
2379 | ||
2440a2c3 | 2380 | DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
2381 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
2382 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
2383 | ||
2384 | if (s->is_snapshot) { | |
2385 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 2386 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
2387 | |
2388 | return -EINVAL; | |
2389 | } | |
2390 | ||
2440a2c3 | 2391 | DPRINTF("%s %s\n", sn_info->name, sn_info->id_str); |
33b1db1c MK |
2392 | |
2393 | s->inode.vm_state_size = sn_info->vm_state_size; | |
2394 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
2395 | /* It appears that inode.tag does not require a NUL terminator, |
2396 | * which means this use of strncpy is ok. | |
2397 | */ | |
33b1db1c MK |
2398 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
2399 | /* we don't need to update entire object */ | |
2400 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
2df5fee2 | 2401 | inode = g_malloc(datalen); |
33b1db1c MK |
2402 | |
2403 | /* refresh inode. */ | |
dfb12bf8 | 2404 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2405 | if (fd < 0) { |
565f65d2 | 2406 | error_report_err(local_err); |
cb595887 | 2407 | ret = fd; |
33b1db1c MK |
2408 | goto cleanup; |
2409 | } | |
2410 | ||
84390bed SH |
2411 | ret = write_object(fd, s->aio_context, (char *)&s->inode, |
2412 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, | |
2413 | datalen, 0, false, s->cache_flags); | |
33b1db1c | 2414 | if (ret < 0) { |
6daf194d | 2415 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
2416 | goto cleanup; |
2417 | } | |
2418 | ||
7d2d3e74 | 2419 | ret = do_sd_create(s, &new_vid, 1, &local_err); |
33b1db1c | 2420 | if (ret < 0) { |
c29b77f9 MA |
2421 | error_reportf_err(local_err, |
2422 | "failed to create inode for snapshot: "); | |
33b1db1c MK |
2423 | goto cleanup; |
2424 | } | |
2425 | ||
84390bed SH |
2426 | ret = read_object(fd, s->aio_context, (char *)inode, |
2427 | vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0, | |
2428 | s->cache_flags); | |
33b1db1c MK |
2429 | |
2430 | if (ret < 0) { | |
6daf194d | 2431 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
2432 | goto cleanup; |
2433 | } | |
2434 | ||
2435 | memcpy(&s->inode, inode, datalen); | |
2440a2c3 | 2436 | DPRINTF("s->inode: name %s snap_id %x oid %x\n", |
33b1db1c MK |
2437 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
2438 | ||
2439 | cleanup: | |
2df5fee2 | 2440 | g_free(inode); |
33b1db1c MK |
2441 | closesocket(fd); |
2442 | return ret; | |
2443 | } | |
2444 | ||
859e5553 LY |
2445 | /* |
2446 | * We implement rollback(loadvm) operation to the specified snapshot by | |
2447 | * 1) switch to the snapshot | |
2448 | * 2) rely on sd_create_branch to delete working VDI and | |
dc6fb73d | 2449 | * 3) create a new working VDI based on the specified snapshot |
859e5553 | 2450 | */ |
33b1db1c MK |
2451 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
2452 | { | |
2453 | BDRVSheepdogState *s = bs->opaque; | |
2454 | BDRVSheepdogState *old_s; | |
9ff53a0e | 2455 | char tag[SD_MAX_VDI_TAG_LEN]; |
33b1db1c | 2456 | uint32_t snapid = 0; |
9ff53a0e | 2457 | int ret = 0; |
33b1db1c | 2458 | |
5839e53b | 2459 | old_s = g_new(BDRVSheepdogState, 1); |
33b1db1c MK |
2460 | |
2461 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
2462 | ||
33b1db1c | 2463 | snapid = strtoul(snapshot_id, NULL, 10); |
3178e275 JM |
2464 | if (snapid) { |
2465 | tag[0] = 0; | |
2466 | } else { | |
b579ffb3 | 2467 | pstrcpy(tag, sizeof(tag), snapshot_id); |
33b1db1c MK |
2468 | } |
2469 | ||
9ff53a0e | 2470 | ret = reload_inode(s, snapid, tag); |
33b1db1c | 2471 | if (ret) { |
33b1db1c MK |
2472 | goto out; |
2473 | } | |
2474 | ||
cede621f LY |
2475 | ret = sd_create_branch(s); |
2476 | if (ret) { | |
33b1db1c MK |
2477 | goto out; |
2478 | } | |
2479 | ||
7267c094 | 2480 | g_free(old_s); |
33b1db1c MK |
2481 | |
2482 | return 0; | |
2483 | out: | |
2484 | /* recover bdrv_sd_state */ | |
2485 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 | 2486 | g_free(old_s); |
33b1db1c | 2487 | |
6daf194d | 2488 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
2489 | |
2490 | return ret; | |
2491 | } | |
2492 | ||
eab8eb8d VT |
2493 | #define NR_BATCHED_DISCARD 128 |
2494 | ||
2495 | static bool remove_objects(BDRVSheepdogState *s) | |
2496 | { | |
2497 | int fd, i = 0, nr_objs = 0; | |
2498 | Error *local_err = NULL; | |
2499 | int ret = 0; | |
2500 | bool result = true; | |
2501 | SheepdogInode *inode = &s->inode; | |
2502 | ||
2503 | fd = connect_to_sdog(s, &local_err); | |
2504 | if (fd < 0) { | |
2505 | error_report_err(local_err); | |
2506 | return false; | |
2507 | } | |
2508 | ||
2509 | nr_objs = count_data_objs(inode); | |
2510 | while (i < nr_objs) { | |
2511 | int start_idx, nr_filled_idx; | |
2512 | ||
2513 | while (i < nr_objs && !inode->data_vdi_id[i]) { | |
2514 | i++; | |
2515 | } | |
2516 | start_idx = i; | |
2517 | ||
2518 | nr_filled_idx = 0; | |
2519 | while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) { | |
2520 | if (inode->data_vdi_id[i]) { | |
2521 | inode->data_vdi_id[i] = 0; | |
2522 | nr_filled_idx++; | |
2523 | } | |
2524 | ||
2525 | i++; | |
2526 | } | |
2527 | ||
2528 | ret = write_object(fd, s->aio_context, | |
2529 | (char *)&inode->data_vdi_id[start_idx], | |
2530 | vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies, | |
2531 | (i - start_idx) * sizeof(uint32_t), | |
2532 | offsetof(struct SheepdogInode, | |
2533 | data_vdi_id[start_idx]), | |
2534 | false, s->cache_flags); | |
2535 | if (ret < 0) { | |
2536 | error_report("failed to discard snapshot inode."); | |
2537 | result = false; | |
2538 | goto out; | |
2539 | } | |
2540 | } | |
2541 | ||
2542 | out: | |
2543 | closesocket(fd); | |
2544 | return result; | |
2545 | } | |
2546 | ||
a89d89d3 WX |
2547 | static int sd_snapshot_delete(BlockDriverState *bs, |
2548 | const char *snapshot_id, | |
2549 | const char *name, | |
2550 | Error **errp) | |
33b1db1c | 2551 | { |
eab8eb8d VT |
2552 | uint32_t snap_id = 0; |
2553 | char snap_tag[SD_MAX_VDI_TAG_LEN]; | |
2554 | Error *local_err = NULL; | |
2555 | int fd, ret; | |
2556 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
2557 | BDRVSheepdogState *s = bs->opaque; | |
2558 | unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0; | |
2559 | uint32_t vid; | |
2560 | SheepdogVdiReq hdr = { | |
2561 | .opcode = SD_OP_DEL_VDI, | |
2562 | .data_length = wlen, | |
2563 | .flags = SD_FLAG_CMD_WRITE, | |
2564 | }; | |
2565 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2566 | ||
2567 | if (!remove_objects(s)) { | |
2568 | return -1; | |
2569 | } | |
2570 | ||
2571 | memset(buf, 0, sizeof(buf)); | |
2572 | memset(snap_tag, 0, sizeof(snap_tag)); | |
2573 | pstrcpy(buf, SD_MAX_VDI_LEN, s->name); | |
2574 | if (qemu_strtoul(snapshot_id, NULL, 10, (unsigned long *)&snap_id)) { | |
2575 | return -1; | |
2576 | } | |
2577 | ||
2578 | if (snap_id) { | |
2579 | hdr.snapid = snap_id; | |
2580 | } else { | |
2581 | pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id); | |
2582 | pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag); | |
2583 | } | |
2584 | ||
2585 | ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, | |
2586 | &local_err); | |
2587 | if (ret) { | |
2588 | return ret; | |
2589 | } | |
2590 | ||
2591 | fd = connect_to_sdog(s, &local_err); | |
2592 | if (fd < 0) { | |
2593 | error_report_err(local_err); | |
2594 | return -1; | |
2595 | } | |
2596 | ||
2597 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&hdr, | |
2598 | buf, &wlen, &rlen); | |
2599 | closesocket(fd); | |
2600 | if (ret) { | |
2601 | return ret; | |
2602 | } | |
2603 | ||
2604 | switch (rsp->result) { | |
2605 | case SD_RES_NO_VDI: | |
2606 | error_report("%s was already deleted", s->name); | |
2607 | case SD_RES_SUCCESS: | |
2608 | break; | |
2609 | default: | |
2610 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2611 | return -1; | |
2612 | } | |
2613 | ||
2614 | return ret; | |
33b1db1c MK |
2615 | } |
2616 | ||
33b1db1c MK |
2617 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
2618 | { | |
dfb12bf8 | 2619 | Error *local_err = NULL; |
33b1db1c MK |
2620 | BDRVSheepdogState *s = bs->opaque; |
2621 | SheepdogReq req; | |
2622 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
2623 | QEMUSnapshotInfo *sn_tab = NULL; | |
2624 | unsigned wlen, rlen; | |
2625 | int found = 0; | |
2626 | static SheepdogInode inode; | |
2627 | unsigned long *vdi_inuse; | |
2628 | unsigned int start_nr; | |
2629 | uint64_t hval; | |
2630 | uint32_t vid; | |
2631 | ||
7267c094 | 2632 | vdi_inuse = g_malloc(max); |
33b1db1c | 2633 | |
dfb12bf8 | 2634 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2635 | if (fd < 0) { |
565f65d2 | 2636 | error_report_err(local_err); |
cb595887 | 2637 | ret = fd; |
33b1db1c MK |
2638 | goto out; |
2639 | } | |
2640 | ||
2641 | rlen = max; | |
2642 | wlen = 0; | |
2643 | ||
2644 | memset(&req, 0, sizeof(req)); | |
2645 | ||
2646 | req.opcode = SD_OP_READ_VDIS; | |
2647 | req.data_length = max; | |
2648 | ||
84390bed SH |
2649 | ret = do_req(fd, s->aio_context, (SheepdogReq *)&req, |
2650 | vdi_inuse, &wlen, &rlen); | |
33b1db1c MK |
2651 | |
2652 | closesocket(fd); | |
2653 | if (ret) { | |
2654 | goto out; | |
2655 | } | |
2656 | ||
02c4f26b | 2657 | sn_tab = g_new0(QEMUSnapshotInfo, nr); |
33b1db1c MK |
2658 | |
2659 | /* calculate a vdi id with hash function */ | |
2660 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2661 | start_nr = hval & (SD_NR_VDIS - 1); | |
2662 | ||
dfb12bf8 | 2663 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2664 | if (fd < 0) { |
565f65d2 | 2665 | error_report_err(local_err); |
cb595887 | 2666 | ret = fd; |
33b1db1c MK |
2667 | goto out; |
2668 | } | |
2669 | ||
2670 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2671 | if (!test_bit(vid, vdi_inuse)) { | |
2672 | break; | |
2673 | } | |
2674 | ||
2675 | /* we don't need to read entire object */ | |
84390bed SH |
2676 | ret = read_object(fd, s->aio_context, (char *)&inode, |
2677 | vid_to_vdi_oid(vid), | |
47622c44 | 2678 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2679 | s->cache_flags); |
33b1db1c MK |
2680 | |
2681 | if (ret) { | |
2682 | continue; | |
2683 | } | |
2684 | ||
2685 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2686 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2687 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2688 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2689 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2690 | ||
521b2b5d HR |
2691 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), |
2692 | "%" PRIu32, inode.snap_id); | |
3178e275 JM |
2693 | pstrcpy(sn_tab[found].name, |
2694 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2695 | inode.tag); | |
33b1db1c MK |
2696 | found++; |
2697 | } | |
2698 | } | |
2699 | ||
2700 | closesocket(fd); | |
2701 | out: | |
2702 | *psn_tab = sn_tab; | |
2703 | ||
7267c094 | 2704 | g_free(vdi_inuse); |
33b1db1c | 2705 | |
cb595887 MK |
2706 | if (ret < 0) { |
2707 | return ret; | |
2708 | } | |
2709 | ||
33b1db1c MK |
2710 | return found; |
2711 | } | |
2712 | ||
2713 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2714 | int64_t pos, int size, int load) | |
2715 | { | |
dfb12bf8 | 2716 | Error *local_err = NULL; |
2f536801 MK |
2717 | bool create; |
2718 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2719 | unsigned int data_len; |
2720 | uint64_t vmstate_oid; | |
33b1db1c | 2721 | uint64_t offset; |
cede621f LY |
2722 | uint32_t vdi_index; |
2723 | uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id; | |
876eb1b0 | 2724 | uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift); |
33b1db1c | 2725 | |
dfb12bf8 | 2726 | fd = connect_to_sdog(s, &local_err); |
33b1db1c | 2727 | if (fd < 0) { |
565f65d2 | 2728 | error_report_err(local_err); |
cb595887 | 2729 | return fd; |
33b1db1c MK |
2730 | } |
2731 | ||
6f3c714e | 2732 | while (remaining) { |
876eb1b0 TI |
2733 | vdi_index = pos / object_size; |
2734 | offset = pos % object_size; | |
33b1db1c | 2735 | |
876eb1b0 | 2736 | data_len = MIN(remaining, object_size - offset); |
33b1db1c | 2737 | |
cede621f | 2738 | vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index); |
33b1db1c MK |
2739 | |
2740 | create = (offset == 0); | |
2741 | if (load) { | |
84390bed | 2742 | ret = read_object(fd, s->aio_context, (char *)data, vmstate_oid, |
47622c44 | 2743 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2744 | s->cache_flags); |
33b1db1c | 2745 | } else { |
84390bed | 2746 | ret = write_object(fd, s->aio_context, (char *)data, vmstate_oid, |
47622c44 | 2747 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2748 | s->cache_flags); |
33b1db1c MK |
2749 | } |
2750 | ||
2751 | if (ret < 0) { | |
6daf194d | 2752 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2753 | goto cleanup; |
2754 | } | |
2755 | ||
2756 | pos += data_len; | |
1f7a48de | 2757 | data += data_len; |
6f3c714e | 2758 | remaining -= data_len; |
33b1db1c | 2759 | } |
6f3c714e | 2760 | ret = size; |
33b1db1c MK |
2761 | cleanup: |
2762 | closesocket(fd); | |
2763 | return ret; | |
2764 | } | |
2765 | ||
cf8074b3 KW |
2766 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2767 | int64_t pos) | |
33b1db1c MK |
2768 | { |
2769 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2770 | void *buf; |
2771 | int ret; | |
33b1db1c | 2772 | |
cf8074b3 KW |
2773 | buf = qemu_blockalign(bs, qiov->size); |
2774 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2775 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2776 | qemu_vfree(buf); | |
2777 | ||
2778 | return ret; | |
33b1db1c MK |
2779 | } |
2780 | ||
2781 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
2782 | int64_t pos, int size) | |
2783 | { | |
2784 | BDRVSheepdogState *s = bs->opaque; | |
2785 | ||
2786 | return do_load_save_vmstate(s, data, pos, size, 1); | |
2787 | } | |
2788 | ||
2789 | ||
cac8f4a6 LY |
2790 | static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num, |
2791 | int nb_sectors) | |
2792 | { | |
2793 | SheepdogAIOCB *acb; | |
cac8f4a6 LY |
2794 | BDRVSheepdogState *s = bs->opaque; |
2795 | int ret; | |
e6fd57ea HM |
2796 | QEMUIOVector discard_iov; |
2797 | struct iovec iov; | |
2798 | uint32_t zero = 0; | |
cac8f4a6 LY |
2799 | |
2800 | if (!s->discard_supported) { | |
2801 | return 0; | |
2802 | } | |
2803 | ||
e6fd57ea HM |
2804 | memset(&discard_iov, 0, sizeof(discard_iov)); |
2805 | memset(&iov, 0, sizeof(iov)); | |
2806 | iov.iov_base = &zero; | |
2807 | iov.iov_len = sizeof(zero); | |
2808 | discard_iov.iov = &iov; | |
2809 | discard_iov.niov = 1; | |
2810 | acb = sd_aio_setup(bs, &discard_iov, sector_num, nb_sectors); | |
cac8f4a6 LY |
2811 | acb->aiocb_type = AIOCB_DISCARD_OBJ; |
2812 | acb->aio_done_func = sd_finish_aiocb; | |
2813 | ||
6a55c82c | 2814 | retry: |
498f2140 HM |
2815 | if (check_overlapping_aiocb(s, acb)) { |
2816 | qemu_co_queue_wait(&s->overlapping_queue); | |
6a55c82c HM |
2817 | goto retry; |
2818 | } | |
2819 | ||
cac8f4a6 LY |
2820 | ret = sd_co_rw_vector(acb); |
2821 | if (ret <= 0) { | |
6a55c82c | 2822 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2823 | qemu_co_queue_restart_all(&s->overlapping_queue); |
8007429a | 2824 | qemu_aio_unref(acb); |
cac8f4a6 LY |
2825 | return ret; |
2826 | } | |
2827 | ||
2828 | qemu_coroutine_yield(); | |
2829 | ||
6a55c82c | 2830 | QLIST_REMOVE(acb, aiocb_siblings); |
498f2140 | 2831 | qemu_co_queue_restart_all(&s->overlapping_queue); |
6a55c82c | 2832 | |
cac8f4a6 LY |
2833 | return acb->ret; |
2834 | } | |
2835 | ||
b6b8a333 PB |
2836 | static coroutine_fn int64_t |
2837 | sd_co_get_block_status(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
67a0fd2a | 2838 | int *pnum, BlockDriverState **file) |
8d71c631 LY |
2839 | { |
2840 | BDRVSheepdogState *s = bs->opaque; | |
2841 | SheepdogInode *inode = &s->inode; | |
876eb1b0 | 2842 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
9cd76737 | 2843 | uint64_t offset = sector_num * BDRV_SECTOR_SIZE; |
876eb1b0 | 2844 | unsigned long start = offset / object_size, |
8d71c631 | 2845 | end = DIV_ROUND_UP((sector_num + nb_sectors) * |
876eb1b0 | 2846 | BDRV_SECTOR_SIZE, object_size); |
8d71c631 | 2847 | unsigned long idx; |
9cd76737 | 2848 | int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset; |
8d71c631 LY |
2849 | |
2850 | for (idx = start; idx < end; idx++) { | |
2851 | if (inode->data_vdi_id[idx] == 0) { | |
2852 | break; | |
2853 | } | |
2854 | } | |
2855 | if (idx == start) { | |
2856 | /* Get the longest length of unallocated sectors */ | |
2857 | ret = 0; | |
2858 | for (idx = start + 1; idx < end; idx++) { | |
2859 | if (inode->data_vdi_id[idx] != 0) { | |
2860 | break; | |
2861 | } | |
2862 | } | |
2863 | } | |
2864 | ||
876eb1b0 | 2865 | *pnum = (idx - start) * object_size / BDRV_SECTOR_SIZE; |
8d71c631 LY |
2866 | if (*pnum > nb_sectors) { |
2867 | *pnum = nb_sectors; | |
2868 | } | |
d234c929 FZ |
2869 | if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) { |
2870 | *file = bs; | |
2871 | } | |
8d71c631 LY |
2872 | return ret; |
2873 | } | |
2874 | ||
85829722 LY |
2875 | static int64_t sd_get_allocated_file_size(BlockDriverState *bs) |
2876 | { | |
2877 | BDRVSheepdogState *s = bs->opaque; | |
2878 | SheepdogInode *inode = &s->inode; | |
876eb1b0 TI |
2879 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); |
2880 | unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size); | |
85829722 LY |
2881 | uint64_t size = 0; |
2882 | ||
2883 | for (i = 0; i < last; i++) { | |
2884 | if (inode->data_vdi_id[i] == 0) { | |
2885 | continue; | |
2886 | } | |
876eb1b0 | 2887 | size += object_size; |
85829722 LY |
2888 | } |
2889 | return size; | |
2890 | } | |
2891 | ||
b222237b CL |
2892 | static QemuOptsList sd_create_opts = { |
2893 | .name = "sheepdog-create-opts", | |
2894 | .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head), | |
2895 | .desc = { | |
2896 | { | |
2897 | .name = BLOCK_OPT_SIZE, | |
2898 | .type = QEMU_OPT_SIZE, | |
2899 | .help = "Virtual disk size" | |
2900 | }, | |
2901 | { | |
2902 | .name = BLOCK_OPT_BACKING_FILE, | |
2903 | .type = QEMU_OPT_STRING, | |
2904 | .help = "File name of a base image" | |
2905 | }, | |
2906 | { | |
2907 | .name = BLOCK_OPT_PREALLOC, | |
2908 | .type = QEMU_OPT_STRING, | |
2909 | .help = "Preallocation mode (allowed values: off, full)" | |
2910 | }, | |
2911 | { | |
2912 | .name = BLOCK_OPT_REDUNDANCY, | |
2913 | .type = QEMU_OPT_STRING, | |
2914 | .help = "Redundancy of the image" | |
2915 | }, | |
876eb1b0 TI |
2916 | { |
2917 | .name = BLOCK_OPT_OBJECT_SIZE, | |
2918 | .type = QEMU_OPT_SIZE, | |
2919 | .help = "Object size of the image" | |
2920 | }, | |
b222237b CL |
2921 | { /* end of list */ } |
2922 | } | |
33b1db1c MK |
2923 | }; |
2924 | ||
5d6768e3 | 2925 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2926 | .format_name = "sheepdog", |
2927 | .protocol_name = "sheepdog", | |
2928 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2929 | .bdrv_needs_filename = true, |
33b1db1c | 2930 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2931 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2932 | .bdrv_reopen_commit = sd_reopen_commit, | |
2933 | .bdrv_reopen_abort = sd_reopen_abort, | |
33b1db1c | 2934 | .bdrv_close = sd_close, |
c282e1fd | 2935 | .bdrv_create = sd_create, |
e4f5c1bf | 2936 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
33b1db1c | 2937 | .bdrv_getlength = sd_getlength, |
85829722 | 2938 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
33b1db1c MK |
2939 | .bdrv_truncate = sd_truncate, |
2940 | ||
2df46246 MK |
2941 | .bdrv_co_readv = sd_co_readv, |
2942 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2943 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
cac8f4a6 | 2944 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2945 | .bdrv_co_get_block_status = sd_co_get_block_status, |
33b1db1c MK |
2946 | |
2947 | .bdrv_snapshot_create = sd_snapshot_create, | |
2948 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2949 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2950 | .bdrv_snapshot_list = sd_snapshot_list, | |
2951 | ||
2952 | .bdrv_save_vmstate = sd_save_vmstate, | |
2953 | .bdrv_load_vmstate = sd_load_vmstate, | |
2954 | ||
84390bed SH |
2955 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2956 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2957 | ||
b222237b | 2958 | .create_opts = &sd_create_opts, |
33b1db1c MK |
2959 | }; |
2960 | ||
5d6768e3 MK |
2961 | static BlockDriver bdrv_sheepdog_tcp = { |
2962 | .format_name = "sheepdog", | |
2963 | .protocol_name = "sheepdog+tcp", | |
2964 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 2965 | .bdrv_needs_filename = true, |
5d6768e3 | 2966 | .bdrv_file_open = sd_open, |
4da65c80 LY |
2967 | .bdrv_reopen_prepare = sd_reopen_prepare, |
2968 | .bdrv_reopen_commit = sd_reopen_commit, | |
2969 | .bdrv_reopen_abort = sd_reopen_abort, | |
5d6768e3 | 2970 | .bdrv_close = sd_close, |
c282e1fd | 2971 | .bdrv_create = sd_create, |
e4f5c1bf | 2972 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
5d6768e3 | 2973 | .bdrv_getlength = sd_getlength, |
85829722 | 2974 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
5d6768e3 MK |
2975 | .bdrv_truncate = sd_truncate, |
2976 | ||
2977 | .bdrv_co_readv = sd_co_readv, | |
2978 | .bdrv_co_writev = sd_co_writev, | |
2979 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2980 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 2981 | .bdrv_co_get_block_status = sd_co_get_block_status, |
5d6768e3 MK |
2982 | |
2983 | .bdrv_snapshot_create = sd_snapshot_create, | |
2984 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2985 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2986 | .bdrv_snapshot_list = sd_snapshot_list, | |
2987 | ||
2988 | .bdrv_save_vmstate = sd_save_vmstate, | |
2989 | .bdrv_load_vmstate = sd_load_vmstate, | |
2990 | ||
84390bed SH |
2991 | .bdrv_detach_aio_context = sd_detach_aio_context, |
2992 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
2993 | ||
b222237b | 2994 | .create_opts = &sd_create_opts, |
5d6768e3 MK |
2995 | }; |
2996 | ||
1b8bbb46 MK |
2997 | static BlockDriver bdrv_sheepdog_unix = { |
2998 | .format_name = "sheepdog", | |
2999 | .protocol_name = "sheepdog+unix", | |
3000 | .instance_size = sizeof(BDRVSheepdogState), | |
030be321 | 3001 | .bdrv_needs_filename = true, |
1b8bbb46 | 3002 | .bdrv_file_open = sd_open, |
4da65c80 LY |
3003 | .bdrv_reopen_prepare = sd_reopen_prepare, |
3004 | .bdrv_reopen_commit = sd_reopen_commit, | |
3005 | .bdrv_reopen_abort = sd_reopen_abort, | |
1b8bbb46 | 3006 | .bdrv_close = sd_close, |
c282e1fd | 3007 | .bdrv_create = sd_create, |
3ac21627 | 3008 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
1b8bbb46 | 3009 | .bdrv_getlength = sd_getlength, |
85829722 | 3010 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, |
1b8bbb46 MK |
3011 | .bdrv_truncate = sd_truncate, |
3012 | ||
3013 | .bdrv_co_readv = sd_co_readv, | |
3014 | .bdrv_co_writev = sd_co_writev, | |
3015 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 3016 | .bdrv_co_discard = sd_co_discard, |
b6b8a333 | 3017 | .bdrv_co_get_block_status = sd_co_get_block_status, |
1b8bbb46 MK |
3018 | |
3019 | .bdrv_snapshot_create = sd_snapshot_create, | |
3020 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
3021 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
3022 | .bdrv_snapshot_list = sd_snapshot_list, | |
3023 | ||
3024 | .bdrv_save_vmstate = sd_save_vmstate, | |
3025 | .bdrv_load_vmstate = sd_load_vmstate, | |
3026 | ||
84390bed SH |
3027 | .bdrv_detach_aio_context = sd_detach_aio_context, |
3028 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
3029 | ||
b222237b | 3030 | .create_opts = &sd_create_opts, |
1b8bbb46 MK |
3031 | }; |
3032 | ||
33b1db1c MK |
3033 | static void bdrv_sheepdog_init(void) |
3034 | { | |
3035 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 3036 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 3037 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
3038 | } |
3039 | block_init(bdrv_sheepdog_init); |