]>
Commit | Line | Data |
---|---|---|
33b1db1c MK |
1 | /* |
2 | * Copyright (C) 2009-2010 Nippon Telegraph and Telephone Corporation. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License version | |
6 | * 2 as published by the Free Software Foundation. | |
7 | * | |
8 | * You should have received a copy of the GNU General Public License | |
9 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | |
6b620ca3 PB |
10 | * |
11 | * Contributions after 2012-01-13 are licensed under the terms of the | |
12 | * GNU GPL, version 2 or (at your option) any later version. | |
33b1db1c | 13 | */ |
33b1db1c MK |
14 | |
15 | #include "qemu-common.h" | |
5d6768e3 | 16 | #include "qemu/uri.h" |
1de7afc9 PB |
17 | #include "qemu/error-report.h" |
18 | #include "qemu/sockets.h" | |
737e150e | 19 | #include "block/block_int.h" |
1de7afc9 | 20 | #include "qemu/bitops.h" |
33b1db1c MK |
21 | |
22 | #define SD_PROTO_VER 0x01 | |
23 | ||
24 | #define SD_DEFAULT_ADDR "localhost" | |
25af257d | 25 | #define SD_DEFAULT_PORT 7000 |
33b1db1c MK |
26 | |
27 | #define SD_OP_CREATE_AND_WRITE_OBJ 0x01 | |
28 | #define SD_OP_READ_OBJ 0x02 | |
29 | #define SD_OP_WRITE_OBJ 0x03 | |
cac8f4a6 LY |
30 | /* 0x04 is used internally by Sheepdog */ |
31 | #define SD_OP_DISCARD_OBJ 0x05 | |
33b1db1c MK |
32 | |
33 | #define SD_OP_NEW_VDI 0x11 | |
34 | #define SD_OP_LOCK_VDI 0x12 | |
35 | #define SD_OP_RELEASE_VDI 0x13 | |
36 | #define SD_OP_GET_VDI_INFO 0x14 | |
37 | #define SD_OP_READ_VDIS 0x15 | |
47622c44 | 38 | #define SD_OP_FLUSH_VDI 0x16 |
33b1db1c MK |
39 | |
40 | #define SD_FLAG_CMD_WRITE 0x01 | |
41 | #define SD_FLAG_CMD_COW 0x02 | |
0e7106d8 LY |
42 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
43 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
33b1db1c MK |
44 | |
45 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
46 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
47 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
48 | #define SD_RES_EIO 0x03 /* I/O error */ | |
49 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
50 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
51 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
52 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
53 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
54 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
55 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
56 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
57 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
58 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
59 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
60 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
61 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
62 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
63 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
64 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
65 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
66 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
67 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
68 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
69 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
fca23f0a | 70 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ |
33b1db1c MK |
71 | |
72 | /* | |
73 | * Object ID rules | |
74 | * | |
75 | * 0 - 19 (20 bits): data object space | |
76 | * 20 - 31 (12 bits): reserved data object space | |
77 | * 32 - 55 (24 bits): vdi object space | |
78 | * 56 - 59 ( 4 bits): reserved vdi object space | |
7acae208 | 79 | * 60 - 63 ( 4 bits): object type identifier space |
33b1db1c MK |
80 | */ |
81 | ||
82 | #define VDI_SPACE_SHIFT 32 | |
83 | #define VDI_BIT (UINT64_C(1) << 63) | |
84 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
85 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
86 | #define MAX_CHILDREN 1024 | |
87 | #define SD_MAX_VDI_LEN 256 | |
88 | #define SD_MAX_VDI_TAG_LEN 256 | |
89 | #define SD_NR_VDIS (1U << 24) | |
90 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
91 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
33b1db1c MK |
92 | |
93 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
94 | #define CURRENT_VDI_ID 0 | |
95 | ||
96 | typedef struct SheepdogReq { | |
97 | uint8_t proto_ver; | |
98 | uint8_t opcode; | |
99 | uint16_t flags; | |
100 | uint32_t epoch; | |
101 | uint32_t id; | |
102 | uint32_t data_length; | |
103 | uint32_t opcode_specific[8]; | |
104 | } SheepdogReq; | |
105 | ||
106 | typedef struct SheepdogRsp { | |
107 | uint8_t proto_ver; | |
108 | uint8_t opcode; | |
109 | uint16_t flags; | |
110 | uint32_t epoch; | |
111 | uint32_t id; | |
112 | uint32_t data_length; | |
113 | uint32_t result; | |
114 | uint32_t opcode_specific[7]; | |
115 | } SheepdogRsp; | |
116 | ||
117 | typedef struct SheepdogObjReq { | |
118 | uint8_t proto_ver; | |
119 | uint8_t opcode; | |
120 | uint16_t flags; | |
121 | uint32_t epoch; | |
122 | uint32_t id; | |
123 | uint32_t data_length; | |
124 | uint64_t oid; | |
125 | uint64_t cow_oid; | |
126 | uint32_t copies; | |
127 | uint32_t rsvd; | |
128 | uint64_t offset; | |
129 | } SheepdogObjReq; | |
130 | ||
131 | typedef struct SheepdogObjRsp { | |
132 | uint8_t proto_ver; | |
133 | uint8_t opcode; | |
134 | uint16_t flags; | |
135 | uint32_t epoch; | |
136 | uint32_t id; | |
137 | uint32_t data_length; | |
138 | uint32_t result; | |
139 | uint32_t copies; | |
140 | uint32_t pad[6]; | |
141 | } SheepdogObjRsp; | |
142 | ||
143 | typedef struct SheepdogVdiReq { | |
144 | uint8_t proto_ver; | |
145 | uint8_t opcode; | |
146 | uint16_t flags; | |
147 | uint32_t epoch; | |
148 | uint32_t id; | |
149 | uint32_t data_length; | |
150 | uint64_t vdi_size; | |
6f74c260 | 151 | uint32_t vdi_id; |
33b1db1c MK |
152 | uint32_t copies; |
153 | uint32_t snapid; | |
154 | uint32_t pad[3]; | |
155 | } SheepdogVdiReq; | |
156 | ||
157 | typedef struct SheepdogVdiRsp { | |
158 | uint8_t proto_ver; | |
159 | uint8_t opcode; | |
160 | uint16_t flags; | |
161 | uint32_t epoch; | |
162 | uint32_t id; | |
163 | uint32_t data_length; | |
164 | uint32_t result; | |
165 | uint32_t rsvd; | |
166 | uint32_t vdi_id; | |
167 | uint32_t pad[5]; | |
168 | } SheepdogVdiRsp; | |
169 | ||
170 | typedef struct SheepdogInode { | |
171 | char name[SD_MAX_VDI_LEN]; | |
172 | char tag[SD_MAX_VDI_TAG_LEN]; | |
173 | uint64_t ctime; | |
174 | uint64_t snap_ctime; | |
175 | uint64_t vm_clock_nsec; | |
176 | uint64_t vdi_size; | |
177 | uint64_t vm_state_size; | |
178 | uint16_t copy_policy; | |
179 | uint8_t nr_copies; | |
180 | uint8_t block_size_shift; | |
181 | uint32_t snap_id; | |
182 | uint32_t vdi_id; | |
183 | uint32_t parent_vdi_id; | |
184 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
185 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
186 | } SheepdogInode; | |
187 | ||
188 | /* | |
189 | * 64 bit FNV-1a non-zero initial basis | |
190 | */ | |
191 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
192 | ||
193 | /* | |
194 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
195 | */ | |
196 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
197 | { | |
198 | unsigned char *bp = buf; | |
199 | unsigned char *be = bp + len; | |
200 | while (bp < be) { | |
201 | hval ^= (uint64_t) *bp++; | |
202 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
203 | (hval << 7) + (hval << 8) + (hval << 40); | |
204 | } | |
205 | return hval; | |
206 | } | |
207 | ||
2f536801 | 208 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
33b1db1c MK |
209 | { |
210 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
211 | } | |
212 | ||
2f536801 | 213 | static inline bool is_data_obj(uint64_t oid) |
33b1db1c MK |
214 | { |
215 | return !(VDI_BIT & oid); | |
216 | } | |
217 | ||
218 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
219 | { | |
220 | return oid & (MAX_DATA_OBJS - 1); | |
221 | } | |
222 | ||
223 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) | |
224 | { | |
225 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
226 | } | |
227 | ||
228 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
229 | { | |
230 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
231 | } | |
232 | ||
233 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
234 | { | |
235 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
236 | } | |
237 | ||
2f536801 | 238 | static inline bool is_snapshot(struct SheepdogInode *inode) |
33b1db1c MK |
239 | { |
240 | return !!inode->snap_ctime; | |
241 | } | |
242 | ||
243 | #undef dprintf | |
244 | #ifdef DEBUG_SDOG | |
245 | #define dprintf(fmt, args...) \ | |
246 | do { \ | |
247 | fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ | |
248 | } while (0) | |
249 | #else | |
250 | #define dprintf(fmt, args...) | |
251 | #endif | |
252 | ||
253 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
254 | ||
255 | typedef struct AIOReq { | |
256 | SheepdogAIOCB *aiocb; | |
257 | unsigned int iov_offset; | |
258 | ||
259 | uint64_t oid; | |
260 | uint64_t base_oid; | |
261 | uint64_t offset; | |
262 | unsigned int data_len; | |
263 | uint8_t flags; | |
264 | uint32_t id; | |
265 | ||
c292ee6a | 266 | QLIST_ENTRY(AIOReq) aio_siblings; |
33b1db1c MK |
267 | } AIOReq; |
268 | ||
269 | enum AIOCBState { | |
270 | AIOCB_WRITE_UDATA, | |
271 | AIOCB_READ_UDATA, | |
47783072 | 272 | AIOCB_FLUSH_CACHE, |
cac8f4a6 | 273 | AIOCB_DISCARD_OBJ, |
33b1db1c MK |
274 | }; |
275 | ||
276 | struct SheepdogAIOCB { | |
277 | BlockDriverAIOCB common; | |
278 | ||
279 | QEMUIOVector *qiov; | |
280 | ||
281 | int64_t sector_num; | |
282 | int nb_sectors; | |
283 | ||
284 | int ret; | |
285 | enum AIOCBState aiocb_type; | |
286 | ||
2df46246 | 287 | Coroutine *coroutine; |
33b1db1c MK |
288 | void (*aio_done_func)(SheepdogAIOCB *); |
289 | ||
2f536801 | 290 | bool canceled; |
1d732d7d | 291 | int nr_pending; |
33b1db1c MK |
292 | }; |
293 | ||
294 | typedef struct BDRVSheepdogState { | |
295 | SheepdogInode inode; | |
296 | ||
297 | uint32_t min_dirty_data_idx; | |
298 | uint32_t max_dirty_data_idx; | |
299 | ||
300 | char name[SD_MAX_VDI_LEN]; | |
2f536801 | 301 | bool is_snapshot; |
0e7106d8 | 302 | uint32_t cache_flags; |
cac8f4a6 | 303 | bool discard_supported; |
33b1db1c | 304 | |
25af257d | 305 | char *host_spec; |
1b8bbb46 | 306 | bool is_unix; |
33b1db1c MK |
307 | int fd; |
308 | ||
2df46246 MK |
309 | CoMutex lock; |
310 | Coroutine *co_send; | |
311 | Coroutine *co_recv; | |
312 | ||
33b1db1c | 313 | uint32_t aioreq_seq_num; |
c292ee6a MK |
314 | QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
315 | QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head; | |
33b1db1c MK |
316 | } BDRVSheepdogState; |
317 | ||
318 | static const char * sd_strerror(int err) | |
319 | { | |
320 | int i; | |
321 | ||
322 | static const struct { | |
323 | int err; | |
324 | const char *desc; | |
325 | } errors[] = { | |
326 | {SD_RES_SUCCESS, "Success"}, | |
327 | {SD_RES_UNKNOWN, "Unknown error"}, | |
328 | {SD_RES_NO_OBJ, "No object found"}, | |
329 | {SD_RES_EIO, "I/O error"}, | |
330 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
331 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
332 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
333 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
334 | {SD_RES_NO_VDI, "No vdi found"}, | |
335 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
336 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
337 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
338 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
339 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
340 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
341 | {SD_RES_STARTUP, "The system is still booting"}, | |
342 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
343 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
344 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
345 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
346 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
347 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
348 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
349 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
350 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
fca23f0a | 351 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, |
33b1db1c MK |
352 | }; |
353 | ||
354 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
355 | if (errors[i].err == err) { | |
356 | return errors[i].desc; | |
357 | } | |
358 | } | |
359 | ||
360 | return "Invalid error code"; | |
361 | } | |
362 | ||
363 | /* | |
364 | * Sheepdog I/O handling: | |
365 | * | |
2df46246 | 366 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and |
c292ee6a | 367 | * link the requests to the inflight_list in the |
2df46246 MK |
368 | * BDRVSheepdogState. The function exits without waiting for |
369 | * receiving the response. | |
33b1db1c | 370 | * |
2df46246 | 371 | * 2. We receive the response in aio_read_response, the fd handler to |
33b1db1c MK |
372 | * the sheepdog connection. If metadata update is needed, we send |
373 | * the write request to the vdi object in sd_write_done, the write | |
2df46246 MK |
374 | * completion function. We switch back to sd_co_readv/writev after |
375 | * all the requests belonging to the AIOCB are finished. | |
33b1db1c MK |
376 | */ |
377 | ||
378 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
379 | uint64_t oid, unsigned int data_len, | |
380 | uint64_t offset, uint8_t flags, | |
381 | uint64_t base_oid, unsigned int iov_offset) | |
382 | { | |
383 | AIOReq *aio_req; | |
384 | ||
7267c094 | 385 | aio_req = g_malloc(sizeof(*aio_req)); |
33b1db1c MK |
386 | aio_req->aiocb = acb; |
387 | aio_req->iov_offset = iov_offset; | |
388 | aio_req->oid = oid; | |
389 | aio_req->base_oid = base_oid; | |
390 | aio_req->offset = offset; | |
391 | aio_req->data_len = data_len; | |
392 | aio_req->flags = flags; | |
393 | aio_req->id = s->aioreq_seq_num++; | |
394 | ||
1d732d7d | 395 | acb->nr_pending++; |
33b1db1c MK |
396 | return aio_req; |
397 | } | |
398 | ||
1d732d7d | 399 | static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
33b1db1c MK |
400 | { |
401 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1d732d7d | 402 | |
c292ee6a | 403 | QLIST_REMOVE(aio_req, aio_siblings); |
7267c094 | 404 | g_free(aio_req); |
33b1db1c | 405 | |
1d732d7d | 406 | acb->nr_pending--; |
33b1db1c MK |
407 | } |
408 | ||
d8716b41 | 409 | static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
33b1db1c MK |
410 | { |
411 | if (!acb->canceled) { | |
2df46246 | 412 | qemu_coroutine_enter(acb->coroutine, NULL); |
33b1db1c MK |
413 | } |
414 | qemu_aio_release(acb); | |
415 | } | |
416 | ||
417 | static void sd_aio_cancel(BlockDriverAIOCB *blockacb) | |
418 | { | |
419 | SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; | |
420 | ||
421 | /* | |
422 | * Sheepdog cannot cancel the requests which are already sent to | |
423 | * the servers, so we just complete the request with -EIO here. | |
424 | */ | |
2df46246 MK |
425 | acb->ret = -EIO; |
426 | qemu_coroutine_enter(acb->coroutine, NULL); | |
2f536801 | 427 | acb->canceled = true; |
33b1db1c MK |
428 | } |
429 | ||
d7331bed | 430 | static const AIOCBInfo sd_aiocb_info = { |
33b1db1c MK |
431 | .aiocb_size = sizeof(SheepdogAIOCB), |
432 | .cancel = sd_aio_cancel, | |
433 | }; | |
434 | ||
435 | static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov, | |
f700f8e3 | 436 | int64_t sector_num, int nb_sectors) |
33b1db1c MK |
437 | { |
438 | SheepdogAIOCB *acb; | |
439 | ||
f700f8e3 | 440 | acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); |
33b1db1c MK |
441 | |
442 | acb->qiov = qiov; | |
443 | ||
444 | acb->sector_num = sector_num; | |
445 | acb->nb_sectors = nb_sectors; | |
446 | ||
447 | acb->aio_done_func = NULL; | |
2f536801 | 448 | acb->canceled = false; |
2df46246 | 449 | acb->coroutine = qemu_coroutine_self(); |
33b1db1c | 450 | acb->ret = 0; |
1d732d7d | 451 | acb->nr_pending = 0; |
33b1db1c MK |
452 | return acb; |
453 | } | |
454 | ||
25af257d | 455 | static int connect_to_sdog(BDRVSheepdogState *s) |
33b1db1c | 456 | { |
25af257d MK |
457 | int fd; |
458 | Error *err = NULL; | |
33b1db1c | 459 | |
1b8bbb46 MK |
460 | if (s->is_unix) { |
461 | fd = unix_connect(s->host_spec, &err); | |
462 | } else { | |
463 | fd = inet_connect(s->host_spec, &err); | |
464 | ||
465 | if (err == NULL) { | |
466 | int ret = socket_set_nodelay(fd); | |
467 | if (ret < 0) { | |
468 | error_report("%s", strerror(errno)); | |
469 | } | |
470 | } | |
471 | } | |
33b1db1c | 472 | |
25af257d MK |
473 | if (err != NULL) { |
474 | qerror_report_err(err); | |
475 | error_free(err); | |
0d6db300 | 476 | } else { |
f9e8cacc | 477 | qemu_set_nonblock(fd); |
33b1db1c MK |
478 | } |
479 | ||
33b1db1c MK |
480 | return fd; |
481 | } | |
482 | ||
e0d93a89 MK |
483 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
484 | unsigned int *wlen) | |
47622c44 LY |
485 | { |
486 | int ret; | |
487 | ||
488 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
489 | if (ret < sizeof(*hdr)) { | |
490 | error_report("failed to send a req, %s", strerror(errno)); | |
eb092180 | 491 | return ret; |
47622c44 LY |
492 | } |
493 | ||
494 | ret = qemu_co_send(sockfd, data, *wlen); | |
495 | if (ret < *wlen) { | |
496 | error_report("failed to send a req, %s", strerror(errno)); | |
497 | } | |
498 | ||
499 | return ret; | |
500 | } | |
e0d93a89 | 501 | |
2dfcca3b MK |
502 | static void restart_co_req(void *opaque) |
503 | { | |
504 | Coroutine *co = opaque; | |
505 | ||
506 | qemu_coroutine_enter(co, NULL); | |
507 | } | |
508 | ||
ed9ba724 MK |
509 | static int have_co_req(void *opaque) |
510 | { | |
511 | /* this handler is set only when there is a pending request, so | |
512 | * always returns 1. */ | |
513 | return 1; | |
514 | } | |
515 | ||
cddd4ac7 MK |
516 | typedef struct SheepdogReqCo { |
517 | int sockfd; | |
518 | SheepdogReq *hdr; | |
519 | void *data; | |
520 | unsigned int *wlen; | |
521 | unsigned int *rlen; | |
522 | int ret; | |
523 | bool finished; | |
524 | } SheepdogReqCo; | |
525 | ||
526 | static coroutine_fn void do_co_req(void *opaque) | |
47622c44 LY |
527 | { |
528 | int ret; | |
2dfcca3b | 529 | Coroutine *co; |
cddd4ac7 MK |
530 | SheepdogReqCo *srco = opaque; |
531 | int sockfd = srco->sockfd; | |
532 | SheepdogReq *hdr = srco->hdr; | |
533 | void *data = srco->data; | |
534 | unsigned int *wlen = srco->wlen; | |
535 | unsigned int *rlen = srco->rlen; | |
2dfcca3b MK |
536 | |
537 | co = qemu_coroutine_self(); | |
ed9ba724 | 538 | qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, have_co_req, co); |
47622c44 | 539 | |
47622c44 LY |
540 | ret = send_co_req(sockfd, hdr, data, wlen); |
541 | if (ret < 0) { | |
542 | goto out; | |
543 | } | |
544 | ||
ed9ba724 | 545 | qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, have_co_req, co); |
2dfcca3b | 546 | |
47622c44 LY |
547 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); |
548 | if (ret < sizeof(*hdr)) { | |
549 | error_report("failed to get a rsp, %s", strerror(errno)); | |
cb595887 | 550 | ret = -errno; |
47622c44 LY |
551 | goto out; |
552 | } | |
553 | ||
554 | if (*rlen > hdr->data_length) { | |
555 | *rlen = hdr->data_length; | |
556 | } | |
557 | ||
558 | if (*rlen) { | |
559 | ret = qemu_co_recv(sockfd, data, *rlen); | |
560 | if (ret < *rlen) { | |
561 | error_report("failed to get the data, %s", strerror(errno)); | |
cb595887 | 562 | ret = -errno; |
47622c44 LY |
563 | goto out; |
564 | } | |
565 | } | |
566 | ret = 0; | |
567 | out: | |
ed9ba724 MK |
568 | /* there is at most one request for this sockfd, so it is safe to |
569 | * set each handler to NULL. */ | |
2dfcca3b | 570 | qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL); |
cddd4ac7 MK |
571 | |
572 | srco->ret = ret; | |
573 | srco->finished = true; | |
574 | } | |
575 | ||
576 | static int do_req(int sockfd, SheepdogReq *hdr, void *data, | |
577 | unsigned int *wlen, unsigned int *rlen) | |
578 | { | |
579 | Coroutine *co; | |
580 | SheepdogReqCo srco = { | |
581 | .sockfd = sockfd, | |
582 | .hdr = hdr, | |
583 | .data = data, | |
584 | .wlen = wlen, | |
585 | .rlen = rlen, | |
586 | .ret = 0, | |
587 | .finished = false, | |
588 | }; | |
589 | ||
590 | if (qemu_in_coroutine()) { | |
591 | do_co_req(&srco); | |
592 | } else { | |
593 | co = qemu_coroutine_create(do_co_req); | |
594 | qemu_coroutine_enter(co, &srco); | |
595 | while (!srco.finished) { | |
596 | qemu_aio_wait(); | |
597 | } | |
598 | } | |
599 | ||
600 | return srco.ret; | |
47622c44 LY |
601 | } |
602 | ||
d8716b41 | 603 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
2f536801 | 604 | struct iovec *iov, int niov, bool create, |
33b1db1c MK |
605 | enum AIOCBState aiocb_type); |
606 | ||
7dc1cde0 MK |
607 | |
608 | static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid) | |
609 | { | |
610 | AIOReq *aio_req; | |
611 | ||
612 | QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) { | |
613 | if (aio_req->oid == oid) { | |
614 | return aio_req; | |
615 | } | |
616 | } | |
617 | ||
618 | return NULL; | |
619 | } | |
620 | ||
33b1db1c MK |
621 | /* |
622 | * This function searchs pending requests to the object `oid', and | |
623 | * sends them. | |
624 | */ | |
c292ee6a | 625 | static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid) |
33b1db1c | 626 | { |
7dc1cde0 | 627 | AIOReq *aio_req; |
33b1db1c MK |
628 | SheepdogAIOCB *acb; |
629 | int ret; | |
630 | ||
7dc1cde0 | 631 | while ((aio_req = find_pending_req(s, oid)) != NULL) { |
33b1db1c | 632 | acb = aio_req->aiocb; |
c292ee6a MK |
633 | /* move aio_req from pending list to inflight one */ |
634 | QLIST_REMOVE(aio_req, aio_siblings); | |
635 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
33b1db1c | 636 | ret = add_aio_request(s, aio_req, acb->qiov->iov, |
2f536801 | 637 | acb->qiov->niov, false, acb->aiocb_type); |
33b1db1c | 638 | if (ret < 0) { |
6daf194d | 639 | error_report("add_aio_request is failed"); |
33b1db1c | 640 | free_aio_req(s, aio_req); |
1d732d7d | 641 | if (!acb->nr_pending) { |
33b1db1c MK |
642 | sd_finish_aiocb(acb); |
643 | } | |
644 | } | |
645 | } | |
646 | } | |
647 | ||
648 | /* | |
649 | * Receive responses of the I/O requests. | |
650 | * | |
651 | * This function is registered as a fd handler, and called from the | |
652 | * main loop when s->fd is ready for reading responses. | |
653 | */ | |
d8716b41 | 654 | static void coroutine_fn aio_read_response(void *opaque) |
33b1db1c MK |
655 | { |
656 | SheepdogObjRsp rsp; | |
657 | BDRVSheepdogState *s = opaque; | |
658 | int fd = s->fd; | |
659 | int ret; | |
660 | AIOReq *aio_req = NULL; | |
661 | SheepdogAIOCB *acb; | |
cac8f4a6 | 662 | uint64_t idx; |
33b1db1c | 663 | |
c292ee6a | 664 | if (QLIST_EMPTY(&s->inflight_aio_head)) { |
2df46246 | 665 | goto out; |
33b1db1c MK |
666 | } |
667 | ||
668 | /* read a header */ | |
8c5135f9 PB |
669 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); |
670 | if (ret < 0) { | |
6daf194d | 671 | error_report("failed to get the header, %s", strerror(errno)); |
2df46246 | 672 | goto out; |
33b1db1c MK |
673 | } |
674 | ||
c292ee6a MK |
675 | /* find the right aio_req from the inflight aio list */ |
676 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
33b1db1c MK |
677 | if (aio_req->id == rsp.id) { |
678 | break; | |
679 | } | |
680 | } | |
681 | if (!aio_req) { | |
6daf194d | 682 | error_report("cannot find aio_req %x", rsp.id); |
2df46246 | 683 | goto out; |
33b1db1c MK |
684 | } |
685 | ||
686 | acb = aio_req->aiocb; | |
687 | ||
688 | switch (acb->aiocb_type) { | |
689 | case AIOCB_WRITE_UDATA: | |
6d1acda8 MK |
690 | /* this coroutine context is no longer suitable for co_recv |
691 | * because we may send data to update vdi objects */ | |
692 | s->co_recv = NULL; | |
33b1db1c MK |
693 | if (!is_data_obj(aio_req->oid)) { |
694 | break; | |
695 | } | |
696 | idx = data_oid_to_idx(aio_req->oid); | |
697 | ||
698 | if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) { | |
699 | /* | |
700 | * If the object is newly created one, we need to update | |
701 | * the vdi object (metadata object). min_dirty_data_idx | |
702 | * and max_dirty_data_idx are changed to include updated | |
703 | * index between them. | |
704 | */ | |
bd751f22 LY |
705 | if (rsp.result == SD_RES_SUCCESS) { |
706 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
707 | s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); | |
708 | s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); | |
709 | } | |
33b1db1c MK |
710 | /* |
711 | * Some requests may be blocked because simultaneous | |
712 | * create requests are not allowed, so we search the | |
713 | * pending requests here. | |
714 | */ | |
d6b1ef89 | 715 | send_pending_req(s, aio_req->oid); |
33b1db1c MK |
716 | } |
717 | break; | |
718 | case AIOCB_READ_UDATA: | |
2fc8ae1d MT |
719 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
720 | aio_req->iov_offset, rsp.data_length); | |
8c5135f9 | 721 | if (ret < 0) { |
6daf194d | 722 | error_report("failed to get the data, %s", strerror(errno)); |
2df46246 | 723 | goto out; |
33b1db1c MK |
724 | } |
725 | break; | |
47783072 LY |
726 | case AIOCB_FLUSH_CACHE: |
727 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
728 | dprintf("disable cache since the server doesn't support it\n"); | |
729 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
730 | rsp.result = SD_RES_SUCCESS; | |
731 | } | |
732 | break; | |
cac8f4a6 LY |
733 | case AIOCB_DISCARD_OBJ: |
734 | switch (rsp.result) { | |
735 | case SD_RES_INVALID_PARMS: | |
736 | error_report("sheep(%s) doesn't support discard command", | |
737 | s->host_spec); | |
738 | rsp.result = SD_RES_SUCCESS; | |
739 | s->discard_supported = false; | |
740 | break; | |
741 | case SD_RES_SUCCESS: | |
742 | idx = data_oid_to_idx(aio_req->oid); | |
743 | s->inode.data_vdi_id[idx] = 0; | |
744 | break; | |
745 | default: | |
746 | break; | |
747 | } | |
33b1db1c MK |
748 | } |
749 | ||
750 | if (rsp.result != SD_RES_SUCCESS) { | |
751 | acb->ret = -EIO; | |
6daf194d | 752 | error_report("%s", sd_strerror(rsp.result)); |
33b1db1c MK |
753 | } |
754 | ||
1d732d7d MK |
755 | free_aio_req(s, aio_req); |
756 | if (!acb->nr_pending) { | |
33b1db1c MK |
757 | /* |
758 | * We've finished all requests which belong to the AIOCB, so | |
2df46246 | 759 | * we can switch back to sd_co_readv/writev now. |
33b1db1c MK |
760 | */ |
761 | acb->aio_done_func(acb); | |
762 | } | |
2df46246 MK |
763 | out: |
764 | s->co_recv = NULL; | |
765 | } | |
766 | ||
767 | static void co_read_response(void *opaque) | |
768 | { | |
769 | BDRVSheepdogState *s = opaque; | |
770 | ||
771 | if (!s->co_recv) { | |
772 | s->co_recv = qemu_coroutine_create(aio_read_response); | |
773 | } | |
774 | ||
775 | qemu_coroutine_enter(s->co_recv, opaque); | |
776 | } | |
777 | ||
778 | static void co_write_request(void *opaque) | |
779 | { | |
780 | BDRVSheepdogState *s = opaque; | |
781 | ||
782 | qemu_coroutine_enter(s->co_send, NULL); | |
33b1db1c MK |
783 | } |
784 | ||
785 | static int aio_flush_request(void *opaque) | |
786 | { | |
787 | BDRVSheepdogState *s = opaque; | |
788 | ||
c292ee6a MK |
789 | return !QLIST_EMPTY(&s->inflight_aio_head) || |
790 | !QLIST_EMPTY(&s->pending_aio_head); | |
33b1db1c MK |
791 | } |
792 | ||
33b1db1c MK |
793 | /* |
794 | * Return a socket discriptor to read/write objects. | |
795 | * | |
796 | * We cannot use this discriptor for other operations because | |
797 | * the block driver may be on waiting response from the server. | |
798 | */ | |
799 | static int get_sheep_fd(BDRVSheepdogState *s) | |
800 | { | |
1b8bbb46 | 801 | int fd; |
33b1db1c | 802 | |
25af257d | 803 | fd = connect_to_sdog(s); |
33b1db1c | 804 | if (fd < 0) { |
cb595887 | 805 | return fd; |
33b1db1c MK |
806 | } |
807 | ||
bafbd6a1 | 808 | qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s); |
33b1db1c MK |
809 | return fd; |
810 | } | |
811 | ||
5d6768e3 MK |
812 | static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
813 | char *vdi, uint32_t *snapid, char *tag) | |
814 | { | |
815 | URI *uri; | |
816 | QueryParams *qp = NULL; | |
817 | int ret = 0; | |
818 | ||
819 | uri = uri_parse(filename); | |
820 | if (!uri) { | |
821 | return -EINVAL; | |
822 | } | |
823 | ||
1b8bbb46 MK |
824 | /* transport */ |
825 | if (!strcmp(uri->scheme, "sheepdog")) { | |
826 | s->is_unix = false; | |
827 | } else if (!strcmp(uri->scheme, "sheepdog+tcp")) { | |
828 | s->is_unix = false; | |
829 | } else if (!strcmp(uri->scheme, "sheepdog+unix")) { | |
830 | s->is_unix = true; | |
831 | } else { | |
832 | ret = -EINVAL; | |
833 | goto out; | |
834 | } | |
835 | ||
5d6768e3 MK |
836 | if (uri->path == NULL || !strcmp(uri->path, "/")) { |
837 | ret = -EINVAL; | |
838 | goto out; | |
839 | } | |
840 | pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1); | |
841 | ||
1b8bbb46 MK |
842 | qp = query_params_parse(uri->query); |
843 | if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { | |
844 | ret = -EINVAL; | |
845 | goto out; | |
846 | } | |
847 | ||
848 | if (s->is_unix) { | |
849 | /* sheepdog+unix:///vdiname?socket=path */ | |
850 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
851 | ret = -EINVAL; | |
852 | goto out; | |
853 | } | |
854 | s->host_spec = g_strdup(qp->p[0].value); | |
855 | } else { | |
856 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
857 | s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR, | |
858 | uri->port ?: SD_DEFAULT_PORT); | |
859 | } | |
5d6768e3 MK |
860 | |
861 | /* snapshot tag */ | |
862 | if (uri->fragment) { | |
863 | *snapid = strtoul(uri->fragment, NULL, 10); | |
864 | if (*snapid == 0) { | |
865 | pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); | |
866 | } | |
867 | } else { | |
868 | *snapid = CURRENT_VDI_ID; /* search current vdi */ | |
869 | } | |
870 | ||
871 | out: | |
872 | if (qp) { | |
873 | query_params_free(qp); | |
874 | } | |
875 | uri_free(uri); | |
876 | return ret; | |
877 | } | |
878 | ||
33b1db1c | 879 | /* |
5d6768e3 | 880 | * Parse a filename (old syntax) |
33b1db1c MK |
881 | * |
882 | * filename must be one of the following formats: | |
883 | * 1. [vdiname] | |
884 | * 2. [vdiname]:[snapid] | |
885 | * 3. [vdiname]:[tag] | |
886 | * 4. [hostname]:[port]:[vdiname] | |
887 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
888 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
889 | * | |
890 | * You can boot from the snapshot images by specifying `snapid` or | |
891 | * `tag'. | |
892 | * | |
893 | * You can run VMs outside the Sheepdog cluster by specifying | |
894 | * `hostname' and `port' (experimental). | |
895 | */ | |
896 | static int parse_vdiname(BDRVSheepdogState *s, const char *filename, | |
897 | char *vdi, uint32_t *snapid, char *tag) | |
898 | { | |
5d6768e3 MK |
899 | char *p, *q, *uri; |
900 | const char *host_spec, *vdi_spec; | |
901 | int nr_sep, ret; | |
33b1db1c | 902 | |
5d6768e3 | 903 | strstart(filename, "sheepdog:", (const char **)&filename); |
7267c094 | 904 | p = q = g_strdup(filename); |
33b1db1c MK |
905 | |
906 | /* count the number of separators */ | |
907 | nr_sep = 0; | |
908 | while (*p) { | |
909 | if (*p == ':') { | |
910 | nr_sep++; | |
911 | } | |
912 | p++; | |
913 | } | |
914 | p = q; | |
915 | ||
5d6768e3 | 916 | /* use the first two tokens as host_spec. */ |
33b1db1c | 917 | if (nr_sep >= 2) { |
5d6768e3 | 918 | host_spec = p; |
33b1db1c | 919 | p = strchr(p, ':'); |
5d6768e3 | 920 | p++; |
33b1db1c MK |
921 | p = strchr(p, ':'); |
922 | *p++ = '\0'; | |
923 | } else { | |
5d6768e3 | 924 | host_spec = ""; |
33b1db1c MK |
925 | } |
926 | ||
5d6768e3 | 927 | vdi_spec = p; |
33b1db1c | 928 | |
5d6768e3 | 929 | p = strchr(vdi_spec, ':'); |
33b1db1c | 930 | if (p) { |
5d6768e3 | 931 | *p++ = '#'; |
33b1db1c MK |
932 | } |
933 | ||
5d6768e3 | 934 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); |
33b1db1c | 935 | |
5d6768e3 MK |
936 | ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
937 | ||
938 | g_free(q); | |
939 | g_free(uri); | |
940 | ||
941 | return ret; | |
33b1db1c MK |
942 | } |
943 | ||
982dcbf4 MK |
944 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, |
945 | uint32_t snapid, const char *tag, uint32_t *vid, | |
946 | bool lock) | |
33b1db1c MK |
947 | { |
948 | int ret, fd; | |
949 | SheepdogVdiReq hdr; | |
950 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
951 | unsigned int wlen, rlen = 0; | |
952 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
953 | ||
25af257d | 954 | fd = connect_to_sdog(s); |
33b1db1c | 955 | if (fd < 0) { |
cb595887 | 956 | return fd; |
33b1db1c MK |
957 | } |
958 | ||
3178e275 JM |
959 | /* This pair of strncpy calls ensures that the buffer is zero-filled, |
960 | * which is desirable since we'll soon be sending those bytes, and | |
961 | * don't want the send_req to read uninitialized data. | |
962 | */ | |
33b1db1c MK |
963 | strncpy(buf, filename, SD_MAX_VDI_LEN); |
964 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
965 | ||
966 | memset(&hdr, 0, sizeof(hdr)); | |
982dcbf4 | 967 | if (lock) { |
33b1db1c | 968 | hdr.opcode = SD_OP_LOCK_VDI; |
982dcbf4 MK |
969 | } else { |
970 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
33b1db1c MK |
971 | } |
972 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
973 | hdr.proto_ver = SD_PROTO_VER; | |
974 | hdr.data_length = wlen; | |
975 | hdr.snapid = snapid; | |
976 | hdr.flags = SD_FLAG_CMD_WRITE; | |
977 | ||
978 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
979 | if (ret) { | |
33b1db1c MK |
980 | goto out; |
981 | } | |
982 | ||
983 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 984 | error_report("cannot get vdi info, %s, %s %d %s", |
33b1db1c | 985 | sd_strerror(rsp->result), filename, snapid, tag); |
cb595887 MK |
986 | if (rsp->result == SD_RES_NO_VDI) { |
987 | ret = -ENOENT; | |
988 | } else { | |
989 | ret = -EIO; | |
990 | } | |
33b1db1c MK |
991 | goto out; |
992 | } | |
993 | *vid = rsp->vdi_id; | |
994 | ||
995 | ret = 0; | |
996 | out: | |
997 | closesocket(fd); | |
998 | return ret; | |
999 | } | |
1000 | ||
d8716b41 | 1001 | static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
2f536801 | 1002 | struct iovec *iov, int niov, bool create, |
33b1db1c MK |
1003 | enum AIOCBState aiocb_type) |
1004 | { | |
1005 | int nr_copies = s->inode.nr_copies; | |
1006 | SheepdogObjReq hdr; | |
47783072 | 1007 | unsigned int wlen = 0; |
33b1db1c MK |
1008 | int ret; |
1009 | uint64_t oid = aio_req->oid; | |
1010 | unsigned int datalen = aio_req->data_len; | |
1011 | uint64_t offset = aio_req->offset; | |
1012 | uint8_t flags = aio_req->flags; | |
1013 | uint64_t old_oid = aio_req->base_oid; | |
1014 | ||
1015 | if (!nr_copies) { | |
6daf194d | 1016 | error_report("bug"); |
33b1db1c MK |
1017 | } |
1018 | ||
1019 | memset(&hdr, 0, sizeof(hdr)); | |
1020 | ||
47783072 LY |
1021 | switch (aiocb_type) { |
1022 | case AIOCB_FLUSH_CACHE: | |
1023 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1024 | break; | |
1025 | case AIOCB_READ_UDATA: | |
33b1db1c MK |
1026 | hdr.opcode = SD_OP_READ_OBJ; |
1027 | hdr.flags = flags; | |
47783072 LY |
1028 | break; |
1029 | case AIOCB_WRITE_UDATA: | |
1030 | if (create) { | |
1031 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1032 | } else { | |
1033 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1034 | } | |
33b1db1c | 1035 | wlen = datalen; |
33b1db1c | 1036 | hdr.flags = SD_FLAG_CMD_WRITE | flags; |
47783072 | 1037 | break; |
cac8f4a6 LY |
1038 | case AIOCB_DISCARD_OBJ: |
1039 | hdr.opcode = SD_OP_DISCARD_OBJ; | |
1040 | break; | |
33b1db1c MK |
1041 | } |
1042 | ||
0e7106d8 LY |
1043 | if (s->cache_flags) { |
1044 | hdr.flags |= s->cache_flags; | |
47622c44 LY |
1045 | } |
1046 | ||
33b1db1c MK |
1047 | hdr.oid = oid; |
1048 | hdr.cow_oid = old_oid; | |
1049 | hdr.copies = s->inode.nr_copies; | |
1050 | ||
1051 | hdr.data_length = datalen; | |
1052 | hdr.offset = offset; | |
1053 | ||
1054 | hdr.id = aio_req->id; | |
1055 | ||
2df46246 MK |
1056 | qemu_co_mutex_lock(&s->lock); |
1057 | s->co_send = qemu_coroutine_self(); | |
1058 | qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, | |
bafbd6a1 | 1059 | aio_flush_request, s); |
128aa589 | 1060 | socket_set_cork(s->fd, 1); |
33b1db1c MK |
1061 | |
1062 | /* send a header */ | |
8c5135f9 PB |
1063 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); |
1064 | if (ret < 0) { | |
c3fecea5 | 1065 | qemu_co_mutex_unlock(&s->lock); |
6daf194d | 1066 | error_report("failed to send a req, %s", strerror(errno)); |
cb595887 | 1067 | return -errno; |
33b1db1c MK |
1068 | } |
1069 | ||
1070 | if (wlen) { | |
2fc8ae1d | 1071 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
8c5135f9 | 1072 | if (ret < 0) { |
c3fecea5 | 1073 | qemu_co_mutex_unlock(&s->lock); |
6daf194d | 1074 | error_report("failed to send a data, %s", strerror(errno)); |
cb595887 | 1075 | return -errno; |
33b1db1c MK |
1076 | } |
1077 | } | |
1078 | ||
128aa589 | 1079 | socket_set_cork(s->fd, 0); |
2df46246 | 1080 | qemu_aio_set_fd_handler(s->fd, co_read_response, NULL, |
bafbd6a1 | 1081 | aio_flush_request, s); |
2df46246 | 1082 | qemu_co_mutex_unlock(&s->lock); |
33b1db1c MK |
1083 | |
1084 | return 0; | |
1085 | } | |
1086 | ||
1087 | static int read_write_object(int fd, char *buf, uint64_t oid, int copies, | |
1088 | unsigned int datalen, uint64_t offset, | |
0e7106d8 | 1089 | bool write, bool create, uint32_t cache_flags) |
33b1db1c MK |
1090 | { |
1091 | SheepdogObjReq hdr; | |
1092 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1093 | unsigned int wlen, rlen; | |
1094 | int ret; | |
1095 | ||
1096 | memset(&hdr, 0, sizeof(hdr)); | |
1097 | ||
1098 | if (write) { | |
1099 | wlen = datalen; | |
1100 | rlen = 0; | |
1101 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1102 | if (create) { | |
1103 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1104 | } else { | |
1105 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1106 | } | |
1107 | } else { | |
1108 | wlen = 0; | |
1109 | rlen = datalen; | |
1110 | hdr.opcode = SD_OP_READ_OBJ; | |
1111 | } | |
47622c44 | 1112 | |
0e7106d8 | 1113 | hdr.flags |= cache_flags; |
47622c44 | 1114 | |
33b1db1c MK |
1115 | hdr.oid = oid; |
1116 | hdr.data_length = datalen; | |
1117 | hdr.offset = offset; | |
1118 | hdr.copies = copies; | |
1119 | ||
1120 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1121 | if (ret) { | |
6daf194d | 1122 | error_report("failed to send a request to the sheep"); |
cb595887 | 1123 | return ret; |
33b1db1c MK |
1124 | } |
1125 | ||
1126 | switch (rsp->result) { | |
1127 | case SD_RES_SUCCESS: | |
1128 | return 0; | |
1129 | default: | |
6daf194d | 1130 | error_report("%s", sd_strerror(rsp->result)); |
cb595887 | 1131 | return -EIO; |
33b1db1c MK |
1132 | } |
1133 | } | |
1134 | ||
1135 | static int read_object(int fd, char *buf, uint64_t oid, int copies, | |
0e7106d8 LY |
1136 | unsigned int datalen, uint64_t offset, |
1137 | uint32_t cache_flags) | |
33b1db1c | 1138 | { |
2f536801 | 1139 | return read_write_object(fd, buf, oid, copies, datalen, offset, false, |
0e7106d8 | 1140 | false, cache_flags); |
33b1db1c MK |
1141 | } |
1142 | ||
1143 | static int write_object(int fd, char *buf, uint64_t oid, int copies, | |
2f536801 | 1144 | unsigned int datalen, uint64_t offset, bool create, |
0e7106d8 | 1145 | uint32_t cache_flags) |
33b1db1c | 1146 | { |
2f536801 | 1147 | return read_write_object(fd, buf, oid, copies, datalen, offset, true, |
0e7106d8 | 1148 | create, cache_flags); |
33b1db1c MK |
1149 | } |
1150 | ||
c8c96350 KW |
1151 | /* TODO Convert to fine grained options */ |
1152 | static QemuOptsList runtime_opts = { | |
1153 | .name = "sheepdog", | |
1154 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1155 | .desc = { | |
1156 | { | |
1157 | .name = "filename", | |
1158 | .type = QEMU_OPT_STRING, | |
1159 | .help = "URL to the sheepdog image", | |
1160 | }, | |
1161 | { /* end of list */ } | |
1162 | }, | |
1163 | }; | |
1164 | ||
56d1b4d2 | 1165 | static int sd_open(BlockDriverState *bs, QDict *options, int flags) |
33b1db1c MK |
1166 | { |
1167 | int ret, fd; | |
1168 | uint32_t vid = 0; | |
1169 | BDRVSheepdogState *s = bs->opaque; | |
1170 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1171 | uint32_t snapid; | |
1172 | char *buf = NULL; | |
c8c96350 KW |
1173 | QemuOpts *opts; |
1174 | Error *local_err = NULL; | |
1175 | const char *filename; | |
1176 | ||
1177 | opts = qemu_opts_create_nofail(&runtime_opts); | |
1178 | qemu_opts_absorb_qdict(opts, options, &local_err); | |
1179 | if (error_is_set(&local_err)) { | |
1180 | qerror_report_err(local_err); | |
1181 | error_free(local_err); | |
1182 | ret = -EINVAL; | |
1183 | goto out; | |
1184 | } | |
1185 | ||
1186 | filename = qemu_opt_get(opts, "filename"); | |
33b1db1c | 1187 | |
c292ee6a MK |
1188 | QLIST_INIT(&s->inflight_aio_head); |
1189 | QLIST_INIT(&s->pending_aio_head); | |
33b1db1c MK |
1190 | s->fd = -1; |
1191 | ||
1192 | memset(vdi, 0, sizeof(vdi)); | |
1193 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1194 | |
1195 | if (strstr(filename, "://")) { | |
1196 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1197 | } else { | |
1198 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1199 | } | |
1200 | if (ret < 0) { | |
33b1db1c MK |
1201 | goto out; |
1202 | } | |
1203 | s->fd = get_sheep_fd(s); | |
1204 | if (s->fd < 0) { | |
cb595887 | 1205 | ret = s->fd; |
33b1db1c MK |
1206 | goto out; |
1207 | } | |
1208 | ||
982dcbf4 | 1209 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, true); |
33b1db1c MK |
1210 | if (ret) { |
1211 | goto out; | |
1212 | } | |
1213 | ||
0e7106d8 LY |
1214 | /* |
1215 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1216 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1217 | */ | |
1218 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1219 | if (flags & BDRV_O_NOCACHE) { | |
1220 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1221 | } | |
cac8f4a6 | 1222 | s->discard_supported = true; |
0e7106d8 | 1223 | |
622b6057 | 1224 | if (snapid || tag[0] != '\0') { |
33b1db1c | 1225 | dprintf("%" PRIx32 " snapshot inode was open.\n", vid); |
2f536801 | 1226 | s->is_snapshot = true; |
33b1db1c MK |
1227 | } |
1228 | ||
25af257d | 1229 | fd = connect_to_sdog(s); |
33b1db1c | 1230 | if (fd < 0) { |
cb595887 | 1231 | ret = fd; |
33b1db1c MK |
1232 | goto out; |
1233 | } | |
1234 | ||
7267c094 | 1235 | buf = g_malloc(SD_INODE_SIZE); |
47622c44 | 1236 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, |
0e7106d8 | 1237 | s->cache_flags); |
33b1db1c MK |
1238 | |
1239 | closesocket(fd); | |
1240 | ||
1241 | if (ret) { | |
1242 | goto out; | |
1243 | } | |
1244 | ||
1245 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1246 | s->min_dirty_data_idx = UINT32_MAX; | |
1247 | s->max_dirty_data_idx = 0; | |
1248 | ||
e8bfaa2f | 1249 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; |
3178e275 | 1250 | pstrcpy(s->name, sizeof(s->name), vdi); |
2df46246 | 1251 | qemu_co_mutex_init(&s->lock); |
c8c96350 | 1252 | qemu_opts_del(opts); |
7267c094 | 1253 | g_free(buf); |
33b1db1c MK |
1254 | return 0; |
1255 | out: | |
bafbd6a1 | 1256 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
33b1db1c MK |
1257 | if (s->fd >= 0) { |
1258 | closesocket(s->fd); | |
1259 | } | |
c8c96350 | 1260 | qemu_opts_del(opts); |
7267c094 | 1261 | g_free(buf); |
cb595887 | 1262 | return ret; |
33b1db1c MK |
1263 | } |
1264 | ||
25af257d MK |
1265 | static int do_sd_create(BDRVSheepdogState *s, char *filename, int64_t vdi_size, |
1266 | uint32_t base_vid, uint32_t *vdi_id, int snapshot) | |
33b1db1c MK |
1267 | { |
1268 | SheepdogVdiReq hdr; | |
1269 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1270 | int fd, ret; | |
1271 | unsigned int wlen, rlen = 0; | |
1272 | char buf[SD_MAX_VDI_LEN]; | |
1273 | ||
25af257d | 1274 | fd = connect_to_sdog(s); |
33b1db1c | 1275 | if (fd < 0) { |
cb595887 | 1276 | return fd; |
33b1db1c MK |
1277 | } |
1278 | ||
3178e275 JM |
1279 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename |
1280 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1281 | */ | |
33b1db1c | 1282 | memset(buf, 0, sizeof(buf)); |
3178e275 | 1283 | pstrcpy(buf, sizeof(buf), filename); |
33b1db1c MK |
1284 | |
1285 | memset(&hdr, 0, sizeof(hdr)); | |
1286 | hdr.opcode = SD_OP_NEW_VDI; | |
6f74c260 | 1287 | hdr.vdi_id = base_vid; |
33b1db1c MK |
1288 | |
1289 | wlen = SD_MAX_VDI_LEN; | |
1290 | ||
1291 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1292 | hdr.snapid = snapshot; | |
1293 | ||
1294 | hdr.data_length = wlen; | |
1295 | hdr.vdi_size = vdi_size; | |
1296 | ||
1297 | ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1298 | ||
1299 | closesocket(fd); | |
1300 | ||
1301 | if (ret) { | |
cb595887 | 1302 | return ret; |
33b1db1c MK |
1303 | } |
1304 | ||
1305 | if (rsp->result != SD_RES_SUCCESS) { | |
6daf194d | 1306 | error_report("%s, %s", sd_strerror(rsp->result), filename); |
33b1db1c MK |
1307 | return -EIO; |
1308 | } | |
1309 | ||
1310 | if (vdi_id) { | |
1311 | *vdi_id = rsp->vdi_id; | |
1312 | } | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
a8e0fdd7 MK |
1317 | static int sd_prealloc(const char *filename) |
1318 | { | |
1319 | BlockDriverState *bs = NULL; | |
1320 | uint32_t idx, max_idx; | |
1321 | int64_t vdi_size; | |
7267c094 | 1322 | void *buf = g_malloc0(SD_DATA_OBJ_SIZE); |
a8e0fdd7 MK |
1323 | int ret; |
1324 | ||
787e4a85 | 1325 | ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR); |
a8e0fdd7 MK |
1326 | if (ret < 0) { |
1327 | goto out; | |
1328 | } | |
1329 | ||
1330 | vdi_size = bdrv_getlength(bs); | |
1331 | if (vdi_size < 0) { | |
1332 | ret = vdi_size; | |
1333 | goto out; | |
1334 | } | |
1335 | max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); | |
1336 | ||
1337 | for (idx = 0; idx < max_idx; idx++) { | |
1338 | /* | |
1339 | * The created image can be a cloned image, so we need to read | |
1340 | * a data from the source image. | |
1341 | */ | |
1342 | ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1343 | if (ret < 0) { | |
1344 | goto out; | |
1345 | } | |
1346 | ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); | |
1347 | if (ret < 0) { | |
1348 | goto out; | |
1349 | } | |
1350 | } | |
1351 | out: | |
1352 | if (bs) { | |
1353 | bdrv_delete(bs); | |
1354 | } | |
7267c094 | 1355 | g_free(buf); |
a8e0fdd7 MK |
1356 | |
1357 | return ret; | |
1358 | } | |
1359 | ||
33b1db1c MK |
1360 | static int sd_create(const char *filename, QEMUOptionParameter *options) |
1361 | { | |
b6fc8245 | 1362 | int ret = 0; |
b4447363 | 1363 | uint32_t vid = 0, base_vid = 0; |
33b1db1c MK |
1364 | int64_t vdi_size = 0; |
1365 | char *backing_file = NULL; | |
b6fc8245 | 1366 | BDRVSheepdogState *s; |
b4447363 MK |
1367 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; |
1368 | uint32_t snapid; | |
2f536801 | 1369 | bool prealloc = false; |
33b1db1c | 1370 | |
b6fc8245 MK |
1371 | s = g_malloc0(sizeof(BDRVSheepdogState)); |
1372 | ||
b4447363 MK |
1373 | memset(vdi, 0, sizeof(vdi)); |
1374 | memset(tag, 0, sizeof(tag)); | |
5d6768e3 MK |
1375 | if (strstr(filename, "://")) { |
1376 | ret = sd_parse_uri(s, filename, vdi, &snapid, tag); | |
1377 | } else { | |
1378 | ret = parse_vdiname(s, filename, vdi, &snapid, tag); | |
1379 | } | |
1380 | if (ret < 0) { | |
b6fc8245 | 1381 | goto out; |
b4447363 MK |
1382 | } |
1383 | ||
33b1db1c MK |
1384 | while (options && options->name) { |
1385 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1386 | vdi_size = options->value.n; | |
1387 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1388 | backing_file = options->value.s; | |
a8e0fdd7 MK |
1389 | } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1390 | if (!options->value.s || !strcmp(options->value.s, "off")) { | |
2f536801 | 1391 | prealloc = false; |
a8e0fdd7 | 1392 | } else if (!strcmp(options->value.s, "full")) { |
2f536801 | 1393 | prealloc = true; |
a8e0fdd7 MK |
1394 | } else { |
1395 | error_report("Invalid preallocation mode: '%s'", | |
1396 | options->value.s); | |
b6fc8245 MK |
1397 | ret = -EINVAL; |
1398 | goto out; | |
a8e0fdd7 | 1399 | } |
33b1db1c MK |
1400 | } |
1401 | options++; | |
1402 | } | |
1403 | ||
1404 | if (vdi_size > SD_MAX_VDI_SIZE) { | |
6daf194d | 1405 | error_report("too big image size"); |
b6fc8245 MK |
1406 | ret = -EINVAL; |
1407 | goto out; | |
33b1db1c MK |
1408 | } |
1409 | ||
1410 | if (backing_file) { | |
1411 | BlockDriverState *bs; | |
1412 | BDRVSheepdogState *s; | |
1413 | BlockDriver *drv; | |
1414 | ||
1415 | /* Currently, only Sheepdog backing image is supported. */ | |
1416 | drv = bdrv_find_protocol(backing_file); | |
1417 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { | |
6daf194d | 1418 | error_report("backing_file must be a sheepdog image"); |
b6fc8245 MK |
1419 | ret = -EINVAL; |
1420 | goto out; | |
33b1db1c MK |
1421 | } |
1422 | ||
787e4a85 | 1423 | ret = bdrv_file_open(&bs, backing_file, NULL, 0); |
cb595887 | 1424 | if (ret < 0) { |
b6fc8245 | 1425 | goto out; |
cb595887 | 1426 | } |
33b1db1c MK |
1427 | |
1428 | s = bs->opaque; | |
1429 | ||
1430 | if (!is_snapshot(&s->inode)) { | |
6daf194d | 1431 | error_report("cannot clone from a non snapshot vdi"); |
33b1db1c | 1432 | bdrv_delete(bs); |
b6fc8245 MK |
1433 | ret = -EINVAL; |
1434 | goto out; | |
33b1db1c MK |
1435 | } |
1436 | ||
b4447363 | 1437 | base_vid = s->inode.vdi_id; |
33b1db1c MK |
1438 | bdrv_delete(bs); |
1439 | } | |
1440 | ||
25af257d | 1441 | ret = do_sd_create(s, vdi, vdi_size, base_vid, &vid, 0); |
a8e0fdd7 | 1442 | if (!prealloc || ret) { |
b6fc8245 | 1443 | goto out; |
a8e0fdd7 MK |
1444 | } |
1445 | ||
b6fc8245 MK |
1446 | ret = sd_prealloc(filename); |
1447 | out: | |
1448 | g_free(s); | |
1449 | return ret; | |
33b1db1c MK |
1450 | } |
1451 | ||
1452 | static void sd_close(BlockDriverState *bs) | |
1453 | { | |
1454 | BDRVSheepdogState *s = bs->opaque; | |
1455 | SheepdogVdiReq hdr; | |
1456 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1457 | unsigned int wlen, rlen = 0; | |
1458 | int fd, ret; | |
1459 | ||
1460 | dprintf("%s\n", s->name); | |
1461 | ||
25af257d | 1462 | fd = connect_to_sdog(s); |
33b1db1c MK |
1463 | if (fd < 0) { |
1464 | return; | |
1465 | } | |
1466 | ||
1467 | memset(&hdr, 0, sizeof(hdr)); | |
1468 | ||
1469 | hdr.opcode = SD_OP_RELEASE_VDI; | |
6f74c260 | 1470 | hdr.vdi_id = s->inode.vdi_id; |
33b1db1c MK |
1471 | wlen = strlen(s->name) + 1; |
1472 | hdr.data_length = wlen; | |
1473 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1474 | ||
1475 | ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); | |
1476 | ||
1477 | closesocket(fd); | |
1478 | ||
1479 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
1480 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
6daf194d | 1481 | error_report("%s, %s", sd_strerror(rsp->result), s->name); |
33b1db1c MK |
1482 | } |
1483 | ||
bafbd6a1 | 1484 | qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
33b1db1c | 1485 | closesocket(s->fd); |
25af257d | 1486 | g_free(s->host_spec); |
33b1db1c MK |
1487 | } |
1488 | ||
1489 | static int64_t sd_getlength(BlockDriverState *bs) | |
1490 | { | |
1491 | BDRVSheepdogState *s = bs->opaque; | |
1492 | ||
1493 | return s->inode.vdi_size; | |
1494 | } | |
1495 | ||
1496 | static int sd_truncate(BlockDriverState *bs, int64_t offset) | |
1497 | { | |
1498 | BDRVSheepdogState *s = bs->opaque; | |
1499 | int ret, fd; | |
1500 | unsigned int datalen; | |
1501 | ||
1502 | if (offset < s->inode.vdi_size) { | |
6daf194d | 1503 | error_report("shrinking is not supported"); |
33b1db1c MK |
1504 | return -EINVAL; |
1505 | } else if (offset > SD_MAX_VDI_SIZE) { | |
6daf194d | 1506 | error_report("too big image size"); |
33b1db1c MK |
1507 | return -EINVAL; |
1508 | } | |
1509 | ||
25af257d | 1510 | fd = connect_to_sdog(s); |
33b1db1c | 1511 | if (fd < 0) { |
cb595887 | 1512 | return fd; |
33b1db1c MK |
1513 | } |
1514 | ||
1515 | /* we don't need to update entire object */ | |
1516 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1517 | s->inode.vdi_size = offset; | |
1518 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
0e7106d8 | 1519 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
33b1db1c MK |
1520 | close(fd); |
1521 | ||
1522 | if (ret < 0) { | |
6daf194d | 1523 | error_report("failed to update an inode."); |
33b1db1c MK |
1524 | } |
1525 | ||
cb595887 | 1526 | return ret; |
33b1db1c MK |
1527 | } |
1528 | ||
1529 | /* | |
1530 | * This function is called after writing data objects. If we need to | |
1531 | * update metadata, this sends a write request to the vdi object. | |
2df46246 | 1532 | * Otherwise, this switches back to sd_co_readv/writev. |
33b1db1c | 1533 | */ |
d8716b41 | 1534 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
33b1db1c MK |
1535 | { |
1536 | int ret; | |
1537 | BDRVSheepdogState *s = acb->common.bs->opaque; | |
1538 | struct iovec iov; | |
1539 | AIOReq *aio_req; | |
1540 | uint32_t offset, data_len, mn, mx; | |
1541 | ||
1542 | mn = s->min_dirty_data_idx; | |
1543 | mx = s->max_dirty_data_idx; | |
1544 | if (mn <= mx) { | |
1545 | /* we need to update the vdi object. */ | |
1546 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
1547 | mn * sizeof(s->inode.data_vdi_id[0]); | |
1548 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
1549 | ||
1550 | s->min_dirty_data_idx = UINT32_MAX; | |
1551 | s->max_dirty_data_idx = 0; | |
1552 | ||
1553 | iov.iov_base = &s->inode; | |
1554 | iov.iov_len = sizeof(s->inode); | |
1555 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
1556 | data_len, offset, 0, 0, offset); | |
c292ee6a | 1557 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
2f536801 | 1558 | ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); |
33b1db1c MK |
1559 | if (ret) { |
1560 | free_aio_req(s, aio_req); | |
1561 | acb->ret = -EIO; | |
1562 | goto out; | |
1563 | } | |
1564 | ||
1565 | acb->aio_done_func = sd_finish_aiocb; | |
1566 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1567 | return; | |
1568 | } | |
1569 | out: | |
1570 | sd_finish_aiocb(acb); | |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * Create a writable VDI from a snapshot | |
1575 | */ | |
1576 | static int sd_create_branch(BDRVSheepdogState *s) | |
1577 | { | |
1578 | int ret, fd; | |
1579 | uint32_t vid; | |
1580 | char *buf; | |
1581 | ||
1582 | dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); | |
1583 | ||
7267c094 | 1584 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 1585 | |
25af257d | 1586 | ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1); |
33b1db1c MK |
1587 | if (ret) { |
1588 | goto out; | |
1589 | } | |
1590 | ||
1591 | dprintf("%" PRIx32 " is created.\n", vid); | |
1592 | ||
25af257d | 1593 | fd = connect_to_sdog(s); |
33b1db1c | 1594 | if (fd < 0) { |
cb595887 | 1595 | ret = fd; |
33b1db1c MK |
1596 | goto out; |
1597 | } | |
1598 | ||
1599 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, | |
0e7106d8 | 1600 | SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1601 | |
1602 | closesocket(fd); | |
1603 | ||
1604 | if (ret < 0) { | |
1605 | goto out; | |
1606 | } | |
1607 | ||
1608 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1609 | ||
2f536801 | 1610 | s->is_snapshot = false; |
33b1db1c MK |
1611 | ret = 0; |
1612 | dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id); | |
1613 | ||
1614 | out: | |
7267c094 | 1615 | g_free(buf); |
33b1db1c MK |
1616 | |
1617 | return ret; | |
1618 | } | |
1619 | ||
1620 | /* | |
1621 | * Send I/O requests to the server. | |
1622 | * | |
1623 | * This function sends requests to the server, links the requests to | |
c292ee6a | 1624 | * the inflight_list in BDRVSheepdogState, and exits without |
33b1db1c MK |
1625 | * waiting the response. The responses are received in the |
1626 | * `aio_read_response' function which is called from the main loop as | |
1627 | * a fd handler. | |
2df46246 MK |
1628 | * |
1629 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
1630 | * request and -errno in error cases. | |
33b1db1c | 1631 | */ |
d8716b41 | 1632 | static int coroutine_fn sd_co_rw_vector(void *p) |
33b1db1c MK |
1633 | { |
1634 | SheepdogAIOCB *acb = p; | |
1635 | int ret = 0; | |
e8bfaa2f LY |
1636 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; |
1637 | unsigned long idx = acb->sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE; | |
33b1db1c | 1638 | uint64_t oid; |
e8bfaa2f | 1639 | uint64_t offset = (acb->sector_num * BDRV_SECTOR_SIZE) % SD_DATA_OBJ_SIZE; |
33b1db1c MK |
1640 | BDRVSheepdogState *s = acb->common.bs->opaque; |
1641 | SheepdogInode *inode = &s->inode; | |
1642 | AIOReq *aio_req; | |
1643 | ||
33b1db1c MK |
1644 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { |
1645 | /* | |
1646 | * In the case we open the snapshot VDI, Sheepdog creates the | |
1647 | * writable VDI when we do a write operation first. | |
1648 | */ | |
1649 | ret = sd_create_branch(s); | |
1650 | if (ret) { | |
1651 | acb->ret = -EIO; | |
1652 | goto out; | |
1653 | } | |
1654 | } | |
1655 | ||
1d732d7d MK |
1656 | /* |
1657 | * Make sure we don't free the aiocb before we are done with all requests. | |
1658 | * This additional reference is dropped at the end of this function. | |
1659 | */ | |
1660 | acb->nr_pending++; | |
1661 | ||
33b1db1c MK |
1662 | while (done != total) { |
1663 | uint8_t flags = 0; | |
1664 | uint64_t old_oid = 0; | |
2f536801 | 1665 | bool create = false; |
33b1db1c MK |
1666 | |
1667 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
1668 | ||
1669 | len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); | |
1670 | ||
19db9b90 CH |
1671 | switch (acb->aiocb_type) { |
1672 | case AIOCB_READ_UDATA: | |
1673 | if (!inode->data_vdi_id[idx]) { | |
1674 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
33b1db1c MK |
1675 | goto done; |
1676 | } | |
19db9b90 CH |
1677 | break; |
1678 | case AIOCB_WRITE_UDATA: | |
1679 | if (!inode->data_vdi_id[idx]) { | |
2f536801 | 1680 | create = true; |
19db9b90 CH |
1681 | } else if (!is_data_obj_writable(inode, idx)) { |
1682 | /* Copy-On-Write */ | |
2f536801 | 1683 | create = true; |
19db9b90 CH |
1684 | old_oid = oid; |
1685 | flags = SD_FLAG_CMD_COW; | |
1686 | } | |
1687 | break; | |
cac8f4a6 LY |
1688 | case AIOCB_DISCARD_OBJ: |
1689 | /* | |
1690 | * We discard the object only when the whole object is | |
1691 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
1692 | */ | |
1693 | if (len != SD_DATA_OBJ_SIZE || inode->data_vdi_id[idx] == 0) { | |
1694 | goto done; | |
1695 | } | |
1696 | break; | |
19db9b90 CH |
1697 | default: |
1698 | break; | |
33b1db1c MK |
1699 | } |
1700 | ||
1701 | if (create) { | |
1b6ac998 MK |
1702 | dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1703 | inode->vdi_id, oid, | |
33b1db1c MK |
1704 | vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
1705 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
1b6ac998 | 1706 | dprintf("new oid %" PRIx64 "\n", oid); |
33b1db1c MK |
1707 | } |
1708 | ||
1709 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); | |
1710 | ||
1711 | if (create) { | |
1712 | AIOReq *areq; | |
c292ee6a | 1713 | QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { |
33b1db1c MK |
1714 | if (areq->oid == oid) { |
1715 | /* | |
1716 | * Sheepdog cannot handle simultaneous create | |
1717 | * requests to the same object. So we cannot send | |
1718 | * the request until the previous request | |
1719 | * finishes. | |
1720 | */ | |
1721 | aio_req->flags = 0; | |
1722 | aio_req->base_oid = 0; | |
c292ee6a MK |
1723 | QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, |
1724 | aio_siblings); | |
33b1db1c MK |
1725 | goto done; |
1726 | } | |
1727 | } | |
1728 | } | |
1729 | ||
c292ee6a | 1730 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
33b1db1c MK |
1731 | ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
1732 | create, acb->aiocb_type); | |
1733 | if (ret < 0) { | |
6daf194d | 1734 | error_report("add_aio_request is failed"); |
33b1db1c MK |
1735 | free_aio_req(s, aio_req); |
1736 | acb->ret = -EIO; | |
1737 | goto out; | |
1738 | } | |
1739 | done: | |
1740 | offset = 0; | |
1741 | idx++; | |
1742 | done += len; | |
1743 | } | |
1744 | out: | |
1d732d7d | 1745 | if (!--acb->nr_pending) { |
2df46246 | 1746 | return acb->ret; |
33b1db1c | 1747 | } |
2df46246 | 1748 | return 1; |
33b1db1c MK |
1749 | } |
1750 | ||
a968168c | 1751 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1752 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1753 | { |
1754 | SheepdogAIOCB *acb; | |
2df46246 | 1755 | int ret; |
33b1db1c MK |
1756 | |
1757 | if (bs->growable && sector_num + nb_sectors > bs->total_sectors) { | |
e8bfaa2f | 1758 | ret = sd_truncate(bs, (sector_num + nb_sectors) * BDRV_SECTOR_SIZE); |
cb595887 MK |
1759 | if (ret < 0) { |
1760 | return ret; | |
33b1db1c MK |
1761 | } |
1762 | bs->total_sectors = sector_num + nb_sectors; | |
1763 | } | |
1764 | ||
f700f8e3 | 1765 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
1766 | acb->aio_done_func = sd_write_done; |
1767 | acb->aiocb_type = AIOCB_WRITE_UDATA; | |
1768 | ||
2df46246 MK |
1769 | ret = sd_co_rw_vector(acb); |
1770 | if (ret <= 0) { | |
1771 | qemu_aio_release(acb); | |
1772 | return ret; | |
1773 | } | |
1774 | ||
1775 | qemu_coroutine_yield(); | |
1776 | ||
1777 | return acb->ret; | |
33b1db1c MK |
1778 | } |
1779 | ||
a968168c | 1780 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
2df46246 | 1781 | int nb_sectors, QEMUIOVector *qiov) |
33b1db1c MK |
1782 | { |
1783 | SheepdogAIOCB *acb; | |
19db9b90 | 1784 | int ret; |
33b1db1c | 1785 | |
f700f8e3 | 1786 | acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
33b1db1c MK |
1787 | acb->aiocb_type = AIOCB_READ_UDATA; |
1788 | acb->aio_done_func = sd_finish_aiocb; | |
1789 | ||
2df46246 MK |
1790 | ret = sd_co_rw_vector(acb); |
1791 | if (ret <= 0) { | |
1792 | qemu_aio_release(acb); | |
1793 | return ret; | |
1794 | } | |
1795 | ||
1796 | qemu_coroutine_yield(); | |
1797 | ||
1798 | return acb->ret; | |
33b1db1c MK |
1799 | } |
1800 | ||
47622c44 LY |
1801 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
1802 | { | |
1803 | BDRVSheepdogState *s = bs->opaque; | |
47783072 LY |
1804 | SheepdogAIOCB *acb; |
1805 | AIOReq *aio_req; | |
47622c44 | 1806 | int ret; |
47622c44 | 1807 | |
0e7106d8 | 1808 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { |
47622c44 LY |
1809 | return 0; |
1810 | } | |
1811 | ||
f700f8e3 | 1812 | acb = sd_aio_setup(bs, NULL, 0, 0); |
47783072 LY |
1813 | acb->aiocb_type = AIOCB_FLUSH_CACHE; |
1814 | acb->aio_done_func = sd_finish_aiocb; | |
47622c44 | 1815 | |
47783072 LY |
1816 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
1817 | 0, 0, 0, 0, 0); | |
1818 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
1819 | ret = add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type); | |
1820 | if (ret < 0) { | |
1821 | error_report("add_aio_request is failed"); | |
1822 | free_aio_req(s, aio_req); | |
1823 | qemu_aio_release(acb); | |
47622c44 LY |
1824 | return ret; |
1825 | } | |
1826 | ||
47783072 LY |
1827 | qemu_coroutine_yield(); |
1828 | return acb->ret; | |
47622c44 LY |
1829 | } |
1830 | ||
33b1db1c MK |
1831 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
1832 | { | |
1833 | BDRVSheepdogState *s = bs->opaque; | |
1834 | int ret, fd; | |
1835 | uint32_t new_vid; | |
1836 | SheepdogInode *inode; | |
1837 | unsigned int datalen; | |
1838 | ||
1b6ac998 | 1839 | dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
33b1db1c MK |
1840 | "is_snapshot %d\n", sn_info->name, sn_info->id_str, |
1841 | s->name, sn_info->vm_state_size, s->is_snapshot); | |
1842 | ||
1843 | if (s->is_snapshot) { | |
1844 | error_report("You can't create a snapshot of a snapshot VDI, " | |
6daf194d | 1845 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
33b1db1c MK |
1846 | |
1847 | return -EINVAL; | |
1848 | } | |
1849 | ||
1850 | dprintf("%s %s\n", sn_info->name, sn_info->id_str); | |
1851 | ||
1852 | s->inode.vm_state_size = sn_info->vm_state_size; | |
1853 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
3178e275 JM |
1854 | /* It appears that inode.tag does not require a NUL terminator, |
1855 | * which means this use of strncpy is ok. | |
1856 | */ | |
33b1db1c MK |
1857 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); |
1858 | /* we don't need to update entire object */ | |
1859 | datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id); | |
1860 | ||
1861 | /* refresh inode. */ | |
25af257d | 1862 | fd = connect_to_sdog(s); |
33b1db1c | 1863 | if (fd < 0) { |
cb595887 | 1864 | ret = fd; |
33b1db1c MK |
1865 | goto cleanup; |
1866 | } | |
1867 | ||
1868 | ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id), | |
0e7106d8 | 1869 | s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
33b1db1c | 1870 | if (ret < 0) { |
6daf194d | 1871 | error_report("failed to write snapshot's inode."); |
33b1db1c MK |
1872 | goto cleanup; |
1873 | } | |
1874 | ||
25af257d MK |
1875 | ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, |
1876 | 1); | |
33b1db1c | 1877 | if (ret < 0) { |
6daf194d | 1878 | error_report("failed to create inode for snapshot. %s", |
33b1db1c | 1879 | strerror(errno)); |
33b1db1c MK |
1880 | goto cleanup; |
1881 | } | |
1882 | ||
7267c094 | 1883 | inode = (SheepdogInode *)g_malloc(datalen); |
33b1db1c MK |
1884 | |
1885 | ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid), | |
0e7106d8 | 1886 | s->inode.nr_copies, datalen, 0, s->cache_flags); |
33b1db1c MK |
1887 | |
1888 | if (ret < 0) { | |
6daf194d | 1889 | error_report("failed to read new inode info. %s", strerror(errno)); |
33b1db1c MK |
1890 | goto cleanup; |
1891 | } | |
1892 | ||
1893 | memcpy(&s->inode, inode, datalen); | |
1894 | dprintf("s->inode: name %s snap_id %x oid %x\n", | |
1895 | s->inode.name, s->inode.snap_id, s->inode.vdi_id); | |
1896 | ||
1897 | cleanup: | |
1898 | closesocket(fd); | |
1899 | return ret; | |
1900 | } | |
1901 | ||
1902 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) | |
1903 | { | |
1904 | BDRVSheepdogState *s = bs->opaque; | |
1905 | BDRVSheepdogState *old_s; | |
1906 | char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN]; | |
1907 | char *buf = NULL; | |
1908 | uint32_t vid; | |
1909 | uint32_t snapid = 0; | |
cb595887 | 1910 | int ret = 0, fd; |
33b1db1c | 1911 | |
7267c094 | 1912 | old_s = g_malloc(sizeof(BDRVSheepdogState)); |
33b1db1c MK |
1913 | |
1914 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
1915 | ||
3178e275 | 1916 | pstrcpy(vdi, sizeof(vdi), s->name); |
33b1db1c | 1917 | |
33b1db1c | 1918 | snapid = strtoul(snapshot_id, NULL, 10); |
3178e275 JM |
1919 | if (snapid) { |
1920 | tag[0] = 0; | |
1921 | } else { | |
1922 | pstrcpy(tag, sizeof(tag), s->name); | |
33b1db1c MK |
1923 | } |
1924 | ||
982dcbf4 | 1925 | ret = find_vdi_name(s, vdi, snapid, tag, &vid, false); |
33b1db1c | 1926 | if (ret) { |
6daf194d | 1927 | error_report("Failed to find_vdi_name"); |
33b1db1c MK |
1928 | goto out; |
1929 | } | |
1930 | ||
25af257d | 1931 | fd = connect_to_sdog(s); |
33b1db1c | 1932 | if (fd < 0) { |
cb595887 | 1933 | ret = fd; |
33b1db1c MK |
1934 | goto out; |
1935 | } | |
1936 | ||
7267c094 | 1937 | buf = g_malloc(SD_INODE_SIZE); |
33b1db1c | 1938 | ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
0e7106d8 | 1939 | SD_INODE_SIZE, 0, s->cache_flags); |
33b1db1c MK |
1940 | |
1941 | closesocket(fd); | |
1942 | ||
1943 | if (ret) { | |
33b1db1c MK |
1944 | goto out; |
1945 | } | |
1946 | ||
1947 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1948 | ||
1949 | if (!s->inode.vm_state_size) { | |
6daf194d | 1950 | error_report("Invalid snapshot"); |
33b1db1c MK |
1951 | ret = -ENOENT; |
1952 | goto out; | |
1953 | } | |
1954 | ||
2f536801 | 1955 | s->is_snapshot = true; |
33b1db1c | 1956 | |
7267c094 AL |
1957 | g_free(buf); |
1958 | g_free(old_s); | |
33b1db1c MK |
1959 | |
1960 | return 0; | |
1961 | out: | |
1962 | /* recover bdrv_sd_state */ | |
1963 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
7267c094 AL |
1964 | g_free(buf); |
1965 | g_free(old_s); | |
33b1db1c | 1966 | |
6daf194d | 1967 | error_report("failed to open. recover old bdrv_sd_state."); |
33b1db1c MK |
1968 | |
1969 | return ret; | |
1970 | } | |
1971 | ||
1972 | static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) | |
1973 | { | |
1974 | /* FIXME: Delete specified snapshot id. */ | |
1975 | return 0; | |
1976 | } | |
1977 | ||
33b1db1c MK |
1978 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
1979 | { | |
1980 | BDRVSheepdogState *s = bs->opaque; | |
1981 | SheepdogReq req; | |
1982 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
1983 | QEMUSnapshotInfo *sn_tab = NULL; | |
1984 | unsigned wlen, rlen; | |
1985 | int found = 0; | |
1986 | static SheepdogInode inode; | |
1987 | unsigned long *vdi_inuse; | |
1988 | unsigned int start_nr; | |
1989 | uint64_t hval; | |
1990 | uint32_t vid; | |
1991 | ||
7267c094 | 1992 | vdi_inuse = g_malloc(max); |
33b1db1c | 1993 | |
25af257d | 1994 | fd = connect_to_sdog(s); |
33b1db1c | 1995 | if (fd < 0) { |
cb595887 | 1996 | ret = fd; |
33b1db1c MK |
1997 | goto out; |
1998 | } | |
1999 | ||
2000 | rlen = max; | |
2001 | wlen = 0; | |
2002 | ||
2003 | memset(&req, 0, sizeof(req)); | |
2004 | ||
2005 | req.opcode = SD_OP_READ_VDIS; | |
2006 | req.data_length = max; | |
2007 | ||
2008 | ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); | |
2009 | ||
2010 | closesocket(fd); | |
2011 | if (ret) { | |
2012 | goto out; | |
2013 | } | |
2014 | ||
7267c094 | 2015 | sn_tab = g_malloc0(nr * sizeof(*sn_tab)); |
33b1db1c MK |
2016 | |
2017 | /* calculate a vdi id with hash function */ | |
2018 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2019 | start_nr = hval & (SD_NR_VDIS - 1); | |
2020 | ||
25af257d | 2021 | fd = connect_to_sdog(s); |
33b1db1c | 2022 | if (fd < 0) { |
cb595887 | 2023 | ret = fd; |
33b1db1c MK |
2024 | goto out; |
2025 | } | |
2026 | ||
2027 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2028 | if (!test_bit(vid, vdi_inuse)) { | |
2029 | break; | |
2030 | } | |
2031 | ||
2032 | /* we don't need to read entire object */ | |
2033 | ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid), | |
47622c44 | 2034 | 0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
0e7106d8 | 2035 | s->cache_flags); |
33b1db1c MK |
2036 | |
2037 | if (ret) { | |
2038 | continue; | |
2039 | } | |
2040 | ||
2041 | if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) { | |
2042 | sn_tab[found].date_sec = inode.snap_ctime >> 32; | |
2043 | sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff; | |
2044 | sn_tab[found].vm_state_size = inode.vm_state_size; | |
2045 | sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; | |
2046 | ||
2047 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", | |
2048 | inode.snap_id); | |
3178e275 JM |
2049 | pstrcpy(sn_tab[found].name, |
2050 | MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), | |
2051 | inode.tag); | |
33b1db1c MK |
2052 | found++; |
2053 | } | |
2054 | } | |
2055 | ||
2056 | closesocket(fd); | |
2057 | out: | |
2058 | *psn_tab = sn_tab; | |
2059 | ||
7267c094 | 2060 | g_free(vdi_inuse); |
33b1db1c | 2061 | |
cb595887 MK |
2062 | if (ret < 0) { |
2063 | return ret; | |
2064 | } | |
2065 | ||
33b1db1c MK |
2066 | return found; |
2067 | } | |
2068 | ||
2069 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
2070 | int64_t pos, int size, int load) | |
2071 | { | |
2f536801 MK |
2072 | bool create; |
2073 | int fd, ret = 0, remaining = size; | |
33b1db1c MK |
2074 | unsigned int data_len; |
2075 | uint64_t vmstate_oid; | |
2076 | uint32_t vdi_index; | |
2077 | uint64_t offset; | |
2078 | ||
25af257d | 2079 | fd = connect_to_sdog(s); |
33b1db1c | 2080 | if (fd < 0) { |
cb595887 | 2081 | return fd; |
33b1db1c MK |
2082 | } |
2083 | ||
6f3c714e | 2084 | while (remaining) { |
33b1db1c MK |
2085 | vdi_index = pos / SD_DATA_OBJ_SIZE; |
2086 | offset = pos % SD_DATA_OBJ_SIZE; | |
2087 | ||
1f7a48de | 2088 | data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset); |
33b1db1c MK |
2089 | |
2090 | vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index); | |
2091 | ||
2092 | create = (offset == 0); | |
2093 | if (load) { | |
2094 | ret = read_object(fd, (char *)data, vmstate_oid, | |
47622c44 | 2095 | s->inode.nr_copies, data_len, offset, |
0e7106d8 | 2096 | s->cache_flags); |
33b1db1c MK |
2097 | } else { |
2098 | ret = write_object(fd, (char *)data, vmstate_oid, | |
47622c44 | 2099 | s->inode.nr_copies, data_len, offset, create, |
0e7106d8 | 2100 | s->cache_flags); |
33b1db1c MK |
2101 | } |
2102 | ||
2103 | if (ret < 0) { | |
6daf194d | 2104 | error_report("failed to save vmstate %s", strerror(errno)); |
33b1db1c MK |
2105 | goto cleanup; |
2106 | } | |
2107 | ||
2108 | pos += data_len; | |
1f7a48de | 2109 | data += data_len; |
6f3c714e | 2110 | remaining -= data_len; |
33b1db1c | 2111 | } |
6f3c714e | 2112 | ret = size; |
33b1db1c MK |
2113 | cleanup: |
2114 | closesocket(fd); | |
2115 | return ret; | |
2116 | } | |
2117 | ||
cf8074b3 KW |
2118 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
2119 | int64_t pos) | |
33b1db1c MK |
2120 | { |
2121 | BDRVSheepdogState *s = bs->opaque; | |
cf8074b3 KW |
2122 | void *buf; |
2123 | int ret; | |
33b1db1c | 2124 | |
cf8074b3 KW |
2125 | buf = qemu_blockalign(bs, qiov->size); |
2126 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
2127 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
2128 | qemu_vfree(buf); | |
2129 | ||
2130 | return ret; | |
33b1db1c MK |
2131 | } |
2132 | ||
2133 | static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, | |
2134 | int64_t pos, int size) | |
2135 | { | |
2136 | BDRVSheepdogState *s = bs->opaque; | |
2137 | ||
2138 | return do_load_save_vmstate(s, data, pos, size, 1); | |
2139 | } | |
2140 | ||
2141 | ||
cac8f4a6 LY |
2142 | static coroutine_fn int sd_co_discard(BlockDriverState *bs, int64_t sector_num, |
2143 | int nb_sectors) | |
2144 | { | |
2145 | SheepdogAIOCB *acb; | |
2146 | QEMUIOVector dummy; | |
2147 | BDRVSheepdogState *s = bs->opaque; | |
2148 | int ret; | |
2149 | ||
2150 | if (!s->discard_supported) { | |
2151 | return 0; | |
2152 | } | |
2153 | ||
2154 | acb = sd_aio_setup(bs, &dummy, sector_num, nb_sectors); | |
2155 | acb->aiocb_type = AIOCB_DISCARD_OBJ; | |
2156 | acb->aio_done_func = sd_finish_aiocb; | |
2157 | ||
2158 | ret = sd_co_rw_vector(acb); | |
2159 | if (ret <= 0) { | |
2160 | qemu_aio_release(acb); | |
2161 | return ret; | |
2162 | } | |
2163 | ||
2164 | qemu_coroutine_yield(); | |
2165 | ||
2166 | return acb->ret; | |
2167 | } | |
2168 | ||
8d71c631 LY |
2169 | static coroutine_fn int |
2170 | sd_co_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
2171 | int *pnum) | |
2172 | { | |
2173 | BDRVSheepdogState *s = bs->opaque; | |
2174 | SheepdogInode *inode = &s->inode; | |
2175 | unsigned long start = sector_num * BDRV_SECTOR_SIZE / SD_DATA_OBJ_SIZE, | |
2176 | end = DIV_ROUND_UP((sector_num + nb_sectors) * | |
2177 | BDRV_SECTOR_SIZE, SD_DATA_OBJ_SIZE); | |
2178 | unsigned long idx; | |
2179 | int ret = 1; | |
2180 | ||
2181 | for (idx = start; idx < end; idx++) { | |
2182 | if (inode->data_vdi_id[idx] == 0) { | |
2183 | break; | |
2184 | } | |
2185 | } | |
2186 | if (idx == start) { | |
2187 | /* Get the longest length of unallocated sectors */ | |
2188 | ret = 0; | |
2189 | for (idx = start + 1; idx < end; idx++) { | |
2190 | if (inode->data_vdi_id[idx] != 0) { | |
2191 | break; | |
2192 | } | |
2193 | } | |
2194 | } | |
2195 | ||
2196 | *pnum = (idx - start) * SD_DATA_OBJ_SIZE / BDRV_SECTOR_SIZE; | |
2197 | if (*pnum > nb_sectors) { | |
2198 | *pnum = nb_sectors; | |
2199 | } | |
2200 | return ret; | |
2201 | } | |
2202 | ||
33b1db1c MK |
2203 | static QEMUOptionParameter sd_create_options[] = { |
2204 | { | |
2205 | .name = BLOCK_OPT_SIZE, | |
2206 | .type = OPT_SIZE, | |
2207 | .help = "Virtual disk size" | |
2208 | }, | |
2209 | { | |
2210 | .name = BLOCK_OPT_BACKING_FILE, | |
2211 | .type = OPT_STRING, | |
2212 | .help = "File name of a base image" | |
2213 | }, | |
a8e0fdd7 MK |
2214 | { |
2215 | .name = BLOCK_OPT_PREALLOC, | |
2216 | .type = OPT_STRING, | |
2217 | .help = "Preallocation mode (allowed values: off, full)" | |
2218 | }, | |
33b1db1c MK |
2219 | { NULL } |
2220 | }; | |
2221 | ||
5d6768e3 | 2222 | static BlockDriver bdrv_sheepdog = { |
33b1db1c MK |
2223 | .format_name = "sheepdog", |
2224 | .protocol_name = "sheepdog", | |
2225 | .instance_size = sizeof(BDRVSheepdogState), | |
2226 | .bdrv_file_open = sd_open, | |
2227 | .bdrv_close = sd_close, | |
2228 | .bdrv_create = sd_create, | |
2229 | .bdrv_getlength = sd_getlength, | |
2230 | .bdrv_truncate = sd_truncate, | |
2231 | ||
2df46246 MK |
2232 | .bdrv_co_readv = sd_co_readv, |
2233 | .bdrv_co_writev = sd_co_writev, | |
47622c44 | 2234 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
cac8f4a6 | 2235 | .bdrv_co_discard = sd_co_discard, |
8d71c631 | 2236 | .bdrv_co_is_allocated = sd_co_is_allocated, |
33b1db1c MK |
2237 | |
2238 | .bdrv_snapshot_create = sd_snapshot_create, | |
2239 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2240 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2241 | .bdrv_snapshot_list = sd_snapshot_list, | |
2242 | ||
2243 | .bdrv_save_vmstate = sd_save_vmstate, | |
2244 | .bdrv_load_vmstate = sd_load_vmstate, | |
2245 | ||
2246 | .create_options = sd_create_options, | |
2247 | }; | |
2248 | ||
5d6768e3 MK |
2249 | static BlockDriver bdrv_sheepdog_tcp = { |
2250 | .format_name = "sheepdog", | |
2251 | .protocol_name = "sheepdog+tcp", | |
2252 | .instance_size = sizeof(BDRVSheepdogState), | |
2253 | .bdrv_file_open = sd_open, | |
2254 | .bdrv_close = sd_close, | |
2255 | .bdrv_create = sd_create, | |
2256 | .bdrv_getlength = sd_getlength, | |
2257 | .bdrv_truncate = sd_truncate, | |
2258 | ||
2259 | .bdrv_co_readv = sd_co_readv, | |
2260 | .bdrv_co_writev = sd_co_writev, | |
2261 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2262 | .bdrv_co_discard = sd_co_discard, |
8d71c631 | 2263 | .bdrv_co_is_allocated = sd_co_is_allocated, |
5d6768e3 MK |
2264 | |
2265 | .bdrv_snapshot_create = sd_snapshot_create, | |
2266 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2267 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2268 | .bdrv_snapshot_list = sd_snapshot_list, | |
2269 | ||
2270 | .bdrv_save_vmstate = sd_save_vmstate, | |
2271 | .bdrv_load_vmstate = sd_load_vmstate, | |
2272 | ||
2273 | .create_options = sd_create_options, | |
2274 | }; | |
2275 | ||
1b8bbb46 MK |
2276 | static BlockDriver bdrv_sheepdog_unix = { |
2277 | .format_name = "sheepdog", | |
2278 | .protocol_name = "sheepdog+unix", | |
2279 | .instance_size = sizeof(BDRVSheepdogState), | |
2280 | .bdrv_file_open = sd_open, | |
2281 | .bdrv_close = sd_close, | |
2282 | .bdrv_create = sd_create, | |
2283 | .bdrv_getlength = sd_getlength, | |
2284 | .bdrv_truncate = sd_truncate, | |
2285 | ||
2286 | .bdrv_co_readv = sd_co_readv, | |
2287 | .bdrv_co_writev = sd_co_writev, | |
2288 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
cac8f4a6 | 2289 | .bdrv_co_discard = sd_co_discard, |
8d71c631 | 2290 | .bdrv_co_is_allocated = sd_co_is_allocated, |
1b8bbb46 MK |
2291 | |
2292 | .bdrv_snapshot_create = sd_snapshot_create, | |
2293 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
2294 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
2295 | .bdrv_snapshot_list = sd_snapshot_list, | |
2296 | ||
2297 | .bdrv_save_vmstate = sd_save_vmstate, | |
2298 | .bdrv_load_vmstate = sd_load_vmstate, | |
2299 | ||
2300 | .create_options = sd_create_options, | |
2301 | }; | |
2302 | ||
33b1db1c MK |
2303 | static void bdrv_sheepdog_init(void) |
2304 | { | |
2305 | bdrv_register(&bdrv_sheepdog); | |
5d6768e3 | 2306 | bdrv_register(&bdrv_sheepdog_tcp); |
1b8bbb46 | 2307 | bdrv_register(&bdrv_sheepdog_unix); |
33b1db1c MK |
2308 | } |
2309 | block_init(bdrv_sheepdog_init); |