root / block / sheepdog.c @ 737e150e
History | View | Annotate | Download (54.4 kB)
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/>.
|
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.
|
13 |
*/
|
14 |
|
15 |
#include "qemu-common.h" |
16 |
#include "qemu-error.h" |
17 |
#include "qemu_socket.h" |
18 |
#include "block/block_int.h" |
19 |
#include "bitops.h" |
20 |
|
21 |
#define SD_PROTO_VER 0x01 |
22 |
|
23 |
#define SD_DEFAULT_ADDR "localhost" |
24 |
#define SD_DEFAULT_PORT "7000" |
25 |
|
26 |
#define SD_OP_CREATE_AND_WRITE_OBJ 0x01 |
27 |
#define SD_OP_READ_OBJ 0x02 |
28 |
#define SD_OP_WRITE_OBJ 0x03 |
29 |
|
30 |
#define SD_OP_NEW_VDI 0x11 |
31 |
#define SD_OP_LOCK_VDI 0x12 |
32 |
#define SD_OP_RELEASE_VDI 0x13 |
33 |
#define SD_OP_GET_VDI_INFO 0x14 |
34 |
#define SD_OP_READ_VDIS 0x15 |
35 |
#define SD_OP_FLUSH_VDI 0x16 |
36 |
|
37 |
#define SD_FLAG_CMD_WRITE 0x01 |
38 |
#define SD_FLAG_CMD_COW 0x02 |
39 |
#define SD_FLAG_CMD_CACHE 0x04 |
40 |
|
41 |
#define SD_RES_SUCCESS 0x00 /* Success */ |
42 |
#define SD_RES_UNKNOWN 0x01 /* Unknown error */ |
43 |
#define SD_RES_NO_OBJ 0x02 /* No object found */ |
44 |
#define SD_RES_EIO 0x03 /* I/O error */ |
45 |
#define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ |
46 |
#define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ |
47 |
#define SD_RES_SYSTEM_ERROR 0x06 /* System error */ |
48 |
#define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ |
49 |
#define SD_RES_NO_VDI 0x08 /* No vdi found */ |
50 |
#define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ |
51 |
#define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ |
52 |
#define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ |
53 |
#define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ |
54 |
#define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ |
55 |
#define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ |
56 |
#define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ |
57 |
#define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ |
58 |
#define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ |
59 |
#define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ |
60 |
#define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ |
61 |
#define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ |
62 |
#define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ |
63 |
#define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ |
64 |
#define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ |
65 |
#define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ |
66 |
|
67 |
/*
|
68 |
* Object ID rules
|
69 |
*
|
70 |
* 0 - 19 (20 bits): data object space
|
71 |
* 20 - 31 (12 bits): reserved data object space
|
72 |
* 32 - 55 (24 bits): vdi object space
|
73 |
* 56 - 59 ( 4 bits): reserved vdi object space
|
74 |
* 60 - 63 ( 4 bits): object type identifier space
|
75 |
*/
|
76 |
|
77 |
#define VDI_SPACE_SHIFT 32 |
78 |
#define VDI_BIT (UINT64_C(1) << 63) |
79 |
#define VMSTATE_BIT (UINT64_C(1) << 62) |
80 |
#define MAX_DATA_OBJS (UINT64_C(1) << 20) |
81 |
#define MAX_CHILDREN 1024 |
82 |
#define SD_MAX_VDI_LEN 256 |
83 |
#define SD_MAX_VDI_TAG_LEN 256 |
84 |
#define SD_NR_VDIS (1U << 24) |
85 |
#define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) |
86 |
#define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
|
87 |
#define SECTOR_SIZE 512 |
88 |
|
89 |
#define SD_INODE_SIZE (sizeof(SheepdogInode)) |
90 |
#define CURRENT_VDI_ID 0 |
91 |
|
92 |
typedef struct SheepdogReq { |
93 |
uint8_t proto_ver; |
94 |
uint8_t opcode; |
95 |
uint16_t flags; |
96 |
uint32_t epoch; |
97 |
uint32_t id; |
98 |
uint32_t data_length; |
99 |
uint32_t opcode_specific[8];
|
100 |
} SheepdogReq; |
101 |
|
102 |
typedef struct SheepdogRsp { |
103 |
uint8_t proto_ver; |
104 |
uint8_t opcode; |
105 |
uint16_t flags; |
106 |
uint32_t epoch; |
107 |
uint32_t id; |
108 |
uint32_t data_length; |
109 |
uint32_t result; |
110 |
uint32_t opcode_specific[7];
|
111 |
} SheepdogRsp; |
112 |
|
113 |
typedef struct SheepdogObjReq { |
114 |
uint8_t proto_ver; |
115 |
uint8_t opcode; |
116 |
uint16_t flags; |
117 |
uint32_t epoch; |
118 |
uint32_t id; |
119 |
uint32_t data_length; |
120 |
uint64_t oid; |
121 |
uint64_t cow_oid; |
122 |
uint32_t copies; |
123 |
uint32_t rsvd; |
124 |
uint64_t offset; |
125 |
} SheepdogObjReq; |
126 |
|
127 |
typedef struct SheepdogObjRsp { |
128 |
uint8_t proto_ver; |
129 |
uint8_t opcode; |
130 |
uint16_t flags; |
131 |
uint32_t epoch; |
132 |
uint32_t id; |
133 |
uint32_t data_length; |
134 |
uint32_t result; |
135 |
uint32_t copies; |
136 |
uint32_t pad[6];
|
137 |
} SheepdogObjRsp; |
138 |
|
139 |
typedef struct SheepdogVdiReq { |
140 |
uint8_t proto_ver; |
141 |
uint8_t opcode; |
142 |
uint16_t flags; |
143 |
uint32_t epoch; |
144 |
uint32_t id; |
145 |
uint32_t data_length; |
146 |
uint64_t vdi_size; |
147 |
uint32_t base_vdi_id; |
148 |
uint32_t copies; |
149 |
uint32_t snapid; |
150 |
uint32_t pad[3];
|
151 |
} SheepdogVdiReq; |
152 |
|
153 |
typedef struct SheepdogVdiRsp { |
154 |
uint8_t proto_ver; |
155 |
uint8_t opcode; |
156 |
uint16_t flags; |
157 |
uint32_t epoch; |
158 |
uint32_t id; |
159 |
uint32_t data_length; |
160 |
uint32_t result; |
161 |
uint32_t rsvd; |
162 |
uint32_t vdi_id; |
163 |
uint32_t pad[5];
|
164 |
} SheepdogVdiRsp; |
165 |
|
166 |
typedef struct SheepdogInode { |
167 |
char name[SD_MAX_VDI_LEN];
|
168 |
char tag[SD_MAX_VDI_TAG_LEN];
|
169 |
uint64_t ctime; |
170 |
uint64_t snap_ctime; |
171 |
uint64_t vm_clock_nsec; |
172 |
uint64_t vdi_size; |
173 |
uint64_t vm_state_size; |
174 |
uint16_t copy_policy; |
175 |
uint8_t nr_copies; |
176 |
uint8_t block_size_shift; |
177 |
uint32_t snap_id; |
178 |
uint32_t vdi_id; |
179 |
uint32_t parent_vdi_id; |
180 |
uint32_t child_vdi_id[MAX_CHILDREN]; |
181 |
uint32_t data_vdi_id[MAX_DATA_OBJS]; |
182 |
} SheepdogInode; |
183 |
|
184 |
/*
|
185 |
* 64 bit FNV-1a non-zero initial basis
|
186 |
*/
|
187 |
#define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) |
188 |
|
189 |
/*
|
190 |
* 64 bit Fowler/Noll/Vo FNV-1a hash code
|
191 |
*/
|
192 |
static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) |
193 |
{ |
194 |
unsigned char *bp = buf; |
195 |
unsigned char *be = bp + len; |
196 |
while (bp < be) {
|
197 |
hval ^= (uint64_t) *bp++; |
198 |
hval += (hval << 1) + (hval << 4) + (hval << 5) + |
199 |
(hval << 7) + (hval << 8) + (hval << 40); |
200 |
} |
201 |
return hval;
|
202 |
} |
203 |
|
204 |
static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
205 |
{ |
206 |
return inode->vdi_id == inode->data_vdi_id[idx];
|
207 |
} |
208 |
|
209 |
static inline bool is_data_obj(uint64_t oid) |
210 |
{ |
211 |
return !(VDI_BIT & oid);
|
212 |
} |
213 |
|
214 |
static inline uint64_t data_oid_to_idx(uint64_t oid) |
215 |
{ |
216 |
return oid & (MAX_DATA_OBJS - 1); |
217 |
} |
218 |
|
219 |
static inline uint64_t vid_to_vdi_oid(uint32_t vid) |
220 |
{ |
221 |
return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
|
222 |
} |
223 |
|
224 |
static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) |
225 |
{ |
226 |
return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
|
227 |
} |
228 |
|
229 |
static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) |
230 |
{ |
231 |
return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
|
232 |
} |
233 |
|
234 |
static inline bool is_snapshot(struct SheepdogInode *inode) |
235 |
{ |
236 |
return !!inode->snap_ctime;
|
237 |
} |
238 |
|
239 |
#undef dprintf
|
240 |
#ifdef DEBUG_SDOG
|
241 |
#define dprintf(fmt, args...) \
|
242 |
do { \
|
243 |
fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ |
244 |
} while (0) |
245 |
#else
|
246 |
#define dprintf(fmt, args...)
|
247 |
#endif
|
248 |
|
249 |
typedef struct SheepdogAIOCB SheepdogAIOCB; |
250 |
|
251 |
typedef struct AIOReq { |
252 |
SheepdogAIOCB *aiocb; |
253 |
unsigned int iov_offset; |
254 |
|
255 |
uint64_t oid; |
256 |
uint64_t base_oid; |
257 |
uint64_t offset; |
258 |
unsigned int data_len; |
259 |
uint8_t flags; |
260 |
uint32_t id; |
261 |
|
262 |
QLIST_ENTRY(AIOReq) aio_siblings; |
263 |
} AIOReq; |
264 |
|
265 |
enum AIOCBState {
|
266 |
AIOCB_WRITE_UDATA, |
267 |
AIOCB_READ_UDATA, |
268 |
}; |
269 |
|
270 |
struct SheepdogAIOCB {
|
271 |
BlockDriverAIOCB common; |
272 |
|
273 |
QEMUIOVector *qiov; |
274 |
|
275 |
int64_t sector_num; |
276 |
int nb_sectors;
|
277 |
|
278 |
int ret;
|
279 |
enum AIOCBState aiocb_type;
|
280 |
|
281 |
Coroutine *coroutine; |
282 |
void (*aio_done_func)(SheepdogAIOCB *);
|
283 |
|
284 |
bool canceled;
|
285 |
int nr_pending;
|
286 |
}; |
287 |
|
288 |
typedef struct BDRVSheepdogState { |
289 |
SheepdogInode inode; |
290 |
|
291 |
uint32_t min_dirty_data_idx; |
292 |
uint32_t max_dirty_data_idx; |
293 |
|
294 |
char name[SD_MAX_VDI_LEN];
|
295 |
bool is_snapshot;
|
296 |
bool cache_enabled;
|
297 |
|
298 |
char *addr;
|
299 |
char *port;
|
300 |
int fd;
|
301 |
int flush_fd;
|
302 |
|
303 |
CoMutex lock; |
304 |
Coroutine *co_send; |
305 |
Coroutine *co_recv; |
306 |
|
307 |
uint32_t aioreq_seq_num; |
308 |
QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
309 |
QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head; |
310 |
} BDRVSheepdogState; |
311 |
|
312 |
static const char * sd_strerror(int err) |
313 |
{ |
314 |
int i;
|
315 |
|
316 |
static const struct { |
317 |
int err;
|
318 |
const char *desc; |
319 |
} errors[] = { |
320 |
{SD_RES_SUCCESS, "Success"},
|
321 |
{SD_RES_UNKNOWN, "Unknown error"},
|
322 |
{SD_RES_NO_OBJ, "No object found"},
|
323 |
{SD_RES_EIO, "I/O error"},
|
324 |
{SD_RES_VDI_EXIST, "VDI exists already"},
|
325 |
{SD_RES_INVALID_PARMS, "Invalid parameters"},
|
326 |
{SD_RES_SYSTEM_ERROR, "System error"},
|
327 |
{SD_RES_VDI_LOCKED, "VDI is already locked"},
|
328 |
{SD_RES_NO_VDI, "No vdi found"},
|
329 |
{SD_RES_NO_BASE_VDI, "No base VDI found"},
|
330 |
{SD_RES_VDI_READ, "Failed read the requested VDI"},
|
331 |
{SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
|
332 |
{SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
|
333 |
{SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
|
334 |
{SD_RES_NO_TAG, "Failed to find the requested tag"},
|
335 |
{SD_RES_STARTUP, "The system is still booting"},
|
336 |
{SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
|
337 |
{SD_RES_SHUTDOWN, "The system is shutting down"},
|
338 |
{SD_RES_NO_MEM, "Out of memory on the server"},
|
339 |
{SD_RES_FULL_VDI, "We already have the maximum vdis"},
|
340 |
{SD_RES_VER_MISMATCH, "Protocol version mismatch"},
|
341 |
{SD_RES_NO_SPACE, "Server has no space for new objects"},
|
342 |
{SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
|
343 |
{SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
|
344 |
{SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
|
345 |
}; |
346 |
|
347 |
for (i = 0; i < ARRAY_SIZE(errors); ++i) { |
348 |
if (errors[i].err == err) {
|
349 |
return errors[i].desc;
|
350 |
} |
351 |
} |
352 |
|
353 |
return "Invalid error code"; |
354 |
} |
355 |
|
356 |
/*
|
357 |
* Sheepdog I/O handling:
|
358 |
*
|
359 |
* 1. In sd_co_rw_vector, we send the I/O requests to the server and
|
360 |
* link the requests to the inflight_list in the
|
361 |
* BDRVSheepdogState. The function exits without waiting for
|
362 |
* receiving the response.
|
363 |
*
|
364 |
* 2. We receive the response in aio_read_response, the fd handler to
|
365 |
* the sheepdog connection. If metadata update is needed, we send
|
366 |
* the write request to the vdi object in sd_write_done, the write
|
367 |
* completion function. We switch back to sd_co_readv/writev after
|
368 |
* all the requests belonging to the AIOCB are finished.
|
369 |
*/
|
370 |
|
371 |
static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, |
372 |
uint64_t oid, unsigned int data_len, |
373 |
uint64_t offset, uint8_t flags, |
374 |
uint64_t base_oid, unsigned int iov_offset) |
375 |
{ |
376 |
AIOReq *aio_req; |
377 |
|
378 |
aio_req = g_malloc(sizeof(*aio_req));
|
379 |
aio_req->aiocb = acb; |
380 |
aio_req->iov_offset = iov_offset; |
381 |
aio_req->oid = oid; |
382 |
aio_req->base_oid = base_oid; |
383 |
aio_req->offset = offset; |
384 |
aio_req->data_len = data_len; |
385 |
aio_req->flags = flags; |
386 |
aio_req->id = s->aioreq_seq_num++; |
387 |
|
388 |
acb->nr_pending++; |
389 |
return aio_req;
|
390 |
} |
391 |
|
392 |
static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
393 |
{ |
394 |
SheepdogAIOCB *acb = aio_req->aiocb; |
395 |
|
396 |
QLIST_REMOVE(aio_req, aio_siblings); |
397 |
g_free(aio_req); |
398 |
|
399 |
acb->nr_pending--; |
400 |
} |
401 |
|
402 |
static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
403 |
{ |
404 |
if (!acb->canceled) {
|
405 |
qemu_coroutine_enter(acb->coroutine, NULL);
|
406 |
} |
407 |
qemu_aio_release(acb); |
408 |
} |
409 |
|
410 |
static void sd_aio_cancel(BlockDriverAIOCB *blockacb) |
411 |
{ |
412 |
SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; |
413 |
|
414 |
/*
|
415 |
* Sheepdog cannot cancel the requests which are already sent to
|
416 |
* the servers, so we just complete the request with -EIO here.
|
417 |
*/
|
418 |
acb->ret = -EIO; |
419 |
qemu_coroutine_enter(acb->coroutine, NULL);
|
420 |
acb->canceled = true;
|
421 |
} |
422 |
|
423 |
static const AIOCBInfo sd_aiocb_info = { |
424 |
.aiocb_size = sizeof(SheepdogAIOCB),
|
425 |
.cancel = sd_aio_cancel, |
426 |
}; |
427 |
|
428 |
static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov,
|
429 |
int64_t sector_num, int nb_sectors,
|
430 |
BlockDriverCompletionFunc *cb, void *opaque)
|
431 |
{ |
432 |
SheepdogAIOCB *acb; |
433 |
|
434 |
acb = qemu_aio_get(&sd_aiocb_info, bs, cb, opaque); |
435 |
|
436 |
acb->qiov = qiov; |
437 |
|
438 |
acb->sector_num = sector_num; |
439 |
acb->nb_sectors = nb_sectors; |
440 |
|
441 |
acb->aio_done_func = NULL;
|
442 |
acb->canceled = false;
|
443 |
acb->coroutine = qemu_coroutine_self(); |
444 |
acb->ret = 0;
|
445 |
acb->nr_pending = 0;
|
446 |
return acb;
|
447 |
} |
448 |
|
449 |
static int connect_to_sdog(const char *addr, const char *port) |
450 |
{ |
451 |
char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
|
452 |
int fd, ret;
|
453 |
struct addrinfo hints, *res, *res0;
|
454 |
|
455 |
if (!addr) {
|
456 |
addr = SD_DEFAULT_ADDR; |
457 |
port = SD_DEFAULT_PORT; |
458 |
} |
459 |
|
460 |
memset(&hints, 0, sizeof(hints)); |
461 |
hints.ai_socktype = SOCK_STREAM; |
462 |
|
463 |
ret = getaddrinfo(addr, port, &hints, &res0); |
464 |
if (ret) {
|
465 |
error_report("unable to get address info %s, %s",
|
466 |
addr, strerror(errno)); |
467 |
return -errno;
|
468 |
} |
469 |
|
470 |
for (res = res0; res; res = res->ai_next) {
|
471 |
ret = getnameinfo(res->ai_addr, res->ai_addrlen, hbuf, sizeof(hbuf),
|
472 |
sbuf, sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
|
473 |
if (ret) {
|
474 |
continue;
|
475 |
} |
476 |
|
477 |
fd = socket(res->ai_family, res->ai_socktype, res->ai_protocol); |
478 |
if (fd < 0) { |
479 |
continue;
|
480 |
} |
481 |
|
482 |
reconnect:
|
483 |
ret = connect(fd, res->ai_addr, res->ai_addrlen); |
484 |
if (ret < 0) { |
485 |
if (errno == EINTR) {
|
486 |
goto reconnect;
|
487 |
} |
488 |
close(fd); |
489 |
break;
|
490 |
} |
491 |
|
492 |
dprintf("connected to %s:%s\n", addr, port);
|
493 |
goto success;
|
494 |
} |
495 |
fd = -errno; |
496 |
error_report("failed connect to %s:%s", addr, port);
|
497 |
success:
|
498 |
freeaddrinfo(res0); |
499 |
return fd;
|
500 |
} |
501 |
|
502 |
static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
503 |
unsigned int *wlen) |
504 |
{ |
505 |
int ret;
|
506 |
|
507 |
ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
|
508 |
if (ret < sizeof(*hdr)) { |
509 |
error_report("failed to send a req, %s", strerror(errno));
|
510 |
return ret;
|
511 |
} |
512 |
|
513 |
ret = qemu_co_send(sockfd, data, *wlen); |
514 |
if (ret < *wlen) {
|
515 |
error_report("failed to send a req, %s", strerror(errno));
|
516 |
} |
517 |
|
518 |
return ret;
|
519 |
} |
520 |
|
521 |
static void restart_co_req(void *opaque) |
522 |
{ |
523 |
Coroutine *co = opaque; |
524 |
|
525 |
qemu_coroutine_enter(co, NULL);
|
526 |
} |
527 |
|
528 |
typedef struct SheepdogReqCo { |
529 |
int sockfd;
|
530 |
SheepdogReq *hdr; |
531 |
void *data;
|
532 |
unsigned int *wlen; |
533 |
unsigned int *rlen; |
534 |
int ret;
|
535 |
bool finished;
|
536 |
} SheepdogReqCo; |
537 |
|
538 |
static coroutine_fn void do_co_req(void *opaque) |
539 |
{ |
540 |
int ret;
|
541 |
Coroutine *co; |
542 |
SheepdogReqCo *srco = opaque; |
543 |
int sockfd = srco->sockfd;
|
544 |
SheepdogReq *hdr = srco->hdr; |
545 |
void *data = srco->data;
|
546 |
unsigned int *wlen = srco->wlen; |
547 |
unsigned int *rlen = srco->rlen; |
548 |
|
549 |
co = qemu_coroutine_self(); |
550 |
qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, NULL, co); |
551 |
|
552 |
socket_set_block(sockfd); |
553 |
ret = send_co_req(sockfd, hdr, data, wlen); |
554 |
if (ret < 0) { |
555 |
goto out;
|
556 |
} |
557 |
|
558 |
qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, NULL, co); |
559 |
|
560 |
ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
|
561 |
if (ret < sizeof(*hdr)) { |
562 |
error_report("failed to get a rsp, %s", strerror(errno));
|
563 |
ret = -errno; |
564 |
goto out;
|
565 |
} |
566 |
|
567 |
if (*rlen > hdr->data_length) {
|
568 |
*rlen = hdr->data_length; |
569 |
} |
570 |
|
571 |
if (*rlen) {
|
572 |
ret = qemu_co_recv(sockfd, data, *rlen); |
573 |
if (ret < *rlen) {
|
574 |
error_report("failed to get the data, %s", strerror(errno));
|
575 |
ret = -errno; |
576 |
goto out;
|
577 |
} |
578 |
} |
579 |
ret = 0;
|
580 |
out:
|
581 |
qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL); |
582 |
socket_set_nonblock(sockfd); |
583 |
|
584 |
srco->ret = ret; |
585 |
srco->finished = true;
|
586 |
} |
587 |
|
588 |
static int do_req(int sockfd, SheepdogReq *hdr, void *data, |
589 |
unsigned int *wlen, unsigned int *rlen) |
590 |
{ |
591 |
Coroutine *co; |
592 |
SheepdogReqCo srco = { |
593 |
.sockfd = sockfd, |
594 |
.hdr = hdr, |
595 |
.data = data, |
596 |
.wlen = wlen, |
597 |
.rlen = rlen, |
598 |
.ret = 0,
|
599 |
.finished = false,
|
600 |
}; |
601 |
|
602 |
if (qemu_in_coroutine()) {
|
603 |
do_co_req(&srco); |
604 |
} else {
|
605 |
co = qemu_coroutine_create(do_co_req); |
606 |
qemu_coroutine_enter(co, &srco); |
607 |
while (!srco.finished) {
|
608 |
qemu_aio_wait(); |
609 |
} |
610 |
} |
611 |
|
612 |
return srco.ret;
|
613 |
} |
614 |
|
615 |
static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
616 |
struct iovec *iov, int niov, bool create, |
617 |
enum AIOCBState aiocb_type);
|
618 |
|
619 |
|
620 |
static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid)
|
621 |
{ |
622 |
AIOReq *aio_req; |
623 |
|
624 |
QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) { |
625 |
if (aio_req->oid == oid) {
|
626 |
return aio_req;
|
627 |
} |
628 |
} |
629 |
|
630 |
return NULL; |
631 |
} |
632 |
|
633 |
/*
|
634 |
* This function searchs pending requests to the object `oid', and
|
635 |
* sends them.
|
636 |
*/
|
637 |
static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid) |
638 |
{ |
639 |
AIOReq *aio_req; |
640 |
SheepdogAIOCB *acb; |
641 |
int ret;
|
642 |
|
643 |
while ((aio_req = find_pending_req(s, oid)) != NULL) { |
644 |
acb = aio_req->aiocb; |
645 |
/* move aio_req from pending list to inflight one */
|
646 |
QLIST_REMOVE(aio_req, aio_siblings); |
647 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
648 |
ret = add_aio_request(s, aio_req, acb->qiov->iov, |
649 |
acb->qiov->niov, false, acb->aiocb_type);
|
650 |
if (ret < 0) { |
651 |
error_report("add_aio_request is failed");
|
652 |
free_aio_req(s, aio_req); |
653 |
if (!acb->nr_pending) {
|
654 |
sd_finish_aiocb(acb); |
655 |
} |
656 |
} |
657 |
} |
658 |
} |
659 |
|
660 |
/*
|
661 |
* Receive responses of the I/O requests.
|
662 |
*
|
663 |
* This function is registered as a fd handler, and called from the
|
664 |
* main loop when s->fd is ready for reading responses.
|
665 |
*/
|
666 |
static void coroutine_fn aio_read_response(void *opaque) |
667 |
{ |
668 |
SheepdogObjRsp rsp; |
669 |
BDRVSheepdogState *s = opaque; |
670 |
int fd = s->fd;
|
671 |
int ret;
|
672 |
AIOReq *aio_req = NULL;
|
673 |
SheepdogAIOCB *acb; |
674 |
unsigned long idx; |
675 |
|
676 |
if (QLIST_EMPTY(&s->inflight_aio_head)) {
|
677 |
goto out;
|
678 |
} |
679 |
|
680 |
/* read a header */
|
681 |
ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
|
682 |
if (ret < 0) { |
683 |
error_report("failed to get the header, %s", strerror(errno));
|
684 |
goto out;
|
685 |
} |
686 |
|
687 |
/* find the right aio_req from the inflight aio list */
|
688 |
QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { |
689 |
if (aio_req->id == rsp.id) {
|
690 |
break;
|
691 |
} |
692 |
} |
693 |
if (!aio_req) {
|
694 |
error_report("cannot find aio_req %x", rsp.id);
|
695 |
goto out;
|
696 |
} |
697 |
|
698 |
acb = aio_req->aiocb; |
699 |
|
700 |
switch (acb->aiocb_type) {
|
701 |
case AIOCB_WRITE_UDATA:
|
702 |
/* this coroutine context is no longer suitable for co_recv
|
703 |
* because we may send data to update vdi objects */
|
704 |
s->co_recv = NULL;
|
705 |
if (!is_data_obj(aio_req->oid)) {
|
706 |
break;
|
707 |
} |
708 |
idx = data_oid_to_idx(aio_req->oid); |
709 |
|
710 |
if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) {
|
711 |
/*
|
712 |
* If the object is newly created one, we need to update
|
713 |
* the vdi object (metadata object). min_dirty_data_idx
|
714 |
* and max_dirty_data_idx are changed to include updated
|
715 |
* index between them.
|
716 |
*/
|
717 |
s->inode.data_vdi_id[idx] = s->inode.vdi_id; |
718 |
s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); |
719 |
s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); |
720 |
|
721 |
/*
|
722 |
* Some requests may be blocked because simultaneous
|
723 |
* create requests are not allowed, so we search the
|
724 |
* pending requests here.
|
725 |
*/
|
726 |
send_pending_req(s, vid_to_data_oid(s->inode.vdi_id, idx)); |
727 |
} |
728 |
break;
|
729 |
case AIOCB_READ_UDATA:
|
730 |
ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
731 |
aio_req->iov_offset, rsp.data_length); |
732 |
if (ret < 0) { |
733 |
error_report("failed to get the data, %s", strerror(errno));
|
734 |
goto out;
|
735 |
} |
736 |
break;
|
737 |
} |
738 |
|
739 |
if (rsp.result != SD_RES_SUCCESS) {
|
740 |
acb->ret = -EIO; |
741 |
error_report("%s", sd_strerror(rsp.result));
|
742 |
} |
743 |
|
744 |
free_aio_req(s, aio_req); |
745 |
if (!acb->nr_pending) {
|
746 |
/*
|
747 |
* We've finished all requests which belong to the AIOCB, so
|
748 |
* we can switch back to sd_co_readv/writev now.
|
749 |
*/
|
750 |
acb->aio_done_func(acb); |
751 |
} |
752 |
out:
|
753 |
s->co_recv = NULL;
|
754 |
} |
755 |
|
756 |
static void co_read_response(void *opaque) |
757 |
{ |
758 |
BDRVSheepdogState *s = opaque; |
759 |
|
760 |
if (!s->co_recv) {
|
761 |
s->co_recv = qemu_coroutine_create(aio_read_response); |
762 |
} |
763 |
|
764 |
qemu_coroutine_enter(s->co_recv, opaque); |
765 |
} |
766 |
|
767 |
static void co_write_request(void *opaque) |
768 |
{ |
769 |
BDRVSheepdogState *s = opaque; |
770 |
|
771 |
qemu_coroutine_enter(s->co_send, NULL);
|
772 |
} |
773 |
|
774 |
static int aio_flush_request(void *opaque) |
775 |
{ |
776 |
BDRVSheepdogState *s = opaque; |
777 |
|
778 |
return !QLIST_EMPTY(&s->inflight_aio_head) ||
|
779 |
!QLIST_EMPTY(&s->pending_aio_head); |
780 |
} |
781 |
|
782 |
static int set_nodelay(int fd) |
783 |
{ |
784 |
int ret, opt;
|
785 |
|
786 |
opt = 1;
|
787 |
ret = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *)&opt, sizeof(opt)); |
788 |
return ret;
|
789 |
} |
790 |
|
791 |
/*
|
792 |
* Return a socket discriptor to read/write objects.
|
793 |
*
|
794 |
* We cannot use this discriptor for other operations because
|
795 |
* the block driver may be on waiting response from the server.
|
796 |
*/
|
797 |
static int get_sheep_fd(BDRVSheepdogState *s) |
798 |
{ |
799 |
int ret, fd;
|
800 |
|
801 |
fd = connect_to_sdog(s->addr, s->port); |
802 |
if (fd < 0) { |
803 |
error_report("%s", strerror(errno));
|
804 |
return fd;
|
805 |
} |
806 |
|
807 |
socket_set_nonblock(fd); |
808 |
|
809 |
ret = set_nodelay(fd); |
810 |
if (ret) {
|
811 |
error_report("%s", strerror(errno));
|
812 |
closesocket(fd); |
813 |
return -errno;
|
814 |
} |
815 |
|
816 |
qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s);
|
817 |
return fd;
|
818 |
} |
819 |
|
820 |
/*
|
821 |
* Parse a filename
|
822 |
*
|
823 |
* filename must be one of the following formats:
|
824 |
* 1. [vdiname]
|
825 |
* 2. [vdiname]:[snapid]
|
826 |
* 3. [vdiname]:[tag]
|
827 |
* 4. [hostname]:[port]:[vdiname]
|
828 |
* 5. [hostname]:[port]:[vdiname]:[snapid]
|
829 |
* 6. [hostname]:[port]:[vdiname]:[tag]
|
830 |
*
|
831 |
* You can boot from the snapshot images by specifying `snapid` or
|
832 |
* `tag'.
|
833 |
*
|
834 |
* You can run VMs outside the Sheepdog cluster by specifying
|
835 |
* `hostname' and `port' (experimental).
|
836 |
*/
|
837 |
static int parse_vdiname(BDRVSheepdogState *s, const char *filename, |
838 |
char *vdi, uint32_t *snapid, char *tag) |
839 |
{ |
840 |
char *p, *q;
|
841 |
int nr_sep;
|
842 |
|
843 |
p = q = g_strdup(filename); |
844 |
|
845 |
/* count the number of separators */
|
846 |
nr_sep = 0;
|
847 |
while (*p) {
|
848 |
if (*p == ':') { |
849 |
nr_sep++; |
850 |
} |
851 |
p++; |
852 |
} |
853 |
p = q; |
854 |
|
855 |
/* use the first two tokens as hostname and port number. */
|
856 |
if (nr_sep >= 2) { |
857 |
s->addr = p; |
858 |
p = strchr(p, ':');
|
859 |
*p++ = '\0';
|
860 |
|
861 |
s->port = p; |
862 |
p = strchr(p, ':');
|
863 |
*p++ = '\0';
|
864 |
} else {
|
865 |
s->addr = NULL;
|
866 |
s->port = 0;
|
867 |
} |
868 |
|
869 |
pstrcpy(vdi, SD_MAX_VDI_LEN, p); |
870 |
|
871 |
p = strchr(vdi, ':');
|
872 |
if (p) {
|
873 |
*p++ = '\0';
|
874 |
*snapid = strtoul(p, NULL, 10); |
875 |
if (*snapid == 0) { |
876 |
pstrcpy(tag, SD_MAX_VDI_TAG_LEN, p); |
877 |
} |
878 |
} else {
|
879 |
*snapid = CURRENT_VDI_ID; /* search current vdi */
|
880 |
} |
881 |
|
882 |
if (s->addr == NULL) { |
883 |
g_free(q); |
884 |
} |
885 |
|
886 |
return 0; |
887 |
} |
888 |
|
889 |
static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid, |
890 |
char *tag, uint32_t *vid, int for_snapshot) |
891 |
{ |
892 |
int ret, fd;
|
893 |
SheepdogVdiReq hdr; |
894 |
SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; |
895 |
unsigned int wlen, rlen = 0; |
896 |
char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
|
897 |
|
898 |
fd = connect_to_sdog(s->addr, s->port); |
899 |
if (fd < 0) { |
900 |
return fd;
|
901 |
} |
902 |
|
903 |
/* This pair of strncpy calls ensures that the buffer is zero-filled,
|
904 |
* which is desirable since we'll soon be sending those bytes, and
|
905 |
* don't want the send_req to read uninitialized data.
|
906 |
*/
|
907 |
strncpy(buf, filename, SD_MAX_VDI_LEN); |
908 |
strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); |
909 |
|
910 |
memset(&hdr, 0, sizeof(hdr)); |
911 |
if (for_snapshot) {
|
912 |
hdr.opcode = SD_OP_GET_VDI_INFO; |
913 |
} else {
|
914 |
hdr.opcode = SD_OP_LOCK_VDI; |
915 |
} |
916 |
wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; |
917 |
hdr.proto_ver = SD_PROTO_VER; |
918 |
hdr.data_length = wlen; |
919 |
hdr.snapid = snapid; |
920 |
hdr.flags = SD_FLAG_CMD_WRITE; |
921 |
|
922 |
ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
923 |
if (ret) {
|
924 |
goto out;
|
925 |
} |
926 |
|
927 |
if (rsp->result != SD_RES_SUCCESS) {
|
928 |
error_report("cannot get vdi info, %s, %s %d %s",
|
929 |
sd_strerror(rsp->result), filename, snapid, tag); |
930 |
if (rsp->result == SD_RES_NO_VDI) {
|
931 |
ret = -ENOENT; |
932 |
} else {
|
933 |
ret = -EIO; |
934 |
} |
935 |
goto out;
|
936 |
} |
937 |
*vid = rsp->vdi_id; |
938 |
|
939 |
ret = 0;
|
940 |
out:
|
941 |
closesocket(fd); |
942 |
return ret;
|
943 |
} |
944 |
|
945 |
static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
946 |
struct iovec *iov, int niov, bool create, |
947 |
enum AIOCBState aiocb_type)
|
948 |
{ |
949 |
int nr_copies = s->inode.nr_copies;
|
950 |
SheepdogObjReq hdr; |
951 |
unsigned int wlen; |
952 |
int ret;
|
953 |
uint64_t oid = aio_req->oid; |
954 |
unsigned int datalen = aio_req->data_len; |
955 |
uint64_t offset = aio_req->offset; |
956 |
uint8_t flags = aio_req->flags; |
957 |
uint64_t old_oid = aio_req->base_oid; |
958 |
|
959 |
if (!nr_copies) {
|
960 |
error_report("bug");
|
961 |
} |
962 |
|
963 |
memset(&hdr, 0, sizeof(hdr)); |
964 |
|
965 |
if (aiocb_type == AIOCB_READ_UDATA) {
|
966 |
wlen = 0;
|
967 |
hdr.opcode = SD_OP_READ_OBJ; |
968 |
hdr.flags = flags; |
969 |
} else if (create) { |
970 |
wlen = datalen; |
971 |
hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; |
972 |
hdr.flags = SD_FLAG_CMD_WRITE | flags; |
973 |
} else {
|
974 |
wlen = datalen; |
975 |
hdr.opcode = SD_OP_WRITE_OBJ; |
976 |
hdr.flags = SD_FLAG_CMD_WRITE | flags; |
977 |
} |
978 |
|
979 |
if (s->cache_enabled) {
|
980 |
hdr.flags |= SD_FLAG_CMD_CACHE; |
981 |
} |
982 |
|
983 |
hdr.oid = oid; |
984 |
hdr.cow_oid = old_oid; |
985 |
hdr.copies = s->inode.nr_copies; |
986 |
|
987 |
hdr.data_length = datalen; |
988 |
hdr.offset = offset; |
989 |
|
990 |
hdr.id = aio_req->id; |
991 |
|
992 |
qemu_co_mutex_lock(&s->lock); |
993 |
s->co_send = qemu_coroutine_self(); |
994 |
qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, |
995 |
aio_flush_request, s); |
996 |
socket_set_cork(s->fd, 1);
|
997 |
|
998 |
/* send a header */
|
999 |
ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
|
1000 |
if (ret < 0) { |
1001 |
qemu_co_mutex_unlock(&s->lock); |
1002 |
error_report("failed to send a req, %s", strerror(errno));
|
1003 |
return -errno;
|
1004 |
} |
1005 |
|
1006 |
if (wlen) {
|
1007 |
ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
1008 |
if (ret < 0) { |
1009 |
qemu_co_mutex_unlock(&s->lock); |
1010 |
error_report("failed to send a data, %s", strerror(errno));
|
1011 |
return -errno;
|
1012 |
} |
1013 |
} |
1014 |
|
1015 |
socket_set_cork(s->fd, 0);
|
1016 |
qemu_aio_set_fd_handler(s->fd, co_read_response, NULL,
|
1017 |
aio_flush_request, s); |
1018 |
qemu_co_mutex_unlock(&s->lock); |
1019 |
|
1020 |
return 0; |
1021 |
} |
1022 |
|
1023 |
static int read_write_object(int fd, char *buf, uint64_t oid, int copies, |
1024 |
unsigned int datalen, uint64_t offset, |
1025 |
bool write, bool create, bool cache) |
1026 |
{ |
1027 |
SheepdogObjReq hdr; |
1028 |
SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; |
1029 |
unsigned int wlen, rlen; |
1030 |
int ret;
|
1031 |
|
1032 |
memset(&hdr, 0, sizeof(hdr)); |
1033 |
|
1034 |
if (write) {
|
1035 |
wlen = datalen; |
1036 |
rlen = 0;
|
1037 |
hdr.flags = SD_FLAG_CMD_WRITE; |
1038 |
if (create) {
|
1039 |
hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; |
1040 |
} else {
|
1041 |
hdr.opcode = SD_OP_WRITE_OBJ; |
1042 |
} |
1043 |
} else {
|
1044 |
wlen = 0;
|
1045 |
rlen = datalen; |
1046 |
hdr.opcode = SD_OP_READ_OBJ; |
1047 |
} |
1048 |
|
1049 |
if (cache) {
|
1050 |
hdr.flags |= SD_FLAG_CMD_CACHE; |
1051 |
} |
1052 |
|
1053 |
hdr.oid = oid; |
1054 |
hdr.data_length = datalen; |
1055 |
hdr.offset = offset; |
1056 |
hdr.copies = copies; |
1057 |
|
1058 |
ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
1059 |
if (ret) {
|
1060 |
error_report("failed to send a request to the sheep");
|
1061 |
return ret;
|
1062 |
} |
1063 |
|
1064 |
switch (rsp->result) {
|
1065 |
case SD_RES_SUCCESS:
|
1066 |
return 0; |
1067 |
default:
|
1068 |
error_report("%s", sd_strerror(rsp->result));
|
1069 |
return -EIO;
|
1070 |
} |
1071 |
} |
1072 |
|
1073 |
static int read_object(int fd, char *buf, uint64_t oid, int copies, |
1074 |
unsigned int datalen, uint64_t offset, bool cache) |
1075 |
{ |
1076 |
return read_write_object(fd, buf, oid, copies, datalen, offset, false, |
1077 |
false, cache);
|
1078 |
} |
1079 |
|
1080 |
static int write_object(int fd, char *buf, uint64_t oid, int copies, |
1081 |
unsigned int datalen, uint64_t offset, bool create, |
1082 |
bool cache)
|
1083 |
{ |
1084 |
return read_write_object(fd, buf, oid, copies, datalen, offset, true, |
1085 |
create, cache); |
1086 |
} |
1087 |
|
1088 |
static int sd_open(BlockDriverState *bs, const char *filename, int flags) |
1089 |
{ |
1090 |
int ret, fd;
|
1091 |
uint32_t vid = 0;
|
1092 |
BDRVSheepdogState *s = bs->opaque; |
1093 |
char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
|
1094 |
uint32_t snapid; |
1095 |
char *buf = NULL; |
1096 |
|
1097 |
strstart(filename, "sheepdog:", (const char **)&filename); |
1098 |
|
1099 |
QLIST_INIT(&s->inflight_aio_head); |
1100 |
QLIST_INIT(&s->pending_aio_head); |
1101 |
s->fd = -1;
|
1102 |
|
1103 |
memset(vdi, 0, sizeof(vdi)); |
1104 |
memset(tag, 0, sizeof(tag)); |
1105 |
if (parse_vdiname(s, filename, vdi, &snapid, tag) < 0) { |
1106 |
ret = -EINVAL; |
1107 |
goto out;
|
1108 |
} |
1109 |
s->fd = get_sheep_fd(s); |
1110 |
if (s->fd < 0) { |
1111 |
ret = s->fd; |
1112 |
goto out;
|
1113 |
} |
1114 |
|
1115 |
ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0);
|
1116 |
if (ret) {
|
1117 |
goto out;
|
1118 |
} |
1119 |
|
1120 |
s->cache_enabled = true;
|
1121 |
s->flush_fd = connect_to_sdog(s->addr, s->port); |
1122 |
if (s->flush_fd < 0) { |
1123 |
error_report("failed to connect");
|
1124 |
ret = s->flush_fd; |
1125 |
goto out;
|
1126 |
} |
1127 |
|
1128 |
if (snapid || tag[0] != '\0') { |
1129 |
dprintf("%" PRIx32 " snapshot inode was open.\n", vid); |
1130 |
s->is_snapshot = true;
|
1131 |
} |
1132 |
|
1133 |
fd = connect_to_sdog(s->addr, s->port); |
1134 |
if (fd < 0) { |
1135 |
error_report("failed to connect");
|
1136 |
ret = fd; |
1137 |
goto out;
|
1138 |
} |
1139 |
|
1140 |
buf = g_malloc(SD_INODE_SIZE); |
1141 |
ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, |
1142 |
s->cache_enabled); |
1143 |
|
1144 |
closesocket(fd); |
1145 |
|
1146 |
if (ret) {
|
1147 |
goto out;
|
1148 |
} |
1149 |
|
1150 |
memcpy(&s->inode, buf, sizeof(s->inode));
|
1151 |
s->min_dirty_data_idx = UINT32_MAX; |
1152 |
s->max_dirty_data_idx = 0;
|
1153 |
|
1154 |
bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE; |
1155 |
pstrcpy(s->name, sizeof(s->name), vdi);
|
1156 |
qemu_co_mutex_init(&s->lock); |
1157 |
g_free(buf); |
1158 |
return 0; |
1159 |
out:
|
1160 |
qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
1161 |
if (s->fd >= 0) { |
1162 |
closesocket(s->fd); |
1163 |
} |
1164 |
g_free(buf); |
1165 |
return ret;
|
1166 |
} |
1167 |
|
1168 |
static int do_sd_create(char *filename, int64_t vdi_size, |
1169 |
uint32_t base_vid, uint32_t *vdi_id, int snapshot,
|
1170 |
const char *addr, const char *port) |
1171 |
{ |
1172 |
SheepdogVdiReq hdr; |
1173 |
SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; |
1174 |
int fd, ret;
|
1175 |
unsigned int wlen, rlen = 0; |
1176 |
char buf[SD_MAX_VDI_LEN];
|
1177 |
|
1178 |
fd = connect_to_sdog(addr, port); |
1179 |
if (fd < 0) { |
1180 |
return fd;
|
1181 |
} |
1182 |
|
1183 |
/* FIXME: would it be better to fail (e.g., return -EIO) when filename
|
1184 |
* does not fit in buf? For now, just truncate and avoid buffer overrun.
|
1185 |
*/
|
1186 |
memset(buf, 0, sizeof(buf)); |
1187 |
pstrcpy(buf, sizeof(buf), filename);
|
1188 |
|
1189 |
memset(&hdr, 0, sizeof(hdr)); |
1190 |
hdr.opcode = SD_OP_NEW_VDI; |
1191 |
hdr.base_vdi_id = base_vid; |
1192 |
|
1193 |
wlen = SD_MAX_VDI_LEN; |
1194 |
|
1195 |
hdr.flags = SD_FLAG_CMD_WRITE; |
1196 |
hdr.snapid = snapshot; |
1197 |
|
1198 |
hdr.data_length = wlen; |
1199 |
hdr.vdi_size = vdi_size; |
1200 |
|
1201 |
ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
1202 |
|
1203 |
closesocket(fd); |
1204 |
|
1205 |
if (ret) {
|
1206 |
return ret;
|
1207 |
} |
1208 |
|
1209 |
if (rsp->result != SD_RES_SUCCESS) {
|
1210 |
error_report("%s, %s", sd_strerror(rsp->result), filename);
|
1211 |
return -EIO;
|
1212 |
} |
1213 |
|
1214 |
if (vdi_id) {
|
1215 |
*vdi_id = rsp->vdi_id; |
1216 |
} |
1217 |
|
1218 |
return 0; |
1219 |
} |
1220 |
|
1221 |
static int sd_prealloc(const char *filename) |
1222 |
{ |
1223 |
BlockDriverState *bs = NULL;
|
1224 |
uint32_t idx, max_idx; |
1225 |
int64_t vdi_size; |
1226 |
void *buf = g_malloc0(SD_DATA_OBJ_SIZE);
|
1227 |
int ret;
|
1228 |
|
1229 |
ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); |
1230 |
if (ret < 0) { |
1231 |
goto out;
|
1232 |
} |
1233 |
|
1234 |
vdi_size = bdrv_getlength(bs); |
1235 |
if (vdi_size < 0) { |
1236 |
ret = vdi_size; |
1237 |
goto out;
|
1238 |
} |
1239 |
max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); |
1240 |
|
1241 |
for (idx = 0; idx < max_idx; idx++) { |
1242 |
/*
|
1243 |
* The created image can be a cloned image, so we need to read
|
1244 |
* a data from the source image.
|
1245 |
*/
|
1246 |
ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); |
1247 |
if (ret < 0) { |
1248 |
goto out;
|
1249 |
} |
1250 |
ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); |
1251 |
if (ret < 0) { |
1252 |
goto out;
|
1253 |
} |
1254 |
} |
1255 |
out:
|
1256 |
if (bs) {
|
1257 |
bdrv_delete(bs); |
1258 |
} |
1259 |
g_free(buf); |
1260 |
|
1261 |
return ret;
|
1262 |
} |
1263 |
|
1264 |
static int sd_create(const char *filename, QEMUOptionParameter *options) |
1265 |
{ |
1266 |
int ret = 0; |
1267 |
uint32_t vid = 0, base_vid = 0; |
1268 |
int64_t vdi_size = 0;
|
1269 |
char *backing_file = NULL; |
1270 |
BDRVSheepdogState *s; |
1271 |
char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
|
1272 |
uint32_t snapid; |
1273 |
bool prealloc = false; |
1274 |
const char *vdiname; |
1275 |
|
1276 |
s = g_malloc0(sizeof(BDRVSheepdogState));
|
1277 |
|
1278 |
strstart(filename, "sheepdog:", &vdiname);
|
1279 |
|
1280 |
memset(vdi, 0, sizeof(vdi)); |
1281 |
memset(tag, 0, sizeof(tag)); |
1282 |
if (parse_vdiname(s, vdiname, vdi, &snapid, tag) < 0) { |
1283 |
error_report("invalid filename");
|
1284 |
ret = -EINVAL; |
1285 |
goto out;
|
1286 |
} |
1287 |
|
1288 |
while (options && options->name) {
|
1289 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
1290 |
vdi_size = options->value.n; |
1291 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
1292 |
backing_file = options->value.s; |
1293 |
} else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1294 |
if (!options->value.s || !strcmp(options->value.s, "off")) { |
1295 |
prealloc = false;
|
1296 |
} else if (!strcmp(options->value.s, "full")) { |
1297 |
prealloc = true;
|
1298 |
} else {
|
1299 |
error_report("Invalid preallocation mode: '%s'",
|
1300 |
options->value.s); |
1301 |
ret = -EINVAL; |
1302 |
goto out;
|
1303 |
} |
1304 |
} |
1305 |
options++; |
1306 |
} |
1307 |
|
1308 |
if (vdi_size > SD_MAX_VDI_SIZE) {
|
1309 |
error_report("too big image size");
|
1310 |
ret = -EINVAL; |
1311 |
goto out;
|
1312 |
} |
1313 |
|
1314 |
if (backing_file) {
|
1315 |
BlockDriverState *bs; |
1316 |
BDRVSheepdogState *s; |
1317 |
BlockDriver *drv; |
1318 |
|
1319 |
/* Currently, only Sheepdog backing image is supported. */
|
1320 |
drv = bdrv_find_protocol(backing_file); |
1321 |
if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
1322 |
error_report("backing_file must be a sheepdog image");
|
1323 |
ret = -EINVAL; |
1324 |
goto out;
|
1325 |
} |
1326 |
|
1327 |
ret = bdrv_file_open(&bs, backing_file, 0);
|
1328 |
if (ret < 0) { |
1329 |
goto out;
|
1330 |
} |
1331 |
|
1332 |
s = bs->opaque; |
1333 |
|
1334 |
if (!is_snapshot(&s->inode)) {
|
1335 |
error_report("cannot clone from a non snapshot vdi");
|
1336 |
bdrv_delete(bs); |
1337 |
ret = -EINVAL; |
1338 |
goto out;
|
1339 |
} |
1340 |
|
1341 |
base_vid = s->inode.vdi_id; |
1342 |
bdrv_delete(bs); |
1343 |
} |
1344 |
|
1345 |
ret = do_sd_create(vdi, vdi_size, base_vid, &vid, 0, s->addr, s->port);
|
1346 |
if (!prealloc || ret) {
|
1347 |
goto out;
|
1348 |
} |
1349 |
|
1350 |
ret = sd_prealloc(filename); |
1351 |
out:
|
1352 |
g_free(s); |
1353 |
return ret;
|
1354 |
} |
1355 |
|
1356 |
static void sd_close(BlockDriverState *bs) |
1357 |
{ |
1358 |
BDRVSheepdogState *s = bs->opaque; |
1359 |
SheepdogVdiReq hdr; |
1360 |
SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; |
1361 |
unsigned int wlen, rlen = 0; |
1362 |
int fd, ret;
|
1363 |
|
1364 |
dprintf("%s\n", s->name);
|
1365 |
|
1366 |
fd = connect_to_sdog(s->addr, s->port); |
1367 |
if (fd < 0) { |
1368 |
return;
|
1369 |
} |
1370 |
|
1371 |
memset(&hdr, 0, sizeof(hdr)); |
1372 |
|
1373 |
hdr.opcode = SD_OP_RELEASE_VDI; |
1374 |
wlen = strlen(s->name) + 1;
|
1375 |
hdr.data_length = wlen; |
1376 |
hdr.flags = SD_FLAG_CMD_WRITE; |
1377 |
|
1378 |
ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); |
1379 |
|
1380 |
closesocket(fd); |
1381 |
|
1382 |
if (!ret && rsp->result != SD_RES_SUCCESS &&
|
1383 |
rsp->result != SD_RES_VDI_NOT_LOCKED) { |
1384 |
error_report("%s, %s", sd_strerror(rsp->result), s->name);
|
1385 |
} |
1386 |
|
1387 |
qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
1388 |
closesocket(s->fd); |
1389 |
if (s->cache_enabled) {
|
1390 |
closesocket(s->flush_fd); |
1391 |
} |
1392 |
g_free(s->addr); |
1393 |
} |
1394 |
|
1395 |
static int64_t sd_getlength(BlockDriverState *bs)
|
1396 |
{ |
1397 |
BDRVSheepdogState *s = bs->opaque; |
1398 |
|
1399 |
return s->inode.vdi_size;
|
1400 |
} |
1401 |
|
1402 |
static int sd_truncate(BlockDriverState *bs, int64_t offset) |
1403 |
{ |
1404 |
BDRVSheepdogState *s = bs->opaque; |
1405 |
int ret, fd;
|
1406 |
unsigned int datalen; |
1407 |
|
1408 |
if (offset < s->inode.vdi_size) {
|
1409 |
error_report("shrinking is not supported");
|
1410 |
return -EINVAL;
|
1411 |
} else if (offset > SD_MAX_VDI_SIZE) { |
1412 |
error_report("too big image size");
|
1413 |
return -EINVAL;
|
1414 |
} |
1415 |
|
1416 |
fd = connect_to_sdog(s->addr, s->port); |
1417 |
if (fd < 0) { |
1418 |
return fd;
|
1419 |
} |
1420 |
|
1421 |
/* we don't need to update entire object */
|
1422 |
datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
|
1423 |
s->inode.vdi_size = offset; |
1424 |
ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
|
1425 |
s->inode.nr_copies, datalen, 0, false, s->cache_enabled); |
1426 |
close(fd); |
1427 |
|
1428 |
if (ret < 0) { |
1429 |
error_report("failed to update an inode.");
|
1430 |
} |
1431 |
|
1432 |
return ret;
|
1433 |
} |
1434 |
|
1435 |
/*
|
1436 |
* This function is called after writing data objects. If we need to
|
1437 |
* update metadata, this sends a write request to the vdi object.
|
1438 |
* Otherwise, this switches back to sd_co_readv/writev.
|
1439 |
*/
|
1440 |
static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
1441 |
{ |
1442 |
int ret;
|
1443 |
BDRVSheepdogState *s = acb->common.bs->opaque; |
1444 |
struct iovec iov;
|
1445 |
AIOReq *aio_req; |
1446 |
uint32_t offset, data_len, mn, mx; |
1447 |
|
1448 |
mn = s->min_dirty_data_idx; |
1449 |
mx = s->max_dirty_data_idx; |
1450 |
if (mn <= mx) {
|
1451 |
/* we need to update the vdi object. */
|
1452 |
offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + |
1453 |
mn * sizeof(s->inode.data_vdi_id[0]); |
1454 |
data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); |
1455 |
|
1456 |
s->min_dirty_data_idx = UINT32_MAX; |
1457 |
s->max_dirty_data_idx = 0;
|
1458 |
|
1459 |
iov.iov_base = &s->inode; |
1460 |
iov.iov_len = sizeof(s->inode);
|
1461 |
aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
1462 |
data_len, offset, 0, 0, offset); |
1463 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1464 |
ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); |
1465 |
if (ret) {
|
1466 |
free_aio_req(s, aio_req); |
1467 |
acb->ret = -EIO; |
1468 |
goto out;
|
1469 |
} |
1470 |
|
1471 |
acb->aio_done_func = sd_finish_aiocb; |
1472 |
acb->aiocb_type = AIOCB_WRITE_UDATA; |
1473 |
return;
|
1474 |
} |
1475 |
out:
|
1476 |
sd_finish_aiocb(acb); |
1477 |
} |
1478 |
|
1479 |
/*
|
1480 |
* Create a writable VDI from a snapshot
|
1481 |
*/
|
1482 |
static int sd_create_branch(BDRVSheepdogState *s) |
1483 |
{ |
1484 |
int ret, fd;
|
1485 |
uint32_t vid; |
1486 |
char *buf;
|
1487 |
|
1488 |
dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
1489 |
|
1490 |
buf = g_malloc(SD_INODE_SIZE); |
1491 |
|
1492 |
ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1,
|
1493 |
s->addr, s->port); |
1494 |
if (ret) {
|
1495 |
goto out;
|
1496 |
} |
1497 |
|
1498 |
dprintf("%" PRIx32 " is created.\n", vid); |
1499 |
|
1500 |
fd = connect_to_sdog(s->addr, s->port); |
1501 |
if (fd < 0) { |
1502 |
error_report("failed to connect");
|
1503 |
ret = fd; |
1504 |
goto out;
|
1505 |
} |
1506 |
|
1507 |
ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
1508 |
SD_INODE_SIZE, 0, s->cache_enabled);
|
1509 |
|
1510 |
closesocket(fd); |
1511 |
|
1512 |
if (ret < 0) { |
1513 |
goto out;
|
1514 |
} |
1515 |
|
1516 |
memcpy(&s->inode, buf, sizeof(s->inode));
|
1517 |
|
1518 |
s->is_snapshot = false;
|
1519 |
ret = 0;
|
1520 |
dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
1521 |
|
1522 |
out:
|
1523 |
g_free(buf); |
1524 |
|
1525 |
return ret;
|
1526 |
} |
1527 |
|
1528 |
/*
|
1529 |
* Send I/O requests to the server.
|
1530 |
*
|
1531 |
* This function sends requests to the server, links the requests to
|
1532 |
* the inflight_list in BDRVSheepdogState, and exits without
|
1533 |
* waiting the response. The responses are received in the
|
1534 |
* `aio_read_response' function which is called from the main loop as
|
1535 |
* a fd handler.
|
1536 |
*
|
1537 |
* Returns 1 when we need to wait a response, 0 when there is no sent
|
1538 |
* request and -errno in error cases.
|
1539 |
*/
|
1540 |
static int coroutine_fn sd_co_rw_vector(void *p) |
1541 |
{ |
1542 |
SheepdogAIOCB *acb = p; |
1543 |
int ret = 0; |
1544 |
unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE; |
1545 |
unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE; |
1546 |
uint64_t oid; |
1547 |
uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE; |
1548 |
BDRVSheepdogState *s = acb->common.bs->opaque; |
1549 |
SheepdogInode *inode = &s->inode; |
1550 |
AIOReq *aio_req; |
1551 |
|
1552 |
if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
|
1553 |
/*
|
1554 |
* In the case we open the snapshot VDI, Sheepdog creates the
|
1555 |
* writable VDI when we do a write operation first.
|
1556 |
*/
|
1557 |
ret = sd_create_branch(s); |
1558 |
if (ret) {
|
1559 |
acb->ret = -EIO; |
1560 |
goto out;
|
1561 |
} |
1562 |
} |
1563 |
|
1564 |
/*
|
1565 |
* Make sure we don't free the aiocb before we are done with all requests.
|
1566 |
* This additional reference is dropped at the end of this function.
|
1567 |
*/
|
1568 |
acb->nr_pending++; |
1569 |
|
1570 |
while (done != total) {
|
1571 |
uint8_t flags = 0;
|
1572 |
uint64_t old_oid = 0;
|
1573 |
bool create = false; |
1574 |
|
1575 |
oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); |
1576 |
|
1577 |
len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); |
1578 |
|
1579 |
switch (acb->aiocb_type) {
|
1580 |
case AIOCB_READ_UDATA:
|
1581 |
if (!inode->data_vdi_id[idx]) {
|
1582 |
qemu_iovec_memset(acb->qiov, done, 0, len);
|
1583 |
goto done;
|
1584 |
} |
1585 |
break;
|
1586 |
case AIOCB_WRITE_UDATA:
|
1587 |
if (!inode->data_vdi_id[idx]) {
|
1588 |
create = true;
|
1589 |
} else if (!is_data_obj_writable(inode, idx)) { |
1590 |
/* Copy-On-Write */
|
1591 |
create = true;
|
1592 |
old_oid = oid; |
1593 |
flags = SD_FLAG_CMD_COW; |
1594 |
} |
1595 |
break;
|
1596 |
default:
|
1597 |
break;
|
1598 |
} |
1599 |
|
1600 |
if (create) {
|
1601 |
dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1602 |
inode->vdi_id, oid, |
1603 |
vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
1604 |
oid = vid_to_data_oid(inode->vdi_id, idx); |
1605 |
dprintf("new oid %" PRIx64 "\n", oid); |
1606 |
} |
1607 |
|
1608 |
aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); |
1609 |
|
1610 |
if (create) {
|
1611 |
AIOReq *areq; |
1612 |
QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { |
1613 |
if (areq->oid == oid) {
|
1614 |
/*
|
1615 |
* Sheepdog cannot handle simultaneous create
|
1616 |
* requests to the same object. So we cannot send
|
1617 |
* the request until the previous request
|
1618 |
* finishes.
|
1619 |
*/
|
1620 |
aio_req->flags = 0;
|
1621 |
aio_req->base_oid = 0;
|
1622 |
QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, |
1623 |
aio_siblings); |
1624 |
goto done;
|
1625 |
} |
1626 |
} |
1627 |
} |
1628 |
|
1629 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1630 |
ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
1631 |
create, acb->aiocb_type); |
1632 |
if (ret < 0) { |
1633 |
error_report("add_aio_request is failed");
|
1634 |
free_aio_req(s, aio_req); |
1635 |
acb->ret = -EIO; |
1636 |
goto out;
|
1637 |
} |
1638 |
done:
|
1639 |
offset = 0;
|
1640 |
idx++; |
1641 |
done += len; |
1642 |
} |
1643 |
out:
|
1644 |
if (!--acb->nr_pending) {
|
1645 |
return acb->ret;
|
1646 |
} |
1647 |
return 1; |
1648 |
} |
1649 |
|
1650 |
static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
1651 |
int nb_sectors, QEMUIOVector *qiov)
|
1652 |
{ |
1653 |
SheepdogAIOCB *acb; |
1654 |
int ret;
|
1655 |
|
1656 |
if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
|
1657 |
ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE); |
1658 |
if (ret < 0) { |
1659 |
return ret;
|
1660 |
} |
1661 |
bs->total_sectors = sector_num + nb_sectors; |
1662 |
} |
1663 |
|
1664 |
acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL); |
1665 |
acb->aio_done_func = sd_write_done; |
1666 |
acb->aiocb_type = AIOCB_WRITE_UDATA; |
1667 |
|
1668 |
ret = sd_co_rw_vector(acb); |
1669 |
if (ret <= 0) { |
1670 |
qemu_aio_release(acb); |
1671 |
return ret;
|
1672 |
} |
1673 |
|
1674 |
qemu_coroutine_yield(); |
1675 |
|
1676 |
return acb->ret;
|
1677 |
} |
1678 |
|
1679 |
static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
1680 |
int nb_sectors, QEMUIOVector *qiov)
|
1681 |
{ |
1682 |
SheepdogAIOCB *acb; |
1683 |
int ret;
|
1684 |
|
1685 |
acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors, NULL, NULL); |
1686 |
acb->aiocb_type = AIOCB_READ_UDATA; |
1687 |
acb->aio_done_func = sd_finish_aiocb; |
1688 |
|
1689 |
ret = sd_co_rw_vector(acb); |
1690 |
if (ret <= 0) { |
1691 |
qemu_aio_release(acb); |
1692 |
return ret;
|
1693 |
} |
1694 |
|
1695 |
qemu_coroutine_yield(); |
1696 |
|
1697 |
return acb->ret;
|
1698 |
} |
1699 |
|
1700 |
static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
1701 |
{ |
1702 |
BDRVSheepdogState *s = bs->opaque; |
1703 |
SheepdogObjReq hdr = { 0 };
|
1704 |
SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; |
1705 |
SheepdogInode *inode = &s->inode; |
1706 |
int ret;
|
1707 |
unsigned int wlen = 0, rlen = 0; |
1708 |
|
1709 |
if (!s->cache_enabled) {
|
1710 |
return 0; |
1711 |
} |
1712 |
|
1713 |
hdr.opcode = SD_OP_FLUSH_VDI; |
1714 |
hdr.oid = vid_to_vdi_oid(inode->vdi_id); |
1715 |
|
1716 |
ret = do_req(s->flush_fd, (SheepdogReq *)&hdr, NULL, &wlen, &rlen);
|
1717 |
if (ret) {
|
1718 |
error_report("failed to send a request to the sheep");
|
1719 |
return ret;
|
1720 |
} |
1721 |
|
1722 |
if (rsp->result == SD_RES_INVALID_PARMS) {
|
1723 |
dprintf("disable write cache since the server doesn't support it\n");
|
1724 |
|
1725 |
s->cache_enabled = false;
|
1726 |
closesocket(s->flush_fd); |
1727 |
return 0; |
1728 |
} |
1729 |
|
1730 |
if (rsp->result != SD_RES_SUCCESS) {
|
1731 |
error_report("%s", sd_strerror(rsp->result));
|
1732 |
return -EIO;
|
1733 |
} |
1734 |
|
1735 |
return 0; |
1736 |
} |
1737 |
|
1738 |
static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
1739 |
{ |
1740 |
BDRVSheepdogState *s = bs->opaque; |
1741 |
int ret, fd;
|
1742 |
uint32_t new_vid; |
1743 |
SheepdogInode *inode; |
1744 |
unsigned int datalen; |
1745 |
|
1746 |
dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
1747 |
"is_snapshot %d\n", sn_info->name, sn_info->id_str,
|
1748 |
s->name, sn_info->vm_state_size, s->is_snapshot); |
1749 |
|
1750 |
if (s->is_snapshot) {
|
1751 |
error_report("You can't create a snapshot of a snapshot VDI, "
|
1752 |
"%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
1753 |
|
1754 |
return -EINVAL;
|
1755 |
} |
1756 |
|
1757 |
dprintf("%s %s\n", sn_info->name, sn_info->id_str);
|
1758 |
|
1759 |
s->inode.vm_state_size = sn_info->vm_state_size; |
1760 |
s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; |
1761 |
/* It appears that inode.tag does not require a NUL terminator,
|
1762 |
* which means this use of strncpy is ok.
|
1763 |
*/
|
1764 |
strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
|
1765 |
/* we don't need to update entire object */
|
1766 |
datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
|
1767 |
|
1768 |
/* refresh inode. */
|
1769 |
fd = connect_to_sdog(s->addr, s->port); |
1770 |
if (fd < 0) { |
1771 |
ret = fd; |
1772 |
goto cleanup;
|
1773 |
} |
1774 |
|
1775 |
ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
|
1776 |
s->inode.nr_copies, datalen, 0, false, s->cache_enabled); |
1777 |
if (ret < 0) { |
1778 |
error_report("failed to write snapshot's inode.");
|
1779 |
goto cleanup;
|
1780 |
} |
1781 |
|
1782 |
ret = do_sd_create(s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, 1,
|
1783 |
s->addr, s->port); |
1784 |
if (ret < 0) { |
1785 |
error_report("failed to create inode for snapshot. %s",
|
1786 |
strerror(errno)); |
1787 |
goto cleanup;
|
1788 |
} |
1789 |
|
1790 |
inode = (SheepdogInode *)g_malloc(datalen); |
1791 |
|
1792 |
ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid),
|
1793 |
s->inode.nr_copies, datalen, 0, s->cache_enabled);
|
1794 |
|
1795 |
if (ret < 0) { |
1796 |
error_report("failed to read new inode info. %s", strerror(errno));
|
1797 |
goto cleanup;
|
1798 |
} |
1799 |
|
1800 |
memcpy(&s->inode, inode, datalen); |
1801 |
dprintf("s->inode: name %s snap_id %x oid %x\n",
|
1802 |
s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
1803 |
|
1804 |
cleanup:
|
1805 |
closesocket(fd); |
1806 |
return ret;
|
1807 |
} |
1808 |
|
1809 |
static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
1810 |
{ |
1811 |
BDRVSheepdogState *s = bs->opaque; |
1812 |
BDRVSheepdogState *old_s; |
1813 |
char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
|
1814 |
char *buf = NULL; |
1815 |
uint32_t vid; |
1816 |
uint32_t snapid = 0;
|
1817 |
int ret = 0, fd; |
1818 |
|
1819 |
old_s = g_malloc(sizeof(BDRVSheepdogState));
|
1820 |
|
1821 |
memcpy(old_s, s, sizeof(BDRVSheepdogState));
|
1822 |
|
1823 |
pstrcpy(vdi, sizeof(vdi), s->name);
|
1824 |
|
1825 |
snapid = strtoul(snapshot_id, NULL, 10); |
1826 |
if (snapid) {
|
1827 |
tag[0] = 0; |
1828 |
} else {
|
1829 |
pstrcpy(tag, sizeof(tag), s->name);
|
1830 |
} |
1831 |
|
1832 |
ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1);
|
1833 |
if (ret) {
|
1834 |
error_report("Failed to find_vdi_name");
|
1835 |
goto out;
|
1836 |
} |
1837 |
|
1838 |
fd = connect_to_sdog(s->addr, s->port); |
1839 |
if (fd < 0) { |
1840 |
error_report("failed to connect");
|
1841 |
ret = fd; |
1842 |
goto out;
|
1843 |
} |
1844 |
|
1845 |
buf = g_malloc(SD_INODE_SIZE); |
1846 |
ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
1847 |
SD_INODE_SIZE, 0, s->cache_enabled);
|
1848 |
|
1849 |
closesocket(fd); |
1850 |
|
1851 |
if (ret) {
|
1852 |
goto out;
|
1853 |
} |
1854 |
|
1855 |
memcpy(&s->inode, buf, sizeof(s->inode));
|
1856 |
|
1857 |
if (!s->inode.vm_state_size) {
|
1858 |
error_report("Invalid snapshot");
|
1859 |
ret = -ENOENT; |
1860 |
goto out;
|
1861 |
} |
1862 |
|
1863 |
s->is_snapshot = true;
|
1864 |
|
1865 |
g_free(buf); |
1866 |
g_free(old_s); |
1867 |
|
1868 |
return 0; |
1869 |
out:
|
1870 |
/* recover bdrv_sd_state */
|
1871 |
memcpy(s, old_s, sizeof(BDRVSheepdogState));
|
1872 |
g_free(buf); |
1873 |
g_free(old_s); |
1874 |
|
1875 |
error_report("failed to open. recover old bdrv_sd_state.");
|
1876 |
|
1877 |
return ret;
|
1878 |
} |
1879 |
|
1880 |
static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) |
1881 |
{ |
1882 |
/* FIXME: Delete specified snapshot id. */
|
1883 |
return 0; |
1884 |
} |
1885 |
|
1886 |
static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
1887 |
{ |
1888 |
BDRVSheepdogState *s = bs->opaque; |
1889 |
SheepdogReq req; |
1890 |
int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); |
1891 |
QEMUSnapshotInfo *sn_tab = NULL;
|
1892 |
unsigned wlen, rlen;
|
1893 |
int found = 0; |
1894 |
static SheepdogInode inode;
|
1895 |
unsigned long *vdi_inuse; |
1896 |
unsigned int start_nr; |
1897 |
uint64_t hval; |
1898 |
uint32_t vid; |
1899 |
|
1900 |
vdi_inuse = g_malloc(max); |
1901 |
|
1902 |
fd = connect_to_sdog(s->addr, s->port); |
1903 |
if (fd < 0) { |
1904 |
ret = fd; |
1905 |
goto out;
|
1906 |
} |
1907 |
|
1908 |
rlen = max; |
1909 |
wlen = 0;
|
1910 |
|
1911 |
memset(&req, 0, sizeof(req)); |
1912 |
|
1913 |
req.opcode = SD_OP_READ_VDIS; |
1914 |
req.data_length = max; |
1915 |
|
1916 |
ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); |
1917 |
|
1918 |
closesocket(fd); |
1919 |
if (ret) {
|
1920 |
goto out;
|
1921 |
} |
1922 |
|
1923 |
sn_tab = g_malloc0(nr * sizeof(*sn_tab));
|
1924 |
|
1925 |
/* calculate a vdi id with hash function */
|
1926 |
hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); |
1927 |
start_nr = hval & (SD_NR_VDIS - 1);
|
1928 |
|
1929 |
fd = connect_to_sdog(s->addr, s->port); |
1930 |
if (fd < 0) { |
1931 |
error_report("failed to connect");
|
1932 |
ret = fd; |
1933 |
goto out;
|
1934 |
} |
1935 |
|
1936 |
for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { |
1937 |
if (!test_bit(vid, vdi_inuse)) {
|
1938 |
break;
|
1939 |
} |
1940 |
|
1941 |
/* we don't need to read entire object */
|
1942 |
ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid),
|
1943 |
0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
1944 |
s->cache_enabled); |
1945 |
|
1946 |
if (ret) {
|
1947 |
continue;
|
1948 |
} |
1949 |
|
1950 |
if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
|
1951 |
sn_tab[found].date_sec = inode.snap_ctime >> 32;
|
1952 |
sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
|
1953 |
sn_tab[found].vm_state_size = inode.vm_state_size; |
1954 |
sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; |
1955 |
|
1956 |
snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", |
1957 |
inode.snap_id); |
1958 |
pstrcpy(sn_tab[found].name, |
1959 |
MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), |
1960 |
inode.tag); |
1961 |
found++; |
1962 |
} |
1963 |
} |
1964 |
|
1965 |
closesocket(fd); |
1966 |
out:
|
1967 |
*psn_tab = sn_tab; |
1968 |
|
1969 |
g_free(vdi_inuse); |
1970 |
|
1971 |
if (ret < 0) { |
1972 |
return ret;
|
1973 |
} |
1974 |
|
1975 |
return found;
|
1976 |
} |
1977 |
|
1978 |
static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, |
1979 |
int64_t pos, int size, int load) |
1980 |
{ |
1981 |
bool create;
|
1982 |
int fd, ret = 0, remaining = size; |
1983 |
unsigned int data_len; |
1984 |
uint64_t vmstate_oid; |
1985 |
uint32_t vdi_index; |
1986 |
uint64_t offset; |
1987 |
|
1988 |
fd = connect_to_sdog(s->addr, s->port); |
1989 |
if (fd < 0) { |
1990 |
return fd;
|
1991 |
} |
1992 |
|
1993 |
while (remaining) {
|
1994 |
vdi_index = pos / SD_DATA_OBJ_SIZE; |
1995 |
offset = pos % SD_DATA_OBJ_SIZE; |
1996 |
|
1997 |
data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset); |
1998 |
|
1999 |
vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index); |
2000 |
|
2001 |
create = (offset == 0);
|
2002 |
if (load) {
|
2003 |
ret = read_object(fd, (char *)data, vmstate_oid,
|
2004 |
s->inode.nr_copies, data_len, offset, |
2005 |
s->cache_enabled); |
2006 |
} else {
|
2007 |
ret = write_object(fd, (char *)data, vmstate_oid,
|
2008 |
s->inode.nr_copies, data_len, offset, create, |
2009 |
s->cache_enabled); |
2010 |
} |
2011 |
|
2012 |
if (ret < 0) { |
2013 |
error_report("failed to save vmstate %s", strerror(errno));
|
2014 |
goto cleanup;
|
2015 |
} |
2016 |
|
2017 |
pos += data_len; |
2018 |
data += data_len; |
2019 |
remaining -= data_len; |
2020 |
} |
2021 |
ret = size; |
2022 |
cleanup:
|
2023 |
closesocket(fd); |
2024 |
return ret;
|
2025 |
} |
2026 |
|
2027 |
static int sd_save_vmstate(BlockDriverState *bs, const uint8_t *data, |
2028 |
int64_t pos, int size)
|
2029 |
{ |
2030 |
BDRVSheepdogState *s = bs->opaque; |
2031 |
|
2032 |
return do_load_save_vmstate(s, (uint8_t *)data, pos, size, 0); |
2033 |
} |
2034 |
|
2035 |
static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, |
2036 |
int64_t pos, int size)
|
2037 |
{ |
2038 |
BDRVSheepdogState *s = bs->opaque; |
2039 |
|
2040 |
return do_load_save_vmstate(s, data, pos, size, 1); |
2041 |
} |
2042 |
|
2043 |
|
2044 |
static QEMUOptionParameter sd_create_options[] = {
|
2045 |
{ |
2046 |
.name = BLOCK_OPT_SIZE, |
2047 |
.type = OPT_SIZE, |
2048 |
.help = "Virtual disk size"
|
2049 |
}, |
2050 |
{ |
2051 |
.name = BLOCK_OPT_BACKING_FILE, |
2052 |
.type = OPT_STRING, |
2053 |
.help = "File name of a base image"
|
2054 |
}, |
2055 |
{ |
2056 |
.name = BLOCK_OPT_PREALLOC, |
2057 |
.type = OPT_STRING, |
2058 |
.help = "Preallocation mode (allowed values: off, full)"
|
2059 |
}, |
2060 |
{ NULL }
|
2061 |
}; |
2062 |
|
2063 |
BlockDriver bdrv_sheepdog = { |
2064 |
.format_name = "sheepdog",
|
2065 |
.protocol_name = "sheepdog",
|
2066 |
.instance_size = sizeof(BDRVSheepdogState),
|
2067 |
.bdrv_file_open = sd_open, |
2068 |
.bdrv_close = sd_close, |
2069 |
.bdrv_create = sd_create, |
2070 |
.bdrv_getlength = sd_getlength, |
2071 |
.bdrv_truncate = sd_truncate, |
2072 |
|
2073 |
.bdrv_co_readv = sd_co_readv, |
2074 |
.bdrv_co_writev = sd_co_writev, |
2075 |
.bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
2076 |
|
2077 |
.bdrv_snapshot_create = sd_snapshot_create, |
2078 |
.bdrv_snapshot_goto = sd_snapshot_goto, |
2079 |
.bdrv_snapshot_delete = sd_snapshot_delete, |
2080 |
.bdrv_snapshot_list = sd_snapshot_list, |
2081 |
|
2082 |
.bdrv_save_vmstate = sd_save_vmstate, |
2083 |
.bdrv_load_vmstate = sd_load_vmstate, |
2084 |
|
2085 |
.create_options = sd_create_options, |
2086 |
}; |
2087 |
|
2088 |
static void bdrv_sheepdog_init(void) |
2089 |
{ |
2090 |
bdrv_register(&bdrv_sheepdog); |
2091 |
} |
2092 |
block_init(bdrv_sheepdog_init); |