root / block / sheepdog.c @ 0d6db300
History | View | Annotate | Download (56.4 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/uri.h" |
17 |
#include "qemu/error-report.h" |
18 |
#include "qemu/sockets.h" |
19 |
#include "block/block_int.h" |
20 |
#include "qemu/bitops.h" |
21 |
|
22 |
#define SD_PROTO_VER 0x01 |
23 |
|
24 |
#define SD_DEFAULT_ADDR "localhost" |
25 |
#define SD_DEFAULT_PORT 7000 |
26 |
|
27 |
#define SD_OP_CREATE_AND_WRITE_OBJ 0x01 |
28 |
#define SD_OP_READ_OBJ 0x02 |
29 |
#define SD_OP_WRITE_OBJ 0x03 |
30 |
|
31 |
#define SD_OP_NEW_VDI 0x11 |
32 |
#define SD_OP_LOCK_VDI 0x12 |
33 |
#define SD_OP_RELEASE_VDI 0x13 |
34 |
#define SD_OP_GET_VDI_INFO 0x14 |
35 |
#define SD_OP_READ_VDIS 0x15 |
36 |
#define SD_OP_FLUSH_VDI 0x16 |
37 |
|
38 |
#define SD_FLAG_CMD_WRITE 0x01 |
39 |
#define SD_FLAG_CMD_COW 0x02 |
40 |
#define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ |
41 |
#define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ |
42 |
|
43 |
#define SD_RES_SUCCESS 0x00 /* Success */ |
44 |
#define SD_RES_UNKNOWN 0x01 /* Unknown error */ |
45 |
#define SD_RES_NO_OBJ 0x02 /* No object found */ |
46 |
#define SD_RES_EIO 0x03 /* I/O error */ |
47 |
#define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ |
48 |
#define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ |
49 |
#define SD_RES_SYSTEM_ERROR 0x06 /* System error */ |
50 |
#define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ |
51 |
#define SD_RES_NO_VDI 0x08 /* No vdi found */ |
52 |
#define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ |
53 |
#define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ |
54 |
#define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ |
55 |
#define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ |
56 |
#define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ |
57 |
#define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ |
58 |
#define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ |
59 |
#define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ |
60 |
#define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ |
61 |
#define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ |
62 |
#define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ |
63 |
#define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ |
64 |
#define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ |
65 |
#define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ |
66 |
#define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ |
67 |
#define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ |
68 |
|
69 |
/*
|
70 |
* Object ID rules
|
71 |
*
|
72 |
* 0 - 19 (20 bits): data object space
|
73 |
* 20 - 31 (12 bits): reserved data object space
|
74 |
* 32 - 55 (24 bits): vdi object space
|
75 |
* 56 - 59 ( 4 bits): reserved vdi object space
|
76 |
* 60 - 63 ( 4 bits): object type identifier space
|
77 |
*/
|
78 |
|
79 |
#define VDI_SPACE_SHIFT 32 |
80 |
#define VDI_BIT (UINT64_C(1) << 63) |
81 |
#define VMSTATE_BIT (UINT64_C(1) << 62) |
82 |
#define MAX_DATA_OBJS (UINT64_C(1) << 20) |
83 |
#define MAX_CHILDREN 1024 |
84 |
#define SD_MAX_VDI_LEN 256 |
85 |
#define SD_MAX_VDI_TAG_LEN 256 |
86 |
#define SD_NR_VDIS (1U << 24) |
87 |
#define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) |
88 |
#define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
|
89 |
#define SECTOR_SIZE 512 |
90 |
|
91 |
#define SD_INODE_SIZE (sizeof(SheepdogInode)) |
92 |
#define CURRENT_VDI_ID 0 |
93 |
|
94 |
typedef struct SheepdogReq { |
95 |
uint8_t proto_ver; |
96 |
uint8_t opcode; |
97 |
uint16_t flags; |
98 |
uint32_t epoch; |
99 |
uint32_t id; |
100 |
uint32_t data_length; |
101 |
uint32_t opcode_specific[8];
|
102 |
} SheepdogReq; |
103 |
|
104 |
typedef struct SheepdogRsp { |
105 |
uint8_t proto_ver; |
106 |
uint8_t opcode; |
107 |
uint16_t flags; |
108 |
uint32_t epoch; |
109 |
uint32_t id; |
110 |
uint32_t data_length; |
111 |
uint32_t result; |
112 |
uint32_t opcode_specific[7];
|
113 |
} SheepdogRsp; |
114 |
|
115 |
typedef struct SheepdogObjReq { |
116 |
uint8_t proto_ver; |
117 |
uint8_t opcode; |
118 |
uint16_t flags; |
119 |
uint32_t epoch; |
120 |
uint32_t id; |
121 |
uint32_t data_length; |
122 |
uint64_t oid; |
123 |
uint64_t cow_oid; |
124 |
uint32_t copies; |
125 |
uint32_t rsvd; |
126 |
uint64_t offset; |
127 |
} SheepdogObjReq; |
128 |
|
129 |
typedef struct SheepdogObjRsp { |
130 |
uint8_t proto_ver; |
131 |
uint8_t opcode; |
132 |
uint16_t flags; |
133 |
uint32_t epoch; |
134 |
uint32_t id; |
135 |
uint32_t data_length; |
136 |
uint32_t result; |
137 |
uint32_t copies; |
138 |
uint32_t pad[6];
|
139 |
} SheepdogObjRsp; |
140 |
|
141 |
typedef struct SheepdogVdiReq { |
142 |
uint8_t proto_ver; |
143 |
uint8_t opcode; |
144 |
uint16_t flags; |
145 |
uint32_t epoch; |
146 |
uint32_t id; |
147 |
uint32_t data_length; |
148 |
uint64_t vdi_size; |
149 |
uint32_t vdi_id; |
150 |
uint32_t copies; |
151 |
uint32_t snapid; |
152 |
uint32_t pad[3];
|
153 |
} SheepdogVdiReq; |
154 |
|
155 |
typedef struct SheepdogVdiRsp { |
156 |
uint8_t proto_ver; |
157 |
uint8_t opcode; |
158 |
uint16_t flags; |
159 |
uint32_t epoch; |
160 |
uint32_t id; |
161 |
uint32_t data_length; |
162 |
uint32_t result; |
163 |
uint32_t rsvd; |
164 |
uint32_t vdi_id; |
165 |
uint32_t pad[5];
|
166 |
} SheepdogVdiRsp; |
167 |
|
168 |
typedef struct SheepdogInode { |
169 |
char name[SD_MAX_VDI_LEN];
|
170 |
char tag[SD_MAX_VDI_TAG_LEN];
|
171 |
uint64_t ctime; |
172 |
uint64_t snap_ctime; |
173 |
uint64_t vm_clock_nsec; |
174 |
uint64_t vdi_size; |
175 |
uint64_t vm_state_size; |
176 |
uint16_t copy_policy; |
177 |
uint8_t nr_copies; |
178 |
uint8_t block_size_shift; |
179 |
uint32_t snap_id; |
180 |
uint32_t vdi_id; |
181 |
uint32_t parent_vdi_id; |
182 |
uint32_t child_vdi_id[MAX_CHILDREN]; |
183 |
uint32_t data_vdi_id[MAX_DATA_OBJS]; |
184 |
} SheepdogInode; |
185 |
|
186 |
/*
|
187 |
* 64 bit FNV-1a non-zero initial basis
|
188 |
*/
|
189 |
#define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) |
190 |
|
191 |
/*
|
192 |
* 64 bit Fowler/Noll/Vo FNV-1a hash code
|
193 |
*/
|
194 |
static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) |
195 |
{ |
196 |
unsigned char *bp = buf; |
197 |
unsigned char *be = bp + len; |
198 |
while (bp < be) {
|
199 |
hval ^= (uint64_t) *bp++; |
200 |
hval += (hval << 1) + (hval << 4) + (hval << 5) + |
201 |
(hval << 7) + (hval << 8) + (hval << 40); |
202 |
} |
203 |
return hval;
|
204 |
} |
205 |
|
206 |
static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) |
207 |
{ |
208 |
return inode->vdi_id == inode->data_vdi_id[idx];
|
209 |
} |
210 |
|
211 |
static inline bool is_data_obj(uint64_t oid) |
212 |
{ |
213 |
return !(VDI_BIT & oid);
|
214 |
} |
215 |
|
216 |
static inline uint64_t data_oid_to_idx(uint64_t oid) |
217 |
{ |
218 |
return oid & (MAX_DATA_OBJS - 1); |
219 |
} |
220 |
|
221 |
static inline uint64_t vid_to_vdi_oid(uint32_t vid) |
222 |
{ |
223 |
return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
|
224 |
} |
225 |
|
226 |
static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) |
227 |
{ |
228 |
return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
|
229 |
} |
230 |
|
231 |
static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) |
232 |
{ |
233 |
return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
|
234 |
} |
235 |
|
236 |
static inline bool is_snapshot(struct SheepdogInode *inode) |
237 |
{ |
238 |
return !!inode->snap_ctime;
|
239 |
} |
240 |
|
241 |
#undef dprintf
|
242 |
#ifdef DEBUG_SDOG
|
243 |
#define dprintf(fmt, args...) \
|
244 |
do { \
|
245 |
fprintf(stdout, "%s %d: " fmt, __func__, __LINE__, ##args); \ |
246 |
} while (0) |
247 |
#else
|
248 |
#define dprintf(fmt, args...)
|
249 |
#endif
|
250 |
|
251 |
typedef struct SheepdogAIOCB SheepdogAIOCB; |
252 |
|
253 |
typedef struct AIOReq { |
254 |
SheepdogAIOCB *aiocb; |
255 |
unsigned int iov_offset; |
256 |
|
257 |
uint64_t oid; |
258 |
uint64_t base_oid; |
259 |
uint64_t offset; |
260 |
unsigned int data_len; |
261 |
uint8_t flags; |
262 |
uint32_t id; |
263 |
|
264 |
QLIST_ENTRY(AIOReq) aio_siblings; |
265 |
} AIOReq; |
266 |
|
267 |
enum AIOCBState {
|
268 |
AIOCB_WRITE_UDATA, |
269 |
AIOCB_READ_UDATA, |
270 |
AIOCB_FLUSH_CACHE, |
271 |
}; |
272 |
|
273 |
struct SheepdogAIOCB {
|
274 |
BlockDriverAIOCB common; |
275 |
|
276 |
QEMUIOVector *qiov; |
277 |
|
278 |
int64_t sector_num; |
279 |
int nb_sectors;
|
280 |
|
281 |
int ret;
|
282 |
enum AIOCBState aiocb_type;
|
283 |
|
284 |
Coroutine *coroutine; |
285 |
void (*aio_done_func)(SheepdogAIOCB *);
|
286 |
|
287 |
bool canceled;
|
288 |
int nr_pending;
|
289 |
}; |
290 |
|
291 |
typedef struct BDRVSheepdogState { |
292 |
SheepdogInode inode; |
293 |
|
294 |
uint32_t min_dirty_data_idx; |
295 |
uint32_t max_dirty_data_idx; |
296 |
|
297 |
char name[SD_MAX_VDI_LEN];
|
298 |
bool is_snapshot;
|
299 |
uint32_t cache_flags; |
300 |
|
301 |
char *host_spec;
|
302 |
bool is_unix;
|
303 |
int fd;
|
304 |
|
305 |
CoMutex lock; |
306 |
Coroutine *co_send; |
307 |
Coroutine *co_recv; |
308 |
|
309 |
uint32_t aioreq_seq_num; |
310 |
QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head; |
311 |
QLIST_HEAD(pending_aio_head, AIOReq) pending_aio_head; |
312 |
} BDRVSheepdogState; |
313 |
|
314 |
static const char * sd_strerror(int err) |
315 |
{ |
316 |
int i;
|
317 |
|
318 |
static const struct { |
319 |
int err;
|
320 |
const char *desc; |
321 |
} errors[] = { |
322 |
{SD_RES_SUCCESS, "Success"},
|
323 |
{SD_RES_UNKNOWN, "Unknown error"},
|
324 |
{SD_RES_NO_OBJ, "No object found"},
|
325 |
{SD_RES_EIO, "I/O error"},
|
326 |
{SD_RES_VDI_EXIST, "VDI exists already"},
|
327 |
{SD_RES_INVALID_PARMS, "Invalid parameters"},
|
328 |
{SD_RES_SYSTEM_ERROR, "System error"},
|
329 |
{SD_RES_VDI_LOCKED, "VDI is already locked"},
|
330 |
{SD_RES_NO_VDI, "No vdi found"},
|
331 |
{SD_RES_NO_BASE_VDI, "No base VDI found"},
|
332 |
{SD_RES_VDI_READ, "Failed read the requested VDI"},
|
333 |
{SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
|
334 |
{SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
|
335 |
{SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
|
336 |
{SD_RES_NO_TAG, "Failed to find the requested tag"},
|
337 |
{SD_RES_STARTUP, "The system is still booting"},
|
338 |
{SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
|
339 |
{SD_RES_SHUTDOWN, "The system is shutting down"},
|
340 |
{SD_RES_NO_MEM, "Out of memory on the server"},
|
341 |
{SD_RES_FULL_VDI, "We already have the maximum vdis"},
|
342 |
{SD_RES_VER_MISMATCH, "Protocol version mismatch"},
|
343 |
{SD_RES_NO_SPACE, "Server has no space for new objects"},
|
344 |
{SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
|
345 |
{SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
|
346 |
{SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
|
347 |
}; |
348 |
|
349 |
for (i = 0; i < ARRAY_SIZE(errors); ++i) { |
350 |
if (errors[i].err == err) {
|
351 |
return errors[i].desc;
|
352 |
} |
353 |
} |
354 |
|
355 |
return "Invalid error code"; |
356 |
} |
357 |
|
358 |
/*
|
359 |
* Sheepdog I/O handling:
|
360 |
*
|
361 |
* 1. In sd_co_rw_vector, we send the I/O requests to the server and
|
362 |
* link the requests to the inflight_list in the
|
363 |
* BDRVSheepdogState. The function exits without waiting for
|
364 |
* receiving the response.
|
365 |
*
|
366 |
* 2. We receive the response in aio_read_response, the fd handler to
|
367 |
* the sheepdog connection. If metadata update is needed, we send
|
368 |
* the write request to the vdi object in sd_write_done, the write
|
369 |
* completion function. We switch back to sd_co_readv/writev after
|
370 |
* all the requests belonging to the AIOCB are finished.
|
371 |
*/
|
372 |
|
373 |
static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, |
374 |
uint64_t oid, unsigned int data_len, |
375 |
uint64_t offset, uint8_t flags, |
376 |
uint64_t base_oid, unsigned int iov_offset) |
377 |
{ |
378 |
AIOReq *aio_req; |
379 |
|
380 |
aio_req = g_malloc(sizeof(*aio_req));
|
381 |
aio_req->aiocb = acb; |
382 |
aio_req->iov_offset = iov_offset; |
383 |
aio_req->oid = oid; |
384 |
aio_req->base_oid = base_oid; |
385 |
aio_req->offset = offset; |
386 |
aio_req->data_len = data_len; |
387 |
aio_req->flags = flags; |
388 |
aio_req->id = s->aioreq_seq_num++; |
389 |
|
390 |
acb->nr_pending++; |
391 |
return aio_req;
|
392 |
} |
393 |
|
394 |
static inline void free_aio_req(BDRVSheepdogState *s, AIOReq *aio_req) |
395 |
{ |
396 |
SheepdogAIOCB *acb = aio_req->aiocb; |
397 |
|
398 |
QLIST_REMOVE(aio_req, aio_siblings); |
399 |
g_free(aio_req); |
400 |
|
401 |
acb->nr_pending--; |
402 |
} |
403 |
|
404 |
static void coroutine_fn sd_finish_aiocb(SheepdogAIOCB *acb) |
405 |
{ |
406 |
if (!acb->canceled) {
|
407 |
qemu_coroutine_enter(acb->coroutine, NULL);
|
408 |
} |
409 |
qemu_aio_release(acb); |
410 |
} |
411 |
|
412 |
static void sd_aio_cancel(BlockDriverAIOCB *blockacb) |
413 |
{ |
414 |
SheepdogAIOCB *acb = (SheepdogAIOCB *)blockacb; |
415 |
|
416 |
/*
|
417 |
* Sheepdog cannot cancel the requests which are already sent to
|
418 |
* the servers, so we just complete the request with -EIO here.
|
419 |
*/
|
420 |
acb->ret = -EIO; |
421 |
qemu_coroutine_enter(acb->coroutine, NULL);
|
422 |
acb->canceled = true;
|
423 |
} |
424 |
|
425 |
static const AIOCBInfo sd_aiocb_info = { |
426 |
.aiocb_size = sizeof(SheepdogAIOCB),
|
427 |
.cancel = sd_aio_cancel, |
428 |
}; |
429 |
|
430 |
static SheepdogAIOCB *sd_aio_setup(BlockDriverState *bs, QEMUIOVector *qiov,
|
431 |
int64_t sector_num, int nb_sectors)
|
432 |
{ |
433 |
SheepdogAIOCB *acb; |
434 |
|
435 |
acb = qemu_aio_get(&sd_aiocb_info, bs, NULL, NULL); |
436 |
|
437 |
acb->qiov = qiov; |
438 |
|
439 |
acb->sector_num = sector_num; |
440 |
acb->nb_sectors = nb_sectors; |
441 |
|
442 |
acb->aio_done_func = NULL;
|
443 |
acb->canceled = false;
|
444 |
acb->coroutine = qemu_coroutine_self(); |
445 |
acb->ret = 0;
|
446 |
acb->nr_pending = 0;
|
447 |
return acb;
|
448 |
} |
449 |
|
450 |
static int connect_to_sdog(BDRVSheepdogState *s) |
451 |
{ |
452 |
int fd;
|
453 |
Error *err = NULL;
|
454 |
|
455 |
if (s->is_unix) {
|
456 |
fd = unix_connect(s->host_spec, &err); |
457 |
} else {
|
458 |
fd = inet_connect(s->host_spec, &err); |
459 |
|
460 |
if (err == NULL) { |
461 |
int ret = socket_set_nodelay(fd);
|
462 |
if (ret < 0) { |
463 |
error_report("%s", strerror(errno));
|
464 |
} |
465 |
} |
466 |
} |
467 |
|
468 |
if (err != NULL) { |
469 |
qerror_report_err(err); |
470 |
error_free(err); |
471 |
} else {
|
472 |
socket_set_nonblock(fd); |
473 |
} |
474 |
|
475 |
return fd;
|
476 |
} |
477 |
|
478 |
static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, |
479 |
unsigned int *wlen) |
480 |
{ |
481 |
int ret;
|
482 |
|
483 |
ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
|
484 |
if (ret < sizeof(*hdr)) { |
485 |
error_report("failed to send a req, %s", strerror(errno));
|
486 |
return ret;
|
487 |
} |
488 |
|
489 |
ret = qemu_co_send(sockfd, data, *wlen); |
490 |
if (ret < *wlen) {
|
491 |
error_report("failed to send a req, %s", strerror(errno));
|
492 |
} |
493 |
|
494 |
return ret;
|
495 |
} |
496 |
|
497 |
static void restart_co_req(void *opaque) |
498 |
{ |
499 |
Coroutine *co = opaque; |
500 |
|
501 |
qemu_coroutine_enter(co, NULL);
|
502 |
} |
503 |
|
504 |
typedef struct SheepdogReqCo { |
505 |
int sockfd;
|
506 |
SheepdogReq *hdr; |
507 |
void *data;
|
508 |
unsigned int *wlen; |
509 |
unsigned int *rlen; |
510 |
int ret;
|
511 |
bool finished;
|
512 |
} SheepdogReqCo; |
513 |
|
514 |
static coroutine_fn void do_co_req(void *opaque) |
515 |
{ |
516 |
int ret;
|
517 |
Coroutine *co; |
518 |
SheepdogReqCo *srco = opaque; |
519 |
int sockfd = srco->sockfd;
|
520 |
SheepdogReq *hdr = srco->hdr; |
521 |
void *data = srco->data;
|
522 |
unsigned int *wlen = srco->wlen; |
523 |
unsigned int *rlen = srco->rlen; |
524 |
|
525 |
co = qemu_coroutine_self(); |
526 |
qemu_aio_set_fd_handler(sockfd, NULL, restart_co_req, NULL, co); |
527 |
|
528 |
ret = send_co_req(sockfd, hdr, data, wlen); |
529 |
if (ret < 0) { |
530 |
goto out;
|
531 |
} |
532 |
|
533 |
qemu_aio_set_fd_handler(sockfd, restart_co_req, NULL, NULL, co); |
534 |
|
535 |
ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
|
536 |
if (ret < sizeof(*hdr)) { |
537 |
error_report("failed to get a rsp, %s", strerror(errno));
|
538 |
ret = -errno; |
539 |
goto out;
|
540 |
} |
541 |
|
542 |
if (*rlen > hdr->data_length) {
|
543 |
*rlen = hdr->data_length; |
544 |
} |
545 |
|
546 |
if (*rlen) {
|
547 |
ret = qemu_co_recv(sockfd, data, *rlen); |
548 |
if (ret < *rlen) {
|
549 |
error_report("failed to get the data, %s", strerror(errno));
|
550 |
ret = -errno; |
551 |
goto out;
|
552 |
} |
553 |
} |
554 |
ret = 0;
|
555 |
out:
|
556 |
qemu_aio_set_fd_handler(sockfd, NULL, NULL, NULL, NULL); |
557 |
|
558 |
srco->ret = ret; |
559 |
srco->finished = true;
|
560 |
} |
561 |
|
562 |
static int do_req(int sockfd, SheepdogReq *hdr, void *data, |
563 |
unsigned int *wlen, unsigned int *rlen) |
564 |
{ |
565 |
Coroutine *co; |
566 |
SheepdogReqCo srco = { |
567 |
.sockfd = sockfd, |
568 |
.hdr = hdr, |
569 |
.data = data, |
570 |
.wlen = wlen, |
571 |
.rlen = rlen, |
572 |
.ret = 0,
|
573 |
.finished = false,
|
574 |
}; |
575 |
|
576 |
if (qemu_in_coroutine()) {
|
577 |
do_co_req(&srco); |
578 |
} else {
|
579 |
co = qemu_coroutine_create(do_co_req); |
580 |
qemu_coroutine_enter(co, &srco); |
581 |
while (!srco.finished) {
|
582 |
qemu_aio_wait(); |
583 |
} |
584 |
} |
585 |
|
586 |
return srco.ret;
|
587 |
} |
588 |
|
589 |
static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
590 |
struct iovec *iov, int niov, bool create, |
591 |
enum AIOCBState aiocb_type);
|
592 |
|
593 |
|
594 |
static AIOReq *find_pending_req(BDRVSheepdogState *s, uint64_t oid)
|
595 |
{ |
596 |
AIOReq *aio_req; |
597 |
|
598 |
QLIST_FOREACH(aio_req, &s->pending_aio_head, aio_siblings) { |
599 |
if (aio_req->oid == oid) {
|
600 |
return aio_req;
|
601 |
} |
602 |
} |
603 |
|
604 |
return NULL; |
605 |
} |
606 |
|
607 |
/*
|
608 |
* This function searchs pending requests to the object `oid', and
|
609 |
* sends them.
|
610 |
*/
|
611 |
static void coroutine_fn send_pending_req(BDRVSheepdogState *s, uint64_t oid) |
612 |
{ |
613 |
AIOReq *aio_req; |
614 |
SheepdogAIOCB *acb; |
615 |
int ret;
|
616 |
|
617 |
while ((aio_req = find_pending_req(s, oid)) != NULL) { |
618 |
acb = aio_req->aiocb; |
619 |
/* move aio_req from pending list to inflight one */
|
620 |
QLIST_REMOVE(aio_req, aio_siblings); |
621 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
622 |
ret = add_aio_request(s, aio_req, acb->qiov->iov, |
623 |
acb->qiov->niov, false, acb->aiocb_type);
|
624 |
if (ret < 0) { |
625 |
error_report("add_aio_request is failed");
|
626 |
free_aio_req(s, aio_req); |
627 |
if (!acb->nr_pending) {
|
628 |
sd_finish_aiocb(acb); |
629 |
} |
630 |
} |
631 |
} |
632 |
} |
633 |
|
634 |
/*
|
635 |
* Receive responses of the I/O requests.
|
636 |
*
|
637 |
* This function is registered as a fd handler, and called from the
|
638 |
* main loop when s->fd is ready for reading responses.
|
639 |
*/
|
640 |
static void coroutine_fn aio_read_response(void *opaque) |
641 |
{ |
642 |
SheepdogObjRsp rsp; |
643 |
BDRVSheepdogState *s = opaque; |
644 |
int fd = s->fd;
|
645 |
int ret;
|
646 |
AIOReq *aio_req = NULL;
|
647 |
SheepdogAIOCB *acb; |
648 |
unsigned long idx; |
649 |
|
650 |
if (QLIST_EMPTY(&s->inflight_aio_head)) {
|
651 |
goto out;
|
652 |
} |
653 |
|
654 |
/* read a header */
|
655 |
ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
|
656 |
if (ret < 0) { |
657 |
error_report("failed to get the header, %s", strerror(errno));
|
658 |
goto out;
|
659 |
} |
660 |
|
661 |
/* find the right aio_req from the inflight aio list */
|
662 |
QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { |
663 |
if (aio_req->id == rsp.id) {
|
664 |
break;
|
665 |
} |
666 |
} |
667 |
if (!aio_req) {
|
668 |
error_report("cannot find aio_req %x", rsp.id);
|
669 |
goto out;
|
670 |
} |
671 |
|
672 |
acb = aio_req->aiocb; |
673 |
|
674 |
switch (acb->aiocb_type) {
|
675 |
case AIOCB_WRITE_UDATA:
|
676 |
/* this coroutine context is no longer suitable for co_recv
|
677 |
* because we may send data to update vdi objects */
|
678 |
s->co_recv = NULL;
|
679 |
if (!is_data_obj(aio_req->oid)) {
|
680 |
break;
|
681 |
} |
682 |
idx = data_oid_to_idx(aio_req->oid); |
683 |
|
684 |
if (s->inode.data_vdi_id[idx] != s->inode.vdi_id) {
|
685 |
/*
|
686 |
* If the object is newly created one, we need to update
|
687 |
* the vdi object (metadata object). min_dirty_data_idx
|
688 |
* and max_dirty_data_idx are changed to include updated
|
689 |
* index between them.
|
690 |
*/
|
691 |
if (rsp.result == SD_RES_SUCCESS) {
|
692 |
s->inode.data_vdi_id[idx] = s->inode.vdi_id; |
693 |
s->max_dirty_data_idx = MAX(idx, s->max_dirty_data_idx); |
694 |
s->min_dirty_data_idx = MIN(idx, s->min_dirty_data_idx); |
695 |
} |
696 |
/*
|
697 |
* Some requests may be blocked because simultaneous
|
698 |
* create requests are not allowed, so we search the
|
699 |
* pending requests here.
|
700 |
*/
|
701 |
send_pending_req(s, aio_req->oid); |
702 |
} |
703 |
break;
|
704 |
case AIOCB_READ_UDATA:
|
705 |
ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, |
706 |
aio_req->iov_offset, rsp.data_length); |
707 |
if (ret < 0) { |
708 |
error_report("failed to get the data, %s", strerror(errno));
|
709 |
goto out;
|
710 |
} |
711 |
break;
|
712 |
case AIOCB_FLUSH_CACHE:
|
713 |
if (rsp.result == SD_RES_INVALID_PARMS) {
|
714 |
dprintf("disable cache since the server doesn't support it\n");
|
715 |
s->cache_flags = SD_FLAG_CMD_DIRECT; |
716 |
rsp.result = SD_RES_SUCCESS; |
717 |
} |
718 |
break;
|
719 |
} |
720 |
|
721 |
if (rsp.result != SD_RES_SUCCESS) {
|
722 |
acb->ret = -EIO; |
723 |
error_report("%s", sd_strerror(rsp.result));
|
724 |
} |
725 |
|
726 |
free_aio_req(s, aio_req); |
727 |
if (!acb->nr_pending) {
|
728 |
/*
|
729 |
* We've finished all requests which belong to the AIOCB, so
|
730 |
* we can switch back to sd_co_readv/writev now.
|
731 |
*/
|
732 |
acb->aio_done_func(acb); |
733 |
} |
734 |
out:
|
735 |
s->co_recv = NULL;
|
736 |
} |
737 |
|
738 |
static void co_read_response(void *opaque) |
739 |
{ |
740 |
BDRVSheepdogState *s = opaque; |
741 |
|
742 |
if (!s->co_recv) {
|
743 |
s->co_recv = qemu_coroutine_create(aio_read_response); |
744 |
} |
745 |
|
746 |
qemu_coroutine_enter(s->co_recv, opaque); |
747 |
} |
748 |
|
749 |
static void co_write_request(void *opaque) |
750 |
{ |
751 |
BDRVSheepdogState *s = opaque; |
752 |
|
753 |
qemu_coroutine_enter(s->co_send, NULL);
|
754 |
} |
755 |
|
756 |
static int aio_flush_request(void *opaque) |
757 |
{ |
758 |
BDRVSheepdogState *s = opaque; |
759 |
|
760 |
return !QLIST_EMPTY(&s->inflight_aio_head) ||
|
761 |
!QLIST_EMPTY(&s->pending_aio_head); |
762 |
} |
763 |
|
764 |
/*
|
765 |
* Return a socket discriptor to read/write objects.
|
766 |
*
|
767 |
* We cannot use this discriptor for other operations because
|
768 |
* the block driver may be on waiting response from the server.
|
769 |
*/
|
770 |
static int get_sheep_fd(BDRVSheepdogState *s) |
771 |
{ |
772 |
int fd;
|
773 |
|
774 |
fd = connect_to_sdog(s); |
775 |
if (fd < 0) { |
776 |
return fd;
|
777 |
} |
778 |
|
779 |
qemu_aio_set_fd_handler(fd, co_read_response, NULL, aio_flush_request, s);
|
780 |
return fd;
|
781 |
} |
782 |
|
783 |
static int sd_parse_uri(BDRVSheepdogState *s, const char *filename, |
784 |
char *vdi, uint32_t *snapid, char *tag) |
785 |
{ |
786 |
URI *uri; |
787 |
QueryParams *qp = NULL;
|
788 |
int ret = 0; |
789 |
|
790 |
uri = uri_parse(filename); |
791 |
if (!uri) {
|
792 |
return -EINVAL;
|
793 |
} |
794 |
|
795 |
/* transport */
|
796 |
if (!strcmp(uri->scheme, "sheepdog")) { |
797 |
s->is_unix = false;
|
798 |
} else if (!strcmp(uri->scheme, "sheepdog+tcp")) { |
799 |
s->is_unix = false;
|
800 |
} else if (!strcmp(uri->scheme, "sheepdog+unix")) { |
801 |
s->is_unix = true;
|
802 |
} else {
|
803 |
ret = -EINVAL; |
804 |
goto out;
|
805 |
} |
806 |
|
807 |
if (uri->path == NULL || !strcmp(uri->path, "/")) { |
808 |
ret = -EINVAL; |
809 |
goto out;
|
810 |
} |
811 |
pstrcpy(vdi, SD_MAX_VDI_LEN, uri->path + 1);
|
812 |
|
813 |
qp = query_params_parse(uri->query); |
814 |
if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) { |
815 |
ret = -EINVAL; |
816 |
goto out;
|
817 |
} |
818 |
|
819 |
if (s->is_unix) {
|
820 |
/* sheepdog+unix:///vdiname?socket=path */
|
821 |
if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { |
822 |
ret = -EINVAL; |
823 |
goto out;
|
824 |
} |
825 |
s->host_spec = g_strdup(qp->p[0].value);
|
826 |
} else {
|
827 |
/* sheepdog[+tcp]://[host:port]/vdiname */
|
828 |
s->host_spec = g_strdup_printf("%s:%d", uri->server ?: SD_DEFAULT_ADDR,
|
829 |
uri->port ?: SD_DEFAULT_PORT); |
830 |
} |
831 |
|
832 |
/* snapshot tag */
|
833 |
if (uri->fragment) {
|
834 |
*snapid = strtoul(uri->fragment, NULL, 10); |
835 |
if (*snapid == 0) { |
836 |
pstrcpy(tag, SD_MAX_VDI_TAG_LEN, uri->fragment); |
837 |
} |
838 |
} else {
|
839 |
*snapid = CURRENT_VDI_ID; /* search current vdi */
|
840 |
} |
841 |
|
842 |
out:
|
843 |
if (qp) {
|
844 |
query_params_free(qp); |
845 |
} |
846 |
uri_free(uri); |
847 |
return ret;
|
848 |
} |
849 |
|
850 |
/*
|
851 |
* Parse a filename (old syntax)
|
852 |
*
|
853 |
* filename must be one of the following formats:
|
854 |
* 1. [vdiname]
|
855 |
* 2. [vdiname]:[snapid]
|
856 |
* 3. [vdiname]:[tag]
|
857 |
* 4. [hostname]:[port]:[vdiname]
|
858 |
* 5. [hostname]:[port]:[vdiname]:[snapid]
|
859 |
* 6. [hostname]:[port]:[vdiname]:[tag]
|
860 |
*
|
861 |
* You can boot from the snapshot images by specifying `snapid` or
|
862 |
* `tag'.
|
863 |
*
|
864 |
* You can run VMs outside the Sheepdog cluster by specifying
|
865 |
* `hostname' and `port' (experimental).
|
866 |
*/
|
867 |
static int parse_vdiname(BDRVSheepdogState *s, const char *filename, |
868 |
char *vdi, uint32_t *snapid, char *tag) |
869 |
{ |
870 |
char *p, *q, *uri;
|
871 |
const char *host_spec, *vdi_spec; |
872 |
int nr_sep, ret;
|
873 |
|
874 |
strstart(filename, "sheepdog:", (const char **)&filename); |
875 |
p = q = g_strdup(filename); |
876 |
|
877 |
/* count the number of separators */
|
878 |
nr_sep = 0;
|
879 |
while (*p) {
|
880 |
if (*p == ':') { |
881 |
nr_sep++; |
882 |
} |
883 |
p++; |
884 |
} |
885 |
p = q; |
886 |
|
887 |
/* use the first two tokens as host_spec. */
|
888 |
if (nr_sep >= 2) { |
889 |
host_spec = p; |
890 |
p = strchr(p, ':');
|
891 |
p++; |
892 |
p = strchr(p, ':');
|
893 |
*p++ = '\0';
|
894 |
} else {
|
895 |
host_spec = "";
|
896 |
} |
897 |
|
898 |
vdi_spec = p; |
899 |
|
900 |
p = strchr(vdi_spec, ':');
|
901 |
if (p) {
|
902 |
*p++ = '#';
|
903 |
} |
904 |
|
905 |
uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec);
|
906 |
|
907 |
ret = sd_parse_uri(s, uri, vdi, snapid, tag); |
908 |
|
909 |
g_free(q); |
910 |
g_free(uri); |
911 |
|
912 |
return ret;
|
913 |
} |
914 |
|
915 |
static int find_vdi_name(BDRVSheepdogState *s, char *filename, uint32_t snapid, |
916 |
char *tag, uint32_t *vid, int for_snapshot) |
917 |
{ |
918 |
int ret, fd;
|
919 |
SheepdogVdiReq hdr; |
920 |
SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; |
921 |
unsigned int wlen, rlen = 0; |
922 |
char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
|
923 |
|
924 |
fd = connect_to_sdog(s); |
925 |
if (fd < 0) { |
926 |
return fd;
|
927 |
} |
928 |
|
929 |
/* This pair of strncpy calls ensures that the buffer is zero-filled,
|
930 |
* which is desirable since we'll soon be sending those bytes, and
|
931 |
* don't want the send_req to read uninitialized data.
|
932 |
*/
|
933 |
strncpy(buf, filename, SD_MAX_VDI_LEN); |
934 |
strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); |
935 |
|
936 |
memset(&hdr, 0, sizeof(hdr)); |
937 |
if (for_snapshot) {
|
938 |
hdr.opcode = SD_OP_GET_VDI_INFO; |
939 |
} else {
|
940 |
hdr.opcode = SD_OP_LOCK_VDI; |
941 |
} |
942 |
wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; |
943 |
hdr.proto_ver = SD_PROTO_VER; |
944 |
hdr.data_length = wlen; |
945 |
hdr.snapid = snapid; |
946 |
hdr.flags = SD_FLAG_CMD_WRITE; |
947 |
|
948 |
ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
949 |
if (ret) {
|
950 |
goto out;
|
951 |
} |
952 |
|
953 |
if (rsp->result != SD_RES_SUCCESS) {
|
954 |
error_report("cannot get vdi info, %s, %s %d %s",
|
955 |
sd_strerror(rsp->result), filename, snapid, tag); |
956 |
if (rsp->result == SD_RES_NO_VDI) {
|
957 |
ret = -ENOENT; |
958 |
} else {
|
959 |
ret = -EIO; |
960 |
} |
961 |
goto out;
|
962 |
} |
963 |
*vid = rsp->vdi_id; |
964 |
|
965 |
ret = 0;
|
966 |
out:
|
967 |
closesocket(fd); |
968 |
return ret;
|
969 |
} |
970 |
|
971 |
static int coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, |
972 |
struct iovec *iov, int niov, bool create, |
973 |
enum AIOCBState aiocb_type)
|
974 |
{ |
975 |
int nr_copies = s->inode.nr_copies;
|
976 |
SheepdogObjReq hdr; |
977 |
unsigned int wlen = 0; |
978 |
int ret;
|
979 |
uint64_t oid = aio_req->oid; |
980 |
unsigned int datalen = aio_req->data_len; |
981 |
uint64_t offset = aio_req->offset; |
982 |
uint8_t flags = aio_req->flags; |
983 |
uint64_t old_oid = aio_req->base_oid; |
984 |
|
985 |
if (!nr_copies) {
|
986 |
error_report("bug");
|
987 |
} |
988 |
|
989 |
memset(&hdr, 0, sizeof(hdr)); |
990 |
|
991 |
switch (aiocb_type) {
|
992 |
case AIOCB_FLUSH_CACHE:
|
993 |
hdr.opcode = SD_OP_FLUSH_VDI; |
994 |
break;
|
995 |
case AIOCB_READ_UDATA:
|
996 |
hdr.opcode = SD_OP_READ_OBJ; |
997 |
hdr.flags = flags; |
998 |
break;
|
999 |
case AIOCB_WRITE_UDATA:
|
1000 |
if (create) {
|
1001 |
hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; |
1002 |
} else {
|
1003 |
hdr.opcode = SD_OP_WRITE_OBJ; |
1004 |
} |
1005 |
wlen = datalen; |
1006 |
hdr.flags = SD_FLAG_CMD_WRITE | flags; |
1007 |
break;
|
1008 |
} |
1009 |
|
1010 |
if (s->cache_flags) {
|
1011 |
hdr.flags |= s->cache_flags; |
1012 |
} |
1013 |
|
1014 |
hdr.oid = oid; |
1015 |
hdr.cow_oid = old_oid; |
1016 |
hdr.copies = s->inode.nr_copies; |
1017 |
|
1018 |
hdr.data_length = datalen; |
1019 |
hdr.offset = offset; |
1020 |
|
1021 |
hdr.id = aio_req->id; |
1022 |
|
1023 |
qemu_co_mutex_lock(&s->lock); |
1024 |
s->co_send = qemu_coroutine_self(); |
1025 |
qemu_aio_set_fd_handler(s->fd, co_read_response, co_write_request, |
1026 |
aio_flush_request, s); |
1027 |
socket_set_cork(s->fd, 1);
|
1028 |
|
1029 |
/* send a header */
|
1030 |
ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
|
1031 |
if (ret < 0) { |
1032 |
qemu_co_mutex_unlock(&s->lock); |
1033 |
error_report("failed to send a req, %s", strerror(errno));
|
1034 |
return -errno;
|
1035 |
} |
1036 |
|
1037 |
if (wlen) {
|
1038 |
ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); |
1039 |
if (ret < 0) { |
1040 |
qemu_co_mutex_unlock(&s->lock); |
1041 |
error_report("failed to send a data, %s", strerror(errno));
|
1042 |
return -errno;
|
1043 |
} |
1044 |
} |
1045 |
|
1046 |
socket_set_cork(s->fd, 0);
|
1047 |
qemu_aio_set_fd_handler(s->fd, co_read_response, NULL,
|
1048 |
aio_flush_request, s); |
1049 |
qemu_co_mutex_unlock(&s->lock); |
1050 |
|
1051 |
return 0; |
1052 |
} |
1053 |
|
1054 |
static int read_write_object(int fd, char *buf, uint64_t oid, int copies, |
1055 |
unsigned int datalen, uint64_t offset, |
1056 |
bool write, bool create, uint32_t cache_flags) |
1057 |
{ |
1058 |
SheepdogObjReq hdr; |
1059 |
SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; |
1060 |
unsigned int wlen, rlen; |
1061 |
int ret;
|
1062 |
|
1063 |
memset(&hdr, 0, sizeof(hdr)); |
1064 |
|
1065 |
if (write) {
|
1066 |
wlen = datalen; |
1067 |
rlen = 0;
|
1068 |
hdr.flags = SD_FLAG_CMD_WRITE; |
1069 |
if (create) {
|
1070 |
hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; |
1071 |
} else {
|
1072 |
hdr.opcode = SD_OP_WRITE_OBJ; |
1073 |
} |
1074 |
} else {
|
1075 |
wlen = 0;
|
1076 |
rlen = datalen; |
1077 |
hdr.opcode = SD_OP_READ_OBJ; |
1078 |
} |
1079 |
|
1080 |
hdr.flags |= cache_flags; |
1081 |
|
1082 |
hdr.oid = oid; |
1083 |
hdr.data_length = datalen; |
1084 |
hdr.offset = offset; |
1085 |
hdr.copies = copies; |
1086 |
|
1087 |
ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
1088 |
if (ret) {
|
1089 |
error_report("failed to send a request to the sheep");
|
1090 |
return ret;
|
1091 |
} |
1092 |
|
1093 |
switch (rsp->result) {
|
1094 |
case SD_RES_SUCCESS:
|
1095 |
return 0; |
1096 |
default:
|
1097 |
error_report("%s", sd_strerror(rsp->result));
|
1098 |
return -EIO;
|
1099 |
} |
1100 |
} |
1101 |
|
1102 |
static int read_object(int fd, char *buf, uint64_t oid, int copies, |
1103 |
unsigned int datalen, uint64_t offset, |
1104 |
uint32_t cache_flags) |
1105 |
{ |
1106 |
return read_write_object(fd, buf, oid, copies, datalen, offset, false, |
1107 |
false, cache_flags);
|
1108 |
} |
1109 |
|
1110 |
static int write_object(int fd, char *buf, uint64_t oid, int copies, |
1111 |
unsigned int datalen, uint64_t offset, bool create, |
1112 |
uint32_t cache_flags) |
1113 |
{ |
1114 |
return read_write_object(fd, buf, oid, copies, datalen, offset, true, |
1115 |
create, cache_flags); |
1116 |
} |
1117 |
|
1118 |
static int sd_open(BlockDriverState *bs, const char *filename, int flags) |
1119 |
{ |
1120 |
int ret, fd;
|
1121 |
uint32_t vid = 0;
|
1122 |
BDRVSheepdogState *s = bs->opaque; |
1123 |
char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
|
1124 |
uint32_t snapid; |
1125 |
char *buf = NULL; |
1126 |
|
1127 |
QLIST_INIT(&s->inflight_aio_head); |
1128 |
QLIST_INIT(&s->pending_aio_head); |
1129 |
s->fd = -1;
|
1130 |
|
1131 |
memset(vdi, 0, sizeof(vdi)); |
1132 |
memset(tag, 0, sizeof(tag)); |
1133 |
|
1134 |
if (strstr(filename, "://")) { |
1135 |
ret = sd_parse_uri(s, filename, vdi, &snapid, tag); |
1136 |
} else {
|
1137 |
ret = parse_vdiname(s, filename, vdi, &snapid, tag); |
1138 |
} |
1139 |
if (ret < 0) { |
1140 |
goto out;
|
1141 |
} |
1142 |
s->fd = get_sheep_fd(s); |
1143 |
if (s->fd < 0) { |
1144 |
ret = s->fd; |
1145 |
goto out;
|
1146 |
} |
1147 |
|
1148 |
ret = find_vdi_name(s, vdi, snapid, tag, &vid, 0);
|
1149 |
if (ret) {
|
1150 |
goto out;
|
1151 |
} |
1152 |
|
1153 |
/*
|
1154 |
* QEMU block layer emulates writethrough cache as 'writeback + flush', so
|
1155 |
* we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
|
1156 |
*/
|
1157 |
s->cache_flags = SD_FLAG_CMD_CACHE; |
1158 |
if (flags & BDRV_O_NOCACHE) {
|
1159 |
s->cache_flags = SD_FLAG_CMD_DIRECT; |
1160 |
} |
1161 |
|
1162 |
if (snapid || tag[0] != '\0') { |
1163 |
dprintf("%" PRIx32 " snapshot inode was open.\n", vid); |
1164 |
s->is_snapshot = true;
|
1165 |
} |
1166 |
|
1167 |
fd = connect_to_sdog(s); |
1168 |
if (fd < 0) { |
1169 |
ret = fd; |
1170 |
goto out;
|
1171 |
} |
1172 |
|
1173 |
buf = g_malloc(SD_INODE_SIZE); |
1174 |
ret = read_object(fd, buf, vid_to_vdi_oid(vid), 0, SD_INODE_SIZE, 0, |
1175 |
s->cache_flags); |
1176 |
|
1177 |
closesocket(fd); |
1178 |
|
1179 |
if (ret) {
|
1180 |
goto out;
|
1181 |
} |
1182 |
|
1183 |
memcpy(&s->inode, buf, sizeof(s->inode));
|
1184 |
s->min_dirty_data_idx = UINT32_MAX; |
1185 |
s->max_dirty_data_idx = 0;
|
1186 |
|
1187 |
bs->total_sectors = s->inode.vdi_size / SECTOR_SIZE; |
1188 |
pstrcpy(s->name, sizeof(s->name), vdi);
|
1189 |
qemu_co_mutex_init(&s->lock); |
1190 |
g_free(buf); |
1191 |
return 0; |
1192 |
out:
|
1193 |
qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
1194 |
if (s->fd >= 0) { |
1195 |
closesocket(s->fd); |
1196 |
} |
1197 |
g_free(buf); |
1198 |
return ret;
|
1199 |
} |
1200 |
|
1201 |
static int do_sd_create(BDRVSheepdogState *s, char *filename, int64_t vdi_size, |
1202 |
uint32_t base_vid, uint32_t *vdi_id, int snapshot)
|
1203 |
{ |
1204 |
SheepdogVdiReq hdr; |
1205 |
SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; |
1206 |
int fd, ret;
|
1207 |
unsigned int wlen, rlen = 0; |
1208 |
char buf[SD_MAX_VDI_LEN];
|
1209 |
|
1210 |
fd = connect_to_sdog(s); |
1211 |
if (fd < 0) { |
1212 |
return fd;
|
1213 |
} |
1214 |
|
1215 |
/* FIXME: would it be better to fail (e.g., return -EIO) when filename
|
1216 |
* does not fit in buf? For now, just truncate and avoid buffer overrun.
|
1217 |
*/
|
1218 |
memset(buf, 0, sizeof(buf)); |
1219 |
pstrcpy(buf, sizeof(buf), filename);
|
1220 |
|
1221 |
memset(&hdr, 0, sizeof(hdr)); |
1222 |
hdr.opcode = SD_OP_NEW_VDI; |
1223 |
hdr.vdi_id = base_vid; |
1224 |
|
1225 |
wlen = SD_MAX_VDI_LEN; |
1226 |
|
1227 |
hdr.flags = SD_FLAG_CMD_WRITE; |
1228 |
hdr.snapid = snapshot; |
1229 |
|
1230 |
hdr.data_length = wlen; |
1231 |
hdr.vdi_size = vdi_size; |
1232 |
|
1233 |
ret = do_req(fd, (SheepdogReq *)&hdr, buf, &wlen, &rlen); |
1234 |
|
1235 |
closesocket(fd); |
1236 |
|
1237 |
if (ret) {
|
1238 |
return ret;
|
1239 |
} |
1240 |
|
1241 |
if (rsp->result != SD_RES_SUCCESS) {
|
1242 |
error_report("%s, %s", sd_strerror(rsp->result), filename);
|
1243 |
return -EIO;
|
1244 |
} |
1245 |
|
1246 |
if (vdi_id) {
|
1247 |
*vdi_id = rsp->vdi_id; |
1248 |
} |
1249 |
|
1250 |
return 0; |
1251 |
} |
1252 |
|
1253 |
static int sd_prealloc(const char *filename) |
1254 |
{ |
1255 |
BlockDriverState *bs = NULL;
|
1256 |
uint32_t idx, max_idx; |
1257 |
int64_t vdi_size; |
1258 |
void *buf = g_malloc0(SD_DATA_OBJ_SIZE);
|
1259 |
int ret;
|
1260 |
|
1261 |
ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); |
1262 |
if (ret < 0) { |
1263 |
goto out;
|
1264 |
} |
1265 |
|
1266 |
vdi_size = bdrv_getlength(bs); |
1267 |
if (vdi_size < 0) { |
1268 |
ret = vdi_size; |
1269 |
goto out;
|
1270 |
} |
1271 |
max_idx = DIV_ROUND_UP(vdi_size, SD_DATA_OBJ_SIZE); |
1272 |
|
1273 |
for (idx = 0; idx < max_idx; idx++) { |
1274 |
/*
|
1275 |
* The created image can be a cloned image, so we need to read
|
1276 |
* a data from the source image.
|
1277 |
*/
|
1278 |
ret = bdrv_pread(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); |
1279 |
if (ret < 0) { |
1280 |
goto out;
|
1281 |
} |
1282 |
ret = bdrv_pwrite(bs, idx * SD_DATA_OBJ_SIZE, buf, SD_DATA_OBJ_SIZE); |
1283 |
if (ret < 0) { |
1284 |
goto out;
|
1285 |
} |
1286 |
} |
1287 |
out:
|
1288 |
if (bs) {
|
1289 |
bdrv_delete(bs); |
1290 |
} |
1291 |
g_free(buf); |
1292 |
|
1293 |
return ret;
|
1294 |
} |
1295 |
|
1296 |
static int sd_create(const char *filename, QEMUOptionParameter *options) |
1297 |
{ |
1298 |
int ret = 0; |
1299 |
uint32_t vid = 0, base_vid = 0; |
1300 |
int64_t vdi_size = 0;
|
1301 |
char *backing_file = NULL; |
1302 |
BDRVSheepdogState *s; |
1303 |
char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
|
1304 |
uint32_t snapid; |
1305 |
bool prealloc = false; |
1306 |
|
1307 |
s = g_malloc0(sizeof(BDRVSheepdogState));
|
1308 |
|
1309 |
memset(vdi, 0, sizeof(vdi)); |
1310 |
memset(tag, 0, sizeof(tag)); |
1311 |
if (strstr(filename, "://")) { |
1312 |
ret = sd_parse_uri(s, filename, vdi, &snapid, tag); |
1313 |
} else {
|
1314 |
ret = parse_vdiname(s, filename, vdi, &snapid, tag); |
1315 |
} |
1316 |
if (ret < 0) { |
1317 |
goto out;
|
1318 |
} |
1319 |
|
1320 |
while (options && options->name) {
|
1321 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
1322 |
vdi_size = options->value.n; |
1323 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
1324 |
backing_file = options->value.s; |
1325 |
} else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
1326 |
if (!options->value.s || !strcmp(options->value.s, "off")) { |
1327 |
prealloc = false;
|
1328 |
} else if (!strcmp(options->value.s, "full")) { |
1329 |
prealloc = true;
|
1330 |
} else {
|
1331 |
error_report("Invalid preallocation mode: '%s'",
|
1332 |
options->value.s); |
1333 |
ret = -EINVAL; |
1334 |
goto out;
|
1335 |
} |
1336 |
} |
1337 |
options++; |
1338 |
} |
1339 |
|
1340 |
if (vdi_size > SD_MAX_VDI_SIZE) {
|
1341 |
error_report("too big image size");
|
1342 |
ret = -EINVAL; |
1343 |
goto out;
|
1344 |
} |
1345 |
|
1346 |
if (backing_file) {
|
1347 |
BlockDriverState *bs; |
1348 |
BDRVSheepdogState *s; |
1349 |
BlockDriver *drv; |
1350 |
|
1351 |
/* Currently, only Sheepdog backing image is supported. */
|
1352 |
drv = bdrv_find_protocol(backing_file); |
1353 |
if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { |
1354 |
error_report("backing_file must be a sheepdog image");
|
1355 |
ret = -EINVAL; |
1356 |
goto out;
|
1357 |
} |
1358 |
|
1359 |
ret = bdrv_file_open(&bs, backing_file, 0);
|
1360 |
if (ret < 0) { |
1361 |
goto out;
|
1362 |
} |
1363 |
|
1364 |
s = bs->opaque; |
1365 |
|
1366 |
if (!is_snapshot(&s->inode)) {
|
1367 |
error_report("cannot clone from a non snapshot vdi");
|
1368 |
bdrv_delete(bs); |
1369 |
ret = -EINVAL; |
1370 |
goto out;
|
1371 |
} |
1372 |
|
1373 |
base_vid = s->inode.vdi_id; |
1374 |
bdrv_delete(bs); |
1375 |
} |
1376 |
|
1377 |
ret = do_sd_create(s, vdi, vdi_size, base_vid, &vid, 0);
|
1378 |
if (!prealloc || ret) {
|
1379 |
goto out;
|
1380 |
} |
1381 |
|
1382 |
ret = sd_prealloc(filename); |
1383 |
out:
|
1384 |
g_free(s); |
1385 |
return ret;
|
1386 |
} |
1387 |
|
1388 |
static void sd_close(BlockDriverState *bs) |
1389 |
{ |
1390 |
BDRVSheepdogState *s = bs->opaque; |
1391 |
SheepdogVdiReq hdr; |
1392 |
SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; |
1393 |
unsigned int wlen, rlen = 0; |
1394 |
int fd, ret;
|
1395 |
|
1396 |
dprintf("%s\n", s->name);
|
1397 |
|
1398 |
fd = connect_to_sdog(s); |
1399 |
if (fd < 0) { |
1400 |
return;
|
1401 |
} |
1402 |
|
1403 |
memset(&hdr, 0, sizeof(hdr)); |
1404 |
|
1405 |
hdr.opcode = SD_OP_RELEASE_VDI; |
1406 |
hdr.vdi_id = s->inode.vdi_id; |
1407 |
wlen = strlen(s->name) + 1;
|
1408 |
hdr.data_length = wlen; |
1409 |
hdr.flags = SD_FLAG_CMD_WRITE; |
1410 |
|
1411 |
ret = do_req(fd, (SheepdogReq *)&hdr, s->name, &wlen, &rlen); |
1412 |
|
1413 |
closesocket(fd); |
1414 |
|
1415 |
if (!ret && rsp->result != SD_RES_SUCCESS &&
|
1416 |
rsp->result != SD_RES_VDI_NOT_LOCKED) { |
1417 |
error_report("%s, %s", sd_strerror(rsp->result), s->name);
|
1418 |
} |
1419 |
|
1420 |
qemu_aio_set_fd_handler(s->fd, NULL, NULL, NULL, NULL); |
1421 |
closesocket(s->fd); |
1422 |
g_free(s->host_spec); |
1423 |
} |
1424 |
|
1425 |
static int64_t sd_getlength(BlockDriverState *bs)
|
1426 |
{ |
1427 |
BDRVSheepdogState *s = bs->opaque; |
1428 |
|
1429 |
return s->inode.vdi_size;
|
1430 |
} |
1431 |
|
1432 |
static int sd_truncate(BlockDriverState *bs, int64_t offset) |
1433 |
{ |
1434 |
BDRVSheepdogState *s = bs->opaque; |
1435 |
int ret, fd;
|
1436 |
unsigned int datalen; |
1437 |
|
1438 |
if (offset < s->inode.vdi_size) {
|
1439 |
error_report("shrinking is not supported");
|
1440 |
return -EINVAL;
|
1441 |
} else if (offset > SD_MAX_VDI_SIZE) { |
1442 |
error_report("too big image size");
|
1443 |
return -EINVAL;
|
1444 |
} |
1445 |
|
1446 |
fd = connect_to_sdog(s); |
1447 |
if (fd < 0) { |
1448 |
return fd;
|
1449 |
} |
1450 |
|
1451 |
/* we don't need to update entire object */
|
1452 |
datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
|
1453 |
s->inode.vdi_size = offset; |
1454 |
ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
|
1455 |
s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
1456 |
close(fd); |
1457 |
|
1458 |
if (ret < 0) { |
1459 |
error_report("failed to update an inode.");
|
1460 |
} |
1461 |
|
1462 |
return ret;
|
1463 |
} |
1464 |
|
1465 |
/*
|
1466 |
* This function is called after writing data objects. If we need to
|
1467 |
* update metadata, this sends a write request to the vdi object.
|
1468 |
* Otherwise, this switches back to sd_co_readv/writev.
|
1469 |
*/
|
1470 |
static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) |
1471 |
{ |
1472 |
int ret;
|
1473 |
BDRVSheepdogState *s = acb->common.bs->opaque; |
1474 |
struct iovec iov;
|
1475 |
AIOReq *aio_req; |
1476 |
uint32_t offset, data_len, mn, mx; |
1477 |
|
1478 |
mn = s->min_dirty_data_idx; |
1479 |
mx = s->max_dirty_data_idx; |
1480 |
if (mn <= mx) {
|
1481 |
/* we need to update the vdi object. */
|
1482 |
offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + |
1483 |
mn * sizeof(s->inode.data_vdi_id[0]); |
1484 |
data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); |
1485 |
|
1486 |
s->min_dirty_data_idx = UINT32_MAX; |
1487 |
s->max_dirty_data_idx = 0;
|
1488 |
|
1489 |
iov.iov_base = &s->inode; |
1490 |
iov.iov_len = sizeof(s->inode);
|
1491 |
aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
1492 |
data_len, offset, 0, 0, offset); |
1493 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1494 |
ret = add_aio_request(s, aio_req, &iov, 1, false, AIOCB_WRITE_UDATA); |
1495 |
if (ret) {
|
1496 |
free_aio_req(s, aio_req); |
1497 |
acb->ret = -EIO; |
1498 |
goto out;
|
1499 |
} |
1500 |
|
1501 |
acb->aio_done_func = sd_finish_aiocb; |
1502 |
acb->aiocb_type = AIOCB_WRITE_UDATA; |
1503 |
return;
|
1504 |
} |
1505 |
out:
|
1506 |
sd_finish_aiocb(acb); |
1507 |
} |
1508 |
|
1509 |
/*
|
1510 |
* Create a writable VDI from a snapshot
|
1511 |
*/
|
1512 |
static int sd_create_branch(BDRVSheepdogState *s) |
1513 |
{ |
1514 |
int ret, fd;
|
1515 |
uint32_t vid; |
1516 |
char *buf;
|
1517 |
|
1518 |
dprintf("%" PRIx32 " is snapshot.\n", s->inode.vdi_id); |
1519 |
|
1520 |
buf = g_malloc(SD_INODE_SIZE); |
1521 |
|
1522 |
ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &vid, 1);
|
1523 |
if (ret) {
|
1524 |
goto out;
|
1525 |
} |
1526 |
|
1527 |
dprintf("%" PRIx32 " is created.\n", vid); |
1528 |
|
1529 |
fd = connect_to_sdog(s); |
1530 |
if (fd < 0) { |
1531 |
ret = fd; |
1532 |
goto out;
|
1533 |
} |
1534 |
|
1535 |
ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
1536 |
SD_INODE_SIZE, 0, s->cache_flags);
|
1537 |
|
1538 |
closesocket(fd); |
1539 |
|
1540 |
if (ret < 0) { |
1541 |
goto out;
|
1542 |
} |
1543 |
|
1544 |
memcpy(&s->inode, buf, sizeof(s->inode));
|
1545 |
|
1546 |
s->is_snapshot = false;
|
1547 |
ret = 0;
|
1548 |
dprintf("%" PRIx32 " was newly created.\n", s->inode.vdi_id); |
1549 |
|
1550 |
out:
|
1551 |
g_free(buf); |
1552 |
|
1553 |
return ret;
|
1554 |
} |
1555 |
|
1556 |
/*
|
1557 |
* Send I/O requests to the server.
|
1558 |
*
|
1559 |
* This function sends requests to the server, links the requests to
|
1560 |
* the inflight_list in BDRVSheepdogState, and exits without
|
1561 |
* waiting the response. The responses are received in the
|
1562 |
* `aio_read_response' function which is called from the main loop as
|
1563 |
* a fd handler.
|
1564 |
*
|
1565 |
* Returns 1 when we need to wait a response, 0 when there is no sent
|
1566 |
* request and -errno in error cases.
|
1567 |
*/
|
1568 |
static int coroutine_fn sd_co_rw_vector(void *p) |
1569 |
{ |
1570 |
SheepdogAIOCB *acb = p; |
1571 |
int ret = 0; |
1572 |
unsigned long len, done = 0, total = acb->nb_sectors * SECTOR_SIZE; |
1573 |
unsigned long idx = acb->sector_num * SECTOR_SIZE / SD_DATA_OBJ_SIZE; |
1574 |
uint64_t oid; |
1575 |
uint64_t offset = (acb->sector_num * SECTOR_SIZE) % SD_DATA_OBJ_SIZE; |
1576 |
BDRVSheepdogState *s = acb->common.bs->opaque; |
1577 |
SheepdogInode *inode = &s->inode; |
1578 |
AIOReq *aio_req; |
1579 |
|
1580 |
if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
|
1581 |
/*
|
1582 |
* In the case we open the snapshot VDI, Sheepdog creates the
|
1583 |
* writable VDI when we do a write operation first.
|
1584 |
*/
|
1585 |
ret = sd_create_branch(s); |
1586 |
if (ret) {
|
1587 |
acb->ret = -EIO; |
1588 |
goto out;
|
1589 |
} |
1590 |
} |
1591 |
|
1592 |
/*
|
1593 |
* Make sure we don't free the aiocb before we are done with all requests.
|
1594 |
* This additional reference is dropped at the end of this function.
|
1595 |
*/
|
1596 |
acb->nr_pending++; |
1597 |
|
1598 |
while (done != total) {
|
1599 |
uint8_t flags = 0;
|
1600 |
uint64_t old_oid = 0;
|
1601 |
bool create = false; |
1602 |
|
1603 |
oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); |
1604 |
|
1605 |
len = MIN(total - done, SD_DATA_OBJ_SIZE - offset); |
1606 |
|
1607 |
switch (acb->aiocb_type) {
|
1608 |
case AIOCB_READ_UDATA:
|
1609 |
if (!inode->data_vdi_id[idx]) {
|
1610 |
qemu_iovec_memset(acb->qiov, done, 0, len);
|
1611 |
goto done;
|
1612 |
} |
1613 |
break;
|
1614 |
case AIOCB_WRITE_UDATA:
|
1615 |
if (!inode->data_vdi_id[idx]) {
|
1616 |
create = true;
|
1617 |
} else if (!is_data_obj_writable(inode, idx)) { |
1618 |
/* Copy-On-Write */
|
1619 |
create = true;
|
1620 |
old_oid = oid; |
1621 |
flags = SD_FLAG_CMD_COW; |
1622 |
} |
1623 |
break;
|
1624 |
default:
|
1625 |
break;
|
1626 |
} |
1627 |
|
1628 |
if (create) {
|
1629 |
dprintf("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n", |
1630 |
inode->vdi_id, oid, |
1631 |
vid_to_data_oid(inode->data_vdi_id[idx], idx), idx); |
1632 |
oid = vid_to_data_oid(inode->vdi_id, idx); |
1633 |
dprintf("new oid %" PRIx64 "\n", oid); |
1634 |
} |
1635 |
|
1636 |
aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, old_oid, done); |
1637 |
|
1638 |
if (create) {
|
1639 |
AIOReq *areq; |
1640 |
QLIST_FOREACH(areq, &s->inflight_aio_head, aio_siblings) { |
1641 |
if (areq->oid == oid) {
|
1642 |
/*
|
1643 |
* Sheepdog cannot handle simultaneous create
|
1644 |
* requests to the same object. So we cannot send
|
1645 |
* the request until the previous request
|
1646 |
* finishes.
|
1647 |
*/
|
1648 |
aio_req->flags = 0;
|
1649 |
aio_req->base_oid = 0;
|
1650 |
QLIST_INSERT_HEAD(&s->pending_aio_head, aio_req, |
1651 |
aio_siblings); |
1652 |
goto done;
|
1653 |
} |
1654 |
} |
1655 |
} |
1656 |
|
1657 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1658 |
ret = add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, |
1659 |
create, acb->aiocb_type); |
1660 |
if (ret < 0) { |
1661 |
error_report("add_aio_request is failed");
|
1662 |
free_aio_req(s, aio_req); |
1663 |
acb->ret = -EIO; |
1664 |
goto out;
|
1665 |
} |
1666 |
done:
|
1667 |
offset = 0;
|
1668 |
idx++; |
1669 |
done += len; |
1670 |
} |
1671 |
out:
|
1672 |
if (!--acb->nr_pending) {
|
1673 |
return acb->ret;
|
1674 |
} |
1675 |
return 1; |
1676 |
} |
1677 |
|
1678 |
static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, |
1679 |
int nb_sectors, QEMUIOVector *qiov)
|
1680 |
{ |
1681 |
SheepdogAIOCB *acb; |
1682 |
int ret;
|
1683 |
|
1684 |
if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
|
1685 |
ret = sd_truncate(bs, (sector_num + nb_sectors) * SECTOR_SIZE); |
1686 |
if (ret < 0) { |
1687 |
return ret;
|
1688 |
} |
1689 |
bs->total_sectors = sector_num + nb_sectors; |
1690 |
} |
1691 |
|
1692 |
acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
1693 |
acb->aio_done_func = sd_write_done; |
1694 |
acb->aiocb_type = AIOCB_WRITE_UDATA; |
1695 |
|
1696 |
ret = sd_co_rw_vector(acb); |
1697 |
if (ret <= 0) { |
1698 |
qemu_aio_release(acb); |
1699 |
return ret;
|
1700 |
} |
1701 |
|
1702 |
qemu_coroutine_yield(); |
1703 |
|
1704 |
return acb->ret;
|
1705 |
} |
1706 |
|
1707 |
static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, |
1708 |
int nb_sectors, QEMUIOVector *qiov)
|
1709 |
{ |
1710 |
SheepdogAIOCB *acb; |
1711 |
int ret;
|
1712 |
|
1713 |
acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors); |
1714 |
acb->aiocb_type = AIOCB_READ_UDATA; |
1715 |
acb->aio_done_func = sd_finish_aiocb; |
1716 |
|
1717 |
ret = sd_co_rw_vector(acb); |
1718 |
if (ret <= 0) { |
1719 |
qemu_aio_release(acb); |
1720 |
return ret;
|
1721 |
} |
1722 |
|
1723 |
qemu_coroutine_yield(); |
1724 |
|
1725 |
return acb->ret;
|
1726 |
} |
1727 |
|
1728 |
static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) |
1729 |
{ |
1730 |
BDRVSheepdogState *s = bs->opaque; |
1731 |
SheepdogAIOCB *acb; |
1732 |
AIOReq *aio_req; |
1733 |
int ret;
|
1734 |
|
1735 |
if (s->cache_flags != SD_FLAG_CMD_CACHE) {
|
1736 |
return 0; |
1737 |
} |
1738 |
|
1739 |
acb = sd_aio_setup(bs, NULL, 0, 0); |
1740 |
acb->aiocb_type = AIOCB_FLUSH_CACHE; |
1741 |
acb->aio_done_func = sd_finish_aiocb; |
1742 |
|
1743 |
aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), |
1744 |
0, 0, 0, 0, 0); |
1745 |
QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); |
1746 |
ret = add_aio_request(s, aio_req, NULL, 0, false, acb->aiocb_type); |
1747 |
if (ret < 0) { |
1748 |
error_report("add_aio_request is failed");
|
1749 |
free_aio_req(s, aio_req); |
1750 |
qemu_aio_release(acb); |
1751 |
return ret;
|
1752 |
} |
1753 |
|
1754 |
qemu_coroutine_yield(); |
1755 |
return acb->ret;
|
1756 |
} |
1757 |
|
1758 |
static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) |
1759 |
{ |
1760 |
BDRVSheepdogState *s = bs->opaque; |
1761 |
int ret, fd;
|
1762 |
uint32_t new_vid; |
1763 |
SheepdogInode *inode; |
1764 |
unsigned int datalen; |
1765 |
|
1766 |
dprintf("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " " |
1767 |
"is_snapshot %d\n", sn_info->name, sn_info->id_str,
|
1768 |
s->name, sn_info->vm_state_size, s->is_snapshot); |
1769 |
|
1770 |
if (s->is_snapshot) {
|
1771 |
error_report("You can't create a snapshot of a snapshot VDI, "
|
1772 |
"%s (%" PRIu32 ").", s->name, s->inode.vdi_id); |
1773 |
|
1774 |
return -EINVAL;
|
1775 |
} |
1776 |
|
1777 |
dprintf("%s %s\n", sn_info->name, sn_info->id_str);
|
1778 |
|
1779 |
s->inode.vm_state_size = sn_info->vm_state_size; |
1780 |
s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; |
1781 |
/* It appears that inode.tag does not require a NUL terminator,
|
1782 |
* which means this use of strncpy is ok.
|
1783 |
*/
|
1784 |
strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
|
1785 |
/* we don't need to update entire object */
|
1786 |
datalen = SD_INODE_SIZE - sizeof(s->inode.data_vdi_id);
|
1787 |
|
1788 |
/* refresh inode. */
|
1789 |
fd = connect_to_sdog(s); |
1790 |
if (fd < 0) { |
1791 |
ret = fd; |
1792 |
goto cleanup;
|
1793 |
} |
1794 |
|
1795 |
ret = write_object(fd, (char *)&s->inode, vid_to_vdi_oid(s->inode.vdi_id),
|
1796 |
s->inode.nr_copies, datalen, 0, false, s->cache_flags); |
1797 |
if (ret < 0) { |
1798 |
error_report("failed to write snapshot's inode.");
|
1799 |
goto cleanup;
|
1800 |
} |
1801 |
|
1802 |
ret = do_sd_create(s, s->name, s->inode.vdi_size, s->inode.vdi_id, &new_vid, |
1803 |
1);
|
1804 |
if (ret < 0) { |
1805 |
error_report("failed to create inode for snapshot. %s",
|
1806 |
strerror(errno)); |
1807 |
goto cleanup;
|
1808 |
} |
1809 |
|
1810 |
inode = (SheepdogInode *)g_malloc(datalen); |
1811 |
|
1812 |
ret = read_object(fd, (char *)inode, vid_to_vdi_oid(new_vid),
|
1813 |
s->inode.nr_copies, datalen, 0, s->cache_flags);
|
1814 |
|
1815 |
if (ret < 0) { |
1816 |
error_report("failed to read new inode info. %s", strerror(errno));
|
1817 |
goto cleanup;
|
1818 |
} |
1819 |
|
1820 |
memcpy(&s->inode, inode, datalen); |
1821 |
dprintf("s->inode: name %s snap_id %x oid %x\n",
|
1822 |
s->inode.name, s->inode.snap_id, s->inode.vdi_id); |
1823 |
|
1824 |
cleanup:
|
1825 |
closesocket(fd); |
1826 |
return ret;
|
1827 |
} |
1828 |
|
1829 |
static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) |
1830 |
{ |
1831 |
BDRVSheepdogState *s = bs->opaque; |
1832 |
BDRVSheepdogState *old_s; |
1833 |
char vdi[SD_MAX_VDI_LEN], tag[SD_MAX_VDI_TAG_LEN];
|
1834 |
char *buf = NULL; |
1835 |
uint32_t vid; |
1836 |
uint32_t snapid = 0;
|
1837 |
int ret = 0, fd; |
1838 |
|
1839 |
old_s = g_malloc(sizeof(BDRVSheepdogState));
|
1840 |
|
1841 |
memcpy(old_s, s, sizeof(BDRVSheepdogState));
|
1842 |
|
1843 |
pstrcpy(vdi, sizeof(vdi), s->name);
|
1844 |
|
1845 |
snapid = strtoul(snapshot_id, NULL, 10); |
1846 |
if (snapid) {
|
1847 |
tag[0] = 0; |
1848 |
} else {
|
1849 |
pstrcpy(tag, sizeof(tag), s->name);
|
1850 |
} |
1851 |
|
1852 |
ret = find_vdi_name(s, vdi, snapid, tag, &vid, 1);
|
1853 |
if (ret) {
|
1854 |
error_report("Failed to find_vdi_name");
|
1855 |
goto out;
|
1856 |
} |
1857 |
|
1858 |
fd = connect_to_sdog(s); |
1859 |
if (fd < 0) { |
1860 |
ret = fd; |
1861 |
goto out;
|
1862 |
} |
1863 |
|
1864 |
buf = g_malloc(SD_INODE_SIZE); |
1865 |
ret = read_object(fd, buf, vid_to_vdi_oid(vid), s->inode.nr_copies, |
1866 |
SD_INODE_SIZE, 0, s->cache_flags);
|
1867 |
|
1868 |
closesocket(fd); |
1869 |
|
1870 |
if (ret) {
|
1871 |
goto out;
|
1872 |
} |
1873 |
|
1874 |
memcpy(&s->inode, buf, sizeof(s->inode));
|
1875 |
|
1876 |
if (!s->inode.vm_state_size) {
|
1877 |
error_report("Invalid snapshot");
|
1878 |
ret = -ENOENT; |
1879 |
goto out;
|
1880 |
} |
1881 |
|
1882 |
s->is_snapshot = true;
|
1883 |
|
1884 |
g_free(buf); |
1885 |
g_free(old_s); |
1886 |
|
1887 |
return 0; |
1888 |
out:
|
1889 |
/* recover bdrv_sd_state */
|
1890 |
memcpy(s, old_s, sizeof(BDRVSheepdogState));
|
1891 |
g_free(buf); |
1892 |
g_free(old_s); |
1893 |
|
1894 |
error_report("failed to open. recover old bdrv_sd_state.");
|
1895 |
|
1896 |
return ret;
|
1897 |
} |
1898 |
|
1899 |
static int sd_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) |
1900 |
{ |
1901 |
/* FIXME: Delete specified snapshot id. */
|
1902 |
return 0; |
1903 |
} |
1904 |
|
1905 |
static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) |
1906 |
{ |
1907 |
BDRVSheepdogState *s = bs->opaque; |
1908 |
SheepdogReq req; |
1909 |
int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); |
1910 |
QEMUSnapshotInfo *sn_tab = NULL;
|
1911 |
unsigned wlen, rlen;
|
1912 |
int found = 0; |
1913 |
static SheepdogInode inode;
|
1914 |
unsigned long *vdi_inuse; |
1915 |
unsigned int start_nr; |
1916 |
uint64_t hval; |
1917 |
uint32_t vid; |
1918 |
|
1919 |
vdi_inuse = g_malloc(max); |
1920 |
|
1921 |
fd = connect_to_sdog(s); |
1922 |
if (fd < 0) { |
1923 |
ret = fd; |
1924 |
goto out;
|
1925 |
} |
1926 |
|
1927 |
rlen = max; |
1928 |
wlen = 0;
|
1929 |
|
1930 |
memset(&req, 0, sizeof(req)); |
1931 |
|
1932 |
req.opcode = SD_OP_READ_VDIS; |
1933 |
req.data_length = max; |
1934 |
|
1935 |
ret = do_req(fd, (SheepdogReq *)&req, vdi_inuse, &wlen, &rlen); |
1936 |
|
1937 |
closesocket(fd); |
1938 |
if (ret) {
|
1939 |
goto out;
|
1940 |
} |
1941 |
|
1942 |
sn_tab = g_malloc0(nr * sizeof(*sn_tab));
|
1943 |
|
1944 |
/* calculate a vdi id with hash function */
|
1945 |
hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); |
1946 |
start_nr = hval & (SD_NR_VDIS - 1);
|
1947 |
|
1948 |
fd = connect_to_sdog(s); |
1949 |
if (fd < 0) { |
1950 |
ret = fd; |
1951 |
goto out;
|
1952 |
} |
1953 |
|
1954 |
for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { |
1955 |
if (!test_bit(vid, vdi_inuse)) {
|
1956 |
break;
|
1957 |
} |
1958 |
|
1959 |
/* we don't need to read entire object */
|
1960 |
ret = read_object(fd, (char *)&inode, vid_to_vdi_oid(vid),
|
1961 |
0, SD_INODE_SIZE - sizeof(inode.data_vdi_id), 0, |
1962 |
s->cache_flags); |
1963 |
|
1964 |
if (ret) {
|
1965 |
continue;
|
1966 |
} |
1967 |
|
1968 |
if (!strcmp(inode.name, s->name) && is_snapshot(&inode)) {
|
1969 |
sn_tab[found].date_sec = inode.snap_ctime >> 32;
|
1970 |
sn_tab[found].date_nsec = inode.snap_ctime & 0xffffffff;
|
1971 |
sn_tab[found].vm_state_size = inode.vm_state_size; |
1972 |
sn_tab[found].vm_clock_nsec = inode.vm_clock_nsec; |
1973 |
|
1974 |
snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), "%u", |
1975 |
inode.snap_id); |
1976 |
pstrcpy(sn_tab[found].name, |
1977 |
MIN(sizeof(sn_tab[found].name), sizeof(inode.tag)), |
1978 |
inode.tag); |
1979 |
found++; |
1980 |
} |
1981 |
} |
1982 |
|
1983 |
closesocket(fd); |
1984 |
out:
|
1985 |
*psn_tab = sn_tab; |
1986 |
|
1987 |
g_free(vdi_inuse); |
1988 |
|
1989 |
if (ret < 0) { |
1990 |
return ret;
|
1991 |
} |
1992 |
|
1993 |
return found;
|
1994 |
} |
1995 |
|
1996 |
static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, |
1997 |
int64_t pos, int size, int load) |
1998 |
{ |
1999 |
bool create;
|
2000 |
int fd, ret = 0, remaining = size; |
2001 |
unsigned int data_len; |
2002 |
uint64_t vmstate_oid; |
2003 |
uint32_t vdi_index; |
2004 |
uint64_t offset; |
2005 |
|
2006 |
fd = connect_to_sdog(s); |
2007 |
if (fd < 0) { |
2008 |
return fd;
|
2009 |
} |
2010 |
|
2011 |
while (remaining) {
|
2012 |
vdi_index = pos / SD_DATA_OBJ_SIZE; |
2013 |
offset = pos % SD_DATA_OBJ_SIZE; |
2014 |
|
2015 |
data_len = MIN(remaining, SD_DATA_OBJ_SIZE - offset); |
2016 |
|
2017 |
vmstate_oid = vid_to_vmstate_oid(s->inode.vdi_id, vdi_index); |
2018 |
|
2019 |
create = (offset == 0);
|
2020 |
if (load) {
|
2021 |
ret = read_object(fd, (char *)data, vmstate_oid,
|
2022 |
s->inode.nr_copies, data_len, offset, |
2023 |
s->cache_flags); |
2024 |
} else {
|
2025 |
ret = write_object(fd, (char *)data, vmstate_oid,
|
2026 |
s->inode.nr_copies, data_len, offset, create, |
2027 |
s->cache_flags); |
2028 |
} |
2029 |
|
2030 |
if (ret < 0) { |
2031 |
error_report("failed to save vmstate %s", strerror(errno));
|
2032 |
goto cleanup;
|
2033 |
} |
2034 |
|
2035 |
pos += data_len; |
2036 |
data += data_len; |
2037 |
remaining -= data_len; |
2038 |
} |
2039 |
ret = size; |
2040 |
cleanup:
|
2041 |
closesocket(fd); |
2042 |
return ret;
|
2043 |
} |
2044 |
|
2045 |
static int sd_save_vmstate(BlockDriverState *bs, const uint8_t *data, |
2046 |
int64_t pos, int size)
|
2047 |
{ |
2048 |
BDRVSheepdogState *s = bs->opaque; |
2049 |
|
2050 |
return do_load_save_vmstate(s, (uint8_t *)data, pos, size, 0); |
2051 |
} |
2052 |
|
2053 |
static int sd_load_vmstate(BlockDriverState *bs, uint8_t *data, |
2054 |
int64_t pos, int size)
|
2055 |
{ |
2056 |
BDRVSheepdogState *s = bs->opaque; |
2057 |
|
2058 |
return do_load_save_vmstate(s, data, pos, size, 1); |
2059 |
} |
2060 |
|
2061 |
|
2062 |
static QEMUOptionParameter sd_create_options[] = {
|
2063 |
{ |
2064 |
.name = BLOCK_OPT_SIZE, |
2065 |
.type = OPT_SIZE, |
2066 |
.help = "Virtual disk size"
|
2067 |
}, |
2068 |
{ |
2069 |
.name = BLOCK_OPT_BACKING_FILE, |
2070 |
.type = OPT_STRING, |
2071 |
.help = "File name of a base image"
|
2072 |
}, |
2073 |
{ |
2074 |
.name = BLOCK_OPT_PREALLOC, |
2075 |
.type = OPT_STRING, |
2076 |
.help = "Preallocation mode (allowed values: off, full)"
|
2077 |
}, |
2078 |
{ NULL }
|
2079 |
}; |
2080 |
|
2081 |
static BlockDriver bdrv_sheepdog = {
|
2082 |
.format_name = "sheepdog",
|
2083 |
.protocol_name = "sheepdog",
|
2084 |
.instance_size = sizeof(BDRVSheepdogState),
|
2085 |
.bdrv_file_open = sd_open, |
2086 |
.bdrv_close = sd_close, |
2087 |
.bdrv_create = sd_create, |
2088 |
.bdrv_getlength = sd_getlength, |
2089 |
.bdrv_truncate = sd_truncate, |
2090 |
|
2091 |
.bdrv_co_readv = sd_co_readv, |
2092 |
.bdrv_co_writev = sd_co_writev, |
2093 |
.bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
2094 |
|
2095 |
.bdrv_snapshot_create = sd_snapshot_create, |
2096 |
.bdrv_snapshot_goto = sd_snapshot_goto, |
2097 |
.bdrv_snapshot_delete = sd_snapshot_delete, |
2098 |
.bdrv_snapshot_list = sd_snapshot_list, |
2099 |
|
2100 |
.bdrv_save_vmstate = sd_save_vmstate, |
2101 |
.bdrv_load_vmstate = sd_load_vmstate, |
2102 |
|
2103 |
.create_options = sd_create_options, |
2104 |
}; |
2105 |
|
2106 |
static BlockDriver bdrv_sheepdog_tcp = {
|
2107 |
.format_name = "sheepdog",
|
2108 |
.protocol_name = "sheepdog+tcp",
|
2109 |
.instance_size = sizeof(BDRVSheepdogState),
|
2110 |
.bdrv_file_open = sd_open, |
2111 |
.bdrv_close = sd_close, |
2112 |
.bdrv_create = sd_create, |
2113 |
.bdrv_getlength = sd_getlength, |
2114 |
.bdrv_truncate = sd_truncate, |
2115 |
|
2116 |
.bdrv_co_readv = sd_co_readv, |
2117 |
.bdrv_co_writev = sd_co_writev, |
2118 |
.bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
2119 |
|
2120 |
.bdrv_snapshot_create = sd_snapshot_create, |
2121 |
.bdrv_snapshot_goto = sd_snapshot_goto, |
2122 |
.bdrv_snapshot_delete = sd_snapshot_delete, |
2123 |
.bdrv_snapshot_list = sd_snapshot_list, |
2124 |
|
2125 |
.bdrv_save_vmstate = sd_save_vmstate, |
2126 |
.bdrv_load_vmstate = sd_load_vmstate, |
2127 |
|
2128 |
.create_options = sd_create_options, |
2129 |
}; |
2130 |
|
2131 |
static BlockDriver bdrv_sheepdog_unix = {
|
2132 |
.format_name = "sheepdog",
|
2133 |
.protocol_name = "sheepdog+unix",
|
2134 |
.instance_size = sizeof(BDRVSheepdogState),
|
2135 |
.bdrv_file_open = sd_open, |
2136 |
.bdrv_close = sd_close, |
2137 |
.bdrv_create = sd_create, |
2138 |
.bdrv_getlength = sd_getlength, |
2139 |
.bdrv_truncate = sd_truncate, |
2140 |
|
2141 |
.bdrv_co_readv = sd_co_readv, |
2142 |
.bdrv_co_writev = sd_co_writev, |
2143 |
.bdrv_co_flush_to_disk = sd_co_flush_to_disk, |
2144 |
|
2145 |
.bdrv_snapshot_create = sd_snapshot_create, |
2146 |
.bdrv_snapshot_goto = sd_snapshot_goto, |
2147 |
.bdrv_snapshot_delete = sd_snapshot_delete, |
2148 |
.bdrv_snapshot_list = sd_snapshot_list, |
2149 |
|
2150 |
.bdrv_save_vmstate = sd_save_vmstate, |
2151 |
.bdrv_load_vmstate = sd_load_vmstate, |
2152 |
|
2153 |
.create_options = sd_create_options, |
2154 |
}; |
2155 |
|
2156 |
static void bdrv_sheepdog_init(void) |
2157 |
{ |
2158 |
bdrv_register(&bdrv_sheepdog); |
2159 |
bdrv_register(&bdrv_sheepdog_tcp); |
2160 |
bdrv_register(&bdrv_sheepdog_unix); |
2161 |
} |
2162 |
block_init(bdrv_sheepdog_init); |