]>
Commit | Line | Data |
---|---|---|
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/osdep.h" | |
16 | #include "qemu-common.h" | |
17 | #include "qapi/error.h" | |
18 | #include "qapi/qapi-visit-sockets.h" | |
19 | #include "qapi/qapi-visit-block-core.h" | |
20 | #include "qapi/qmp/qdict.h" | |
21 | #include "qapi/qobject-input-visitor.h" | |
22 | #include "qapi/qobject-output-visitor.h" | |
23 | #include "qemu/uri.h" | |
24 | #include "qemu/error-report.h" | |
25 | #include "qemu/module.h" | |
26 | #include "qemu/option.h" | |
27 | #include "qemu/sockets.h" | |
28 | #include "block/block_int.h" | |
29 | #include "block/qdict.h" | |
30 | #include "sysemu/block-backend.h" | |
31 | #include "qemu/bitops.h" | |
32 | #include "qemu/cutils.h" | |
33 | #include "trace.h" | |
34 | ||
35 | #define SD_PROTO_VER 0x01 | |
36 | ||
37 | #define SD_DEFAULT_ADDR "localhost" | |
38 | #define SD_DEFAULT_PORT 7000 | |
39 | ||
40 | #define SD_OP_CREATE_AND_WRITE_OBJ 0x01 | |
41 | #define SD_OP_READ_OBJ 0x02 | |
42 | #define SD_OP_WRITE_OBJ 0x03 | |
43 | /* 0x04 is used internally by Sheepdog */ | |
44 | ||
45 | #define SD_OP_NEW_VDI 0x11 | |
46 | #define SD_OP_LOCK_VDI 0x12 | |
47 | #define SD_OP_RELEASE_VDI 0x13 | |
48 | #define SD_OP_GET_VDI_INFO 0x14 | |
49 | #define SD_OP_READ_VDIS 0x15 | |
50 | #define SD_OP_FLUSH_VDI 0x16 | |
51 | #define SD_OP_DEL_VDI 0x17 | |
52 | #define SD_OP_GET_CLUSTER_DEFAULT 0x18 | |
53 | ||
54 | #define SD_FLAG_CMD_WRITE 0x01 | |
55 | #define SD_FLAG_CMD_COW 0x02 | |
56 | #define SD_FLAG_CMD_CACHE 0x04 /* Writeback mode for cache */ | |
57 | #define SD_FLAG_CMD_DIRECT 0x08 /* Don't use cache */ | |
58 | ||
59 | #define SD_RES_SUCCESS 0x00 /* Success */ | |
60 | #define SD_RES_UNKNOWN 0x01 /* Unknown error */ | |
61 | #define SD_RES_NO_OBJ 0x02 /* No object found */ | |
62 | #define SD_RES_EIO 0x03 /* I/O error */ | |
63 | #define SD_RES_VDI_EXIST 0x04 /* Vdi exists already */ | |
64 | #define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */ | |
65 | #define SD_RES_SYSTEM_ERROR 0x06 /* System error */ | |
66 | #define SD_RES_VDI_LOCKED 0x07 /* Vdi is locked */ | |
67 | #define SD_RES_NO_VDI 0x08 /* No vdi found */ | |
68 | #define SD_RES_NO_BASE_VDI 0x09 /* No base vdi found */ | |
69 | #define SD_RES_VDI_READ 0x0A /* Cannot read requested vdi */ | |
70 | #define SD_RES_VDI_WRITE 0x0B /* Cannot write requested vdi */ | |
71 | #define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */ | |
72 | #define SD_RES_BASE_VDI_WRITE 0x0D /* Cannot write base vdi */ | |
73 | #define SD_RES_NO_TAG 0x0E /* Requested tag is not found */ | |
74 | #define SD_RES_STARTUP 0x0F /* Sheepdog is on starting up */ | |
75 | #define SD_RES_VDI_NOT_LOCKED 0x10 /* Vdi is not locked */ | |
76 | #define SD_RES_SHUTDOWN 0x11 /* Sheepdog is shutting down */ | |
77 | #define SD_RES_NO_MEM 0x12 /* Cannot allocate memory */ | |
78 | #define SD_RES_FULL_VDI 0x13 /* we already have the maximum vdis */ | |
79 | #define SD_RES_VER_MISMATCH 0x14 /* Protocol version mismatch */ | |
80 | #define SD_RES_NO_SPACE 0x15 /* Server has no room for new objects */ | |
81 | #define SD_RES_WAIT_FOR_FORMAT 0x16 /* Waiting for a format operation */ | |
82 | #define SD_RES_WAIT_FOR_JOIN 0x17 /* Waiting for other nodes joining */ | |
83 | #define SD_RES_JOIN_FAILED 0x18 /* Target node had failed to join sheepdog */ | |
84 | #define SD_RES_HALT 0x19 /* Sheepdog is stopped serving IO request */ | |
85 | #define SD_RES_READONLY 0x1A /* Object is read-only */ | |
86 | ||
87 | /* | |
88 | * Object ID rules | |
89 | * | |
90 | * 0 - 19 (20 bits): data object space | |
91 | * 20 - 31 (12 bits): reserved data object space | |
92 | * 32 - 55 (24 bits): vdi object space | |
93 | * 56 - 59 ( 4 bits): reserved vdi object space | |
94 | * 60 - 63 ( 4 bits): object type identifier space | |
95 | */ | |
96 | ||
97 | #define VDI_SPACE_SHIFT 32 | |
98 | #define VDI_BIT (UINT64_C(1) << 63) | |
99 | #define VMSTATE_BIT (UINT64_C(1) << 62) | |
100 | #define MAX_DATA_OBJS (UINT64_C(1) << 20) | |
101 | #define MAX_CHILDREN 1024 | |
102 | #define SD_MAX_VDI_LEN 256 | |
103 | #define SD_MAX_VDI_TAG_LEN 256 | |
104 | #define SD_NR_VDIS (1U << 24) | |
105 | #define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22) | |
106 | #define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS) | |
107 | #define SD_DEFAULT_BLOCK_SIZE_SHIFT 22 | |
108 | /* | |
109 | * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and | |
110 | * (SD_EC_MAX_STRIP - 1) for parity strips | |
111 | * | |
112 | * SD_MAX_COPIES is sum of number of data strips and parity strips. | |
113 | */ | |
114 | #define SD_EC_MAX_STRIP 16 | |
115 | #define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1) | |
116 | ||
117 | #define SD_INODE_SIZE (sizeof(SheepdogInode)) | |
118 | #define CURRENT_VDI_ID 0 | |
119 | ||
120 | #define LOCK_TYPE_NORMAL 0 | |
121 | #define LOCK_TYPE_SHARED 1 /* for iSCSI multipath */ | |
122 | ||
123 | typedef struct SheepdogReq { | |
124 | uint8_t proto_ver; | |
125 | uint8_t opcode; | |
126 | uint16_t flags; | |
127 | uint32_t epoch; | |
128 | uint32_t id; | |
129 | uint32_t data_length; | |
130 | uint32_t opcode_specific[8]; | |
131 | } SheepdogReq; | |
132 | ||
133 | typedef struct SheepdogRsp { | |
134 | uint8_t proto_ver; | |
135 | uint8_t opcode; | |
136 | uint16_t flags; | |
137 | uint32_t epoch; | |
138 | uint32_t id; | |
139 | uint32_t data_length; | |
140 | uint32_t result; | |
141 | uint32_t opcode_specific[7]; | |
142 | } SheepdogRsp; | |
143 | ||
144 | typedef struct SheepdogObjReq { | |
145 | uint8_t proto_ver; | |
146 | uint8_t opcode; | |
147 | uint16_t flags; | |
148 | uint32_t epoch; | |
149 | uint32_t id; | |
150 | uint32_t data_length; | |
151 | uint64_t oid; | |
152 | uint64_t cow_oid; | |
153 | uint8_t copies; | |
154 | uint8_t copy_policy; | |
155 | uint8_t reserved[6]; | |
156 | uint64_t offset; | |
157 | } SheepdogObjReq; | |
158 | ||
159 | typedef struct SheepdogObjRsp { | |
160 | uint8_t proto_ver; | |
161 | uint8_t opcode; | |
162 | uint16_t flags; | |
163 | uint32_t epoch; | |
164 | uint32_t id; | |
165 | uint32_t data_length; | |
166 | uint32_t result; | |
167 | uint8_t copies; | |
168 | uint8_t copy_policy; | |
169 | uint8_t reserved[2]; | |
170 | uint32_t pad[6]; | |
171 | } SheepdogObjRsp; | |
172 | ||
173 | typedef struct SheepdogVdiReq { | |
174 | uint8_t proto_ver; | |
175 | uint8_t opcode; | |
176 | uint16_t flags; | |
177 | uint32_t epoch; | |
178 | uint32_t id; | |
179 | uint32_t data_length; | |
180 | uint64_t vdi_size; | |
181 | uint32_t base_vdi_id; | |
182 | uint8_t copies; | |
183 | uint8_t copy_policy; | |
184 | uint8_t store_policy; | |
185 | uint8_t block_size_shift; | |
186 | uint32_t snapid; | |
187 | uint32_t type; | |
188 | uint32_t pad[2]; | |
189 | } SheepdogVdiReq; | |
190 | ||
191 | typedef struct SheepdogVdiRsp { | |
192 | uint8_t proto_ver; | |
193 | uint8_t opcode; | |
194 | uint16_t flags; | |
195 | uint32_t epoch; | |
196 | uint32_t id; | |
197 | uint32_t data_length; | |
198 | uint32_t result; | |
199 | uint32_t rsvd; | |
200 | uint32_t vdi_id; | |
201 | uint32_t pad[5]; | |
202 | } SheepdogVdiRsp; | |
203 | ||
204 | typedef struct SheepdogClusterRsp { | |
205 | uint8_t proto_ver; | |
206 | uint8_t opcode; | |
207 | uint16_t flags; | |
208 | uint32_t epoch; | |
209 | uint32_t id; | |
210 | uint32_t data_length; | |
211 | uint32_t result; | |
212 | uint8_t nr_copies; | |
213 | uint8_t copy_policy; | |
214 | uint8_t block_size_shift; | |
215 | uint8_t __pad1; | |
216 | uint32_t __pad2[6]; | |
217 | } SheepdogClusterRsp; | |
218 | ||
219 | typedef struct SheepdogInode { | |
220 | char name[SD_MAX_VDI_LEN]; | |
221 | char tag[SD_MAX_VDI_TAG_LEN]; | |
222 | uint64_t ctime; | |
223 | uint64_t snap_ctime; | |
224 | uint64_t vm_clock_nsec; | |
225 | uint64_t vdi_size; | |
226 | uint64_t vm_state_size; | |
227 | uint16_t copy_policy; | |
228 | uint8_t nr_copies; | |
229 | uint8_t block_size_shift; | |
230 | uint32_t snap_id; | |
231 | uint32_t vdi_id; | |
232 | uint32_t parent_vdi_id; | |
233 | uint32_t child_vdi_id[MAX_CHILDREN]; | |
234 | uint32_t data_vdi_id[MAX_DATA_OBJS]; | |
235 | } SheepdogInode; | |
236 | ||
237 | #define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id) | |
238 | ||
239 | /* | |
240 | * 64 bit FNV-1a non-zero initial basis | |
241 | */ | |
242 | #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL) | |
243 | ||
244 | /* | |
245 | * 64 bit Fowler/Noll/Vo FNV-1a hash code | |
246 | */ | |
247 | static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval) | |
248 | { | |
249 | unsigned char *bp = buf; | |
250 | unsigned char *be = bp + len; | |
251 | while (bp < be) { | |
252 | hval ^= (uint64_t) *bp++; | |
253 | hval += (hval << 1) + (hval << 4) + (hval << 5) + | |
254 | (hval << 7) + (hval << 8) + (hval << 40); | |
255 | } | |
256 | return hval; | |
257 | } | |
258 | ||
259 | static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx) | |
260 | { | |
261 | return inode->vdi_id == inode->data_vdi_id[idx]; | |
262 | } | |
263 | ||
264 | static inline bool is_data_obj(uint64_t oid) | |
265 | { | |
266 | return !(VDI_BIT & oid); | |
267 | } | |
268 | ||
269 | static inline uint64_t data_oid_to_idx(uint64_t oid) | |
270 | { | |
271 | return oid & (MAX_DATA_OBJS - 1); | |
272 | } | |
273 | ||
274 | static inline uint32_t oid_to_vid(uint64_t oid) | |
275 | { | |
276 | return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT; | |
277 | } | |
278 | ||
279 | static inline uint64_t vid_to_vdi_oid(uint32_t vid) | |
280 | { | |
281 | return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT); | |
282 | } | |
283 | ||
284 | static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx) | |
285 | { | |
286 | return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
287 | } | |
288 | ||
289 | static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx) | |
290 | { | |
291 | return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx; | |
292 | } | |
293 | ||
294 | static inline bool is_snapshot(struct SheepdogInode *inode) | |
295 | { | |
296 | return !!inode->snap_ctime; | |
297 | } | |
298 | ||
299 | static inline size_t count_data_objs(const struct SheepdogInode *inode) | |
300 | { | |
301 | return DIV_ROUND_UP(inode->vdi_size, | |
302 | (1UL << inode->block_size_shift)); | |
303 | } | |
304 | ||
305 | typedef struct SheepdogAIOCB SheepdogAIOCB; | |
306 | typedef struct BDRVSheepdogState BDRVSheepdogState; | |
307 | ||
308 | typedef struct AIOReq { | |
309 | SheepdogAIOCB *aiocb; | |
310 | unsigned int iov_offset; | |
311 | ||
312 | uint64_t oid; | |
313 | uint64_t base_oid; | |
314 | uint64_t offset; | |
315 | unsigned int data_len; | |
316 | uint8_t flags; | |
317 | uint32_t id; | |
318 | bool create; | |
319 | ||
320 | QLIST_ENTRY(AIOReq) aio_siblings; | |
321 | } AIOReq; | |
322 | ||
323 | enum AIOCBState { | |
324 | AIOCB_WRITE_UDATA, | |
325 | AIOCB_READ_UDATA, | |
326 | AIOCB_FLUSH_CACHE, | |
327 | AIOCB_DISCARD_OBJ, | |
328 | }; | |
329 | ||
330 | #define AIOCBOverlapping(x, y) \ | |
331 | (!(x->max_affect_data_idx < y->min_affect_data_idx \ | |
332 | || y->max_affect_data_idx < x->min_affect_data_idx)) | |
333 | ||
334 | struct SheepdogAIOCB { | |
335 | BDRVSheepdogState *s; | |
336 | ||
337 | QEMUIOVector *qiov; | |
338 | ||
339 | int64_t sector_num; | |
340 | int nb_sectors; | |
341 | ||
342 | int ret; | |
343 | enum AIOCBState aiocb_type; | |
344 | ||
345 | Coroutine *coroutine; | |
346 | int nr_pending; | |
347 | ||
348 | uint32_t min_affect_data_idx; | |
349 | uint32_t max_affect_data_idx; | |
350 | ||
351 | /* | |
352 | * The difference between affect_data_idx and dirty_data_idx: | |
353 | * affect_data_idx represents range of index of all request types. | |
354 | * dirty_data_idx represents range of index updated by COW requests. | |
355 | * dirty_data_idx is used for updating an inode object. | |
356 | */ | |
357 | uint32_t min_dirty_data_idx; | |
358 | uint32_t max_dirty_data_idx; | |
359 | ||
360 | QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings; | |
361 | }; | |
362 | ||
363 | struct BDRVSheepdogState { | |
364 | BlockDriverState *bs; | |
365 | AioContext *aio_context; | |
366 | ||
367 | SheepdogInode inode; | |
368 | ||
369 | char name[SD_MAX_VDI_LEN]; | |
370 | bool is_snapshot; | |
371 | uint32_t cache_flags; | |
372 | bool discard_supported; | |
373 | ||
374 | SocketAddress *addr; | |
375 | int fd; | |
376 | ||
377 | CoMutex lock; | |
378 | Coroutine *co_send; | |
379 | Coroutine *co_recv; | |
380 | ||
381 | uint32_t aioreq_seq_num; | |
382 | ||
383 | /* Every aio request must be linked to either of these queues. */ | |
384 | QLIST_HEAD(, AIOReq) inflight_aio_head; | |
385 | QLIST_HEAD(, AIOReq) failed_aio_head; | |
386 | ||
387 | CoMutex queue_lock; | |
388 | CoQueue overlapping_queue; | |
389 | QLIST_HEAD(, SheepdogAIOCB) inflight_aiocb_head; | |
390 | }; | |
391 | ||
392 | typedef struct BDRVSheepdogReopenState { | |
393 | int fd; | |
394 | int cache_flags; | |
395 | } BDRVSheepdogReopenState; | |
396 | ||
397 | static const char *sd_strerror(int err) | |
398 | { | |
399 | int i; | |
400 | ||
401 | static const struct { | |
402 | int err; | |
403 | const char *desc; | |
404 | } errors[] = { | |
405 | {SD_RES_SUCCESS, "Success"}, | |
406 | {SD_RES_UNKNOWN, "Unknown error"}, | |
407 | {SD_RES_NO_OBJ, "No object found"}, | |
408 | {SD_RES_EIO, "I/O error"}, | |
409 | {SD_RES_VDI_EXIST, "VDI exists already"}, | |
410 | {SD_RES_INVALID_PARMS, "Invalid parameters"}, | |
411 | {SD_RES_SYSTEM_ERROR, "System error"}, | |
412 | {SD_RES_VDI_LOCKED, "VDI is already locked"}, | |
413 | {SD_RES_NO_VDI, "No vdi found"}, | |
414 | {SD_RES_NO_BASE_VDI, "No base VDI found"}, | |
415 | {SD_RES_VDI_READ, "Failed read the requested VDI"}, | |
416 | {SD_RES_VDI_WRITE, "Failed to write the requested VDI"}, | |
417 | {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"}, | |
418 | {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"}, | |
419 | {SD_RES_NO_TAG, "Failed to find the requested tag"}, | |
420 | {SD_RES_STARTUP, "The system is still booting"}, | |
421 | {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"}, | |
422 | {SD_RES_SHUTDOWN, "The system is shutting down"}, | |
423 | {SD_RES_NO_MEM, "Out of memory on the server"}, | |
424 | {SD_RES_FULL_VDI, "We already have the maximum vdis"}, | |
425 | {SD_RES_VER_MISMATCH, "Protocol version mismatch"}, | |
426 | {SD_RES_NO_SPACE, "Server has no space for new objects"}, | |
427 | {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"}, | |
428 | {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"}, | |
429 | {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"}, | |
430 | {SD_RES_HALT, "Sheepdog is stopped serving IO request"}, | |
431 | {SD_RES_READONLY, "Object is read-only"}, | |
432 | }; | |
433 | ||
434 | for (i = 0; i < ARRAY_SIZE(errors); ++i) { | |
435 | if (errors[i].err == err) { | |
436 | return errors[i].desc; | |
437 | } | |
438 | } | |
439 | ||
440 | return "Invalid error code"; | |
441 | } | |
442 | ||
443 | /* | |
444 | * Sheepdog I/O handling: | |
445 | * | |
446 | * 1. In sd_co_rw_vector, we send the I/O requests to the server and | |
447 | * link the requests to the inflight_list in the | |
448 | * BDRVSheepdogState. The function yields while waiting for | |
449 | * receiving the response. | |
450 | * | |
451 | * 2. We receive the response in aio_read_response, the fd handler to | |
452 | * the sheepdog connection. We switch back to sd_co_readv/sd_writev | |
453 | * after all the requests belonging to the AIOCB are finished. If | |
454 | * needed, sd_co_writev will send another requests for the vdi object. | |
455 | */ | |
456 | ||
457 | static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb, | |
458 | uint64_t oid, unsigned int data_len, | |
459 | uint64_t offset, uint8_t flags, bool create, | |
460 | uint64_t base_oid, unsigned int iov_offset) | |
461 | { | |
462 | AIOReq *aio_req; | |
463 | ||
464 | aio_req = g_malloc(sizeof(*aio_req)); | |
465 | aio_req->aiocb = acb; | |
466 | aio_req->iov_offset = iov_offset; | |
467 | aio_req->oid = oid; | |
468 | aio_req->base_oid = base_oid; | |
469 | aio_req->offset = offset; | |
470 | aio_req->data_len = data_len; | |
471 | aio_req->flags = flags; | |
472 | aio_req->id = s->aioreq_seq_num++; | |
473 | aio_req->create = create; | |
474 | ||
475 | acb->nr_pending++; | |
476 | return aio_req; | |
477 | } | |
478 | ||
479 | static void wait_for_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *acb) | |
480 | { | |
481 | SheepdogAIOCB *cb; | |
482 | ||
483 | retry: | |
484 | QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) { | |
485 | if (AIOCBOverlapping(acb, cb)) { | |
486 | qemu_co_queue_wait(&s->overlapping_queue, &s->queue_lock); | |
487 | goto retry; | |
488 | } | |
489 | } | |
490 | } | |
491 | ||
492 | static void sd_aio_setup(SheepdogAIOCB *acb, BDRVSheepdogState *s, | |
493 | QEMUIOVector *qiov, int64_t sector_num, int nb_sectors, | |
494 | int type) | |
495 | { | |
496 | uint32_t object_size; | |
497 | ||
498 | object_size = (UINT32_C(1) << s->inode.block_size_shift); | |
499 | ||
500 | acb->s = s; | |
501 | ||
502 | acb->qiov = qiov; | |
503 | ||
504 | acb->sector_num = sector_num; | |
505 | acb->nb_sectors = nb_sectors; | |
506 | ||
507 | acb->coroutine = qemu_coroutine_self(); | |
508 | acb->ret = 0; | |
509 | acb->nr_pending = 0; | |
510 | ||
511 | acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
512 | acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE + | |
513 | acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size; | |
514 | ||
515 | acb->min_dirty_data_idx = UINT32_MAX; | |
516 | acb->max_dirty_data_idx = 0; | |
517 | acb->aiocb_type = type; | |
518 | ||
519 | if (type == AIOCB_FLUSH_CACHE) { | |
520 | return; | |
521 | } | |
522 | ||
523 | qemu_co_mutex_lock(&s->queue_lock); | |
524 | wait_for_overlapping_aiocb(s, acb); | |
525 | QLIST_INSERT_HEAD(&s->inflight_aiocb_head, acb, aiocb_siblings); | |
526 | qemu_co_mutex_unlock(&s->queue_lock); | |
527 | } | |
528 | ||
529 | static SocketAddress *sd_server_config(QDict *options, Error **errp) | |
530 | { | |
531 | QDict *server = NULL; | |
532 | Visitor *iv = NULL; | |
533 | SocketAddress *saddr = NULL; | |
534 | Error *local_err = NULL; | |
535 | ||
536 | qdict_extract_subqdict(options, &server, "server."); | |
537 | ||
538 | iv = qobject_input_visitor_new_flat_confused(server, errp); | |
539 | if (!iv) { | |
540 | goto done; | |
541 | } | |
542 | ||
543 | visit_type_SocketAddress(iv, NULL, &saddr, &local_err); | |
544 | if (local_err) { | |
545 | error_propagate(errp, local_err); | |
546 | goto done; | |
547 | } | |
548 | ||
549 | done: | |
550 | visit_free(iv); | |
551 | qobject_unref(server); | |
552 | return saddr; | |
553 | } | |
554 | ||
555 | /* Return -EIO in case of error, file descriptor on success */ | |
556 | static int connect_to_sdog(BDRVSheepdogState *s, Error **errp) | |
557 | { | |
558 | int fd; | |
559 | ||
560 | fd = socket_connect(s->addr, errp); | |
561 | ||
562 | if (s->addr->type == SOCKET_ADDRESS_TYPE_INET && fd >= 0) { | |
563 | int ret = socket_set_nodelay(fd); | |
564 | if (ret < 0) { | |
565 | warn_report("can't set TCP_NODELAY: %s", strerror(errno)); | |
566 | } | |
567 | } | |
568 | ||
569 | if (fd >= 0) { | |
570 | qemu_set_nonblock(fd); | |
571 | } else { | |
572 | fd = -EIO; | |
573 | } | |
574 | ||
575 | return fd; | |
576 | } | |
577 | ||
578 | /* Return 0 on success and -errno in case of error */ | |
579 | static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data, | |
580 | unsigned int *wlen) | |
581 | { | |
582 | int ret; | |
583 | ||
584 | ret = qemu_co_send(sockfd, hdr, sizeof(*hdr)); | |
585 | if (ret != sizeof(*hdr)) { | |
586 | error_report("failed to send a req, %s", strerror(errno)); | |
587 | return -errno; | |
588 | } | |
589 | ||
590 | ret = qemu_co_send(sockfd, data, *wlen); | |
591 | if (ret != *wlen) { | |
592 | error_report("failed to send a req, %s", strerror(errno)); | |
593 | return -errno; | |
594 | } | |
595 | ||
596 | return ret; | |
597 | } | |
598 | ||
599 | typedef struct SheepdogReqCo { | |
600 | int sockfd; | |
601 | BlockDriverState *bs; | |
602 | AioContext *aio_context; | |
603 | SheepdogReq *hdr; | |
604 | void *data; | |
605 | unsigned int *wlen; | |
606 | unsigned int *rlen; | |
607 | int ret; | |
608 | bool finished; | |
609 | Coroutine *co; | |
610 | } SheepdogReqCo; | |
611 | ||
612 | static void restart_co_req(void *opaque) | |
613 | { | |
614 | SheepdogReqCo *srco = opaque; | |
615 | ||
616 | aio_co_wake(srco->co); | |
617 | } | |
618 | ||
619 | static coroutine_fn void do_co_req(void *opaque) | |
620 | { | |
621 | int ret; | |
622 | SheepdogReqCo *srco = opaque; | |
623 | int sockfd = srco->sockfd; | |
624 | SheepdogReq *hdr = srco->hdr; | |
625 | void *data = srco->data; | |
626 | unsigned int *wlen = srco->wlen; | |
627 | unsigned int *rlen = srco->rlen; | |
628 | ||
629 | srco->co = qemu_coroutine_self(); | |
630 | aio_set_fd_handler(srco->aio_context, sockfd, false, | |
631 | NULL, restart_co_req, NULL, srco); | |
632 | ||
633 | ret = send_co_req(sockfd, hdr, data, wlen); | |
634 | if (ret < 0) { | |
635 | goto out; | |
636 | } | |
637 | ||
638 | aio_set_fd_handler(srco->aio_context, sockfd, false, | |
639 | restart_co_req, NULL, NULL, srco); | |
640 | ||
641 | ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr)); | |
642 | if (ret != sizeof(*hdr)) { | |
643 | error_report("failed to get a rsp, %s", strerror(errno)); | |
644 | ret = -errno; | |
645 | goto out; | |
646 | } | |
647 | ||
648 | if (*rlen > hdr->data_length) { | |
649 | *rlen = hdr->data_length; | |
650 | } | |
651 | ||
652 | if (*rlen) { | |
653 | ret = qemu_co_recv(sockfd, data, *rlen); | |
654 | if (ret != *rlen) { | |
655 | error_report("failed to get the data, %s", strerror(errno)); | |
656 | ret = -errno; | |
657 | goto out; | |
658 | } | |
659 | } | |
660 | ret = 0; | |
661 | out: | |
662 | /* there is at most one request for this sockfd, so it is safe to | |
663 | * set each handler to NULL. */ | |
664 | aio_set_fd_handler(srco->aio_context, sockfd, false, | |
665 | NULL, NULL, NULL, NULL); | |
666 | ||
667 | srco->co = NULL; | |
668 | srco->ret = ret; | |
669 | /* Set srco->finished before reading bs->wakeup. */ | |
670 | atomic_mb_set(&srco->finished, true); | |
671 | if (srco->bs) { | |
672 | bdrv_wakeup(srco->bs); | |
673 | } | |
674 | } | |
675 | ||
676 | /* | |
677 | * Send the request to the sheep in a synchronous manner. | |
678 | * | |
679 | * Return 0 on success, -errno in case of error. | |
680 | */ | |
681 | static int do_req(int sockfd, BlockDriverState *bs, SheepdogReq *hdr, | |
682 | void *data, unsigned int *wlen, unsigned int *rlen) | |
683 | { | |
684 | Coroutine *co; | |
685 | SheepdogReqCo srco = { | |
686 | .sockfd = sockfd, | |
687 | .aio_context = bs ? bdrv_get_aio_context(bs) : qemu_get_aio_context(), | |
688 | .bs = bs, | |
689 | .hdr = hdr, | |
690 | .data = data, | |
691 | .wlen = wlen, | |
692 | .rlen = rlen, | |
693 | .ret = 0, | |
694 | .finished = false, | |
695 | }; | |
696 | ||
697 | if (qemu_in_coroutine()) { | |
698 | do_co_req(&srco); | |
699 | } else { | |
700 | co = qemu_coroutine_create(do_co_req, &srco); | |
701 | if (bs) { | |
702 | bdrv_coroutine_enter(bs, co); | |
703 | BDRV_POLL_WHILE(bs, !srco.finished); | |
704 | } else { | |
705 | qemu_coroutine_enter(co); | |
706 | while (!srco.finished) { | |
707 | aio_poll(qemu_get_aio_context(), true); | |
708 | } | |
709 | } | |
710 | } | |
711 | ||
712 | return srco.ret; | |
713 | } | |
714 | ||
715 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, | |
716 | struct iovec *iov, int niov, | |
717 | enum AIOCBState aiocb_type); | |
718 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req); | |
719 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag); | |
720 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp); | |
721 | static void co_write_request(void *opaque); | |
722 | ||
723 | static coroutine_fn void reconnect_to_sdog(void *opaque) | |
724 | { | |
725 | BDRVSheepdogState *s = opaque; | |
726 | AIOReq *aio_req, *next; | |
727 | ||
728 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, | |
729 | NULL, NULL, NULL); | |
730 | close(s->fd); | |
731 | s->fd = -1; | |
732 | ||
733 | /* Wait for outstanding write requests to be completed. */ | |
734 | while (s->co_send != NULL) { | |
735 | co_write_request(opaque); | |
736 | } | |
737 | ||
738 | /* Try to reconnect the sheepdog server every one second. */ | |
739 | while (s->fd < 0) { | |
740 | Error *local_err = NULL; | |
741 | s->fd = get_sheep_fd(s, &local_err); | |
742 | if (s->fd < 0) { | |
743 | trace_sheepdog_reconnect_to_sdog(); | |
744 | error_report_err(local_err); | |
745 | qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000000ULL); | |
746 | } | |
747 | }; | |
748 | ||
749 | /* | |
750 | * Now we have to resend all the request in the inflight queue. However, | |
751 | * resend_aioreq() can yield and newly created requests can be added to the | |
752 | * inflight queue before the coroutine is resumed. To avoid mixing them, we | |
753 | * have to move all the inflight requests to the failed queue before | |
754 | * resend_aioreq() is called. | |
755 | */ | |
756 | qemu_co_mutex_lock(&s->queue_lock); | |
757 | QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) { | |
758 | QLIST_REMOVE(aio_req, aio_siblings); | |
759 | QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings); | |
760 | } | |
761 | ||
762 | /* Resend all the failed aio requests. */ | |
763 | while (!QLIST_EMPTY(&s->failed_aio_head)) { | |
764 | aio_req = QLIST_FIRST(&s->failed_aio_head); | |
765 | QLIST_REMOVE(aio_req, aio_siblings); | |
766 | qemu_co_mutex_unlock(&s->queue_lock); | |
767 | resend_aioreq(s, aio_req); | |
768 | qemu_co_mutex_lock(&s->queue_lock); | |
769 | } | |
770 | qemu_co_mutex_unlock(&s->queue_lock); | |
771 | } | |
772 | ||
773 | /* | |
774 | * Receive responses of the I/O requests. | |
775 | * | |
776 | * This function is registered as a fd handler, and called from the | |
777 | * main loop when s->fd is ready for reading responses. | |
778 | */ | |
779 | static void coroutine_fn aio_read_response(void *opaque) | |
780 | { | |
781 | SheepdogObjRsp rsp; | |
782 | BDRVSheepdogState *s = opaque; | |
783 | int fd = s->fd; | |
784 | int ret; | |
785 | AIOReq *aio_req = NULL; | |
786 | SheepdogAIOCB *acb; | |
787 | uint64_t idx; | |
788 | ||
789 | /* read a header */ | |
790 | ret = qemu_co_recv(fd, &rsp, sizeof(rsp)); | |
791 | if (ret != sizeof(rsp)) { | |
792 | error_report("failed to get the header, %s", strerror(errno)); | |
793 | goto err; | |
794 | } | |
795 | ||
796 | /* find the right aio_req from the inflight aio list */ | |
797 | QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) { | |
798 | if (aio_req->id == rsp.id) { | |
799 | break; | |
800 | } | |
801 | } | |
802 | if (!aio_req) { | |
803 | error_report("cannot find aio_req %x", rsp.id); | |
804 | goto err; | |
805 | } | |
806 | ||
807 | acb = aio_req->aiocb; | |
808 | ||
809 | switch (acb->aiocb_type) { | |
810 | case AIOCB_WRITE_UDATA: | |
811 | if (!is_data_obj(aio_req->oid)) { | |
812 | break; | |
813 | } | |
814 | idx = data_oid_to_idx(aio_req->oid); | |
815 | ||
816 | if (aio_req->create) { | |
817 | /* | |
818 | * If the object is newly created one, we need to update | |
819 | * the vdi object (metadata object). min_dirty_data_idx | |
820 | * and max_dirty_data_idx are changed to include updated | |
821 | * index between them. | |
822 | */ | |
823 | if (rsp.result == SD_RES_SUCCESS) { | |
824 | s->inode.data_vdi_id[idx] = s->inode.vdi_id; | |
825 | acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx); | |
826 | acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx); | |
827 | } | |
828 | } | |
829 | break; | |
830 | case AIOCB_READ_UDATA: | |
831 | ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov, | |
832 | aio_req->iov_offset, rsp.data_length); | |
833 | if (ret != rsp.data_length) { | |
834 | error_report("failed to get the data, %s", strerror(errno)); | |
835 | goto err; | |
836 | } | |
837 | break; | |
838 | case AIOCB_FLUSH_CACHE: | |
839 | if (rsp.result == SD_RES_INVALID_PARMS) { | |
840 | trace_sheepdog_aio_read_response(); | |
841 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
842 | rsp.result = SD_RES_SUCCESS; | |
843 | } | |
844 | break; | |
845 | case AIOCB_DISCARD_OBJ: | |
846 | switch (rsp.result) { | |
847 | case SD_RES_INVALID_PARMS: | |
848 | error_report("server doesn't support discard command"); | |
849 | rsp.result = SD_RES_SUCCESS; | |
850 | s->discard_supported = false; | |
851 | break; | |
852 | default: | |
853 | break; | |
854 | } | |
855 | } | |
856 | ||
857 | /* No more data for this aio_req (reload_inode below uses its own file | |
858 | * descriptor handler which doesn't use co_recv). | |
859 | */ | |
860 | s->co_recv = NULL; | |
861 | ||
862 | qemu_co_mutex_lock(&s->queue_lock); | |
863 | QLIST_REMOVE(aio_req, aio_siblings); | |
864 | qemu_co_mutex_unlock(&s->queue_lock); | |
865 | ||
866 | switch (rsp.result) { | |
867 | case SD_RES_SUCCESS: | |
868 | break; | |
869 | case SD_RES_READONLY: | |
870 | if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) { | |
871 | ret = reload_inode(s, 0, ""); | |
872 | if (ret < 0) { | |
873 | goto err; | |
874 | } | |
875 | } | |
876 | if (is_data_obj(aio_req->oid)) { | |
877 | aio_req->oid = vid_to_data_oid(s->inode.vdi_id, | |
878 | data_oid_to_idx(aio_req->oid)); | |
879 | } else { | |
880 | aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id); | |
881 | } | |
882 | resend_aioreq(s, aio_req); | |
883 | return; | |
884 | default: | |
885 | acb->ret = -EIO; | |
886 | error_report("%s", sd_strerror(rsp.result)); | |
887 | break; | |
888 | } | |
889 | ||
890 | g_free(aio_req); | |
891 | ||
892 | if (!--acb->nr_pending) { | |
893 | /* | |
894 | * We've finished all requests which belong to the AIOCB, so | |
895 | * we can switch back to sd_co_readv/writev now. | |
896 | */ | |
897 | aio_co_wake(acb->coroutine); | |
898 | } | |
899 | ||
900 | return; | |
901 | ||
902 | err: | |
903 | reconnect_to_sdog(opaque); | |
904 | } | |
905 | ||
906 | static void co_read_response(void *opaque) | |
907 | { | |
908 | BDRVSheepdogState *s = opaque; | |
909 | ||
910 | if (!s->co_recv) { | |
911 | s->co_recv = qemu_coroutine_create(aio_read_response, opaque); | |
912 | } | |
913 | ||
914 | aio_co_enter(s->aio_context, s->co_recv); | |
915 | } | |
916 | ||
917 | static void co_write_request(void *opaque) | |
918 | { | |
919 | BDRVSheepdogState *s = opaque; | |
920 | ||
921 | aio_co_wake(s->co_send); | |
922 | } | |
923 | ||
924 | /* | |
925 | * Return a socket descriptor to read/write objects. | |
926 | * | |
927 | * We cannot use this descriptor for other operations because | |
928 | * the block driver may be on waiting response from the server. | |
929 | */ | |
930 | static int get_sheep_fd(BDRVSheepdogState *s, Error **errp) | |
931 | { | |
932 | int fd; | |
933 | ||
934 | fd = connect_to_sdog(s, errp); | |
935 | if (fd < 0) { | |
936 | return fd; | |
937 | } | |
938 | ||
939 | aio_set_fd_handler(s->aio_context, fd, false, | |
940 | co_read_response, NULL, NULL, s); | |
941 | return fd; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Parse numeric snapshot ID in @str | |
946 | * If @str can't be parsed as number, return false. | |
947 | * Else, if the number is zero or too large, set *@snapid to zero and | |
948 | * return true. | |
949 | * Else, set *@snapid to the number and return true. | |
950 | */ | |
951 | static bool sd_parse_snapid(const char *str, uint32_t *snapid) | |
952 | { | |
953 | unsigned long ul; | |
954 | int ret; | |
955 | ||
956 | ret = qemu_strtoul(str, NULL, 10, &ul); | |
957 | if (ret == -ERANGE) { | |
958 | ul = ret = 0; | |
959 | } | |
960 | if (ret) { | |
961 | return false; | |
962 | } | |
963 | if (ul > UINT32_MAX) { | |
964 | ul = 0; | |
965 | } | |
966 | ||
967 | *snapid = ul; | |
968 | return true; | |
969 | } | |
970 | ||
971 | static bool sd_parse_snapid_or_tag(const char *str, | |
972 | uint32_t *snapid, char tag[]) | |
973 | { | |
974 | if (!sd_parse_snapid(str, snapid)) { | |
975 | *snapid = 0; | |
976 | if (g_strlcpy(tag, str, SD_MAX_VDI_TAG_LEN) >= SD_MAX_VDI_TAG_LEN) { | |
977 | return false; | |
978 | } | |
979 | } else if (!*snapid) { | |
980 | return false; | |
981 | } else { | |
982 | tag[0] = 0; | |
983 | } | |
984 | return true; | |
985 | } | |
986 | ||
987 | typedef struct { | |
988 | const char *path; /* non-null iff transport is tcp */ | |
989 | const char *host; /* valid when transport is tcp */ | |
990 | int port; /* valid when transport is tcp */ | |
991 | char vdi[SD_MAX_VDI_LEN]; | |
992 | char tag[SD_MAX_VDI_TAG_LEN]; | |
993 | uint32_t snap_id; | |
994 | /* Remainder is only for sd_config_done() */ | |
995 | URI *uri; | |
996 | QueryParams *qp; | |
997 | } SheepdogConfig; | |
998 | ||
999 | static void sd_config_done(SheepdogConfig *cfg) | |
1000 | { | |
1001 | if (cfg->qp) { | |
1002 | query_params_free(cfg->qp); | |
1003 | } | |
1004 | uri_free(cfg->uri); | |
1005 | } | |
1006 | ||
1007 | static void sd_parse_uri(SheepdogConfig *cfg, const char *filename, | |
1008 | Error **errp) | |
1009 | { | |
1010 | Error *err = NULL; | |
1011 | QueryParams *qp = NULL; | |
1012 | bool is_unix; | |
1013 | URI *uri; | |
1014 | ||
1015 | memset(cfg, 0, sizeof(*cfg)); | |
1016 | ||
1017 | cfg->uri = uri = uri_parse(filename); | |
1018 | if (!uri) { | |
1019 | error_setg(&err, "invalid URI '%s'", filename); | |
1020 | goto out; | |
1021 | } | |
1022 | ||
1023 | /* transport */ | |
1024 | if (!g_strcmp0(uri->scheme, "sheepdog")) { | |
1025 | is_unix = false; | |
1026 | } else if (!g_strcmp0(uri->scheme, "sheepdog+tcp")) { | |
1027 | is_unix = false; | |
1028 | } else if (!g_strcmp0(uri->scheme, "sheepdog+unix")) { | |
1029 | is_unix = true; | |
1030 | } else { | |
1031 | error_setg(&err, "URI scheme must be 'sheepdog', 'sheepdog+tcp'," | |
1032 | " or 'sheepdog+unix'"); | |
1033 | goto out; | |
1034 | } | |
1035 | ||
1036 | if (uri->path == NULL || !strcmp(uri->path, "/")) { | |
1037 | error_setg(&err, "missing file path in URI"); | |
1038 | goto out; | |
1039 | } | |
1040 | if (g_strlcpy(cfg->vdi, uri->path + 1, SD_MAX_VDI_LEN) | |
1041 | >= SD_MAX_VDI_LEN) { | |
1042 | error_setg(&err, "VDI name is too long"); | |
1043 | goto out; | |
1044 | } | |
1045 | ||
1046 | cfg->qp = qp = query_params_parse(uri->query); | |
1047 | ||
1048 | if (is_unix) { | |
1049 | /* sheepdog+unix:///vdiname?socket=path */ | |
1050 | if (uri->server || uri->port) { | |
1051 | error_setg(&err, "URI scheme %s doesn't accept a server address", | |
1052 | uri->scheme); | |
1053 | goto out; | |
1054 | } | |
1055 | if (!qp->n) { | |
1056 | error_setg(&err, | |
1057 | "URI scheme %s requires query parameter 'socket'", | |
1058 | uri->scheme); | |
1059 | goto out; | |
1060 | } | |
1061 | if (qp->n != 1 || strcmp(qp->p[0].name, "socket")) { | |
1062 | error_setg(&err, "unexpected query parameters"); | |
1063 | goto out; | |
1064 | } | |
1065 | cfg->path = qp->p[0].value; | |
1066 | } else { | |
1067 | /* sheepdog[+tcp]://[host:port]/vdiname */ | |
1068 | if (qp->n) { | |
1069 | error_setg(&err, "unexpected query parameters"); | |
1070 | goto out; | |
1071 | } | |
1072 | cfg->host = uri->server; | |
1073 | cfg->port = uri->port; | |
1074 | } | |
1075 | ||
1076 | /* snapshot tag */ | |
1077 | if (uri->fragment) { | |
1078 | if (!sd_parse_snapid_or_tag(uri->fragment, | |
1079 | &cfg->snap_id, cfg->tag)) { | |
1080 | error_setg(&err, "'%s' is not a valid snapshot ID", | |
1081 | uri->fragment); | |
1082 | goto out; | |
1083 | } | |
1084 | } else { | |
1085 | cfg->snap_id = CURRENT_VDI_ID; /* search current vdi */ | |
1086 | } | |
1087 | ||
1088 | out: | |
1089 | if (err) { | |
1090 | error_propagate(errp, err); | |
1091 | sd_config_done(cfg); | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Parse a filename (old syntax) | |
1097 | * | |
1098 | * filename must be one of the following formats: | |
1099 | * 1. [vdiname] | |
1100 | * 2. [vdiname]:[snapid] | |
1101 | * 3. [vdiname]:[tag] | |
1102 | * 4. [hostname]:[port]:[vdiname] | |
1103 | * 5. [hostname]:[port]:[vdiname]:[snapid] | |
1104 | * 6. [hostname]:[port]:[vdiname]:[tag] | |
1105 | * | |
1106 | * You can boot from the snapshot images by specifying `snapid` or | |
1107 | * `tag'. | |
1108 | * | |
1109 | * You can run VMs outside the Sheepdog cluster by specifying | |
1110 | * `hostname' and `port' (experimental). | |
1111 | */ | |
1112 | static void parse_vdiname(SheepdogConfig *cfg, const char *filename, | |
1113 | Error **errp) | |
1114 | { | |
1115 | Error *err = NULL; | |
1116 | char *p, *q, *uri; | |
1117 | const char *host_spec, *vdi_spec; | |
1118 | int nr_sep; | |
1119 | ||
1120 | strstart(filename, "sheepdog:", &filename); | |
1121 | p = q = g_strdup(filename); | |
1122 | ||
1123 | /* count the number of separators */ | |
1124 | nr_sep = 0; | |
1125 | while (*p) { | |
1126 | if (*p == ':') { | |
1127 | nr_sep++; | |
1128 | } | |
1129 | p++; | |
1130 | } | |
1131 | p = q; | |
1132 | ||
1133 | /* use the first two tokens as host_spec. */ | |
1134 | if (nr_sep >= 2) { | |
1135 | host_spec = p; | |
1136 | p = strchr(p, ':'); | |
1137 | p++; | |
1138 | p = strchr(p, ':'); | |
1139 | *p++ = '\0'; | |
1140 | } else { | |
1141 | host_spec = ""; | |
1142 | } | |
1143 | ||
1144 | vdi_spec = p; | |
1145 | ||
1146 | p = strchr(vdi_spec, ':'); | |
1147 | if (p) { | |
1148 | *p++ = '#'; | |
1149 | } | |
1150 | ||
1151 | uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec); | |
1152 | ||
1153 | /* | |
1154 | * FIXME We to escape URI meta-characters, e.g. "x?y=z" | |
1155 | * produces "sheepdog://x?y=z". Because of that ... | |
1156 | */ | |
1157 | sd_parse_uri(cfg, uri, &err); | |
1158 | if (err) { | |
1159 | /* | |
1160 | * ... this can fail, but the error message is misleading. | |
1161 | * Replace it by the traditional useless one until the | |
1162 | * escaping is fixed. | |
1163 | */ | |
1164 | error_free(err); | |
1165 | error_setg(errp, "Can't parse filename"); | |
1166 | } | |
1167 | ||
1168 | g_free(q); | |
1169 | g_free(uri); | |
1170 | } | |
1171 | ||
1172 | static void sd_parse_filename(const char *filename, QDict *options, | |
1173 | Error **errp) | |
1174 | { | |
1175 | Error *err = NULL; | |
1176 | SheepdogConfig cfg; | |
1177 | char buf[32]; | |
1178 | ||
1179 | if (strstr(filename, "://")) { | |
1180 | sd_parse_uri(&cfg, filename, &err); | |
1181 | } else { | |
1182 | parse_vdiname(&cfg, filename, &err); | |
1183 | } | |
1184 | if (err) { | |
1185 | error_propagate(errp, err); | |
1186 | return; | |
1187 | } | |
1188 | ||
1189 | if (cfg.path) { | |
1190 | qdict_set_default_str(options, "server.path", cfg.path); | |
1191 | qdict_set_default_str(options, "server.type", "unix"); | |
1192 | } else { | |
1193 | qdict_set_default_str(options, "server.type", "inet"); | |
1194 | qdict_set_default_str(options, "server.host", | |
1195 | cfg.host ?: SD_DEFAULT_ADDR); | |
1196 | snprintf(buf, sizeof(buf), "%d", cfg.port ?: SD_DEFAULT_PORT); | |
1197 | qdict_set_default_str(options, "server.port", buf); | |
1198 | } | |
1199 | qdict_set_default_str(options, "vdi", cfg.vdi); | |
1200 | qdict_set_default_str(options, "tag", cfg.tag); | |
1201 | if (cfg.snap_id) { | |
1202 | snprintf(buf, sizeof(buf), "%d", cfg.snap_id); | |
1203 | qdict_set_default_str(options, "snap-id", buf); | |
1204 | } | |
1205 | ||
1206 | sd_config_done(&cfg); | |
1207 | } | |
1208 | ||
1209 | static int find_vdi_name(BDRVSheepdogState *s, const char *filename, | |
1210 | uint32_t snapid, const char *tag, uint32_t *vid, | |
1211 | bool lock, Error **errp) | |
1212 | { | |
1213 | int ret, fd; | |
1214 | SheepdogVdiReq hdr; | |
1215 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1216 | unsigned int wlen, rlen = 0; | |
1217 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN] QEMU_NONSTRING; | |
1218 | ||
1219 | fd = connect_to_sdog(s, errp); | |
1220 | if (fd < 0) { | |
1221 | return fd; | |
1222 | } | |
1223 | ||
1224 | /* This pair of strncpy calls ensures that the buffer is zero-filled, | |
1225 | * which is desirable since we'll soon be sending those bytes, and | |
1226 | * don't want the send_req to read uninitialized data. | |
1227 | */ | |
1228 | strncpy(buf, filename, SD_MAX_VDI_LEN); | |
1229 | strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN); | |
1230 | ||
1231 | memset(&hdr, 0, sizeof(hdr)); | |
1232 | if (lock) { | |
1233 | hdr.opcode = SD_OP_LOCK_VDI; | |
1234 | hdr.type = LOCK_TYPE_NORMAL; | |
1235 | } else { | |
1236 | hdr.opcode = SD_OP_GET_VDI_INFO; | |
1237 | } | |
1238 | wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN; | |
1239 | hdr.proto_ver = SD_PROTO_VER; | |
1240 | hdr.data_length = wlen; | |
1241 | hdr.snapid = snapid; | |
1242 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1243 | ||
1244 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1245 | if (ret) { | |
1246 | error_setg_errno(errp, -ret, "cannot get vdi info"); | |
1247 | goto out; | |
1248 | } | |
1249 | ||
1250 | if (rsp->result != SD_RES_SUCCESS) { | |
1251 | error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s", | |
1252 | sd_strerror(rsp->result), filename, snapid, tag); | |
1253 | if (rsp->result == SD_RES_NO_VDI) { | |
1254 | ret = -ENOENT; | |
1255 | } else if (rsp->result == SD_RES_VDI_LOCKED) { | |
1256 | ret = -EBUSY; | |
1257 | } else { | |
1258 | ret = -EIO; | |
1259 | } | |
1260 | goto out; | |
1261 | } | |
1262 | *vid = rsp->vdi_id; | |
1263 | ||
1264 | ret = 0; | |
1265 | out: | |
1266 | closesocket(fd); | |
1267 | return ret; | |
1268 | } | |
1269 | ||
1270 | static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req, | |
1271 | struct iovec *iov, int niov, | |
1272 | enum AIOCBState aiocb_type) | |
1273 | { | |
1274 | int nr_copies = s->inode.nr_copies; | |
1275 | SheepdogObjReq hdr; | |
1276 | unsigned int wlen = 0; | |
1277 | int ret; | |
1278 | uint64_t oid = aio_req->oid; | |
1279 | unsigned int datalen = aio_req->data_len; | |
1280 | uint64_t offset = aio_req->offset; | |
1281 | uint8_t flags = aio_req->flags; | |
1282 | uint64_t old_oid = aio_req->base_oid; | |
1283 | bool create = aio_req->create; | |
1284 | ||
1285 | qemu_co_mutex_lock(&s->queue_lock); | |
1286 | QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings); | |
1287 | qemu_co_mutex_unlock(&s->queue_lock); | |
1288 | ||
1289 | if (!nr_copies) { | |
1290 | error_report("bug"); | |
1291 | } | |
1292 | ||
1293 | memset(&hdr, 0, sizeof(hdr)); | |
1294 | ||
1295 | switch (aiocb_type) { | |
1296 | case AIOCB_FLUSH_CACHE: | |
1297 | hdr.opcode = SD_OP_FLUSH_VDI; | |
1298 | break; | |
1299 | case AIOCB_READ_UDATA: | |
1300 | hdr.opcode = SD_OP_READ_OBJ; | |
1301 | hdr.flags = flags; | |
1302 | break; | |
1303 | case AIOCB_WRITE_UDATA: | |
1304 | if (create) { | |
1305 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1306 | } else { | |
1307 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1308 | } | |
1309 | wlen = datalen; | |
1310 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
1311 | break; | |
1312 | case AIOCB_DISCARD_OBJ: | |
1313 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1314 | hdr.flags = SD_FLAG_CMD_WRITE | flags; | |
1315 | s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0; | |
1316 | offset = offsetof(SheepdogInode, | |
1317 | data_vdi_id[data_oid_to_idx(oid)]); | |
1318 | oid = vid_to_vdi_oid(s->inode.vdi_id); | |
1319 | wlen = datalen = sizeof(uint32_t); | |
1320 | break; | |
1321 | } | |
1322 | ||
1323 | if (s->cache_flags) { | |
1324 | hdr.flags |= s->cache_flags; | |
1325 | } | |
1326 | ||
1327 | hdr.oid = oid; | |
1328 | hdr.cow_oid = old_oid; | |
1329 | hdr.copies = s->inode.nr_copies; | |
1330 | ||
1331 | hdr.data_length = datalen; | |
1332 | hdr.offset = offset; | |
1333 | ||
1334 | hdr.id = aio_req->id; | |
1335 | ||
1336 | qemu_co_mutex_lock(&s->lock); | |
1337 | s->co_send = qemu_coroutine_self(); | |
1338 | aio_set_fd_handler(s->aio_context, s->fd, false, | |
1339 | co_read_response, co_write_request, NULL, s); | |
1340 | socket_set_cork(s->fd, 1); | |
1341 | ||
1342 | /* send a header */ | |
1343 | ret = qemu_co_send(s->fd, &hdr, sizeof(hdr)); | |
1344 | if (ret != sizeof(hdr)) { | |
1345 | error_report("failed to send a req, %s", strerror(errno)); | |
1346 | goto out; | |
1347 | } | |
1348 | ||
1349 | if (wlen) { | |
1350 | ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen); | |
1351 | if (ret != wlen) { | |
1352 | error_report("failed to send a data, %s", strerror(errno)); | |
1353 | } | |
1354 | } | |
1355 | out: | |
1356 | socket_set_cork(s->fd, 0); | |
1357 | aio_set_fd_handler(s->aio_context, s->fd, false, | |
1358 | co_read_response, NULL, NULL, s); | |
1359 | s->co_send = NULL; | |
1360 | qemu_co_mutex_unlock(&s->lock); | |
1361 | } | |
1362 | ||
1363 | static int read_write_object(int fd, BlockDriverState *bs, char *buf, | |
1364 | uint64_t oid, uint8_t copies, | |
1365 | unsigned int datalen, uint64_t offset, | |
1366 | bool write, bool create, uint32_t cache_flags) | |
1367 | { | |
1368 | SheepdogObjReq hdr; | |
1369 | SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr; | |
1370 | unsigned int wlen, rlen; | |
1371 | int ret; | |
1372 | ||
1373 | memset(&hdr, 0, sizeof(hdr)); | |
1374 | ||
1375 | if (write) { | |
1376 | wlen = datalen; | |
1377 | rlen = 0; | |
1378 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1379 | if (create) { | |
1380 | hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ; | |
1381 | } else { | |
1382 | hdr.opcode = SD_OP_WRITE_OBJ; | |
1383 | } | |
1384 | } else { | |
1385 | wlen = 0; | |
1386 | rlen = datalen; | |
1387 | hdr.opcode = SD_OP_READ_OBJ; | |
1388 | } | |
1389 | ||
1390 | hdr.flags |= cache_flags; | |
1391 | ||
1392 | hdr.oid = oid; | |
1393 | hdr.data_length = datalen; | |
1394 | hdr.offset = offset; | |
1395 | hdr.copies = copies; | |
1396 | ||
1397 | ret = do_req(fd, bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1398 | if (ret) { | |
1399 | error_report("failed to send a request to the sheep"); | |
1400 | return ret; | |
1401 | } | |
1402 | ||
1403 | switch (rsp->result) { | |
1404 | case SD_RES_SUCCESS: | |
1405 | return 0; | |
1406 | default: | |
1407 | error_report("%s", sd_strerror(rsp->result)); | |
1408 | return -EIO; | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | static int read_object(int fd, BlockDriverState *bs, char *buf, | |
1413 | uint64_t oid, uint8_t copies, | |
1414 | unsigned int datalen, uint64_t offset, | |
1415 | uint32_t cache_flags) | |
1416 | { | |
1417 | return read_write_object(fd, bs, buf, oid, copies, | |
1418 | datalen, offset, false, | |
1419 | false, cache_flags); | |
1420 | } | |
1421 | ||
1422 | static int write_object(int fd, BlockDriverState *bs, char *buf, | |
1423 | uint64_t oid, uint8_t copies, | |
1424 | unsigned int datalen, uint64_t offset, bool create, | |
1425 | uint32_t cache_flags) | |
1426 | { | |
1427 | return read_write_object(fd, bs, buf, oid, copies, | |
1428 | datalen, offset, true, | |
1429 | create, cache_flags); | |
1430 | } | |
1431 | ||
1432 | /* update inode with the latest state */ | |
1433 | static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag) | |
1434 | { | |
1435 | Error *local_err = NULL; | |
1436 | SheepdogInode *inode; | |
1437 | int ret = 0, fd; | |
1438 | uint32_t vid = 0; | |
1439 | ||
1440 | fd = connect_to_sdog(s, &local_err); | |
1441 | if (fd < 0) { | |
1442 | error_report_err(local_err); | |
1443 | return -EIO; | |
1444 | } | |
1445 | ||
1446 | inode = g_malloc(SD_INODE_HEADER_SIZE); | |
1447 | ||
1448 | ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err); | |
1449 | if (ret) { | |
1450 | error_report_err(local_err); | |
1451 | goto out; | |
1452 | } | |
1453 | ||
1454 | ret = read_object(fd, s->bs, (char *)inode, vid_to_vdi_oid(vid), | |
1455 | s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0, | |
1456 | s->cache_flags); | |
1457 | if (ret < 0) { | |
1458 | goto out; | |
1459 | } | |
1460 | ||
1461 | if (inode->vdi_id != s->inode.vdi_id) { | |
1462 | memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE); | |
1463 | } | |
1464 | ||
1465 | out: | |
1466 | g_free(inode); | |
1467 | closesocket(fd); | |
1468 | ||
1469 | return ret; | |
1470 | } | |
1471 | ||
1472 | static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req) | |
1473 | { | |
1474 | SheepdogAIOCB *acb = aio_req->aiocb; | |
1475 | ||
1476 | aio_req->create = false; | |
1477 | ||
1478 | /* check whether this request becomes a CoW one */ | |
1479 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) { | |
1480 | int idx = data_oid_to_idx(aio_req->oid); | |
1481 | ||
1482 | if (is_data_obj_writable(&s->inode, idx)) { | |
1483 | goto out; | |
1484 | } | |
1485 | ||
1486 | if (s->inode.data_vdi_id[idx]) { | |
1487 | aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx); | |
1488 | aio_req->flags |= SD_FLAG_CMD_COW; | |
1489 | } | |
1490 | aio_req->create = true; | |
1491 | } | |
1492 | out: | |
1493 | if (is_data_obj(aio_req->oid)) { | |
1494 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, | |
1495 | acb->aiocb_type); | |
1496 | } else { | |
1497 | struct iovec iov; | |
1498 | iov.iov_base = &s->inode; | |
1499 | iov.iov_len = sizeof(s->inode); | |
1500 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); | |
1501 | } | |
1502 | } | |
1503 | ||
1504 | static void sd_detach_aio_context(BlockDriverState *bs) | |
1505 | { | |
1506 | BDRVSheepdogState *s = bs->opaque; | |
1507 | ||
1508 | aio_set_fd_handler(s->aio_context, s->fd, false, NULL, | |
1509 | NULL, NULL, NULL); | |
1510 | } | |
1511 | ||
1512 | static void sd_attach_aio_context(BlockDriverState *bs, | |
1513 | AioContext *new_context) | |
1514 | { | |
1515 | BDRVSheepdogState *s = bs->opaque; | |
1516 | ||
1517 | s->aio_context = new_context; | |
1518 | aio_set_fd_handler(new_context, s->fd, false, | |
1519 | co_read_response, NULL, NULL, s); | |
1520 | } | |
1521 | ||
1522 | static QemuOptsList runtime_opts = { | |
1523 | .name = "sheepdog", | |
1524 | .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), | |
1525 | .desc = { | |
1526 | { | |
1527 | .name = "vdi", | |
1528 | .type = QEMU_OPT_STRING, | |
1529 | }, | |
1530 | { | |
1531 | .name = "snap-id", | |
1532 | .type = QEMU_OPT_NUMBER, | |
1533 | }, | |
1534 | { | |
1535 | .name = "tag", | |
1536 | .type = QEMU_OPT_STRING, | |
1537 | }, | |
1538 | { /* end of list */ } | |
1539 | }, | |
1540 | }; | |
1541 | ||
1542 | static int sd_open(BlockDriverState *bs, QDict *options, int flags, | |
1543 | Error **errp) | |
1544 | { | |
1545 | int ret, fd; | |
1546 | uint32_t vid = 0; | |
1547 | BDRVSheepdogState *s = bs->opaque; | |
1548 | const char *vdi, *snap_id_str, *tag; | |
1549 | uint64_t snap_id; | |
1550 | char *buf = NULL; | |
1551 | QemuOpts *opts; | |
1552 | Error *local_err = NULL; | |
1553 | ||
1554 | s->bs = bs; | |
1555 | s->aio_context = bdrv_get_aio_context(bs); | |
1556 | ||
1557 | opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); | |
1558 | qemu_opts_absorb_qdict(opts, options, &local_err); | |
1559 | if (local_err) { | |
1560 | error_propagate(errp, local_err); | |
1561 | ret = -EINVAL; | |
1562 | goto err_no_fd; | |
1563 | } | |
1564 | ||
1565 | s->addr = sd_server_config(options, errp); | |
1566 | if (!s->addr) { | |
1567 | ret = -EINVAL; | |
1568 | goto err_no_fd; | |
1569 | } | |
1570 | ||
1571 | vdi = qemu_opt_get(opts, "vdi"); | |
1572 | snap_id_str = qemu_opt_get(opts, "snap-id"); | |
1573 | snap_id = qemu_opt_get_number(opts, "snap-id", CURRENT_VDI_ID); | |
1574 | tag = qemu_opt_get(opts, "tag"); | |
1575 | ||
1576 | if (!vdi) { | |
1577 | error_setg(errp, "parameter 'vdi' is missing"); | |
1578 | ret = -EINVAL; | |
1579 | goto err_no_fd; | |
1580 | } | |
1581 | if (strlen(vdi) >= SD_MAX_VDI_LEN) { | |
1582 | error_setg(errp, "value of parameter 'vdi' is too long"); | |
1583 | ret = -EINVAL; | |
1584 | goto err_no_fd; | |
1585 | } | |
1586 | ||
1587 | if (snap_id > UINT32_MAX) { | |
1588 | snap_id = 0; | |
1589 | } | |
1590 | if (snap_id_str && !snap_id) { | |
1591 | error_setg(errp, "'snap-id=%s' is not a valid snapshot ID", | |
1592 | snap_id_str); | |
1593 | ret = -EINVAL; | |
1594 | goto err_no_fd; | |
1595 | } | |
1596 | ||
1597 | if (!tag) { | |
1598 | tag = ""; | |
1599 | } | |
1600 | if (strlen(tag) >= SD_MAX_VDI_TAG_LEN) { | |
1601 | error_setg(errp, "value of parameter 'tag' is too long"); | |
1602 | ret = -EINVAL; | |
1603 | goto err_no_fd; | |
1604 | } | |
1605 | ||
1606 | QLIST_INIT(&s->inflight_aio_head); | |
1607 | QLIST_INIT(&s->failed_aio_head); | |
1608 | QLIST_INIT(&s->inflight_aiocb_head); | |
1609 | ||
1610 | s->fd = get_sheep_fd(s, errp); | |
1611 | if (s->fd < 0) { | |
1612 | ret = s->fd; | |
1613 | goto err_no_fd; | |
1614 | } | |
1615 | ||
1616 | ret = find_vdi_name(s, vdi, (uint32_t)snap_id, tag, &vid, true, errp); | |
1617 | if (ret) { | |
1618 | goto err; | |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * QEMU block layer emulates writethrough cache as 'writeback + flush', so | |
1623 | * we always set SD_FLAG_CMD_CACHE (writeback cache) as default. | |
1624 | */ | |
1625 | s->cache_flags = SD_FLAG_CMD_CACHE; | |
1626 | if (flags & BDRV_O_NOCACHE) { | |
1627 | s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1628 | } | |
1629 | s->discard_supported = true; | |
1630 | ||
1631 | if (snap_id || tag[0]) { | |
1632 | trace_sheepdog_open(vid); | |
1633 | s->is_snapshot = true; | |
1634 | } | |
1635 | ||
1636 | fd = connect_to_sdog(s, errp); | |
1637 | if (fd < 0) { | |
1638 | ret = fd; | |
1639 | goto err; | |
1640 | } | |
1641 | ||
1642 | buf = g_malloc(SD_INODE_SIZE); | |
1643 | ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid), | |
1644 | 0, SD_INODE_SIZE, 0, s->cache_flags); | |
1645 | ||
1646 | closesocket(fd); | |
1647 | ||
1648 | if (ret) { | |
1649 | error_setg(errp, "Can't read snapshot inode"); | |
1650 | goto err; | |
1651 | } | |
1652 | ||
1653 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
1654 | ||
1655 | bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE; | |
1656 | pstrcpy(s->name, sizeof(s->name), vdi); | |
1657 | qemu_co_mutex_init(&s->lock); | |
1658 | qemu_co_mutex_init(&s->queue_lock); | |
1659 | qemu_co_queue_init(&s->overlapping_queue); | |
1660 | qemu_opts_del(opts); | |
1661 | g_free(buf); | |
1662 | return 0; | |
1663 | ||
1664 | err: | |
1665 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, | |
1666 | false, NULL, NULL, NULL, NULL); | |
1667 | closesocket(s->fd); | |
1668 | err_no_fd: | |
1669 | qemu_opts_del(opts); | |
1670 | g_free(buf); | |
1671 | return ret; | |
1672 | } | |
1673 | ||
1674 | static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue, | |
1675 | Error **errp) | |
1676 | { | |
1677 | BDRVSheepdogState *s = state->bs->opaque; | |
1678 | BDRVSheepdogReopenState *re_s; | |
1679 | int ret = 0; | |
1680 | ||
1681 | re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1); | |
1682 | ||
1683 | re_s->cache_flags = SD_FLAG_CMD_CACHE; | |
1684 | if (state->flags & BDRV_O_NOCACHE) { | |
1685 | re_s->cache_flags = SD_FLAG_CMD_DIRECT; | |
1686 | } | |
1687 | ||
1688 | re_s->fd = get_sheep_fd(s, errp); | |
1689 | if (re_s->fd < 0) { | |
1690 | ret = re_s->fd; | |
1691 | return ret; | |
1692 | } | |
1693 | ||
1694 | return ret; | |
1695 | } | |
1696 | ||
1697 | static void sd_reopen_commit(BDRVReopenState *state) | |
1698 | { | |
1699 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1700 | BDRVSheepdogState *s = state->bs->opaque; | |
1701 | ||
1702 | if (s->fd) { | |
1703 | aio_set_fd_handler(s->aio_context, s->fd, false, | |
1704 | NULL, NULL, NULL, NULL); | |
1705 | closesocket(s->fd); | |
1706 | } | |
1707 | ||
1708 | s->fd = re_s->fd; | |
1709 | s->cache_flags = re_s->cache_flags; | |
1710 | ||
1711 | g_free(state->opaque); | |
1712 | state->opaque = NULL; | |
1713 | ||
1714 | return; | |
1715 | } | |
1716 | ||
1717 | static void sd_reopen_abort(BDRVReopenState *state) | |
1718 | { | |
1719 | BDRVSheepdogReopenState *re_s = state->opaque; | |
1720 | BDRVSheepdogState *s = state->bs->opaque; | |
1721 | ||
1722 | if (re_s == NULL) { | |
1723 | return; | |
1724 | } | |
1725 | ||
1726 | if (re_s->fd) { | |
1727 | aio_set_fd_handler(s->aio_context, re_s->fd, false, | |
1728 | NULL, NULL, NULL, NULL); | |
1729 | closesocket(re_s->fd); | |
1730 | } | |
1731 | ||
1732 | g_free(state->opaque); | |
1733 | state->opaque = NULL; | |
1734 | ||
1735 | return; | |
1736 | } | |
1737 | ||
1738 | static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot, | |
1739 | Error **errp) | |
1740 | { | |
1741 | SheepdogVdiReq hdr; | |
1742 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
1743 | int fd, ret; | |
1744 | unsigned int wlen, rlen = 0; | |
1745 | char buf[SD_MAX_VDI_LEN]; | |
1746 | ||
1747 | fd = connect_to_sdog(s, errp); | |
1748 | if (fd < 0) { | |
1749 | return fd; | |
1750 | } | |
1751 | ||
1752 | /* FIXME: would it be better to fail (e.g., return -EIO) when filename | |
1753 | * does not fit in buf? For now, just truncate and avoid buffer overrun. | |
1754 | */ | |
1755 | memset(buf, 0, sizeof(buf)); | |
1756 | pstrcpy(buf, sizeof(buf), s->name); | |
1757 | ||
1758 | memset(&hdr, 0, sizeof(hdr)); | |
1759 | hdr.opcode = SD_OP_NEW_VDI; | |
1760 | hdr.base_vdi_id = s->inode.vdi_id; | |
1761 | ||
1762 | wlen = SD_MAX_VDI_LEN; | |
1763 | ||
1764 | hdr.flags = SD_FLAG_CMD_WRITE; | |
1765 | hdr.snapid = snapshot; | |
1766 | ||
1767 | hdr.data_length = wlen; | |
1768 | hdr.vdi_size = s->inode.vdi_size; | |
1769 | hdr.copy_policy = s->inode.copy_policy; | |
1770 | hdr.copies = s->inode.nr_copies; | |
1771 | hdr.block_size_shift = s->inode.block_size_shift; | |
1772 | ||
1773 | ret = do_req(fd, NULL, (SheepdogReq *)&hdr, buf, &wlen, &rlen); | |
1774 | ||
1775 | closesocket(fd); | |
1776 | ||
1777 | if (ret) { | |
1778 | error_setg_errno(errp, -ret, "create failed"); | |
1779 | return ret; | |
1780 | } | |
1781 | ||
1782 | if (rsp->result != SD_RES_SUCCESS) { | |
1783 | error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name); | |
1784 | return -EIO; | |
1785 | } | |
1786 | ||
1787 | if (vdi_id) { | |
1788 | *vdi_id = rsp->vdi_id; | |
1789 | } | |
1790 | ||
1791 | return 0; | |
1792 | } | |
1793 | ||
1794 | static int sd_prealloc(BlockDriverState *bs, int64_t old_size, int64_t new_size, | |
1795 | Error **errp) | |
1796 | { | |
1797 | BlockBackend *blk = NULL; | |
1798 | BDRVSheepdogState *base = bs->opaque; | |
1799 | unsigned long buf_size; | |
1800 | uint32_t idx, max_idx; | |
1801 | uint32_t object_size; | |
1802 | void *buf = NULL; | |
1803 | int ret; | |
1804 | ||
1805 | blk = blk_new(bdrv_get_aio_context(bs), | |
1806 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE, | |
1807 | BLK_PERM_ALL); | |
1808 | ||
1809 | ret = blk_insert_bs(blk, bs, errp); | |
1810 | if (ret < 0) { | |
1811 | goto out_with_err_set; | |
1812 | } | |
1813 | ||
1814 | blk_set_allow_write_beyond_eof(blk, true); | |
1815 | ||
1816 | object_size = (UINT32_C(1) << base->inode.block_size_shift); | |
1817 | buf_size = MIN(object_size, SD_DATA_OBJ_SIZE); | |
1818 | buf = g_malloc0(buf_size); | |
1819 | ||
1820 | max_idx = DIV_ROUND_UP(new_size, buf_size); | |
1821 | ||
1822 | for (idx = old_size / buf_size; idx < max_idx; idx++) { | |
1823 | /* | |
1824 | * The created image can be a cloned image, so we need to read | |
1825 | * a data from the source image. | |
1826 | */ | |
1827 | ret = blk_pread(blk, idx * buf_size, buf, buf_size); | |
1828 | if (ret < 0) { | |
1829 | goto out; | |
1830 | } | |
1831 | ret = blk_pwrite(blk, idx * buf_size, buf, buf_size, 0); | |
1832 | if (ret < 0) { | |
1833 | goto out; | |
1834 | } | |
1835 | } | |
1836 | ||
1837 | ret = 0; | |
1838 | out: | |
1839 | if (ret < 0) { | |
1840 | error_setg_errno(errp, -ret, "Can't pre-allocate"); | |
1841 | } | |
1842 | out_with_err_set: | |
1843 | blk_unref(blk); | |
1844 | g_free(buf); | |
1845 | ||
1846 | return ret; | |
1847 | } | |
1848 | ||
1849 | static int sd_create_prealloc(BlockdevOptionsSheepdog *location, int64_t size, | |
1850 | Error **errp) | |
1851 | { | |
1852 | BlockDriverState *bs; | |
1853 | Visitor *v; | |
1854 | QObject *obj = NULL; | |
1855 | QDict *qdict; | |
1856 | Error *local_err = NULL; | |
1857 | int ret; | |
1858 | ||
1859 | v = qobject_output_visitor_new(&obj); | |
1860 | visit_type_BlockdevOptionsSheepdog(v, NULL, &location, &local_err); | |
1861 | visit_free(v); | |
1862 | ||
1863 | if (local_err) { | |
1864 | error_propagate(errp, local_err); | |
1865 | qobject_unref(obj); | |
1866 | return -EINVAL; | |
1867 | } | |
1868 | ||
1869 | qdict = qobject_to(QDict, obj); | |
1870 | qdict_flatten(qdict); | |
1871 | ||
1872 | qdict_put_str(qdict, "driver", "sheepdog"); | |
1873 | ||
1874 | bs = bdrv_open(NULL, NULL, qdict, BDRV_O_PROTOCOL | BDRV_O_RDWR, errp); | |
1875 | if (bs == NULL) { | |
1876 | ret = -EIO; | |
1877 | goto fail; | |
1878 | } | |
1879 | ||
1880 | ret = sd_prealloc(bs, 0, size, errp); | |
1881 | fail: | |
1882 | bdrv_unref(bs); | |
1883 | qobject_unref(qdict); | |
1884 | return ret; | |
1885 | } | |
1886 | ||
1887 | static int parse_redundancy(BDRVSheepdogState *s, SheepdogRedundancy *opt) | |
1888 | { | |
1889 | struct SheepdogInode *inode = &s->inode; | |
1890 | ||
1891 | switch (opt->type) { | |
1892 | case SHEEPDOG_REDUNDANCY_TYPE_FULL: | |
1893 | if (opt->u.full.copies > SD_MAX_COPIES || opt->u.full.copies < 1) { | |
1894 | return -EINVAL; | |
1895 | } | |
1896 | inode->copy_policy = 0; | |
1897 | inode->nr_copies = opt->u.full.copies; | |
1898 | return 0; | |
1899 | ||
1900 | case SHEEPDOG_REDUNDANCY_TYPE_ERASURE_CODED: | |
1901 | { | |
1902 | int64_t copy = opt->u.erasure_coded.data_strips; | |
1903 | int64_t parity = opt->u.erasure_coded.parity_strips; | |
1904 | ||
1905 | if (copy != 2 && copy != 4 && copy != 8 && copy != 16) { | |
1906 | return -EINVAL; | |
1907 | } | |
1908 | ||
1909 | if (parity >= SD_EC_MAX_STRIP || parity < 1) { | |
1910 | return -EINVAL; | |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * 4 bits for parity and 4 bits for data. | |
1915 | * We have to compress upper data bits because it can't represent 16 | |
1916 | */ | |
1917 | inode->copy_policy = ((copy / 2) << 4) + parity; | |
1918 | inode->nr_copies = copy + parity; | |
1919 | return 0; | |
1920 | } | |
1921 | ||
1922 | default: | |
1923 | g_assert_not_reached(); | |
1924 | } | |
1925 | ||
1926 | return -EINVAL; | |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * Sheepdog support two kinds of redundancy, full replication and erasure | |
1931 | * coding. | |
1932 | * | |
1933 | * # create a fully replicated vdi with x copies | |
1934 | * -o redundancy=x (1 <= x <= SD_MAX_COPIES) | |
1935 | * | |
1936 | * # create a erasure coded vdi with x data strips and y parity strips | |
1937 | * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP) | |
1938 | */ | |
1939 | static SheepdogRedundancy *parse_redundancy_str(const char *opt) | |
1940 | { | |
1941 | SheepdogRedundancy *redundancy; | |
1942 | const char *n1, *n2; | |
1943 | long copy, parity; | |
1944 | char p[10]; | |
1945 | int ret; | |
1946 | ||
1947 | pstrcpy(p, sizeof(p), opt); | |
1948 | n1 = strtok(p, ":"); | |
1949 | n2 = strtok(NULL, ":"); | |
1950 | ||
1951 | if (!n1) { | |
1952 | return NULL; | |
1953 | } | |
1954 | ||
1955 | ret = qemu_strtol(n1, NULL, 10, ©); | |
1956 | if (ret < 0) { | |
1957 | return NULL; | |
1958 | } | |
1959 | ||
1960 | redundancy = g_new0(SheepdogRedundancy, 1); | |
1961 | if (!n2) { | |
1962 | *redundancy = (SheepdogRedundancy) { | |
1963 | .type = SHEEPDOG_REDUNDANCY_TYPE_FULL, | |
1964 | .u.full.copies = copy, | |
1965 | }; | |
1966 | } else { | |
1967 | ret = qemu_strtol(n2, NULL, 10, &parity); | |
1968 | if (ret < 0) { | |
1969 | g_free(redundancy); | |
1970 | return NULL; | |
1971 | } | |
1972 | ||
1973 | *redundancy = (SheepdogRedundancy) { | |
1974 | .type = SHEEPDOG_REDUNDANCY_TYPE_ERASURE_CODED, | |
1975 | .u.erasure_coded = { | |
1976 | .data_strips = copy, | |
1977 | .parity_strips = parity, | |
1978 | }, | |
1979 | }; | |
1980 | } | |
1981 | ||
1982 | return redundancy; | |
1983 | } | |
1984 | ||
1985 | static int parse_block_size_shift(BDRVSheepdogState *s, | |
1986 | BlockdevCreateOptionsSheepdog *opts) | |
1987 | { | |
1988 | struct SheepdogInode *inode = &s->inode; | |
1989 | uint64_t object_size; | |
1990 | int obj_order; | |
1991 | ||
1992 | if (opts->has_object_size) { | |
1993 | object_size = opts->object_size; | |
1994 | ||
1995 | if ((object_size - 1) & object_size) { /* not a power of 2? */ | |
1996 | return -EINVAL; | |
1997 | } | |
1998 | obj_order = ctz32(object_size); | |
1999 | if (obj_order < 20 || obj_order > 31) { | |
2000 | return -EINVAL; | |
2001 | } | |
2002 | inode->block_size_shift = (uint8_t)obj_order; | |
2003 | } | |
2004 | ||
2005 | return 0; | |
2006 | } | |
2007 | ||
2008 | static int sd_co_create(BlockdevCreateOptions *options, Error **errp) | |
2009 | { | |
2010 | BlockdevCreateOptionsSheepdog *opts = &options->u.sheepdog; | |
2011 | int ret = 0; | |
2012 | uint32_t vid = 0; | |
2013 | char *backing_file = NULL; | |
2014 | char *buf = NULL; | |
2015 | BDRVSheepdogState *s; | |
2016 | uint64_t max_vdi_size; | |
2017 | bool prealloc = false; | |
2018 | ||
2019 | assert(options->driver == BLOCKDEV_DRIVER_SHEEPDOG); | |
2020 | ||
2021 | s = g_new0(BDRVSheepdogState, 1); | |
2022 | ||
2023 | /* Steal SocketAddress from QAPI, set NULL to prevent double free */ | |
2024 | s->addr = opts->location->server; | |
2025 | opts->location->server = NULL; | |
2026 | ||
2027 | if (strlen(opts->location->vdi) >= sizeof(s->name)) { | |
2028 | error_setg(errp, "'vdi' string too long"); | |
2029 | ret = -EINVAL; | |
2030 | goto out; | |
2031 | } | |
2032 | pstrcpy(s->name, sizeof(s->name), opts->location->vdi); | |
2033 | ||
2034 | s->inode.vdi_size = opts->size; | |
2035 | backing_file = opts->backing_file; | |
2036 | ||
2037 | if (!opts->has_preallocation) { | |
2038 | opts->preallocation = PREALLOC_MODE_OFF; | |
2039 | } | |
2040 | switch (opts->preallocation) { | |
2041 | case PREALLOC_MODE_OFF: | |
2042 | prealloc = false; | |
2043 | break; | |
2044 | case PREALLOC_MODE_FULL: | |
2045 | prealloc = true; | |
2046 | break; | |
2047 | default: | |
2048 | error_setg(errp, "Preallocation mode not supported for Sheepdog"); | |
2049 | ret = -EINVAL; | |
2050 | goto out; | |
2051 | } | |
2052 | ||
2053 | if (opts->has_redundancy) { | |
2054 | ret = parse_redundancy(s, opts->redundancy); | |
2055 | if (ret < 0) { | |
2056 | error_setg(errp, "Invalid redundancy mode"); | |
2057 | goto out; | |
2058 | } | |
2059 | } | |
2060 | ret = parse_block_size_shift(s, opts); | |
2061 | if (ret < 0) { | |
2062 | error_setg(errp, "Invalid object_size." | |
2063 | " obect_size needs to be power of 2" | |
2064 | " and be limited from 2^20 to 2^31"); | |
2065 | goto out; | |
2066 | } | |
2067 | ||
2068 | if (opts->has_backing_file) { | |
2069 | BlockBackend *blk; | |
2070 | BDRVSheepdogState *base; | |
2071 | BlockDriver *drv; | |
2072 | ||
2073 | /* Currently, only Sheepdog backing image is supported. */ | |
2074 | drv = bdrv_find_protocol(opts->backing_file, true, NULL); | |
2075 | if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) { | |
2076 | error_setg(errp, "backing_file must be a sheepdog image"); | |
2077 | ret = -EINVAL; | |
2078 | goto out; | |
2079 | } | |
2080 | ||
2081 | blk = blk_new_open(opts->backing_file, NULL, NULL, | |
2082 | BDRV_O_PROTOCOL, errp); | |
2083 | if (blk == NULL) { | |
2084 | ret = -EIO; | |
2085 | goto out; | |
2086 | } | |
2087 | ||
2088 | base = blk_bs(blk)->opaque; | |
2089 | ||
2090 | if (!is_snapshot(&base->inode)) { | |
2091 | error_setg(errp, "cannot clone from a non snapshot vdi"); | |
2092 | blk_unref(blk); | |
2093 | ret = -EINVAL; | |
2094 | goto out; | |
2095 | } | |
2096 | s->inode.vdi_id = base->inode.vdi_id; | |
2097 | blk_unref(blk); | |
2098 | } | |
2099 | ||
2100 | s->aio_context = qemu_get_aio_context(); | |
2101 | ||
2102 | /* if block_size_shift is not specified, get cluster default value */ | |
2103 | if (s->inode.block_size_shift == 0) { | |
2104 | SheepdogVdiReq hdr; | |
2105 | SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr; | |
2106 | int fd; | |
2107 | unsigned int wlen = 0, rlen = 0; | |
2108 | ||
2109 | fd = connect_to_sdog(s, errp); | |
2110 | if (fd < 0) { | |
2111 | ret = fd; | |
2112 | goto out; | |
2113 | } | |
2114 | ||
2115 | memset(&hdr, 0, sizeof(hdr)); | |
2116 | hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT; | |
2117 | hdr.proto_ver = SD_PROTO_VER; | |
2118 | ||
2119 | ret = do_req(fd, NULL, (SheepdogReq *)&hdr, | |
2120 | NULL, &wlen, &rlen); | |
2121 | closesocket(fd); | |
2122 | if (ret) { | |
2123 | error_setg_errno(errp, -ret, "failed to get cluster default"); | |
2124 | goto out; | |
2125 | } | |
2126 | if (rsp->result == SD_RES_SUCCESS) { | |
2127 | s->inode.block_size_shift = rsp->block_size_shift; | |
2128 | } else { | |
2129 | s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT; | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; | |
2134 | ||
2135 | if (s->inode.vdi_size > max_vdi_size) { | |
2136 | error_setg(errp, "An image is too large." | |
2137 | " The maximum image size is %"PRIu64 "GB", | |
2138 | max_vdi_size / 1024 / 1024 / 1024); | |
2139 | ret = -EINVAL; | |
2140 | goto out; | |
2141 | } | |
2142 | ||
2143 | ret = do_sd_create(s, &vid, 0, errp); | |
2144 | if (ret) { | |
2145 | goto out; | |
2146 | } | |
2147 | ||
2148 | if (prealloc) { | |
2149 | ret = sd_create_prealloc(opts->location, opts->size, errp); | |
2150 | } | |
2151 | out: | |
2152 | g_free(backing_file); | |
2153 | g_free(buf); | |
2154 | g_free(s->addr); | |
2155 | g_free(s); | |
2156 | return ret; | |
2157 | } | |
2158 | ||
2159 | static int coroutine_fn sd_co_create_opts(const char *filename, QemuOpts *opts, | |
2160 | Error **errp) | |
2161 | { | |
2162 | BlockdevCreateOptions *create_options = NULL; | |
2163 | QDict *qdict, *location_qdict; | |
2164 | Visitor *v; | |
2165 | char *redundancy; | |
2166 | Error *local_err = NULL; | |
2167 | int ret; | |
2168 | ||
2169 | redundancy = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY); | |
2170 | ||
2171 | qdict = qemu_opts_to_qdict(opts, NULL); | |
2172 | qdict_put_str(qdict, "driver", "sheepdog"); | |
2173 | ||
2174 | location_qdict = qdict_new(); | |
2175 | qdict_put(qdict, "location", location_qdict); | |
2176 | ||
2177 | sd_parse_filename(filename, location_qdict, &local_err); | |
2178 | if (local_err) { | |
2179 | error_propagate(errp, local_err); | |
2180 | ret = -EINVAL; | |
2181 | goto fail; | |
2182 | } | |
2183 | ||
2184 | qdict_flatten(qdict); | |
2185 | ||
2186 | /* Change legacy command line options into QMP ones */ | |
2187 | static const QDictRenames opt_renames[] = { | |
2188 | { BLOCK_OPT_BACKING_FILE, "backing-file" }, | |
2189 | { BLOCK_OPT_OBJECT_SIZE, "object-size" }, | |
2190 | { NULL, NULL }, | |
2191 | }; | |
2192 | ||
2193 | if (!qdict_rename_keys(qdict, opt_renames, errp)) { | |
2194 | ret = -EINVAL; | |
2195 | goto fail; | |
2196 | } | |
2197 | ||
2198 | /* Get the QAPI object */ | |
2199 | v = qobject_input_visitor_new_flat_confused(qdict, errp); | |
2200 | if (!v) { | |
2201 | ret = -EINVAL; | |
2202 | goto fail; | |
2203 | } | |
2204 | ||
2205 | visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); | |
2206 | visit_free(v); | |
2207 | ||
2208 | if (local_err) { | |
2209 | error_propagate(errp, local_err); | |
2210 | ret = -EINVAL; | |
2211 | goto fail; | |
2212 | } | |
2213 | ||
2214 | assert(create_options->driver == BLOCKDEV_DRIVER_SHEEPDOG); | |
2215 | create_options->u.sheepdog.size = | |
2216 | ROUND_UP(create_options->u.sheepdog.size, BDRV_SECTOR_SIZE); | |
2217 | ||
2218 | if (redundancy) { | |
2219 | create_options->u.sheepdog.has_redundancy = true; | |
2220 | create_options->u.sheepdog.redundancy = | |
2221 | parse_redundancy_str(redundancy); | |
2222 | if (create_options->u.sheepdog.redundancy == NULL) { | |
2223 | error_setg(errp, "Invalid redundancy mode"); | |
2224 | ret = -EINVAL; | |
2225 | goto fail; | |
2226 | } | |
2227 | } | |
2228 | ||
2229 | ret = sd_co_create(create_options, errp); | |
2230 | fail: | |
2231 | qapi_free_BlockdevCreateOptions(create_options); | |
2232 | qobject_unref(qdict); | |
2233 | g_free(redundancy); | |
2234 | return ret; | |
2235 | } | |
2236 | ||
2237 | static void sd_close(BlockDriverState *bs) | |
2238 | { | |
2239 | Error *local_err = NULL; | |
2240 | BDRVSheepdogState *s = bs->opaque; | |
2241 | SheepdogVdiReq hdr; | |
2242 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2243 | unsigned int wlen, rlen = 0; | |
2244 | int fd, ret; | |
2245 | ||
2246 | trace_sheepdog_close(s->name); | |
2247 | ||
2248 | fd = connect_to_sdog(s, &local_err); | |
2249 | if (fd < 0) { | |
2250 | error_report_err(local_err); | |
2251 | return; | |
2252 | } | |
2253 | ||
2254 | memset(&hdr, 0, sizeof(hdr)); | |
2255 | ||
2256 | hdr.opcode = SD_OP_RELEASE_VDI; | |
2257 | hdr.type = LOCK_TYPE_NORMAL; | |
2258 | hdr.base_vdi_id = s->inode.vdi_id; | |
2259 | wlen = strlen(s->name) + 1; | |
2260 | hdr.data_length = wlen; | |
2261 | hdr.flags = SD_FLAG_CMD_WRITE; | |
2262 | ||
2263 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, | |
2264 | s->name, &wlen, &rlen); | |
2265 | ||
2266 | closesocket(fd); | |
2267 | ||
2268 | if (!ret && rsp->result != SD_RES_SUCCESS && | |
2269 | rsp->result != SD_RES_VDI_NOT_LOCKED) { | |
2270 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2271 | } | |
2272 | ||
2273 | aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd, | |
2274 | false, NULL, NULL, NULL, NULL); | |
2275 | closesocket(s->fd); | |
2276 | qapi_free_SocketAddress(s->addr); | |
2277 | } | |
2278 | ||
2279 | static int64_t sd_getlength(BlockDriverState *bs) | |
2280 | { | |
2281 | BDRVSheepdogState *s = bs->opaque; | |
2282 | ||
2283 | return s->inode.vdi_size; | |
2284 | } | |
2285 | ||
2286 | static int coroutine_fn sd_co_truncate(BlockDriverState *bs, int64_t offset, | |
2287 | PreallocMode prealloc, Error **errp) | |
2288 | { | |
2289 | BDRVSheepdogState *s = bs->opaque; | |
2290 | int ret, fd; | |
2291 | unsigned int datalen; | |
2292 | uint64_t max_vdi_size; | |
2293 | int64_t old_size = s->inode.vdi_size; | |
2294 | ||
2295 | if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_FULL) { | |
2296 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
2297 | PreallocMode_str(prealloc)); | |
2298 | return -ENOTSUP; | |
2299 | } | |
2300 | ||
2301 | max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS; | |
2302 | if (offset < old_size) { | |
2303 | error_setg(errp, "shrinking is not supported"); | |
2304 | return -EINVAL; | |
2305 | } else if (offset > max_vdi_size) { | |
2306 | error_setg(errp, "too big image size"); | |
2307 | return -EINVAL; | |
2308 | } | |
2309 | ||
2310 | fd = connect_to_sdog(s, errp); | |
2311 | if (fd < 0) { | |
2312 | return fd; | |
2313 | } | |
2314 | ||
2315 | /* we don't need to update entire object */ | |
2316 | datalen = SD_INODE_HEADER_SIZE; | |
2317 | s->inode.vdi_size = offset; | |
2318 | ret = write_object(fd, s->bs, (char *)&s->inode, | |
2319 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, | |
2320 | datalen, 0, false, s->cache_flags); | |
2321 | close(fd); | |
2322 | ||
2323 | if (ret < 0) { | |
2324 | error_setg_errno(errp, -ret, "failed to update an inode"); | |
2325 | return ret; | |
2326 | } | |
2327 | ||
2328 | if (prealloc == PREALLOC_MODE_FULL) { | |
2329 | ret = sd_prealloc(bs, old_size, offset, errp); | |
2330 | if (ret < 0) { | |
2331 | return ret; | |
2332 | } | |
2333 | } | |
2334 | ||
2335 | return 0; | |
2336 | } | |
2337 | ||
2338 | /* | |
2339 | * This function is called after writing data objects. If we need to | |
2340 | * update metadata, this sends a write request to the vdi object. | |
2341 | */ | |
2342 | static void coroutine_fn sd_write_done(SheepdogAIOCB *acb) | |
2343 | { | |
2344 | BDRVSheepdogState *s = acb->s; | |
2345 | struct iovec iov; | |
2346 | AIOReq *aio_req; | |
2347 | uint32_t offset, data_len, mn, mx; | |
2348 | ||
2349 | mn = acb->min_dirty_data_idx; | |
2350 | mx = acb->max_dirty_data_idx; | |
2351 | if (mn <= mx) { | |
2352 | /* we need to update the vdi object. */ | |
2353 | ++acb->nr_pending; | |
2354 | offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) + | |
2355 | mn * sizeof(s->inode.data_vdi_id[0]); | |
2356 | data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]); | |
2357 | ||
2358 | acb->min_dirty_data_idx = UINT32_MAX; | |
2359 | acb->max_dirty_data_idx = 0; | |
2360 | ||
2361 | iov.iov_base = &s->inode; | |
2362 | iov.iov_len = sizeof(s->inode); | |
2363 | aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id), | |
2364 | data_len, offset, 0, false, 0, offset); | |
2365 | add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA); | |
2366 | if (--acb->nr_pending) { | |
2367 | qemu_coroutine_yield(); | |
2368 | } | |
2369 | } | |
2370 | } | |
2371 | ||
2372 | /* Delete current working VDI on the snapshot chain */ | |
2373 | static bool sd_delete(BDRVSheepdogState *s) | |
2374 | { | |
2375 | Error *local_err = NULL; | |
2376 | unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0; | |
2377 | SheepdogVdiReq hdr = { | |
2378 | .opcode = SD_OP_DEL_VDI, | |
2379 | .base_vdi_id = s->inode.vdi_id, | |
2380 | .data_length = wlen, | |
2381 | .flags = SD_FLAG_CMD_WRITE, | |
2382 | }; | |
2383 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2384 | int fd, ret; | |
2385 | ||
2386 | fd = connect_to_sdog(s, &local_err); | |
2387 | if (fd < 0) { | |
2388 | error_report_err(local_err); | |
2389 | return false; | |
2390 | } | |
2391 | ||
2392 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, | |
2393 | s->name, &wlen, &rlen); | |
2394 | closesocket(fd); | |
2395 | if (ret) { | |
2396 | return false; | |
2397 | } | |
2398 | switch (rsp->result) { | |
2399 | case SD_RES_NO_VDI: | |
2400 | error_report("%s was already deleted", s->name); | |
2401 | /* fall through */ | |
2402 | case SD_RES_SUCCESS: | |
2403 | break; | |
2404 | default: | |
2405 | error_report("%s, %s", sd_strerror(rsp->result), s->name); | |
2406 | return false; | |
2407 | } | |
2408 | ||
2409 | return true; | |
2410 | } | |
2411 | ||
2412 | /* | |
2413 | * Create a writable VDI from a snapshot | |
2414 | */ | |
2415 | static int sd_create_branch(BDRVSheepdogState *s) | |
2416 | { | |
2417 | Error *local_err = NULL; | |
2418 | int ret, fd; | |
2419 | uint32_t vid; | |
2420 | char *buf; | |
2421 | bool deleted; | |
2422 | ||
2423 | trace_sheepdog_create_branch_snapshot(s->inode.vdi_id); | |
2424 | ||
2425 | buf = g_malloc(SD_INODE_SIZE); | |
2426 | ||
2427 | /* | |
2428 | * Even If deletion fails, we will just create extra snapshot based on | |
2429 | * the working VDI which was supposed to be deleted. So no need to | |
2430 | * false bail out. | |
2431 | */ | |
2432 | deleted = sd_delete(s); | |
2433 | ret = do_sd_create(s, &vid, !deleted, &local_err); | |
2434 | if (ret) { | |
2435 | error_report_err(local_err); | |
2436 | goto out; | |
2437 | } | |
2438 | ||
2439 | trace_sheepdog_create_branch_created(vid); | |
2440 | ||
2441 | fd = connect_to_sdog(s, &local_err); | |
2442 | if (fd < 0) { | |
2443 | error_report_err(local_err); | |
2444 | ret = fd; | |
2445 | goto out; | |
2446 | } | |
2447 | ||
2448 | ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid), | |
2449 | s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags); | |
2450 | ||
2451 | closesocket(fd); | |
2452 | ||
2453 | if (ret < 0) { | |
2454 | goto out; | |
2455 | } | |
2456 | ||
2457 | memcpy(&s->inode, buf, sizeof(s->inode)); | |
2458 | ||
2459 | s->is_snapshot = false; | |
2460 | ret = 0; | |
2461 | trace_sheepdog_create_branch_new(s->inode.vdi_id); | |
2462 | ||
2463 | out: | |
2464 | g_free(buf); | |
2465 | ||
2466 | return ret; | |
2467 | } | |
2468 | ||
2469 | /* | |
2470 | * Send I/O requests to the server. | |
2471 | * | |
2472 | * This function sends requests to the server, links the requests to | |
2473 | * the inflight_list in BDRVSheepdogState, and exits without | |
2474 | * waiting the response. The responses are received in the | |
2475 | * `aio_read_response' function which is called from the main loop as | |
2476 | * a fd handler. | |
2477 | * | |
2478 | * Returns 1 when we need to wait a response, 0 when there is no sent | |
2479 | * request and -errno in error cases. | |
2480 | */ | |
2481 | static void coroutine_fn sd_co_rw_vector(SheepdogAIOCB *acb) | |
2482 | { | |
2483 | int ret = 0; | |
2484 | unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE; | |
2485 | unsigned long idx; | |
2486 | uint32_t object_size; | |
2487 | uint64_t oid; | |
2488 | uint64_t offset; | |
2489 | BDRVSheepdogState *s = acb->s; | |
2490 | SheepdogInode *inode = &s->inode; | |
2491 | AIOReq *aio_req; | |
2492 | ||
2493 | if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) { | |
2494 | /* | |
2495 | * In the case we open the snapshot VDI, Sheepdog creates the | |
2496 | * writable VDI when we do a write operation first. | |
2497 | */ | |
2498 | ret = sd_create_branch(s); | |
2499 | if (ret) { | |
2500 | acb->ret = -EIO; | |
2501 | return; | |
2502 | } | |
2503 | } | |
2504 | ||
2505 | object_size = (UINT32_C(1) << inode->block_size_shift); | |
2506 | idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size; | |
2507 | offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size; | |
2508 | ||
2509 | /* | |
2510 | * Make sure we don't free the aiocb before we are done with all requests. | |
2511 | * This additional reference is dropped at the end of this function. | |
2512 | */ | |
2513 | acb->nr_pending++; | |
2514 | ||
2515 | while (done != total) { | |
2516 | uint8_t flags = 0; | |
2517 | uint64_t old_oid = 0; | |
2518 | bool create = false; | |
2519 | ||
2520 | oid = vid_to_data_oid(inode->data_vdi_id[idx], idx); | |
2521 | ||
2522 | len = MIN(total - done, object_size - offset); | |
2523 | ||
2524 | switch (acb->aiocb_type) { | |
2525 | case AIOCB_READ_UDATA: | |
2526 | if (!inode->data_vdi_id[idx]) { | |
2527 | qemu_iovec_memset(acb->qiov, done, 0, len); | |
2528 | goto done; | |
2529 | } | |
2530 | break; | |
2531 | case AIOCB_WRITE_UDATA: | |
2532 | if (!inode->data_vdi_id[idx]) { | |
2533 | create = true; | |
2534 | } else if (!is_data_obj_writable(inode, idx)) { | |
2535 | /* Copy-On-Write */ | |
2536 | create = true; | |
2537 | old_oid = oid; | |
2538 | flags = SD_FLAG_CMD_COW; | |
2539 | } | |
2540 | break; | |
2541 | case AIOCB_DISCARD_OBJ: | |
2542 | /* | |
2543 | * We discard the object only when the whole object is | |
2544 | * 1) allocated 2) trimmed. Otherwise, simply skip it. | |
2545 | */ | |
2546 | if (len != object_size || inode->data_vdi_id[idx] == 0) { | |
2547 | goto done; | |
2548 | } | |
2549 | break; | |
2550 | default: | |
2551 | break; | |
2552 | } | |
2553 | ||
2554 | if (create) { | |
2555 | trace_sheepdog_co_rw_vector_update(inode->vdi_id, oid, | |
2556 | vid_to_data_oid(inode->data_vdi_id[idx], idx), | |
2557 | idx); | |
2558 | oid = vid_to_data_oid(inode->vdi_id, idx); | |
2559 | trace_sheepdog_co_rw_vector_new(oid); | |
2560 | } | |
2561 | ||
2562 | aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create, | |
2563 | old_oid, | |
2564 | acb->aiocb_type == AIOCB_DISCARD_OBJ ? | |
2565 | 0 : done); | |
2566 | add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov, | |
2567 | acb->aiocb_type); | |
2568 | done: | |
2569 | offset = 0; | |
2570 | idx++; | |
2571 | done += len; | |
2572 | } | |
2573 | if (--acb->nr_pending) { | |
2574 | qemu_coroutine_yield(); | |
2575 | } | |
2576 | } | |
2577 | ||
2578 | static void sd_aio_complete(SheepdogAIOCB *acb) | |
2579 | { | |
2580 | BDRVSheepdogState *s; | |
2581 | if (acb->aiocb_type == AIOCB_FLUSH_CACHE) { | |
2582 | return; | |
2583 | } | |
2584 | ||
2585 | s = acb->s; | |
2586 | qemu_co_mutex_lock(&s->queue_lock); | |
2587 | QLIST_REMOVE(acb, aiocb_siblings); | |
2588 | qemu_co_queue_restart_all(&s->overlapping_queue); | |
2589 | qemu_co_mutex_unlock(&s->queue_lock); | |
2590 | } | |
2591 | ||
2592 | static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num, | |
2593 | int nb_sectors, QEMUIOVector *qiov, | |
2594 | int flags) | |
2595 | { | |
2596 | SheepdogAIOCB acb; | |
2597 | int ret; | |
2598 | int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE; | |
2599 | BDRVSheepdogState *s = bs->opaque; | |
2600 | ||
2601 | assert(!flags); | |
2602 | if (offset > s->inode.vdi_size) { | |
2603 | ret = sd_co_truncate(bs, offset, PREALLOC_MODE_OFF, NULL); | |
2604 | if (ret < 0) { | |
2605 | return ret; | |
2606 | } | |
2607 | } | |
2608 | ||
2609 | sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_WRITE_UDATA); | |
2610 | sd_co_rw_vector(&acb); | |
2611 | sd_write_done(&acb); | |
2612 | sd_aio_complete(&acb); | |
2613 | ||
2614 | return acb.ret; | |
2615 | } | |
2616 | ||
2617 | static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num, | |
2618 | int nb_sectors, QEMUIOVector *qiov) | |
2619 | { | |
2620 | SheepdogAIOCB acb; | |
2621 | BDRVSheepdogState *s = bs->opaque; | |
2622 | ||
2623 | sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_READ_UDATA); | |
2624 | sd_co_rw_vector(&acb); | |
2625 | sd_aio_complete(&acb); | |
2626 | ||
2627 | return acb.ret; | |
2628 | } | |
2629 | ||
2630 | static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs) | |
2631 | { | |
2632 | BDRVSheepdogState *s = bs->opaque; | |
2633 | SheepdogAIOCB acb; | |
2634 | AIOReq *aio_req; | |
2635 | ||
2636 | if (s->cache_flags != SD_FLAG_CMD_CACHE) { | |
2637 | return 0; | |
2638 | } | |
2639 | ||
2640 | sd_aio_setup(&acb, s, NULL, 0, 0, AIOCB_FLUSH_CACHE); | |
2641 | ||
2642 | acb.nr_pending++; | |
2643 | aio_req = alloc_aio_req(s, &acb, vid_to_vdi_oid(s->inode.vdi_id), | |
2644 | 0, 0, 0, false, 0, 0); | |
2645 | add_aio_request(s, aio_req, NULL, 0, acb.aiocb_type); | |
2646 | ||
2647 | if (--acb.nr_pending) { | |
2648 | qemu_coroutine_yield(); | |
2649 | } | |
2650 | ||
2651 | sd_aio_complete(&acb); | |
2652 | return acb.ret; | |
2653 | } | |
2654 | ||
2655 | static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info) | |
2656 | { | |
2657 | Error *local_err = NULL; | |
2658 | BDRVSheepdogState *s = bs->opaque; | |
2659 | int ret, fd; | |
2660 | uint32_t new_vid; | |
2661 | SheepdogInode *inode; | |
2662 | unsigned int datalen; | |
2663 | ||
2664 | trace_sheepdog_snapshot_create_info(sn_info->name, sn_info->id_str, s->name, | |
2665 | sn_info->vm_state_size, s->is_snapshot); | |
2666 | ||
2667 | if (s->is_snapshot) { | |
2668 | error_report("You can't create a snapshot of a snapshot VDI, " | |
2669 | "%s (%" PRIu32 ").", s->name, s->inode.vdi_id); | |
2670 | ||
2671 | return -EINVAL; | |
2672 | } | |
2673 | ||
2674 | trace_sheepdog_snapshot_create(sn_info->name, sn_info->id_str); | |
2675 | ||
2676 | s->inode.vm_state_size = sn_info->vm_state_size; | |
2677 | s->inode.vm_clock_nsec = sn_info->vm_clock_nsec; | |
2678 | /* It appears that inode.tag does not require a NUL terminator, | |
2679 | * which means this use of strncpy is ok. | |
2680 | */ | |
2681 | strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag)); | |
2682 | /* we don't need to update entire object */ | |
2683 | datalen = SD_INODE_HEADER_SIZE; | |
2684 | inode = g_malloc(datalen); | |
2685 | ||
2686 | /* refresh inode. */ | |
2687 | fd = connect_to_sdog(s, &local_err); | |
2688 | if (fd < 0) { | |
2689 | error_report_err(local_err); | |
2690 | ret = fd; | |
2691 | goto cleanup; | |
2692 | } | |
2693 | ||
2694 | ret = write_object(fd, s->bs, (char *)&s->inode, | |
2695 | vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies, | |
2696 | datalen, 0, false, s->cache_flags); | |
2697 | if (ret < 0) { | |
2698 | error_report("failed to write snapshot's inode."); | |
2699 | goto cleanup; | |
2700 | } | |
2701 | ||
2702 | ret = do_sd_create(s, &new_vid, 1, &local_err); | |
2703 | if (ret < 0) { | |
2704 | error_reportf_err(local_err, | |
2705 | "failed to create inode for snapshot: "); | |
2706 | goto cleanup; | |
2707 | } | |
2708 | ||
2709 | ret = read_object(fd, s->bs, (char *)inode, | |
2710 | vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0, | |
2711 | s->cache_flags); | |
2712 | ||
2713 | if (ret < 0) { | |
2714 | error_report("failed to read new inode info. %s", strerror(errno)); | |
2715 | goto cleanup; | |
2716 | } | |
2717 | ||
2718 | memcpy(&s->inode, inode, datalen); | |
2719 | trace_sheepdog_snapshot_create_inode(s->inode.name, s->inode.snap_id, | |
2720 | s->inode.vdi_id); | |
2721 | ||
2722 | cleanup: | |
2723 | g_free(inode); | |
2724 | closesocket(fd); | |
2725 | return ret; | |
2726 | } | |
2727 | ||
2728 | /* | |
2729 | * We implement rollback(loadvm) operation to the specified snapshot by | |
2730 | * 1) switch to the snapshot | |
2731 | * 2) rely on sd_create_branch to delete working VDI and | |
2732 | * 3) create a new working VDI based on the specified snapshot | |
2733 | */ | |
2734 | static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id) | |
2735 | { | |
2736 | BDRVSheepdogState *s = bs->opaque; | |
2737 | BDRVSheepdogState *old_s; | |
2738 | char tag[SD_MAX_VDI_TAG_LEN]; | |
2739 | uint32_t snapid = 0; | |
2740 | int ret; | |
2741 | ||
2742 | if (!sd_parse_snapid_or_tag(snapshot_id, &snapid, tag)) { | |
2743 | return -EINVAL; | |
2744 | } | |
2745 | ||
2746 | old_s = g_new(BDRVSheepdogState, 1); | |
2747 | ||
2748 | memcpy(old_s, s, sizeof(BDRVSheepdogState)); | |
2749 | ||
2750 | ret = reload_inode(s, snapid, tag); | |
2751 | if (ret) { | |
2752 | goto out; | |
2753 | } | |
2754 | ||
2755 | ret = sd_create_branch(s); | |
2756 | if (ret) { | |
2757 | goto out; | |
2758 | } | |
2759 | ||
2760 | g_free(old_s); | |
2761 | ||
2762 | return 0; | |
2763 | out: | |
2764 | /* recover bdrv_sd_state */ | |
2765 | memcpy(s, old_s, sizeof(BDRVSheepdogState)); | |
2766 | g_free(old_s); | |
2767 | ||
2768 | error_report("failed to open. recover old bdrv_sd_state."); | |
2769 | ||
2770 | return ret; | |
2771 | } | |
2772 | ||
2773 | #define NR_BATCHED_DISCARD 128 | |
2774 | ||
2775 | static int remove_objects(BDRVSheepdogState *s, Error **errp) | |
2776 | { | |
2777 | int fd, i = 0, nr_objs = 0; | |
2778 | int ret; | |
2779 | SheepdogInode *inode = &s->inode; | |
2780 | ||
2781 | fd = connect_to_sdog(s, errp); | |
2782 | if (fd < 0) { | |
2783 | return fd; | |
2784 | } | |
2785 | ||
2786 | nr_objs = count_data_objs(inode); | |
2787 | while (i < nr_objs) { | |
2788 | int start_idx, nr_filled_idx; | |
2789 | ||
2790 | while (i < nr_objs && !inode->data_vdi_id[i]) { | |
2791 | i++; | |
2792 | } | |
2793 | start_idx = i; | |
2794 | ||
2795 | nr_filled_idx = 0; | |
2796 | while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) { | |
2797 | if (inode->data_vdi_id[i]) { | |
2798 | inode->data_vdi_id[i] = 0; | |
2799 | nr_filled_idx++; | |
2800 | } | |
2801 | ||
2802 | i++; | |
2803 | } | |
2804 | ||
2805 | ret = write_object(fd, s->bs, | |
2806 | (char *)&inode->data_vdi_id[start_idx], | |
2807 | vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies, | |
2808 | (i - start_idx) * sizeof(uint32_t), | |
2809 | offsetof(struct SheepdogInode, | |
2810 | data_vdi_id[start_idx]), | |
2811 | false, s->cache_flags); | |
2812 | if (ret < 0) { | |
2813 | error_setg(errp, "Failed to discard snapshot inode"); | |
2814 | goto out; | |
2815 | } | |
2816 | } | |
2817 | ||
2818 | ret = 0; | |
2819 | out: | |
2820 | closesocket(fd); | |
2821 | return ret; | |
2822 | } | |
2823 | ||
2824 | static int sd_snapshot_delete(BlockDriverState *bs, | |
2825 | const char *snapshot_id, | |
2826 | const char *name, | |
2827 | Error **errp) | |
2828 | { | |
2829 | /* | |
2830 | * FIXME should delete the snapshot matching both @snapshot_id and | |
2831 | * @name, but @name not used here | |
2832 | */ | |
2833 | unsigned long snap_id = 0; | |
2834 | char snap_tag[SD_MAX_VDI_TAG_LEN]; | |
2835 | int fd, ret; | |
2836 | char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN]; | |
2837 | BDRVSheepdogState *s = bs->opaque; | |
2838 | unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0; | |
2839 | uint32_t vid; | |
2840 | SheepdogVdiReq hdr = { | |
2841 | .opcode = SD_OP_DEL_VDI, | |
2842 | .data_length = wlen, | |
2843 | .flags = SD_FLAG_CMD_WRITE, | |
2844 | }; | |
2845 | SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr; | |
2846 | ||
2847 | ret = remove_objects(s, errp); | |
2848 | if (ret) { | |
2849 | return ret; | |
2850 | } | |
2851 | ||
2852 | memset(buf, 0, sizeof(buf)); | |
2853 | memset(snap_tag, 0, sizeof(snap_tag)); | |
2854 | pstrcpy(buf, SD_MAX_VDI_LEN, s->name); | |
2855 | /* TODO Use sd_parse_snapid() once this mess is cleaned up */ | |
2856 | ret = qemu_strtoul(snapshot_id, NULL, 10, &snap_id); | |
2857 | if (ret || snap_id > UINT32_MAX) { | |
2858 | /* | |
2859 | * FIXME Since qemu_strtoul() returns -EINVAL when | |
2860 | * @snapshot_id is null, @snapshot_id is mandatory. Correct | |
2861 | * would be to require at least one of @snapshot_id and @name. | |
2862 | */ | |
2863 | error_setg(errp, "Invalid snapshot ID: %s", | |
2864 | snapshot_id ? snapshot_id : "<null>"); | |
2865 | return -EINVAL; | |
2866 | } | |
2867 | ||
2868 | if (snap_id) { | |
2869 | hdr.snapid = (uint32_t) snap_id; | |
2870 | } else { | |
2871 | /* FIXME I suspect we should use @name here */ | |
2872 | /* FIXME don't truncate silently */ | |
2873 | pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id); | |
2874 | pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag); | |
2875 | } | |
2876 | ||
2877 | ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, errp); | |
2878 | if (ret) { | |
2879 | return ret; | |
2880 | } | |
2881 | ||
2882 | fd = connect_to_sdog(s, errp); | |
2883 | if (fd < 0) { | |
2884 | return fd; | |
2885 | } | |
2886 | ||
2887 | ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, | |
2888 | buf, &wlen, &rlen); | |
2889 | closesocket(fd); | |
2890 | if (ret) { | |
2891 | error_setg_errno(errp, -ret, "Couldn't send request to server"); | |
2892 | return ret; | |
2893 | } | |
2894 | ||
2895 | switch (rsp->result) { | |
2896 | case SD_RES_NO_VDI: | |
2897 | error_setg(errp, "Can't find the snapshot"); | |
2898 | return -ENOENT; | |
2899 | case SD_RES_SUCCESS: | |
2900 | break; | |
2901 | default: | |
2902 | error_setg(errp, "%s", sd_strerror(rsp->result)); | |
2903 | return -EIO; | |
2904 | } | |
2905 | ||
2906 | return 0; | |
2907 | } | |
2908 | ||
2909 | static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab) | |
2910 | { | |
2911 | Error *local_err = NULL; | |
2912 | BDRVSheepdogState *s = bs->opaque; | |
2913 | SheepdogReq req; | |
2914 | int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long); | |
2915 | QEMUSnapshotInfo *sn_tab = NULL; | |
2916 | unsigned wlen, rlen; | |
2917 | int found = 0; | |
2918 | SheepdogInode *inode; | |
2919 | unsigned long *vdi_inuse; | |
2920 | unsigned int start_nr; | |
2921 | uint64_t hval; | |
2922 | uint32_t vid; | |
2923 | ||
2924 | vdi_inuse = g_malloc(max); | |
2925 | inode = g_malloc(SD_INODE_HEADER_SIZE); | |
2926 | ||
2927 | fd = connect_to_sdog(s, &local_err); | |
2928 | if (fd < 0) { | |
2929 | error_report_err(local_err); | |
2930 | ret = fd; | |
2931 | goto out; | |
2932 | } | |
2933 | ||
2934 | rlen = max; | |
2935 | wlen = 0; | |
2936 | ||
2937 | memset(&req, 0, sizeof(req)); | |
2938 | ||
2939 | req.opcode = SD_OP_READ_VDIS; | |
2940 | req.data_length = max; | |
2941 | ||
2942 | ret = do_req(fd, s->bs, &req, vdi_inuse, &wlen, &rlen); | |
2943 | ||
2944 | closesocket(fd); | |
2945 | if (ret) { | |
2946 | goto out; | |
2947 | } | |
2948 | ||
2949 | sn_tab = g_new0(QEMUSnapshotInfo, nr); | |
2950 | ||
2951 | /* calculate a vdi id with hash function */ | |
2952 | hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT); | |
2953 | start_nr = hval & (SD_NR_VDIS - 1); | |
2954 | ||
2955 | fd = connect_to_sdog(s, &local_err); | |
2956 | if (fd < 0) { | |
2957 | error_report_err(local_err); | |
2958 | ret = fd; | |
2959 | goto out; | |
2960 | } | |
2961 | ||
2962 | for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) { | |
2963 | if (!test_bit(vid, vdi_inuse)) { | |
2964 | break; | |
2965 | } | |
2966 | ||
2967 | /* we don't need to read entire object */ | |
2968 | ret = read_object(fd, s->bs, (char *)inode, | |
2969 | vid_to_vdi_oid(vid), | |
2970 | 0, SD_INODE_HEADER_SIZE, 0, | |
2971 | s->cache_flags); | |
2972 | ||
2973 | if (ret) { | |
2974 | continue; | |
2975 | } | |
2976 | ||
2977 | if (!strcmp(inode->name, s->name) && is_snapshot(inode)) { | |
2978 | sn_tab[found].date_sec = inode->snap_ctime >> 32; | |
2979 | sn_tab[found].date_nsec = inode->snap_ctime & 0xffffffff; | |
2980 | sn_tab[found].vm_state_size = inode->vm_state_size; | |
2981 | sn_tab[found].vm_clock_nsec = inode->vm_clock_nsec; | |
2982 | ||
2983 | snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str), | |
2984 | "%" PRIu32, inode->snap_id); | |
2985 | pstrcpy(sn_tab[found].name, | |
2986 | MIN(sizeof(sn_tab[found].name), sizeof(inode->tag)), | |
2987 | inode->tag); | |
2988 | found++; | |
2989 | } | |
2990 | } | |
2991 | ||
2992 | closesocket(fd); | |
2993 | out: | |
2994 | *psn_tab = sn_tab; | |
2995 | ||
2996 | g_free(vdi_inuse); | |
2997 | g_free(inode); | |
2998 | ||
2999 | if (ret < 0) { | |
3000 | return ret; | |
3001 | } | |
3002 | ||
3003 | return found; | |
3004 | } | |
3005 | ||
3006 | static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data, | |
3007 | int64_t pos, int size, int load) | |
3008 | { | |
3009 | Error *local_err = NULL; | |
3010 | bool create; | |
3011 | int fd, ret = 0, remaining = size; | |
3012 | unsigned int data_len; | |
3013 | uint64_t vmstate_oid; | |
3014 | uint64_t offset; | |
3015 | uint32_t vdi_index; | |
3016 | uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id; | |
3017 | uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift); | |
3018 | ||
3019 | fd = connect_to_sdog(s, &local_err); | |
3020 | if (fd < 0) { | |
3021 | error_report_err(local_err); | |
3022 | return fd; | |
3023 | } | |
3024 | ||
3025 | while (remaining) { | |
3026 | vdi_index = pos / object_size; | |
3027 | offset = pos % object_size; | |
3028 | ||
3029 | data_len = MIN(remaining, object_size - offset); | |
3030 | ||
3031 | vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index); | |
3032 | ||
3033 | create = (offset == 0); | |
3034 | if (load) { | |
3035 | ret = read_object(fd, s->bs, (char *)data, vmstate_oid, | |
3036 | s->inode.nr_copies, data_len, offset, | |
3037 | s->cache_flags); | |
3038 | } else { | |
3039 | ret = write_object(fd, s->bs, (char *)data, vmstate_oid, | |
3040 | s->inode.nr_copies, data_len, offset, create, | |
3041 | s->cache_flags); | |
3042 | } | |
3043 | ||
3044 | if (ret < 0) { | |
3045 | error_report("failed to save vmstate %s", strerror(errno)); | |
3046 | goto cleanup; | |
3047 | } | |
3048 | ||
3049 | pos += data_len; | |
3050 | data += data_len; | |
3051 | remaining -= data_len; | |
3052 | } | |
3053 | ret = size; | |
3054 | cleanup: | |
3055 | closesocket(fd); | |
3056 | return ret; | |
3057 | } | |
3058 | ||
3059 | static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, | |
3060 | int64_t pos) | |
3061 | { | |
3062 | BDRVSheepdogState *s = bs->opaque; | |
3063 | void *buf; | |
3064 | int ret; | |
3065 | ||
3066 | buf = qemu_blockalign(bs, qiov->size); | |
3067 | qemu_iovec_to_buf(qiov, 0, buf, qiov->size); | |
3068 | ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0); | |
3069 | qemu_vfree(buf); | |
3070 | ||
3071 | return ret; | |
3072 | } | |
3073 | ||
3074 | static int sd_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, | |
3075 | int64_t pos) | |
3076 | { | |
3077 | BDRVSheepdogState *s = bs->opaque; | |
3078 | void *buf; | |
3079 | int ret; | |
3080 | ||
3081 | buf = qemu_blockalign(bs, qiov->size); | |
3082 | ret = do_load_save_vmstate(s, buf, pos, qiov->size, 1); | |
3083 | qemu_iovec_from_buf(qiov, 0, buf, qiov->size); | |
3084 | qemu_vfree(buf); | |
3085 | ||
3086 | return ret; | |
3087 | } | |
3088 | ||
3089 | ||
3090 | static coroutine_fn int sd_co_pdiscard(BlockDriverState *bs, int64_t offset, | |
3091 | int bytes) | |
3092 | { | |
3093 | SheepdogAIOCB acb; | |
3094 | BDRVSheepdogState *s = bs->opaque; | |
3095 | QEMUIOVector discard_iov; | |
3096 | struct iovec iov; | |
3097 | uint32_t zero = 0; | |
3098 | ||
3099 | if (!s->discard_supported) { | |
3100 | return 0; | |
3101 | } | |
3102 | ||
3103 | memset(&discard_iov, 0, sizeof(discard_iov)); | |
3104 | memset(&iov, 0, sizeof(iov)); | |
3105 | iov.iov_base = &zero; | |
3106 | iov.iov_len = sizeof(zero); | |
3107 | discard_iov.iov = &iov; | |
3108 | discard_iov.niov = 1; | |
3109 | if (!QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE)) { | |
3110 | return -ENOTSUP; | |
3111 | } | |
3112 | sd_aio_setup(&acb, s, &discard_iov, offset >> BDRV_SECTOR_BITS, | |
3113 | bytes >> BDRV_SECTOR_BITS, AIOCB_DISCARD_OBJ); | |
3114 | sd_co_rw_vector(&acb); | |
3115 | sd_aio_complete(&acb); | |
3116 | ||
3117 | return acb.ret; | |
3118 | } | |
3119 | ||
3120 | static coroutine_fn int | |
3121 | sd_co_block_status(BlockDriverState *bs, bool want_zero, int64_t offset, | |
3122 | int64_t bytes, int64_t *pnum, int64_t *map, | |
3123 | BlockDriverState **file) | |
3124 | { | |
3125 | BDRVSheepdogState *s = bs->opaque; | |
3126 | SheepdogInode *inode = &s->inode; | |
3127 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); | |
3128 | unsigned long start = offset / object_size, | |
3129 | end = DIV_ROUND_UP(offset + bytes, object_size); | |
3130 | unsigned long idx; | |
3131 | *map = offset; | |
3132 | int ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
3133 | ||
3134 | for (idx = start; idx < end; idx++) { | |
3135 | if (inode->data_vdi_id[idx] == 0) { | |
3136 | break; | |
3137 | } | |
3138 | } | |
3139 | if (idx == start) { | |
3140 | /* Get the longest length of unallocated sectors */ | |
3141 | ret = 0; | |
3142 | for (idx = start + 1; idx < end; idx++) { | |
3143 | if (inode->data_vdi_id[idx] != 0) { | |
3144 | break; | |
3145 | } | |
3146 | } | |
3147 | } | |
3148 | ||
3149 | *pnum = (idx - start) * object_size; | |
3150 | if (*pnum > bytes) { | |
3151 | *pnum = bytes; | |
3152 | } | |
3153 | if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) { | |
3154 | *file = bs; | |
3155 | } | |
3156 | return ret; | |
3157 | } | |
3158 | ||
3159 | static int64_t sd_get_allocated_file_size(BlockDriverState *bs) | |
3160 | { | |
3161 | BDRVSheepdogState *s = bs->opaque; | |
3162 | SheepdogInode *inode = &s->inode; | |
3163 | uint32_t object_size = (UINT32_C(1) << inode->block_size_shift); | |
3164 | unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size); | |
3165 | uint64_t size = 0; | |
3166 | ||
3167 | for (i = 0; i < last; i++) { | |
3168 | if (inode->data_vdi_id[i] == 0) { | |
3169 | continue; | |
3170 | } | |
3171 | size += object_size; | |
3172 | } | |
3173 | return size; | |
3174 | } | |
3175 | ||
3176 | static QemuOptsList sd_create_opts = { | |
3177 | .name = "sheepdog-create-opts", | |
3178 | .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head), | |
3179 | .desc = { | |
3180 | { | |
3181 | .name = BLOCK_OPT_SIZE, | |
3182 | .type = QEMU_OPT_SIZE, | |
3183 | .help = "Virtual disk size" | |
3184 | }, | |
3185 | { | |
3186 | .name = BLOCK_OPT_BACKING_FILE, | |
3187 | .type = QEMU_OPT_STRING, | |
3188 | .help = "File name of a base image" | |
3189 | }, | |
3190 | { | |
3191 | .name = BLOCK_OPT_PREALLOC, | |
3192 | .type = QEMU_OPT_STRING, | |
3193 | .help = "Preallocation mode (allowed values: off, full)" | |
3194 | }, | |
3195 | { | |
3196 | .name = BLOCK_OPT_REDUNDANCY, | |
3197 | .type = QEMU_OPT_STRING, | |
3198 | .help = "Redundancy of the image" | |
3199 | }, | |
3200 | { | |
3201 | .name = BLOCK_OPT_OBJECT_SIZE, | |
3202 | .type = QEMU_OPT_SIZE, | |
3203 | .help = "Object size of the image" | |
3204 | }, | |
3205 | { /* end of list */ } | |
3206 | } | |
3207 | }; | |
3208 | ||
3209 | static const char *const sd_strong_runtime_opts[] = { | |
3210 | "vdi", | |
3211 | "snap-id", | |
3212 | "tag", | |
3213 | "server.", | |
3214 | ||
3215 | NULL | |
3216 | }; | |
3217 | ||
3218 | static BlockDriver bdrv_sheepdog = { | |
3219 | .format_name = "sheepdog", | |
3220 | .protocol_name = "sheepdog", | |
3221 | .instance_size = sizeof(BDRVSheepdogState), | |
3222 | .bdrv_parse_filename = sd_parse_filename, | |
3223 | .bdrv_file_open = sd_open, | |
3224 | .bdrv_reopen_prepare = sd_reopen_prepare, | |
3225 | .bdrv_reopen_commit = sd_reopen_commit, | |
3226 | .bdrv_reopen_abort = sd_reopen_abort, | |
3227 | .bdrv_close = sd_close, | |
3228 | .bdrv_co_create = sd_co_create, | |
3229 | .bdrv_co_create_opts = sd_co_create_opts, | |
3230 | .bdrv_has_zero_init = bdrv_has_zero_init_1, | |
3231 | .bdrv_getlength = sd_getlength, | |
3232 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, | |
3233 | .bdrv_co_truncate = sd_co_truncate, | |
3234 | ||
3235 | .bdrv_co_readv = sd_co_readv, | |
3236 | .bdrv_co_writev = sd_co_writev, | |
3237 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
3238 | .bdrv_co_pdiscard = sd_co_pdiscard, | |
3239 | .bdrv_co_block_status = sd_co_block_status, | |
3240 | ||
3241 | .bdrv_snapshot_create = sd_snapshot_create, | |
3242 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
3243 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
3244 | .bdrv_snapshot_list = sd_snapshot_list, | |
3245 | ||
3246 | .bdrv_save_vmstate = sd_save_vmstate, | |
3247 | .bdrv_load_vmstate = sd_load_vmstate, | |
3248 | ||
3249 | .bdrv_detach_aio_context = sd_detach_aio_context, | |
3250 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
3251 | ||
3252 | .create_opts = &sd_create_opts, | |
3253 | .strong_runtime_opts = sd_strong_runtime_opts, | |
3254 | }; | |
3255 | ||
3256 | static BlockDriver bdrv_sheepdog_tcp = { | |
3257 | .format_name = "sheepdog", | |
3258 | .protocol_name = "sheepdog+tcp", | |
3259 | .instance_size = sizeof(BDRVSheepdogState), | |
3260 | .bdrv_parse_filename = sd_parse_filename, | |
3261 | .bdrv_file_open = sd_open, | |
3262 | .bdrv_reopen_prepare = sd_reopen_prepare, | |
3263 | .bdrv_reopen_commit = sd_reopen_commit, | |
3264 | .bdrv_reopen_abort = sd_reopen_abort, | |
3265 | .bdrv_close = sd_close, | |
3266 | .bdrv_co_create = sd_co_create, | |
3267 | .bdrv_co_create_opts = sd_co_create_opts, | |
3268 | .bdrv_has_zero_init = bdrv_has_zero_init_1, | |
3269 | .bdrv_getlength = sd_getlength, | |
3270 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, | |
3271 | .bdrv_co_truncate = sd_co_truncate, | |
3272 | ||
3273 | .bdrv_co_readv = sd_co_readv, | |
3274 | .bdrv_co_writev = sd_co_writev, | |
3275 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
3276 | .bdrv_co_pdiscard = sd_co_pdiscard, | |
3277 | .bdrv_co_block_status = sd_co_block_status, | |
3278 | ||
3279 | .bdrv_snapshot_create = sd_snapshot_create, | |
3280 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
3281 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
3282 | .bdrv_snapshot_list = sd_snapshot_list, | |
3283 | ||
3284 | .bdrv_save_vmstate = sd_save_vmstate, | |
3285 | .bdrv_load_vmstate = sd_load_vmstate, | |
3286 | ||
3287 | .bdrv_detach_aio_context = sd_detach_aio_context, | |
3288 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
3289 | ||
3290 | .create_opts = &sd_create_opts, | |
3291 | .strong_runtime_opts = sd_strong_runtime_opts, | |
3292 | }; | |
3293 | ||
3294 | static BlockDriver bdrv_sheepdog_unix = { | |
3295 | .format_name = "sheepdog", | |
3296 | .protocol_name = "sheepdog+unix", | |
3297 | .instance_size = sizeof(BDRVSheepdogState), | |
3298 | .bdrv_parse_filename = sd_parse_filename, | |
3299 | .bdrv_file_open = sd_open, | |
3300 | .bdrv_reopen_prepare = sd_reopen_prepare, | |
3301 | .bdrv_reopen_commit = sd_reopen_commit, | |
3302 | .bdrv_reopen_abort = sd_reopen_abort, | |
3303 | .bdrv_close = sd_close, | |
3304 | .bdrv_co_create = sd_co_create, | |
3305 | .bdrv_co_create_opts = sd_co_create_opts, | |
3306 | .bdrv_has_zero_init = bdrv_has_zero_init_1, | |
3307 | .bdrv_getlength = sd_getlength, | |
3308 | .bdrv_get_allocated_file_size = sd_get_allocated_file_size, | |
3309 | .bdrv_co_truncate = sd_co_truncate, | |
3310 | ||
3311 | .bdrv_co_readv = sd_co_readv, | |
3312 | .bdrv_co_writev = sd_co_writev, | |
3313 | .bdrv_co_flush_to_disk = sd_co_flush_to_disk, | |
3314 | .bdrv_co_pdiscard = sd_co_pdiscard, | |
3315 | .bdrv_co_block_status = sd_co_block_status, | |
3316 | ||
3317 | .bdrv_snapshot_create = sd_snapshot_create, | |
3318 | .bdrv_snapshot_goto = sd_snapshot_goto, | |
3319 | .bdrv_snapshot_delete = sd_snapshot_delete, | |
3320 | .bdrv_snapshot_list = sd_snapshot_list, | |
3321 | ||
3322 | .bdrv_save_vmstate = sd_save_vmstate, | |
3323 | .bdrv_load_vmstate = sd_load_vmstate, | |
3324 | ||
3325 | .bdrv_detach_aio_context = sd_detach_aio_context, | |
3326 | .bdrv_attach_aio_context = sd_attach_aio_context, | |
3327 | ||
3328 | .create_opts = &sd_create_opts, | |
3329 | .strong_runtime_opts = sd_strong_runtime_opts, | |
3330 | }; | |
3331 | ||
3332 | static void bdrv_sheepdog_init(void) | |
3333 | { | |
3334 | bdrv_register(&bdrv_sheepdog); | |
3335 | bdrv_register(&bdrv_sheepdog_tcp); | |
3336 | bdrv_register(&bdrv_sheepdog_unix); | |
3337 | } | |
3338 | block_init(bdrv_sheepdog_init); |