]>
Commit | Line | Data |
---|---|---|
f27aaf4b CB |
1 | /* |
2 | * QEMU Block driver for RADOS (Ceph) | |
3 | * | |
ad32e9c0 JD |
4 | * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>, |
5 | * Josh Durgin <josh.durgin@dreamhost.com> | |
f27aaf4b CB |
6 | * |
7 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
8 | * the COPYING file in the top-level directory. | |
9 | * | |
6b620ca3 PB |
10 | * Contributions after 2012-01-13 are licensed under the terms of the |
11 | * GNU GPL, version 2 or (at your option) any later version. | |
f27aaf4b CB |
12 | */ |
13 | ||
ad32e9c0 JD |
14 | #include <inttypes.h> |
15 | ||
f27aaf4b CB |
16 | #include "qemu-common.h" |
17 | #include "qemu-error.h" | |
f27aaf4b CB |
18 | #include "block_int.h" |
19 | ||
ad32e9c0 | 20 | #include <rbd/librbd.h> |
f27aaf4b | 21 | |
f27aaf4b CB |
22 | /* |
23 | * When specifying the image filename use: | |
24 | * | |
fab5cf59 | 25 | * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]] |
f27aaf4b | 26 | * |
9e1fbcde | 27 | * poolname must be the name of an existing rados pool. |
f27aaf4b | 28 | * |
9e1fbcde | 29 | * devicename is the name of the rbd image. |
f27aaf4b | 30 | * |
9e1fbcde SW |
31 | * Each option given is used to configure rados, and may be any valid |
32 | * Ceph option, "id", or "conf". | |
fab5cf59 | 33 | * |
9e1fbcde SW |
34 | * The "id" option indicates what user we should authenticate as to |
35 | * the Ceph cluster. If it is excluded we will use the Ceph default | |
36 | * (normally 'admin'). | |
f27aaf4b | 37 | * |
9e1fbcde SW |
38 | * The "conf" option specifies a Ceph configuration file to read. If |
39 | * it is not specified, we will read from the default Ceph locations | |
40 | * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration | |
41 | * file, specify conf=/dev/null. | |
f27aaf4b | 42 | * |
9e1fbcde SW |
43 | * Configuration values containing :, @, or = can be escaped with a |
44 | * leading "\". | |
f27aaf4b CB |
45 | */ |
46 | ||
47 | #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER) | |
48 | ||
ad32e9c0 JD |
49 | #define RBD_MAX_CONF_NAME_SIZE 128 |
50 | #define RBD_MAX_CONF_VAL_SIZE 512 | |
51 | #define RBD_MAX_CONF_SIZE 1024 | |
52 | #define RBD_MAX_POOL_NAME_SIZE 128 | |
53 | #define RBD_MAX_SNAP_NAME_SIZE 128 | |
54 | #define RBD_MAX_SNAPS 100 | |
55 | ||
f27aaf4b CB |
56 | typedef struct RBDAIOCB { |
57 | BlockDriverAIOCB common; | |
58 | QEMUBH *bh; | |
59 | int ret; | |
60 | QEMUIOVector *qiov; | |
61 | char *bounce; | |
62 | int write; | |
63 | int64_t sector_num; | |
f27aaf4b CB |
64 | int error; |
65 | struct BDRVRBDState *s; | |
66 | int cancelled; | |
67 | } RBDAIOCB; | |
68 | ||
69 | typedef struct RADOSCB { | |
70 | int rcbid; | |
71 | RBDAIOCB *acb; | |
72 | struct BDRVRBDState *s; | |
73 | int done; | |
ad32e9c0 | 74 | int64_t size; |
f27aaf4b CB |
75 | char *buf; |
76 | int ret; | |
77 | } RADOSCB; | |
78 | ||
79 | #define RBD_FD_READ 0 | |
80 | #define RBD_FD_WRITE 1 | |
81 | ||
82 | typedef struct BDRVRBDState { | |
83 | int fds[2]; | |
ad32e9c0 JD |
84 | rados_t cluster; |
85 | rados_ioctx_t io_ctx; | |
86 | rbd_image_t image; | |
87 | char name[RBD_MAX_IMAGE_NAME_SIZE]; | |
f27aaf4b | 88 | int qemu_aio_count; |
ad32e9c0 | 89 | char *snap; |
f27aaf4b CB |
90 | int event_reader_pos; |
91 | RADOSCB *event_rcb; | |
92 | } BDRVRBDState; | |
93 | ||
f27aaf4b CB |
94 | static void rbd_aio_bh_cb(void *opaque); |
95 | ||
ad32e9c0 JD |
96 | static int qemu_rbd_next_tok(char *dst, int dst_len, |
97 | char *src, char delim, | |
98 | const char *name, | |
99 | char **p) | |
f27aaf4b CB |
100 | { |
101 | int l; | |
102 | char *end; | |
103 | ||
104 | *p = NULL; | |
105 | ||
106 | if (delim != '\0') { | |
16a06b24 SW |
107 | for (end = src; *end; ++end) { |
108 | if (*end == delim) { | |
109 | break; | |
110 | } | |
111 | if (*end == '\\' && end[1] != '\0') { | |
112 | end++; | |
113 | } | |
114 | } | |
115 | if (*end == delim) { | |
f27aaf4b CB |
116 | *p = end + 1; |
117 | *end = '\0'; | |
118 | } | |
119 | } | |
120 | l = strlen(src); | |
121 | if (l >= dst_len) { | |
122 | error_report("%s too long", name); | |
123 | return -EINVAL; | |
124 | } else if (l == 0) { | |
125 | error_report("%s too short", name); | |
126 | return -EINVAL; | |
127 | } | |
128 | ||
129 | pstrcpy(dst, dst_len, src); | |
130 | ||
131 | return 0; | |
132 | } | |
133 | ||
16a06b24 SW |
134 | static void qemu_rbd_unescape(char *src) |
135 | { | |
136 | char *p; | |
137 | ||
138 | for (p = src; *src; ++src, ++p) { | |
139 | if (*src == '\\' && src[1] != '\0') { | |
140 | src++; | |
141 | } | |
142 | *p = *src; | |
143 | } | |
144 | *p = '\0'; | |
145 | } | |
146 | ||
ad32e9c0 JD |
147 | static int qemu_rbd_parsename(const char *filename, |
148 | char *pool, int pool_len, | |
149 | char *snap, int snap_len, | |
fab5cf59 JD |
150 | char *name, int name_len, |
151 | char *conf, int conf_len) | |
f27aaf4b CB |
152 | { |
153 | const char *start; | |
154 | char *p, *buf; | |
155 | int ret; | |
156 | ||
157 | if (!strstart(filename, "rbd:", &start)) { | |
158 | return -EINVAL; | |
159 | } | |
160 | ||
7267c094 | 161 | buf = g_strdup(start); |
f27aaf4b | 162 | p = buf; |
fab5cf59 JD |
163 | *snap = '\0'; |
164 | *conf = '\0'; | |
f27aaf4b | 165 | |
ad32e9c0 | 166 | ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p); |
f27aaf4b CB |
167 | if (ret < 0 || !p) { |
168 | ret = -EINVAL; | |
169 | goto done; | |
170 | } | |
16a06b24 | 171 | qemu_rbd_unescape(pool); |
fab5cf59 JD |
172 | |
173 | if (strchr(p, '@')) { | |
174 | ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p); | |
175 | if (ret < 0) { | |
176 | goto done; | |
177 | } | |
178 | ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p); | |
16a06b24 | 179 | qemu_rbd_unescape(snap); |
fab5cf59 JD |
180 | } else { |
181 | ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p); | |
f27aaf4b | 182 | } |
16a06b24 | 183 | qemu_rbd_unescape(name); |
fab5cf59 | 184 | if (ret < 0 || !p) { |
f27aaf4b CB |
185 | goto done; |
186 | } | |
187 | ||
fab5cf59 | 188 | ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p); |
f27aaf4b CB |
189 | |
190 | done: | |
7267c094 | 191 | g_free(buf); |
f27aaf4b CB |
192 | return ret; |
193 | } | |
194 | ||
7c7e9df0 SW |
195 | static char *qemu_rbd_parse_clientname(const char *conf, char *clientname) |
196 | { | |
197 | const char *p = conf; | |
198 | ||
199 | while (*p) { | |
200 | int len; | |
201 | const char *end = strchr(p, ':'); | |
202 | ||
203 | if (end) { | |
204 | len = end - p; | |
205 | } else { | |
206 | len = strlen(p); | |
207 | } | |
208 | ||
209 | if (strncmp(p, "id=", 3) == 0) { | |
210 | len -= 3; | |
211 | strncpy(clientname, p + 3, len); | |
212 | clientname[len] = '\0'; | |
213 | return clientname; | |
214 | } | |
215 | if (end == NULL) { | |
216 | break; | |
217 | } | |
218 | p = end + 1; | |
219 | } | |
220 | return NULL; | |
221 | } | |
222 | ||
fab5cf59 JD |
223 | static int qemu_rbd_set_conf(rados_t cluster, const char *conf) |
224 | { | |
225 | char *p, *buf; | |
226 | char name[RBD_MAX_CONF_NAME_SIZE]; | |
227 | char value[RBD_MAX_CONF_VAL_SIZE]; | |
228 | int ret = 0; | |
229 | ||
7267c094 | 230 | buf = g_strdup(conf); |
fab5cf59 JD |
231 | p = buf; |
232 | ||
233 | while (p) { | |
234 | ret = qemu_rbd_next_tok(name, sizeof(name), p, | |
235 | '=', "conf option name", &p); | |
236 | if (ret < 0) { | |
237 | break; | |
238 | } | |
16a06b24 | 239 | qemu_rbd_unescape(name); |
fab5cf59 JD |
240 | |
241 | if (!p) { | |
242 | error_report("conf option %s has no value", name); | |
243 | ret = -EINVAL; | |
244 | break; | |
245 | } | |
246 | ||
247 | ret = qemu_rbd_next_tok(value, sizeof(value), p, | |
248 | ':', "conf option value", &p); | |
249 | if (ret < 0) { | |
250 | break; | |
251 | } | |
16a06b24 | 252 | qemu_rbd_unescape(value); |
fab5cf59 | 253 | |
7c7e9df0 SW |
254 | if (strcmp(name, "conf") == 0) { |
255 | ret = rados_conf_read_file(cluster, value); | |
fab5cf59 | 256 | if (ret < 0) { |
7c7e9df0 | 257 | error_report("error reading conf file %s", value); |
fab5cf59 JD |
258 | break; |
259 | } | |
7c7e9df0 SW |
260 | } else if (strcmp(name, "id") == 0) { |
261 | /* ignore, this is parsed by qemu_rbd_parse_clientname() */ | |
fab5cf59 | 262 | } else { |
7c7e9df0 | 263 | ret = rados_conf_set(cluster, name, value); |
fab5cf59 | 264 | if (ret < 0) { |
7c7e9df0 SW |
265 | error_report("invalid conf option %s", name); |
266 | ret = -EINVAL; | |
fab5cf59 JD |
267 | break; |
268 | } | |
269 | } | |
270 | } | |
271 | ||
7267c094 | 272 | g_free(buf); |
fab5cf59 JD |
273 | return ret; |
274 | } | |
275 | ||
ad32e9c0 | 276 | static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options) |
f27aaf4b CB |
277 | { |
278 | int64_t bytes = 0; | |
279 | int64_t objsize; | |
ad32e9c0 JD |
280 | int obj_order = 0; |
281 | char pool[RBD_MAX_POOL_NAME_SIZE]; | |
282 | char name[RBD_MAX_IMAGE_NAME_SIZE]; | |
283 | char snap_buf[RBD_MAX_SNAP_NAME_SIZE]; | |
fab5cf59 | 284 | char conf[RBD_MAX_CONF_SIZE]; |
7c7e9df0 SW |
285 | char clientname_buf[RBD_MAX_CONF_SIZE]; |
286 | char *clientname; | |
ad32e9c0 JD |
287 | rados_t cluster; |
288 | rados_ioctx_t io_ctx; | |
f27aaf4b CB |
289 | int ret; |
290 | ||
ad32e9c0 JD |
291 | if (qemu_rbd_parsename(filename, pool, sizeof(pool), |
292 | snap_buf, sizeof(snap_buf), | |
fab5cf59 JD |
293 | name, sizeof(name), |
294 | conf, sizeof(conf)) < 0) { | |
f27aaf4b CB |
295 | return -EINVAL; |
296 | } | |
f27aaf4b | 297 | |
f27aaf4b CB |
298 | /* Read out options */ |
299 | while (options && options->name) { | |
300 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
301 | bytes = options->value.n; | |
302 | } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { | |
303 | if (options->value.n) { | |
304 | objsize = options->value.n; | |
305 | if ((objsize - 1) & objsize) { /* not a power of 2? */ | |
306 | error_report("obj size needs to be power of 2"); | |
307 | return -EINVAL; | |
308 | } | |
309 | if (objsize < 4096) { | |
310 | error_report("obj size too small"); | |
311 | return -EINVAL; | |
312 | } | |
ad32e9c0 | 313 | obj_order = ffs(objsize) - 1; |
f27aaf4b CB |
314 | } |
315 | } | |
316 | options++; | |
317 | } | |
318 | ||
7c7e9df0 SW |
319 | clientname = qemu_rbd_parse_clientname(conf, clientname_buf); |
320 | if (rados_create(&cluster, clientname) < 0) { | |
f27aaf4b CB |
321 | error_report("error initializing"); |
322 | return -EIO; | |
323 | } | |
324 | ||
fab5cf59 | 325 | if (strstr(conf, "conf=") == NULL) { |
f9fe18ec SW |
326 | /* try default location, but ignore failure */ |
327 | rados_conf_read_file(cluster, NULL); | |
fab5cf59 JD |
328 | } |
329 | ||
330 | if (conf[0] != '\0' && | |
331 | qemu_rbd_set_conf(cluster, conf) < 0) { | |
332 | error_report("error setting config options"); | |
ad32e9c0 | 333 | rados_shutdown(cluster); |
f27aaf4b CB |
334 | return -EIO; |
335 | } | |
336 | ||
ad32e9c0 JD |
337 | if (rados_connect(cluster) < 0) { |
338 | error_report("error connecting"); | |
339 | rados_shutdown(cluster); | |
f27aaf4b CB |
340 | return -EIO; |
341 | } | |
f27aaf4b | 342 | |
ad32e9c0 JD |
343 | if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) { |
344 | error_report("error opening pool %s", pool); | |
345 | rados_shutdown(cluster); | |
346 | return -EIO; | |
f27aaf4b CB |
347 | } |
348 | ||
ad32e9c0 JD |
349 | ret = rbd_create(io_ctx, name, bytes, &obj_order); |
350 | rados_ioctx_destroy(io_ctx); | |
351 | rados_shutdown(cluster); | |
f27aaf4b CB |
352 | |
353 | return ret; | |
354 | } | |
355 | ||
356 | /* | |
ad32e9c0 JD |
357 | * This aio completion is being called from qemu_rbd_aio_event_reader() |
358 | * and runs in qemu context. It schedules a bh, but just in case the aio | |
f27aaf4b CB |
359 | * was not cancelled before. |
360 | */ | |
ad32e9c0 | 361 | static void qemu_rbd_complete_aio(RADOSCB *rcb) |
f27aaf4b CB |
362 | { |
363 | RBDAIOCB *acb = rcb->acb; | |
364 | int64_t r; | |
365 | ||
f27aaf4b | 366 | if (acb->cancelled) { |
ad32e9c0 JD |
367 | qemu_vfree(acb->bounce); |
368 | qemu_aio_release(acb); | |
f27aaf4b CB |
369 | goto done; |
370 | } | |
371 | ||
372 | r = rcb->ret; | |
373 | ||
374 | if (acb->write) { | |
375 | if (r < 0) { | |
376 | acb->ret = r; | |
377 | acb->error = 1; | |
378 | } else if (!acb->error) { | |
ad32e9c0 | 379 | acb->ret = rcb->size; |
f27aaf4b CB |
380 | } |
381 | } else { | |
ad32e9c0 JD |
382 | if (r < 0) { |
383 | memset(rcb->buf, 0, rcb->size); | |
f27aaf4b CB |
384 | acb->ret = r; |
385 | acb->error = 1; | |
ad32e9c0 JD |
386 | } else if (r < rcb->size) { |
387 | memset(rcb->buf + r, 0, rcb->size - r); | |
f27aaf4b | 388 | if (!acb->error) { |
ad32e9c0 | 389 | acb->ret = rcb->size; |
f27aaf4b CB |
390 | } |
391 | } else if (!acb->error) { | |
ad32e9c0 | 392 | acb->ret = r; |
f27aaf4b CB |
393 | } |
394 | } | |
395 | /* Note that acb->bh can be NULL in case where the aio was cancelled */ | |
ad32e9c0 JD |
396 | acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb); |
397 | qemu_bh_schedule(acb->bh); | |
f27aaf4b | 398 | done: |
7267c094 | 399 | g_free(rcb); |
f27aaf4b CB |
400 | } |
401 | ||
402 | /* | |
403 | * aio fd read handler. It runs in the qemu context and calls the | |
404 | * completion handling of completed rados aio operations. | |
405 | */ | |
ad32e9c0 | 406 | static void qemu_rbd_aio_event_reader(void *opaque) |
f27aaf4b CB |
407 | { |
408 | BDRVRBDState *s = opaque; | |
409 | ||
410 | ssize_t ret; | |
411 | ||
412 | do { | |
413 | char *p = (char *)&s->event_rcb; | |
414 | ||
415 | /* now read the rcb pointer that was sent from a non qemu thread */ | |
dfe80b07 SW |
416 | ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos, |
417 | sizeof(s->event_rcb) - s->event_reader_pos); | |
418 | if (ret > 0) { | |
419 | s->event_reader_pos += ret; | |
420 | if (s->event_reader_pos == sizeof(s->event_rcb)) { | |
421 | s->event_reader_pos = 0; | |
422 | qemu_rbd_complete_aio(s->event_rcb); | |
423 | s->qemu_aio_count--; | |
f27aaf4b CB |
424 | } |
425 | } | |
426 | } while (ret < 0 && errno == EINTR); | |
427 | } | |
428 | ||
ad32e9c0 | 429 | static int qemu_rbd_aio_flush_cb(void *opaque) |
f27aaf4b CB |
430 | { |
431 | BDRVRBDState *s = opaque; | |
432 | ||
433 | return (s->qemu_aio_count > 0); | |
434 | } | |
435 | ||
ad32e9c0 | 436 | static int qemu_rbd_open(BlockDriverState *bs, const char *filename, int flags) |
f27aaf4b CB |
437 | { |
438 | BDRVRBDState *s = bs->opaque; | |
ad32e9c0 JD |
439 | char pool[RBD_MAX_POOL_NAME_SIZE]; |
440 | char snap_buf[RBD_MAX_SNAP_NAME_SIZE]; | |
fab5cf59 | 441 | char conf[RBD_MAX_CONF_SIZE]; |
7c7e9df0 SW |
442 | char clientname_buf[RBD_MAX_CONF_SIZE]; |
443 | char *clientname; | |
f27aaf4b CB |
444 | int r; |
445 | ||
ad32e9c0 JD |
446 | if (qemu_rbd_parsename(filename, pool, sizeof(pool), |
447 | snap_buf, sizeof(snap_buf), | |
fab5cf59 JD |
448 | s->name, sizeof(s->name), |
449 | conf, sizeof(conf)) < 0) { | |
f27aaf4b CB |
450 | return -EINVAL; |
451 | } | |
f27aaf4b | 452 | |
7c7e9df0 SW |
453 | clientname = qemu_rbd_parse_clientname(conf, clientname_buf); |
454 | r = rados_create(&s->cluster, clientname); | |
ad32e9c0 | 455 | if (r < 0) { |
f27aaf4b CB |
456 | error_report("error initializing"); |
457 | return r; | |
458 | } | |
459 | ||
eb93d5d9 SW |
460 | s->snap = NULL; |
461 | if (snap_buf[0] != '\0') { | |
462 | s->snap = g_strdup(snap_buf); | |
463 | } | |
464 | ||
fab5cf59 | 465 | if (strstr(conf, "conf=") == NULL) { |
f9fe18ec SW |
466 | /* try default location, but ignore failure */ |
467 | rados_conf_read_file(s->cluster, NULL); | |
fab5cf59 JD |
468 | } |
469 | ||
470 | if (conf[0] != '\0') { | |
471 | r = qemu_rbd_set_conf(s->cluster, conf); | |
472 | if (r < 0) { | |
473 | error_report("error setting config options"); | |
eb93d5d9 | 474 | goto failed_shutdown; |
fab5cf59 | 475 | } |
f27aaf4b CB |
476 | } |
477 | ||
ad32e9c0 JD |
478 | r = rados_connect(s->cluster); |
479 | if (r < 0) { | |
480 | error_report("error connecting"); | |
eb93d5d9 | 481 | goto failed_shutdown; |
f27aaf4b CB |
482 | } |
483 | ||
ad32e9c0 JD |
484 | r = rados_ioctx_create(s->cluster, pool, &s->io_ctx); |
485 | if (r < 0) { | |
486 | error_report("error opening pool %s", pool); | |
eb93d5d9 | 487 | goto failed_shutdown; |
f27aaf4b CB |
488 | } |
489 | ||
ad32e9c0 | 490 | r = rbd_open(s->io_ctx, s->name, &s->image, s->snap); |
f27aaf4b | 491 | if (r < 0) { |
ad32e9c0 | 492 | error_report("error reading header from %s", s->name); |
eb93d5d9 | 493 | goto failed_open; |
f27aaf4b CB |
494 | } |
495 | ||
ad32e9c0 | 496 | bs->read_only = (s->snap != NULL); |
f27aaf4b CB |
497 | |
498 | s->event_reader_pos = 0; | |
499 | r = qemu_pipe(s->fds); | |
500 | if (r < 0) { | |
501 | error_report("error opening eventfd"); | |
502 | goto failed; | |
503 | } | |
504 | fcntl(s->fds[0], F_SETFL, O_NONBLOCK); | |
505 | fcntl(s->fds[1], F_SETFL, O_NONBLOCK); | |
ad32e9c0 JD |
506 | qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader, |
507 | NULL, qemu_rbd_aio_flush_cb, NULL, s); | |
f27aaf4b | 508 | |
f27aaf4b CB |
509 | |
510 | return 0; | |
511 | ||
512 | failed: | |
ad32e9c0 | 513 | rbd_close(s->image); |
eb93d5d9 | 514 | failed_open: |
ad32e9c0 | 515 | rados_ioctx_destroy(s->io_ctx); |
eb93d5d9 | 516 | failed_shutdown: |
ad32e9c0 | 517 | rados_shutdown(s->cluster); |
eb93d5d9 | 518 | g_free(s->snap); |
f27aaf4b CB |
519 | return r; |
520 | } | |
521 | ||
ad32e9c0 | 522 | static void qemu_rbd_close(BlockDriverState *bs) |
f27aaf4b CB |
523 | { |
524 | BDRVRBDState *s = bs->opaque; | |
525 | ||
526 | close(s->fds[0]); | |
527 | close(s->fds[1]); | |
528 | qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL , NULL, NULL, NULL, | |
529 | NULL); | |
530 | ||
ad32e9c0 JD |
531 | rbd_close(s->image); |
532 | rados_ioctx_destroy(s->io_ctx); | |
7267c094 | 533 | g_free(s->snap); |
ad32e9c0 | 534 | rados_shutdown(s->cluster); |
f27aaf4b CB |
535 | } |
536 | ||
537 | /* | |
538 | * Cancel aio. Since we don't reference acb in a non qemu threads, | |
539 | * it is safe to access it here. | |
540 | */ | |
ad32e9c0 | 541 | static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb) |
f27aaf4b CB |
542 | { |
543 | RBDAIOCB *acb = (RBDAIOCB *) blockacb; | |
544 | acb->cancelled = 1; | |
545 | } | |
546 | ||
547 | static AIOPool rbd_aio_pool = { | |
548 | .aiocb_size = sizeof(RBDAIOCB), | |
ad32e9c0 | 549 | .cancel = qemu_rbd_aio_cancel, |
f27aaf4b CB |
550 | }; |
551 | ||
ad32e9c0 | 552 | static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb) |
f27aaf4b | 553 | { |
ad32e9c0 | 554 | int ret = 0; |
f27aaf4b CB |
555 | while (1) { |
556 | fd_set wfd; | |
ad32e9c0 | 557 | int fd = s->fds[RBD_FD_WRITE]; |
f27aaf4b | 558 | |
ad32e9c0 JD |
559 | /* send the op pointer to the qemu thread that is responsible |
560 | for the aio/op completion. Must do it in a qemu thread context */ | |
f27aaf4b CB |
561 | ret = write(fd, (void *)&rcb, sizeof(rcb)); |
562 | if (ret >= 0) { | |
563 | break; | |
564 | } | |
565 | if (errno == EINTR) { | |
566 | continue; | |
ad32e9c0 | 567 | } |
f27aaf4b CB |
568 | if (errno != EAGAIN) { |
569 | break; | |
ad32e9c0 | 570 | } |
f27aaf4b CB |
571 | |
572 | FD_ZERO(&wfd); | |
573 | FD_SET(fd, &wfd); | |
574 | do { | |
575 | ret = select(fd + 1, NULL, &wfd, NULL, NULL); | |
576 | } while (ret < 0 && errno == EINTR); | |
577 | } | |
578 | ||
ad32e9c0 JD |
579 | return ret; |
580 | } | |
581 | ||
582 | /* | |
583 | * This is the callback function for rbd_aio_read and _write | |
584 | * | |
585 | * Note: this function is being called from a non qemu thread so | |
586 | * we need to be careful about what we do here. Generally we only | |
587 | * write to the block notification pipe, and do the rest of the | |
588 | * io completion handling from qemu_rbd_aio_event_reader() which | |
589 | * runs in a qemu context. | |
590 | */ | |
591 | static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb) | |
592 | { | |
593 | int ret; | |
594 | rcb->ret = rbd_aio_get_return_value(c); | |
595 | rbd_aio_release(c); | |
596 | ret = qemu_rbd_send_pipe(rcb->s, rcb); | |
f27aaf4b | 597 | if (ret < 0) { |
ad32e9c0 | 598 | error_report("failed writing to acb->s->fds"); |
7267c094 | 599 | g_free(rcb); |
f27aaf4b CB |
600 | } |
601 | } | |
602 | ||
ad32e9c0 | 603 | /* Callback when all queued rbd_aio requests are complete */ |
f27aaf4b CB |
604 | |
605 | static void rbd_aio_bh_cb(void *opaque) | |
606 | { | |
607 | RBDAIOCB *acb = opaque; | |
608 | ||
609 | if (!acb->write) { | |
610 | qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size); | |
611 | } | |
612 | qemu_vfree(acb->bounce); | |
613 | acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret)); | |
614 | qemu_bh_delete(acb->bh); | |
615 | acb->bh = NULL; | |
616 | ||
617 | qemu_aio_release(acb); | |
618 | } | |
619 | ||
620 | static BlockDriverAIOCB *rbd_aio_rw_vector(BlockDriverState *bs, | |
621 | int64_t sector_num, | |
622 | QEMUIOVector *qiov, | |
623 | int nb_sectors, | |
624 | BlockDriverCompletionFunc *cb, | |
625 | void *opaque, int write) | |
626 | { | |
627 | RBDAIOCB *acb; | |
628 | RADOSCB *rcb; | |
ad32e9c0 | 629 | rbd_completion_t c; |
f27aaf4b CB |
630 | int64_t off, size; |
631 | char *buf; | |
51a13528 | 632 | int r; |
f27aaf4b CB |
633 | |
634 | BDRVRBDState *s = bs->opaque; | |
635 | ||
636 | acb = qemu_aio_get(&rbd_aio_pool, bs, cb, opaque); | |
637 | acb->write = write; | |
638 | acb->qiov = qiov; | |
639 | acb->bounce = qemu_blockalign(bs, qiov->size); | |
f27aaf4b CB |
640 | acb->ret = 0; |
641 | acb->error = 0; | |
642 | acb->s = s; | |
643 | acb->cancelled = 0; | |
644 | acb->bh = NULL; | |
645 | ||
646 | if (write) { | |
647 | qemu_iovec_to_buffer(acb->qiov, acb->bounce); | |
648 | } | |
649 | ||
650 | buf = acb->bounce; | |
651 | ||
652 | off = sector_num * BDRV_SECTOR_SIZE; | |
653 | size = nb_sectors * BDRV_SECTOR_SIZE; | |
f27aaf4b | 654 | |
ad32e9c0 | 655 | s->qemu_aio_count++; /* All the RADOSCB */ |
f27aaf4b | 656 | |
7267c094 | 657 | rcb = g_malloc(sizeof(RADOSCB)); |
ad32e9c0 JD |
658 | rcb->done = 0; |
659 | rcb->acb = acb; | |
660 | rcb->buf = buf; | |
661 | rcb->s = acb->s; | |
662 | rcb->size = size; | |
51a13528 JD |
663 | r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c); |
664 | if (r < 0) { | |
665 | goto failed; | |
666 | } | |
f27aaf4b | 667 | |
ad32e9c0 | 668 | if (write) { |
51a13528 | 669 | r = rbd_aio_write(s->image, off, size, buf, c); |
ad32e9c0 | 670 | } else { |
51a13528 JD |
671 | r = rbd_aio_read(s->image, off, size, buf, c); |
672 | } | |
673 | ||
674 | if (r < 0) { | |
675 | goto failed; | |
f27aaf4b CB |
676 | } |
677 | ||
678 | return &acb->common; | |
51a13528 JD |
679 | |
680 | failed: | |
7267c094 | 681 | g_free(rcb); |
51a13528 JD |
682 | s->qemu_aio_count--; |
683 | qemu_aio_release(acb); | |
684 | return NULL; | |
f27aaf4b CB |
685 | } |
686 | ||
ad32e9c0 JD |
687 | static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs, |
688 | int64_t sector_num, | |
689 | QEMUIOVector *qiov, | |
690 | int nb_sectors, | |
691 | BlockDriverCompletionFunc *cb, | |
692 | void *opaque) | |
f27aaf4b CB |
693 | { |
694 | return rbd_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); | |
695 | } | |
696 | ||
ad32e9c0 JD |
697 | static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs, |
698 | int64_t sector_num, | |
699 | QEMUIOVector *qiov, | |
700 | int nb_sectors, | |
701 | BlockDriverCompletionFunc *cb, | |
702 | void *opaque) | |
f27aaf4b CB |
703 | { |
704 | return rbd_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); | |
705 | } | |
706 | ||
8b94ff85 | 707 | static int qemu_rbd_co_flush(BlockDriverState *bs) |
7a3f5fe9 SW |
708 | { |
709 | #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1) | |
710 | /* rbd_flush added in 0.1.1 */ | |
711 | BDRVRBDState *s = bs->opaque; | |
712 | return rbd_flush(s->image); | |
713 | #else | |
714 | return 0; | |
715 | #endif | |
716 | } | |
717 | ||
ad32e9c0 | 718 | static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi) |
f27aaf4b CB |
719 | { |
720 | BDRVRBDState *s = bs->opaque; | |
ad32e9c0 JD |
721 | rbd_image_info_t info; |
722 | int r; | |
723 | ||
724 | r = rbd_stat(s->image, &info, sizeof(info)); | |
725 | if (r < 0) { | |
726 | return r; | |
727 | } | |
728 | ||
729 | bdi->cluster_size = info.obj_size; | |
f27aaf4b CB |
730 | return 0; |
731 | } | |
732 | ||
ad32e9c0 | 733 | static int64_t qemu_rbd_getlength(BlockDriverState *bs) |
f27aaf4b CB |
734 | { |
735 | BDRVRBDState *s = bs->opaque; | |
ad32e9c0 JD |
736 | rbd_image_info_t info; |
737 | int r; | |
f27aaf4b | 738 | |
ad32e9c0 JD |
739 | r = rbd_stat(s->image, &info, sizeof(info)); |
740 | if (r < 0) { | |
741 | return r; | |
742 | } | |
743 | ||
744 | return info.size; | |
f27aaf4b CB |
745 | } |
746 | ||
30cdc48c JD |
747 | static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset) |
748 | { | |
749 | BDRVRBDState *s = bs->opaque; | |
750 | int r; | |
751 | ||
752 | r = rbd_resize(s->image, offset); | |
753 | if (r < 0) { | |
754 | return r; | |
755 | } | |
756 | ||
757 | return 0; | |
758 | } | |
759 | ||
ad32e9c0 JD |
760 | static int qemu_rbd_snap_create(BlockDriverState *bs, |
761 | QEMUSnapshotInfo *sn_info) | |
f27aaf4b CB |
762 | { |
763 | BDRVRBDState *s = bs->opaque; | |
f27aaf4b | 764 | int r; |
f27aaf4b CB |
765 | |
766 | if (sn_info->name[0] == '\0') { | |
767 | return -EINVAL; /* we need a name for rbd snapshots */ | |
768 | } | |
769 | ||
770 | /* | |
771 | * rbd snapshots are using the name as the user controlled unique identifier | |
772 | * we can't use the rbd snapid for that purpose, as it can't be set | |
773 | */ | |
774 | if (sn_info->id_str[0] != '\0' && | |
775 | strcmp(sn_info->id_str, sn_info->name) != 0) { | |
776 | return -EINVAL; | |
777 | } | |
778 | ||
779 | if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) { | |
780 | return -ERANGE; | |
781 | } | |
782 | ||
ad32e9c0 | 783 | r = rbd_snap_create(s->image, sn_info->name); |
f27aaf4b | 784 | if (r < 0) { |
ad32e9c0 | 785 | error_report("failed to create snap: %s", strerror(-r)); |
f27aaf4b CB |
786 | return r; |
787 | } | |
788 | ||
f27aaf4b CB |
789 | return 0; |
790 | } | |
791 | ||
bd603247 GF |
792 | static int qemu_rbd_snap_remove(BlockDriverState *bs, |
793 | const char *snapshot_name) | |
794 | { | |
795 | BDRVRBDState *s = bs->opaque; | |
796 | int r; | |
797 | ||
798 | r = rbd_snap_remove(s->image, snapshot_name); | |
799 | return r; | |
800 | } | |
801 | ||
802 | static int qemu_rbd_snap_rollback(BlockDriverState *bs, | |
803 | const char *snapshot_name) | |
804 | { | |
805 | BDRVRBDState *s = bs->opaque; | |
806 | int r; | |
807 | ||
808 | r = rbd_snap_rollback(s->image, snapshot_name); | |
809 | return r; | |
810 | } | |
811 | ||
ad32e9c0 JD |
812 | static int qemu_rbd_snap_list(BlockDriverState *bs, |
813 | QEMUSnapshotInfo **psn_tab) | |
f27aaf4b CB |
814 | { |
815 | BDRVRBDState *s = bs->opaque; | |
f27aaf4b | 816 | QEMUSnapshotInfo *sn_info, *sn_tab = NULL; |
ad32e9c0 JD |
817 | int i, snap_count; |
818 | rbd_snap_info_t *snaps; | |
819 | int max_snaps = RBD_MAX_SNAPS; | |
f27aaf4b | 820 | |
ad32e9c0 | 821 | do { |
7267c094 | 822 | snaps = g_malloc(sizeof(*snaps) * max_snaps); |
ad32e9c0 JD |
823 | snap_count = rbd_snap_list(s->image, snaps, &max_snaps); |
824 | if (snap_count < 0) { | |
7267c094 | 825 | g_free(snaps); |
f27aaf4b | 826 | } |
ad32e9c0 | 827 | } while (snap_count == -ERANGE); |
f27aaf4b | 828 | |
ad32e9c0 | 829 | if (snap_count <= 0) { |
b9c53290 | 830 | goto done; |
f27aaf4b CB |
831 | } |
832 | ||
7267c094 | 833 | sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo)); |
f27aaf4b | 834 | |
ad32e9c0 JD |
835 | for (i = 0; i < snap_count; i++) { |
836 | const char *snap_name = snaps[i].name; | |
f27aaf4b CB |
837 | |
838 | sn_info = sn_tab + i; | |
839 | pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name); | |
840 | pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name); | |
f27aaf4b | 841 | |
ad32e9c0 | 842 | sn_info->vm_state_size = snaps[i].size; |
f27aaf4b CB |
843 | sn_info->date_sec = 0; |
844 | sn_info->date_nsec = 0; | |
845 | sn_info->vm_clock_nsec = 0; | |
846 | } | |
ad32e9c0 JD |
847 | rbd_snap_list_end(snaps); |
848 | ||
b9c53290 | 849 | done: |
f27aaf4b | 850 | *psn_tab = sn_tab; |
f27aaf4b | 851 | return snap_count; |
f27aaf4b CB |
852 | } |
853 | ||
ad32e9c0 | 854 | static QEMUOptionParameter qemu_rbd_create_options[] = { |
f27aaf4b CB |
855 | { |
856 | .name = BLOCK_OPT_SIZE, | |
857 | .type = OPT_SIZE, | |
858 | .help = "Virtual disk size" | |
859 | }, | |
860 | { | |
861 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
862 | .type = OPT_SIZE, | |
863 | .help = "RBD object size" | |
864 | }, | |
865 | {NULL} | |
866 | }; | |
867 | ||
868 | static BlockDriver bdrv_rbd = { | |
869 | .format_name = "rbd", | |
870 | .instance_size = sizeof(BDRVRBDState), | |
ad32e9c0 JD |
871 | .bdrv_file_open = qemu_rbd_open, |
872 | .bdrv_close = qemu_rbd_close, | |
873 | .bdrv_create = qemu_rbd_create, | |
874 | .bdrv_get_info = qemu_rbd_getinfo, | |
875 | .create_options = qemu_rbd_create_options, | |
876 | .bdrv_getlength = qemu_rbd_getlength, | |
30cdc48c | 877 | .bdrv_truncate = qemu_rbd_truncate, |
f27aaf4b CB |
878 | .protocol_name = "rbd", |
879 | ||
c68b89ac KW |
880 | .bdrv_aio_readv = qemu_rbd_aio_readv, |
881 | .bdrv_aio_writev = qemu_rbd_aio_writev, | |
882 | .bdrv_co_flush_to_disk = qemu_rbd_co_flush, | |
f27aaf4b | 883 | |
c68b89ac | 884 | .bdrv_snapshot_create = qemu_rbd_snap_create, |
bd603247 | 885 | .bdrv_snapshot_delete = qemu_rbd_snap_remove, |
c68b89ac | 886 | .bdrv_snapshot_list = qemu_rbd_snap_list, |
bd603247 | 887 | .bdrv_snapshot_goto = qemu_rbd_snap_rollback, |
f27aaf4b CB |
888 | }; |
889 | ||
890 | static void bdrv_rbd_init(void) | |
891 | { | |
892 | bdrv_register(&bdrv_rbd); | |
893 | } | |
894 | ||
895 | block_init(bdrv_rbd_init); |