Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2017 Intel Corporation.
3 : * All rights reserved.
4 : */
5 :
6 : #include "spdk/stdinc.h"
7 :
8 : #include "bdev_rbd.h"
9 :
10 : #include <rbd/librbd.h>
11 : #include <rados/librados.h>
12 :
13 : #include "spdk/env.h"
14 : #include "spdk/bdev.h"
15 : #include "spdk/thread.h"
16 : #include "spdk/json.h"
17 : #include "spdk/string.h"
18 : #include "spdk/util.h"
19 : #include "spdk/likely.h"
20 :
21 : #include "spdk/bdev_module.h"
22 : #include "spdk/log.h"
23 :
24 : static int bdev_rbd_count = 0;
25 :
26 : struct bdev_rbd_pool_ctx {
27 : rados_t *cluster_p;
28 : char *name;
29 : rados_ioctx_t io_ctx;
30 : uint32_t ref;
31 : STAILQ_ENTRY(bdev_rbd_pool_ctx) link;
32 : };
33 :
34 : static STAILQ_HEAD(, bdev_rbd_pool_ctx) g_map_bdev_rbd_pool_ctx = STAILQ_HEAD_INITIALIZER(
35 : g_map_bdev_rbd_pool_ctx);
36 :
37 : struct bdev_rbd {
38 : struct spdk_bdev disk;
39 : char *rbd_name;
40 : char *user_id;
41 : char *pool_name;
42 : char **config;
43 :
44 : rados_t cluster;
45 : rados_t *cluster_p;
46 : char *cluster_name;
47 :
48 : union rbd_ctx {
49 : rados_ioctx_t io_ctx;
50 : struct bdev_rbd_pool_ctx *ctx;
51 : } rados_ctx;
52 :
53 : rbd_image_t image;
54 :
55 : rbd_image_info_t info;
56 : struct spdk_thread *destruct_td;
57 :
58 : TAILQ_ENTRY(bdev_rbd) tailq;
59 : struct spdk_poller *reset_timer;
60 : struct spdk_bdev_io *reset_bdev_io;
61 :
62 : uint64_t rbd_watch_handle;
63 : };
64 :
65 : struct bdev_rbd_io_channel {
66 : struct bdev_rbd *disk;
67 : struct spdk_io_channel *group_ch;
68 : };
69 :
70 : struct bdev_rbd_io {
71 : struct spdk_thread *submit_td;
72 : enum spdk_bdev_io_status status;
73 : rbd_completion_t comp;
74 : size_t total_len;
75 : };
76 :
77 : struct bdev_rbd_cluster {
78 : char *name;
79 : char *user_id;
80 : char **config_param;
81 : char *config_file;
82 : char *key_file;
83 : char *core_mask;
84 : rados_t cluster;
85 : uint32_t ref;
86 : STAILQ_ENTRY(bdev_rbd_cluster) link;
87 : };
88 :
89 : static STAILQ_HEAD(, bdev_rbd_cluster) g_map_bdev_rbd_cluster = STAILQ_HEAD_INITIALIZER(
90 : g_map_bdev_rbd_cluster);
91 : static pthread_mutex_t g_map_bdev_rbd_cluster_mutex = PTHREAD_MUTEX_INITIALIZER;
92 :
93 : static struct spdk_io_channel *bdev_rbd_get_io_channel(void *ctx);
94 :
95 : static void
96 0 : _rbd_update_callback(void *arg)
97 : {
98 0 : struct bdev_rbd *rbd = arg;
99 0 : uint64_t current_size_in_bytes = 0;
100 0 : int rc;
101 :
102 0 : rc = rbd_get_size(rbd->image, ¤t_size_in_bytes);
103 0 : if (rc < 0) {
104 0 : SPDK_ERRLOG("Failed getting size %d\n", rc);
105 0 : return;
106 : }
107 :
108 0 : rc = spdk_bdev_notify_blockcnt_change(&rbd->disk, current_size_in_bytes / rbd->disk.blocklen);
109 0 : if (rc != 0) {
110 0 : SPDK_ERRLOG("failed to notify block cnt change.\n");
111 0 : }
112 0 : }
113 :
114 : static void
115 0 : rbd_update_callback(void *arg)
116 : {
117 0 : spdk_thread_send_msg(spdk_thread_get_app_thread(), _rbd_update_callback, arg);
118 0 : }
119 :
120 : static void
121 0 : bdev_rbd_cluster_free(struct bdev_rbd_cluster *entry)
122 : {
123 0 : assert(entry != NULL);
124 :
125 0 : bdev_rbd_free_config(entry->config_param);
126 0 : free(entry->config_file);
127 0 : free(entry->key_file);
128 0 : free(entry->user_id);
129 0 : free(entry->name);
130 0 : free(entry->core_mask);
131 0 : free(entry);
132 0 : }
133 :
134 : static void
135 0 : bdev_rbd_put_cluster(rados_t **cluster)
136 : {
137 0 : struct bdev_rbd_cluster *entry;
138 :
139 0 : assert(cluster != NULL);
140 :
141 : /* No need go through the map if *cluster equals to NULL */
142 0 : if (*cluster == NULL) {
143 0 : return;
144 : }
145 :
146 0 : pthread_mutex_lock(&g_map_bdev_rbd_cluster_mutex);
147 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
148 0 : if (*cluster != &entry->cluster) {
149 0 : continue;
150 : }
151 :
152 0 : assert(entry->ref > 0);
153 0 : entry->ref--;
154 0 : *cluster = NULL;
155 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
156 0 : return;
157 : }
158 :
159 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
160 0 : SPDK_ERRLOG("Cannot find the entry for cluster=%p\n", cluster);
161 0 : }
162 :
163 : static void
164 0 : bdev_rbd_put_pool_ctx(struct bdev_rbd_pool_ctx *entry)
165 : {
166 0 : assert(spdk_get_thread() == spdk_thread_get_app_thread());
167 :
168 0 : assert(entry != NULL);
169 0 : assert(entry->ref > 0);
170 0 : entry->ref--;
171 0 : if (entry->ref == 0) {
172 0 : STAILQ_REMOVE(&g_map_bdev_rbd_pool_ctx, entry, bdev_rbd_pool_ctx, link);
173 0 : rados_ioctx_destroy(entry->io_ctx);
174 0 : free(entry->name);
175 0 : free(entry);
176 0 : }
177 0 : }
178 :
179 : static void
180 0 : bdev_rbd_free(struct bdev_rbd *rbd)
181 : {
182 0 : if (!rbd) {
183 0 : return;
184 : }
185 :
186 0 : if (rbd->image) {
187 0 : rbd_update_unwatch(rbd->image, rbd->rbd_watch_handle);
188 0 : rbd_flush(rbd->image);
189 0 : rbd_close(rbd->image);
190 0 : }
191 :
192 0 : free(rbd->disk.name);
193 0 : free(rbd->rbd_name);
194 0 : free(rbd->user_id);
195 0 : free(rbd->pool_name);
196 0 : bdev_rbd_free_config(rbd->config);
197 :
198 0 : if (rbd->cluster_name) {
199 : /* When rbd is destructed by bdev_rbd_destruct, it will not enter here
200 : * because the ctx will already freed by bdev_rbd_free_cb in async manner.
201 : * This path only happens during the rbd initialization procedure of rbd */
202 0 : if (rbd->rados_ctx.ctx) {
203 0 : bdev_rbd_put_pool_ctx(rbd->rados_ctx.ctx);
204 0 : rbd->rados_ctx.ctx = NULL;
205 0 : }
206 :
207 0 : bdev_rbd_put_cluster(&rbd->cluster_p);
208 0 : free(rbd->cluster_name);
209 0 : } else if (rbd->cluster) {
210 0 : if (rbd->rados_ctx.io_ctx) {
211 0 : rados_ioctx_destroy(rbd->rados_ctx.io_ctx);
212 0 : }
213 0 : rados_shutdown(rbd->cluster);
214 0 : }
215 :
216 0 : free(rbd);
217 0 : }
218 :
219 : void
220 0 : bdev_rbd_free_config(char **config)
221 : {
222 0 : char **entry;
223 :
224 0 : if (config) {
225 0 : for (entry = config; *entry; entry++) {
226 0 : free(*entry);
227 0 : }
228 0 : free(config);
229 0 : }
230 0 : }
231 :
232 : char **
233 0 : bdev_rbd_dup_config(const char *const *config)
234 : {
235 0 : size_t count;
236 0 : char **copy;
237 :
238 0 : if (!config) {
239 0 : return NULL;
240 : }
241 0 : for (count = 0; config[count]; count++) {}
242 0 : copy = calloc(count + 1, sizeof(*copy));
243 0 : if (!copy) {
244 0 : return NULL;
245 : }
246 0 : for (count = 0; config[count]; count++) {
247 0 : if (!(copy[count] = strdup(config[count]))) {
248 0 : bdev_rbd_free_config(copy);
249 0 : return NULL;
250 : }
251 0 : }
252 0 : return copy;
253 0 : }
254 :
255 : static int
256 0 : bdev_rados_cluster_init(const char *user_id, const char *const *config,
257 : rados_t *cluster)
258 : {
259 0 : int ret;
260 :
261 0 : ret = rados_create(cluster, user_id);
262 0 : if (ret < 0) {
263 0 : SPDK_ERRLOG("Failed to create rados_t struct\n");
264 0 : return -1;
265 : }
266 :
267 0 : if (config) {
268 0 : const char *const *entry = config;
269 0 : while (*entry) {
270 0 : ret = rados_conf_set(*cluster, entry[0], entry[1]);
271 0 : if (ret < 0) {
272 0 : SPDK_ERRLOG("Failed to set %s = %s\n", entry[0], entry[1]);
273 0 : rados_shutdown(*cluster);
274 0 : *cluster = NULL;
275 0 : return -1;
276 : }
277 0 : entry += 2;
278 : }
279 0 : } else {
280 0 : ret = rados_conf_read_file(*cluster, NULL);
281 0 : if (ret < 0) {
282 0 : SPDK_ERRLOG("Failed to read conf file\n");
283 0 : rados_shutdown(*cluster);
284 0 : *cluster = NULL;
285 0 : return -1;
286 : }
287 : }
288 :
289 0 : ret = rados_connect(*cluster);
290 0 : if (ret < 0) {
291 0 : SPDK_ERRLOG("Failed to connect to rbd_pool\n");
292 0 : rados_shutdown(*cluster);
293 0 : *cluster = NULL;
294 0 : return -1;
295 : }
296 :
297 0 : return 0;
298 0 : }
299 :
300 : static int
301 0 : bdev_rbd_get_cluster(const char *cluster_name, rados_t **cluster)
302 : {
303 0 : struct bdev_rbd_cluster *entry;
304 :
305 0 : if (cluster == NULL) {
306 0 : SPDK_ERRLOG("cluster should not be NULL\n");
307 0 : return -1;
308 : }
309 :
310 0 : pthread_mutex_lock(&g_map_bdev_rbd_cluster_mutex);
311 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
312 0 : if (strcmp(cluster_name, entry->name) == 0) {
313 0 : entry->ref++;
314 0 : *cluster = &entry->cluster;
315 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
316 0 : return 0;
317 : }
318 0 : }
319 :
320 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
321 0 : return -1;
322 0 : }
323 :
324 : static int
325 0 : bdev_rbd_shared_cluster_init(const char *cluster_name, rados_t **cluster)
326 : {
327 0 : int ret;
328 :
329 0 : ret = bdev_rbd_get_cluster(cluster_name, cluster);
330 0 : if (ret < 0) {
331 0 : SPDK_ERRLOG("Failed to create rados_t struct\n");
332 0 : return -1;
333 : }
334 :
335 0 : return ret;
336 0 : }
337 :
338 : static void *
339 0 : bdev_rbd_cluster_handle(void *arg)
340 : {
341 0 : void *ret = arg;
342 0 : struct bdev_rbd *rbd = arg;
343 0 : int rc;
344 :
345 0 : rc = bdev_rados_cluster_init(rbd->user_id, (const char *const *)rbd->config,
346 0 : &rbd->cluster);
347 0 : if (rc < 0) {
348 0 : SPDK_ERRLOG("Failed to create rados cluster for user_id=%s and rbd_pool=%s\n",
349 : rbd->user_id ? rbd->user_id : "admin (the default)", rbd->pool_name);
350 0 : ret = NULL;
351 0 : }
352 :
353 0 : return ret;
354 0 : }
355 :
356 : static int
357 0 : bdev_rbd_get_pool_ctx(rados_t *cluster_p, const char *name, struct bdev_rbd_pool_ctx **ctx)
358 : {
359 0 : struct bdev_rbd_pool_ctx *entry;
360 :
361 0 : assert(spdk_get_thread() == spdk_thread_get_app_thread());
362 :
363 0 : if (name == NULL || ctx == NULL) {
364 0 : return -1;
365 : }
366 :
367 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_pool_ctx, link) {
368 0 : if (strcmp(name, entry->name) == 0 && cluster_p == entry->cluster_p) {
369 0 : entry->ref++;
370 0 : *ctx = entry;
371 0 : return 0;
372 : }
373 0 : }
374 :
375 0 : entry = calloc(1, sizeof(*entry));
376 0 : if (!entry) {
377 0 : SPDK_ERRLOG("Cannot allocate an entry for name=%s\n", name);
378 0 : return -1;
379 : }
380 :
381 0 : entry->name = strdup(name);
382 0 : if (entry->name == NULL) {
383 0 : SPDK_ERRLOG("Failed to allocate the name =%s space on entry =%p\n", name, entry);
384 0 : goto err_handle;
385 : }
386 :
387 0 : if (rados_ioctx_create(*cluster_p, name, &entry->io_ctx) < 0) {
388 0 : goto err_handle1;
389 : }
390 :
391 0 : entry->cluster_p = cluster_p;
392 0 : entry->ref = 1;
393 0 : *ctx = entry;
394 0 : STAILQ_INSERT_TAIL(&g_map_bdev_rbd_pool_ctx, entry, link);
395 :
396 0 : return 0;
397 :
398 : err_handle1:
399 0 : free(entry->name);
400 : err_handle:
401 0 : free(entry);
402 :
403 0 : return -1;
404 0 : }
405 :
406 : static void *
407 0 : bdev_rbd_init_context(void *arg)
408 : {
409 0 : struct bdev_rbd *rbd = arg;
410 0 : int rc;
411 0 : rados_ioctx_t *io_ctx = NULL;
412 :
413 0 : if (rbd->cluster_name) {
414 0 : if (bdev_rbd_get_pool_ctx(rbd->cluster_p, rbd->pool_name, &rbd->rados_ctx.ctx) < 0) {
415 0 : SPDK_ERRLOG("Failed to create ioctx on rbd=%p with cluster_name=%s\n",
416 : rbd, rbd->cluster_name);
417 0 : return NULL;
418 : }
419 0 : io_ctx = &rbd->rados_ctx.ctx->io_ctx;
420 0 : } else {
421 0 : if (rados_ioctx_create(*(rbd->cluster_p), rbd->pool_name, &rbd->rados_ctx.io_ctx) < 0) {
422 0 : SPDK_ERRLOG("Failed to create ioctx on rbd=%p\n", rbd);
423 0 : return NULL;
424 : }
425 0 : io_ctx = &rbd->rados_ctx.io_ctx;
426 : }
427 :
428 0 : assert(io_ctx != NULL);
429 0 : rc = rbd_open(*io_ctx, rbd->rbd_name, &rbd->image, NULL);
430 0 : if (rc < 0) {
431 0 : SPDK_ERRLOG("Failed to open specified rbd device\n");
432 0 : return NULL;
433 : }
434 :
435 0 : rc = rbd_update_watch(rbd->image, &rbd->rbd_watch_handle, rbd_update_callback, (void *)rbd);
436 0 : if (rc < 0) {
437 0 : SPDK_ERRLOG("Failed to set up watch %d\n", rc);
438 0 : }
439 :
440 0 : rc = rbd_stat(rbd->image, &rbd->info, sizeof(rbd->info));
441 0 : if (rc < 0) {
442 0 : SPDK_ERRLOG("Failed to stat specified rbd device\n");
443 0 : return NULL;
444 : }
445 :
446 0 : return arg;
447 0 : }
448 :
449 : static int
450 0 : bdev_rbd_init(struct bdev_rbd *rbd)
451 : {
452 0 : int ret = 0;
453 :
454 0 : if (!rbd->cluster_name) {
455 0 : rbd->cluster_p = &rbd->cluster;
456 : /* Cluster should be created in non-SPDK thread to avoid conflict between
457 : * Rados and SPDK thread */
458 0 : if (spdk_call_unaffinitized(bdev_rbd_cluster_handle, rbd) == NULL) {
459 0 : SPDK_ERRLOG("Cannot create the rados object on rbd=%p\n", rbd);
460 0 : return -1;
461 : }
462 0 : } else {
463 0 : ret = bdev_rbd_shared_cluster_init(rbd->cluster_name, &rbd->cluster_p);
464 0 : if (ret < 0) {
465 0 : SPDK_ERRLOG("Failed to create rados object for rbd =%p on cluster_name=%s\n",
466 : rbd, rbd->cluster_name);
467 0 : return -1;
468 : }
469 : }
470 :
471 0 : if (spdk_call_unaffinitized(bdev_rbd_init_context, rbd) == NULL) {
472 0 : SPDK_ERRLOG("Cannot init rbd context for rbd=%p\n", rbd);
473 0 : return -1;
474 : }
475 :
476 0 : return ret;
477 0 : }
478 :
479 : static void
480 0 : _bdev_rbd_io_complete(void *_rbd_io)
481 : {
482 0 : struct bdev_rbd_io *rbd_io = _rbd_io;
483 :
484 0 : spdk_bdev_io_complete(spdk_bdev_io_from_ctx(rbd_io), rbd_io->status);
485 0 : }
486 :
487 : static void
488 0 : bdev_rbd_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
489 : {
490 0 : struct bdev_rbd_io *rbd_io = (struct bdev_rbd_io *)bdev_io->driver_ctx;
491 0 : struct spdk_thread *current_thread = spdk_get_thread();
492 :
493 0 : rbd_io->status = status;
494 0 : assert(rbd_io->submit_td != NULL);
495 0 : if (rbd_io->submit_td != current_thread) {
496 0 : spdk_thread_send_msg(rbd_io->submit_td, _bdev_rbd_io_complete, rbd_io);
497 0 : } else {
498 0 : _bdev_rbd_io_complete(rbd_io);
499 : }
500 0 : }
501 :
502 : static void
503 0 : bdev_rbd_finish_aiocb(rbd_completion_t cb, void *arg)
504 : {
505 0 : int io_status;
506 0 : struct spdk_bdev_io *bdev_io;
507 0 : struct bdev_rbd_io *rbd_io;
508 0 : enum spdk_bdev_io_status bio_status;
509 :
510 0 : bdev_io = rbd_aio_get_arg(cb);
511 0 : rbd_io = (struct bdev_rbd_io *)bdev_io->driver_ctx;
512 0 : io_status = rbd_aio_get_return_value(cb);
513 0 : bio_status = SPDK_BDEV_IO_STATUS_SUCCESS;
514 :
515 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
516 0 : if ((int)rbd_io->total_len != io_status) {
517 0 : bio_status = SPDK_BDEV_IO_STATUS_FAILED;
518 0 : }
519 : #ifdef LIBRBD_SUPPORTS_COMPARE_AND_WRITE_IOVEC
520 0 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE && io_status == -EILSEQ) {
521 0 : bio_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
522 : #endif
523 0 : } else if (io_status != 0) { /* For others, 0 means success */
524 0 : bio_status = SPDK_BDEV_IO_STATUS_FAILED;
525 0 : }
526 :
527 0 : rbd_aio_release(cb);
528 :
529 0 : bdev_rbd_io_complete(bdev_io, bio_status);
530 0 : }
531 :
532 : static void
533 0 : _bdev_rbd_start_aio(struct bdev_rbd *disk, struct spdk_bdev_io *bdev_io,
534 : struct iovec *iov, int iovcnt, uint64_t offset, size_t len)
535 : {
536 0 : int ret;
537 0 : struct bdev_rbd_io *rbd_io = (struct bdev_rbd_io *)bdev_io->driver_ctx;
538 0 : rbd_image_t image = disk->image;
539 :
540 0 : ret = rbd_aio_create_completion(bdev_io, bdev_rbd_finish_aiocb,
541 0 : &rbd_io->comp);
542 0 : if (ret < 0) {
543 0 : goto err;
544 : }
545 :
546 0 : switch (bdev_io->type) {
547 : case SPDK_BDEV_IO_TYPE_READ:
548 0 : rbd_io->total_len = len;
549 0 : if (spdk_likely(iovcnt == 1)) {
550 0 : ret = rbd_aio_read(image, offset, iov[0].iov_len, iov[0].iov_base,
551 0 : rbd_io->comp);
552 0 : } else {
553 0 : ret = rbd_aio_readv(image, iov, iovcnt, offset, rbd_io->comp);
554 : }
555 0 : break;
556 : case SPDK_BDEV_IO_TYPE_WRITE:
557 0 : if (spdk_likely(iovcnt == 1)) {
558 0 : ret = rbd_aio_write(image, offset, iov[0].iov_len, iov[0].iov_base,
559 0 : rbd_io->comp);
560 0 : } else {
561 0 : ret = rbd_aio_writev(image, iov, iovcnt, offset, rbd_io->comp);
562 : }
563 0 : break;
564 : case SPDK_BDEV_IO_TYPE_UNMAP:
565 0 : ret = rbd_aio_discard(image, offset, len, rbd_io->comp);
566 0 : break;
567 : case SPDK_BDEV_IO_TYPE_FLUSH:
568 0 : ret = rbd_aio_flush(image, rbd_io->comp);
569 0 : break;
570 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
571 0 : ret = rbd_aio_write_zeroes(image, offset, len, rbd_io->comp, /* zero_flags */ 0,
572 : /* op_flags */ 0);
573 0 : break;
574 : #ifdef LIBRBD_SUPPORTS_COMPARE_AND_WRITE_IOVEC
575 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
576 0 : ret = rbd_aio_compare_and_writev(image, offset, iov /* cmp */, iovcnt,
577 0 : bdev_io->u.bdev.fused_iovs /* write */,
578 0 : bdev_io->u.bdev.fused_iovcnt,
579 0 : rbd_io->comp, NULL,
580 : /* op_flags */ 0);
581 0 : break;
582 : #endif
583 : default:
584 : /* This should not happen.
585 : * Function should only be called with supported io types in bdev_rbd_submit_request
586 : */
587 0 : SPDK_ERRLOG("Unsupported IO type =%d\n", bdev_io->type);
588 0 : ret = -ENOTSUP;
589 0 : break;
590 : }
591 :
592 0 : if (ret < 0) {
593 0 : rbd_aio_release(rbd_io->comp);
594 0 : goto err;
595 : }
596 :
597 0 : return;
598 :
599 : err:
600 0 : bdev_rbd_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
601 0 : }
602 :
603 : static void
604 0 : bdev_rbd_start_aio(void *ctx)
605 : {
606 0 : struct spdk_bdev_io *bdev_io = ctx;
607 0 : struct bdev_rbd *disk = (struct bdev_rbd *)bdev_io->bdev->ctxt;
608 :
609 0 : _bdev_rbd_start_aio(disk,
610 0 : bdev_io,
611 0 : bdev_io->u.bdev.iovs,
612 0 : bdev_io->u.bdev.iovcnt,
613 0 : bdev_io->u.bdev.offset_blocks * bdev_io->bdev->blocklen,
614 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
615 0 : }
616 :
617 : static int bdev_rbd_library_init(void);
618 : static void bdev_rbd_library_fini(void);
619 :
620 : static int
621 0 : bdev_rbd_get_ctx_size(void)
622 : {
623 0 : return sizeof(struct bdev_rbd_io);
624 : }
625 :
626 : static struct spdk_bdev_module rbd_if = {
627 : .name = "rbd",
628 : .module_init = bdev_rbd_library_init,
629 : .module_fini = bdev_rbd_library_fini,
630 : .get_ctx_size = bdev_rbd_get_ctx_size,
631 :
632 : };
633 0 : SPDK_BDEV_MODULE_REGISTER(rbd, &rbd_if)
634 :
635 : static int bdev_rbd_reset_timer(void *arg);
636 :
637 : static void
638 0 : bdev_rbd_check_outstanding_ios(struct spdk_bdev *bdev, uint64_t current_qd,
639 : void *cb_arg, int rc)
640 : {
641 0 : struct bdev_rbd *disk = cb_arg;
642 0 : enum spdk_bdev_io_status bio_status;
643 :
644 0 : if (rc == 0 && current_qd > 0) {
645 0 : disk->reset_timer = SPDK_POLLER_REGISTER(bdev_rbd_reset_timer, disk, 1000);
646 0 : return;
647 : }
648 :
649 0 : if (rc != 0) {
650 0 : bio_status = SPDK_BDEV_IO_STATUS_FAILED;
651 0 : } else {
652 0 : bio_status = SPDK_BDEV_IO_STATUS_SUCCESS;
653 : }
654 :
655 0 : bdev_rbd_io_complete(disk->reset_bdev_io, bio_status);
656 0 : disk->reset_bdev_io = NULL;
657 0 : }
658 :
659 : static int
660 0 : bdev_rbd_reset_timer(void *arg)
661 : {
662 0 : struct bdev_rbd *disk = arg;
663 :
664 0 : spdk_poller_unregister(&disk->reset_timer);
665 :
666 0 : spdk_bdev_get_current_qd(&disk->disk, bdev_rbd_check_outstanding_ios, disk);
667 :
668 0 : return SPDK_POLLER_BUSY;
669 0 : }
670 :
671 : static void
672 0 : bdev_rbd_reset(void *ctx)
673 : {
674 0 : struct spdk_bdev_io *bdev_io = ctx;
675 0 : struct bdev_rbd *disk = (struct bdev_rbd *)bdev_io->bdev->ctxt;
676 :
677 : /*
678 : * HACK: Since librbd doesn't provide any way to cancel outstanding aio, just kick off a
679 : * poller to wait for in-flight I/O to complete.
680 : */
681 0 : assert(disk->reset_bdev_io == NULL);
682 0 : disk->reset_bdev_io = bdev_io;
683 :
684 0 : bdev_rbd_reset_timer(disk);
685 0 : }
686 :
687 : static void
688 0 : _bdev_rbd_destruct_done(void *io_device)
689 : {
690 0 : struct bdev_rbd *rbd = io_device;
691 :
692 0 : assert(rbd != NULL);
693 :
694 0 : spdk_bdev_destruct_done(&rbd->disk, 0);
695 0 : bdev_rbd_free(rbd);
696 0 : }
697 :
698 : static void
699 0 : bdev_rbd_free_cb(void *io_device)
700 : {
701 0 : struct bdev_rbd *rbd = io_device;
702 :
703 0 : assert(spdk_get_thread() == spdk_thread_get_app_thread());
704 :
705 : /* free the ctx */
706 0 : if (rbd->cluster_name && rbd->rados_ctx.ctx) {
707 0 : bdev_rbd_put_pool_ctx(rbd->rados_ctx.ctx);
708 0 : rbd->rados_ctx.ctx = NULL;
709 0 : }
710 :
711 : /* The io device has been unregistered. Send a message back to the
712 : * original thread that started the destruct operation, so that the
713 : * bdev unregister callback is invoked on the same thread that started
714 : * this whole process.
715 : */
716 0 : spdk_thread_send_msg(rbd->destruct_td, _bdev_rbd_destruct_done, rbd);
717 0 : }
718 :
719 : static void
720 0 : _bdev_rbd_destruct(void *ctx)
721 : {
722 0 : struct bdev_rbd *rbd = ctx;
723 :
724 0 : spdk_io_device_unregister(rbd, bdev_rbd_free_cb);
725 0 : }
726 :
727 : static int
728 0 : bdev_rbd_destruct(void *ctx)
729 : {
730 0 : struct bdev_rbd *rbd = ctx;
731 :
732 : /* Start the destruct operation on the rbd bdev's
733 : * main thread. This guarantees it will only start
734 : * executing after any messages related to channel
735 : * deletions have finished completing. *Always*
736 : * send a message, even if this function gets called
737 : * from the main thread, in case there are pending
738 : * channel delete messages in flight to this thread.
739 : */
740 0 : assert(rbd->destruct_td == NULL);
741 0 : rbd->destruct_td = spdk_get_thread();
742 0 : spdk_thread_send_msg(spdk_thread_get_app_thread(), _bdev_rbd_destruct, rbd);
743 :
744 : /* Return 1 to indicate the destruct path is asynchronous. */
745 0 : return 1;
746 0 : }
747 :
748 : static void
749 0 : bdev_rbd_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
750 : bool success)
751 : {
752 0 : if (!success) {
753 0 : bdev_rbd_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
754 0 : return;
755 : }
756 :
757 0 : bdev_rbd_start_aio(bdev_io);
758 0 : }
759 :
760 : static void
761 0 : bdev_rbd_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
762 : {
763 0 : struct spdk_thread *submit_td = spdk_io_channel_get_thread(ch);
764 0 : struct bdev_rbd_io *rbd_io = (struct bdev_rbd_io *)bdev_io->driver_ctx;
765 :
766 0 : rbd_io->submit_td = submit_td;
767 0 : switch (bdev_io->type) {
768 : case SPDK_BDEV_IO_TYPE_READ:
769 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rbd_get_buf_cb,
770 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
771 0 : break;
772 :
773 : case SPDK_BDEV_IO_TYPE_WRITE:
774 : case SPDK_BDEV_IO_TYPE_UNMAP:
775 : case SPDK_BDEV_IO_TYPE_FLUSH:
776 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
777 : #ifdef LIBRBD_SUPPORTS_COMPARE_AND_WRITE_IOVEC
778 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
779 : #endif
780 0 : bdev_rbd_start_aio(bdev_io);
781 0 : break;
782 :
783 : case SPDK_BDEV_IO_TYPE_RESET:
784 0 : spdk_thread_exec_msg(spdk_thread_get_app_thread(), bdev_rbd_reset, bdev_io);
785 0 : break;
786 :
787 : default:
788 0 : SPDK_ERRLOG("Unsupported IO type =%d\n", bdev_io->type);
789 0 : bdev_rbd_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
790 0 : break;
791 : }
792 0 : }
793 :
794 : static bool
795 0 : bdev_rbd_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
796 : {
797 0 : switch (io_type) {
798 : case SPDK_BDEV_IO_TYPE_READ:
799 : case SPDK_BDEV_IO_TYPE_WRITE:
800 : case SPDK_BDEV_IO_TYPE_UNMAP:
801 : case SPDK_BDEV_IO_TYPE_FLUSH:
802 : case SPDK_BDEV_IO_TYPE_RESET:
803 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
804 : #ifdef LIBRBD_SUPPORTS_COMPARE_AND_WRITE_IOVEC
805 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
806 : #endif
807 0 : return true;
808 :
809 : default:
810 0 : return false;
811 : }
812 0 : }
813 :
814 : static int
815 0 : bdev_rbd_create_cb(void *io_device, void *ctx_buf)
816 : {
817 0 : struct bdev_rbd_io_channel *ch = ctx_buf;
818 0 : struct bdev_rbd *disk = io_device;
819 :
820 0 : ch->disk = disk;
821 0 : ch->group_ch = spdk_get_io_channel(&rbd_if);
822 0 : assert(ch->group_ch != NULL);
823 :
824 0 : return 0;
825 0 : }
826 :
827 : static void
828 0 : bdev_rbd_destroy_cb(void *io_device, void *ctx_buf)
829 : {
830 0 : struct bdev_rbd_io_channel *ch = ctx_buf;
831 :
832 0 : spdk_put_io_channel(ch->group_ch);
833 0 : }
834 :
835 : static struct spdk_io_channel *
836 0 : bdev_rbd_get_io_channel(void *ctx)
837 : {
838 0 : struct bdev_rbd *rbd_bdev = ctx;
839 :
840 0 : return spdk_get_io_channel(rbd_bdev);
841 0 : }
842 :
843 : static void
844 0 : bdev_rbd_cluster_dump_entry(const char *cluster_name, struct spdk_json_write_ctx *w)
845 : {
846 0 : struct bdev_rbd_cluster *entry;
847 :
848 0 : pthread_mutex_lock(&g_map_bdev_rbd_cluster_mutex);
849 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
850 0 : if (strcmp(cluster_name, entry->name)) {
851 0 : continue;
852 : }
853 0 : if (entry->user_id) {
854 0 : spdk_json_write_named_string(w, "user_id", entry->user_id);
855 0 : }
856 :
857 0 : if (entry->config_param) {
858 0 : char **config_entry = entry->config_param;
859 :
860 0 : spdk_json_write_named_object_begin(w, "config_param");
861 0 : while (*config_entry) {
862 0 : spdk_json_write_named_string(w, config_entry[0], config_entry[1]);
863 0 : config_entry += 2;
864 : }
865 0 : spdk_json_write_object_end(w);
866 0 : }
867 0 : if (entry->config_file) {
868 0 : spdk_json_write_named_string(w, "config_file", entry->config_file);
869 0 : }
870 0 : if (entry->key_file) {
871 0 : spdk_json_write_named_string(w, "key_file", entry->key_file);
872 0 : }
873 :
874 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
875 0 : return;
876 : }
877 :
878 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
879 0 : }
880 :
881 : static int
882 0 : bdev_rbd_dump_info_json(void *ctx, struct spdk_json_write_ctx *w)
883 : {
884 0 : struct bdev_rbd *rbd_bdev = ctx;
885 :
886 0 : spdk_json_write_named_object_begin(w, "rbd");
887 :
888 0 : spdk_json_write_named_string(w, "pool_name", rbd_bdev->pool_name);
889 :
890 0 : spdk_json_write_named_string(w, "rbd_name", rbd_bdev->rbd_name);
891 :
892 0 : if (rbd_bdev->cluster_name) {
893 0 : bdev_rbd_cluster_dump_entry(rbd_bdev->cluster_name, w);
894 0 : goto end;
895 : }
896 :
897 0 : if (rbd_bdev->user_id) {
898 0 : spdk_json_write_named_string(w, "user_id", rbd_bdev->user_id);
899 0 : }
900 :
901 0 : if (rbd_bdev->config) {
902 0 : char **entry = rbd_bdev->config;
903 :
904 0 : spdk_json_write_named_object_begin(w, "config");
905 0 : while (*entry) {
906 0 : spdk_json_write_named_string(w, entry[0], entry[1]);
907 0 : entry += 2;
908 : }
909 0 : spdk_json_write_object_end(w);
910 0 : }
911 :
912 : end:
913 0 : spdk_json_write_object_end(w);
914 :
915 0 : return 0;
916 0 : }
917 :
918 : static void
919 0 : bdev_rbd_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
920 : {
921 0 : struct bdev_rbd *rbd = bdev->ctxt;
922 :
923 0 : spdk_json_write_object_begin(w);
924 :
925 0 : spdk_json_write_named_string(w, "method", "bdev_rbd_create");
926 :
927 0 : spdk_json_write_named_object_begin(w, "params");
928 0 : spdk_json_write_named_string(w, "name", bdev->name);
929 0 : spdk_json_write_named_string(w, "pool_name", rbd->pool_name);
930 0 : spdk_json_write_named_string(w, "rbd_name", rbd->rbd_name);
931 0 : spdk_json_write_named_uint32(w, "block_size", bdev->blocklen);
932 0 : if (rbd->user_id) {
933 0 : spdk_json_write_named_string(w, "user_id", rbd->user_id);
934 0 : }
935 :
936 0 : if (rbd->config) {
937 0 : char **entry = rbd->config;
938 :
939 0 : spdk_json_write_named_object_begin(w, "config");
940 0 : while (*entry) {
941 0 : spdk_json_write_named_string(w, entry[0], entry[1]);
942 0 : entry += 2;
943 : }
944 0 : spdk_json_write_object_end(w);
945 0 : }
946 :
947 0 : spdk_json_write_named_uuid(w, "uuid", &bdev->uuid);
948 :
949 0 : spdk_json_write_object_end(w);
950 :
951 0 : spdk_json_write_object_end(w);
952 0 : }
953 :
954 : static void
955 0 : dump_single_cluster_entry(struct bdev_rbd_cluster *entry, struct spdk_json_write_ctx *w)
956 : {
957 0 : assert(entry != NULL);
958 :
959 0 : spdk_json_write_object_begin(w);
960 0 : spdk_json_write_named_string(w, "cluster_name", entry->name);
961 :
962 0 : if (entry->user_id) {
963 0 : spdk_json_write_named_string(w, "user_id", entry->user_id);
964 0 : }
965 :
966 0 : if (entry->config_param) {
967 0 : char **config_entry = entry->config_param;
968 :
969 0 : spdk_json_write_named_object_begin(w, "config_param");
970 0 : while (*config_entry) {
971 0 : spdk_json_write_named_string(w, config_entry[0], config_entry[1]);
972 0 : config_entry += 2;
973 : }
974 0 : spdk_json_write_object_end(w);
975 0 : }
976 0 : if (entry->config_file) {
977 0 : spdk_json_write_named_string(w, "config_file", entry->config_file);
978 0 : }
979 0 : if (entry->key_file) {
980 0 : spdk_json_write_named_string(w, "key_file", entry->key_file);
981 0 : }
982 :
983 0 : if (entry->core_mask) {
984 0 : spdk_json_write_named_string(w, "core_mask", entry->core_mask);
985 0 : }
986 :
987 0 : spdk_json_write_object_end(w);
988 0 : }
989 :
990 : int
991 0 : bdev_rbd_get_clusters_info(struct spdk_jsonrpc_request *request, const char *name)
992 : {
993 0 : struct bdev_rbd_cluster *entry;
994 0 : struct spdk_json_write_ctx *w;
995 :
996 0 : pthread_mutex_lock(&g_map_bdev_rbd_cluster_mutex);
997 :
998 0 : if (STAILQ_EMPTY(&g_map_bdev_rbd_cluster)) {
999 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1000 0 : return -ENOENT;
1001 : }
1002 :
1003 : /* If cluster name is provided */
1004 0 : if (name) {
1005 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
1006 0 : if (strcmp(name, entry->name) == 0) {
1007 0 : w = spdk_jsonrpc_begin_result(request);
1008 0 : dump_single_cluster_entry(entry, w);
1009 0 : spdk_jsonrpc_end_result(request, w);
1010 :
1011 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1012 0 : return 0;
1013 : }
1014 0 : }
1015 :
1016 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1017 0 : return -ENOENT;
1018 : }
1019 :
1020 0 : w = spdk_jsonrpc_begin_result(request);
1021 0 : spdk_json_write_array_begin(w);
1022 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
1023 0 : dump_single_cluster_entry(entry, w);
1024 0 : }
1025 0 : spdk_json_write_array_end(w);
1026 0 : spdk_jsonrpc_end_result(request, w);
1027 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1028 :
1029 0 : return 0;
1030 0 : }
1031 :
1032 : static const struct spdk_bdev_fn_table rbd_fn_table = {
1033 : .destruct = bdev_rbd_destruct,
1034 : .submit_request = bdev_rbd_submit_request,
1035 : .io_type_supported = bdev_rbd_io_type_supported,
1036 : .get_io_channel = bdev_rbd_get_io_channel,
1037 : .dump_info_json = bdev_rbd_dump_info_json,
1038 : .write_config_json = bdev_rbd_write_config_json,
1039 : };
1040 :
1041 : static int
1042 0 : rbd_thread_set_cpumask(struct spdk_cpuset *set)
1043 : {
1044 : #ifdef __linux__
1045 0 : uint32_t lcore;
1046 0 : cpu_set_t mask;
1047 :
1048 0 : assert(set != NULL);
1049 0 : CPU_ZERO(&mask);
1050 :
1051 : /* get the core id on current spdk_cpuset and set to cpu_set_t */
1052 0 : for (lcore = 0; lcore < SPDK_CPUSET_SIZE; lcore++) {
1053 0 : if (spdk_cpuset_get_cpu(set, lcore)) {
1054 0 : CPU_SET(lcore, &mask);
1055 0 : }
1056 0 : }
1057 :
1058 : /* change current thread core mask */
1059 0 : if (sched_setaffinity(0, sizeof(mask), &mask) < 0) {
1060 0 : SPDK_ERRLOG("Set non SPDK thread cpu mask error (errno=%d)\n", errno);
1061 0 : return -1;
1062 : }
1063 :
1064 0 : return 0;
1065 : #else
1066 : SPDK_ERRLOG("SPDK non spdk thread cpumask setup supports only Linux platform now.\n");
1067 : return -ENOTSUP;
1068 : #endif
1069 0 : }
1070 :
1071 :
1072 : static int
1073 0 : rbd_register_cluster(const char *name, const char *user_id, const char *const *config_param,
1074 : const char *config_file, const char *key_file, const char *core_mask)
1075 : {
1076 0 : struct bdev_rbd_cluster *entry;
1077 0 : struct spdk_cpuset rbd_core_mask = {};
1078 0 : int rc;
1079 :
1080 0 : pthread_mutex_lock(&g_map_bdev_rbd_cluster_mutex);
1081 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
1082 0 : if (strcmp(name, entry->name) == 0) {
1083 0 : SPDK_ERRLOG("Cluster name=%s already exists\n", name);
1084 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1085 0 : return -1;
1086 : }
1087 0 : }
1088 :
1089 0 : entry = calloc(1, sizeof(*entry));
1090 0 : if (!entry) {
1091 0 : SPDK_ERRLOG("Cannot allocate an entry for name=%s\n", name);
1092 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1093 0 : return -1;
1094 : }
1095 :
1096 0 : entry->name = strdup(name);
1097 0 : if (entry->name == NULL) {
1098 0 : SPDK_ERRLOG("Failed to save the name =%s on entry =%p\n", name, entry);
1099 0 : goto err_handle;
1100 : }
1101 :
1102 0 : if (user_id) {
1103 0 : entry->user_id = strdup(user_id);
1104 0 : if (entry->user_id == NULL) {
1105 0 : SPDK_ERRLOG("Failed to save the str =%s on entry =%p\n", user_id, entry);
1106 0 : goto err_handle;
1107 : }
1108 0 : }
1109 :
1110 : /* Support specify config_param or config_file separately, or both of them. */
1111 0 : if (config_param) {
1112 0 : entry->config_param = bdev_rbd_dup_config(config_param);
1113 0 : if (entry->config_param == NULL) {
1114 0 : SPDK_ERRLOG("Failed to save the config_param=%p on entry = %p\n", config_param, entry);
1115 0 : goto err_handle;
1116 : }
1117 0 : }
1118 :
1119 0 : if (config_file) {
1120 0 : entry->config_file = strdup(config_file);
1121 0 : if (entry->config_file == NULL) {
1122 0 : SPDK_ERRLOG("Failed to save the config_file=%s on entry = %p\n", config_file, entry);
1123 0 : goto err_handle;
1124 : }
1125 0 : }
1126 :
1127 0 : if (key_file) {
1128 0 : entry->key_file = strdup(key_file);
1129 0 : if (entry->key_file == NULL) {
1130 0 : SPDK_ERRLOG("Failed to save the key_file=%s on entry = %p\n", key_file, entry);
1131 0 : goto err_handle;
1132 : }
1133 0 : }
1134 :
1135 0 : if (core_mask) {
1136 0 : entry->core_mask = strdup(core_mask);
1137 0 : if (entry->core_mask == NULL) {
1138 0 : SPDK_ERRLOG("Core_mask=%s allocation failed on entry = %p\n", core_mask, entry);
1139 0 : goto err_handle;
1140 : }
1141 :
1142 0 : if (spdk_cpuset_parse(&rbd_core_mask, entry->core_mask) < 0) {
1143 0 : SPDK_ERRLOG("Invalid cpumask=%s on entry = %p\n", entry->core_mask, entry);
1144 0 : goto err_handle;
1145 : }
1146 :
1147 0 : if (rbd_thread_set_cpumask(&rbd_core_mask) < 0) {
1148 0 : SPDK_ERRLOG("Failed to change rbd threads to core_mask %s on entry = %p\n", core_mask, entry);
1149 0 : goto err_handle;
1150 : }
1151 0 : }
1152 :
1153 :
1154 : /* If rbd thread core mask is given, rados_create() must execute with
1155 : * the affinity set by rbd_thread_set_cpumask(). The affinity set
1156 : * by rbd_thread_set_cpumask() will be reverted once rbd_register_cluster() returns
1157 : * and when we leave the spdk_call_unaffinitized context. */
1158 0 : rc = rados_create(&entry->cluster, user_id);
1159 0 : if (rc < 0) {
1160 0 : SPDK_ERRLOG("Failed to create rados_t struct\n");
1161 0 : goto err_handle;
1162 : }
1163 :
1164 : /* Try default location when entry->config_file is NULL, but ignore failure when it is NULL */
1165 0 : rc = rados_conf_read_file(entry->cluster, entry->config_file);
1166 0 : if (entry->config_file && rc < 0) {
1167 0 : SPDK_ERRLOG("Failed to read conf file %s\n", entry->config_file);
1168 0 : rados_shutdown(entry->cluster);
1169 0 : goto err_handle;
1170 : }
1171 :
1172 0 : if (config_param) {
1173 0 : const char *const *config_entry = config_param;
1174 0 : while (*config_entry) {
1175 0 : rc = rados_conf_set(entry->cluster, config_entry[0], config_entry[1]);
1176 0 : if (rc < 0) {
1177 0 : SPDK_ERRLOG("Failed to set %s = %s\n", config_entry[0], config_entry[1]);
1178 0 : rados_shutdown(entry->cluster);
1179 0 : goto err_handle;
1180 : }
1181 0 : config_entry += 2;
1182 : }
1183 0 : }
1184 :
1185 0 : if (key_file) {
1186 0 : rc = rados_conf_set(entry->cluster, "keyring", key_file);
1187 0 : if (rc < 0) {
1188 0 : SPDK_ERRLOG("Failed to set keyring = %s\n", key_file);
1189 0 : rados_shutdown(entry->cluster);
1190 0 : goto err_handle;
1191 : }
1192 0 : }
1193 :
1194 0 : rc = rados_connect(entry->cluster);
1195 0 : if (rc < 0) {
1196 0 : SPDK_ERRLOG("Failed to connect to rbd_pool on cluster=%p\n", entry->cluster);
1197 0 : rados_shutdown(entry->cluster);
1198 0 : goto err_handle;
1199 : }
1200 :
1201 0 : STAILQ_INSERT_TAIL(&g_map_bdev_rbd_cluster, entry, link);
1202 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1203 :
1204 0 : return 0;
1205 :
1206 : err_handle:
1207 0 : bdev_rbd_cluster_free(entry);
1208 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1209 0 : return -1;
1210 0 : }
1211 :
1212 : int
1213 0 : bdev_rbd_unregister_cluster(const char *name)
1214 : {
1215 0 : struct bdev_rbd_cluster *entry;
1216 0 : int rc = 0;
1217 :
1218 0 : if (name == NULL) {
1219 0 : return -1;
1220 : }
1221 :
1222 0 : pthread_mutex_lock(&g_map_bdev_rbd_cluster_mutex);
1223 0 : STAILQ_FOREACH(entry, &g_map_bdev_rbd_cluster, link) {
1224 0 : if (strcmp(name, entry->name) == 0) {
1225 0 : if (entry->ref == 0) {
1226 0 : STAILQ_REMOVE(&g_map_bdev_rbd_cluster, entry, bdev_rbd_cluster, link);
1227 0 : rados_shutdown(entry->cluster);
1228 0 : bdev_rbd_cluster_free(entry);
1229 0 : } else {
1230 0 : SPDK_ERRLOG("Cluster with name=%p is still used and we cannot delete it\n",
1231 : entry->name);
1232 0 : rc = -1;
1233 : }
1234 :
1235 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1236 0 : return rc;
1237 : }
1238 0 : }
1239 :
1240 0 : pthread_mutex_unlock(&g_map_bdev_rbd_cluster_mutex);
1241 :
1242 0 : SPDK_ERRLOG("Could not find the cluster name =%p\n", name);
1243 :
1244 0 : return -1;
1245 0 : }
1246 :
1247 : static void *
1248 0 : _bdev_rbd_register_cluster(void *arg)
1249 : {
1250 0 : struct cluster_register_info *info = arg;
1251 0 : void *ret = arg;
1252 0 : int rc;
1253 :
1254 0 : rc = rbd_register_cluster((const char *)info->name, (const char *)info->user_id,
1255 0 : (const char *const *)info->config_param, (const char *)info->config_file,
1256 0 : (const char *)info->key_file, info->core_mask);
1257 0 : if (rc) {
1258 0 : ret = NULL;
1259 0 : }
1260 :
1261 0 : return ret;
1262 0 : }
1263 :
1264 : int
1265 0 : bdev_rbd_register_cluster(struct cluster_register_info *info)
1266 : {
1267 0 : assert(info != NULL);
1268 :
1269 : /* Rados cluster info need to be created in non SPDK-thread to avoid CPU
1270 : * resource contention */
1271 0 : if (spdk_call_unaffinitized(_bdev_rbd_register_cluster, info) == NULL) {
1272 0 : return -1;
1273 : }
1274 :
1275 0 : return 0;
1276 0 : }
1277 :
1278 : int
1279 0 : bdev_rbd_create(struct spdk_bdev **bdev, const char *name, const char *user_id,
1280 : const char *pool_name,
1281 : const char *const *config,
1282 : const char *rbd_name,
1283 : uint32_t block_size,
1284 : const char *cluster_name,
1285 : const struct spdk_uuid *uuid)
1286 : {
1287 0 : struct bdev_rbd *rbd;
1288 0 : int ret;
1289 :
1290 0 : if ((pool_name == NULL) || (rbd_name == NULL) || (block_size == 0)) {
1291 0 : return -EINVAL;
1292 : }
1293 :
1294 0 : rbd = calloc(1, sizeof(struct bdev_rbd));
1295 0 : if (rbd == NULL) {
1296 0 : SPDK_ERRLOG("Failed to allocate bdev_rbd struct\n");
1297 0 : return -ENOMEM;
1298 : }
1299 :
1300 0 : rbd->rbd_name = strdup(rbd_name);
1301 0 : if (!rbd->rbd_name) {
1302 0 : bdev_rbd_free(rbd);
1303 0 : return -ENOMEM;
1304 : }
1305 :
1306 0 : if (user_id) {
1307 0 : rbd->user_id = strdup(user_id);
1308 0 : if (!rbd->user_id) {
1309 0 : bdev_rbd_free(rbd);
1310 0 : return -ENOMEM;
1311 : }
1312 0 : }
1313 :
1314 0 : if (cluster_name) {
1315 0 : rbd->cluster_name = strdup(cluster_name);
1316 0 : if (!rbd->cluster_name) {
1317 0 : bdev_rbd_free(rbd);
1318 0 : return -ENOMEM;
1319 : }
1320 0 : }
1321 0 : rbd->pool_name = strdup(pool_name);
1322 0 : if (!rbd->pool_name) {
1323 0 : bdev_rbd_free(rbd);
1324 0 : return -ENOMEM;
1325 : }
1326 :
1327 0 : if (config && !(rbd->config = bdev_rbd_dup_config(config))) {
1328 0 : bdev_rbd_free(rbd);
1329 0 : return -ENOMEM;
1330 : }
1331 :
1332 0 : ret = bdev_rbd_init(rbd);
1333 0 : if (ret < 0) {
1334 0 : bdev_rbd_free(rbd);
1335 0 : SPDK_ERRLOG("Failed to init rbd device\n");
1336 0 : return ret;
1337 : }
1338 :
1339 0 : rbd->disk.uuid = *uuid;
1340 0 : if (name) {
1341 0 : rbd->disk.name = strdup(name);
1342 0 : } else {
1343 0 : rbd->disk.name = spdk_sprintf_alloc("Ceph%d", bdev_rbd_count);
1344 : }
1345 0 : if (!rbd->disk.name) {
1346 0 : bdev_rbd_free(rbd);
1347 0 : return -ENOMEM;
1348 : }
1349 0 : rbd->disk.product_name = "Ceph Rbd Disk";
1350 0 : bdev_rbd_count++;
1351 :
1352 0 : rbd->disk.write_cache = 0;
1353 0 : rbd->disk.blocklen = block_size;
1354 0 : rbd->disk.blockcnt = rbd->info.size / rbd->disk.blocklen;
1355 0 : rbd->disk.ctxt = rbd;
1356 0 : rbd->disk.fn_table = &rbd_fn_table;
1357 0 : rbd->disk.module = &rbd_if;
1358 :
1359 0 : SPDK_NOTICELOG("Add %s rbd disk to lun\n", rbd->disk.name);
1360 :
1361 0 : spdk_io_device_register(rbd, bdev_rbd_create_cb,
1362 : bdev_rbd_destroy_cb,
1363 : sizeof(struct bdev_rbd_io_channel),
1364 0 : rbd_name);
1365 0 : ret = spdk_bdev_register(&rbd->disk);
1366 0 : if (ret) {
1367 0 : spdk_io_device_unregister(rbd, NULL);
1368 0 : bdev_rbd_free(rbd);
1369 0 : return ret;
1370 : }
1371 :
1372 0 : *bdev = &(rbd->disk);
1373 :
1374 0 : return ret;
1375 0 : }
1376 :
1377 : void
1378 0 : bdev_rbd_delete(const char *name, spdk_delete_rbd_complete cb_fn, void *cb_arg)
1379 : {
1380 0 : int rc;
1381 :
1382 0 : rc = spdk_bdev_unregister_by_name(name, &rbd_if, cb_fn, cb_arg);
1383 0 : if (rc != 0) {
1384 0 : cb_fn(cb_arg, rc);
1385 0 : }
1386 0 : }
1387 :
1388 : static void
1389 0 : dummy_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
1390 : {
1391 0 : }
1392 :
1393 : int
1394 0 : bdev_rbd_resize(const char *name, const uint64_t new_size_in_mb)
1395 : {
1396 0 : struct spdk_bdev_desc *desc;
1397 0 : struct spdk_bdev *bdev;
1398 0 : struct bdev_rbd *rbd;
1399 0 : int rc = 0;
1400 0 : uint64_t new_size_in_byte;
1401 0 : uint64_t current_size_in_mb;
1402 :
1403 0 : rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &desc);
1404 0 : if (rc != 0) {
1405 0 : return rc;
1406 : }
1407 :
1408 0 : bdev = spdk_bdev_desc_get_bdev(desc);
1409 :
1410 0 : if (bdev->module != &rbd_if) {
1411 0 : rc = -EINVAL;
1412 0 : goto exit;
1413 : }
1414 :
1415 0 : current_size_in_mb = bdev->blocklen * bdev->blockcnt / (1024 * 1024);
1416 0 : if (current_size_in_mb > new_size_in_mb) {
1417 0 : SPDK_ERRLOG("The new bdev size must be larger than current bdev size.\n");
1418 0 : rc = -EINVAL;
1419 0 : goto exit;
1420 : }
1421 :
1422 0 : rbd = SPDK_CONTAINEROF(bdev, struct bdev_rbd, disk);
1423 0 : new_size_in_byte = new_size_in_mb * 1024 * 1024;
1424 0 : rc = rbd_resize(rbd->image, new_size_in_byte);
1425 0 : if (rc != 0) {
1426 0 : SPDK_ERRLOG("failed to resize the ceph bdev.\n");
1427 0 : goto exit;
1428 : }
1429 :
1430 0 : rc = spdk_bdev_notify_blockcnt_change(bdev, new_size_in_byte / bdev->blocklen);
1431 0 : if (rc != 0) {
1432 0 : SPDK_ERRLOG("failed to notify block cnt change.\n");
1433 0 : }
1434 :
1435 : exit:
1436 0 : spdk_bdev_close(desc);
1437 0 : return rc;
1438 0 : }
1439 :
1440 : static int
1441 0 : bdev_rbd_group_create_cb(void *io_device, void *ctx_buf)
1442 : {
1443 0 : return 0;
1444 : }
1445 :
1446 : static void
1447 0 : bdev_rbd_group_destroy_cb(void *io_device, void *ctx_buf)
1448 : {
1449 0 : }
1450 :
1451 : static int
1452 0 : bdev_rbd_library_init(void)
1453 : {
1454 0 : spdk_io_device_register(&rbd_if, bdev_rbd_group_create_cb, bdev_rbd_group_destroy_cb,
1455 : 0, "bdev_rbd_poll_groups");
1456 0 : return 0;
1457 : }
1458 :
1459 : static void
1460 0 : bdev_rbd_library_fini(void)
1461 : {
1462 0 : spdk_io_device_unregister(&rbd_if, NULL);
1463 0 : }
1464 :
1465 0 : SPDK_LOG_REGISTER_COMPONENT(bdev_rbd)
|
