Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2016 Intel Corporation. All rights reserved.
3 : * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
4 : * Copyright (c) 2021-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : * Copyright (c) 2022 Dell Inc, or its subsidiaries. All rights reserved.
6 : */
7 :
8 : #include "spdk/stdinc.h"
9 :
10 : #include "bdev_nvme.h"
11 :
12 : #include "spdk/accel.h"
13 : #include "spdk/config.h"
14 : #include "spdk/endian.h"
15 : #include "spdk/bdev.h"
16 : #include "spdk/json.h"
17 : #include "spdk/keyring.h"
18 : #include "spdk/likely.h"
19 : #include "spdk/nvme.h"
20 : #include "spdk/nvme_ocssd.h"
21 : #include "spdk/nvme_zns.h"
22 : #include "spdk/opal.h"
23 : #include "spdk/thread.h"
24 : #include "spdk/trace.h"
25 : #include "spdk/string.h"
26 : #include "spdk/util.h"
27 : #include "spdk/uuid.h"
28 :
29 : #include "spdk/bdev_module.h"
30 : #include "spdk/log.h"
31 :
32 : #include "spdk_internal/usdt.h"
33 : #include "spdk_internal/trace_defs.h"
34 :
35 : #define SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT true
36 : #define SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS (10000)
37 :
38 : #define NSID_STR_LEN 10
39 :
40 : #define SPDK_CONTROLLER_NAME_MAX 512
41 :
42 : static int bdev_nvme_config_json(struct spdk_json_write_ctx *w);
43 :
44 : struct nvme_bdev_io {
45 : /** array of iovecs to transfer. */
46 : struct iovec *iovs;
47 :
48 : /** Number of iovecs in iovs array. */
49 : int iovcnt;
50 :
51 : /** Current iovec position. */
52 : int iovpos;
53 :
54 : /** Offset in current iovec. */
55 : uint32_t iov_offset;
56 :
57 : /** I/O path the current I/O or admin passthrough is submitted on, or the I/O path
58 : * being reset in a reset I/O.
59 : */
60 : struct nvme_io_path *io_path;
61 :
62 : /** array of iovecs to transfer. */
63 : struct iovec *fused_iovs;
64 :
65 : /** Number of iovecs in iovs array. */
66 : int fused_iovcnt;
67 :
68 : /** Current iovec position. */
69 : int fused_iovpos;
70 :
71 : /** Offset in current iovec. */
72 : uint32_t fused_iov_offset;
73 :
74 : /** Saved status for admin passthru completion event, PI error verification, or intermediate compare-and-write status */
75 : struct spdk_nvme_cpl cpl;
76 :
77 : /** Extended IO opts passed by the user to bdev layer and mapped to NVME format */
78 : struct spdk_nvme_ns_cmd_ext_io_opts ext_opts;
79 :
80 : /** Keeps track if first of fused commands was submitted */
81 : bool first_fused_submitted;
82 :
83 : /** Keeps track if first of fused commands was completed */
84 : bool first_fused_completed;
85 :
86 : /** Temporary pointer to zone report buffer */
87 : struct spdk_nvme_zns_zone_report *zone_report_buf;
88 :
89 : /** Keep track of how many zones that have been copied to the spdk_bdev_zone_info struct */
90 : uint64_t handled_zones;
91 :
92 : /** Expiration value in ticks to retry the current I/O. */
93 : uint64_t retry_ticks;
94 :
95 : /* How many times the current I/O was retried. */
96 : int32_t retry_count;
97 :
98 : /* Current tsc at submit time. */
99 : uint64_t submit_tsc;
100 : };
101 :
102 : struct nvme_probe_skip_entry {
103 : struct spdk_nvme_transport_id trid;
104 : TAILQ_ENTRY(nvme_probe_skip_entry) tailq;
105 : };
106 : /* All the controllers deleted by users via RPC are skipped by hotplug monitor */
107 : static TAILQ_HEAD(, nvme_probe_skip_entry) g_skipped_nvme_ctrlrs = TAILQ_HEAD_INITIALIZER(
108 : g_skipped_nvme_ctrlrs);
109 :
110 : #define BDEV_NVME_DEFAULT_DIGESTS (SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA256) | \
111 : SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA384) | \
112 : SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA512))
113 :
114 : #define BDEV_NVME_DEFAULT_DHGROUPS (SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_NULL) | \
115 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_2048) | \
116 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_3072) | \
117 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_4096) | \
118 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_6144) | \
119 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_8192))
120 :
121 : static struct spdk_bdev_nvme_opts g_opts = {
122 : .action_on_timeout = SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE,
123 : .timeout_us = 0,
124 : .timeout_admin_us = 0,
125 : .keep_alive_timeout_ms = SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS,
126 : .transport_retry_count = 4,
127 : .arbitration_burst = 0,
128 : .low_priority_weight = 0,
129 : .medium_priority_weight = 0,
130 : .high_priority_weight = 0,
131 : .nvme_adminq_poll_period_us = 10000ULL,
132 : .nvme_ioq_poll_period_us = 0,
133 : .io_queue_requests = 0,
134 : .delay_cmd_submit = SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT,
135 : .bdev_retry_count = 3,
136 : .transport_ack_timeout = 0,
137 : .ctrlr_loss_timeout_sec = 0,
138 : .reconnect_delay_sec = 0,
139 : .fast_io_fail_timeout_sec = 0,
140 : .disable_auto_failback = false,
141 : .generate_uuids = false,
142 : .transport_tos = 0,
143 : .nvme_error_stat = false,
144 : .io_path_stat = false,
145 : .allow_accel_sequence = false,
146 : .dhchap_digests = BDEV_NVME_DEFAULT_DIGESTS,
147 : .dhchap_dhgroups = BDEV_NVME_DEFAULT_DHGROUPS,
148 : };
149 :
150 : #define NVME_HOTPLUG_POLL_PERIOD_MAX 10000000ULL
151 : #define NVME_HOTPLUG_POLL_PERIOD_DEFAULT 100000ULL
152 :
153 : static int g_hot_insert_nvme_controller_index = 0;
154 : static uint64_t g_nvme_hotplug_poll_period_us = NVME_HOTPLUG_POLL_PERIOD_DEFAULT;
155 : static bool g_nvme_hotplug_enabled = false;
156 : struct spdk_thread *g_bdev_nvme_init_thread;
157 : static struct spdk_poller *g_hotplug_poller;
158 : static struct spdk_poller *g_hotplug_probe_poller;
159 : static struct spdk_nvme_probe_ctx *g_hotplug_probe_ctx;
160 :
161 : static void nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
162 : struct nvme_async_probe_ctx *ctx);
163 : static void nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
164 : struct nvme_async_probe_ctx *ctx);
165 : static int bdev_nvme_library_init(void);
166 : static void bdev_nvme_library_fini(void);
167 : static void _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch,
168 : struct spdk_bdev_io *bdev_io);
169 : static void bdev_nvme_submit_request(struct spdk_io_channel *ch,
170 : struct spdk_bdev_io *bdev_io);
171 : static int bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
172 : void *md, uint64_t lba_count, uint64_t lba,
173 : uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
174 : struct spdk_accel_sequence *seq);
175 : static int bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
176 : void *md, uint64_t lba_count, uint64_t lba);
177 : static int bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
178 : void *md, uint64_t lba_count, uint64_t lba,
179 : uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
180 : struct spdk_accel_sequence *seq,
181 : union spdk_bdev_nvme_cdw12 cdw12, union spdk_bdev_nvme_cdw13 cdw13);
182 : static int bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
183 : void *md, uint64_t lba_count,
184 : uint64_t zslba, uint32_t flags);
185 : static int bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
186 : void *md, uint64_t lba_count, uint64_t lba,
187 : uint32_t flags);
188 : static int bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio,
189 : struct iovec *cmp_iov, int cmp_iovcnt, struct iovec *write_iov,
190 : int write_iovcnt, void *md, uint64_t lba_count, uint64_t lba,
191 : uint32_t flags);
192 : static int bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id,
193 : uint32_t num_zones, struct spdk_bdev_zone_info *info);
194 : static int bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
195 : enum spdk_bdev_zone_action action);
196 : static void bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch,
197 : struct nvme_bdev_io *bio,
198 : struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes);
199 : static int bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
200 : void *buf, size_t nbytes);
201 : static int bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
202 : void *buf, size_t nbytes, void *md_buf, size_t md_len);
203 : static int bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
204 : struct iovec *iov, int iovcnt, size_t nbytes,
205 : void *md_buf, size_t md_len);
206 : static void bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch,
207 : struct nvme_bdev_io *bio, struct nvme_bdev_io *bio_to_abort);
208 : static void bdev_nvme_reset_io(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio);
209 : static int bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
210 : static int bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
211 : static void remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr);
212 : static int nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr);
213 :
214 : static struct nvme_ns *nvme_ns_alloc(void);
215 : static void nvme_ns_free(struct nvme_ns *ns);
216 :
217 : static int
218 173 : nvme_ns_cmp(struct nvme_ns *ns1, struct nvme_ns *ns2)
219 : {
220 173 : return ns1->id < ns2->id ? -1 : ns1->id > ns2->id;
221 : }
222 :
223 902 : RB_GENERATE_STATIC(nvme_ns_tree, nvme_ns, node, nvme_ns_cmp);
224 :
225 : struct spdk_nvme_qpair *
226 1 : bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch)
227 : {
228 : struct nvme_ctrlr_channel *ctrlr_ch;
229 :
230 1 : assert(ctrlr_io_ch != NULL);
231 :
232 1 : ctrlr_ch = spdk_io_channel_get_ctx(ctrlr_io_ch);
233 :
234 1 : return ctrlr_ch->qpair->qpair;
235 : }
236 :
237 : static int
238 0 : bdev_nvme_get_ctx_size(void)
239 : {
240 0 : return sizeof(struct nvme_bdev_io);
241 : }
242 :
243 : static struct spdk_bdev_module nvme_if = {
244 : .name = "nvme",
245 : .async_fini = true,
246 : .module_init = bdev_nvme_library_init,
247 : .module_fini = bdev_nvme_library_fini,
248 : .config_json = bdev_nvme_config_json,
249 : .get_ctx_size = bdev_nvme_get_ctx_size,
250 :
251 : };
252 1 : SPDK_BDEV_MODULE_REGISTER(nvme, &nvme_if)
253 :
254 : struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs = TAILQ_HEAD_INITIALIZER(g_nvme_bdev_ctrlrs);
255 : pthread_mutex_t g_bdev_nvme_mutex = PTHREAD_MUTEX_INITIALIZER;
256 : bool g_bdev_nvme_module_finish;
257 :
258 : struct nvme_bdev_ctrlr *
259 270 : nvme_bdev_ctrlr_get_by_name(const char *name)
260 : {
261 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
262 :
263 270 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
264 148 : if (strcmp(name, nbdev_ctrlr->name) == 0) {
265 148 : break;
266 : }
267 : }
268 :
269 270 : return nbdev_ctrlr;
270 : }
271 :
272 : static struct nvme_ctrlr *
273 58 : nvme_bdev_ctrlr_get_ctrlr(struct nvme_bdev_ctrlr *nbdev_ctrlr,
274 : const struct spdk_nvme_transport_id *trid)
275 : {
276 : struct nvme_ctrlr *nvme_ctrlr;
277 :
278 99 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
279 74 : if (spdk_nvme_transport_id_compare(trid, &nvme_ctrlr->active_path_id->trid) == 0) {
280 33 : break;
281 : }
282 : }
283 :
284 58 : return nvme_ctrlr;
285 : }
286 :
287 : struct nvme_ctrlr *
288 0 : nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr,
289 : uint16_t cntlid)
290 : {
291 : struct nvme_ctrlr *nvme_ctrlr;
292 : const struct spdk_nvme_ctrlr_data *cdata;
293 :
294 0 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
295 0 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
296 0 : if (cdata->cntlid == cntlid) {
297 0 : break;
298 : }
299 : }
300 :
301 0 : return nvme_ctrlr;
302 : }
303 :
304 : static struct nvme_bdev *
305 72 : nvme_bdev_ctrlr_get_bdev(struct nvme_bdev_ctrlr *nbdev_ctrlr, uint32_t nsid)
306 : {
307 : struct nvme_bdev *bdev;
308 :
309 72 : pthread_mutex_lock(&g_bdev_nvme_mutex);
310 106 : TAILQ_FOREACH(bdev, &nbdev_ctrlr->bdevs, tailq) {
311 68 : if (bdev->nsid == nsid) {
312 34 : break;
313 : }
314 : }
315 72 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
316 :
317 72 : return bdev;
318 : }
319 :
320 : struct nvme_ns *
321 139 : nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid)
322 : {
323 139 : struct nvme_ns ns;
324 :
325 139 : assert(nsid > 0);
326 :
327 139 : ns.id = nsid;
328 139 : return RB_FIND(nvme_ns_tree, &nvme_ctrlr->namespaces, &ns);
329 : }
330 :
331 : struct nvme_ns *
332 152 : nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr)
333 : {
334 152 : return RB_MIN(nvme_ns_tree, &nvme_ctrlr->namespaces);
335 : }
336 :
337 : struct nvme_ns *
338 63 : nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns)
339 : {
340 63 : if (ns == NULL) {
341 0 : return NULL;
342 : }
343 :
344 63 : return RB_NEXT(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
345 : }
346 :
347 : static struct nvme_ctrlr *
348 51 : nvme_ctrlr_get(const struct spdk_nvme_transport_id *trid)
349 : {
350 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
351 51 : struct nvme_ctrlr *nvme_ctrlr = NULL;
352 :
353 51 : pthread_mutex_lock(&g_bdev_nvme_mutex);
354 70 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
355 19 : nvme_ctrlr = nvme_bdev_ctrlr_get_ctrlr(nbdev_ctrlr, trid);
356 19 : if (nvme_ctrlr != NULL) {
357 0 : break;
358 : }
359 : }
360 51 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
361 :
362 51 : return nvme_ctrlr;
363 : }
364 :
365 : struct nvme_ctrlr *
366 71 : nvme_ctrlr_get_by_name(const char *name)
367 : {
368 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
369 71 : struct nvme_ctrlr *nvme_ctrlr = NULL;
370 :
371 71 : if (name == NULL) {
372 0 : return NULL;
373 : }
374 :
375 71 : pthread_mutex_lock(&g_bdev_nvme_mutex);
376 71 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
377 71 : if (nbdev_ctrlr != NULL) {
378 40 : nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
379 : }
380 71 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
381 :
382 71 : return nvme_ctrlr;
383 : }
384 :
385 : void
386 0 : nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx)
387 : {
388 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
389 :
390 0 : pthread_mutex_lock(&g_bdev_nvme_mutex);
391 0 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
392 0 : fn(nbdev_ctrlr, ctx);
393 : }
394 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
395 0 : }
396 :
397 : void
398 0 : nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid, struct spdk_json_write_ctx *w)
399 : {
400 : const char *trtype_str;
401 : const char *adrfam_str;
402 :
403 0 : trtype_str = spdk_nvme_transport_id_trtype_str(trid->trtype);
404 0 : if (trtype_str) {
405 0 : spdk_json_write_named_string(w, "trtype", trtype_str);
406 : }
407 :
408 0 : adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
409 0 : if (adrfam_str) {
410 0 : spdk_json_write_named_string(w, "adrfam", adrfam_str);
411 : }
412 :
413 0 : if (trid->traddr[0] != '\0') {
414 0 : spdk_json_write_named_string(w, "traddr", trid->traddr);
415 : }
416 :
417 0 : if (trid->trsvcid[0] != '\0') {
418 0 : spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
419 : }
420 :
421 0 : if (trid->subnqn[0] != '\0') {
422 0 : spdk_json_write_named_string(w, "subnqn", trid->subnqn);
423 : }
424 0 : }
425 :
426 : static void
427 59 : nvme_bdev_ctrlr_delete(struct nvme_bdev_ctrlr *nbdev_ctrlr,
428 : struct nvme_ctrlr *nvme_ctrlr)
429 : {
430 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_delete, nvme_ctrlr->nbdev_ctrlr->name);
431 59 : pthread_mutex_lock(&g_bdev_nvme_mutex);
432 :
433 59 : TAILQ_REMOVE(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
434 59 : if (!TAILQ_EMPTY(&nbdev_ctrlr->ctrlrs)) {
435 15 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
436 :
437 15 : return;
438 : }
439 44 : TAILQ_REMOVE(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
440 :
441 44 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
442 :
443 44 : assert(TAILQ_EMPTY(&nbdev_ctrlr->bdevs));
444 :
445 44 : free(nbdev_ctrlr->name);
446 44 : free(nbdev_ctrlr);
447 : }
448 :
449 : static void
450 60 : _nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
451 : {
452 : struct nvme_path_id *path_id, *tmp_path;
453 : struct nvme_ns *ns, *tmp_ns;
454 :
455 60 : free(nvme_ctrlr->copied_ana_desc);
456 60 : spdk_free(nvme_ctrlr->ana_log_page);
457 :
458 60 : if (nvme_ctrlr->opal_dev) {
459 0 : spdk_opal_dev_destruct(nvme_ctrlr->opal_dev);
460 0 : nvme_ctrlr->opal_dev = NULL;
461 : }
462 :
463 60 : if (nvme_ctrlr->nbdev_ctrlr) {
464 59 : nvme_bdev_ctrlr_delete(nvme_ctrlr->nbdev_ctrlr, nvme_ctrlr);
465 : }
466 :
467 60 : RB_FOREACH_SAFE(ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp_ns) {
468 0 : RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
469 0 : nvme_ns_free(ns);
470 : }
471 :
472 120 : TAILQ_FOREACH_SAFE(path_id, &nvme_ctrlr->trids, link, tmp_path) {
473 60 : TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
474 60 : free(path_id);
475 : }
476 :
477 60 : pthread_mutex_destroy(&nvme_ctrlr->mutex);
478 60 : spdk_keyring_put_key(nvme_ctrlr->psk);
479 60 : spdk_keyring_put_key(nvme_ctrlr->dhchap_key);
480 60 : spdk_keyring_put_key(nvme_ctrlr->dhchap_ctrlr_key);
481 60 : free(nvme_ctrlr);
482 :
483 60 : pthread_mutex_lock(&g_bdev_nvme_mutex);
484 60 : if (g_bdev_nvme_module_finish && TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
485 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
486 0 : spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
487 0 : spdk_bdev_module_fini_done();
488 0 : return;
489 : }
490 60 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
491 : }
492 :
493 : static int
494 60 : nvme_detach_poller(void *arg)
495 : {
496 60 : struct nvme_ctrlr *nvme_ctrlr = arg;
497 : int rc;
498 :
499 60 : rc = spdk_nvme_detach_poll_async(nvme_ctrlr->detach_ctx);
500 60 : if (rc != -EAGAIN) {
501 60 : spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
502 60 : _nvme_ctrlr_delete(nvme_ctrlr);
503 : }
504 :
505 60 : return SPDK_POLLER_BUSY;
506 : }
507 :
508 : static void
509 60 : nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
510 : {
511 : int rc;
512 :
513 60 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
514 :
515 : /* First, unregister the adminq poller, as the driver will poll adminq if necessary */
516 60 : spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);
517 :
518 : /* If we got here, the reset/detach poller cannot be active */
519 60 : assert(nvme_ctrlr->reset_detach_poller == NULL);
520 60 : nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(nvme_detach_poller,
521 : nvme_ctrlr, 1000);
522 60 : if (nvme_ctrlr->reset_detach_poller == NULL) {
523 0 : SPDK_ERRLOG("Failed to register detach poller\n");
524 0 : goto error;
525 : }
526 :
527 60 : rc = spdk_nvme_detach_async(nvme_ctrlr->ctrlr, &nvme_ctrlr->detach_ctx);
528 60 : if (rc != 0) {
529 0 : SPDK_ERRLOG("Failed to detach the NVMe controller\n");
530 0 : goto error;
531 : }
532 :
533 60 : return;
534 0 : error:
535 : /* We don't have a good way to handle errors here, so just do what we can and delete the
536 : * controller without detaching the underlying NVMe device.
537 : */
538 0 : spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
539 0 : _nvme_ctrlr_delete(nvme_ctrlr);
540 : }
541 :
542 : static void
543 59 : nvme_ctrlr_unregister_cb(void *io_device)
544 : {
545 59 : struct nvme_ctrlr *nvme_ctrlr = io_device;
546 :
547 59 : nvme_ctrlr_delete(nvme_ctrlr);
548 59 : }
549 :
550 : static void
551 59 : nvme_ctrlr_unregister(void *ctx)
552 : {
553 59 : struct nvme_ctrlr *nvme_ctrlr = ctx;
554 :
555 59 : spdk_io_device_unregister(nvme_ctrlr, nvme_ctrlr_unregister_cb);
556 59 : }
557 :
558 : static bool
559 220 : nvme_ctrlr_can_be_unregistered(struct nvme_ctrlr *nvme_ctrlr)
560 : {
561 220 : if (!nvme_ctrlr->destruct) {
562 105 : return false;
563 : }
564 :
565 115 : if (nvme_ctrlr->ref > 0) {
566 56 : return false;
567 : }
568 :
569 59 : if (nvme_ctrlr->resetting) {
570 0 : return false;
571 : }
572 :
573 59 : if (nvme_ctrlr->ana_log_page_updating) {
574 0 : return false;
575 : }
576 :
577 59 : if (nvme_ctrlr->io_path_cache_clearing) {
578 0 : return false;
579 : }
580 :
581 59 : return true;
582 : }
583 :
584 : static void
585 164 : nvme_ctrlr_release(struct nvme_ctrlr *nvme_ctrlr)
586 : {
587 164 : pthread_mutex_lock(&nvme_ctrlr->mutex);
588 : SPDK_DTRACE_PROBE2(bdev_nvme_ctrlr_release, nvme_ctrlr->nbdev_ctrlr->name, nvme_ctrlr->ref);
589 :
590 164 : assert(nvme_ctrlr->ref > 0);
591 164 : nvme_ctrlr->ref--;
592 :
593 164 : if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
594 105 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
595 105 : return;
596 : }
597 :
598 59 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
599 :
600 59 : spdk_thread_exec_msg(nvme_ctrlr->thread, nvme_ctrlr_unregister, nvme_ctrlr);
601 : }
602 :
603 : static void
604 161 : bdev_nvme_clear_current_io_path(struct nvme_bdev_channel *nbdev_ch)
605 : {
606 161 : nbdev_ch->current_io_path = NULL;
607 161 : nbdev_ch->rr_counter = 0;
608 161 : }
609 :
610 : static struct nvme_io_path *
611 8 : _bdev_nvme_get_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
612 : {
613 : struct nvme_io_path *io_path;
614 :
615 16 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
616 15 : if (io_path->nvme_ns == nvme_ns) {
617 7 : break;
618 : }
619 : }
620 :
621 8 : return io_path;
622 : }
623 :
624 : static struct nvme_io_path *
625 35 : nvme_io_path_alloc(void)
626 : {
627 : struct nvme_io_path *io_path;
628 :
629 35 : io_path = calloc(1, sizeof(*io_path));
630 35 : if (io_path == NULL) {
631 0 : SPDK_ERRLOG("Failed to alloc io_path.\n");
632 0 : return NULL;
633 : }
634 :
635 35 : if (g_opts.io_path_stat) {
636 0 : io_path->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
637 0 : if (io_path->stat == NULL) {
638 0 : free(io_path);
639 0 : SPDK_ERRLOG("Failed to alloc io_path stat.\n");
640 0 : return NULL;
641 : }
642 0 : spdk_bdev_reset_io_stat(io_path->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
643 : }
644 :
645 35 : return io_path;
646 : }
647 :
648 : static void
649 35 : nvme_io_path_free(struct nvme_io_path *io_path)
650 : {
651 35 : free(io_path->stat);
652 35 : free(io_path);
653 35 : }
654 :
655 : static int
656 35 : _bdev_nvme_add_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
657 : {
658 : struct nvme_io_path *io_path;
659 : struct spdk_io_channel *ch;
660 : struct nvme_ctrlr_channel *ctrlr_ch;
661 : struct nvme_qpair *nvme_qpair;
662 :
663 35 : io_path = nvme_io_path_alloc();
664 35 : if (io_path == NULL) {
665 0 : return -ENOMEM;
666 : }
667 :
668 35 : io_path->nvme_ns = nvme_ns;
669 :
670 35 : ch = spdk_get_io_channel(nvme_ns->ctrlr);
671 35 : if (ch == NULL) {
672 0 : nvme_io_path_free(io_path);
673 0 : SPDK_ERRLOG("Failed to alloc io_channel.\n");
674 0 : return -ENOMEM;
675 : }
676 :
677 35 : ctrlr_ch = spdk_io_channel_get_ctx(ch);
678 :
679 35 : nvme_qpair = ctrlr_ch->qpair;
680 35 : assert(nvme_qpair != NULL);
681 :
682 35 : io_path->qpair = nvme_qpair;
683 35 : TAILQ_INSERT_TAIL(&nvme_qpair->io_path_list, io_path, tailq);
684 :
685 35 : io_path->nbdev_ch = nbdev_ch;
686 35 : STAILQ_INSERT_TAIL(&nbdev_ch->io_path_list, io_path, stailq);
687 :
688 35 : bdev_nvme_clear_current_io_path(nbdev_ch);
689 :
690 35 : return 0;
691 : }
692 :
693 : static void
694 35 : bdev_nvme_clear_retry_io_path(struct nvme_bdev_channel *nbdev_ch,
695 : struct nvme_io_path *io_path)
696 : {
697 : struct spdk_bdev_io *bdev_io;
698 : struct nvme_bdev_io *bio;
699 :
700 36 : TAILQ_FOREACH(bdev_io, &nbdev_ch->retry_io_list, module_link) {
701 1 : bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
702 1 : if (bio->io_path == io_path) {
703 1 : bio->io_path = NULL;
704 : }
705 : }
706 35 : }
707 :
708 : static void
709 35 : _bdev_nvme_delete_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *io_path)
710 : {
711 : struct spdk_io_channel *ch;
712 : struct nvme_qpair *nvme_qpair;
713 : struct nvme_ctrlr_channel *ctrlr_ch;
714 : struct nvme_bdev *nbdev;
715 :
716 35 : nbdev = spdk_io_channel_get_io_device(spdk_io_channel_from_ctx(nbdev_ch));
717 :
718 : /* Add the statistics to nvme_ns before this path is destroyed. */
719 35 : pthread_mutex_lock(&nbdev->mutex);
720 35 : if (nbdev->ref != 0 && io_path->nvme_ns->stat != NULL && io_path->stat != NULL) {
721 0 : spdk_bdev_add_io_stat(io_path->nvme_ns->stat, io_path->stat);
722 : }
723 35 : pthread_mutex_unlock(&nbdev->mutex);
724 :
725 35 : bdev_nvme_clear_current_io_path(nbdev_ch);
726 35 : bdev_nvme_clear_retry_io_path(nbdev_ch, io_path);
727 :
728 35 : STAILQ_REMOVE(&nbdev_ch->io_path_list, io_path, nvme_io_path, stailq);
729 35 : io_path->nbdev_ch = NULL;
730 :
731 35 : nvme_qpair = io_path->qpair;
732 35 : assert(nvme_qpair != NULL);
733 :
734 35 : ctrlr_ch = nvme_qpair->ctrlr_ch;
735 35 : assert(ctrlr_ch != NULL);
736 :
737 35 : ch = spdk_io_channel_from_ctx(ctrlr_ch);
738 35 : spdk_put_io_channel(ch);
739 :
740 : /* After an io_path is removed, I/Os submitted to it may complete and update statistics
741 : * of the io_path. To avoid heap-use-after-free error from this case, do not free the
742 : * io_path here but free the io_path when the associated qpair is freed. It is ensured
743 : * that all I/Os submitted to the io_path are completed when the associated qpair is freed.
744 : */
745 35 : }
746 :
747 : static void
748 22 : _bdev_nvme_delete_io_paths(struct nvme_bdev_channel *nbdev_ch)
749 : {
750 : struct nvme_io_path *io_path, *tmp_io_path;
751 :
752 55 : STAILQ_FOREACH_SAFE(io_path, &nbdev_ch->io_path_list, stailq, tmp_io_path) {
753 33 : _bdev_nvme_delete_io_path(nbdev_ch, io_path);
754 : }
755 22 : }
756 :
757 : static int
758 22 : bdev_nvme_create_bdev_channel_cb(void *io_device, void *ctx_buf)
759 : {
760 22 : struct nvme_bdev_channel *nbdev_ch = ctx_buf;
761 22 : struct nvme_bdev *nbdev = io_device;
762 : struct nvme_ns *nvme_ns;
763 : int rc;
764 :
765 22 : STAILQ_INIT(&nbdev_ch->io_path_list);
766 22 : TAILQ_INIT(&nbdev_ch->retry_io_list);
767 :
768 22 : pthread_mutex_lock(&nbdev->mutex);
769 :
770 22 : nbdev_ch->mp_policy = nbdev->mp_policy;
771 22 : nbdev_ch->mp_selector = nbdev->mp_selector;
772 22 : nbdev_ch->rr_min_io = nbdev->rr_min_io;
773 :
774 55 : TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
775 33 : rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
776 33 : if (rc != 0) {
777 0 : pthread_mutex_unlock(&nbdev->mutex);
778 :
779 0 : _bdev_nvme_delete_io_paths(nbdev_ch);
780 0 : return rc;
781 : }
782 : }
783 22 : pthread_mutex_unlock(&nbdev->mutex);
784 :
785 22 : return 0;
786 : }
787 :
788 : /* If cpl != NULL, complete the bdev_io with nvme status based on 'cpl'.
789 : * If cpl == NULL, complete the bdev_io with bdev status based on 'status'.
790 : */
791 : static inline void
792 47 : __bdev_nvme_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status,
793 : const struct spdk_nvme_cpl *cpl)
794 : {
795 47 : spdk_trace_record(TRACE_BDEV_NVME_IO_DONE, 0, 0, (uintptr_t)bdev_io->driver_ctx,
796 : (uintptr_t)bdev_io);
797 47 : if (cpl) {
798 29 : spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc);
799 : } else {
800 18 : spdk_bdev_io_complete(bdev_io, status);
801 : }
802 47 : }
803 :
804 : static void bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch);
805 :
806 : static void
807 22 : bdev_nvme_destroy_bdev_channel_cb(void *io_device, void *ctx_buf)
808 : {
809 22 : struct nvme_bdev_channel *nbdev_ch = ctx_buf;
810 :
811 22 : bdev_nvme_abort_retry_ios(nbdev_ch);
812 22 : _bdev_nvme_delete_io_paths(nbdev_ch);
813 22 : }
814 :
815 : static inline bool
816 58 : bdev_nvme_io_type_is_admin(enum spdk_bdev_io_type io_type)
817 : {
818 58 : switch (io_type) {
819 5 : case SPDK_BDEV_IO_TYPE_RESET:
820 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
821 : case SPDK_BDEV_IO_TYPE_ABORT:
822 5 : return true;
823 53 : default:
824 53 : break;
825 : }
826 :
827 53 : return false;
828 : }
829 :
830 : static inline bool
831 77 : nvme_ns_is_active(struct nvme_ns *nvme_ns)
832 : {
833 77 : if (spdk_unlikely(nvme_ns->ana_state_updating)) {
834 1 : return false;
835 : }
836 :
837 76 : if (spdk_unlikely(nvme_ns->ns == NULL)) {
838 0 : return false;
839 : }
840 :
841 76 : return true;
842 : }
843 :
844 : static inline bool
845 65 : nvme_ns_is_accessible(struct nvme_ns *nvme_ns)
846 : {
847 65 : if (spdk_unlikely(!nvme_ns_is_active(nvme_ns))) {
848 1 : return false;
849 : }
850 :
851 64 : switch (nvme_ns->ana_state) {
852 57 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
853 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
854 57 : return true;
855 7 : default:
856 7 : break;
857 : }
858 :
859 7 : return false;
860 : }
861 :
862 : static inline bool
863 102 : nvme_qpair_is_connected(struct nvme_qpair *nvme_qpair)
864 : {
865 102 : if (spdk_unlikely(nvme_qpair->qpair == NULL)) {
866 20 : return false;
867 : }
868 :
869 82 : if (spdk_unlikely(spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
870 : SPDK_NVME_QPAIR_FAILURE_NONE)) {
871 0 : return false;
872 : }
873 :
874 82 : if (spdk_unlikely(nvme_qpair->ctrlr_ch->reset_iter != NULL)) {
875 0 : return false;
876 : }
877 :
878 82 : return true;
879 : }
880 :
881 : static inline bool
882 77 : nvme_io_path_is_available(struct nvme_io_path *io_path)
883 : {
884 77 : if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
885 12 : return false;
886 : }
887 :
888 65 : if (spdk_unlikely(!nvme_ns_is_accessible(io_path->nvme_ns))) {
889 8 : return false;
890 : }
891 :
892 57 : return true;
893 : }
894 :
895 : static inline bool
896 8 : nvme_ctrlr_is_failed(struct nvme_ctrlr *nvme_ctrlr)
897 : {
898 8 : if (nvme_ctrlr->destruct) {
899 0 : return true;
900 : }
901 :
902 8 : if (nvme_ctrlr->fast_io_fail_timedout) {
903 2 : return true;
904 : }
905 :
906 6 : if (nvme_ctrlr->resetting) {
907 4 : if (nvme_ctrlr->opts.reconnect_delay_sec != 0) {
908 4 : return false;
909 : } else {
910 0 : return true;
911 : }
912 : }
913 :
914 2 : if (nvme_ctrlr->reconnect_is_delayed) {
915 2 : return false;
916 : }
917 :
918 0 : if (nvme_ctrlr->disabled) {
919 0 : return true;
920 : }
921 :
922 0 : if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
923 0 : return true;
924 : } else {
925 0 : return false;
926 : }
927 : }
928 :
929 : static bool
930 20 : nvme_ctrlr_is_available(struct nvme_ctrlr *nvme_ctrlr)
931 : {
932 20 : if (nvme_ctrlr->destruct) {
933 0 : return false;
934 : }
935 :
936 20 : if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
937 3 : return false;
938 : }
939 :
940 17 : if (nvme_ctrlr->resetting || nvme_ctrlr->reconnect_is_delayed) {
941 1 : return false;
942 : }
943 :
944 16 : if (nvme_ctrlr->disabled) {
945 0 : return false;
946 : }
947 :
948 16 : return true;
949 : }
950 :
951 : /* Simulate circular linked list. */
952 : static inline struct nvme_io_path *
953 87 : nvme_io_path_get_next(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *prev_path)
954 : {
955 : struct nvme_io_path *next_path;
956 :
957 87 : if (prev_path != NULL) {
958 37 : next_path = STAILQ_NEXT(prev_path, stailq);
959 37 : if (next_path != NULL) {
960 14 : return next_path;
961 : }
962 : }
963 :
964 73 : return STAILQ_FIRST(&nbdev_ch->io_path_list);
965 : }
966 :
967 : static struct nvme_io_path *
968 57 : _bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
969 : {
970 57 : struct nvme_io_path *io_path, *start, *non_optimized = NULL;
971 :
972 57 : start = nvme_io_path_get_next(nbdev_ch, nbdev_ch->current_io_path);
973 :
974 57 : io_path = start;
975 : do {
976 69 : if (spdk_likely(nvme_io_path_is_available(io_path))) {
977 49 : switch (io_path->nvme_ns->ana_state) {
978 39 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
979 39 : nbdev_ch->current_io_path = io_path;
980 39 : return io_path;
981 10 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
982 10 : if (non_optimized == NULL) {
983 7 : non_optimized = io_path;
984 : }
985 10 : break;
986 0 : default:
987 0 : assert(false);
988 : break;
989 : }
990 20 : }
991 30 : io_path = nvme_io_path_get_next(nbdev_ch, io_path);
992 30 : } while (io_path != start);
993 :
994 18 : if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
995 : /* We come here only if there is no optimized path. Cache even non_optimized
996 : * path for load balance across multiple non_optimized paths.
997 : */
998 1 : nbdev_ch->current_io_path = non_optimized;
999 : }
1000 :
1001 18 : return non_optimized;
1002 : }
1003 :
1004 : static struct nvme_io_path *
1005 4 : _bdev_nvme_find_io_path_min_qd(struct nvme_bdev_channel *nbdev_ch)
1006 : {
1007 : struct nvme_io_path *io_path;
1008 4 : struct nvme_io_path *optimized = NULL, *non_optimized = NULL;
1009 4 : uint32_t opt_min_qd = UINT32_MAX, non_opt_min_qd = UINT32_MAX;
1010 : uint32_t num_outstanding_reqs;
1011 :
1012 16 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
1013 12 : if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
1014 : /* The device is currently resetting. */
1015 0 : continue;
1016 : }
1017 :
1018 12 : if (spdk_unlikely(!nvme_ns_is_active(io_path->nvme_ns))) {
1019 0 : continue;
1020 : }
1021 :
1022 12 : num_outstanding_reqs = spdk_nvme_qpair_get_num_outstanding_reqs(io_path->qpair->qpair);
1023 12 : switch (io_path->nvme_ns->ana_state) {
1024 6 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
1025 6 : if (num_outstanding_reqs < opt_min_qd) {
1026 5 : opt_min_qd = num_outstanding_reqs;
1027 5 : optimized = io_path;
1028 : }
1029 6 : break;
1030 3 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1031 3 : if (num_outstanding_reqs < non_opt_min_qd) {
1032 3 : non_opt_min_qd = num_outstanding_reqs;
1033 3 : non_optimized = io_path;
1034 : }
1035 3 : break;
1036 3 : default:
1037 3 : break;
1038 : }
1039 : }
1040 :
1041 : /* don't cache io path for BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH selector */
1042 4 : if (optimized != NULL) {
1043 3 : return optimized;
1044 : }
1045 :
1046 1 : return non_optimized;
1047 : }
1048 :
1049 : static inline struct nvme_io_path *
1050 95 : bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
1051 : {
1052 95 : if (spdk_likely(nbdev_ch->current_io_path != NULL)) {
1053 41 : if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE) {
1054 31 : return nbdev_ch->current_io_path;
1055 10 : } else if (nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
1056 10 : if (++nbdev_ch->rr_counter < nbdev_ch->rr_min_io) {
1057 3 : return nbdev_ch->current_io_path;
1058 : }
1059 7 : nbdev_ch->rr_counter = 0;
1060 : }
1061 : }
1062 :
1063 61 : if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE ||
1064 14 : nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
1065 57 : return _bdev_nvme_find_io_path(nbdev_ch);
1066 : } else {
1067 4 : return _bdev_nvme_find_io_path_min_qd(nbdev_ch);
1068 : }
1069 : }
1070 :
1071 : /* Return true if there is any io_path whose qpair is active or ctrlr is not failed,
1072 : * or false otherwise.
1073 : *
1074 : * If any io_path has an active qpair but find_io_path() returned NULL, its namespace
1075 : * is likely to be non-accessible now but may become accessible.
1076 : *
1077 : * If any io_path has an unfailed ctrlr but find_io_path() returned NULL, the ctrlr
1078 : * is likely to be resetting now but the reset may succeed. A ctrlr is set to unfailed
1079 : * when starting to reset it but it is set to failed when the reset failed. Hence, if
1080 : * a ctrlr is unfailed, it is likely that it works fine or is resetting.
1081 : */
1082 : static bool
1083 13 : any_io_path_may_become_available(struct nvme_bdev_channel *nbdev_ch)
1084 : {
1085 : struct nvme_io_path *io_path;
1086 :
1087 15 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
1088 13 : if (io_path->nvme_ns->ana_transition_timedout) {
1089 0 : continue;
1090 : }
1091 :
1092 13 : if (nvme_qpair_is_connected(io_path->qpair) ||
1093 8 : !nvme_ctrlr_is_failed(io_path->qpair->ctrlr)) {
1094 11 : return true;
1095 : }
1096 : }
1097 :
1098 2 : return false;
1099 : }
1100 :
1101 : static void
1102 14 : bdev_nvme_retry_io(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
1103 : {
1104 14 : struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
1105 : struct spdk_io_channel *ch;
1106 :
1107 14 : if (nbdev_io->io_path != NULL && nvme_io_path_is_available(nbdev_io->io_path)) {
1108 3 : _bdev_nvme_submit_request(nbdev_ch, bdev_io);
1109 : } else {
1110 11 : ch = spdk_io_channel_from_ctx(nbdev_ch);
1111 11 : bdev_nvme_submit_request(ch, bdev_io);
1112 : }
1113 14 : }
1114 :
1115 : static int
1116 14 : bdev_nvme_retry_ios(void *arg)
1117 : {
1118 14 : struct nvme_bdev_channel *nbdev_ch = arg;
1119 : struct spdk_bdev_io *bdev_io, *tmp_bdev_io;
1120 : struct nvme_bdev_io *bio;
1121 : uint64_t now, delay_us;
1122 :
1123 14 : now = spdk_get_ticks();
1124 :
1125 28 : TAILQ_FOREACH_SAFE(bdev_io, &nbdev_ch->retry_io_list, module_link, tmp_bdev_io) {
1126 15 : bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
1127 15 : if (bio->retry_ticks > now) {
1128 1 : break;
1129 : }
1130 :
1131 14 : TAILQ_REMOVE(&nbdev_ch->retry_io_list, bdev_io, module_link);
1132 :
1133 14 : bdev_nvme_retry_io(nbdev_ch, bdev_io);
1134 : }
1135 :
1136 14 : spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1137 :
1138 14 : bdev_io = TAILQ_FIRST(&nbdev_ch->retry_io_list);
1139 14 : if (bdev_io != NULL) {
1140 4 : bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
1141 :
1142 4 : delay_us = (bio->retry_ticks - now) * SPDK_SEC_TO_USEC / spdk_get_ticks_hz();
1143 :
1144 4 : nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
1145 : delay_us);
1146 : }
1147 :
1148 14 : return SPDK_POLLER_BUSY;
1149 : }
1150 :
1151 : static void
1152 15 : bdev_nvme_queue_retry_io(struct nvme_bdev_channel *nbdev_ch,
1153 : struct nvme_bdev_io *bio, uint64_t delay_ms)
1154 : {
1155 15 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1156 : struct spdk_bdev_io *tmp_bdev_io;
1157 : struct nvme_bdev_io *tmp_bio;
1158 :
1159 15 : bio->retry_ticks = spdk_get_ticks() + delay_ms * spdk_get_ticks_hz() / 1000ULL;
1160 :
1161 15 : TAILQ_FOREACH_REVERSE(tmp_bdev_io, &nbdev_ch->retry_io_list, retry_io_head, module_link) {
1162 1 : tmp_bio = (struct nvme_bdev_io *)tmp_bdev_io->driver_ctx;
1163 :
1164 1 : if (tmp_bio->retry_ticks <= bio->retry_ticks) {
1165 1 : TAILQ_INSERT_AFTER(&nbdev_ch->retry_io_list, tmp_bdev_io, bdev_io,
1166 : module_link);
1167 1 : return;
1168 : }
1169 : }
1170 :
1171 : /* No earlier I/Os were found. This I/O must be the new head. */
1172 14 : TAILQ_INSERT_HEAD(&nbdev_ch->retry_io_list, bdev_io, module_link);
1173 :
1174 14 : spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1175 :
1176 14 : nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
1177 : delay_ms * 1000ULL);
1178 : }
1179 :
1180 : static void
1181 36 : bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch)
1182 : {
1183 : struct spdk_bdev_io *bdev_io, *tmp_io;
1184 :
1185 36 : TAILQ_FOREACH_SAFE(bdev_io, &nbdev_ch->retry_io_list, module_link, tmp_io) {
1186 0 : TAILQ_REMOVE(&nbdev_ch->retry_io_list, bdev_io, module_link);
1187 0 : __bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED, NULL);
1188 : }
1189 :
1190 36 : spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1191 36 : }
1192 :
1193 : static int
1194 6 : bdev_nvme_abort_retry_io(struct nvme_bdev_channel *nbdev_ch,
1195 : struct nvme_bdev_io *bio_to_abort)
1196 : {
1197 : struct spdk_bdev_io *bdev_io_to_abort;
1198 :
1199 6 : TAILQ_FOREACH(bdev_io_to_abort, &nbdev_ch->retry_io_list, module_link) {
1200 1 : if ((struct nvme_bdev_io *)bdev_io_to_abort->driver_ctx == bio_to_abort) {
1201 1 : TAILQ_REMOVE(&nbdev_ch->retry_io_list, bdev_io_to_abort, module_link);
1202 1 : __bdev_nvme_io_complete(bdev_io_to_abort, SPDK_BDEV_IO_STATUS_ABORTED, NULL);
1203 1 : return 0;
1204 : }
1205 : }
1206 :
1207 5 : return -ENOENT;
1208 : }
1209 :
1210 : static void
1211 12 : bdev_nvme_update_nvme_error_stat(struct spdk_bdev_io *bdev_io, const struct spdk_nvme_cpl *cpl)
1212 : {
1213 : struct nvme_bdev *nbdev;
1214 : uint16_t sct, sc;
1215 :
1216 12 : assert(spdk_nvme_cpl_is_error(cpl));
1217 :
1218 12 : nbdev = bdev_io->bdev->ctxt;
1219 :
1220 12 : if (nbdev->err_stat == NULL) {
1221 12 : return;
1222 : }
1223 :
1224 0 : sct = cpl->status.sct;
1225 0 : sc = cpl->status.sc;
1226 :
1227 0 : pthread_mutex_lock(&nbdev->mutex);
1228 :
1229 0 : nbdev->err_stat->status_type[sct]++;
1230 0 : switch (sct) {
1231 0 : case SPDK_NVME_SCT_GENERIC:
1232 : case SPDK_NVME_SCT_COMMAND_SPECIFIC:
1233 : case SPDK_NVME_SCT_MEDIA_ERROR:
1234 : case SPDK_NVME_SCT_PATH:
1235 0 : nbdev->err_stat->status[sct][sc]++;
1236 0 : break;
1237 0 : default:
1238 0 : break;
1239 : }
1240 :
1241 0 : pthread_mutex_unlock(&nbdev->mutex);
1242 : }
1243 :
1244 : static inline void
1245 20 : bdev_nvme_update_io_path_stat(struct nvme_bdev_io *bio)
1246 : {
1247 20 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1248 20 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
1249 20 : uint32_t blocklen = bdev_io->bdev->blocklen;
1250 : struct spdk_bdev_io_stat *stat;
1251 : uint64_t tsc_diff;
1252 :
1253 20 : if (bio->io_path->stat == NULL) {
1254 20 : return;
1255 : }
1256 :
1257 0 : tsc_diff = spdk_get_ticks() - bio->submit_tsc;
1258 0 : stat = bio->io_path->stat;
1259 :
1260 0 : switch (bdev_io->type) {
1261 0 : case SPDK_BDEV_IO_TYPE_READ:
1262 0 : stat->bytes_read += num_blocks * blocklen;
1263 0 : stat->num_read_ops++;
1264 0 : stat->read_latency_ticks += tsc_diff;
1265 0 : if (stat->max_read_latency_ticks < tsc_diff) {
1266 0 : stat->max_read_latency_ticks = tsc_diff;
1267 : }
1268 0 : if (stat->min_read_latency_ticks > tsc_diff) {
1269 0 : stat->min_read_latency_ticks = tsc_diff;
1270 : }
1271 0 : break;
1272 0 : case SPDK_BDEV_IO_TYPE_WRITE:
1273 0 : stat->bytes_written += num_blocks * blocklen;
1274 0 : stat->num_write_ops++;
1275 0 : stat->write_latency_ticks += tsc_diff;
1276 0 : if (stat->max_write_latency_ticks < tsc_diff) {
1277 0 : stat->max_write_latency_ticks = tsc_diff;
1278 : }
1279 0 : if (stat->min_write_latency_ticks > tsc_diff) {
1280 0 : stat->min_write_latency_ticks = tsc_diff;
1281 : }
1282 0 : break;
1283 0 : case SPDK_BDEV_IO_TYPE_UNMAP:
1284 0 : stat->bytes_unmapped += num_blocks * blocklen;
1285 0 : stat->num_unmap_ops++;
1286 0 : stat->unmap_latency_ticks += tsc_diff;
1287 0 : if (stat->max_unmap_latency_ticks < tsc_diff) {
1288 0 : stat->max_unmap_latency_ticks = tsc_diff;
1289 : }
1290 0 : if (stat->min_unmap_latency_ticks > tsc_diff) {
1291 0 : stat->min_unmap_latency_ticks = tsc_diff;
1292 : }
1293 0 : break;
1294 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
1295 : /* Track the data in the start phase only */
1296 0 : if (!bdev_io->u.bdev.zcopy.start) {
1297 0 : break;
1298 : }
1299 0 : if (bdev_io->u.bdev.zcopy.populate) {
1300 0 : stat->bytes_read += num_blocks * blocklen;
1301 0 : stat->num_read_ops++;
1302 0 : stat->read_latency_ticks += tsc_diff;
1303 0 : if (stat->max_read_latency_ticks < tsc_diff) {
1304 0 : stat->max_read_latency_ticks = tsc_diff;
1305 : }
1306 0 : if (stat->min_read_latency_ticks > tsc_diff) {
1307 0 : stat->min_read_latency_ticks = tsc_diff;
1308 : }
1309 : } else {
1310 0 : stat->bytes_written += num_blocks * blocklen;
1311 0 : stat->num_write_ops++;
1312 0 : stat->write_latency_ticks += tsc_diff;
1313 0 : if (stat->max_write_latency_ticks < tsc_diff) {
1314 0 : stat->max_write_latency_ticks = tsc_diff;
1315 : }
1316 0 : if (stat->min_write_latency_ticks > tsc_diff) {
1317 0 : stat->min_write_latency_ticks = tsc_diff;
1318 : }
1319 : }
1320 0 : break;
1321 0 : case SPDK_BDEV_IO_TYPE_COPY:
1322 0 : stat->bytes_copied += num_blocks * blocklen;
1323 0 : stat->num_copy_ops++;
1324 0 : stat->copy_latency_ticks += tsc_diff;
1325 0 : if (stat->max_copy_latency_ticks < tsc_diff) {
1326 0 : stat->max_copy_latency_ticks = tsc_diff;
1327 : }
1328 0 : if (stat->min_copy_latency_ticks > tsc_diff) {
1329 0 : stat->min_copy_latency_ticks = tsc_diff;
1330 : }
1331 0 : break;
1332 0 : default:
1333 0 : break;
1334 : }
1335 : }
1336 :
1337 : static bool
1338 7 : bdev_nvme_check_retry_io(struct nvme_bdev_io *bio,
1339 : const struct spdk_nvme_cpl *cpl,
1340 : struct nvme_bdev_channel *nbdev_ch,
1341 : uint64_t *_delay_ms)
1342 : {
1343 7 : struct nvme_io_path *io_path = bio->io_path;
1344 7 : struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
1345 : const struct spdk_nvme_ctrlr_data *cdata;
1346 :
1347 7 : if (spdk_nvme_cpl_is_path_error(cpl) ||
1348 5 : spdk_nvme_cpl_is_aborted_sq_deletion(cpl) ||
1349 4 : !nvme_io_path_is_available(io_path) ||
1350 4 : !nvme_ctrlr_is_available(nvme_ctrlr)) {
1351 3 : bdev_nvme_clear_current_io_path(nbdev_ch);
1352 3 : bio->io_path = NULL;
1353 3 : if (spdk_nvme_cpl_is_ana_error(cpl)) {
1354 1 : if (nvme_ctrlr_read_ana_log_page(nvme_ctrlr) == 0) {
1355 1 : io_path->nvme_ns->ana_state_updating = true;
1356 : }
1357 : }
1358 3 : if (!any_io_path_may_become_available(nbdev_ch)) {
1359 0 : return false;
1360 : }
1361 3 : *_delay_ms = 0;
1362 : } else {
1363 4 : bio->retry_count++;
1364 :
1365 4 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
1366 :
1367 4 : if (cpl->status.crd != 0) {
1368 1 : *_delay_ms = cdata->crdt[cpl->status.crd] * 100;
1369 : } else {
1370 3 : *_delay_ms = 0;
1371 : }
1372 : }
1373 :
1374 7 : return true;
1375 : }
1376 :
1377 : static inline void
1378 32 : bdev_nvme_io_complete_nvme_status(struct nvme_bdev_io *bio,
1379 : const struct spdk_nvme_cpl *cpl)
1380 : {
1381 32 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1382 : struct nvme_bdev_channel *nbdev_ch;
1383 32 : uint64_t delay_ms;
1384 :
1385 32 : assert(!bdev_nvme_io_type_is_admin(bdev_io->type));
1386 :
1387 32 : if (spdk_likely(spdk_nvme_cpl_is_success(cpl))) {
1388 20 : bdev_nvme_update_io_path_stat(bio);
1389 20 : goto complete;
1390 : }
1391 :
1392 : /* Update error counts before deciding if retry is needed.
1393 : * Hence, error counts may be more than the number of I/O errors.
1394 : */
1395 12 : bdev_nvme_update_nvme_error_stat(bdev_io, cpl);
1396 :
1397 12 : if (cpl->status.dnr != 0 || spdk_nvme_cpl_is_aborted_by_request(cpl) ||
1398 8 : (g_opts.bdev_retry_count != -1 && bio->retry_count >= g_opts.bdev_retry_count)) {
1399 5 : goto complete;
1400 : }
1401 :
1402 : /* At this point we don't know whether the sequence was successfully executed or not, so we
1403 : * cannot retry the IO */
1404 7 : if (bdev_io->u.bdev.accel_sequence != NULL) {
1405 0 : goto complete;
1406 : }
1407 :
1408 7 : nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
1409 :
1410 7 : if (bdev_nvme_check_retry_io(bio, cpl, nbdev_ch, &delay_ms)) {
1411 7 : bdev_nvme_queue_retry_io(nbdev_ch, bio, delay_ms);
1412 7 : return;
1413 : }
1414 :
1415 25 : complete:
1416 25 : bio->retry_count = 0;
1417 25 : bio->submit_tsc = 0;
1418 25 : bdev_io->u.bdev.accel_sequence = NULL;
1419 25 : __bdev_nvme_io_complete(bdev_io, 0, cpl);
1420 : }
1421 :
1422 : static inline void
1423 11 : bdev_nvme_io_complete(struct nvme_bdev_io *bio, int rc)
1424 : {
1425 11 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1426 : struct nvme_bdev_channel *nbdev_ch;
1427 : enum spdk_bdev_io_status io_status;
1428 :
1429 11 : assert(!bdev_nvme_io_type_is_admin(bdev_io->type));
1430 :
1431 11 : switch (rc) {
1432 1 : case 0:
1433 1 : io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1434 1 : break;
1435 0 : case -ENOMEM:
1436 0 : io_status = SPDK_BDEV_IO_STATUS_NOMEM;
1437 0 : break;
1438 10 : case -ENXIO:
1439 10 : if (g_opts.bdev_retry_count == -1 || bio->retry_count < g_opts.bdev_retry_count) {
1440 10 : nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
1441 :
1442 10 : bdev_nvme_clear_current_io_path(nbdev_ch);
1443 10 : bio->io_path = NULL;
1444 :
1445 10 : if (any_io_path_may_become_available(nbdev_ch)) {
1446 8 : bdev_nvme_queue_retry_io(nbdev_ch, bio, 1000ULL);
1447 8 : return;
1448 : }
1449 : }
1450 :
1451 : /* fallthrough */
1452 : default:
1453 2 : spdk_accel_sequence_abort(bdev_io->u.bdev.accel_sequence);
1454 2 : bdev_io->u.bdev.accel_sequence = NULL;
1455 2 : io_status = SPDK_BDEV_IO_STATUS_FAILED;
1456 2 : break;
1457 : }
1458 :
1459 3 : bio->retry_count = 0;
1460 3 : bio->submit_tsc = 0;
1461 3 : __bdev_nvme_io_complete(bdev_io, io_status, NULL);
1462 : }
1463 :
1464 : static inline void
1465 4 : bdev_nvme_admin_complete(struct nvme_bdev_io *bio, int rc)
1466 : {
1467 4 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1468 : enum spdk_bdev_io_status io_status;
1469 :
1470 4 : switch (rc) {
1471 1 : case 0:
1472 1 : io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1473 1 : break;
1474 0 : case -ENOMEM:
1475 0 : io_status = SPDK_BDEV_IO_STATUS_NOMEM;
1476 0 : break;
1477 3 : case -ENXIO:
1478 : /* fallthrough */
1479 : default:
1480 3 : io_status = SPDK_BDEV_IO_STATUS_FAILED;
1481 3 : break;
1482 : }
1483 :
1484 4 : __bdev_nvme_io_complete(bdev_io, io_status, NULL);
1485 4 : }
1486 :
1487 : static void
1488 3 : bdev_nvme_clear_io_path_caches_done(struct spdk_io_channel_iter *i, int status)
1489 : {
1490 3 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
1491 :
1492 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
1493 :
1494 3 : assert(nvme_ctrlr->io_path_cache_clearing == true);
1495 3 : nvme_ctrlr->io_path_cache_clearing = false;
1496 :
1497 3 : if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
1498 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1499 3 : return;
1500 : }
1501 :
1502 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1503 :
1504 0 : nvme_ctrlr_unregister(nvme_ctrlr);
1505 : }
1506 :
1507 : static void
1508 320 : _bdev_nvme_clear_io_path_cache(struct nvme_qpair *nvme_qpair)
1509 : {
1510 : struct nvme_io_path *io_path;
1511 :
1512 459 : TAILQ_FOREACH(io_path, &nvme_qpair->io_path_list, tailq) {
1513 139 : if (io_path->nbdev_ch == NULL) {
1514 64 : continue;
1515 : }
1516 75 : bdev_nvme_clear_current_io_path(io_path->nbdev_ch);
1517 : }
1518 320 : }
1519 :
1520 : static void
1521 1 : bdev_nvme_clear_io_path_cache(struct spdk_io_channel_iter *i)
1522 : {
1523 1 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
1524 1 : struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(_ch);
1525 :
1526 1 : assert(ctrlr_ch->qpair != NULL);
1527 :
1528 1 : _bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);
1529 :
1530 1 : spdk_for_each_channel_continue(i, 0);
1531 1 : }
1532 :
1533 : static void
1534 3 : bdev_nvme_clear_io_path_caches(struct nvme_ctrlr *nvme_ctrlr)
1535 : {
1536 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
1537 3 : if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
1538 : nvme_ctrlr->io_path_cache_clearing) {
1539 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1540 0 : return;
1541 : }
1542 :
1543 3 : nvme_ctrlr->io_path_cache_clearing = true;
1544 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1545 :
1546 3 : spdk_for_each_channel(nvme_ctrlr,
1547 : bdev_nvme_clear_io_path_cache,
1548 : NULL,
1549 : bdev_nvme_clear_io_path_caches_done);
1550 : }
1551 :
1552 : static struct nvme_qpair *
1553 99 : nvme_poll_group_get_qpair(struct nvme_poll_group *group, struct spdk_nvme_qpair *qpair)
1554 : {
1555 : struct nvme_qpair *nvme_qpair;
1556 :
1557 108 : TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
1558 108 : if (nvme_qpair->qpair == qpair) {
1559 99 : break;
1560 : }
1561 : }
1562 :
1563 99 : return nvme_qpair;
1564 : }
1565 :
1566 : static void nvme_qpair_delete(struct nvme_qpair *nvme_qpair);
1567 :
1568 : static void
1569 99 : bdev_nvme_disconnected_qpair_cb(struct spdk_nvme_qpair *qpair, void *poll_group_ctx)
1570 : {
1571 99 : struct nvme_poll_group *group = poll_group_ctx;
1572 : struct nvme_qpair *nvme_qpair;
1573 : struct nvme_ctrlr_channel *ctrlr_ch;
1574 : int status;
1575 :
1576 99 : nvme_qpair = nvme_poll_group_get_qpair(group, qpair);
1577 99 : if (nvme_qpair == NULL) {
1578 0 : return;
1579 : }
1580 :
1581 99 : if (nvme_qpair->qpair != NULL) {
1582 99 : spdk_nvme_ctrlr_free_io_qpair(nvme_qpair->qpair);
1583 99 : nvme_qpair->qpair = NULL;
1584 : }
1585 :
1586 99 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
1587 :
1588 99 : ctrlr_ch = nvme_qpair->ctrlr_ch;
1589 :
1590 99 : if (ctrlr_ch != NULL) {
1591 56 : if (ctrlr_ch->reset_iter != NULL) {
1592 : /* We are in a full reset sequence. */
1593 52 : if (ctrlr_ch->connect_poller != NULL) {
1594 : /* qpair was failed to connect. Abort the reset sequence. */
1595 0 : SPDK_DEBUGLOG(bdev_nvme, "qpair %p was failed to connect. abort the reset ctrlr sequence.\n",
1596 : qpair);
1597 0 : spdk_poller_unregister(&ctrlr_ch->connect_poller);
1598 0 : status = -1;
1599 : } else {
1600 : /* qpair was completed to disconnect. Just move to the next ctrlr_channel. */
1601 52 : SPDK_DEBUGLOG(bdev_nvme, "qpair %p was disconnected and freed in a reset ctrlr sequence.\n",
1602 : qpair);
1603 52 : status = 0;
1604 : }
1605 52 : spdk_for_each_channel_continue(ctrlr_ch->reset_iter, status);
1606 52 : ctrlr_ch->reset_iter = NULL;
1607 : } else {
1608 : /* qpair was disconnected unexpectedly. Reset controller for recovery. */
1609 4 : SPDK_NOTICELOG("qpair %p was disconnected and freed. reset controller.\n", qpair);
1610 4 : bdev_nvme_failover_ctrlr(nvme_qpair->ctrlr);
1611 : }
1612 : } else {
1613 : /* In this case, ctrlr_channel is already deleted. */
1614 43 : SPDK_DEBUGLOG(bdev_nvme, "qpair %p was disconnected and freed. delete nvme_qpair.\n", qpair);
1615 43 : nvme_qpair_delete(nvme_qpair);
1616 : }
1617 : }
1618 :
1619 : static void
1620 0 : bdev_nvme_check_io_qpairs(struct nvme_poll_group *group)
1621 : {
1622 : struct nvme_qpair *nvme_qpair;
1623 :
1624 0 : TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
1625 0 : if (nvme_qpair->qpair == NULL || nvme_qpair->ctrlr_ch == NULL) {
1626 0 : continue;
1627 : }
1628 :
1629 0 : if (spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
1630 : SPDK_NVME_QPAIR_FAILURE_NONE) {
1631 0 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
1632 : }
1633 : }
1634 0 : }
1635 :
1636 : static int
1637 1025 : bdev_nvme_poll(void *arg)
1638 : {
1639 1025 : struct nvme_poll_group *group = arg;
1640 : int64_t num_completions;
1641 :
1642 1025 : if (group->collect_spin_stat && group->start_ticks == 0) {
1643 0 : group->start_ticks = spdk_get_ticks();
1644 : }
1645 :
1646 1025 : num_completions = spdk_nvme_poll_group_process_completions(group->group, 0,
1647 : bdev_nvme_disconnected_qpair_cb);
1648 1025 : if (group->collect_spin_stat) {
1649 0 : if (num_completions > 0) {
1650 0 : if (group->end_ticks != 0) {
1651 0 : group->spin_ticks += (group->end_ticks - group->start_ticks);
1652 0 : group->end_ticks = 0;
1653 : }
1654 0 : group->start_ticks = 0;
1655 : } else {
1656 0 : group->end_ticks = spdk_get_ticks();
1657 : }
1658 : }
1659 :
1660 1025 : if (spdk_unlikely(num_completions < 0)) {
1661 0 : bdev_nvme_check_io_qpairs(group);
1662 : }
1663 :
1664 1025 : return num_completions > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
1665 : }
1666 :
1667 : static int bdev_nvme_poll_adminq(void *arg);
1668 :
1669 : static void
1670 100 : bdev_nvme_change_adminq_poll_period(struct nvme_ctrlr *nvme_ctrlr, uint64_t new_period_us)
1671 : {
1672 100 : spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);
1673 :
1674 100 : nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq,
1675 : nvme_ctrlr, new_period_us);
1676 100 : }
1677 :
1678 : static int
1679 146 : bdev_nvme_poll_adminq(void *arg)
1680 : {
1681 : int32_t rc;
1682 146 : struct nvme_ctrlr *nvme_ctrlr = arg;
1683 : nvme_ctrlr_disconnected_cb disconnected_cb;
1684 :
1685 146 : assert(nvme_ctrlr != NULL);
1686 :
1687 146 : rc = spdk_nvme_ctrlr_process_admin_completions(nvme_ctrlr->ctrlr);
1688 146 : if (rc < 0) {
1689 53 : disconnected_cb = nvme_ctrlr->disconnected_cb;
1690 53 : nvme_ctrlr->disconnected_cb = NULL;
1691 :
1692 53 : if (disconnected_cb != NULL) {
1693 50 : bdev_nvme_change_adminq_poll_period(nvme_ctrlr,
1694 : g_opts.nvme_adminq_poll_period_us);
1695 50 : disconnected_cb(nvme_ctrlr);
1696 : } else {
1697 3 : bdev_nvme_failover_ctrlr(nvme_ctrlr);
1698 : }
1699 93 : } else if (spdk_nvme_ctrlr_get_admin_qp_failure_reason(nvme_ctrlr->ctrlr) !=
1700 : SPDK_NVME_QPAIR_FAILURE_NONE) {
1701 0 : bdev_nvme_clear_io_path_caches(nvme_ctrlr);
1702 : }
1703 :
1704 146 : return rc == 0 ? SPDK_POLLER_IDLE : SPDK_POLLER_BUSY;
1705 : }
1706 :
1707 : static void
1708 37 : nvme_bdev_free(void *io_device)
1709 : {
1710 37 : struct nvme_bdev *nvme_disk = io_device;
1711 :
1712 37 : pthread_mutex_destroy(&nvme_disk->mutex);
1713 37 : free(nvme_disk->disk.name);
1714 37 : free(nvme_disk->err_stat);
1715 37 : free(nvme_disk);
1716 37 : }
1717 :
1718 : static int
1719 36 : bdev_nvme_destruct(void *ctx)
1720 : {
1721 36 : struct nvme_bdev *nvme_disk = ctx;
1722 : struct nvme_ns *nvme_ns, *tmp_nvme_ns;
1723 :
1724 : SPDK_DTRACE_PROBE2(bdev_nvme_destruct, nvme_disk->nbdev_ctrlr->name, nvme_disk->nsid);
1725 :
1726 73 : TAILQ_FOREACH_SAFE(nvme_ns, &nvme_disk->nvme_ns_list, tailq, tmp_nvme_ns) {
1727 37 : pthread_mutex_lock(&nvme_ns->ctrlr->mutex);
1728 :
1729 37 : nvme_ns->bdev = NULL;
1730 :
1731 37 : assert(nvme_ns->id > 0);
1732 :
1733 37 : if (nvme_ctrlr_get_ns(nvme_ns->ctrlr, nvme_ns->id) == NULL) {
1734 0 : pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
1735 :
1736 0 : nvme_ctrlr_release(nvme_ns->ctrlr);
1737 0 : nvme_ns_free(nvme_ns);
1738 : } else {
1739 37 : pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
1740 : }
1741 : }
1742 :
1743 36 : pthread_mutex_lock(&g_bdev_nvme_mutex);
1744 36 : TAILQ_REMOVE(&nvme_disk->nbdev_ctrlr->bdevs, nvme_disk, tailq);
1745 36 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
1746 :
1747 36 : spdk_io_device_unregister(nvme_disk, nvme_bdev_free);
1748 :
1749 36 : return 0;
1750 : }
1751 :
1752 : static int
1753 100 : bdev_nvme_create_qpair(struct nvme_qpair *nvme_qpair)
1754 : {
1755 : struct nvme_ctrlr *nvme_ctrlr;
1756 100 : struct spdk_nvme_io_qpair_opts opts;
1757 : struct spdk_nvme_qpair *qpair;
1758 : int rc;
1759 :
1760 100 : nvme_ctrlr = nvme_qpair->ctrlr;
1761 :
1762 100 : spdk_nvme_ctrlr_get_default_io_qpair_opts(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
1763 100 : opts.delay_cmd_submit = g_opts.delay_cmd_submit;
1764 100 : opts.create_only = true;
1765 100 : opts.async_mode = true;
1766 100 : opts.io_queue_requests = spdk_max(g_opts.io_queue_requests, opts.io_queue_requests);
1767 100 : g_opts.io_queue_requests = opts.io_queue_requests;
1768 :
1769 100 : qpair = spdk_nvme_ctrlr_alloc_io_qpair(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
1770 100 : if (qpair == NULL) {
1771 0 : return -1;
1772 : }
1773 :
1774 : SPDK_DTRACE_PROBE3(bdev_nvme_create_qpair, nvme_ctrlr->nbdev_ctrlr->name,
1775 : spdk_nvme_qpair_get_id(qpair), spdk_thread_get_id(nvme_ctrlr->thread));
1776 :
1777 100 : assert(nvme_qpair->group != NULL);
1778 :
1779 100 : rc = spdk_nvme_poll_group_add(nvme_qpair->group->group, qpair);
1780 100 : if (rc != 0) {
1781 0 : SPDK_ERRLOG("Unable to begin polling on NVMe Channel.\n");
1782 0 : goto err;
1783 : }
1784 :
1785 100 : rc = spdk_nvme_ctrlr_connect_io_qpair(nvme_ctrlr->ctrlr, qpair);
1786 100 : if (rc != 0) {
1787 0 : SPDK_ERRLOG("Unable to connect I/O qpair.\n");
1788 0 : goto err;
1789 : }
1790 :
1791 100 : nvme_qpair->qpair = qpair;
1792 :
1793 100 : if (!g_opts.disable_auto_failback) {
1794 71 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
1795 : }
1796 :
1797 100 : return 0;
1798 :
1799 0 : err:
1800 0 : spdk_nvme_ctrlr_free_io_qpair(qpair);
1801 :
1802 0 : return rc;
1803 : }
1804 :
1805 : static void bdev_nvme_reset_io_continue(void *cb_arg, int rc);
1806 :
1807 : static void
1808 82 : bdev_nvme_complete_pending_resets(struct spdk_io_channel_iter *i)
1809 : {
1810 82 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
1811 82 : struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(_ch);
1812 82 : int rc = 0;
1813 : struct spdk_bdev_io *bdev_io;
1814 : struct nvme_bdev_io *bio;
1815 :
1816 82 : if (spdk_io_channel_iter_get_ctx(i) != NULL) {
1817 35 : rc = -1;
1818 : }
1819 :
1820 86 : while (!TAILQ_EMPTY(&ctrlr_ch->pending_resets)) {
1821 4 : bdev_io = TAILQ_FIRST(&ctrlr_ch->pending_resets);
1822 4 : TAILQ_REMOVE(&ctrlr_ch->pending_resets, bdev_io, module_link);
1823 :
1824 4 : bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
1825 4 : bdev_nvme_reset_io_continue(bio, rc);
1826 : }
1827 :
1828 82 : spdk_for_each_channel_continue(i, 0);
1829 82 : }
1830 :
1831 : /* This function marks the current trid as failed by storing the current ticks
1832 : * and then sets the next trid to the active trid within a controller if exists.
1833 : *
1834 : * The purpose of the boolean return value is to request the caller to disconnect
1835 : * the current trid now to try connecting the next trid.
1836 : */
1837 : static bool
1838 36 : bdev_nvme_failover_trid(struct nvme_ctrlr *nvme_ctrlr, bool remove, bool start)
1839 : {
1840 : struct nvme_path_id *path_id, *next_path;
1841 : int rc __attribute__((unused));
1842 :
1843 36 : path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
1844 36 : assert(path_id);
1845 36 : assert(path_id == nvme_ctrlr->active_path_id);
1846 36 : next_path = TAILQ_NEXT(path_id, link);
1847 :
1848 : /* Update the last failed time. It means the trid is failed if its last
1849 : * failed time is non-zero.
1850 : */
1851 36 : path_id->last_failed_tsc = spdk_get_ticks();
1852 :
1853 36 : if (next_path == NULL) {
1854 : /* There is no alternate trid within a controller. */
1855 25 : return false;
1856 : }
1857 :
1858 11 : if (!start && nvme_ctrlr->opts.reconnect_delay_sec == 0) {
1859 : /* Connect is not retried in a controller reset sequence. Connecting
1860 : * the next trid will be done by the next bdev_nvme_failover_ctrlr() call.
1861 : */
1862 3 : return false;
1863 : }
1864 :
1865 8 : assert(path_id->trid.trtype != SPDK_NVME_TRANSPORT_PCIE);
1866 :
1867 8 : SPDK_NOTICELOG("Start failover from %s:%s to %s:%s\n", path_id->trid.traddr,
1868 : path_id->trid.trsvcid, next_path->trid.traddr, next_path->trid.trsvcid);
1869 :
1870 8 : spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
1871 8 : nvme_ctrlr->active_path_id = next_path;
1872 8 : rc = spdk_nvme_ctrlr_set_trid(nvme_ctrlr->ctrlr, &next_path->trid);
1873 8 : assert(rc == 0);
1874 8 : TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
1875 8 : if (!remove) {
1876 : /** Shuffle the old trid to the end of the list and use the new one.
1877 : * Allows for round robin through multiple connections.
1878 : */
1879 6 : TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, path_id, link);
1880 : } else {
1881 2 : free(path_id);
1882 : }
1883 :
1884 8 : if (start || next_path->last_failed_tsc == 0) {
1885 : /* bdev_nvme_failover_ctrlr() is just called or the next trid is not failed
1886 : * or used yet. Try the next trid now.
1887 : */
1888 7 : return true;
1889 : }
1890 :
1891 1 : if (spdk_get_ticks() > next_path->last_failed_tsc + spdk_get_ticks_hz() *
1892 1 : nvme_ctrlr->opts.reconnect_delay_sec) {
1893 : /* Enough backoff passed since the next trid failed. Try the next trid now. */
1894 0 : return true;
1895 : }
1896 :
1897 : /* The next trid will be tried after reconnect_delay_sec seconds. */
1898 1 : return false;
1899 : }
1900 :
1901 : static bool
1902 68 : bdev_nvme_check_ctrlr_loss_timeout(struct nvme_ctrlr *nvme_ctrlr)
1903 : {
1904 : int32_t elapsed;
1905 :
1906 68 : if (nvme_ctrlr->opts.ctrlr_loss_timeout_sec == 0 ||
1907 36 : nvme_ctrlr->opts.ctrlr_loss_timeout_sec == -1) {
1908 42 : return false;
1909 : }
1910 :
1911 26 : elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
1912 26 : if (elapsed >= nvme_ctrlr->opts.ctrlr_loss_timeout_sec) {
1913 6 : return true;
1914 : } else {
1915 20 : return false;
1916 : }
1917 : }
1918 :
1919 : static bool
1920 12 : bdev_nvme_check_fast_io_fail_timeout(struct nvme_ctrlr *nvme_ctrlr)
1921 : {
1922 : uint32_t elapsed;
1923 :
1924 12 : if (nvme_ctrlr->opts.fast_io_fail_timeout_sec == 0) {
1925 8 : return false;
1926 : }
1927 :
1928 4 : elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
1929 4 : if (elapsed >= nvme_ctrlr->opts.fast_io_fail_timeout_sec) {
1930 2 : return true;
1931 : } else {
1932 2 : return false;
1933 : }
1934 : }
1935 :
1936 : static void bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success);
1937 :
1938 : static void
1939 51 : nvme_ctrlr_disconnect(struct nvme_ctrlr *nvme_ctrlr, nvme_ctrlr_disconnected_cb cb_fn)
1940 : {
1941 : int rc;
1942 :
1943 51 : rc = spdk_nvme_ctrlr_disconnect(nvme_ctrlr->ctrlr);
1944 51 : if (rc != 0) {
1945 : /* Disconnect fails if ctrlr is already resetting or removed. In this case,
1946 : * fail the reset sequence immediately.
1947 : */
1948 1 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
1949 1 : return;
1950 : }
1951 :
1952 : /* spdk_nvme_ctrlr_disconnect() may complete asynchronously later by polling adminq.
1953 : * Set callback here to execute the specified operation after ctrlr is really disconnected.
1954 : */
1955 50 : assert(nvme_ctrlr->disconnected_cb == NULL);
1956 50 : nvme_ctrlr->disconnected_cb = cb_fn;
1957 :
1958 : /* During disconnection, reduce the period to poll adminq more often. */
1959 50 : bdev_nvme_change_adminq_poll_period(nvme_ctrlr, 0);
1960 : }
1961 :
1962 : enum bdev_nvme_op_after_reset {
1963 : OP_NONE,
1964 : OP_COMPLETE_PENDING_DESTRUCT,
1965 : OP_DESTRUCT,
1966 : OP_DELAYED_RECONNECT,
1967 : OP_FAILOVER,
1968 : };
1969 :
1970 : typedef enum bdev_nvme_op_after_reset _bdev_nvme_op_after_reset;
1971 :
1972 : static _bdev_nvme_op_after_reset
1973 50 : bdev_nvme_check_op_after_reset(struct nvme_ctrlr *nvme_ctrlr, bool success)
1974 : {
1975 50 : if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
1976 : /* Complete pending destruct after reset completes. */
1977 0 : return OP_COMPLETE_PENDING_DESTRUCT;
1978 50 : } else if (nvme_ctrlr->pending_failover) {
1979 3 : nvme_ctrlr->pending_failover = false;
1980 3 : nvme_ctrlr->reset_start_tsc = 0;
1981 3 : return OP_FAILOVER;
1982 47 : } else if (success || nvme_ctrlr->opts.reconnect_delay_sec == 0) {
1983 33 : nvme_ctrlr->reset_start_tsc = 0;
1984 33 : return OP_NONE;
1985 14 : } else if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
1986 2 : return OP_DESTRUCT;
1987 : } else {
1988 12 : if (bdev_nvme_check_fast_io_fail_timeout(nvme_ctrlr)) {
1989 2 : nvme_ctrlr->fast_io_fail_timedout = true;
1990 : }
1991 12 : return OP_DELAYED_RECONNECT;
1992 : }
1993 : }
1994 :
1995 : static int bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug);
1996 : static void bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
1997 :
1998 : static int
1999 9 : bdev_nvme_reconnect_delay_timer_expired(void *ctx)
2000 : {
2001 9 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2002 :
2003 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect_delay, nvme_ctrlr->nbdev_ctrlr->name);
2004 9 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2005 :
2006 9 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2007 :
2008 9 : if (!nvme_ctrlr->reconnect_is_delayed) {
2009 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2010 0 : return SPDK_POLLER_BUSY;
2011 : }
2012 :
2013 9 : nvme_ctrlr->reconnect_is_delayed = false;
2014 :
2015 9 : if (nvme_ctrlr->destruct) {
2016 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2017 0 : return SPDK_POLLER_BUSY;
2018 : }
2019 :
2020 9 : assert(nvme_ctrlr->resetting == false);
2021 9 : nvme_ctrlr->resetting = true;
2022 :
2023 9 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2024 :
2025 9 : spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);
2026 :
2027 9 : bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2028 9 : return SPDK_POLLER_BUSY;
2029 : }
2030 :
2031 : static void
2032 12 : bdev_nvme_start_reconnect_delay_timer(struct nvme_ctrlr *nvme_ctrlr)
2033 : {
2034 12 : spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);
2035 :
2036 12 : assert(nvme_ctrlr->reconnect_is_delayed == false);
2037 12 : nvme_ctrlr->reconnect_is_delayed = true;
2038 :
2039 12 : assert(nvme_ctrlr->reconnect_delay_timer == NULL);
2040 12 : nvme_ctrlr->reconnect_delay_timer = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_delay_timer_expired,
2041 : nvme_ctrlr,
2042 : nvme_ctrlr->opts.reconnect_delay_sec * SPDK_SEC_TO_USEC);
2043 12 : }
2044 :
2045 : static void remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr);
2046 :
2047 : static void
2048 48 : _bdev_nvme_reset_ctrlr_complete(struct spdk_io_channel_iter *i, int status)
2049 : {
2050 48 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
2051 48 : bool success = spdk_io_channel_iter_get_ctx(i) == NULL;
2052 48 : bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
2053 48 : void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
2054 : enum bdev_nvme_op_after_reset op_after_reset;
2055 :
2056 48 : assert(nvme_ctrlr->thread == spdk_get_thread());
2057 :
2058 48 : nvme_ctrlr->ctrlr_op_cb_fn = NULL;
2059 48 : nvme_ctrlr->ctrlr_op_cb_arg = NULL;
2060 :
2061 48 : if (!success) {
2062 21 : SPDK_ERRLOG("Resetting controller failed.\n");
2063 : } else {
2064 27 : SPDK_NOTICELOG("Resetting controller successful.\n");
2065 : }
2066 :
2067 48 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2068 48 : nvme_ctrlr->resetting = false;
2069 48 : nvme_ctrlr->dont_retry = false;
2070 48 : nvme_ctrlr->in_failover = false;
2071 :
2072 48 : op_after_reset = bdev_nvme_check_op_after_reset(nvme_ctrlr, success);
2073 48 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2074 :
2075 : /* Delay callbacks when the next operation is a failover. */
2076 48 : if (ctrlr_op_cb_fn && op_after_reset != OP_FAILOVER) {
2077 10 : ctrlr_op_cb_fn(ctrlr_op_cb_arg, success ? 0 : -1);
2078 : }
2079 :
2080 48 : switch (op_after_reset) {
2081 0 : case OP_COMPLETE_PENDING_DESTRUCT:
2082 0 : nvme_ctrlr_unregister(nvme_ctrlr);
2083 0 : break;
2084 2 : case OP_DESTRUCT:
2085 2 : bdev_nvme_delete_ctrlr(nvme_ctrlr, false);
2086 2 : remove_discovery_entry(nvme_ctrlr);
2087 2 : break;
2088 12 : case OP_DELAYED_RECONNECT:
2089 12 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_start_reconnect_delay_timer);
2090 12 : break;
2091 3 : case OP_FAILOVER:
2092 3 : nvme_ctrlr->ctrlr_op_cb_fn = ctrlr_op_cb_fn;
2093 3 : nvme_ctrlr->ctrlr_op_cb_arg = ctrlr_op_cb_arg;
2094 3 : bdev_nvme_failover_ctrlr(nvme_ctrlr);
2095 3 : break;
2096 31 : default:
2097 31 : break;
2098 : }
2099 48 : }
2100 :
2101 : static void
2102 50 : bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success)
2103 : {
2104 50 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2105 50 : if (!success) {
2106 : /* Connecting the active trid failed. Set the next alternate trid to the
2107 : * active trid if it exists.
2108 : */
2109 23 : if (bdev_nvme_failover_trid(nvme_ctrlr, false, false)) {
2110 : /* The next alternate trid exists and is ready to try. Try it now. */
2111 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2112 :
2113 2 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
2114 2 : return;
2115 : }
2116 :
2117 : /* We came here if there is no alternate trid or if the next trid exists but
2118 : * is not ready to try. We will try the active trid after reconnect_delay_sec
2119 : * seconds if it is non-zero or at the next reset call otherwise.
2120 : */
2121 : } else {
2122 : /* Connecting the active trid succeeded. Clear the last failed time because it
2123 : * means the trid is failed if its last failed time is non-zero.
2124 : */
2125 27 : nvme_ctrlr->active_path_id->last_failed_tsc = 0;
2126 : }
2127 48 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2128 :
2129 : /* Make sure we clear any pending resets before returning. */
2130 48 : spdk_for_each_channel(nvme_ctrlr,
2131 : bdev_nvme_complete_pending_resets,
2132 : success ? NULL : (void *)0x1,
2133 : _bdev_nvme_reset_ctrlr_complete);
2134 : }
2135 :
2136 : static void
2137 0 : bdev_nvme_reset_create_qpairs_failed(struct spdk_io_channel_iter *i, int status)
2138 : {
2139 0 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
2140 :
2141 0 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2142 0 : }
2143 :
2144 : static void
2145 62 : bdev_nvme_reset_destroy_qpair(struct spdk_io_channel_iter *i)
2146 : {
2147 62 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
2148 62 : struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(ch);
2149 : struct nvme_qpair *nvme_qpair;
2150 :
2151 62 : nvme_qpair = ctrlr_ch->qpair;
2152 62 : assert(nvme_qpair != NULL);
2153 :
2154 62 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
2155 :
2156 62 : if (nvme_qpair->qpair != NULL) {
2157 52 : if (nvme_qpair->ctrlr->dont_retry) {
2158 39 : spdk_nvme_qpair_set_abort_dnr(nvme_qpair->qpair, true);
2159 : }
2160 52 : spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);
2161 :
2162 : /* The current full reset sequence will move to the next
2163 : * ctrlr_channel after the qpair is actually disconnected.
2164 : */
2165 52 : assert(ctrlr_ch->reset_iter == NULL);
2166 52 : ctrlr_ch->reset_iter = i;
2167 : } else {
2168 10 : spdk_for_each_channel_continue(i, 0);
2169 : }
2170 62 : }
2171 :
2172 : static void
2173 27 : bdev_nvme_reset_create_qpairs_done(struct spdk_io_channel_iter *i, int status)
2174 : {
2175 27 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
2176 :
2177 27 : if (status == 0) {
2178 27 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, true);
2179 : } else {
2180 : /* Delete the added qpairs and quiesce ctrlr to make the states clean. */
2181 0 : spdk_for_each_channel(nvme_ctrlr,
2182 : bdev_nvme_reset_destroy_qpair,
2183 : NULL,
2184 : bdev_nvme_reset_create_qpairs_failed);
2185 : }
2186 27 : }
2187 :
2188 : static int
2189 43 : bdev_nvme_reset_check_qpair_connected(void *ctx)
2190 : {
2191 43 : struct nvme_ctrlr_channel *ctrlr_ch = ctx;
2192 :
2193 43 : if (ctrlr_ch->reset_iter == NULL) {
2194 : /* qpair was already failed to connect and the reset sequence is being aborted. */
2195 0 : assert(ctrlr_ch->connect_poller == NULL);
2196 0 : assert(ctrlr_ch->qpair->qpair == NULL);
2197 0 : return SPDK_POLLER_BUSY;
2198 : }
2199 :
2200 43 : assert(ctrlr_ch->qpair->qpair != NULL);
2201 :
2202 43 : if (!spdk_nvme_qpair_is_connected(ctrlr_ch->qpair->qpair)) {
2203 0 : return SPDK_POLLER_BUSY;
2204 : }
2205 :
2206 43 : spdk_poller_unregister(&ctrlr_ch->connect_poller);
2207 :
2208 : /* qpair was completed to connect. Move to the next ctrlr_channel */
2209 43 : spdk_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
2210 43 : ctrlr_ch->reset_iter = NULL;
2211 :
2212 43 : if (!g_opts.disable_auto_failback) {
2213 30 : _bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);
2214 : }
2215 :
2216 43 : return SPDK_POLLER_BUSY;
2217 : }
2218 :
2219 : static void
2220 43 : bdev_nvme_reset_create_qpair(struct spdk_io_channel_iter *i)
2221 : {
2222 43 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
2223 43 : struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(_ch);
2224 : int rc;
2225 :
2226 43 : rc = bdev_nvme_create_qpair(ctrlr_ch->qpair);
2227 43 : if (rc == 0) {
2228 43 : ctrlr_ch->connect_poller = SPDK_POLLER_REGISTER(bdev_nvme_reset_check_qpair_connected,
2229 : ctrlr_ch, 0);
2230 :
2231 : /* The current full reset sequence will move to the next
2232 : * ctrlr_channel after the qpair is actually connected.
2233 : */
2234 43 : assert(ctrlr_ch->reset_iter == NULL);
2235 43 : ctrlr_ch->reset_iter = i;
2236 : } else {
2237 0 : spdk_for_each_channel_continue(i, rc);
2238 : }
2239 43 : }
2240 :
2241 : static void
2242 27 : nvme_ctrlr_check_namespaces(struct nvme_ctrlr *nvme_ctrlr)
2243 : {
2244 27 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
2245 : struct nvme_ns *nvme_ns;
2246 :
2247 39 : for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
2248 : nvme_ns != NULL;
2249 12 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
2250 12 : if (!spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
2251 1 : SPDK_DEBUGLOG(bdev_nvme, "NSID %u was removed during reset.\n", nvme_ns->id);
2252 : /* NS can be added again. Just nullify nvme_ns->ns. */
2253 1 : nvme_ns->ns = NULL;
2254 : }
2255 : }
2256 27 : }
2257 :
2258 :
2259 : static int
2260 49 : bdev_nvme_reconnect_ctrlr_poll(void *arg)
2261 : {
2262 49 : struct nvme_ctrlr *nvme_ctrlr = arg;
2263 49 : int rc = -ETIMEDOUT;
2264 :
2265 49 : if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
2266 : /* Mark the ctrlr as failed. The next call to
2267 : * spdk_nvme_ctrlr_reconnect_poll_async() will then
2268 : * do the necessary cleanup and return failure.
2269 : */
2270 2 : spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
2271 : }
2272 :
2273 49 : rc = spdk_nvme_ctrlr_reconnect_poll_async(nvme_ctrlr->ctrlr);
2274 49 : if (rc == -EAGAIN) {
2275 0 : return SPDK_POLLER_BUSY;
2276 : }
2277 :
2278 49 : spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
2279 49 : if (rc == 0) {
2280 27 : nvme_ctrlr_check_namespaces(nvme_ctrlr);
2281 :
2282 : /* Recreate all of the I/O queue pairs */
2283 27 : spdk_for_each_channel(nvme_ctrlr,
2284 : bdev_nvme_reset_create_qpair,
2285 : NULL,
2286 : bdev_nvme_reset_create_qpairs_done);
2287 : } else {
2288 22 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2289 : }
2290 49 : return SPDK_POLLER_BUSY;
2291 : }
2292 :
2293 : static void
2294 49 : bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2295 : {
2296 49 : spdk_nvme_ctrlr_reconnect_async(nvme_ctrlr->ctrlr);
2297 :
2298 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect, nvme_ctrlr->nbdev_ctrlr->name);
2299 49 : assert(nvme_ctrlr->reset_detach_poller == NULL);
2300 49 : nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_ctrlr_poll,
2301 : nvme_ctrlr, 0);
2302 49 : }
2303 :
2304 : static void
2305 36 : bdev_nvme_reset_destroy_qpair_done(struct spdk_io_channel_iter *i, int status)
2306 : {
2307 36 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
2308 :
2309 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reset, nvme_ctrlr->nbdev_ctrlr->name);
2310 36 : assert(status == 0);
2311 :
2312 36 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2313 0 : bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2314 : } else {
2315 36 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
2316 : }
2317 36 : }
2318 :
2319 : static void
2320 36 : bdev_nvme_reset_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
2321 : {
2322 36 : spdk_for_each_channel(nvme_ctrlr,
2323 : bdev_nvme_reset_destroy_qpair,
2324 : NULL,
2325 : bdev_nvme_reset_destroy_qpair_done);
2326 36 : }
2327 :
2328 : static void
2329 3 : bdev_nvme_reconnect_ctrlr_now(void *ctx)
2330 : {
2331 3 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2332 :
2333 3 : assert(nvme_ctrlr->resetting == true);
2334 3 : assert(nvme_ctrlr->thread == spdk_get_thread());
2335 :
2336 3 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2337 :
2338 3 : spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);
2339 :
2340 3 : bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2341 3 : }
2342 :
2343 : static void
2344 36 : _bdev_nvme_reset_ctrlr(void *ctx)
2345 : {
2346 36 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2347 :
2348 36 : assert(nvme_ctrlr->resetting == true);
2349 36 : assert(nvme_ctrlr->thread == spdk_get_thread());
2350 :
2351 36 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2352 0 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reset_destroy_qpairs);
2353 : } else {
2354 36 : bdev_nvme_reset_destroy_qpairs(nvme_ctrlr);
2355 : }
2356 36 : }
2357 :
2358 : static int
2359 34 : bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2360 : {
2361 : spdk_msg_fn msg_fn;
2362 :
2363 34 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2364 34 : if (nvme_ctrlr->destruct) {
2365 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2366 3 : return -ENXIO;
2367 : }
2368 :
2369 31 : if (nvme_ctrlr->resetting) {
2370 6 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2371 6 : SPDK_NOTICELOG("Unable to perform reset, already in progress.\n");
2372 6 : return -EBUSY;
2373 : }
2374 :
2375 25 : if (nvme_ctrlr->disabled) {
2376 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2377 0 : SPDK_NOTICELOG("Unable to perform reset. Controller is disabled.\n");
2378 0 : return -EALREADY;
2379 : }
2380 :
2381 25 : nvme_ctrlr->resetting = true;
2382 25 : nvme_ctrlr->dont_retry = true;
2383 :
2384 25 : if (nvme_ctrlr->reconnect_is_delayed) {
2385 1 : SPDK_DEBUGLOG(bdev_nvme, "Reconnect is already scheduled.\n");
2386 1 : msg_fn = bdev_nvme_reconnect_ctrlr_now;
2387 1 : nvme_ctrlr->reconnect_is_delayed = false;
2388 : } else {
2389 24 : msg_fn = _bdev_nvme_reset_ctrlr;
2390 24 : assert(nvme_ctrlr->reset_start_tsc == 0);
2391 : }
2392 :
2393 25 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2394 :
2395 25 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2396 :
2397 25 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
2398 25 : return 0;
2399 : }
2400 :
2401 : static int
2402 3 : bdev_nvme_enable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2403 : {
2404 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2405 3 : if (nvme_ctrlr->destruct) {
2406 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2407 0 : return -ENXIO;
2408 : }
2409 :
2410 3 : if (nvme_ctrlr->resetting) {
2411 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2412 0 : return -EBUSY;
2413 : }
2414 :
2415 3 : if (!nvme_ctrlr->disabled) {
2416 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2417 1 : return -EALREADY;
2418 : }
2419 :
2420 2 : nvme_ctrlr->disabled = false;
2421 2 : nvme_ctrlr->resetting = true;
2422 :
2423 2 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2424 :
2425 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2426 :
2427 2 : spdk_thread_send_msg(nvme_ctrlr->thread, bdev_nvme_reconnect_ctrlr_now, nvme_ctrlr);
2428 2 : return 0;
2429 : }
2430 :
2431 : static void
2432 2 : _bdev_nvme_disable_ctrlr_complete(struct spdk_io_channel_iter *i, int status)
2433 : {
2434 2 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
2435 2 : bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
2436 2 : void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
2437 : enum bdev_nvme_op_after_reset op_after_disable;
2438 :
2439 2 : assert(nvme_ctrlr->thread == spdk_get_thread());
2440 :
2441 2 : nvme_ctrlr->ctrlr_op_cb_fn = NULL;
2442 2 : nvme_ctrlr->ctrlr_op_cb_arg = NULL;
2443 :
2444 2 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2445 :
2446 2 : nvme_ctrlr->resetting = false;
2447 2 : nvme_ctrlr->dont_retry = false;
2448 :
2449 2 : op_after_disable = bdev_nvme_check_op_after_reset(nvme_ctrlr, true);
2450 :
2451 2 : nvme_ctrlr->disabled = true;
2452 2 : spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);
2453 :
2454 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2455 :
2456 2 : if (ctrlr_op_cb_fn) {
2457 0 : ctrlr_op_cb_fn(ctrlr_op_cb_arg, 0);
2458 : }
2459 :
2460 2 : switch (op_after_disable) {
2461 0 : case OP_COMPLETE_PENDING_DESTRUCT:
2462 0 : nvme_ctrlr_unregister(nvme_ctrlr);
2463 0 : break;
2464 2 : default:
2465 2 : break;
2466 : }
2467 :
2468 2 : }
2469 :
2470 : static void
2471 2 : bdev_nvme_disable_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr)
2472 : {
2473 : /* Make sure we clear any pending resets before returning. */
2474 2 : spdk_for_each_channel(nvme_ctrlr,
2475 : bdev_nvme_complete_pending_resets,
2476 : NULL,
2477 : _bdev_nvme_disable_ctrlr_complete);
2478 2 : }
2479 :
2480 : static void
2481 1 : bdev_nvme_disable_destroy_qpairs_done(struct spdk_io_channel_iter *i, int status)
2482 : {
2483 1 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
2484 :
2485 1 : assert(status == 0);
2486 :
2487 1 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2488 0 : bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
2489 : } else {
2490 1 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_ctrlr_complete);
2491 : }
2492 1 : }
2493 :
2494 : static void
2495 1 : bdev_nvme_disable_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
2496 : {
2497 1 : spdk_for_each_channel(nvme_ctrlr,
2498 : bdev_nvme_reset_destroy_qpair,
2499 : NULL,
2500 : bdev_nvme_disable_destroy_qpairs_done);
2501 1 : }
2502 :
2503 : static void
2504 1 : _bdev_nvme_cancel_reconnect_and_disable_ctrlr(void *ctx)
2505 : {
2506 1 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2507 :
2508 1 : assert(nvme_ctrlr->resetting == true);
2509 1 : assert(nvme_ctrlr->thread == spdk_get_thread());
2510 :
2511 1 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2512 :
2513 1 : bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
2514 1 : }
2515 :
2516 : static void
2517 1 : _bdev_nvme_disconnect_and_disable_ctrlr(void *ctx)
2518 : {
2519 1 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2520 :
2521 1 : assert(nvme_ctrlr->resetting == true);
2522 1 : assert(nvme_ctrlr->thread == spdk_get_thread());
2523 :
2524 1 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2525 0 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_destroy_qpairs);
2526 : } else {
2527 1 : bdev_nvme_disable_destroy_qpairs(nvme_ctrlr);
2528 : }
2529 1 : }
2530 :
2531 : static int
2532 5 : bdev_nvme_disable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2533 : {
2534 : spdk_msg_fn msg_fn;
2535 :
2536 5 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2537 5 : if (nvme_ctrlr->destruct) {
2538 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2539 1 : return -ENXIO;
2540 : }
2541 :
2542 4 : if (nvme_ctrlr->resetting) {
2543 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2544 1 : return -EBUSY;
2545 : }
2546 :
2547 3 : if (nvme_ctrlr->disabled) {
2548 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2549 1 : return -EALREADY;
2550 : }
2551 :
2552 2 : nvme_ctrlr->resetting = true;
2553 2 : nvme_ctrlr->dont_retry = true;
2554 :
2555 2 : if (nvme_ctrlr->reconnect_is_delayed) {
2556 1 : msg_fn = _bdev_nvme_cancel_reconnect_and_disable_ctrlr;
2557 1 : nvme_ctrlr->reconnect_is_delayed = false;
2558 : } else {
2559 1 : msg_fn = _bdev_nvme_disconnect_and_disable_ctrlr;
2560 : }
2561 :
2562 2 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2563 :
2564 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2565 :
2566 2 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
2567 2 : return 0;
2568 : }
2569 :
2570 : static int
2571 16 : nvme_ctrlr_op(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
2572 : bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2573 : {
2574 : int rc;
2575 :
2576 16 : switch (op) {
2577 15 : case NVME_CTRLR_OP_RESET:
2578 15 : rc = bdev_nvme_reset_ctrlr(nvme_ctrlr);
2579 15 : break;
2580 0 : case NVME_CTRLR_OP_ENABLE:
2581 0 : rc = bdev_nvme_enable_ctrlr(nvme_ctrlr);
2582 0 : break;
2583 0 : case NVME_CTRLR_OP_DISABLE:
2584 0 : rc = bdev_nvme_disable_ctrlr(nvme_ctrlr);
2585 0 : break;
2586 1 : default:
2587 1 : rc = -EINVAL;
2588 1 : break;
2589 : }
2590 :
2591 16 : if (rc == 0) {
2592 9 : assert(nvme_ctrlr->ctrlr_op_cb_fn == NULL);
2593 9 : assert(nvme_ctrlr->ctrlr_op_cb_arg == NULL);
2594 9 : nvme_ctrlr->ctrlr_op_cb_fn = cb_fn;
2595 9 : nvme_ctrlr->ctrlr_op_cb_arg = cb_arg;
2596 : }
2597 16 : return rc;
2598 : }
2599 :
2600 : struct nvme_ctrlr_op_rpc_ctx {
2601 : struct nvme_ctrlr *nvme_ctrlr;
2602 : struct spdk_thread *orig_thread;
2603 : enum nvme_ctrlr_op op;
2604 : int rc;
2605 : bdev_nvme_ctrlr_op_cb cb_fn;
2606 : void *cb_arg;
2607 : };
2608 :
2609 : static void
2610 4 : _nvme_ctrlr_op_rpc_complete(void *_ctx)
2611 : {
2612 4 : struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
2613 :
2614 4 : assert(ctx != NULL);
2615 4 : assert(ctx->cb_fn != NULL);
2616 :
2617 4 : ctx->cb_fn(ctx->cb_arg, ctx->rc);
2618 :
2619 4 : free(ctx);
2620 4 : }
2621 :
2622 : static void
2623 4 : nvme_ctrlr_op_rpc_complete(void *cb_arg, int rc)
2624 : {
2625 4 : struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;
2626 :
2627 4 : ctx->rc = rc;
2628 :
2629 4 : spdk_thread_send_msg(ctx->orig_thread, _nvme_ctrlr_op_rpc_complete, ctx);
2630 4 : }
2631 :
2632 : void
2633 4 : nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
2634 : bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2635 : {
2636 : struct nvme_ctrlr_op_rpc_ctx *ctx;
2637 : int rc;
2638 :
2639 4 : assert(cb_fn != NULL);
2640 :
2641 4 : ctx = calloc(1, sizeof(*ctx));
2642 4 : if (ctx == NULL) {
2643 0 : SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
2644 0 : cb_fn(cb_arg, -ENOMEM);
2645 0 : return;
2646 : }
2647 :
2648 4 : ctx->orig_thread = spdk_get_thread();
2649 4 : ctx->cb_fn = cb_fn;
2650 4 : ctx->cb_arg = cb_arg;
2651 :
2652 4 : rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_ctrlr_op_rpc_complete, ctx);
2653 4 : if (rc == 0) {
2654 1 : return;
2655 3 : } else if (rc == -EALREADY) {
2656 0 : rc = 0;
2657 : }
2658 :
2659 3 : nvme_ctrlr_op_rpc_complete(ctx, rc);
2660 : }
2661 :
2662 : static void nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc);
2663 :
2664 : static void
2665 2 : _nvme_bdev_ctrlr_op_rpc_continue(void *_ctx)
2666 : {
2667 2 : struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
2668 : struct nvme_ctrlr *prev_nvme_ctrlr, *next_nvme_ctrlr;
2669 : int rc;
2670 :
2671 2 : prev_nvme_ctrlr = ctx->nvme_ctrlr;
2672 2 : ctx->nvme_ctrlr = NULL;
2673 :
2674 2 : if (ctx->rc != 0) {
2675 0 : goto complete;
2676 : }
2677 :
2678 2 : next_nvme_ctrlr = TAILQ_NEXT(prev_nvme_ctrlr, tailq);
2679 2 : if (next_nvme_ctrlr == NULL) {
2680 1 : goto complete;
2681 : }
2682 :
2683 1 : rc = nvme_ctrlr_op(next_nvme_ctrlr, ctx->op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
2684 1 : if (rc == 0) {
2685 1 : ctx->nvme_ctrlr = next_nvme_ctrlr;
2686 1 : return;
2687 0 : } else if (rc == -EALREADY) {
2688 0 : ctx->nvme_ctrlr = next_nvme_ctrlr;
2689 0 : rc = 0;
2690 : }
2691 :
2692 0 : ctx->rc = rc;
2693 :
2694 1 : complete:
2695 1 : ctx->cb_fn(ctx->cb_arg, ctx->rc);
2696 1 : free(ctx);
2697 : }
2698 :
2699 : static void
2700 2 : nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc)
2701 : {
2702 2 : struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;
2703 :
2704 2 : ctx->rc = rc;
2705 :
2706 2 : spdk_thread_send_msg(ctx->orig_thread, _nvme_bdev_ctrlr_op_rpc_continue, ctx);
2707 2 : }
2708 :
2709 : void
2710 1 : nvme_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op,
2711 : bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2712 : {
2713 : struct nvme_ctrlr_op_rpc_ctx *ctx;
2714 : struct nvme_ctrlr *nvme_ctrlr;
2715 : int rc;
2716 :
2717 1 : assert(cb_fn != NULL);
2718 :
2719 1 : ctx = calloc(1, sizeof(*ctx));
2720 1 : if (ctx == NULL) {
2721 0 : SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
2722 0 : cb_fn(cb_arg, -ENOMEM);
2723 0 : return;
2724 : }
2725 :
2726 1 : ctx->orig_thread = spdk_get_thread();
2727 1 : ctx->op = op;
2728 1 : ctx->cb_fn = cb_fn;
2729 1 : ctx->cb_arg = cb_arg;
2730 :
2731 1 : nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
2732 1 : assert(nvme_ctrlr != NULL);
2733 :
2734 1 : rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
2735 1 : if (rc == 0) {
2736 1 : ctx->nvme_ctrlr = nvme_ctrlr;
2737 1 : return;
2738 0 : } else if (rc == -EALREADY) {
2739 0 : ctx->nvme_ctrlr = nvme_ctrlr;
2740 0 : rc = 0;
2741 : }
2742 :
2743 0 : nvme_bdev_ctrlr_op_rpc_continue(ctx, rc);
2744 : }
2745 :
2746 : static int _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio);
2747 :
2748 : static void
2749 7 : _bdev_nvme_reset_io_complete(struct spdk_io_channel_iter *i, int status)
2750 : {
2751 7 : struct nvme_bdev_io *bio = spdk_io_channel_iter_get_ctx(i);
2752 : enum spdk_bdev_io_status io_status;
2753 :
2754 7 : if (bio->cpl.cdw0 == 0) {
2755 5 : io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2756 : } else {
2757 2 : io_status = SPDK_BDEV_IO_STATUS_FAILED;
2758 : }
2759 :
2760 7 : __bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), io_status, NULL);
2761 7 : }
2762 :
2763 : static void
2764 14 : bdev_nvme_abort_bdev_channel(struct spdk_io_channel_iter *i)
2765 : {
2766 14 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
2767 14 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
2768 :
2769 14 : bdev_nvme_abort_retry_ios(nbdev_ch);
2770 :
2771 14 : spdk_for_each_channel_continue(i, 0);
2772 14 : }
2773 :
2774 : static void
2775 7 : bdev_nvme_reset_io_complete(struct nvme_bdev_io *bio)
2776 : {
2777 7 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2778 7 : struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
2779 :
2780 : /* Abort all queued I/Os for retry. */
2781 7 : spdk_for_each_channel(nbdev,
2782 : bdev_nvme_abort_bdev_channel,
2783 : bio,
2784 : _bdev_nvme_reset_io_complete);
2785 7 : }
2786 :
2787 : static void
2788 10 : _bdev_nvme_reset_io_continue(void *ctx)
2789 : {
2790 10 : struct nvme_bdev_io *bio = ctx;
2791 : struct nvme_io_path *prev_io_path, *next_io_path;
2792 : int rc;
2793 :
2794 10 : prev_io_path = bio->io_path;
2795 10 : bio->io_path = NULL;
2796 :
2797 10 : if (bio->cpl.cdw0 != 0) {
2798 2 : goto complete;
2799 : }
2800 :
2801 8 : next_io_path = STAILQ_NEXT(prev_io_path, stailq);
2802 8 : if (next_io_path == NULL) {
2803 5 : goto complete;
2804 : }
2805 :
2806 3 : rc = _bdev_nvme_reset_io(next_io_path, bio);
2807 3 : if (rc == 0) {
2808 3 : return;
2809 : }
2810 :
2811 0 : bio->cpl.cdw0 = 1;
2812 :
2813 7 : complete:
2814 7 : bdev_nvme_reset_io_complete(bio);
2815 : }
2816 :
2817 : static void
2818 10 : bdev_nvme_reset_io_continue(void *cb_arg, int rc)
2819 : {
2820 10 : struct nvme_bdev_io *bio = cb_arg;
2821 10 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2822 :
2823 10 : bio->cpl.cdw0 = (rc == 0) ? 0 : 1;
2824 :
2825 10 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), _bdev_nvme_reset_io_continue, bio);
2826 10 : }
2827 :
2828 : static int
2829 10 : _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio)
2830 : {
2831 : struct nvme_ctrlr_channel *ctrlr_ch;
2832 : struct spdk_bdev_io *bdev_io;
2833 : int rc;
2834 :
2835 10 : rc = nvme_ctrlr_op(io_path->qpair->ctrlr, NVME_CTRLR_OP_RESET,
2836 : bdev_nvme_reset_io_continue, bio);
2837 10 : if (rc != 0 && rc != -EBUSY) {
2838 0 : return rc;
2839 : }
2840 :
2841 10 : assert(bio->io_path == NULL);
2842 10 : bio->io_path = io_path;
2843 :
2844 10 : if (rc == -EBUSY) {
2845 4 : ctrlr_ch = io_path->qpair->ctrlr_ch;
2846 4 : assert(ctrlr_ch != NULL);
2847 : /*
2848 : * Reset call is queued only if it is from the app framework. This is on purpose so that
2849 : * we don't interfere with the app framework reset strategy. i.e. we are deferring to the
2850 : * upper level. If they are in the middle of a reset, we won't try to schedule another one.
2851 : */
2852 4 : bdev_io = spdk_bdev_io_from_ctx(bio);
2853 4 : TAILQ_INSERT_TAIL(&ctrlr_ch->pending_resets, bdev_io, module_link);
2854 : }
2855 :
2856 10 : return 0;
2857 : }
2858 :
2859 : static void
2860 7 : bdev_nvme_reset_io(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio)
2861 : {
2862 : struct nvme_io_path *io_path;
2863 : int rc;
2864 :
2865 7 : bio->cpl.cdw0 = 0;
2866 :
2867 : /* Reset all nvme_ctrlrs of a bdev controller sequentially. */
2868 7 : io_path = STAILQ_FIRST(&nbdev_ch->io_path_list);
2869 7 : assert(io_path != NULL);
2870 :
2871 7 : rc = _bdev_nvme_reset_io(io_path, bio);
2872 7 : if (rc != 0) {
2873 : /* If the current nvme_ctrlr is disabled, skip it and move to the next nvme_ctrlr. */
2874 0 : rc = (rc == -EALREADY) ? 0 : rc;
2875 :
2876 0 : bdev_nvme_reset_io_continue(bio, rc);
2877 : }
2878 7 : }
2879 :
2880 : static int
2881 18 : bdev_nvme_failover_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool remove)
2882 : {
2883 18 : if (nvme_ctrlr->destruct) {
2884 : /* Don't bother resetting if the controller is in the process of being destructed. */
2885 2 : return -ENXIO;
2886 : }
2887 :
2888 16 : if (nvme_ctrlr->resetting) {
2889 3 : if (!nvme_ctrlr->in_failover) {
2890 3 : SPDK_NOTICELOG("Reset is already in progress. Defer failover until reset completes.\n");
2891 :
2892 : /* Defer failover until reset completes. */
2893 3 : nvme_ctrlr->pending_failover = true;
2894 3 : return -EINPROGRESS;
2895 : } else {
2896 0 : SPDK_NOTICELOG("Unable to perform failover, already in progress.\n");
2897 0 : return -EBUSY;
2898 : }
2899 : }
2900 :
2901 13 : bdev_nvme_failover_trid(nvme_ctrlr, remove, true);
2902 :
2903 13 : if (nvme_ctrlr->reconnect_is_delayed) {
2904 1 : SPDK_NOTICELOG("Reconnect is already scheduled.\n");
2905 :
2906 : /* We rely on the next reconnect for the failover. */
2907 1 : return -EALREADY;
2908 : }
2909 :
2910 12 : if (nvme_ctrlr->disabled) {
2911 0 : SPDK_NOTICELOG("Controller is disabled.\n");
2912 :
2913 : /* We rely on the enablement for the failover. */
2914 0 : return -EALREADY;
2915 : }
2916 :
2917 12 : nvme_ctrlr->resetting = true;
2918 12 : nvme_ctrlr->in_failover = true;
2919 :
2920 12 : assert(nvme_ctrlr->reset_start_tsc == 0);
2921 12 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2922 :
2923 12 : return 0;
2924 : }
2925 :
2926 : static int
2927 16 : bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2928 : {
2929 : int rc;
2930 :
2931 16 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2932 16 : rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, false);
2933 16 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2934 :
2935 16 : if (rc == 0) {
2936 11 : spdk_thread_send_msg(nvme_ctrlr->thread, _bdev_nvme_reset_ctrlr, nvme_ctrlr);
2937 5 : } else if (rc == -EALREADY) {
2938 0 : rc = 0;
2939 : }
2940 :
2941 16 : return rc;
2942 : }
2943 :
2944 : static int bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks,
2945 : uint64_t num_blocks);
2946 :
2947 : static int bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks,
2948 : uint64_t num_blocks);
2949 :
2950 : static int bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks,
2951 : uint64_t src_offset_blocks,
2952 : uint64_t num_blocks);
2953 :
2954 : static void
2955 1 : bdev_nvme_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
2956 : bool success)
2957 : {
2958 1 : struct nvme_bdev_io *bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
2959 : int ret;
2960 :
2961 1 : if (!success) {
2962 0 : ret = -EINVAL;
2963 0 : goto exit;
2964 : }
2965 :
2966 1 : if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
2967 0 : ret = -ENXIO;
2968 0 : goto exit;
2969 : }
2970 :
2971 1 : ret = bdev_nvme_readv(bio,
2972 : bdev_io->u.bdev.iovs,
2973 : bdev_io->u.bdev.iovcnt,
2974 : bdev_io->u.bdev.md_buf,
2975 : bdev_io->u.bdev.num_blocks,
2976 : bdev_io->u.bdev.offset_blocks,
2977 : bdev_io->u.bdev.dif_check_flags,
2978 : bdev_io->u.bdev.memory_domain,
2979 : bdev_io->u.bdev.memory_domain_ctx,
2980 : bdev_io->u.bdev.accel_sequence);
2981 :
2982 1 : exit:
2983 1 : if (spdk_unlikely(ret != 0)) {
2984 0 : bdev_nvme_io_complete(bio, ret);
2985 : }
2986 1 : }
2987 :
2988 : static inline void
2989 51 : _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
2990 : {
2991 51 : struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
2992 51 : struct spdk_bdev *bdev = bdev_io->bdev;
2993 : struct nvme_bdev_io *nbdev_io_to_abort;
2994 51 : int rc = 0;
2995 :
2996 51 : switch (bdev_io->type) {
2997 3 : case SPDK_BDEV_IO_TYPE_READ:
2998 3 : if (bdev_io->u.bdev.iovs && bdev_io->u.bdev.iovs[0].iov_base) {
2999 :
3000 2 : rc = bdev_nvme_readv(nbdev_io,
3001 : bdev_io->u.bdev.iovs,
3002 : bdev_io->u.bdev.iovcnt,
3003 : bdev_io->u.bdev.md_buf,
3004 : bdev_io->u.bdev.num_blocks,
3005 : bdev_io->u.bdev.offset_blocks,
3006 : bdev_io->u.bdev.dif_check_flags,
3007 : bdev_io->u.bdev.memory_domain,
3008 : bdev_io->u.bdev.memory_domain_ctx,
3009 : bdev_io->u.bdev.accel_sequence);
3010 : } else {
3011 1 : spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb,
3012 1 : bdev_io->u.bdev.num_blocks * bdev->blocklen);
3013 1 : rc = 0;
3014 : }
3015 3 : break;
3016 25 : case SPDK_BDEV_IO_TYPE_WRITE:
3017 25 : rc = bdev_nvme_writev(nbdev_io,
3018 : bdev_io->u.bdev.iovs,
3019 : bdev_io->u.bdev.iovcnt,
3020 : bdev_io->u.bdev.md_buf,
3021 : bdev_io->u.bdev.num_blocks,
3022 : bdev_io->u.bdev.offset_blocks,
3023 : bdev_io->u.bdev.dif_check_flags,
3024 : bdev_io->u.bdev.memory_domain,
3025 : bdev_io->u.bdev.memory_domain_ctx,
3026 : bdev_io->u.bdev.accel_sequence,
3027 : bdev_io->u.bdev.nvme_cdw12,
3028 : bdev_io->u.bdev.nvme_cdw13);
3029 25 : break;
3030 1 : case SPDK_BDEV_IO_TYPE_COMPARE:
3031 1 : rc = bdev_nvme_comparev(nbdev_io,
3032 : bdev_io->u.bdev.iovs,
3033 : bdev_io->u.bdev.iovcnt,
3034 : bdev_io->u.bdev.md_buf,
3035 : bdev_io->u.bdev.num_blocks,
3036 : bdev_io->u.bdev.offset_blocks,
3037 : bdev_io->u.bdev.dif_check_flags);
3038 1 : break;
3039 2 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
3040 2 : rc = bdev_nvme_comparev_and_writev(nbdev_io,
3041 : bdev_io->u.bdev.iovs,
3042 : bdev_io->u.bdev.iovcnt,
3043 : bdev_io->u.bdev.fused_iovs,
3044 : bdev_io->u.bdev.fused_iovcnt,
3045 : bdev_io->u.bdev.md_buf,
3046 : bdev_io->u.bdev.num_blocks,
3047 : bdev_io->u.bdev.offset_blocks,
3048 : bdev_io->u.bdev.dif_check_flags);
3049 2 : break;
3050 1 : case SPDK_BDEV_IO_TYPE_UNMAP:
3051 1 : rc = bdev_nvme_unmap(nbdev_io,
3052 : bdev_io->u.bdev.offset_blocks,
3053 : bdev_io->u.bdev.num_blocks);
3054 1 : break;
3055 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3056 0 : rc = bdev_nvme_write_zeroes(nbdev_io,
3057 : bdev_io->u.bdev.offset_blocks,
3058 : bdev_io->u.bdev.num_blocks);
3059 0 : break;
3060 7 : case SPDK_BDEV_IO_TYPE_RESET:
3061 7 : nbdev_io->io_path = NULL;
3062 7 : bdev_nvme_reset_io(nbdev_ch, nbdev_io);
3063 7 : return;
3064 :
3065 1 : case SPDK_BDEV_IO_TYPE_FLUSH:
3066 1 : bdev_nvme_io_complete(nbdev_io, 0);
3067 1 : return;
3068 :
3069 0 : case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
3070 0 : rc = bdev_nvme_zone_appendv(nbdev_io,
3071 : bdev_io->u.bdev.iovs,
3072 : bdev_io->u.bdev.iovcnt,
3073 : bdev_io->u.bdev.md_buf,
3074 : bdev_io->u.bdev.num_blocks,
3075 : bdev_io->u.bdev.offset_blocks,
3076 : bdev_io->u.bdev.dif_check_flags);
3077 0 : break;
3078 0 : case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
3079 0 : rc = bdev_nvme_get_zone_info(nbdev_io,
3080 : bdev_io->u.zone_mgmt.zone_id,
3081 : bdev_io->u.zone_mgmt.num_zones,
3082 0 : bdev_io->u.zone_mgmt.buf);
3083 0 : break;
3084 0 : case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
3085 0 : rc = bdev_nvme_zone_management(nbdev_io,
3086 : bdev_io->u.zone_mgmt.zone_id,
3087 : bdev_io->u.zone_mgmt.zone_action);
3088 0 : break;
3089 5 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3090 5 : nbdev_io->io_path = NULL;
3091 5 : bdev_nvme_admin_passthru(nbdev_ch,
3092 : nbdev_io,
3093 : &bdev_io->u.nvme_passthru.cmd,
3094 : bdev_io->u.nvme_passthru.buf,
3095 : bdev_io->u.nvme_passthru.nbytes);
3096 5 : return;
3097 :
3098 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3099 0 : rc = bdev_nvme_io_passthru(nbdev_io,
3100 : &bdev_io->u.nvme_passthru.cmd,
3101 : bdev_io->u.nvme_passthru.buf,
3102 : bdev_io->u.nvme_passthru.nbytes);
3103 0 : break;
3104 0 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3105 0 : rc = bdev_nvme_io_passthru_md(nbdev_io,
3106 : &bdev_io->u.nvme_passthru.cmd,
3107 : bdev_io->u.nvme_passthru.buf,
3108 : bdev_io->u.nvme_passthru.nbytes,
3109 : bdev_io->u.nvme_passthru.md_buf,
3110 : bdev_io->u.nvme_passthru.md_len);
3111 0 : break;
3112 0 : case SPDK_BDEV_IO_TYPE_NVME_IOV_MD:
3113 0 : rc = bdev_nvme_iov_passthru_md(nbdev_io,
3114 : &bdev_io->u.nvme_passthru.cmd,
3115 : bdev_io->u.nvme_passthru.iovs,
3116 : bdev_io->u.nvme_passthru.iovcnt,
3117 : bdev_io->u.nvme_passthru.nbytes,
3118 : bdev_io->u.nvme_passthru.md_buf,
3119 : bdev_io->u.nvme_passthru.md_len);
3120 0 : break;
3121 6 : case SPDK_BDEV_IO_TYPE_ABORT:
3122 6 : nbdev_io->io_path = NULL;
3123 6 : nbdev_io_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx;
3124 6 : bdev_nvme_abort(nbdev_ch,
3125 : nbdev_io,
3126 : nbdev_io_to_abort);
3127 6 : return;
3128 :
3129 0 : case SPDK_BDEV_IO_TYPE_COPY:
3130 0 : rc = bdev_nvme_copy(nbdev_io,
3131 : bdev_io->u.bdev.offset_blocks,
3132 : bdev_io->u.bdev.copy.src_offset_blocks,
3133 : bdev_io->u.bdev.num_blocks);
3134 0 : break;
3135 0 : default:
3136 0 : rc = -EINVAL;
3137 0 : break;
3138 : }
3139 :
3140 32 : if (spdk_unlikely(rc != 0)) {
3141 0 : bdev_nvme_io_complete(nbdev_io, rc);
3142 : }
3143 : }
3144 :
3145 : static void
3146 58 : bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
3147 : {
3148 58 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
3149 58 : struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3150 :
3151 58 : if (spdk_likely(nbdev_io->submit_tsc == 0)) {
3152 58 : nbdev_io->submit_tsc = spdk_bdev_io_get_submit_tsc(bdev_io);
3153 : } else {
3154 : /* There are cases where submit_tsc != 0, i.e. retry I/O.
3155 : * We need to update submit_tsc here.
3156 : */
3157 0 : nbdev_io->submit_tsc = spdk_get_ticks();
3158 : }
3159 :
3160 58 : spdk_trace_record(TRACE_BDEV_NVME_IO_START, 0, 0, (uintptr_t)nbdev_io, (uintptr_t)bdev_io);
3161 58 : nbdev_io->io_path = bdev_nvme_find_io_path(nbdev_ch);
3162 58 : if (spdk_unlikely(!nbdev_io->io_path)) {
3163 11 : if (!bdev_nvme_io_type_is_admin(bdev_io->type)) {
3164 10 : bdev_nvme_io_complete(nbdev_io, -ENXIO);
3165 10 : return;
3166 : }
3167 :
3168 : /* Admin commands do not use the optimal I/O path.
3169 : * Simply fall through even if it is not found.
3170 : */
3171 : }
3172 :
3173 48 : _bdev_nvme_submit_request(nbdev_ch, bdev_io);
3174 : }
3175 :
3176 : static bool
3177 0 : bdev_nvme_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
3178 : {
3179 0 : struct nvme_bdev *nbdev = ctx;
3180 : struct nvme_ns *nvme_ns;
3181 : struct spdk_nvme_ns *ns;
3182 : struct spdk_nvme_ctrlr *ctrlr;
3183 : const struct spdk_nvme_ctrlr_data *cdata;
3184 :
3185 0 : nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
3186 0 : assert(nvme_ns != NULL);
3187 0 : ns = nvme_ns->ns;
3188 0 : if (ns == NULL) {
3189 0 : return false;
3190 : }
3191 :
3192 0 : ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3193 :
3194 0 : switch (io_type) {
3195 0 : case SPDK_BDEV_IO_TYPE_READ:
3196 : case SPDK_BDEV_IO_TYPE_WRITE:
3197 : case SPDK_BDEV_IO_TYPE_RESET:
3198 : case SPDK_BDEV_IO_TYPE_FLUSH:
3199 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3200 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3201 : case SPDK_BDEV_IO_TYPE_ABORT:
3202 0 : return true;
3203 :
3204 0 : case SPDK_BDEV_IO_TYPE_COMPARE:
3205 0 : return spdk_nvme_ns_supports_compare(ns);
3206 :
3207 0 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3208 0 : return spdk_nvme_ns_get_md_size(ns) ? true : false;
3209 :
3210 0 : case SPDK_BDEV_IO_TYPE_UNMAP:
3211 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3212 0 : return cdata->oncs.dsm;
3213 :
3214 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3215 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3216 0 : return cdata->oncs.write_zeroes;
3217 :
3218 0 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
3219 0 : if (spdk_nvme_ctrlr_get_flags(ctrlr) &
3220 : SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED) {
3221 0 : return true;
3222 : }
3223 0 : return false;
3224 :
3225 0 : case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
3226 : case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
3227 0 : return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS;
3228 :
3229 0 : case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
3230 0 : return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS &&
3231 0 : spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ZONE_APPEND_SUPPORTED;
3232 :
3233 0 : case SPDK_BDEV_IO_TYPE_COPY:
3234 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3235 0 : return cdata->oncs.copy;
3236 :
3237 0 : default:
3238 0 : return false;
3239 : }
3240 : }
3241 :
3242 : static int
3243 57 : nvme_qpair_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ctrlr_channel *ctrlr_ch)
3244 : {
3245 : struct nvme_qpair *nvme_qpair;
3246 : struct spdk_io_channel *pg_ch;
3247 : int rc;
3248 :
3249 57 : nvme_qpair = calloc(1, sizeof(*nvme_qpair));
3250 57 : if (!nvme_qpair) {
3251 0 : SPDK_ERRLOG("Failed to alloc nvme_qpair.\n");
3252 0 : return -1;
3253 : }
3254 :
3255 57 : TAILQ_INIT(&nvme_qpair->io_path_list);
3256 :
3257 57 : nvme_qpair->ctrlr = nvme_ctrlr;
3258 57 : nvme_qpair->ctrlr_ch = ctrlr_ch;
3259 :
3260 57 : pg_ch = spdk_get_io_channel(&g_nvme_bdev_ctrlrs);
3261 57 : if (!pg_ch) {
3262 0 : free(nvme_qpair);
3263 0 : return -1;
3264 : }
3265 :
3266 57 : nvme_qpair->group = spdk_io_channel_get_ctx(pg_ch);
3267 :
3268 : #ifdef SPDK_CONFIG_VTUNE
3269 : nvme_qpair->group->collect_spin_stat = true;
3270 : #else
3271 57 : nvme_qpair->group->collect_spin_stat = false;
3272 : #endif
3273 :
3274 57 : if (!nvme_ctrlr->disabled) {
3275 : /* If a nvme_ctrlr is disabled, don't try to create qpair for it. Qpair will
3276 : * be created when it's enabled.
3277 : */
3278 57 : rc = bdev_nvme_create_qpair(nvme_qpair);
3279 57 : if (rc != 0) {
3280 : /* nvme_ctrlr can't create IO qpair if connection is down.
3281 : * If reconnect_delay_sec is non-zero, creating IO qpair is retried
3282 : * after reconnect_delay_sec seconds. If bdev_retry_count is non-zero,
3283 : * submitted IO will be queued until IO qpair is successfully created.
3284 : *
3285 : * Hence, if both are satisfied, ignore the failure.
3286 : */
3287 0 : if (nvme_ctrlr->opts.reconnect_delay_sec == 0 || g_opts.bdev_retry_count == 0) {
3288 0 : spdk_put_io_channel(pg_ch);
3289 0 : free(nvme_qpair);
3290 0 : return rc;
3291 : }
3292 : }
3293 : }
3294 :
3295 57 : TAILQ_INSERT_TAIL(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);
3296 :
3297 57 : ctrlr_ch->qpair = nvme_qpair;
3298 :
3299 57 : pthread_mutex_lock(&nvme_qpair->ctrlr->mutex);
3300 57 : nvme_qpair->ctrlr->ref++;
3301 57 : pthread_mutex_unlock(&nvme_qpair->ctrlr->mutex);
3302 :
3303 57 : return 0;
3304 : }
3305 :
3306 : static int
3307 57 : bdev_nvme_create_ctrlr_channel_cb(void *io_device, void *ctx_buf)
3308 : {
3309 57 : struct nvme_ctrlr *nvme_ctrlr = io_device;
3310 57 : struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
3311 :
3312 57 : TAILQ_INIT(&ctrlr_ch->pending_resets);
3313 :
3314 57 : return nvme_qpair_create(nvme_ctrlr, ctrlr_ch);
3315 : }
3316 :
3317 : static void
3318 57 : nvme_qpair_delete(struct nvme_qpair *nvme_qpair)
3319 : {
3320 : struct nvme_io_path *io_path, *next;
3321 :
3322 57 : assert(nvme_qpair->group != NULL);
3323 :
3324 92 : TAILQ_FOREACH_SAFE(io_path, &nvme_qpair->io_path_list, tailq, next) {
3325 35 : TAILQ_REMOVE(&nvme_qpair->io_path_list, io_path, tailq);
3326 35 : nvme_io_path_free(io_path);
3327 : }
3328 :
3329 57 : TAILQ_REMOVE(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);
3330 :
3331 57 : spdk_put_io_channel(spdk_io_channel_from_ctx(nvme_qpair->group));
3332 :
3333 57 : nvme_ctrlr_release(nvme_qpair->ctrlr);
3334 :
3335 57 : free(nvme_qpair);
3336 57 : }
3337 :
3338 : static void
3339 57 : bdev_nvme_destroy_ctrlr_channel_cb(void *io_device, void *ctx_buf)
3340 : {
3341 57 : struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
3342 : struct nvme_qpair *nvme_qpair;
3343 :
3344 57 : nvme_qpair = ctrlr_ch->qpair;
3345 57 : assert(nvme_qpair != NULL);
3346 :
3347 57 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
3348 :
3349 57 : if (nvme_qpair->qpair != NULL) {
3350 43 : if (ctrlr_ch->reset_iter == NULL) {
3351 43 : spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);
3352 : } else {
3353 : /* Skip current ctrlr_channel in a full reset sequence because
3354 : * it is being deleted now. The qpair is already being disconnected.
3355 : * We do not have to restart disconnecting it.
3356 : */
3357 0 : spdk_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
3358 : }
3359 :
3360 : /* We cannot release a reference to the poll group now.
3361 : * The qpair may be disconnected asynchronously later.
3362 : * We need to poll it until it is actually disconnected.
3363 : * Just detach the qpair from the deleting ctrlr_channel.
3364 : */
3365 43 : nvme_qpair->ctrlr_ch = NULL;
3366 : } else {
3367 14 : assert(ctrlr_ch->reset_iter == NULL);
3368 :
3369 14 : nvme_qpair_delete(nvme_qpair);
3370 : }
3371 57 : }
3372 :
3373 : static inline struct spdk_io_channel *
3374 0 : bdev_nvme_get_accel_channel(struct nvme_poll_group *group)
3375 : {
3376 0 : if (spdk_unlikely(!group->accel_channel)) {
3377 0 : group->accel_channel = spdk_accel_get_io_channel();
3378 0 : if (!group->accel_channel) {
3379 0 : SPDK_ERRLOG("Cannot get the accel_channel for bdev nvme polling group=%p\n",
3380 : group);
3381 0 : return NULL;
3382 : }
3383 : }
3384 :
3385 0 : return group->accel_channel;
3386 : }
3387 :
3388 : static void
3389 0 : bdev_nvme_submit_accel_crc32c(void *ctx, uint32_t *dst, struct iovec *iov,
3390 : uint32_t iov_cnt, uint32_t seed,
3391 : spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
3392 : {
3393 : struct spdk_io_channel *accel_ch;
3394 0 : struct nvme_poll_group *group = ctx;
3395 : int rc;
3396 :
3397 0 : assert(cb_fn != NULL);
3398 :
3399 0 : accel_ch = bdev_nvme_get_accel_channel(group);
3400 0 : if (spdk_unlikely(accel_ch == NULL)) {
3401 0 : cb_fn(cb_arg, -ENOMEM);
3402 0 : return;
3403 : }
3404 :
3405 0 : rc = spdk_accel_submit_crc32cv(accel_ch, dst, iov, iov_cnt, seed, cb_fn, cb_arg);
3406 0 : if (rc) {
3407 : /* For the two cases, spdk_accel_submit_crc32cv does not call the user's cb_fn */
3408 0 : if (rc == -ENOMEM || rc == -EINVAL) {
3409 0 : cb_fn(cb_arg, rc);
3410 : }
3411 0 : SPDK_ERRLOG("Cannot complete the accelerated crc32c operation with iov=%p\n", iov);
3412 : }
3413 : }
3414 :
3415 : static void
3416 0 : bdev_nvme_finish_sequence(void *seq, spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
3417 : {
3418 0 : spdk_accel_sequence_finish(seq, cb_fn, cb_arg);
3419 0 : }
3420 :
3421 : static void
3422 0 : bdev_nvme_abort_sequence(void *seq)
3423 : {
3424 0 : spdk_accel_sequence_abort(seq);
3425 0 : }
3426 :
3427 : static void
3428 0 : bdev_nvme_reverse_sequence(void *seq)
3429 : {
3430 0 : spdk_accel_sequence_reverse(seq);
3431 0 : }
3432 :
3433 : static int
3434 0 : bdev_nvme_append_crc32c(void *ctx, void **seq, uint32_t *dst, struct iovec *iovs, uint32_t iovcnt,
3435 : struct spdk_memory_domain *domain, void *domain_ctx, uint32_t seed,
3436 : spdk_nvme_accel_step_cb cb_fn, void *cb_arg)
3437 : {
3438 : struct spdk_io_channel *ch;
3439 0 : struct nvme_poll_group *group = ctx;
3440 :
3441 0 : ch = bdev_nvme_get_accel_channel(group);
3442 0 : if (spdk_unlikely(ch == NULL)) {
3443 0 : return -ENOMEM;
3444 : }
3445 :
3446 0 : return spdk_accel_append_crc32c((struct spdk_accel_sequence **)seq, ch, dst, iovs, iovcnt,
3447 : domain, domain_ctx, seed, cb_fn, cb_arg);
3448 : }
3449 :
3450 : static struct spdk_nvme_accel_fn_table g_bdev_nvme_accel_fn_table = {
3451 : .table_size = sizeof(struct spdk_nvme_accel_fn_table),
3452 : .submit_accel_crc32c = bdev_nvme_submit_accel_crc32c,
3453 : .append_crc32c = bdev_nvme_append_crc32c,
3454 : .finish_sequence = bdev_nvme_finish_sequence,
3455 : .reverse_sequence = bdev_nvme_reverse_sequence,
3456 : .abort_sequence = bdev_nvme_abort_sequence,
3457 : };
3458 :
3459 : static int
3460 42 : bdev_nvme_create_poll_group_cb(void *io_device, void *ctx_buf)
3461 : {
3462 42 : struct nvme_poll_group *group = ctx_buf;
3463 :
3464 42 : TAILQ_INIT(&group->qpair_list);
3465 :
3466 42 : group->group = spdk_nvme_poll_group_create(group, &g_bdev_nvme_accel_fn_table);
3467 42 : if (group->group == NULL) {
3468 0 : return -1;
3469 : }
3470 :
3471 42 : group->poller = SPDK_POLLER_REGISTER(bdev_nvme_poll, group, g_opts.nvme_ioq_poll_period_us);
3472 :
3473 42 : if (group->poller == NULL) {
3474 0 : spdk_nvme_poll_group_destroy(group->group);
3475 0 : return -1;
3476 : }
3477 :
3478 42 : return 0;
3479 : }
3480 :
3481 : static void
3482 42 : bdev_nvme_destroy_poll_group_cb(void *io_device, void *ctx_buf)
3483 : {
3484 42 : struct nvme_poll_group *group = ctx_buf;
3485 :
3486 42 : assert(TAILQ_EMPTY(&group->qpair_list));
3487 :
3488 42 : if (group->accel_channel) {
3489 0 : spdk_put_io_channel(group->accel_channel);
3490 : }
3491 :
3492 42 : spdk_poller_unregister(&group->poller);
3493 42 : if (spdk_nvme_poll_group_destroy(group->group)) {
3494 0 : SPDK_ERRLOG("Unable to destroy a poll group for the NVMe bdev module.\n");
3495 0 : assert(false);
3496 : }
3497 42 : }
3498 :
3499 : static struct spdk_io_channel *
3500 0 : bdev_nvme_get_io_channel(void *ctx)
3501 : {
3502 0 : struct nvme_bdev *nvme_bdev = ctx;
3503 :
3504 0 : return spdk_get_io_channel(nvme_bdev);
3505 : }
3506 :
3507 : static void *
3508 0 : bdev_nvme_get_module_ctx(void *ctx)
3509 : {
3510 0 : struct nvme_bdev *nvme_bdev = ctx;
3511 : struct nvme_ns *nvme_ns;
3512 :
3513 0 : if (!nvme_bdev || nvme_bdev->disk.module != &nvme_if) {
3514 0 : return NULL;
3515 : }
3516 :
3517 0 : nvme_ns = TAILQ_FIRST(&nvme_bdev->nvme_ns_list);
3518 0 : if (!nvme_ns) {
3519 0 : return NULL;
3520 : }
3521 :
3522 0 : return nvme_ns->ns;
3523 : }
3524 :
3525 : static const char *
3526 0 : _nvme_ana_state_str(enum spdk_nvme_ana_state ana_state)
3527 : {
3528 0 : switch (ana_state) {
3529 0 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
3530 0 : return "optimized";
3531 0 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
3532 0 : return "non_optimized";
3533 0 : case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3534 0 : return "inaccessible";
3535 0 : case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3536 0 : return "persistent_loss";
3537 0 : case SPDK_NVME_ANA_CHANGE_STATE:
3538 0 : return "change";
3539 0 : default:
3540 0 : return NULL;
3541 : }
3542 : }
3543 :
3544 : static int
3545 8 : bdev_nvme_get_memory_domains(void *ctx, struct spdk_memory_domain **domains, int array_size)
3546 : {
3547 8 : struct spdk_memory_domain **_domains = NULL;
3548 8 : struct nvme_bdev *nbdev = ctx;
3549 : struct nvme_ns *nvme_ns;
3550 8 : int i = 0, _array_size = array_size;
3551 8 : int rc = 0;
3552 :
3553 22 : TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
3554 14 : if (domains && array_size >= i) {
3555 11 : _domains = &domains[i];
3556 : } else {
3557 3 : _domains = NULL;
3558 : }
3559 14 : rc = spdk_nvme_ctrlr_get_memory_domains(nvme_ns->ctrlr->ctrlr, _domains, _array_size);
3560 14 : if (rc > 0) {
3561 13 : i += rc;
3562 13 : if (_array_size >= rc) {
3563 9 : _array_size -= rc;
3564 : } else {
3565 4 : _array_size = 0;
3566 : }
3567 1 : } else if (rc < 0) {
3568 0 : return rc;
3569 : }
3570 : }
3571 :
3572 8 : return i;
3573 : }
3574 :
3575 : static const char *
3576 0 : nvme_ctrlr_get_state_str(struct nvme_ctrlr *nvme_ctrlr)
3577 : {
3578 0 : if (nvme_ctrlr->destruct) {
3579 0 : return "deleting";
3580 0 : } else if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
3581 0 : return "failed";
3582 0 : } else if (nvme_ctrlr->resetting) {
3583 0 : return "resetting";
3584 0 : } else if (nvme_ctrlr->reconnect_is_delayed > 0) {
3585 0 : return "reconnect_is_delayed";
3586 0 : } else if (nvme_ctrlr->disabled) {
3587 0 : return "disabled";
3588 : } else {
3589 0 : return "enabled";
3590 : }
3591 : }
3592 :
3593 : void
3594 0 : nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr)
3595 0 : {
3596 : struct spdk_nvme_transport_id *trid;
3597 : const struct spdk_nvme_ctrlr_opts *opts;
3598 : const struct spdk_nvme_ctrlr_data *cdata;
3599 : struct nvme_path_id *path_id;
3600 :
3601 0 : spdk_json_write_object_begin(w);
3602 :
3603 0 : spdk_json_write_named_string(w, "state", nvme_ctrlr_get_state_str(nvme_ctrlr));
3604 :
3605 : #ifdef SPDK_CONFIG_NVME_CUSE
3606 0 : size_t cuse_name_size = 128;
3607 0 : char cuse_name[cuse_name_size];
3608 :
3609 0 : int rc = spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr, cuse_name, &cuse_name_size);
3610 0 : if (rc == 0) {
3611 0 : spdk_json_write_named_string(w, "cuse_device", cuse_name);
3612 : }
3613 : #endif
3614 0 : trid = &nvme_ctrlr->active_path_id->trid;
3615 0 : spdk_json_write_named_object_begin(w, "trid");
3616 0 : nvme_bdev_dump_trid_json(trid, w);
3617 0 : spdk_json_write_object_end(w);
3618 :
3619 0 : path_id = TAILQ_NEXT(nvme_ctrlr->active_path_id, link);
3620 0 : if (path_id != NULL) {
3621 0 : spdk_json_write_named_array_begin(w, "alternate_trids");
3622 : do {
3623 0 : trid = &path_id->trid;
3624 0 : spdk_json_write_object_begin(w);
3625 0 : nvme_bdev_dump_trid_json(trid, w);
3626 0 : spdk_json_write_object_end(w);
3627 :
3628 0 : path_id = TAILQ_NEXT(path_id, link);
3629 0 : } while (path_id != NULL);
3630 0 : spdk_json_write_array_end(w);
3631 : }
3632 :
3633 0 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
3634 0 : spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);
3635 :
3636 0 : opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
3637 0 : spdk_json_write_named_object_begin(w, "host");
3638 0 : spdk_json_write_named_string(w, "nqn", opts->hostnqn);
3639 0 : spdk_json_write_named_string(w, "addr", opts->src_addr);
3640 0 : spdk_json_write_named_string(w, "svcid", opts->src_svcid);
3641 0 : spdk_json_write_object_end(w);
3642 :
3643 0 : spdk_json_write_object_end(w);
3644 0 : }
3645 :
3646 : static void
3647 0 : nvme_namespace_info_json(struct spdk_json_write_ctx *w,
3648 : struct nvme_ns *nvme_ns)
3649 0 : {
3650 : struct spdk_nvme_ns *ns;
3651 : struct spdk_nvme_ctrlr *ctrlr;
3652 : const struct spdk_nvme_ctrlr_data *cdata;
3653 : const struct spdk_nvme_transport_id *trid;
3654 : union spdk_nvme_vs_register vs;
3655 : const struct spdk_nvme_ns_data *nsdata;
3656 0 : char buf[128];
3657 :
3658 0 : ns = nvme_ns->ns;
3659 0 : if (ns == NULL) {
3660 0 : return;
3661 : }
3662 :
3663 0 : ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3664 :
3665 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3666 0 : trid = spdk_nvme_ctrlr_get_transport_id(ctrlr);
3667 0 : vs = spdk_nvme_ctrlr_get_regs_vs(ctrlr);
3668 :
3669 0 : spdk_json_write_object_begin(w);
3670 :
3671 0 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
3672 0 : spdk_json_write_named_string(w, "pci_address", trid->traddr);
3673 : }
3674 :
3675 0 : spdk_json_write_named_object_begin(w, "trid");
3676 :
3677 0 : nvme_bdev_dump_trid_json(trid, w);
3678 :
3679 0 : spdk_json_write_object_end(w);
3680 :
3681 : #ifdef SPDK_CONFIG_NVME_CUSE
3682 0 : size_t cuse_name_size = 128;
3683 0 : char cuse_name[cuse_name_size];
3684 :
3685 0 : int rc = spdk_nvme_cuse_get_ns_name(ctrlr, spdk_nvme_ns_get_id(ns),
3686 : cuse_name, &cuse_name_size);
3687 0 : if (rc == 0) {
3688 0 : spdk_json_write_named_string(w, "cuse_device", cuse_name);
3689 : }
3690 : #endif
3691 :
3692 0 : spdk_json_write_named_object_begin(w, "ctrlr_data");
3693 :
3694 0 : spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);
3695 :
3696 0 : spdk_json_write_named_string_fmt(w, "vendor_id", "0x%04x", cdata->vid);
3697 :
3698 0 : snprintf(buf, sizeof(cdata->mn) + 1, "%s", cdata->mn);
3699 0 : spdk_str_trim(buf);
3700 0 : spdk_json_write_named_string(w, "model_number", buf);
3701 :
3702 0 : snprintf(buf, sizeof(cdata->sn) + 1, "%s", cdata->sn);
3703 0 : spdk_str_trim(buf);
3704 0 : spdk_json_write_named_string(w, "serial_number", buf);
3705 :
3706 0 : snprintf(buf, sizeof(cdata->fr) + 1, "%s", cdata->fr);
3707 0 : spdk_str_trim(buf);
3708 0 : spdk_json_write_named_string(w, "firmware_revision", buf);
3709 :
3710 0 : if (cdata->subnqn[0] != '\0') {
3711 0 : spdk_json_write_named_string(w, "subnqn", cdata->subnqn);
3712 : }
3713 :
3714 0 : spdk_json_write_named_object_begin(w, "oacs");
3715 :
3716 0 : spdk_json_write_named_uint32(w, "security", cdata->oacs.security);
3717 0 : spdk_json_write_named_uint32(w, "format", cdata->oacs.format);
3718 0 : spdk_json_write_named_uint32(w, "firmware", cdata->oacs.firmware);
3719 0 : spdk_json_write_named_uint32(w, "ns_manage", cdata->oacs.ns_manage);
3720 :
3721 0 : spdk_json_write_object_end(w);
3722 :
3723 0 : spdk_json_write_named_bool(w, "multi_ctrlr", cdata->cmic.multi_ctrlr);
3724 0 : spdk_json_write_named_bool(w, "ana_reporting", cdata->cmic.ana_reporting);
3725 :
3726 0 : spdk_json_write_object_end(w);
3727 :
3728 0 : spdk_json_write_named_object_begin(w, "vs");
3729 :
3730 0 : spdk_json_write_name(w, "nvme_version");
3731 0 : if (vs.bits.ter) {
3732 0 : spdk_json_write_string_fmt(w, "%u.%u.%u", vs.bits.mjr, vs.bits.mnr, vs.bits.ter);
3733 : } else {
3734 0 : spdk_json_write_string_fmt(w, "%u.%u", vs.bits.mjr, vs.bits.mnr);
3735 : }
3736 :
3737 0 : spdk_json_write_object_end(w);
3738 :
3739 0 : nsdata = spdk_nvme_ns_get_data(ns);
3740 :
3741 0 : spdk_json_write_named_object_begin(w, "ns_data");
3742 :
3743 0 : spdk_json_write_named_uint32(w, "id", spdk_nvme_ns_get_id(ns));
3744 :
3745 0 : if (cdata->cmic.ana_reporting) {
3746 0 : spdk_json_write_named_string(w, "ana_state",
3747 : _nvme_ana_state_str(nvme_ns->ana_state));
3748 : }
3749 :
3750 0 : spdk_json_write_named_bool(w, "can_share", nsdata->nmic.can_share);
3751 :
3752 0 : spdk_json_write_object_end(w);
3753 :
3754 0 : if (cdata->oacs.security) {
3755 0 : spdk_json_write_named_object_begin(w, "security");
3756 :
3757 0 : spdk_json_write_named_bool(w, "opal", nvme_ns->bdev->opal);
3758 :
3759 0 : spdk_json_write_object_end(w);
3760 : }
3761 :
3762 0 : spdk_json_write_object_end(w);
3763 : }
3764 :
3765 : static const char *
3766 0 : nvme_bdev_get_mp_policy_str(struct nvme_bdev *nbdev)
3767 : {
3768 0 : switch (nbdev->mp_policy) {
3769 0 : case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
3770 0 : return "active_passive";
3771 0 : case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
3772 0 : return "active_active";
3773 0 : default:
3774 0 : assert(false);
3775 : return "invalid";
3776 : }
3777 : }
3778 :
3779 : static const char *
3780 0 : nvme_bdev_get_mp_selector_str(struct nvme_bdev *nbdev)
3781 : {
3782 0 : switch (nbdev->mp_selector) {
3783 0 : case BDEV_NVME_MP_SELECTOR_ROUND_ROBIN:
3784 0 : return "round_robin";
3785 0 : case BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH:
3786 0 : return "queue_depth";
3787 0 : default:
3788 0 : assert(false);
3789 : return "invalid";
3790 : }
3791 : }
3792 :
3793 : static int
3794 0 : bdev_nvme_dump_info_json(void *ctx, struct spdk_json_write_ctx *w)
3795 : {
3796 0 : struct nvme_bdev *nvme_bdev = ctx;
3797 : struct nvme_ns *nvme_ns;
3798 :
3799 0 : pthread_mutex_lock(&nvme_bdev->mutex);
3800 0 : spdk_json_write_named_array_begin(w, "nvme");
3801 0 : TAILQ_FOREACH(nvme_ns, &nvme_bdev->nvme_ns_list, tailq) {
3802 0 : nvme_namespace_info_json(w, nvme_ns);
3803 : }
3804 0 : spdk_json_write_array_end(w);
3805 0 : spdk_json_write_named_string(w, "mp_policy", nvme_bdev_get_mp_policy_str(nvme_bdev));
3806 0 : if (nvme_bdev->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
3807 0 : spdk_json_write_named_string(w, "selector", nvme_bdev_get_mp_selector_str(nvme_bdev));
3808 0 : if (nvme_bdev->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
3809 0 : spdk_json_write_named_uint32(w, "rr_min_io", nvme_bdev->rr_min_io);
3810 : }
3811 : }
3812 0 : pthread_mutex_unlock(&nvme_bdev->mutex);
3813 :
3814 0 : return 0;
3815 : }
3816 :
3817 : static void
3818 0 : bdev_nvme_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
3819 : {
3820 : /* No config per bdev needed */
3821 0 : }
3822 :
3823 : static uint64_t
3824 0 : bdev_nvme_get_spin_time(struct spdk_io_channel *ch)
3825 : {
3826 0 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
3827 : struct nvme_io_path *io_path;
3828 : struct nvme_poll_group *group;
3829 0 : uint64_t spin_time = 0;
3830 :
3831 0 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
3832 0 : group = io_path->qpair->group;
3833 :
3834 0 : if (!group || !group->collect_spin_stat) {
3835 0 : continue;
3836 : }
3837 :
3838 0 : if (group->end_ticks != 0) {
3839 0 : group->spin_ticks += (group->end_ticks - group->start_ticks);
3840 0 : group->end_ticks = 0;
3841 : }
3842 :
3843 0 : spin_time += group->spin_ticks;
3844 0 : group->start_ticks = 0;
3845 0 : group->spin_ticks = 0;
3846 : }
3847 :
3848 0 : return (spin_time * 1000000ULL) / spdk_get_ticks_hz();
3849 : }
3850 :
3851 : static void
3852 0 : bdev_nvme_reset_device_stat(void *ctx)
3853 : {
3854 0 : struct nvme_bdev *nbdev = ctx;
3855 :
3856 0 : if (nbdev->err_stat != NULL) {
3857 0 : memset(nbdev->err_stat, 0, sizeof(struct nvme_error_stat));
3858 : }
3859 0 : }
3860 :
3861 : /* JSON string should be lowercases and underscore delimited string. */
3862 : static void
3863 0 : bdev_nvme_format_nvme_status(char *dst, const char *src)
3864 : {
3865 0 : char tmp[256];
3866 :
3867 0 : spdk_strcpy_replace(dst, 256, src, " - ", "_");
3868 0 : spdk_strcpy_replace(tmp, 256, dst, "-", "_");
3869 0 : spdk_strcpy_replace(dst, 256, tmp, " ", "_");
3870 0 : spdk_strlwr(dst);
3871 0 : }
3872 :
3873 : static void
3874 0 : bdev_nvme_dump_device_stat_json(void *ctx, struct spdk_json_write_ctx *w)
3875 : {
3876 0 : struct nvme_bdev *nbdev = ctx;
3877 0 : struct spdk_nvme_status status = {};
3878 : uint16_t sct, sc;
3879 0 : char status_json[256];
3880 : const char *status_str;
3881 :
3882 0 : if (nbdev->err_stat == NULL) {
3883 0 : return;
3884 : }
3885 :
3886 0 : spdk_json_write_named_object_begin(w, "nvme_error");
3887 :
3888 0 : spdk_json_write_named_object_begin(w, "status_type");
3889 0 : for (sct = 0; sct < 8; sct++) {
3890 0 : if (nbdev->err_stat->status_type[sct] == 0) {
3891 0 : continue;
3892 : }
3893 0 : status.sct = sct;
3894 :
3895 0 : status_str = spdk_nvme_cpl_get_status_type_string(&status);
3896 0 : assert(status_str != NULL);
3897 0 : bdev_nvme_format_nvme_status(status_json, status_str);
3898 :
3899 0 : spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status_type[sct]);
3900 : }
3901 0 : spdk_json_write_object_end(w);
3902 :
3903 0 : spdk_json_write_named_object_begin(w, "status_code");
3904 0 : for (sct = 0; sct < 4; sct++) {
3905 0 : status.sct = sct;
3906 0 : for (sc = 0; sc < 256; sc++) {
3907 0 : if (nbdev->err_stat->status[sct][sc] == 0) {
3908 0 : continue;
3909 : }
3910 0 : status.sc = sc;
3911 :
3912 0 : status_str = spdk_nvme_cpl_get_status_string(&status);
3913 0 : assert(status_str != NULL);
3914 0 : bdev_nvme_format_nvme_status(status_json, status_str);
3915 :
3916 0 : spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status[sct][sc]);
3917 : }
3918 : }
3919 0 : spdk_json_write_object_end(w);
3920 :
3921 0 : spdk_json_write_object_end(w);
3922 : }
3923 :
3924 : static bool
3925 0 : bdev_nvme_accel_sequence_supported(void *ctx, enum spdk_bdev_io_type type)
3926 : {
3927 0 : struct nvme_bdev *nbdev = ctx;
3928 : struct spdk_nvme_ctrlr *ctrlr;
3929 :
3930 0 : if (!g_opts.allow_accel_sequence) {
3931 0 : return false;
3932 : }
3933 :
3934 0 : switch (type) {
3935 0 : case SPDK_BDEV_IO_TYPE_WRITE:
3936 : case SPDK_BDEV_IO_TYPE_READ:
3937 0 : break;
3938 0 : default:
3939 0 : return false;
3940 : }
3941 :
3942 0 : ctrlr = bdev_nvme_get_ctrlr(&nbdev->disk);
3943 0 : assert(ctrlr != NULL);
3944 :
3945 0 : return spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ACCEL_SEQUENCE_SUPPORTED;
3946 : }
3947 :
3948 : static const struct spdk_bdev_fn_table nvmelib_fn_table = {
3949 : .destruct = bdev_nvme_destruct,
3950 : .submit_request = bdev_nvme_submit_request,
3951 : .io_type_supported = bdev_nvme_io_type_supported,
3952 : .get_io_channel = bdev_nvme_get_io_channel,
3953 : .dump_info_json = bdev_nvme_dump_info_json,
3954 : .write_config_json = bdev_nvme_write_config_json,
3955 : .get_spin_time = bdev_nvme_get_spin_time,
3956 : .get_module_ctx = bdev_nvme_get_module_ctx,
3957 : .get_memory_domains = bdev_nvme_get_memory_domains,
3958 : .accel_sequence_supported = bdev_nvme_accel_sequence_supported,
3959 : .reset_device_stat = bdev_nvme_reset_device_stat,
3960 : .dump_device_stat_json = bdev_nvme_dump_device_stat_json,
3961 : };
3962 :
3963 : typedef int (*bdev_nvme_parse_ana_log_page_cb)(
3964 : const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg);
3965 :
3966 : static int
3967 40 : bdev_nvme_parse_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
3968 : bdev_nvme_parse_ana_log_page_cb cb_fn, void *cb_arg)
3969 : {
3970 : struct spdk_nvme_ana_group_descriptor *copied_desc;
3971 : uint8_t *orig_desc;
3972 : uint32_t i, desc_size, copy_len;
3973 40 : int rc = 0;
3974 :
3975 40 : if (nvme_ctrlr->ana_log_page == NULL) {
3976 0 : return -EINVAL;
3977 : }
3978 :
3979 40 : copied_desc = nvme_ctrlr->copied_ana_desc;
3980 :
3981 40 : orig_desc = (uint8_t *)nvme_ctrlr->ana_log_page + sizeof(struct spdk_nvme_ana_page);
3982 40 : copy_len = nvme_ctrlr->max_ana_log_page_size - sizeof(struct spdk_nvme_ana_page);
3983 :
3984 69 : for (i = 0; i < nvme_ctrlr->ana_log_page->num_ana_group_desc; i++) {
3985 65 : memcpy(copied_desc, orig_desc, copy_len);
3986 :
3987 65 : rc = cb_fn(copied_desc, cb_arg);
3988 65 : if (rc != 0) {
3989 36 : break;
3990 : }
3991 :
3992 29 : desc_size = sizeof(struct spdk_nvme_ana_group_descriptor) +
3993 29 : copied_desc->num_of_nsid * sizeof(uint32_t);
3994 29 : orig_desc += desc_size;
3995 29 : copy_len -= desc_size;
3996 : }
3997 :
3998 40 : return rc;
3999 : }
4000 :
4001 : static int
4002 5 : nvme_ns_ana_transition_timedout(void *ctx)
4003 : {
4004 5 : struct nvme_ns *nvme_ns = ctx;
4005 :
4006 5 : spdk_poller_unregister(&nvme_ns->anatt_timer);
4007 5 : nvme_ns->ana_transition_timedout = true;
4008 :
4009 5 : return SPDK_POLLER_BUSY;
4010 : }
4011 :
4012 : static void
4013 45 : _nvme_ns_set_ana_state(struct nvme_ns *nvme_ns,
4014 : const struct spdk_nvme_ana_group_descriptor *desc)
4015 : {
4016 : const struct spdk_nvme_ctrlr_data *cdata;
4017 :
4018 45 : nvme_ns->ana_group_id = desc->ana_group_id;
4019 45 : nvme_ns->ana_state = desc->ana_state;
4020 45 : nvme_ns->ana_state_updating = false;
4021 :
4022 45 : switch (nvme_ns->ana_state) {
4023 38 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
4024 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
4025 38 : nvme_ns->ana_transition_timedout = false;
4026 38 : spdk_poller_unregister(&nvme_ns->anatt_timer);
4027 38 : break;
4028 :
4029 6 : case SPDK_NVME_ANA_INACCESSIBLE_STATE:
4030 : case SPDK_NVME_ANA_CHANGE_STATE:
4031 6 : if (nvme_ns->anatt_timer != NULL) {
4032 1 : break;
4033 : }
4034 :
4035 5 : cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
4036 5 : nvme_ns->anatt_timer = SPDK_POLLER_REGISTER(nvme_ns_ana_transition_timedout,
4037 : nvme_ns,
4038 : cdata->anatt * SPDK_SEC_TO_USEC);
4039 5 : break;
4040 1 : default:
4041 1 : break;
4042 : }
4043 45 : }
4044 :
4045 : static int
4046 59 : nvme_ns_set_ana_state(const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg)
4047 : {
4048 59 : struct nvme_ns *nvme_ns = cb_arg;
4049 : uint32_t i;
4050 :
4051 59 : assert(nvme_ns->ns != NULL);
4052 :
4053 81 : for (i = 0; i < desc->num_of_nsid; i++) {
4054 58 : if (desc->nsid[i] != spdk_nvme_ns_get_id(nvme_ns->ns)) {
4055 22 : continue;
4056 : }
4057 :
4058 36 : _nvme_ns_set_ana_state(nvme_ns, desc);
4059 36 : return 1;
4060 : }
4061 :
4062 23 : return 0;
4063 : }
4064 :
4065 : static int
4066 5 : nvme_generate_uuid(const char *sn, uint32_t nsid, struct spdk_uuid *uuid)
4067 : {
4068 5 : int rc = 0;
4069 5 : struct spdk_uuid new_uuid, namespace_uuid;
4070 5 : char merged_str[SPDK_NVME_CTRLR_SN_LEN + NSID_STR_LEN + 1] = {'\0'};
4071 : /* This namespace UUID was generated using uuid_generate() method. */
4072 5 : const char *namespace_str = {"edaed2de-24bc-4b07-b559-f47ecbe730fd"};
4073 : int size;
4074 :
4075 5 : assert(strlen(sn) <= SPDK_NVME_CTRLR_SN_LEN);
4076 :
4077 5 : spdk_uuid_set_null(&new_uuid);
4078 5 : spdk_uuid_set_null(&namespace_uuid);
4079 :
4080 5 : size = snprintf(merged_str, sizeof(merged_str), "%s%"PRIu32, sn, nsid);
4081 5 : if (size <= 0 || (unsigned long)size >= sizeof(merged_str)) {
4082 0 : return -EINVAL;
4083 : }
4084 :
4085 5 : spdk_uuid_parse(&namespace_uuid, namespace_str);
4086 :
4087 5 : rc = spdk_uuid_generate_sha1(&new_uuid, &namespace_uuid, merged_str, size);
4088 5 : if (rc == 0) {
4089 5 : memcpy(uuid, &new_uuid, sizeof(struct spdk_uuid));
4090 : }
4091 :
4092 5 : return rc;
4093 : }
4094 :
4095 : static int
4096 37 : nvme_disk_create(struct spdk_bdev *disk, const char *base_name,
4097 : struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns,
4098 : uint32_t prchk_flags, void *ctx)
4099 : {
4100 : const struct spdk_uuid *uuid;
4101 : const uint8_t *nguid;
4102 : const struct spdk_nvme_ctrlr_data *cdata;
4103 : const struct spdk_nvme_ns_data *nsdata;
4104 : const struct spdk_nvme_ctrlr_opts *opts;
4105 : enum spdk_nvme_csi csi;
4106 : uint32_t atomic_bs, phys_bs, bs;
4107 37 : char sn_tmp[SPDK_NVME_CTRLR_SN_LEN + 1] = {'\0'};
4108 : int rc;
4109 :
4110 37 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
4111 37 : csi = spdk_nvme_ns_get_csi(ns);
4112 37 : opts = spdk_nvme_ctrlr_get_opts(ctrlr);
4113 :
4114 37 : switch (csi) {
4115 37 : case SPDK_NVME_CSI_NVM:
4116 37 : disk->product_name = "NVMe disk";
4117 37 : break;
4118 0 : case SPDK_NVME_CSI_ZNS:
4119 0 : disk->product_name = "NVMe ZNS disk";
4120 0 : disk->zoned = true;
4121 0 : disk->zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
4122 0 : disk->max_zone_append_size = spdk_nvme_zns_ctrlr_get_max_zone_append_size(ctrlr) /
4123 0 : spdk_nvme_ns_get_extended_sector_size(ns);
4124 0 : disk->max_open_zones = spdk_nvme_zns_ns_get_max_open_zones(ns);
4125 0 : disk->max_active_zones = spdk_nvme_zns_ns_get_max_active_zones(ns);
4126 0 : break;
4127 0 : default:
4128 0 : SPDK_ERRLOG("unsupported CSI: %u\n", csi);
4129 0 : return -ENOTSUP;
4130 : }
4131 :
4132 37 : nguid = spdk_nvme_ns_get_nguid(ns);
4133 37 : if (!nguid) {
4134 37 : uuid = spdk_nvme_ns_get_uuid(ns);
4135 37 : if (uuid) {
4136 12 : disk->uuid = *uuid;
4137 25 : } else if (g_opts.generate_uuids) {
4138 0 : spdk_strcpy_pad(sn_tmp, cdata->sn, SPDK_NVME_CTRLR_SN_LEN, '\0');
4139 0 : rc = nvme_generate_uuid(sn_tmp, spdk_nvme_ns_get_id(ns), &disk->uuid);
4140 0 : if (rc < 0) {
4141 0 : SPDK_ERRLOG("UUID generation failed (%s)\n", spdk_strerror(-rc));
4142 0 : return rc;
4143 : }
4144 : }
4145 : } else {
4146 0 : memcpy(&disk->uuid, nguid, sizeof(disk->uuid));
4147 : }
4148 :
4149 37 : disk->name = spdk_sprintf_alloc("%sn%d", base_name, spdk_nvme_ns_get_id(ns));
4150 37 : if (!disk->name) {
4151 0 : return -ENOMEM;
4152 : }
4153 :
4154 37 : disk->write_cache = 0;
4155 37 : if (cdata->vwc.present) {
4156 : /* Enable if the Volatile Write Cache exists */
4157 0 : disk->write_cache = 1;
4158 : }
4159 37 : if (cdata->oncs.write_zeroes) {
4160 0 : disk->max_write_zeroes = UINT16_MAX + 1;
4161 : }
4162 37 : disk->blocklen = spdk_nvme_ns_get_extended_sector_size(ns);
4163 37 : disk->blockcnt = spdk_nvme_ns_get_num_sectors(ns);
4164 37 : disk->max_segment_size = spdk_nvme_ctrlr_get_max_xfer_size(ctrlr);
4165 37 : disk->ctratt.raw = cdata->ctratt.raw;
4166 : /* NVMe driver will split one request into multiple requests
4167 : * based on MDTS and stripe boundary, the bdev layer will use
4168 : * max_segment_size and max_num_segments to split one big IO
4169 : * into multiple requests, then small request can't run out
4170 : * of NVMe internal requests data structure.
4171 : */
4172 37 : if (opts && opts->io_queue_requests) {
4173 0 : disk->max_num_segments = opts->io_queue_requests / 2;
4174 : }
4175 37 : if (spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_SGL_SUPPORTED) {
4176 : /* The nvme driver will try to split I/O that have too many
4177 : * SGEs, but it doesn't work if that last SGE doesn't end on
4178 : * an aggregate total that is block aligned. The bdev layer has
4179 : * a more robust splitting framework, so use that instead for
4180 : * this case. (See issue #3269.)
4181 : */
4182 0 : uint16_t max_sges = spdk_nvme_ctrlr_get_max_sges(ctrlr);
4183 :
4184 0 : if (disk->max_num_segments == 0) {
4185 0 : disk->max_num_segments = max_sges;
4186 : } else {
4187 0 : disk->max_num_segments = spdk_min(disk->max_num_segments, max_sges);
4188 : }
4189 : }
4190 37 : disk->optimal_io_boundary = spdk_nvme_ns_get_optimal_io_boundary(ns);
4191 :
4192 37 : nsdata = spdk_nvme_ns_get_data(ns);
4193 37 : bs = spdk_nvme_ns_get_sector_size(ns);
4194 37 : atomic_bs = bs;
4195 37 : phys_bs = bs;
4196 37 : if (nsdata->nabo == 0) {
4197 37 : if (nsdata->nsfeat.ns_atomic_write_unit && nsdata->nawupf) {
4198 0 : atomic_bs = bs * (1 + nsdata->nawupf);
4199 : } else {
4200 37 : atomic_bs = bs * (1 + cdata->awupf);
4201 : }
4202 : }
4203 37 : if (nsdata->nsfeat.optperf) {
4204 0 : phys_bs = bs * (1 + nsdata->npwg);
4205 : }
4206 37 : disk->phys_blocklen = spdk_min(phys_bs, atomic_bs);
4207 :
4208 37 : disk->md_len = spdk_nvme_ns_get_md_size(ns);
4209 37 : if (disk->md_len != 0) {
4210 0 : disk->md_interleave = nsdata->flbas.extended;
4211 0 : disk->dif_type = (enum spdk_dif_type)spdk_nvme_ns_get_pi_type(ns);
4212 0 : if (disk->dif_type != SPDK_DIF_DISABLE) {
4213 0 : disk->dif_is_head_of_md = nsdata->dps.md_start;
4214 0 : disk->dif_check_flags = prchk_flags;
4215 : }
4216 : }
4217 :
4218 37 : if (!(spdk_nvme_ctrlr_get_flags(ctrlr) &
4219 : SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED)) {
4220 37 : disk->acwu = 0;
4221 0 : } else if (nsdata->nsfeat.ns_atomic_write_unit) {
4222 0 : disk->acwu = nsdata->nacwu + 1; /* 0-based */
4223 : } else {
4224 0 : disk->acwu = cdata->acwu + 1; /* 0-based */
4225 : }
4226 :
4227 37 : if (cdata->oncs.copy) {
4228 : /* For now bdev interface allows only single segment copy */
4229 0 : disk->max_copy = nsdata->mssrl;
4230 : }
4231 :
4232 37 : disk->ctxt = ctx;
4233 37 : disk->fn_table = &nvmelib_fn_table;
4234 37 : disk->module = &nvme_if;
4235 :
4236 37 : return 0;
4237 : }
4238 :
4239 : static struct nvme_bdev *
4240 37 : nvme_bdev_alloc(void)
4241 : {
4242 : struct nvme_bdev *bdev;
4243 : int rc;
4244 :
4245 37 : bdev = calloc(1, sizeof(*bdev));
4246 37 : if (!bdev) {
4247 0 : SPDK_ERRLOG("bdev calloc() failed\n");
4248 0 : return NULL;
4249 : }
4250 :
4251 37 : if (g_opts.nvme_error_stat) {
4252 0 : bdev->err_stat = calloc(1, sizeof(struct nvme_error_stat));
4253 0 : if (!bdev->err_stat) {
4254 0 : SPDK_ERRLOG("err_stat calloc() failed\n");
4255 0 : free(bdev);
4256 0 : return NULL;
4257 : }
4258 : }
4259 :
4260 37 : rc = pthread_mutex_init(&bdev->mutex, NULL);
4261 37 : if (rc != 0) {
4262 0 : free(bdev->err_stat);
4263 0 : free(bdev);
4264 0 : return NULL;
4265 : }
4266 :
4267 37 : bdev->ref = 1;
4268 37 : bdev->mp_policy = BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE;
4269 37 : bdev->mp_selector = BDEV_NVME_MP_SELECTOR_ROUND_ROBIN;
4270 37 : bdev->rr_min_io = UINT32_MAX;
4271 37 : TAILQ_INIT(&bdev->nvme_ns_list);
4272 :
4273 37 : return bdev;
4274 : }
4275 :
4276 : static int
4277 37 : nvme_bdev_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4278 : {
4279 : struct nvme_bdev *bdev;
4280 37 : struct nvme_bdev_ctrlr *nbdev_ctrlr = nvme_ctrlr->nbdev_ctrlr;
4281 : int rc;
4282 :
4283 37 : bdev = nvme_bdev_alloc();
4284 37 : if (bdev == NULL) {
4285 0 : SPDK_ERRLOG("Failed to allocate NVMe bdev\n");
4286 0 : return -ENOMEM;
4287 : }
4288 :
4289 37 : bdev->opal = nvme_ctrlr->opal_dev != NULL;
4290 :
4291 37 : rc = nvme_disk_create(&bdev->disk, nbdev_ctrlr->name, nvme_ctrlr->ctrlr,
4292 : nvme_ns->ns, nvme_ctrlr->opts.prchk_flags, bdev);
4293 37 : if (rc != 0) {
4294 0 : SPDK_ERRLOG("Failed to create NVMe disk\n");
4295 0 : nvme_bdev_free(bdev);
4296 0 : return rc;
4297 : }
4298 :
4299 37 : spdk_io_device_register(bdev,
4300 : bdev_nvme_create_bdev_channel_cb,
4301 : bdev_nvme_destroy_bdev_channel_cb,
4302 : sizeof(struct nvme_bdev_channel),
4303 37 : bdev->disk.name);
4304 :
4305 37 : nvme_ns->bdev = bdev;
4306 37 : bdev->nsid = nvme_ns->id;
4307 37 : TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
4308 :
4309 37 : bdev->nbdev_ctrlr = nbdev_ctrlr;
4310 37 : TAILQ_INSERT_TAIL(&nbdev_ctrlr->bdevs, bdev, tailq);
4311 :
4312 37 : rc = spdk_bdev_register(&bdev->disk);
4313 37 : if (rc != 0) {
4314 1 : SPDK_ERRLOG("spdk_bdev_register() failed\n");
4315 1 : spdk_io_device_unregister(bdev, NULL);
4316 1 : nvme_ns->bdev = NULL;
4317 1 : TAILQ_REMOVE(&nbdev_ctrlr->bdevs, bdev, tailq);
4318 1 : nvme_bdev_free(bdev);
4319 1 : return rc;
4320 : }
4321 :
4322 36 : return 0;
4323 : }
4324 :
4325 : static bool
4326 23 : bdev_nvme_compare_ns(struct spdk_nvme_ns *ns1, struct spdk_nvme_ns *ns2)
4327 : {
4328 : const struct spdk_nvme_ns_data *nsdata1, *nsdata2;
4329 : const struct spdk_uuid *uuid1, *uuid2;
4330 :
4331 23 : nsdata1 = spdk_nvme_ns_get_data(ns1);
4332 23 : nsdata2 = spdk_nvme_ns_get_data(ns2);
4333 23 : uuid1 = spdk_nvme_ns_get_uuid(ns1);
4334 23 : uuid2 = spdk_nvme_ns_get_uuid(ns2);
4335 :
4336 45 : return memcmp(nsdata1->nguid, nsdata2->nguid, sizeof(nsdata1->nguid)) == 0 &&
4337 22 : nsdata1->eui64 == nsdata2->eui64 &&
4338 21 : ((uuid1 == NULL && uuid2 == NULL) ||
4339 59 : (uuid1 != NULL && uuid2 != NULL && spdk_uuid_compare(uuid1, uuid2) == 0)) &&
4340 18 : spdk_nvme_ns_get_csi(ns1) == spdk_nvme_ns_get_csi(ns2);
4341 : }
4342 :
4343 : static bool
4344 0 : hotplug_probe_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
4345 : struct spdk_nvme_ctrlr_opts *opts)
4346 : {
4347 : struct nvme_probe_skip_entry *entry;
4348 :
4349 0 : TAILQ_FOREACH(entry, &g_skipped_nvme_ctrlrs, tailq) {
4350 0 : if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
4351 0 : return false;
4352 : }
4353 : }
4354 :
4355 0 : opts->arbitration_burst = (uint8_t)g_opts.arbitration_burst;
4356 0 : opts->low_priority_weight = (uint8_t)g_opts.low_priority_weight;
4357 0 : opts->medium_priority_weight = (uint8_t)g_opts.medium_priority_weight;
4358 0 : opts->high_priority_weight = (uint8_t)g_opts.high_priority_weight;
4359 0 : opts->disable_read_ana_log_page = true;
4360 :
4361 0 : SPDK_DEBUGLOG(bdev_nvme, "Attaching to %s\n", trid->traddr);
4362 :
4363 0 : return true;
4364 : }
4365 :
4366 : static void
4367 0 : nvme_abort_cpl(void *ctx, const struct spdk_nvme_cpl *cpl)
4368 : {
4369 0 : struct nvme_ctrlr *nvme_ctrlr = ctx;
4370 :
4371 0 : if (spdk_nvme_cpl_is_error(cpl)) {
4372 0 : SPDK_WARNLOG("Abort failed. Resetting controller. sc is %u, sct is %u.\n", cpl->status.sc,
4373 : cpl->status.sct);
4374 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4375 0 : } else if (cpl->cdw0 & 0x1) {
4376 0 : SPDK_WARNLOG("Specified command could not be aborted.\n");
4377 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4378 : }
4379 0 : }
4380 :
4381 : static void
4382 0 : timeout_cb(void *cb_arg, struct spdk_nvme_ctrlr *ctrlr,
4383 : struct spdk_nvme_qpair *qpair, uint16_t cid)
4384 : {
4385 0 : struct nvme_ctrlr *nvme_ctrlr = cb_arg;
4386 : union spdk_nvme_csts_register csts;
4387 : int rc;
4388 :
4389 0 : assert(nvme_ctrlr->ctrlr == ctrlr);
4390 :
4391 0 : SPDK_WARNLOG("Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n", ctrlr, qpair, cid);
4392 :
4393 : /* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O
4394 : * queue. (Note: qpair == NULL when there's an admin cmd timeout.) Otherwise we
4395 : * would submit another fabrics cmd on the admin queue to read CSTS and check for its
4396 : * completion recursively.
4397 : */
4398 0 : if (nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE || qpair != NULL) {
4399 0 : csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
4400 0 : if (csts.bits.cfs) {
4401 0 : SPDK_ERRLOG("Controller Fatal Status, reset required\n");
4402 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4403 0 : return;
4404 : }
4405 : }
4406 :
4407 0 : switch (g_opts.action_on_timeout) {
4408 0 : case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT:
4409 0 : if (qpair) {
4410 : /* Don't send abort to ctrlr when ctrlr is not available. */
4411 0 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4412 0 : if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
4413 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4414 0 : SPDK_NOTICELOG("Quit abort. Ctrlr is not available.\n");
4415 0 : return;
4416 : }
4417 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4418 :
4419 0 : rc = spdk_nvme_ctrlr_cmd_abort(ctrlr, qpair, cid,
4420 : nvme_abort_cpl, nvme_ctrlr);
4421 0 : if (rc == 0) {
4422 0 : return;
4423 : }
4424 :
4425 0 : SPDK_ERRLOG("Unable to send abort. Resetting, rc is %d.\n", rc);
4426 : }
4427 :
4428 : /* FALLTHROUGH */
4429 : case SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET:
4430 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4431 0 : break;
4432 0 : case SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE:
4433 0 : SPDK_DEBUGLOG(bdev_nvme, "No action for nvme controller timeout.\n");
4434 0 : break;
4435 0 : default:
4436 0 : SPDK_ERRLOG("An invalid timeout action value is found.\n");
4437 0 : break;
4438 : }
4439 : }
4440 :
4441 : static struct nvme_ns *
4442 50 : nvme_ns_alloc(void)
4443 : {
4444 : struct nvme_ns *nvme_ns;
4445 :
4446 50 : nvme_ns = calloc(1, sizeof(struct nvme_ns));
4447 50 : if (nvme_ns == NULL) {
4448 0 : return NULL;
4449 : }
4450 :
4451 50 : if (g_opts.io_path_stat) {
4452 0 : nvme_ns->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
4453 0 : if (nvme_ns->stat == NULL) {
4454 0 : free(nvme_ns);
4455 0 : return NULL;
4456 : }
4457 0 : spdk_bdev_reset_io_stat(nvme_ns->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
4458 : }
4459 :
4460 50 : return nvme_ns;
4461 : }
4462 :
4463 : static void
4464 50 : nvme_ns_free(struct nvme_ns *nvme_ns)
4465 : {
4466 50 : free(nvme_ns->stat);
4467 50 : free(nvme_ns);
4468 50 : }
4469 :
4470 : static void
4471 50 : nvme_ctrlr_populate_namespace_done(struct nvme_ns *nvme_ns, int rc)
4472 : {
4473 50 : struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
4474 50 : struct nvme_async_probe_ctx *ctx = nvme_ns->probe_ctx;
4475 :
4476 50 : if (rc == 0) {
4477 48 : nvme_ns->probe_ctx = NULL;
4478 48 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4479 48 : nvme_ctrlr->ref++;
4480 48 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4481 : } else {
4482 2 : RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4483 2 : nvme_ns_free(nvme_ns);
4484 : }
4485 :
4486 50 : if (ctx) {
4487 49 : ctx->populates_in_progress--;
4488 49 : if (ctx->populates_in_progress == 0) {
4489 12 : nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
4490 : }
4491 : }
4492 50 : }
4493 :
4494 : static void
4495 2 : bdev_nvme_add_io_path(struct spdk_io_channel_iter *i)
4496 : {
4497 2 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
4498 2 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
4499 2 : struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
4500 : int rc;
4501 :
4502 2 : rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
4503 2 : if (rc != 0) {
4504 0 : SPDK_ERRLOG("Failed to add I/O path to bdev_channel dynamically.\n");
4505 : }
4506 :
4507 2 : spdk_for_each_channel_continue(i, rc);
4508 2 : }
4509 :
4510 : static void
4511 2 : bdev_nvme_delete_io_path(struct spdk_io_channel_iter *i)
4512 : {
4513 2 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
4514 2 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
4515 2 : struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
4516 : struct nvme_io_path *io_path;
4517 :
4518 2 : io_path = _bdev_nvme_get_io_path(nbdev_ch, nvme_ns);
4519 2 : if (io_path != NULL) {
4520 2 : _bdev_nvme_delete_io_path(nbdev_ch, io_path);
4521 : }
4522 :
4523 2 : spdk_for_each_channel_continue(i, 0);
4524 2 : }
4525 :
4526 : static void
4527 0 : bdev_nvme_add_io_path_failed(struct spdk_io_channel_iter *i, int status)
4528 : {
4529 0 : struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
4530 :
4531 0 : nvme_ctrlr_populate_namespace_done(nvme_ns, -1);
4532 0 : }
4533 :
4534 : static void
4535 12 : bdev_nvme_add_io_path_done(struct spdk_io_channel_iter *i, int status)
4536 : {
4537 12 : struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
4538 12 : struct nvme_bdev *bdev = spdk_io_channel_iter_get_io_device(i);
4539 :
4540 12 : if (status == 0) {
4541 12 : nvme_ctrlr_populate_namespace_done(nvme_ns, 0);
4542 : } else {
4543 : /* Delete the added io_paths and fail populating the namespace. */
4544 0 : spdk_for_each_channel(bdev,
4545 : bdev_nvme_delete_io_path,
4546 : nvme_ns,
4547 : bdev_nvme_add_io_path_failed);
4548 : }
4549 12 : }
4550 :
4551 : static int
4552 13 : nvme_bdev_add_ns(struct nvme_bdev *bdev, struct nvme_ns *nvme_ns)
4553 : {
4554 : struct nvme_ns *tmp_ns;
4555 : const struct spdk_nvme_ns_data *nsdata;
4556 :
4557 13 : nsdata = spdk_nvme_ns_get_data(nvme_ns->ns);
4558 13 : if (!nsdata->nmic.can_share) {
4559 0 : SPDK_ERRLOG("Namespace cannot be shared.\n");
4560 0 : return -EINVAL;
4561 : }
4562 :
4563 13 : pthread_mutex_lock(&bdev->mutex);
4564 :
4565 13 : tmp_ns = TAILQ_FIRST(&bdev->nvme_ns_list);
4566 13 : assert(tmp_ns != NULL);
4567 :
4568 13 : if (tmp_ns->ns != NULL && !bdev_nvme_compare_ns(nvme_ns->ns, tmp_ns->ns)) {
4569 1 : pthread_mutex_unlock(&bdev->mutex);
4570 1 : SPDK_ERRLOG("Namespaces are not identical.\n");
4571 1 : return -EINVAL;
4572 : }
4573 :
4574 12 : bdev->ref++;
4575 12 : TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
4576 12 : nvme_ns->bdev = bdev;
4577 :
4578 12 : pthread_mutex_unlock(&bdev->mutex);
4579 :
4580 : /* Add nvme_io_path to nvme_bdev_channels dynamically. */
4581 12 : spdk_for_each_channel(bdev,
4582 : bdev_nvme_add_io_path,
4583 : nvme_ns,
4584 : bdev_nvme_add_io_path_done);
4585 :
4586 12 : return 0;
4587 : }
4588 :
4589 : static void
4590 50 : nvme_ctrlr_populate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4591 : {
4592 : struct spdk_nvme_ns *ns;
4593 : struct nvme_bdev *bdev;
4594 50 : int rc = 0;
4595 :
4596 50 : ns = spdk_nvme_ctrlr_get_ns(nvme_ctrlr->ctrlr, nvme_ns->id);
4597 50 : if (!ns) {
4598 0 : SPDK_DEBUGLOG(bdev_nvme, "Invalid NS %d\n", nvme_ns->id);
4599 0 : rc = -EINVAL;
4600 0 : goto done;
4601 : }
4602 :
4603 50 : nvme_ns->ns = ns;
4604 50 : nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
4605 :
4606 50 : if (nvme_ctrlr->ana_log_page != NULL) {
4607 37 : bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ns_set_ana_state, nvme_ns);
4608 : }
4609 :
4610 50 : bdev = nvme_bdev_ctrlr_get_bdev(nvme_ctrlr->nbdev_ctrlr, nvme_ns->id);
4611 50 : if (bdev == NULL) {
4612 37 : rc = nvme_bdev_create(nvme_ctrlr, nvme_ns);
4613 : } else {
4614 13 : rc = nvme_bdev_add_ns(bdev, nvme_ns);
4615 13 : if (rc == 0) {
4616 12 : return;
4617 : }
4618 : }
4619 1 : done:
4620 38 : nvme_ctrlr_populate_namespace_done(nvme_ns, rc);
4621 : }
4622 :
4623 : static void
4624 48 : nvme_ctrlr_depopulate_namespace_done(struct nvme_ns *nvme_ns)
4625 : {
4626 48 : struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
4627 :
4628 48 : assert(nvme_ctrlr != NULL);
4629 :
4630 48 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4631 :
4632 48 : RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4633 :
4634 48 : if (nvme_ns->bdev != NULL) {
4635 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4636 0 : return;
4637 : }
4638 :
4639 48 : nvme_ns_free(nvme_ns);
4640 48 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4641 :
4642 48 : nvme_ctrlr_release(nvme_ctrlr);
4643 : }
4644 :
4645 : static void
4646 11 : bdev_nvme_delete_io_path_done(struct spdk_io_channel_iter *i, int status)
4647 : {
4648 11 : struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
4649 :
4650 11 : nvme_ctrlr_depopulate_namespace_done(nvme_ns);
4651 11 : }
4652 :
4653 : static void
4654 48 : nvme_ctrlr_depopulate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4655 : {
4656 : struct nvme_bdev *bdev;
4657 :
4658 48 : spdk_poller_unregister(&nvme_ns->anatt_timer);
4659 :
4660 48 : bdev = nvme_ns->bdev;
4661 48 : if (bdev != NULL) {
4662 44 : pthread_mutex_lock(&bdev->mutex);
4663 :
4664 44 : assert(bdev->ref > 0);
4665 44 : bdev->ref--;
4666 44 : if (bdev->ref == 0) {
4667 33 : pthread_mutex_unlock(&bdev->mutex);
4668 :
4669 33 : spdk_bdev_unregister(&bdev->disk, NULL, NULL);
4670 : } else {
4671 : /* spdk_bdev_unregister() is not called until the last nvme_ns is
4672 : * depopulated. Hence we need to remove nvme_ns from bdev->nvme_ns_list
4673 : * and clear nvme_ns->bdev here.
4674 : */
4675 11 : TAILQ_REMOVE(&bdev->nvme_ns_list, nvme_ns, tailq);
4676 11 : nvme_ns->bdev = NULL;
4677 :
4678 11 : pthread_mutex_unlock(&bdev->mutex);
4679 :
4680 : /* Delete nvme_io_paths from nvme_bdev_channels dynamically. After that,
4681 : * we call depopulate_namespace_done() to avoid use-after-free.
4682 : */
4683 11 : spdk_for_each_channel(bdev,
4684 : bdev_nvme_delete_io_path,
4685 : nvme_ns,
4686 : bdev_nvme_delete_io_path_done);
4687 11 : return;
4688 : }
4689 : }
4690 :
4691 37 : nvme_ctrlr_depopulate_namespace_done(nvme_ns);
4692 : }
4693 :
4694 : static void
4695 61 : nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
4696 : struct nvme_async_probe_ctx *ctx)
4697 : {
4698 61 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
4699 : struct nvme_ns *nvme_ns, *next;
4700 : struct spdk_nvme_ns *ns;
4701 : struct nvme_bdev *bdev;
4702 : uint32_t nsid;
4703 : int rc;
4704 : uint64_t num_sectors;
4705 :
4706 61 : if (ctx) {
4707 : /* Initialize this count to 1 to handle the populate functions
4708 : * calling nvme_ctrlr_populate_namespace_done() immediately.
4709 : */
4710 45 : ctx->populates_in_progress = 1;
4711 : }
4712 :
4713 : /* First loop over our existing namespaces and see if they have been
4714 : * removed. */
4715 61 : nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
4716 65 : while (nvme_ns != NULL) {
4717 4 : next = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
4718 :
4719 4 : if (spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
4720 : /* NS is still there or added again. Its attributes may have changed. */
4721 3 : ns = spdk_nvme_ctrlr_get_ns(ctrlr, nvme_ns->id);
4722 3 : if (nvme_ns->ns != ns) {
4723 1 : assert(nvme_ns->ns == NULL);
4724 1 : nvme_ns->ns = ns;
4725 1 : SPDK_DEBUGLOG(bdev_nvme, "NSID %u was added\n", nvme_ns->id);
4726 : }
4727 :
4728 3 : num_sectors = spdk_nvme_ns_get_num_sectors(ns);
4729 3 : bdev = nvme_ns->bdev;
4730 3 : assert(bdev != NULL);
4731 3 : if (bdev->disk.blockcnt != num_sectors) {
4732 1 : SPDK_NOTICELOG("NSID %u is resized: bdev name %s, old size %" PRIu64 ", new size %" PRIu64 "\n",
4733 : nvme_ns->id,
4734 : bdev->disk.name,
4735 : bdev->disk.blockcnt,
4736 : num_sectors);
4737 1 : rc = spdk_bdev_notify_blockcnt_change(&bdev->disk, num_sectors);
4738 1 : if (rc != 0) {
4739 0 : SPDK_ERRLOG("Could not change num blocks for nvme bdev: name %s, errno: %d.\n",
4740 : bdev->disk.name, rc);
4741 : }
4742 : }
4743 : } else {
4744 : /* Namespace was removed */
4745 1 : nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
4746 : }
4747 :
4748 4 : nvme_ns = next;
4749 : }
4750 :
4751 : /* Loop through all of the namespaces at the nvme level and see if any of them are new */
4752 61 : nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
4753 114 : while (nsid != 0) {
4754 53 : nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);
4755 :
4756 53 : if (nvme_ns == NULL) {
4757 : /* Found a new one */
4758 50 : nvme_ns = nvme_ns_alloc();
4759 50 : if (nvme_ns == NULL) {
4760 0 : SPDK_ERRLOG("Failed to allocate namespace\n");
4761 : /* This just fails to attach the namespace. It may work on a future attempt. */
4762 0 : continue;
4763 : }
4764 :
4765 50 : nvme_ns->id = nsid;
4766 50 : nvme_ns->ctrlr = nvme_ctrlr;
4767 :
4768 50 : nvme_ns->bdev = NULL;
4769 :
4770 50 : if (ctx) {
4771 49 : ctx->populates_in_progress++;
4772 : }
4773 50 : nvme_ns->probe_ctx = ctx;
4774 :
4775 50 : RB_INSERT(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4776 :
4777 50 : nvme_ctrlr_populate_namespace(nvme_ctrlr, nvme_ns);
4778 : }
4779 :
4780 53 : nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid);
4781 : }
4782 :
4783 61 : if (ctx) {
4784 : /* Decrement this count now that the loop is over to account
4785 : * for the one we started with. If the count is then 0, we
4786 : * know any populate_namespace functions completed immediately,
4787 : * so we'll kick the callback here.
4788 : */
4789 45 : ctx->populates_in_progress--;
4790 45 : if (ctx->populates_in_progress == 0) {
4791 33 : nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
4792 : }
4793 : }
4794 :
4795 61 : }
4796 :
4797 : static void
4798 59 : nvme_ctrlr_depopulate_namespaces(struct nvme_ctrlr *nvme_ctrlr)
4799 : {
4800 : struct nvme_ns *nvme_ns, *tmp;
4801 :
4802 106 : RB_FOREACH_SAFE(nvme_ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp) {
4803 47 : nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
4804 : }
4805 59 : }
4806 :
4807 : static uint32_t
4808 36 : nvme_ctrlr_get_ana_log_page_size(struct nvme_ctrlr *nvme_ctrlr)
4809 : {
4810 36 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
4811 : const struct spdk_nvme_ctrlr_data *cdata;
4812 36 : uint32_t nsid, ns_count = 0;
4813 :
4814 36 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
4815 :
4816 80 : for (nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
4817 44 : nsid != 0; nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid)) {
4818 44 : ns_count++;
4819 : }
4820 :
4821 36 : return sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
4822 36 : sizeof(struct spdk_nvme_ana_group_descriptor) + ns_count *
4823 : sizeof(uint32_t);
4824 : }
4825 :
4826 : static int
4827 6 : nvme_ctrlr_set_ana_states(const struct spdk_nvme_ana_group_descriptor *desc,
4828 : void *cb_arg)
4829 : {
4830 6 : struct nvme_ctrlr *nvme_ctrlr = cb_arg;
4831 : struct nvme_ns *nvme_ns;
4832 : uint32_t i, nsid;
4833 :
4834 11 : for (i = 0; i < desc->num_of_nsid; i++) {
4835 5 : nsid = desc->nsid[i];
4836 5 : if (nsid == 0) {
4837 0 : continue;
4838 : }
4839 :
4840 5 : nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);
4841 :
4842 5 : assert(nvme_ns != NULL);
4843 5 : if (nvme_ns == NULL) {
4844 : /* Target told us that an inactive namespace had an ANA change */
4845 0 : continue;
4846 : }
4847 :
4848 5 : _nvme_ns_set_ana_state(nvme_ns, desc);
4849 : }
4850 :
4851 6 : return 0;
4852 : }
4853 :
4854 : static void
4855 0 : bdev_nvme_disable_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
4856 : {
4857 : struct nvme_ns *nvme_ns;
4858 :
4859 0 : spdk_free(nvme_ctrlr->ana_log_page);
4860 0 : nvme_ctrlr->ana_log_page = NULL;
4861 :
4862 0 : for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
4863 : nvme_ns != NULL;
4864 0 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
4865 0 : nvme_ns->ana_state_updating = false;
4866 0 : nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
4867 : }
4868 0 : }
4869 :
4870 : static void
4871 3 : nvme_ctrlr_read_ana_log_page_done(void *ctx, const struct spdk_nvme_cpl *cpl)
4872 : {
4873 3 : struct nvme_ctrlr *nvme_ctrlr = ctx;
4874 :
4875 3 : if (cpl != NULL && spdk_nvme_cpl_is_success(cpl)) {
4876 3 : bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ctrlr_set_ana_states,
4877 : nvme_ctrlr);
4878 : } else {
4879 0 : bdev_nvme_disable_read_ana_log_page(nvme_ctrlr);
4880 : }
4881 :
4882 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4883 :
4884 3 : assert(nvme_ctrlr->ana_log_page_updating == true);
4885 3 : nvme_ctrlr->ana_log_page_updating = false;
4886 :
4887 3 : if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
4888 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4889 :
4890 0 : nvme_ctrlr_unregister(nvme_ctrlr);
4891 : } else {
4892 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4893 :
4894 3 : bdev_nvme_clear_io_path_caches(nvme_ctrlr);
4895 : }
4896 3 : }
4897 :
4898 : static int
4899 6 : nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
4900 : {
4901 : uint32_t ana_log_page_size;
4902 : int rc;
4903 :
4904 6 : if (nvme_ctrlr->ana_log_page == NULL) {
4905 0 : return -EINVAL;
4906 : }
4907 :
4908 6 : ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);
4909 :
4910 6 : if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
4911 0 : SPDK_ERRLOG("ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
4912 : ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
4913 0 : return -EINVAL;
4914 : }
4915 :
4916 6 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4917 6 : if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
4918 : nvme_ctrlr->ana_log_page_updating) {
4919 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4920 3 : return -EBUSY;
4921 : }
4922 :
4923 3 : nvme_ctrlr->ana_log_page_updating = true;
4924 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4925 :
4926 3 : rc = spdk_nvme_ctrlr_cmd_get_log_page(nvme_ctrlr->ctrlr,
4927 : SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
4928 : SPDK_NVME_GLOBAL_NS_TAG,
4929 3 : nvme_ctrlr->ana_log_page,
4930 : ana_log_page_size, 0,
4931 : nvme_ctrlr_read_ana_log_page_done,
4932 : nvme_ctrlr);
4933 3 : if (rc != 0) {
4934 0 : nvme_ctrlr_read_ana_log_page_done(nvme_ctrlr, NULL);
4935 : }
4936 :
4937 3 : return rc;
4938 : }
4939 :
4940 : static void
4941 0 : dummy_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
4942 : {
4943 0 : }
4944 :
4945 : struct bdev_nvme_set_preferred_path_ctx {
4946 : struct spdk_bdev_desc *desc;
4947 : struct nvme_ns *nvme_ns;
4948 : bdev_nvme_set_preferred_path_cb cb_fn;
4949 : void *cb_arg;
4950 : };
4951 :
4952 : static void
4953 3 : bdev_nvme_set_preferred_path_done(struct spdk_io_channel_iter *i, int status)
4954 : {
4955 3 : struct bdev_nvme_set_preferred_path_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
4956 :
4957 3 : assert(ctx != NULL);
4958 3 : assert(ctx->desc != NULL);
4959 3 : assert(ctx->cb_fn != NULL);
4960 :
4961 3 : spdk_bdev_close(ctx->desc);
4962 :
4963 3 : ctx->cb_fn(ctx->cb_arg, status);
4964 :
4965 3 : free(ctx);
4966 3 : }
4967 :
4968 : static void
4969 2 : _bdev_nvme_set_preferred_path(struct spdk_io_channel_iter *i)
4970 : {
4971 2 : struct bdev_nvme_set_preferred_path_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
4972 2 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
4973 2 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
4974 : struct nvme_io_path *io_path, *prev;
4975 :
4976 2 : prev = NULL;
4977 3 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
4978 3 : if (io_path->nvme_ns == ctx->nvme_ns) {
4979 2 : break;
4980 : }
4981 1 : prev = io_path;
4982 : }
4983 :
4984 2 : if (io_path != NULL) {
4985 2 : if (prev != NULL) {
4986 1 : STAILQ_REMOVE_AFTER(&nbdev_ch->io_path_list, prev, stailq);
4987 1 : STAILQ_INSERT_HEAD(&nbdev_ch->io_path_list, io_path, stailq);
4988 : }
4989 :
4990 : /* We can set io_path to nbdev_ch->current_io_path directly here.
4991 : * However, it needs to be conditional. To simplify the code,
4992 : * just clear nbdev_ch->current_io_path and let find_io_path()
4993 : * fill it.
4994 : *
4995 : * Automatic failback may be disabled. Hence even if the io_path is
4996 : * already at the head, clear nbdev_ch->current_io_path.
4997 : */
4998 2 : bdev_nvme_clear_current_io_path(nbdev_ch);
4999 : }
5000 :
5001 2 : spdk_for_each_channel_continue(i, 0);
5002 2 : }
5003 :
5004 : static struct nvme_ns *
5005 3 : bdev_nvme_set_preferred_ns(struct nvme_bdev *nbdev, uint16_t cntlid)
5006 : {
5007 : struct nvme_ns *nvme_ns, *prev;
5008 : const struct spdk_nvme_ctrlr_data *cdata;
5009 :
5010 3 : prev = NULL;
5011 6 : TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
5012 6 : cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
5013 :
5014 6 : if (cdata->cntlid == cntlid) {
5015 3 : break;
5016 : }
5017 3 : prev = nvme_ns;
5018 : }
5019 :
5020 3 : if (nvme_ns != NULL && prev != NULL) {
5021 2 : TAILQ_REMOVE(&nbdev->nvme_ns_list, nvme_ns, tailq);
5022 2 : TAILQ_INSERT_HEAD(&nbdev->nvme_ns_list, nvme_ns, tailq);
5023 : }
5024 :
5025 3 : return nvme_ns;
5026 : }
5027 :
5028 : /* This function supports only multipath mode. There is only a single I/O path
5029 : * for each NVMe-oF controller. Hence, just move the matched I/O path to the
5030 : * head of the I/O path list for each NVMe bdev channel.
5031 : *
5032 : * NVMe bdev channel may be acquired after completing this function. move the
5033 : * matched namespace to the head of the namespace list for the NVMe bdev too.
5034 : */
5035 : void
5036 3 : bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid,
5037 : bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg)
5038 : {
5039 : struct bdev_nvme_set_preferred_path_ctx *ctx;
5040 : struct spdk_bdev *bdev;
5041 : struct nvme_bdev *nbdev;
5042 3 : int rc = 0;
5043 :
5044 3 : assert(cb_fn != NULL);
5045 :
5046 3 : ctx = calloc(1, sizeof(*ctx));
5047 3 : if (ctx == NULL) {
5048 0 : SPDK_ERRLOG("Failed to alloc context.\n");
5049 0 : rc = -ENOMEM;
5050 0 : goto err_alloc;
5051 : }
5052 :
5053 3 : ctx->cb_fn = cb_fn;
5054 3 : ctx->cb_arg = cb_arg;
5055 :
5056 3 : rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
5057 3 : if (rc != 0) {
5058 0 : SPDK_ERRLOG("Failed to open bdev %s.\n", name);
5059 0 : goto err_open;
5060 : }
5061 :
5062 3 : bdev = spdk_bdev_desc_get_bdev(ctx->desc);
5063 :
5064 3 : if (bdev->module != &nvme_if) {
5065 0 : SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
5066 0 : rc = -ENODEV;
5067 0 : goto err_bdev;
5068 : }
5069 :
5070 3 : nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
5071 :
5072 3 : pthread_mutex_lock(&nbdev->mutex);
5073 :
5074 3 : ctx->nvme_ns = bdev_nvme_set_preferred_ns(nbdev, cntlid);
5075 3 : if (ctx->nvme_ns == NULL) {
5076 0 : pthread_mutex_unlock(&nbdev->mutex);
5077 :
5078 0 : SPDK_ERRLOG("bdev %s does not have namespace to controller %u.\n", name, cntlid);
5079 0 : rc = -ENODEV;
5080 0 : goto err_bdev;
5081 : }
5082 :
5083 3 : pthread_mutex_unlock(&nbdev->mutex);
5084 :
5085 3 : spdk_for_each_channel(nbdev,
5086 : _bdev_nvme_set_preferred_path,
5087 : ctx,
5088 : bdev_nvme_set_preferred_path_done);
5089 3 : return;
5090 :
5091 0 : err_bdev:
5092 0 : spdk_bdev_close(ctx->desc);
5093 0 : err_open:
5094 0 : free(ctx);
5095 0 : err_alloc:
5096 0 : cb_fn(cb_arg, rc);
5097 : }
5098 :
5099 : struct bdev_nvme_set_multipath_policy_ctx {
5100 : struct spdk_bdev_desc *desc;
5101 : bdev_nvme_set_multipath_policy_cb cb_fn;
5102 : void *cb_arg;
5103 : };
5104 :
5105 : static void
5106 3 : bdev_nvme_set_multipath_policy_done(struct spdk_io_channel_iter *i, int status)
5107 : {
5108 3 : struct bdev_nvme_set_multipath_policy_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
5109 :
5110 3 : assert(ctx != NULL);
5111 3 : assert(ctx->desc != NULL);
5112 3 : assert(ctx->cb_fn != NULL);
5113 :
5114 3 : spdk_bdev_close(ctx->desc);
5115 :
5116 3 : ctx->cb_fn(ctx->cb_arg, status);
5117 :
5118 3 : free(ctx);
5119 3 : }
5120 :
5121 : static void
5122 1 : _bdev_nvme_set_multipath_policy(struct spdk_io_channel_iter *i)
5123 : {
5124 1 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
5125 1 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
5126 1 : struct nvme_bdev *nbdev = spdk_io_channel_get_io_device(_ch);
5127 :
5128 1 : nbdev_ch->mp_policy = nbdev->mp_policy;
5129 1 : nbdev_ch->mp_selector = nbdev->mp_selector;
5130 1 : nbdev_ch->rr_min_io = nbdev->rr_min_io;
5131 1 : bdev_nvme_clear_current_io_path(nbdev_ch);
5132 :
5133 1 : spdk_for_each_channel_continue(i, 0);
5134 1 : }
5135 :
5136 : void
5137 3 : bdev_nvme_set_multipath_policy(const char *name, enum bdev_nvme_multipath_policy policy,
5138 : enum bdev_nvme_multipath_selector selector, uint32_t rr_min_io,
5139 : bdev_nvme_set_multipath_policy_cb cb_fn, void *cb_arg)
5140 : {
5141 : struct bdev_nvme_set_multipath_policy_ctx *ctx;
5142 : struct spdk_bdev *bdev;
5143 : struct nvme_bdev *nbdev;
5144 : int rc;
5145 :
5146 3 : assert(cb_fn != NULL);
5147 :
5148 3 : switch (policy) {
5149 1 : case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
5150 1 : break;
5151 2 : case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
5152 : switch (selector) {
5153 1 : case BDEV_NVME_MP_SELECTOR_ROUND_ROBIN:
5154 1 : if (rr_min_io == UINT32_MAX) {
5155 0 : rr_min_io = 1;
5156 1 : } else if (rr_min_io == 0) {
5157 0 : rc = -EINVAL;
5158 0 : goto exit;
5159 : }
5160 1 : break;
5161 1 : case BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH:
5162 1 : break;
5163 0 : default:
5164 0 : rc = -EINVAL;
5165 0 : goto exit;
5166 : }
5167 2 : break;
5168 0 : default:
5169 0 : rc = -EINVAL;
5170 0 : goto exit;
5171 : }
5172 :
5173 3 : ctx = calloc(1, sizeof(*ctx));
5174 3 : if (ctx == NULL) {
5175 0 : SPDK_ERRLOG("Failed to alloc context.\n");
5176 0 : rc = -ENOMEM;
5177 0 : goto exit;
5178 : }
5179 :
5180 3 : ctx->cb_fn = cb_fn;
5181 3 : ctx->cb_arg = cb_arg;
5182 :
5183 3 : rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
5184 3 : if (rc != 0) {
5185 0 : SPDK_ERRLOG("Failed to open bdev %s.\n", name);
5186 0 : rc = -ENODEV;
5187 0 : goto err_open;
5188 : }
5189 :
5190 3 : bdev = spdk_bdev_desc_get_bdev(ctx->desc);
5191 3 : if (bdev->module != &nvme_if) {
5192 0 : SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
5193 0 : rc = -ENODEV;
5194 0 : goto err_module;
5195 : }
5196 3 : nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
5197 :
5198 3 : pthread_mutex_lock(&nbdev->mutex);
5199 3 : nbdev->mp_policy = policy;
5200 3 : nbdev->mp_selector = selector;
5201 3 : nbdev->rr_min_io = rr_min_io;
5202 3 : pthread_mutex_unlock(&nbdev->mutex);
5203 :
5204 3 : spdk_for_each_channel(nbdev,
5205 : _bdev_nvme_set_multipath_policy,
5206 : ctx,
5207 : bdev_nvme_set_multipath_policy_done);
5208 3 : return;
5209 :
5210 0 : err_module:
5211 0 : spdk_bdev_close(ctx->desc);
5212 0 : err_open:
5213 0 : free(ctx);
5214 0 : exit:
5215 0 : cb_fn(cb_arg, rc);
5216 : }
5217 :
5218 : static void
5219 3 : aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
5220 : {
5221 3 : struct nvme_ctrlr *nvme_ctrlr = arg;
5222 : union spdk_nvme_async_event_completion event;
5223 :
5224 3 : if (spdk_nvme_cpl_is_error(cpl)) {
5225 0 : SPDK_WARNLOG("AER request execute failed\n");
5226 0 : return;
5227 : }
5228 :
5229 3 : event.raw = cpl->cdw0;
5230 3 : if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
5231 3 : (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) {
5232 2 : nvme_ctrlr_populate_namespaces(nvme_ctrlr, NULL);
5233 1 : } else if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
5234 1 : (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_ANA_CHANGE)) {
5235 1 : nvme_ctrlr_read_ana_log_page(nvme_ctrlr);
5236 : }
5237 : }
5238 :
5239 : static void
5240 51 : free_nvme_async_probe_ctx(struct nvme_async_probe_ctx *ctx)
5241 : {
5242 51 : spdk_keyring_put_key(ctx->drv_opts.tls_psk);
5243 51 : spdk_keyring_put_key(ctx->drv_opts.dhchap_key);
5244 51 : spdk_keyring_put_key(ctx->drv_opts.dhchap_ctrlr_key);
5245 51 : free(ctx);
5246 51 : }
5247 :
5248 : static void
5249 51 : populate_namespaces_cb(struct nvme_async_probe_ctx *ctx, int rc)
5250 : {
5251 51 : if (ctx->cb_fn) {
5252 51 : ctx->cb_fn(ctx->cb_ctx, ctx->reported_bdevs, rc);
5253 : }
5254 :
5255 51 : ctx->namespaces_populated = true;
5256 51 : if (ctx->probe_done) {
5257 : /* The probe was already completed, so we need to free the context
5258 : * here. This can happen for cases like OCSSD, where we need to
5259 : * send additional commands to the SSD after attach.
5260 : */
5261 31 : free_nvme_async_probe_ctx(ctx);
5262 : }
5263 51 : }
5264 :
5265 : static void
5266 59 : nvme_ctrlr_create_done(struct nvme_ctrlr *nvme_ctrlr,
5267 : struct nvme_async_probe_ctx *ctx)
5268 : {
5269 59 : spdk_io_device_register(nvme_ctrlr,
5270 : bdev_nvme_create_ctrlr_channel_cb,
5271 : bdev_nvme_destroy_ctrlr_channel_cb,
5272 : sizeof(struct nvme_ctrlr_channel),
5273 59 : nvme_ctrlr->nbdev_ctrlr->name);
5274 :
5275 59 : nvme_ctrlr_populate_namespaces(nvme_ctrlr, ctx);
5276 59 : }
5277 :
5278 : static void
5279 30 : nvme_ctrlr_init_ana_log_page_done(void *_ctx, const struct spdk_nvme_cpl *cpl)
5280 : {
5281 30 : struct nvme_ctrlr *nvme_ctrlr = _ctx;
5282 30 : struct nvme_async_probe_ctx *ctx = nvme_ctrlr->probe_ctx;
5283 :
5284 30 : nvme_ctrlr->probe_ctx = NULL;
5285 :
5286 30 : if (spdk_nvme_cpl_is_error(cpl)) {
5287 0 : nvme_ctrlr_delete(nvme_ctrlr);
5288 :
5289 0 : if (ctx != NULL) {
5290 0 : ctx->reported_bdevs = 0;
5291 0 : populate_namespaces_cb(ctx, -1);
5292 : }
5293 0 : return;
5294 : }
5295 :
5296 30 : nvme_ctrlr_create_done(nvme_ctrlr, ctx);
5297 : }
5298 :
5299 : static int
5300 30 : nvme_ctrlr_init_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
5301 : struct nvme_async_probe_ctx *ctx)
5302 : {
5303 30 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5304 : const struct spdk_nvme_ctrlr_data *cdata;
5305 : uint32_t ana_log_page_size;
5306 :
5307 30 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5308 :
5309 : /* Set buffer size enough to include maximum number of allowed namespaces. */
5310 30 : ana_log_page_size = sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
5311 30 : sizeof(struct spdk_nvme_ana_group_descriptor) + cdata->mnan *
5312 : sizeof(uint32_t);
5313 :
5314 30 : nvme_ctrlr->ana_log_page = spdk_zmalloc(ana_log_page_size, 64, NULL,
5315 : SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
5316 30 : if (nvme_ctrlr->ana_log_page == NULL) {
5317 0 : SPDK_ERRLOG("could not allocate ANA log page buffer\n");
5318 0 : return -ENXIO;
5319 : }
5320 :
5321 : /* Each descriptor in a ANA log page is not ensured to be 8-bytes aligned.
5322 : * Hence copy each descriptor to a temporary area when parsing it.
5323 : *
5324 : * Allocate a buffer whose size is as large as ANA log page buffer because
5325 : * we do not know the size of a descriptor until actually reading it.
5326 : */
5327 30 : nvme_ctrlr->copied_ana_desc = calloc(1, ana_log_page_size);
5328 30 : if (nvme_ctrlr->copied_ana_desc == NULL) {
5329 0 : SPDK_ERRLOG("could not allocate a buffer to parse ANA descriptor\n");
5330 0 : return -ENOMEM;
5331 : }
5332 :
5333 30 : nvme_ctrlr->max_ana_log_page_size = ana_log_page_size;
5334 :
5335 30 : nvme_ctrlr->probe_ctx = ctx;
5336 :
5337 : /* Then, set the read size only to include the current active namespaces. */
5338 30 : ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);
5339 :
5340 30 : if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
5341 0 : SPDK_ERRLOG("ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
5342 : ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
5343 0 : return -EINVAL;
5344 : }
5345 :
5346 30 : return spdk_nvme_ctrlr_cmd_get_log_page(ctrlr,
5347 : SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
5348 : SPDK_NVME_GLOBAL_NS_TAG,
5349 30 : nvme_ctrlr->ana_log_page,
5350 : ana_log_page_size, 0,
5351 : nvme_ctrlr_init_ana_log_page_done,
5352 : nvme_ctrlr);
5353 : }
5354 :
5355 : /* hostnqn and subnqn were already verified before attaching a controller.
5356 : * Hence check only the multipath capability and cntlid here.
5357 : */
5358 : static bool
5359 16 : bdev_nvme_check_multipath(struct nvme_bdev_ctrlr *nbdev_ctrlr, struct spdk_nvme_ctrlr *ctrlr)
5360 : {
5361 : struct nvme_ctrlr *tmp;
5362 : const struct spdk_nvme_ctrlr_data *cdata, *tmp_cdata;
5363 :
5364 16 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5365 :
5366 16 : if (!cdata->cmic.multi_ctrlr) {
5367 0 : SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
5368 0 : return false;
5369 : }
5370 :
5371 33 : TAILQ_FOREACH(tmp, &nbdev_ctrlr->ctrlrs, tailq) {
5372 18 : tmp_cdata = spdk_nvme_ctrlr_get_data(tmp->ctrlr);
5373 :
5374 18 : if (!tmp_cdata->cmic.multi_ctrlr) {
5375 0 : SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
5376 0 : return false;
5377 : }
5378 18 : if (cdata->cntlid == tmp_cdata->cntlid) {
5379 1 : SPDK_ERRLOG("cntlid %u are duplicated.\n", tmp_cdata->cntlid);
5380 1 : return false;
5381 : }
5382 : }
5383 :
5384 15 : return true;
5385 : }
5386 :
5387 : static int
5388 60 : nvme_bdev_ctrlr_create(const char *name, struct nvme_ctrlr *nvme_ctrlr)
5389 : {
5390 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
5391 60 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5392 60 : int rc = 0;
5393 :
5394 60 : pthread_mutex_lock(&g_bdev_nvme_mutex);
5395 :
5396 60 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
5397 60 : if (nbdev_ctrlr != NULL) {
5398 16 : if (!bdev_nvme_check_multipath(nbdev_ctrlr, ctrlr)) {
5399 1 : rc = -EINVAL;
5400 1 : goto exit;
5401 : }
5402 : } else {
5403 44 : nbdev_ctrlr = calloc(1, sizeof(*nbdev_ctrlr));
5404 44 : if (nbdev_ctrlr == NULL) {
5405 0 : SPDK_ERRLOG("Failed to allocate nvme_bdev_ctrlr.\n");
5406 0 : rc = -ENOMEM;
5407 0 : goto exit;
5408 : }
5409 44 : nbdev_ctrlr->name = strdup(name);
5410 44 : if (nbdev_ctrlr->name == NULL) {
5411 0 : SPDK_ERRLOG("Failed to allocate name of nvme_bdev_ctrlr.\n");
5412 0 : free(nbdev_ctrlr);
5413 0 : goto exit;
5414 : }
5415 44 : TAILQ_INIT(&nbdev_ctrlr->ctrlrs);
5416 44 : TAILQ_INIT(&nbdev_ctrlr->bdevs);
5417 44 : TAILQ_INSERT_TAIL(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
5418 : }
5419 59 : nvme_ctrlr->nbdev_ctrlr = nbdev_ctrlr;
5420 59 : TAILQ_INSERT_TAIL(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
5421 60 : exit:
5422 60 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
5423 60 : return rc;
5424 : }
5425 :
5426 : static int
5427 60 : nvme_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr,
5428 : const char *name,
5429 : const struct spdk_nvme_transport_id *trid,
5430 : struct nvme_async_probe_ctx *ctx)
5431 : {
5432 : struct nvme_ctrlr *nvme_ctrlr;
5433 : struct nvme_path_id *path_id;
5434 : const struct spdk_nvme_ctrlr_data *cdata;
5435 : int rc;
5436 :
5437 60 : nvme_ctrlr = calloc(1, sizeof(*nvme_ctrlr));
5438 60 : if (nvme_ctrlr == NULL) {
5439 0 : SPDK_ERRLOG("Failed to allocate device struct\n");
5440 0 : return -ENOMEM;
5441 : }
5442 :
5443 60 : rc = pthread_mutex_init(&nvme_ctrlr->mutex, NULL);
5444 60 : if (rc != 0) {
5445 0 : free(nvme_ctrlr);
5446 0 : return rc;
5447 : }
5448 :
5449 60 : TAILQ_INIT(&nvme_ctrlr->trids);
5450 60 : RB_INIT(&nvme_ctrlr->namespaces);
5451 :
5452 : /* Get another reference to the key, so the first one can be released from probe_ctx */
5453 60 : if (ctx != NULL) {
5454 46 : if (ctx->drv_opts.tls_psk != NULL) {
5455 0 : nvme_ctrlr->psk = spdk_keyring_get_key(
5456 : spdk_key_get_name(ctx->drv_opts.tls_psk));
5457 0 : if (nvme_ctrlr->psk == NULL) {
5458 : /* Could only happen if the key was removed in the meantime */
5459 0 : SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5460 : spdk_key_get_name(ctx->drv_opts.tls_psk));
5461 0 : rc = -ENOKEY;
5462 0 : goto err;
5463 : }
5464 : }
5465 :
5466 46 : if (ctx->drv_opts.dhchap_key != NULL) {
5467 0 : nvme_ctrlr->dhchap_key = spdk_keyring_get_key(
5468 : spdk_key_get_name(ctx->drv_opts.dhchap_key));
5469 0 : if (nvme_ctrlr->dhchap_key == NULL) {
5470 0 : SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5471 : spdk_key_get_name(ctx->drv_opts.dhchap_key));
5472 0 : rc = -ENOKEY;
5473 0 : goto err;
5474 : }
5475 : }
5476 :
5477 46 : if (ctx->drv_opts.dhchap_ctrlr_key != NULL) {
5478 0 : nvme_ctrlr->dhchap_ctrlr_key =
5479 0 : spdk_keyring_get_key(
5480 : spdk_key_get_name(ctx->drv_opts.dhchap_ctrlr_key));
5481 0 : if (nvme_ctrlr->dhchap_ctrlr_key == NULL) {
5482 0 : SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5483 : spdk_key_get_name(ctx->drv_opts.dhchap_ctrlr_key));
5484 0 : rc = -ENOKEY;
5485 0 : goto err;
5486 : }
5487 : }
5488 : }
5489 :
5490 60 : path_id = calloc(1, sizeof(*path_id));
5491 60 : if (path_id == NULL) {
5492 0 : SPDK_ERRLOG("Failed to allocate trid entry pointer\n");
5493 0 : rc = -ENOMEM;
5494 0 : goto err;
5495 : }
5496 :
5497 60 : path_id->trid = *trid;
5498 60 : if (ctx != NULL) {
5499 46 : memcpy(path_id->hostid.hostaddr, ctx->drv_opts.src_addr, sizeof(path_id->hostid.hostaddr));
5500 46 : memcpy(path_id->hostid.hostsvcid, ctx->drv_opts.src_svcid, sizeof(path_id->hostid.hostsvcid));
5501 : }
5502 60 : nvme_ctrlr->active_path_id = path_id;
5503 60 : TAILQ_INSERT_HEAD(&nvme_ctrlr->trids, path_id, link);
5504 :
5505 60 : nvme_ctrlr->thread = spdk_get_thread();
5506 60 : nvme_ctrlr->ctrlr = ctrlr;
5507 60 : nvme_ctrlr->ref = 1;
5508 :
5509 60 : if (spdk_nvme_ctrlr_is_ocssd_supported(ctrlr)) {
5510 0 : SPDK_ERRLOG("OCSSDs are not supported");
5511 0 : rc = -ENOTSUP;
5512 0 : goto err;
5513 : }
5514 :
5515 60 : if (ctx != NULL) {
5516 46 : memcpy(&nvme_ctrlr->opts, &ctx->bdev_opts, sizeof(ctx->bdev_opts));
5517 : } else {
5518 14 : bdev_nvme_get_default_ctrlr_opts(&nvme_ctrlr->opts);
5519 : }
5520 :
5521 60 : nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq, nvme_ctrlr,
5522 : g_opts.nvme_adminq_poll_period_us);
5523 :
5524 60 : if (g_opts.timeout_us > 0) {
5525 : /* Register timeout callback. Timeout values for IO vs. admin reqs can be different. */
5526 : /* If timeout_admin_us is 0 (not specified), admin uses same timeout as IO. */
5527 0 : uint64_t adm_timeout_us = (g_opts.timeout_admin_us == 0) ?
5528 0 : g_opts.timeout_us : g_opts.timeout_admin_us;
5529 0 : spdk_nvme_ctrlr_register_timeout_callback(ctrlr, g_opts.timeout_us,
5530 : adm_timeout_us, timeout_cb, nvme_ctrlr);
5531 : }
5532 :
5533 60 : spdk_nvme_ctrlr_register_aer_callback(ctrlr, aer_cb, nvme_ctrlr);
5534 60 : spdk_nvme_ctrlr_set_remove_cb(ctrlr, remove_cb, nvme_ctrlr);
5535 :
5536 60 : if (spdk_nvme_ctrlr_get_flags(ctrlr) &
5537 : SPDK_NVME_CTRLR_SECURITY_SEND_RECV_SUPPORTED) {
5538 0 : nvme_ctrlr->opal_dev = spdk_opal_dev_construct(ctrlr);
5539 : }
5540 :
5541 60 : rc = nvme_bdev_ctrlr_create(name, nvme_ctrlr);
5542 60 : if (rc != 0) {
5543 1 : goto err;
5544 : }
5545 :
5546 59 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5547 :
5548 59 : if (cdata->cmic.ana_reporting) {
5549 30 : rc = nvme_ctrlr_init_ana_log_page(nvme_ctrlr, ctx);
5550 30 : if (rc == 0) {
5551 30 : return 0;
5552 : }
5553 : } else {
5554 29 : nvme_ctrlr_create_done(nvme_ctrlr, ctx);
5555 29 : return 0;
5556 : }
5557 :
5558 1 : err:
5559 1 : nvme_ctrlr_delete(nvme_ctrlr);
5560 1 : return rc;
5561 : }
5562 :
5563 : void
5564 56 : bdev_nvme_get_default_ctrlr_opts(struct nvme_ctrlr_opts *opts)
5565 : {
5566 56 : opts->prchk_flags = 0;
5567 56 : opts->ctrlr_loss_timeout_sec = g_opts.ctrlr_loss_timeout_sec;
5568 56 : opts->reconnect_delay_sec = g_opts.reconnect_delay_sec;
5569 56 : opts->fast_io_fail_timeout_sec = g_opts.fast_io_fail_timeout_sec;
5570 56 : }
5571 :
5572 : static void
5573 0 : attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
5574 : struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *drv_opts)
5575 : {
5576 : char *name;
5577 :
5578 0 : name = spdk_sprintf_alloc("HotInNvme%d", g_hot_insert_nvme_controller_index++);
5579 0 : if (!name) {
5580 0 : SPDK_ERRLOG("Failed to assign name to NVMe device\n");
5581 0 : return;
5582 : }
5583 :
5584 0 : if (nvme_ctrlr_create(ctrlr, name, trid, NULL) == 0) {
5585 0 : SPDK_DEBUGLOG(bdev_nvme, "Attached to %s (%s)\n", trid->traddr, name);
5586 : } else {
5587 0 : SPDK_ERRLOG("Failed to attach to %s (%s)\n", trid->traddr, name);
5588 : }
5589 :
5590 0 : free(name);
5591 : }
5592 :
5593 : static void
5594 59 : _nvme_ctrlr_destruct(void *ctx)
5595 : {
5596 59 : struct nvme_ctrlr *nvme_ctrlr = ctx;
5597 :
5598 59 : nvme_ctrlr_depopulate_namespaces(nvme_ctrlr);
5599 59 : nvme_ctrlr_release(nvme_ctrlr);
5600 59 : }
5601 :
5602 : static int
5603 56 : bdev_nvme_delete_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
5604 : {
5605 : struct nvme_probe_skip_entry *entry;
5606 :
5607 : /* The controller's destruction was already started */
5608 56 : if (nvme_ctrlr->destruct) {
5609 0 : return -EALREADY;
5610 : }
5611 :
5612 56 : if (!hotplug &&
5613 56 : nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
5614 0 : entry = calloc(1, sizeof(*entry));
5615 0 : if (!entry) {
5616 0 : return -ENOMEM;
5617 : }
5618 0 : entry->trid = nvme_ctrlr->active_path_id->trid;
5619 0 : TAILQ_INSERT_TAIL(&g_skipped_nvme_ctrlrs, entry, tailq);
5620 : }
5621 :
5622 56 : nvme_ctrlr->destruct = true;
5623 56 : return 0;
5624 : }
5625 :
5626 : static int
5627 2 : bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
5628 : {
5629 : int rc;
5630 :
5631 2 : pthread_mutex_lock(&nvme_ctrlr->mutex);
5632 2 : rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, hotplug);
5633 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
5634 :
5635 2 : if (rc == 0) {
5636 2 : _nvme_ctrlr_destruct(nvme_ctrlr);
5637 0 : } else if (rc == -EALREADY) {
5638 0 : rc = 0;
5639 : }
5640 :
5641 2 : return rc;
5642 : }
5643 :
5644 : static void
5645 0 : remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr)
5646 : {
5647 0 : struct nvme_ctrlr *nvme_ctrlr = cb_ctx;
5648 :
5649 0 : bdev_nvme_delete_ctrlr(nvme_ctrlr, true);
5650 0 : }
5651 :
5652 : static int
5653 0 : bdev_nvme_hotplug_probe(void *arg)
5654 : {
5655 0 : if (g_hotplug_probe_ctx == NULL) {
5656 0 : spdk_poller_unregister(&g_hotplug_probe_poller);
5657 0 : return SPDK_POLLER_IDLE;
5658 : }
5659 :
5660 0 : if (spdk_nvme_probe_poll_async(g_hotplug_probe_ctx) != -EAGAIN) {
5661 0 : g_hotplug_probe_ctx = NULL;
5662 0 : spdk_poller_unregister(&g_hotplug_probe_poller);
5663 : }
5664 :
5665 0 : return SPDK_POLLER_BUSY;
5666 : }
5667 :
5668 : static int
5669 0 : bdev_nvme_hotplug(void *arg)
5670 : {
5671 0 : struct spdk_nvme_transport_id trid_pcie;
5672 :
5673 0 : if (g_hotplug_probe_ctx) {
5674 0 : return SPDK_POLLER_BUSY;
5675 : }
5676 :
5677 0 : memset(&trid_pcie, 0, sizeof(trid_pcie));
5678 0 : spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);
5679 :
5680 0 : g_hotplug_probe_ctx = spdk_nvme_probe_async(&trid_pcie, NULL,
5681 : hotplug_probe_cb, attach_cb, NULL);
5682 :
5683 0 : if (g_hotplug_probe_ctx) {
5684 0 : assert(g_hotplug_probe_poller == NULL);
5685 0 : g_hotplug_probe_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug_probe, NULL, 1000);
5686 : }
5687 :
5688 0 : return SPDK_POLLER_BUSY;
5689 : }
5690 :
5691 : void
5692 0 : bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts)
5693 : {
5694 0 : *opts = g_opts;
5695 0 : }
5696 :
5697 : static bool bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
5698 : uint32_t reconnect_delay_sec,
5699 : uint32_t fast_io_fail_timeout_sec);
5700 :
5701 : static int
5702 0 : bdev_nvme_validate_opts(const struct spdk_bdev_nvme_opts *opts)
5703 : {
5704 0 : if ((opts->timeout_us == 0) && (opts->timeout_admin_us != 0)) {
5705 : /* Can't set timeout_admin_us without also setting timeout_us */
5706 0 : SPDK_WARNLOG("Invalid options: Can't have (timeout_us == 0) with (timeout_admin_us > 0)\n");
5707 0 : return -EINVAL;
5708 : }
5709 :
5710 0 : if (opts->bdev_retry_count < -1) {
5711 0 : SPDK_WARNLOG("Invalid option: bdev_retry_count can't be less than -1.\n");
5712 0 : return -EINVAL;
5713 : }
5714 :
5715 0 : if (!bdev_nvme_check_io_error_resiliency_params(opts->ctrlr_loss_timeout_sec,
5716 : opts->reconnect_delay_sec,
5717 : opts->fast_io_fail_timeout_sec)) {
5718 0 : return -EINVAL;
5719 : }
5720 :
5721 0 : return 0;
5722 : }
5723 :
5724 : int
5725 0 : bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts)
5726 : {
5727 : int ret;
5728 :
5729 0 : ret = bdev_nvme_validate_opts(opts);
5730 0 : if (ret) {
5731 0 : SPDK_WARNLOG("Failed to set nvme opts.\n");
5732 0 : return ret;
5733 : }
5734 :
5735 0 : if (g_bdev_nvme_init_thread != NULL) {
5736 0 : if (!TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
5737 0 : return -EPERM;
5738 : }
5739 : }
5740 :
5741 0 : if (opts->rdma_srq_size != 0 ||
5742 0 : opts->rdma_max_cq_size != 0 ||
5743 0 : opts->rdma_cm_event_timeout_ms != 0) {
5744 0 : struct spdk_nvme_transport_opts drv_opts;
5745 :
5746 0 : spdk_nvme_transport_get_opts(&drv_opts, sizeof(drv_opts));
5747 0 : if (opts->rdma_srq_size != 0) {
5748 0 : drv_opts.rdma_srq_size = opts->rdma_srq_size;
5749 : }
5750 0 : if (opts->rdma_max_cq_size != 0) {
5751 0 : drv_opts.rdma_max_cq_size = opts->rdma_max_cq_size;
5752 : }
5753 0 : if (opts->rdma_cm_event_timeout_ms != 0) {
5754 0 : drv_opts.rdma_cm_event_timeout_ms = opts->rdma_cm_event_timeout_ms;
5755 : }
5756 :
5757 0 : ret = spdk_nvme_transport_set_opts(&drv_opts, sizeof(drv_opts));
5758 0 : if (ret) {
5759 0 : SPDK_ERRLOG("Failed to set NVMe transport opts.\n");
5760 0 : return ret;
5761 : }
5762 : }
5763 :
5764 0 : g_opts = *opts;
5765 :
5766 0 : return 0;
5767 : }
5768 :
5769 : struct set_nvme_hotplug_ctx {
5770 : uint64_t period_us;
5771 : bool enabled;
5772 : spdk_msg_fn fn;
5773 : void *fn_ctx;
5774 : };
5775 :
5776 : static void
5777 0 : set_nvme_hotplug_period_cb(void *_ctx)
5778 : {
5779 0 : struct set_nvme_hotplug_ctx *ctx = _ctx;
5780 :
5781 0 : spdk_poller_unregister(&g_hotplug_poller);
5782 0 : if (ctx->enabled) {
5783 0 : g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug, NULL, ctx->period_us);
5784 : }
5785 :
5786 0 : g_nvme_hotplug_poll_period_us = ctx->period_us;
5787 0 : g_nvme_hotplug_enabled = ctx->enabled;
5788 0 : if (ctx->fn) {
5789 0 : ctx->fn(ctx->fn_ctx);
5790 : }
5791 :
5792 0 : free(ctx);
5793 0 : }
5794 :
5795 : int
5796 0 : bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx)
5797 : {
5798 : struct set_nvme_hotplug_ctx *ctx;
5799 :
5800 0 : if (enabled == true && !spdk_process_is_primary()) {
5801 0 : return -EPERM;
5802 : }
5803 :
5804 0 : ctx = calloc(1, sizeof(*ctx));
5805 0 : if (ctx == NULL) {
5806 0 : return -ENOMEM;
5807 : }
5808 :
5809 0 : period_us = period_us == 0 ? NVME_HOTPLUG_POLL_PERIOD_DEFAULT : period_us;
5810 0 : ctx->period_us = spdk_min(period_us, NVME_HOTPLUG_POLL_PERIOD_MAX);
5811 0 : ctx->enabled = enabled;
5812 0 : ctx->fn = cb;
5813 0 : ctx->fn_ctx = cb_ctx;
5814 :
5815 0 : spdk_thread_send_msg(g_bdev_nvme_init_thread, set_nvme_hotplug_period_cb, ctx);
5816 0 : return 0;
5817 : }
5818 :
5819 : static void
5820 45 : nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
5821 : struct nvme_async_probe_ctx *ctx)
5822 : {
5823 : struct nvme_ns *nvme_ns;
5824 : struct nvme_bdev *nvme_bdev;
5825 : size_t j;
5826 :
5827 45 : assert(nvme_ctrlr != NULL);
5828 :
5829 45 : if (ctx->names == NULL) {
5830 0 : ctx->reported_bdevs = 0;
5831 0 : populate_namespaces_cb(ctx, 0);
5832 0 : return;
5833 : }
5834 :
5835 : /*
5836 : * Report the new bdevs that were created in this call.
5837 : * There can be more than one bdev per NVMe controller.
5838 : */
5839 45 : j = 0;
5840 45 : nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
5841 92 : while (nvme_ns != NULL) {
5842 47 : nvme_bdev = nvme_ns->bdev;
5843 47 : if (j < ctx->max_bdevs) {
5844 47 : ctx->names[j] = nvme_bdev->disk.name;
5845 47 : j++;
5846 : } else {
5847 0 : SPDK_ERRLOG("Maximum number of namespaces supported per NVMe controller is %du. Unable to return all names of created bdevs\n",
5848 : ctx->max_bdevs);
5849 0 : ctx->reported_bdevs = 0;
5850 0 : populate_namespaces_cb(ctx, -ERANGE);
5851 0 : return;
5852 : }
5853 :
5854 47 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
5855 : }
5856 :
5857 45 : ctx->reported_bdevs = j;
5858 45 : populate_namespaces_cb(ctx, 0);
5859 : }
5860 :
5861 : static int
5862 9 : bdev_nvme_check_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
5863 : struct spdk_nvme_ctrlr *new_ctrlr,
5864 : struct spdk_nvme_transport_id *trid)
5865 : {
5866 : struct nvme_path_id *tmp_trid;
5867 :
5868 9 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
5869 0 : SPDK_ERRLOG("PCIe failover is not supported.\n");
5870 0 : return -ENOTSUP;
5871 : }
5872 :
5873 : /* Currently we only support failover to the same transport type. */
5874 9 : if (nvme_ctrlr->active_path_id->trid.trtype != trid->trtype) {
5875 0 : SPDK_WARNLOG("Failover from trtype: %s to a different trtype: %s is not supported currently\n",
5876 : spdk_nvme_transport_id_trtype_str(nvme_ctrlr->active_path_id->trid.trtype),
5877 : spdk_nvme_transport_id_trtype_str(trid->trtype));
5878 0 : return -EINVAL;
5879 : }
5880 :
5881 :
5882 : /* Currently we only support failover to the same NQN. */
5883 9 : if (strncmp(trid->subnqn, nvme_ctrlr->active_path_id->trid.subnqn, SPDK_NVMF_NQN_MAX_LEN)) {
5884 0 : SPDK_WARNLOG("Failover from subnqn: %s to a different subnqn: %s is not supported currently\n",
5885 : nvme_ctrlr->active_path_id->trid.subnqn, trid->subnqn);
5886 0 : return -EINVAL;
5887 : }
5888 :
5889 : /* Skip all the other checks if we've already registered this path. */
5890 21 : TAILQ_FOREACH(tmp_trid, &nvme_ctrlr->trids, link) {
5891 12 : if (!spdk_nvme_transport_id_compare(&tmp_trid->trid, trid)) {
5892 0 : SPDK_WARNLOG("This path (traddr: %s subnqn: %s) is already registered\n", trid->traddr,
5893 : trid->subnqn);
5894 0 : return -EALREADY;
5895 : }
5896 : }
5897 :
5898 9 : return 0;
5899 : }
5900 :
5901 : static int
5902 9 : bdev_nvme_check_secondary_namespace(struct nvme_ctrlr *nvme_ctrlr,
5903 : struct spdk_nvme_ctrlr *new_ctrlr)
5904 : {
5905 : struct nvme_ns *nvme_ns;
5906 : struct spdk_nvme_ns *new_ns;
5907 :
5908 9 : nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
5909 9 : while (nvme_ns != NULL) {
5910 0 : new_ns = spdk_nvme_ctrlr_get_ns(new_ctrlr, nvme_ns->id);
5911 0 : assert(new_ns != NULL);
5912 :
5913 0 : if (!bdev_nvme_compare_ns(nvme_ns->ns, new_ns)) {
5914 0 : return -EINVAL;
5915 : }
5916 :
5917 0 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
5918 : }
5919 :
5920 9 : return 0;
5921 : }
5922 :
5923 : static int
5924 9 : _bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
5925 : struct spdk_nvme_transport_id *trid)
5926 : {
5927 : struct nvme_path_id *active_id, *new_trid, *tmp_trid;
5928 :
5929 9 : new_trid = calloc(1, sizeof(*new_trid));
5930 9 : if (new_trid == NULL) {
5931 0 : return -ENOMEM;
5932 : }
5933 9 : new_trid->trid = *trid;
5934 :
5935 9 : active_id = nvme_ctrlr->active_path_id;
5936 9 : assert(active_id != NULL);
5937 9 : assert(active_id == TAILQ_FIRST(&nvme_ctrlr->trids));
5938 :
5939 : /* Skip the active trid not to replace it until it is failed. */
5940 9 : tmp_trid = TAILQ_NEXT(active_id, link);
5941 9 : if (tmp_trid == NULL) {
5942 6 : goto add_tail;
5943 : }
5944 :
5945 : /* It means the trid is faled if its last failed time is non-zero.
5946 : * Insert the new alternate trid before any failed trid.
5947 : */
5948 5 : TAILQ_FOREACH_FROM(tmp_trid, &nvme_ctrlr->trids, link) {
5949 3 : if (tmp_trid->last_failed_tsc != 0) {
5950 1 : TAILQ_INSERT_BEFORE(tmp_trid, new_trid, link);
5951 1 : return 0;
5952 : }
5953 : }
5954 :
5955 2 : add_tail:
5956 8 : TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, new_trid, link);
5957 8 : return 0;
5958 : }
5959 :
5960 : /* This is the case that a secondary path is added to an existing
5961 : * nvme_ctrlr for failover. After checking if it can access the same
5962 : * namespaces as the primary path, it is disconnected until failover occurs.
5963 : */
5964 : static int
5965 9 : bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
5966 : struct spdk_nvme_ctrlr *new_ctrlr,
5967 : struct spdk_nvme_transport_id *trid)
5968 : {
5969 : int rc;
5970 :
5971 9 : assert(nvme_ctrlr != NULL);
5972 :
5973 9 : pthread_mutex_lock(&nvme_ctrlr->mutex);
5974 :
5975 9 : rc = bdev_nvme_check_secondary_trid(nvme_ctrlr, new_ctrlr, trid);
5976 9 : if (rc != 0) {
5977 0 : goto exit;
5978 : }
5979 :
5980 9 : rc = bdev_nvme_check_secondary_namespace(nvme_ctrlr, new_ctrlr);
5981 9 : if (rc != 0) {
5982 0 : goto exit;
5983 : }
5984 :
5985 9 : rc = _bdev_nvme_add_secondary_trid(nvme_ctrlr, trid);
5986 :
5987 9 : exit:
5988 9 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
5989 :
5990 9 : spdk_nvme_detach(new_ctrlr);
5991 :
5992 9 : return rc;
5993 : }
5994 :
5995 : static void
5996 46 : connect_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
5997 : struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
5998 : {
5999 46 : struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6000 : struct nvme_async_probe_ctx *ctx;
6001 : int rc;
6002 :
6003 46 : ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
6004 46 : ctx->ctrlr_attached = true;
6005 :
6006 46 : rc = nvme_ctrlr_create(ctrlr, ctx->base_name, &ctx->trid, ctx);
6007 46 : if (rc != 0) {
6008 1 : ctx->reported_bdevs = 0;
6009 1 : populate_namespaces_cb(ctx, rc);
6010 : }
6011 46 : }
6012 :
6013 : static void
6014 4 : connect_set_failover_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
6015 : struct spdk_nvme_ctrlr *ctrlr,
6016 : const struct spdk_nvme_ctrlr_opts *opts)
6017 : {
6018 4 : struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6019 : struct nvme_ctrlr *nvme_ctrlr;
6020 : struct nvme_async_probe_ctx *ctx;
6021 : int rc;
6022 :
6023 4 : ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
6024 4 : ctx->ctrlr_attached = true;
6025 :
6026 4 : nvme_ctrlr = nvme_ctrlr_get_by_name(ctx->base_name);
6027 4 : if (nvme_ctrlr) {
6028 4 : rc = bdev_nvme_add_secondary_trid(nvme_ctrlr, ctrlr, &ctx->trid);
6029 : } else {
6030 0 : rc = -ENODEV;
6031 : }
6032 :
6033 4 : ctx->reported_bdevs = 0;
6034 4 : populate_namespaces_cb(ctx, rc);
6035 4 : }
6036 :
6037 : static int
6038 51 : bdev_nvme_async_poll(void *arg)
6039 : {
6040 51 : struct nvme_async_probe_ctx *ctx = arg;
6041 : int rc;
6042 :
6043 51 : rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
6044 51 : if (spdk_unlikely(rc != -EAGAIN)) {
6045 51 : ctx->probe_done = true;
6046 51 : spdk_poller_unregister(&ctx->poller);
6047 51 : if (!ctx->ctrlr_attached) {
6048 : /* The probe is done, but no controller was attached.
6049 : * That means we had a failure, so report -EIO back to
6050 : * the caller (usually the RPC). populate_namespaces_cb()
6051 : * will take care of freeing the nvme_async_probe_ctx.
6052 : */
6053 1 : ctx->reported_bdevs = 0;
6054 1 : populate_namespaces_cb(ctx, -EIO);
6055 50 : } else if (ctx->namespaces_populated) {
6056 : /* The namespaces for the attached controller were all
6057 : * populated and the response was already sent to the
6058 : * caller (usually the RPC). So free the context here.
6059 : */
6060 20 : free_nvme_async_probe_ctx(ctx);
6061 : }
6062 : }
6063 :
6064 51 : return SPDK_POLLER_BUSY;
6065 : }
6066 :
6067 : static bool
6068 28 : bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
6069 : uint32_t reconnect_delay_sec,
6070 : uint32_t fast_io_fail_timeout_sec)
6071 : {
6072 28 : if (ctrlr_loss_timeout_sec < -1) {
6073 1 : SPDK_ERRLOG("ctrlr_loss_timeout_sec can't be less than -1.\n");
6074 1 : return false;
6075 27 : } else if (ctrlr_loss_timeout_sec == -1) {
6076 13 : if (reconnect_delay_sec == 0) {
6077 1 : SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
6078 1 : return false;
6079 12 : } else if (fast_io_fail_timeout_sec != 0 &&
6080 : fast_io_fail_timeout_sec < reconnect_delay_sec) {
6081 1 : SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io-fail_timeout_sec.\n");
6082 1 : return false;
6083 : }
6084 14 : } else if (ctrlr_loss_timeout_sec != 0) {
6085 11 : if (reconnect_delay_sec == 0) {
6086 1 : SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
6087 1 : return false;
6088 10 : } else if (reconnect_delay_sec > (uint32_t)ctrlr_loss_timeout_sec) {
6089 1 : SPDK_ERRLOG("reconnect_delay_sec can't be more than ctrlr_loss_timeout_sec.\n");
6090 1 : return false;
6091 9 : } else if (fast_io_fail_timeout_sec != 0) {
6092 6 : if (fast_io_fail_timeout_sec < reconnect_delay_sec) {
6093 1 : SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io_fail_timeout_sec.\n");
6094 1 : return false;
6095 5 : } else if (fast_io_fail_timeout_sec > (uint32_t)ctrlr_loss_timeout_sec) {
6096 1 : SPDK_ERRLOG("fast_io_fail_timeout_sec can't be more than ctrlr_loss_timeout_sec.\n");
6097 1 : return false;
6098 : }
6099 : }
6100 3 : } else if (reconnect_delay_sec != 0 || fast_io_fail_timeout_sec != 0) {
6101 2 : SPDK_ERRLOG("Both reconnect_delay_sec and fast_io_fail_timeout_sec must be 0 if ctrlr_loss_timeout_sec is 0.\n");
6102 2 : return false;
6103 : }
6104 :
6105 19 : return true;
6106 : }
6107 :
6108 : static int
6109 0 : bdev_nvme_load_psk(const char *fname, char *buf, size_t bufsz)
6110 : {
6111 : FILE *psk_file;
6112 0 : struct stat statbuf;
6113 : int rc;
6114 : #define TCP_PSK_INVALID_PERMISSIONS 0177
6115 :
6116 0 : if (stat(fname, &statbuf) != 0) {
6117 0 : SPDK_ERRLOG("Could not read permissions for PSK file\n");
6118 0 : return -EACCES;
6119 : }
6120 :
6121 0 : if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) {
6122 0 : SPDK_ERRLOG("Incorrect permissions for PSK file\n");
6123 0 : return -EPERM;
6124 : }
6125 0 : if ((size_t)statbuf.st_size >= bufsz) {
6126 0 : SPDK_ERRLOG("Invalid PSK: too long\n");
6127 0 : return -EINVAL;
6128 : }
6129 0 : psk_file = fopen(fname, "r");
6130 0 : if (psk_file == NULL) {
6131 0 : SPDK_ERRLOG("Could not open PSK file\n");
6132 0 : return -EINVAL;
6133 : }
6134 :
6135 0 : memset(buf, 0, bufsz);
6136 0 : rc = fread(buf, 1, statbuf.st_size, psk_file);
6137 0 : if (rc != statbuf.st_size) {
6138 0 : SPDK_ERRLOG("Failed to read PSK\n");
6139 0 : fclose(psk_file);
6140 0 : return -EINVAL;
6141 : }
6142 :
6143 0 : fclose(psk_file);
6144 0 : return 0;
6145 : }
6146 :
6147 : int
6148 51 : bdev_nvme_create(struct spdk_nvme_transport_id *trid,
6149 : const char *base_name,
6150 : const char **names,
6151 : uint32_t count,
6152 : spdk_bdev_create_nvme_fn cb_fn,
6153 : void *cb_ctx,
6154 : struct spdk_nvme_ctrlr_opts *drv_opts,
6155 : struct nvme_ctrlr_opts *bdev_opts,
6156 : bool multipath)
6157 : {
6158 : struct nvme_probe_skip_entry *entry, *tmp;
6159 : struct nvme_async_probe_ctx *ctx;
6160 : spdk_nvme_attach_cb attach_cb;
6161 : int rc, len;
6162 :
6163 : /* TODO expand this check to include both the host and target TRIDs.
6164 : * Only if both are the same should we fail.
6165 : */
6166 51 : if (nvme_ctrlr_get(trid) != NULL) {
6167 0 : SPDK_ERRLOG("A controller with the provided trid (traddr: %s) already exists.\n", trid->traddr);
6168 0 : return -EEXIST;
6169 : }
6170 :
6171 51 : len = strnlen(base_name, SPDK_CONTROLLER_NAME_MAX);
6172 :
6173 51 : if (len == 0 || len == SPDK_CONTROLLER_NAME_MAX) {
6174 0 : SPDK_ERRLOG("controller name must be between 1 and %d characters\n", SPDK_CONTROLLER_NAME_MAX - 1);
6175 0 : return -EINVAL;
6176 : }
6177 :
6178 51 : if (bdev_opts != NULL &&
6179 9 : !bdev_nvme_check_io_error_resiliency_params(bdev_opts->ctrlr_loss_timeout_sec,
6180 : bdev_opts->reconnect_delay_sec,
6181 : bdev_opts->fast_io_fail_timeout_sec)) {
6182 0 : return -EINVAL;
6183 : }
6184 :
6185 51 : ctx = calloc(1, sizeof(*ctx));
6186 51 : if (!ctx) {
6187 0 : return -ENOMEM;
6188 : }
6189 51 : ctx->base_name = base_name;
6190 51 : ctx->names = names;
6191 51 : ctx->max_bdevs = count;
6192 51 : ctx->cb_fn = cb_fn;
6193 51 : ctx->cb_ctx = cb_ctx;
6194 51 : ctx->trid = *trid;
6195 :
6196 51 : if (bdev_opts) {
6197 9 : memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
6198 : } else {
6199 42 : bdev_nvme_get_default_ctrlr_opts(&ctx->bdev_opts);
6200 : }
6201 :
6202 51 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
6203 0 : TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, tmp) {
6204 0 : if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
6205 0 : TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
6206 0 : free(entry);
6207 0 : break;
6208 : }
6209 : }
6210 : }
6211 :
6212 51 : if (drv_opts) {
6213 0 : memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
6214 : } else {
6215 51 : spdk_nvme_ctrlr_get_default_ctrlr_opts(&ctx->drv_opts, sizeof(ctx->drv_opts));
6216 : }
6217 :
6218 51 : ctx->drv_opts.transport_retry_count = g_opts.transport_retry_count;
6219 51 : ctx->drv_opts.transport_ack_timeout = g_opts.transport_ack_timeout;
6220 51 : ctx->drv_opts.keep_alive_timeout_ms = g_opts.keep_alive_timeout_ms;
6221 51 : ctx->drv_opts.disable_read_ana_log_page = true;
6222 51 : ctx->drv_opts.transport_tos = g_opts.transport_tos;
6223 :
6224 51 : if (ctx->bdev_opts.psk[0] != '\0') {
6225 : /* Try to use the keyring first */
6226 0 : ctx->drv_opts.tls_psk = spdk_keyring_get_key(ctx->bdev_opts.psk);
6227 0 : if (ctx->drv_opts.tls_psk == NULL) {
6228 0 : rc = bdev_nvme_load_psk(ctx->bdev_opts.psk,
6229 0 : ctx->drv_opts.psk, sizeof(ctx->drv_opts.psk));
6230 0 : if (rc != 0) {
6231 0 : SPDK_ERRLOG("Could not load PSK from %s\n", ctx->bdev_opts.psk);
6232 0 : free_nvme_async_probe_ctx(ctx);
6233 0 : return rc;
6234 : }
6235 : }
6236 : }
6237 :
6238 51 : if (ctx->bdev_opts.dhchap_key != NULL) {
6239 0 : ctx->drv_opts.dhchap_key = spdk_keyring_get_key(ctx->bdev_opts.dhchap_key);
6240 0 : if (ctx->drv_opts.dhchap_key == NULL) {
6241 0 : SPDK_ERRLOG("Could not load DH-HMAC-CHAP key: %s\n",
6242 : ctx->bdev_opts.dhchap_key);
6243 0 : free_nvme_async_probe_ctx(ctx);
6244 0 : return -ENOKEY;
6245 : }
6246 :
6247 0 : ctx->drv_opts.dhchap_digests = g_opts.dhchap_digests;
6248 0 : ctx->drv_opts.dhchap_dhgroups = g_opts.dhchap_dhgroups;
6249 : }
6250 51 : if (ctx->bdev_opts.dhchap_ctrlr_key != NULL) {
6251 0 : ctx->drv_opts.dhchap_ctrlr_key =
6252 0 : spdk_keyring_get_key(ctx->bdev_opts.dhchap_ctrlr_key);
6253 0 : if (ctx->drv_opts.dhchap_ctrlr_key == NULL) {
6254 0 : SPDK_ERRLOG("Could not load DH-HMAC-CHAP controller key: %s\n",
6255 : ctx->bdev_opts.dhchap_ctrlr_key);
6256 0 : free_nvme_async_probe_ctx(ctx);
6257 0 : return -ENOKEY;
6258 : }
6259 : }
6260 :
6261 51 : if (nvme_bdev_ctrlr_get_by_name(base_name) == NULL || multipath) {
6262 47 : attach_cb = connect_attach_cb;
6263 : } else {
6264 4 : attach_cb = connect_set_failover_cb;
6265 : }
6266 :
6267 51 : ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, attach_cb);
6268 51 : if (ctx->probe_ctx == NULL) {
6269 0 : SPDK_ERRLOG("No controller was found with provided trid (traddr: %s)\n", trid->traddr);
6270 0 : free_nvme_async_probe_ctx(ctx);
6271 0 : return -ENODEV;
6272 : }
6273 51 : ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_async_poll, ctx, 1000);
6274 :
6275 51 : return 0;
6276 : }
6277 :
6278 : struct bdev_nvme_delete_ctx {
6279 : char *name;
6280 : struct nvme_path_id path_id;
6281 : bdev_nvme_delete_done_fn delete_done;
6282 : void *delete_done_ctx;
6283 : uint64_t timeout_ticks;
6284 : struct spdk_poller *poller;
6285 : };
6286 :
6287 : static void
6288 2 : free_bdev_nvme_delete_ctx(struct bdev_nvme_delete_ctx *ctx)
6289 : {
6290 2 : if (ctx != NULL) {
6291 1 : free(ctx->name);
6292 1 : free(ctx);
6293 : }
6294 2 : }
6295 :
6296 : static bool
6297 74 : nvme_path_id_compare(struct nvme_path_id *p, const struct nvme_path_id *path_id)
6298 : {
6299 74 : if (path_id->trid.trtype != 0) {
6300 21 : if (path_id->trid.trtype == SPDK_NVME_TRANSPORT_CUSTOM) {
6301 0 : if (strcasecmp(path_id->trid.trstring, p->trid.trstring) != 0) {
6302 0 : return false;
6303 : }
6304 : } else {
6305 21 : if (path_id->trid.trtype != p->trid.trtype) {
6306 0 : return false;
6307 : }
6308 : }
6309 : }
6310 :
6311 74 : if (!spdk_mem_all_zero(path_id->trid.traddr, sizeof(path_id->trid.traddr))) {
6312 21 : if (strcasecmp(path_id->trid.traddr, p->trid.traddr) != 0) {
6313 11 : return false;
6314 : }
6315 : }
6316 :
6317 63 : if (path_id->trid.adrfam != 0) {
6318 0 : if (path_id->trid.adrfam != p->trid.adrfam) {
6319 0 : return false;
6320 : }
6321 : }
6322 :
6323 63 : if (!spdk_mem_all_zero(path_id->trid.trsvcid, sizeof(path_id->trid.trsvcid))) {
6324 10 : if (strcasecmp(path_id->trid.trsvcid, p->trid.trsvcid) != 0) {
6325 0 : return false;
6326 : }
6327 : }
6328 :
6329 63 : if (!spdk_mem_all_zero(path_id->trid.subnqn, sizeof(path_id->trid.subnqn))) {
6330 10 : if (strcmp(path_id->trid.subnqn, p->trid.subnqn) != 0) {
6331 0 : return false;
6332 : }
6333 : }
6334 :
6335 63 : if (!spdk_mem_all_zero(path_id->hostid.hostaddr, sizeof(path_id->hostid.hostaddr))) {
6336 0 : if (strcmp(path_id->hostid.hostaddr, p->hostid.hostaddr) != 0) {
6337 0 : return false;
6338 : }
6339 : }
6340 :
6341 63 : if (!spdk_mem_all_zero(path_id->hostid.hostsvcid, sizeof(path_id->hostid.hostsvcid))) {
6342 0 : if (strcmp(path_id->hostid.hostsvcid, p->hostid.hostsvcid) != 0) {
6343 0 : return false;
6344 : }
6345 : }
6346 :
6347 63 : return true;
6348 : }
6349 :
6350 : static bool
6351 2 : nvme_path_id_exists(const char *name, const struct nvme_path_id *path_id)
6352 : {
6353 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
6354 : struct nvme_ctrlr *ctrlr;
6355 : struct nvme_path_id *p;
6356 :
6357 2 : pthread_mutex_lock(&g_bdev_nvme_mutex);
6358 2 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
6359 2 : if (!nbdev_ctrlr) {
6360 1 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6361 1 : return false;
6362 : }
6363 :
6364 1 : TAILQ_FOREACH(ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
6365 1 : pthread_mutex_lock(&ctrlr->mutex);
6366 1 : TAILQ_FOREACH(p, &ctrlr->trids, link) {
6367 1 : if (nvme_path_id_compare(p, path_id)) {
6368 1 : pthread_mutex_unlock(&ctrlr->mutex);
6369 1 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6370 1 : return true;
6371 : }
6372 : }
6373 0 : pthread_mutex_unlock(&ctrlr->mutex);
6374 : }
6375 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6376 :
6377 0 : return false;
6378 : }
6379 :
6380 : static int
6381 2 : bdev_nvme_delete_complete_poll(void *arg)
6382 : {
6383 2 : struct bdev_nvme_delete_ctx *ctx = arg;
6384 2 : int rc = 0;
6385 :
6386 2 : if (nvme_path_id_exists(ctx->name, &ctx->path_id)) {
6387 1 : if (ctx->timeout_ticks > spdk_get_ticks()) {
6388 1 : return SPDK_POLLER_BUSY;
6389 : }
6390 :
6391 0 : SPDK_ERRLOG("NVMe path '%s' still exists after delete\n", ctx->name);
6392 0 : rc = -ETIMEDOUT;
6393 : }
6394 :
6395 1 : spdk_poller_unregister(&ctx->poller);
6396 :
6397 1 : ctx->delete_done(ctx->delete_done_ctx, rc);
6398 1 : free_bdev_nvme_delete_ctx(ctx);
6399 :
6400 1 : return SPDK_POLLER_BUSY;
6401 : }
6402 :
6403 : static int
6404 63 : _bdev_nvme_delete(struct nvme_ctrlr *nvme_ctrlr, const struct nvme_path_id *path_id)
6405 : {
6406 : struct nvme_path_id *p, *t;
6407 : spdk_msg_fn msg_fn;
6408 63 : int rc = -ENXIO;
6409 :
6410 63 : pthread_mutex_lock(&nvme_ctrlr->mutex);
6411 :
6412 73 : TAILQ_FOREACH_REVERSE_SAFE(p, &nvme_ctrlr->trids, nvme_paths, link, t) {
6413 73 : if (p == TAILQ_FIRST(&nvme_ctrlr->trids)) {
6414 63 : break;
6415 : }
6416 :
6417 10 : if (!nvme_path_id_compare(p, path_id)) {
6418 3 : continue;
6419 : }
6420 :
6421 : /* We are not using the specified path. */
6422 7 : TAILQ_REMOVE(&nvme_ctrlr->trids, p, link);
6423 7 : free(p);
6424 7 : rc = 0;
6425 : }
6426 :
6427 63 : if (p == NULL || !nvme_path_id_compare(p, path_id)) {
6428 8 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
6429 8 : return rc;
6430 : }
6431 :
6432 : /* If we made it here, then this path is a match! Now we need to remove it. */
6433 :
6434 : /* This is the active path in use right now. The active path is always the first in the list. */
6435 55 : assert(p == nvme_ctrlr->active_path_id);
6436 :
6437 55 : if (!TAILQ_NEXT(p, link)) {
6438 : /* The current path is the only path. */
6439 54 : msg_fn = _nvme_ctrlr_destruct;
6440 54 : rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, false);
6441 : } else {
6442 : /* There is an alternative path. */
6443 1 : msg_fn = _bdev_nvme_reset_ctrlr;
6444 1 : rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, true);
6445 : }
6446 :
6447 55 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
6448 :
6449 55 : if (rc == 0) {
6450 55 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
6451 0 : } else if (rc == -EALREADY) {
6452 0 : rc = 0;
6453 : }
6454 :
6455 55 : return rc;
6456 : }
6457 :
6458 : int
6459 48 : bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id,
6460 : bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx)
6461 : {
6462 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
6463 : struct nvme_ctrlr *nvme_ctrlr, *tmp_nvme_ctrlr;
6464 48 : struct bdev_nvme_delete_ctx *ctx = NULL;
6465 48 : int rc = -ENXIO, _rc;
6466 :
6467 48 : if (name == NULL || path_id == NULL) {
6468 0 : rc = -EINVAL;
6469 0 : goto exit;
6470 : }
6471 :
6472 48 : pthread_mutex_lock(&g_bdev_nvme_mutex);
6473 :
6474 48 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
6475 48 : if (nbdev_ctrlr == NULL) {
6476 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6477 :
6478 0 : SPDK_ERRLOG("Failed to find NVMe bdev controller\n");
6479 0 : rc = -ENODEV;
6480 0 : goto exit;
6481 : }
6482 :
6483 111 : TAILQ_FOREACH_SAFE(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq, tmp_nvme_ctrlr) {
6484 63 : _rc = _bdev_nvme_delete(nvme_ctrlr, path_id);
6485 63 : if (_rc < 0 && _rc != -ENXIO) {
6486 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6487 0 : rc = _rc;
6488 0 : goto exit;
6489 63 : } else if (_rc == 0) {
6490 : /* We traverse all remaining nvme_ctrlrs even if one nvme_ctrlr
6491 : * was deleted successfully. To remember the successful deletion,
6492 : * overwrite rc only if _rc is zero.
6493 : */
6494 57 : rc = 0;
6495 : }
6496 : }
6497 :
6498 48 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6499 :
6500 48 : if (rc != 0 || delete_done == NULL) {
6501 47 : goto exit;
6502 : }
6503 :
6504 1 : ctx = calloc(1, sizeof(*ctx));
6505 1 : if (ctx == NULL) {
6506 0 : SPDK_ERRLOG("Failed to allocate context for bdev_nvme_delete\n");
6507 0 : rc = -ENOMEM;
6508 0 : goto exit;
6509 : }
6510 :
6511 1 : ctx->name = strdup(name);
6512 1 : if (ctx->name == NULL) {
6513 0 : SPDK_ERRLOG("Failed to copy controller name for deletion\n");
6514 0 : rc = -ENOMEM;
6515 0 : goto exit;
6516 : }
6517 :
6518 1 : ctx->delete_done = delete_done;
6519 1 : ctx->delete_done_ctx = delete_done_ctx;
6520 1 : ctx->path_id = *path_id;
6521 1 : ctx->timeout_ticks = spdk_get_ticks() + 10 * spdk_get_ticks_hz();
6522 1 : ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_delete_complete_poll, ctx, 1000);
6523 1 : if (ctx->poller == NULL) {
6524 0 : SPDK_ERRLOG("Failed to register bdev_nvme_delete poller\n");
6525 0 : rc = -ENOMEM;
6526 0 : goto exit;
6527 : }
6528 :
6529 1 : exit:
6530 48 : if (rc != 0) {
6531 1 : free_bdev_nvme_delete_ctx(ctx);
6532 : }
6533 :
6534 48 : return rc;
6535 : }
6536 :
6537 : #define DISCOVERY_INFOLOG(ctx, format, ...) \
6538 : SPDK_INFOLOG(bdev_nvme, "Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);
6539 :
6540 : #define DISCOVERY_ERRLOG(ctx, format, ...) \
6541 : SPDK_ERRLOG("Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);
6542 :
6543 : struct discovery_entry_ctx {
6544 : char name[128];
6545 : struct spdk_nvme_transport_id trid;
6546 : struct spdk_nvme_ctrlr_opts drv_opts;
6547 : struct spdk_nvmf_discovery_log_page_entry entry;
6548 : TAILQ_ENTRY(discovery_entry_ctx) tailq;
6549 : struct discovery_ctx *ctx;
6550 : };
6551 :
6552 : struct discovery_ctx {
6553 : char *name;
6554 : spdk_bdev_nvme_start_discovery_fn start_cb_fn;
6555 : spdk_bdev_nvme_stop_discovery_fn stop_cb_fn;
6556 : void *cb_ctx;
6557 : struct spdk_nvme_probe_ctx *probe_ctx;
6558 : struct spdk_nvme_detach_ctx *detach_ctx;
6559 : struct spdk_nvme_ctrlr *ctrlr;
6560 : struct spdk_nvme_transport_id trid;
6561 : struct discovery_entry_ctx *entry_ctx_in_use;
6562 : struct spdk_poller *poller;
6563 : struct spdk_nvme_ctrlr_opts drv_opts;
6564 : struct nvme_ctrlr_opts bdev_opts;
6565 : struct spdk_nvmf_discovery_log_page *log_page;
6566 : TAILQ_ENTRY(discovery_ctx) tailq;
6567 : TAILQ_HEAD(, discovery_entry_ctx) nvm_entry_ctxs;
6568 : TAILQ_HEAD(, discovery_entry_ctx) discovery_entry_ctxs;
6569 : int rc;
6570 : bool wait_for_attach;
6571 : uint64_t timeout_ticks;
6572 : /* Denotes that the discovery service is being started. We're waiting
6573 : * for the initial connection to the discovery controller to be
6574 : * established and attach discovered NVM ctrlrs.
6575 : */
6576 : bool initializing;
6577 : /* Denotes if a discovery is currently in progress for this context.
6578 : * That includes connecting to newly discovered subsystems. Used to
6579 : * ensure we do not start a new discovery until an existing one is
6580 : * complete.
6581 : */
6582 : bool in_progress;
6583 :
6584 : /* Denotes if another discovery is needed after the one in progress
6585 : * completes. Set when we receive an AER completion while a discovery
6586 : * is already in progress.
6587 : */
6588 : bool pending;
6589 :
6590 : /* Signal to the discovery context poller that it should stop the
6591 : * discovery service, including detaching from the current discovery
6592 : * controller.
6593 : */
6594 : bool stop;
6595 :
6596 : struct spdk_thread *calling_thread;
6597 : uint32_t index;
6598 : uint32_t attach_in_progress;
6599 : char *hostnqn;
6600 :
6601 : /* Denotes if the discovery service was started by the mdns discovery.
6602 : */
6603 : bool from_mdns_discovery_service;
6604 : };
6605 :
6606 : TAILQ_HEAD(discovery_ctxs, discovery_ctx);
6607 : static struct discovery_ctxs g_discovery_ctxs = TAILQ_HEAD_INITIALIZER(g_discovery_ctxs);
6608 :
6609 : static void get_discovery_log_page(struct discovery_ctx *ctx);
6610 :
6611 : static void
6612 0 : free_discovery_ctx(struct discovery_ctx *ctx)
6613 : {
6614 0 : free(ctx->log_page);
6615 0 : free(ctx->hostnqn);
6616 0 : free(ctx->name);
6617 0 : free(ctx);
6618 0 : }
6619 :
6620 : static void
6621 0 : discovery_complete(struct discovery_ctx *ctx)
6622 : {
6623 0 : ctx->initializing = false;
6624 0 : ctx->in_progress = false;
6625 0 : if (ctx->pending) {
6626 0 : ctx->pending = false;
6627 0 : get_discovery_log_page(ctx);
6628 : }
6629 0 : }
6630 :
6631 : static void
6632 0 : build_trid_from_log_page_entry(struct spdk_nvme_transport_id *trid,
6633 : struct spdk_nvmf_discovery_log_page_entry *entry)
6634 : {
6635 : char *space;
6636 :
6637 0 : trid->trtype = entry->trtype;
6638 0 : trid->adrfam = entry->adrfam;
6639 0 : memcpy(trid->traddr, entry->traddr, sizeof(entry->traddr));
6640 0 : memcpy(trid->trsvcid, entry->trsvcid, sizeof(entry->trsvcid));
6641 : /* Because the source buffer (entry->subnqn) is longer than trid->subnqn, and
6642 : * before call to this function trid->subnqn is zeroed out, we need
6643 : * to copy sizeof(trid->subnqn) minus one byte to make sure the last character
6644 : * remains 0. Then we can shorten the string (replace ' ' with 0) if required
6645 : */
6646 0 : memcpy(trid->subnqn, entry->subnqn, sizeof(trid->subnqn) - 1);
6647 :
6648 : /* We want the traddr, trsvcid and subnqn fields to be NULL-terminated.
6649 : * But the log page entries typically pad them with spaces, not zeroes.
6650 : * So add a NULL terminator to each of these fields at the appropriate
6651 : * location.
6652 : */
6653 0 : space = strchr(trid->traddr, ' ');
6654 0 : if (space) {
6655 0 : *space = 0;
6656 : }
6657 0 : space = strchr(trid->trsvcid, ' ');
6658 0 : if (space) {
6659 0 : *space = 0;
6660 : }
6661 0 : space = strchr(trid->subnqn, ' ');
6662 0 : if (space) {
6663 0 : *space = 0;
6664 : }
6665 0 : }
6666 :
6667 : static void
6668 0 : _stop_discovery(void *_ctx)
6669 : {
6670 0 : struct discovery_ctx *ctx = _ctx;
6671 :
6672 0 : if (ctx->attach_in_progress > 0) {
6673 0 : spdk_thread_send_msg(spdk_get_thread(), _stop_discovery, ctx);
6674 0 : return;
6675 : }
6676 :
6677 0 : ctx->stop = true;
6678 :
6679 0 : while (!TAILQ_EMPTY(&ctx->nvm_entry_ctxs)) {
6680 : struct discovery_entry_ctx *entry_ctx;
6681 0 : struct nvme_path_id path = {};
6682 :
6683 0 : entry_ctx = TAILQ_FIRST(&ctx->nvm_entry_ctxs);
6684 0 : path.trid = entry_ctx->trid;
6685 0 : bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
6686 0 : TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
6687 0 : free(entry_ctx);
6688 : }
6689 :
6690 0 : while (!TAILQ_EMPTY(&ctx->discovery_entry_ctxs)) {
6691 : struct discovery_entry_ctx *entry_ctx;
6692 :
6693 0 : entry_ctx = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
6694 0 : TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
6695 0 : free(entry_ctx);
6696 : }
6697 :
6698 0 : free(ctx->entry_ctx_in_use);
6699 0 : ctx->entry_ctx_in_use = NULL;
6700 : }
6701 :
6702 : static void
6703 0 : stop_discovery(struct discovery_ctx *ctx, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
6704 : {
6705 0 : ctx->stop_cb_fn = cb_fn;
6706 0 : ctx->cb_ctx = cb_ctx;
6707 :
6708 0 : if (ctx->attach_in_progress > 0) {
6709 0 : DISCOVERY_INFOLOG(ctx, "stopping discovery with attach_in_progress: %"PRIu32"\n",
6710 : ctx->attach_in_progress);
6711 : }
6712 :
6713 0 : _stop_discovery(ctx);
6714 0 : }
6715 :
6716 : static void
6717 2 : remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr)
6718 : {
6719 : struct discovery_ctx *d_ctx;
6720 : struct nvme_path_id *path_id;
6721 2 : struct spdk_nvme_transport_id trid = {};
6722 : struct discovery_entry_ctx *entry_ctx, *tmp;
6723 :
6724 2 : path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
6725 :
6726 2 : TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
6727 0 : TAILQ_FOREACH_SAFE(entry_ctx, &d_ctx->nvm_entry_ctxs, tailq, tmp) {
6728 0 : build_trid_from_log_page_entry(&trid, &entry_ctx->entry);
6729 0 : if (spdk_nvme_transport_id_compare(&trid, &path_id->trid) != 0) {
6730 0 : continue;
6731 : }
6732 :
6733 0 : TAILQ_REMOVE(&d_ctx->nvm_entry_ctxs, entry_ctx, tailq);
6734 0 : free(entry_ctx);
6735 0 : DISCOVERY_INFOLOG(d_ctx, "Remove discovery entry: %s:%s:%s\n",
6736 : trid.subnqn, trid.traddr, trid.trsvcid);
6737 :
6738 : /* Fail discovery ctrlr to force reattach attempt */
6739 0 : spdk_nvme_ctrlr_fail(d_ctx->ctrlr);
6740 : }
6741 : }
6742 2 : }
6743 :
6744 : static void
6745 0 : discovery_remove_controllers(struct discovery_ctx *ctx)
6746 : {
6747 0 : struct spdk_nvmf_discovery_log_page *log_page = ctx->log_page;
6748 : struct discovery_entry_ctx *entry_ctx, *tmp;
6749 : struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
6750 0 : struct spdk_nvme_transport_id old_trid = {};
6751 : uint64_t numrec, i;
6752 : bool found;
6753 :
6754 0 : numrec = from_le64(&log_page->numrec);
6755 0 : TAILQ_FOREACH_SAFE(entry_ctx, &ctx->nvm_entry_ctxs, tailq, tmp) {
6756 0 : found = false;
6757 0 : old_entry = &entry_ctx->entry;
6758 0 : build_trid_from_log_page_entry(&old_trid, old_entry);
6759 0 : for (i = 0; i < numrec; i++) {
6760 0 : new_entry = &log_page->entries[i];
6761 0 : if (!memcmp(old_entry, new_entry, sizeof(*old_entry))) {
6762 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s found again\n",
6763 : old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
6764 0 : found = true;
6765 0 : break;
6766 : }
6767 : }
6768 0 : if (!found) {
6769 0 : struct nvme_path_id path = {};
6770 :
6771 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s not found\n",
6772 : old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
6773 :
6774 0 : path.trid = entry_ctx->trid;
6775 0 : bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
6776 0 : TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
6777 0 : free(entry_ctx);
6778 : }
6779 : }
6780 0 : free(log_page);
6781 0 : ctx->log_page = NULL;
6782 0 : discovery_complete(ctx);
6783 0 : }
6784 :
6785 : static void
6786 0 : complete_discovery_start(struct discovery_ctx *ctx, int status)
6787 : {
6788 0 : ctx->timeout_ticks = 0;
6789 0 : ctx->rc = status;
6790 0 : if (ctx->start_cb_fn) {
6791 0 : ctx->start_cb_fn(ctx->cb_ctx, status);
6792 0 : ctx->start_cb_fn = NULL;
6793 0 : ctx->cb_ctx = NULL;
6794 : }
6795 0 : }
6796 :
6797 : static void
6798 0 : discovery_attach_controller_done(void *cb_ctx, size_t bdev_count, int rc)
6799 : {
6800 0 : struct discovery_entry_ctx *entry_ctx = cb_ctx;
6801 0 : struct discovery_ctx *ctx = entry_ctx->ctx;
6802 :
6803 0 : DISCOVERY_INFOLOG(ctx, "attach %s done\n", entry_ctx->name);
6804 0 : ctx->attach_in_progress--;
6805 0 : if (ctx->attach_in_progress == 0) {
6806 0 : complete_discovery_start(ctx, ctx->rc);
6807 0 : if (ctx->initializing && ctx->rc != 0) {
6808 0 : DISCOVERY_ERRLOG(ctx, "stopping discovery due to errors: %d\n", ctx->rc);
6809 0 : stop_discovery(ctx, NULL, ctx->cb_ctx);
6810 : } else {
6811 0 : discovery_remove_controllers(ctx);
6812 : }
6813 : }
6814 0 : }
6815 :
6816 : static struct discovery_entry_ctx *
6817 0 : create_discovery_entry_ctx(struct discovery_ctx *ctx, struct spdk_nvme_transport_id *trid)
6818 : {
6819 : struct discovery_entry_ctx *new_ctx;
6820 :
6821 0 : new_ctx = calloc(1, sizeof(*new_ctx));
6822 0 : if (new_ctx == NULL) {
6823 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
6824 0 : return NULL;
6825 : }
6826 :
6827 0 : new_ctx->ctx = ctx;
6828 0 : memcpy(&new_ctx->trid, trid, sizeof(*trid));
6829 0 : spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
6830 0 : snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
6831 0 : return new_ctx;
6832 : }
6833 :
6834 : static void
6835 0 : discovery_log_page_cb(void *cb_arg, int rc, const struct spdk_nvme_cpl *cpl,
6836 : struct spdk_nvmf_discovery_log_page *log_page)
6837 : {
6838 0 : struct discovery_ctx *ctx = cb_arg;
6839 : struct discovery_entry_ctx *entry_ctx, *tmp;
6840 : struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
6841 : uint64_t numrec, i;
6842 : bool found;
6843 :
6844 0 : if (rc || spdk_nvme_cpl_is_error(cpl)) {
6845 0 : DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
6846 0 : return;
6847 : }
6848 :
6849 0 : ctx->log_page = log_page;
6850 0 : assert(ctx->attach_in_progress == 0);
6851 0 : numrec = from_le64(&log_page->numrec);
6852 0 : TAILQ_FOREACH_SAFE(entry_ctx, &ctx->discovery_entry_ctxs, tailq, tmp) {
6853 0 : TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
6854 0 : free(entry_ctx);
6855 : }
6856 0 : for (i = 0; i < numrec; i++) {
6857 0 : found = false;
6858 0 : new_entry = &log_page->entries[i];
6859 0 : if (new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
6860 0 : new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
6861 : struct discovery_entry_ctx *new_ctx;
6862 0 : struct spdk_nvme_transport_id trid = {};
6863 :
6864 0 : build_trid_from_log_page_entry(&trid, new_entry);
6865 0 : new_ctx = create_discovery_entry_ctx(ctx, &trid);
6866 0 : if (new_ctx == NULL) {
6867 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
6868 0 : break;
6869 : }
6870 :
6871 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, new_ctx, tailq);
6872 0 : continue;
6873 : }
6874 0 : TAILQ_FOREACH(entry_ctx, &ctx->nvm_entry_ctxs, tailq) {
6875 0 : old_entry = &entry_ctx->entry;
6876 0 : if (!memcmp(new_entry, old_entry, sizeof(*new_entry))) {
6877 0 : found = true;
6878 0 : break;
6879 : }
6880 : }
6881 0 : if (!found) {
6882 0 : struct discovery_entry_ctx *subnqn_ctx = NULL, *new_ctx;
6883 : struct discovery_ctx *d_ctx;
6884 :
6885 0 : TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
6886 0 : TAILQ_FOREACH(subnqn_ctx, &d_ctx->nvm_entry_ctxs, tailq) {
6887 0 : if (!memcmp(subnqn_ctx->entry.subnqn, new_entry->subnqn,
6888 : sizeof(new_entry->subnqn))) {
6889 0 : break;
6890 : }
6891 : }
6892 0 : if (subnqn_ctx) {
6893 0 : break;
6894 : }
6895 : }
6896 :
6897 0 : new_ctx = calloc(1, sizeof(*new_ctx));
6898 0 : if (new_ctx == NULL) {
6899 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
6900 0 : break;
6901 : }
6902 :
6903 0 : new_ctx->ctx = ctx;
6904 0 : memcpy(&new_ctx->entry, new_entry, sizeof(*new_entry));
6905 0 : build_trid_from_log_page_entry(&new_ctx->trid, new_entry);
6906 0 : if (subnqn_ctx) {
6907 0 : snprintf(new_ctx->name, sizeof(new_ctx->name), "%s", subnqn_ctx->name);
6908 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new path for %s\n",
6909 : new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
6910 : new_ctx->name);
6911 : } else {
6912 0 : snprintf(new_ctx->name, sizeof(new_ctx->name), "%s%d", ctx->name, ctx->index++);
6913 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new subsystem %s\n",
6914 : new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
6915 : new_ctx->name);
6916 : }
6917 0 : spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
6918 0 : snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
6919 0 : rc = bdev_nvme_create(&new_ctx->trid, new_ctx->name, NULL, 0,
6920 : discovery_attach_controller_done, new_ctx,
6921 : &new_ctx->drv_opts, &ctx->bdev_opts, true);
6922 0 : if (rc == 0) {
6923 0 : TAILQ_INSERT_TAIL(&ctx->nvm_entry_ctxs, new_ctx, tailq);
6924 0 : ctx->attach_in_progress++;
6925 : } else {
6926 0 : DISCOVERY_ERRLOG(ctx, "bdev_nvme_create failed (%s)\n", spdk_strerror(-rc));
6927 : }
6928 : }
6929 : }
6930 :
6931 0 : if (ctx->attach_in_progress == 0) {
6932 0 : discovery_remove_controllers(ctx);
6933 : }
6934 : }
6935 :
6936 : static void
6937 0 : get_discovery_log_page(struct discovery_ctx *ctx)
6938 : {
6939 : int rc;
6940 :
6941 0 : assert(ctx->in_progress == false);
6942 0 : ctx->in_progress = true;
6943 0 : rc = spdk_nvme_ctrlr_get_discovery_log_page(ctx->ctrlr, discovery_log_page_cb, ctx);
6944 0 : if (rc != 0) {
6945 0 : DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
6946 : }
6947 0 : DISCOVERY_INFOLOG(ctx, "sent discovery log page command\n");
6948 0 : }
6949 :
6950 : static void
6951 0 : discovery_aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
6952 : {
6953 0 : struct discovery_ctx *ctx = arg;
6954 0 : uint32_t log_page_id = (cpl->cdw0 & 0xFF0000) >> 16;
6955 :
6956 0 : if (spdk_nvme_cpl_is_error(cpl)) {
6957 0 : DISCOVERY_ERRLOG(ctx, "aer failed\n");
6958 0 : return;
6959 : }
6960 :
6961 0 : if (log_page_id != SPDK_NVME_LOG_DISCOVERY) {
6962 0 : DISCOVERY_ERRLOG(ctx, "unexpected log page 0x%x\n", log_page_id);
6963 0 : return;
6964 : }
6965 :
6966 0 : DISCOVERY_INFOLOG(ctx, "got aer\n");
6967 0 : if (ctx->in_progress) {
6968 0 : ctx->pending = true;
6969 0 : return;
6970 : }
6971 :
6972 0 : get_discovery_log_page(ctx);
6973 : }
6974 :
6975 : static void
6976 0 : discovery_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
6977 : struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
6978 : {
6979 0 : struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6980 : struct discovery_ctx *ctx;
6981 :
6982 0 : ctx = SPDK_CONTAINEROF(user_opts, struct discovery_ctx, drv_opts);
6983 :
6984 0 : DISCOVERY_INFOLOG(ctx, "discovery ctrlr attached\n");
6985 0 : ctx->probe_ctx = NULL;
6986 0 : ctx->ctrlr = ctrlr;
6987 :
6988 0 : if (ctx->rc != 0) {
6989 0 : DISCOVERY_ERRLOG(ctx, "encountered error while attaching discovery ctrlr: %d\n",
6990 : ctx->rc);
6991 0 : return;
6992 : }
6993 :
6994 0 : spdk_nvme_ctrlr_register_aer_callback(ctx->ctrlr, discovery_aer_cb, ctx);
6995 : }
6996 :
6997 : static int
6998 0 : discovery_poller(void *arg)
6999 : {
7000 0 : struct discovery_ctx *ctx = arg;
7001 : struct spdk_nvme_transport_id *trid;
7002 : int rc;
7003 :
7004 0 : if (ctx->detach_ctx) {
7005 0 : rc = spdk_nvme_detach_poll_async(ctx->detach_ctx);
7006 0 : if (rc != -EAGAIN) {
7007 0 : ctx->detach_ctx = NULL;
7008 0 : ctx->ctrlr = NULL;
7009 : }
7010 0 : } else if (ctx->stop) {
7011 0 : if (ctx->ctrlr != NULL) {
7012 0 : rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
7013 0 : if (rc == 0) {
7014 0 : return SPDK_POLLER_BUSY;
7015 : }
7016 0 : DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
7017 : }
7018 0 : spdk_poller_unregister(&ctx->poller);
7019 0 : TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
7020 0 : assert(ctx->start_cb_fn == NULL);
7021 0 : if (ctx->stop_cb_fn != NULL) {
7022 0 : ctx->stop_cb_fn(ctx->cb_ctx);
7023 : }
7024 0 : free_discovery_ctx(ctx);
7025 0 : } else if (ctx->probe_ctx == NULL && ctx->ctrlr == NULL) {
7026 0 : if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7027 0 : DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
7028 0 : assert(ctx->initializing);
7029 0 : spdk_poller_unregister(&ctx->poller);
7030 0 : TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
7031 0 : complete_discovery_start(ctx, -ETIMEDOUT);
7032 0 : stop_discovery(ctx, NULL, NULL);
7033 0 : free_discovery_ctx(ctx);
7034 0 : return SPDK_POLLER_BUSY;
7035 : }
7036 :
7037 0 : assert(ctx->entry_ctx_in_use == NULL);
7038 0 : ctx->entry_ctx_in_use = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
7039 0 : TAILQ_REMOVE(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7040 0 : trid = &ctx->entry_ctx_in_use->trid;
7041 0 : ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, discovery_attach_cb);
7042 0 : if (ctx->probe_ctx) {
7043 0 : spdk_poller_unregister(&ctx->poller);
7044 0 : ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000);
7045 : } else {
7046 0 : DISCOVERY_ERRLOG(ctx, "could not start discovery connect\n");
7047 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7048 0 : ctx->entry_ctx_in_use = NULL;
7049 : }
7050 0 : } else if (ctx->probe_ctx) {
7051 0 : if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7052 0 : DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
7053 0 : complete_discovery_start(ctx, -ETIMEDOUT);
7054 0 : return SPDK_POLLER_BUSY;
7055 : }
7056 :
7057 0 : rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
7058 0 : if (rc != -EAGAIN) {
7059 0 : if (ctx->rc != 0) {
7060 0 : assert(ctx->initializing);
7061 0 : stop_discovery(ctx, NULL, ctx->cb_ctx);
7062 : } else {
7063 0 : assert(rc == 0);
7064 0 : DISCOVERY_INFOLOG(ctx, "discovery ctrlr connected\n");
7065 0 : ctx->rc = rc;
7066 0 : get_discovery_log_page(ctx);
7067 : }
7068 : }
7069 : } else {
7070 0 : if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7071 0 : DISCOVERY_ERRLOG(ctx, "timed out while attaching NVM ctrlrs\n");
7072 0 : complete_discovery_start(ctx, -ETIMEDOUT);
7073 : /* We need to wait until all NVM ctrlrs are attached before we stop the
7074 : * discovery service to make sure we don't detach a ctrlr that is still
7075 : * being attached.
7076 : */
7077 0 : if (ctx->attach_in_progress == 0) {
7078 0 : stop_discovery(ctx, NULL, ctx->cb_ctx);
7079 0 : return SPDK_POLLER_BUSY;
7080 : }
7081 : }
7082 :
7083 0 : rc = spdk_nvme_ctrlr_process_admin_completions(ctx->ctrlr);
7084 0 : if (rc < 0) {
7085 0 : spdk_poller_unregister(&ctx->poller);
7086 0 : ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
7087 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7088 0 : ctx->entry_ctx_in_use = NULL;
7089 :
7090 0 : rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
7091 0 : if (rc != 0) {
7092 0 : DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
7093 0 : ctx->ctrlr = NULL;
7094 : }
7095 : }
7096 : }
7097 :
7098 0 : return SPDK_POLLER_BUSY;
7099 : }
7100 :
7101 : static void
7102 0 : start_discovery_poller(void *arg)
7103 : {
7104 0 : struct discovery_ctx *ctx = arg;
7105 :
7106 0 : TAILQ_INSERT_TAIL(&g_discovery_ctxs, ctx, tailq);
7107 0 : ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
7108 0 : }
7109 :
7110 : int
7111 0 : bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid,
7112 : const char *base_name,
7113 : struct spdk_nvme_ctrlr_opts *drv_opts,
7114 : struct nvme_ctrlr_opts *bdev_opts,
7115 : uint64_t attach_timeout,
7116 : bool from_mdns,
7117 : spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx)
7118 : {
7119 : struct discovery_ctx *ctx;
7120 : struct discovery_entry_ctx *discovery_entry_ctx;
7121 :
7122 0 : snprintf(trid->subnqn, sizeof(trid->subnqn), "%s", SPDK_NVMF_DISCOVERY_NQN);
7123 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7124 0 : if (strcmp(ctx->name, base_name) == 0) {
7125 0 : return -EEXIST;
7126 : }
7127 :
7128 0 : if (ctx->entry_ctx_in_use != NULL) {
7129 0 : if (!spdk_nvme_transport_id_compare(trid, &ctx->entry_ctx_in_use->trid)) {
7130 0 : return -EEXIST;
7131 : }
7132 : }
7133 :
7134 0 : TAILQ_FOREACH(discovery_entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
7135 0 : if (!spdk_nvme_transport_id_compare(trid, &discovery_entry_ctx->trid)) {
7136 0 : return -EEXIST;
7137 : }
7138 : }
7139 : }
7140 :
7141 0 : ctx = calloc(1, sizeof(*ctx));
7142 0 : if (ctx == NULL) {
7143 0 : return -ENOMEM;
7144 : }
7145 :
7146 0 : ctx->name = strdup(base_name);
7147 0 : if (ctx->name == NULL) {
7148 0 : free_discovery_ctx(ctx);
7149 0 : return -ENOMEM;
7150 : }
7151 0 : memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
7152 0 : memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
7153 0 : ctx->from_mdns_discovery_service = from_mdns;
7154 0 : ctx->bdev_opts.from_discovery_service = true;
7155 0 : ctx->calling_thread = spdk_get_thread();
7156 0 : ctx->start_cb_fn = cb_fn;
7157 0 : ctx->cb_ctx = cb_ctx;
7158 0 : ctx->initializing = true;
7159 0 : if (ctx->start_cb_fn) {
7160 : /* We can use this when dumping json to denote if this RPC parameter
7161 : * was specified or not.
7162 : */
7163 0 : ctx->wait_for_attach = true;
7164 : }
7165 0 : if (attach_timeout != 0) {
7166 0 : ctx->timeout_ticks = spdk_get_ticks() + attach_timeout *
7167 0 : spdk_get_ticks_hz() / 1000ull;
7168 : }
7169 0 : TAILQ_INIT(&ctx->nvm_entry_ctxs);
7170 0 : TAILQ_INIT(&ctx->discovery_entry_ctxs);
7171 0 : memcpy(&ctx->trid, trid, sizeof(*trid));
7172 : /* Even if user did not specify hostnqn, we can still strdup("\0"); */
7173 0 : ctx->hostnqn = strdup(ctx->drv_opts.hostnqn);
7174 0 : if (ctx->hostnqn == NULL) {
7175 0 : free_discovery_ctx(ctx);
7176 0 : return -ENOMEM;
7177 : }
7178 0 : discovery_entry_ctx = create_discovery_entry_ctx(ctx, trid);
7179 0 : if (discovery_entry_ctx == NULL) {
7180 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7181 0 : free_discovery_ctx(ctx);
7182 0 : return -ENOMEM;
7183 : }
7184 :
7185 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, discovery_entry_ctx, tailq);
7186 0 : spdk_thread_send_msg(g_bdev_nvme_init_thread, start_discovery_poller, ctx);
7187 0 : return 0;
7188 : }
7189 :
7190 : int
7191 0 : bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
7192 : {
7193 : struct discovery_ctx *ctx;
7194 :
7195 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7196 0 : if (strcmp(name, ctx->name) == 0) {
7197 0 : if (ctx->stop) {
7198 0 : return -EALREADY;
7199 : }
7200 : /* If we're still starting the discovery service and ->rc is non-zero, we're
7201 : * going to stop it as soon as we can
7202 : */
7203 0 : if (ctx->initializing && ctx->rc != 0) {
7204 0 : return -EALREADY;
7205 : }
7206 0 : stop_discovery(ctx, cb_fn, cb_ctx);
7207 0 : return 0;
7208 : }
7209 : }
7210 :
7211 0 : return -ENOENT;
7212 : }
7213 :
7214 : static int
7215 1 : bdev_nvme_library_init(void)
7216 : {
7217 1 : g_bdev_nvme_init_thread = spdk_get_thread();
7218 :
7219 1 : spdk_io_device_register(&g_nvme_bdev_ctrlrs, bdev_nvme_create_poll_group_cb,
7220 : bdev_nvme_destroy_poll_group_cb,
7221 : sizeof(struct nvme_poll_group), "nvme_poll_groups");
7222 :
7223 1 : return 0;
7224 : }
7225 :
7226 : static void
7227 1 : bdev_nvme_fini_destruct_ctrlrs(void)
7228 : {
7229 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
7230 : struct nvme_ctrlr *nvme_ctrlr;
7231 :
7232 1 : pthread_mutex_lock(&g_bdev_nvme_mutex);
7233 1 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
7234 0 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
7235 0 : pthread_mutex_lock(&nvme_ctrlr->mutex);
7236 0 : if (nvme_ctrlr->destruct) {
7237 : /* This controller's destruction was already started
7238 : * before the application started shutting down
7239 : */
7240 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
7241 0 : continue;
7242 : }
7243 0 : nvme_ctrlr->destruct = true;
7244 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
7245 :
7246 0 : spdk_thread_send_msg(nvme_ctrlr->thread, _nvme_ctrlr_destruct,
7247 : nvme_ctrlr);
7248 : }
7249 : }
7250 :
7251 1 : g_bdev_nvme_module_finish = true;
7252 1 : if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
7253 1 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
7254 1 : spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
7255 1 : spdk_bdev_module_fini_done();
7256 1 : return;
7257 : }
7258 :
7259 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
7260 : }
7261 :
7262 : static void
7263 0 : check_discovery_fini(void *arg)
7264 : {
7265 0 : if (TAILQ_EMPTY(&g_discovery_ctxs)) {
7266 0 : bdev_nvme_fini_destruct_ctrlrs();
7267 : }
7268 0 : }
7269 :
7270 : static void
7271 1 : bdev_nvme_library_fini(void)
7272 : {
7273 : struct nvme_probe_skip_entry *entry, *entry_tmp;
7274 : struct discovery_ctx *ctx;
7275 :
7276 1 : spdk_poller_unregister(&g_hotplug_poller);
7277 1 : free(g_hotplug_probe_ctx);
7278 1 : g_hotplug_probe_ctx = NULL;
7279 :
7280 1 : TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, entry_tmp) {
7281 0 : TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
7282 0 : free(entry);
7283 : }
7284 :
7285 1 : assert(spdk_get_thread() == g_bdev_nvme_init_thread);
7286 1 : if (TAILQ_EMPTY(&g_discovery_ctxs)) {
7287 1 : bdev_nvme_fini_destruct_ctrlrs();
7288 : } else {
7289 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7290 0 : stop_discovery(ctx, check_discovery_fini, NULL);
7291 : }
7292 : }
7293 1 : }
7294 :
7295 : static void
7296 0 : bdev_nvme_verify_pi_error(struct nvme_bdev_io *bio)
7297 : {
7298 0 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7299 0 : struct spdk_bdev *bdev = bdev_io->bdev;
7300 0 : struct spdk_dif_ctx dif_ctx;
7301 0 : struct spdk_dif_error err_blk = {};
7302 : int rc;
7303 0 : struct spdk_dif_ctx_init_ext_opts dif_opts;
7304 :
7305 0 : dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
7306 0 : dif_opts.dif_pi_format = SPDK_DIF_PI_FORMAT_16;
7307 0 : rc = spdk_dif_ctx_init(&dif_ctx,
7308 0 : bdev->blocklen, bdev->md_len, bdev->md_interleave,
7309 0 : bdev->dif_is_head_of_md, bdev->dif_type,
7310 : bdev_io->u.bdev.dif_check_flags,
7311 0 : bdev_io->u.bdev.offset_blocks, 0, 0, 0, 0, &dif_opts);
7312 0 : if (rc != 0) {
7313 0 : SPDK_ERRLOG("Initialization of DIF context failed\n");
7314 0 : return;
7315 : }
7316 :
7317 0 : if (bdev->md_interleave) {
7318 0 : rc = spdk_dif_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
7319 0 : bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
7320 : } else {
7321 0 : struct iovec md_iov = {
7322 0 : .iov_base = bdev_io->u.bdev.md_buf,
7323 0 : .iov_len = bdev_io->u.bdev.num_blocks * bdev->md_len,
7324 : };
7325 :
7326 0 : rc = spdk_dix_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
7327 0 : &md_iov, bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
7328 : }
7329 :
7330 0 : if (rc != 0) {
7331 0 : SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
7332 : err_blk.err_type, err_blk.err_offset);
7333 : } else {
7334 0 : SPDK_ERRLOG("Hardware reported PI error but SPDK could not find any.\n");
7335 : }
7336 : }
7337 :
7338 : static void
7339 0 : bdev_nvme_no_pi_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7340 : {
7341 0 : struct nvme_bdev_io *bio = ref;
7342 :
7343 0 : if (spdk_nvme_cpl_is_success(cpl)) {
7344 : /* Run PI verification for read data buffer. */
7345 0 : bdev_nvme_verify_pi_error(bio);
7346 : }
7347 :
7348 : /* Return original completion status */
7349 0 : bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
7350 0 : }
7351 :
7352 : static void
7353 3 : bdev_nvme_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7354 : {
7355 3 : struct nvme_bdev_io *bio = ref;
7356 3 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7357 : int ret;
7358 :
7359 3 : if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
7360 0 : SPDK_ERRLOG("readv completed with PI error (sct=%d, sc=%d)\n",
7361 : cpl->status.sct, cpl->status.sc);
7362 :
7363 : /* Save completion status to use after verifying PI error. */
7364 0 : bio->cpl = *cpl;
7365 :
7366 0 : if (spdk_likely(nvme_io_path_is_available(bio->io_path))) {
7367 : /* Read without PI checking to verify PI error. */
7368 0 : ret = bdev_nvme_no_pi_readv(bio,
7369 : bdev_io->u.bdev.iovs,
7370 : bdev_io->u.bdev.iovcnt,
7371 : bdev_io->u.bdev.md_buf,
7372 : bdev_io->u.bdev.num_blocks,
7373 : bdev_io->u.bdev.offset_blocks);
7374 0 : if (ret == 0) {
7375 0 : return;
7376 : }
7377 : }
7378 : }
7379 :
7380 3 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7381 : }
7382 :
7383 : static void
7384 25 : bdev_nvme_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7385 : {
7386 25 : struct nvme_bdev_io *bio = ref;
7387 :
7388 25 : if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
7389 0 : SPDK_ERRLOG("writev completed with PI error (sct=%d, sc=%d)\n",
7390 : cpl->status.sct, cpl->status.sc);
7391 : /* Run PI verification for write data buffer if PI error is detected. */
7392 0 : bdev_nvme_verify_pi_error(bio);
7393 : }
7394 :
7395 25 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7396 25 : }
7397 :
7398 : static void
7399 0 : bdev_nvme_zone_appendv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7400 : {
7401 0 : struct nvme_bdev_io *bio = ref;
7402 0 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7403 :
7404 : /* spdk_bdev_io_get_append_location() requires that the ALBA is stored in offset_blocks.
7405 : * Additionally, offset_blocks has to be set before calling bdev_nvme_verify_pi_error().
7406 : */
7407 0 : bdev_io->u.bdev.offset_blocks = *(uint64_t *)&cpl->cdw0;
7408 :
7409 0 : if (spdk_nvme_cpl_is_pi_error(cpl)) {
7410 0 : SPDK_ERRLOG("zone append completed with PI error (sct=%d, sc=%d)\n",
7411 : cpl->status.sct, cpl->status.sc);
7412 : /* Run PI verification for zone append data buffer if PI error is detected. */
7413 0 : bdev_nvme_verify_pi_error(bio);
7414 : }
7415 :
7416 0 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7417 0 : }
7418 :
7419 : static void
7420 1 : bdev_nvme_comparev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7421 : {
7422 1 : struct nvme_bdev_io *bio = ref;
7423 :
7424 1 : if (spdk_nvme_cpl_is_pi_error(cpl)) {
7425 0 : SPDK_ERRLOG("comparev completed with PI error (sct=%d, sc=%d)\n",
7426 : cpl->status.sct, cpl->status.sc);
7427 : /* Run PI verification for compare data buffer if PI error is detected. */
7428 0 : bdev_nvme_verify_pi_error(bio);
7429 : }
7430 :
7431 1 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7432 1 : }
7433 :
7434 : static void
7435 4 : bdev_nvme_comparev_and_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7436 : {
7437 4 : struct nvme_bdev_io *bio = ref;
7438 :
7439 : /* Compare operation completion */
7440 4 : if (!bio->first_fused_completed) {
7441 : /* Save compare result for write callback */
7442 2 : bio->cpl = *cpl;
7443 2 : bio->first_fused_completed = true;
7444 2 : return;
7445 : }
7446 :
7447 : /* Write operation completion */
7448 2 : if (spdk_nvme_cpl_is_error(&bio->cpl)) {
7449 : /* If bio->cpl is already an error, it means the compare operation failed. In that case,
7450 : * complete the IO with the compare operation's status.
7451 : */
7452 1 : if (!spdk_nvme_cpl_is_error(cpl)) {
7453 1 : SPDK_ERRLOG("Unexpected write success after compare failure.\n");
7454 : }
7455 :
7456 1 : bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
7457 : } else {
7458 1 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7459 : }
7460 : }
7461 :
7462 : static void
7463 1 : bdev_nvme_queued_done(void *ref, const struct spdk_nvme_cpl *cpl)
7464 : {
7465 1 : struct nvme_bdev_io *bio = ref;
7466 :
7467 1 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7468 1 : }
7469 :
7470 : static int
7471 0 : fill_zone_from_report(struct spdk_bdev_zone_info *info, struct spdk_nvme_zns_zone_desc *desc)
7472 : {
7473 0 : switch (desc->zt) {
7474 0 : case SPDK_NVME_ZONE_TYPE_SEQWR:
7475 0 : info->type = SPDK_BDEV_ZONE_TYPE_SEQWR;
7476 0 : break;
7477 0 : default:
7478 0 : SPDK_ERRLOG("Invalid zone type: %#x in zone report\n", desc->zt);
7479 0 : return -EIO;
7480 : }
7481 :
7482 0 : switch (desc->zs) {
7483 0 : case SPDK_NVME_ZONE_STATE_EMPTY:
7484 0 : info->state = SPDK_BDEV_ZONE_STATE_EMPTY;
7485 0 : break;
7486 0 : case SPDK_NVME_ZONE_STATE_IOPEN:
7487 0 : info->state = SPDK_BDEV_ZONE_STATE_IMP_OPEN;
7488 0 : break;
7489 0 : case SPDK_NVME_ZONE_STATE_EOPEN:
7490 0 : info->state = SPDK_BDEV_ZONE_STATE_EXP_OPEN;
7491 0 : break;
7492 0 : case SPDK_NVME_ZONE_STATE_CLOSED:
7493 0 : info->state = SPDK_BDEV_ZONE_STATE_CLOSED;
7494 0 : break;
7495 0 : case SPDK_NVME_ZONE_STATE_RONLY:
7496 0 : info->state = SPDK_BDEV_ZONE_STATE_READ_ONLY;
7497 0 : break;
7498 0 : case SPDK_NVME_ZONE_STATE_FULL:
7499 0 : info->state = SPDK_BDEV_ZONE_STATE_FULL;
7500 0 : break;
7501 0 : case SPDK_NVME_ZONE_STATE_OFFLINE:
7502 0 : info->state = SPDK_BDEV_ZONE_STATE_OFFLINE;
7503 0 : break;
7504 0 : default:
7505 0 : SPDK_ERRLOG("Invalid zone state: %#x in zone report\n", desc->zs);
7506 0 : return -EIO;
7507 : }
7508 :
7509 0 : info->zone_id = desc->zslba;
7510 0 : info->write_pointer = desc->wp;
7511 0 : info->capacity = desc->zcap;
7512 :
7513 0 : return 0;
7514 : }
7515 :
7516 : static void
7517 0 : bdev_nvme_get_zone_info_done(void *ref, const struct spdk_nvme_cpl *cpl)
7518 : {
7519 0 : struct nvme_bdev_io *bio = ref;
7520 0 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7521 0 : uint64_t zone_id = bdev_io->u.zone_mgmt.zone_id;
7522 0 : uint32_t zones_to_copy = bdev_io->u.zone_mgmt.num_zones;
7523 0 : struct spdk_bdev_zone_info *info = bdev_io->u.zone_mgmt.buf;
7524 : uint64_t max_zones_per_buf, i;
7525 : uint32_t zone_report_bufsize;
7526 : struct spdk_nvme_ns *ns;
7527 : struct spdk_nvme_qpair *qpair;
7528 : int ret;
7529 :
7530 0 : if (spdk_nvme_cpl_is_error(cpl)) {
7531 0 : goto out_complete_io_nvme_cpl;
7532 : }
7533 :
7534 0 : if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
7535 0 : ret = -ENXIO;
7536 0 : goto out_complete_io_ret;
7537 : }
7538 :
7539 0 : ns = bio->io_path->nvme_ns->ns;
7540 0 : qpair = bio->io_path->qpair->qpair;
7541 :
7542 0 : zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
7543 0 : max_zones_per_buf = (zone_report_bufsize - sizeof(*bio->zone_report_buf)) /
7544 : sizeof(bio->zone_report_buf->descs[0]);
7545 :
7546 0 : if (bio->zone_report_buf->nr_zones > max_zones_per_buf) {
7547 0 : ret = -EINVAL;
7548 0 : goto out_complete_io_ret;
7549 : }
7550 :
7551 0 : if (!bio->zone_report_buf->nr_zones) {
7552 0 : ret = -EINVAL;
7553 0 : goto out_complete_io_ret;
7554 : }
7555 :
7556 0 : for (i = 0; i < bio->zone_report_buf->nr_zones && bio->handled_zones < zones_to_copy; i++) {
7557 0 : ret = fill_zone_from_report(&info[bio->handled_zones],
7558 0 : &bio->zone_report_buf->descs[i]);
7559 0 : if (ret) {
7560 0 : goto out_complete_io_ret;
7561 : }
7562 0 : bio->handled_zones++;
7563 : }
7564 :
7565 0 : if (bio->handled_zones < zones_to_copy) {
7566 0 : uint64_t zone_size_lba = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
7567 0 : uint64_t slba = zone_id + (zone_size_lba * bio->handled_zones);
7568 :
7569 0 : memset(bio->zone_report_buf, 0, zone_report_bufsize);
7570 0 : ret = spdk_nvme_zns_report_zones(ns, qpair,
7571 0 : bio->zone_report_buf, zone_report_bufsize,
7572 : slba, SPDK_NVME_ZRA_LIST_ALL, true,
7573 : bdev_nvme_get_zone_info_done, bio);
7574 0 : if (!ret) {
7575 0 : return;
7576 : } else {
7577 0 : goto out_complete_io_ret;
7578 : }
7579 : }
7580 :
7581 0 : out_complete_io_nvme_cpl:
7582 0 : free(bio->zone_report_buf);
7583 0 : bio->zone_report_buf = NULL;
7584 0 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7585 0 : return;
7586 :
7587 0 : out_complete_io_ret:
7588 0 : free(bio->zone_report_buf);
7589 0 : bio->zone_report_buf = NULL;
7590 0 : bdev_nvme_io_complete(bio, ret);
7591 : }
7592 :
7593 : static void
7594 0 : bdev_nvme_zone_management_done(void *ref, const struct spdk_nvme_cpl *cpl)
7595 : {
7596 0 : struct nvme_bdev_io *bio = ref;
7597 :
7598 0 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7599 0 : }
7600 :
7601 : static void
7602 4 : bdev_nvme_admin_passthru_complete_nvme_status(void *ctx)
7603 : {
7604 4 : struct nvme_bdev_io *bio = ctx;
7605 4 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7606 4 : const struct spdk_nvme_cpl *cpl = &bio->cpl;
7607 :
7608 4 : assert(bdev_nvme_io_type_is_admin(bdev_io->type));
7609 :
7610 4 : __bdev_nvme_io_complete(bdev_io, 0, cpl);
7611 4 : }
7612 :
7613 : static void
7614 3 : bdev_nvme_abort_complete(void *ctx)
7615 : {
7616 3 : struct nvme_bdev_io *bio = ctx;
7617 3 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7618 :
7619 3 : if (spdk_nvme_cpl_is_abort_success(&bio->cpl)) {
7620 3 : __bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS, NULL);
7621 : } else {
7622 0 : __bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED, NULL);
7623 : }
7624 3 : }
7625 :
7626 : static void
7627 3 : bdev_nvme_abort_done(void *ref, const struct spdk_nvme_cpl *cpl)
7628 : {
7629 3 : struct nvme_bdev_io *bio = ref;
7630 3 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7631 :
7632 3 : bio->cpl = *cpl;
7633 3 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), bdev_nvme_abort_complete, bio);
7634 3 : }
7635 :
7636 : static void
7637 4 : bdev_nvme_admin_passthru_done(void *ref, const struct spdk_nvme_cpl *cpl)
7638 : {
7639 4 : struct nvme_bdev_io *bio = ref;
7640 4 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7641 :
7642 4 : bio->cpl = *cpl;
7643 4 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7644 : bdev_nvme_admin_passthru_complete_nvme_status, bio);
7645 4 : }
7646 :
7647 : static void
7648 0 : bdev_nvme_queued_reset_sgl(void *ref, uint32_t sgl_offset)
7649 : {
7650 0 : struct nvme_bdev_io *bio = ref;
7651 : struct iovec *iov;
7652 :
7653 0 : bio->iov_offset = sgl_offset;
7654 0 : for (bio->iovpos = 0; bio->iovpos < bio->iovcnt; bio->iovpos++) {
7655 0 : iov = &bio->iovs[bio->iovpos];
7656 0 : if (bio->iov_offset < iov->iov_len) {
7657 0 : break;
7658 : }
7659 :
7660 0 : bio->iov_offset -= iov->iov_len;
7661 : }
7662 0 : }
7663 :
7664 : static int
7665 0 : bdev_nvme_queued_next_sge(void *ref, void **address, uint32_t *length)
7666 : {
7667 0 : struct nvme_bdev_io *bio = ref;
7668 : struct iovec *iov;
7669 :
7670 0 : assert(bio->iovpos < bio->iovcnt);
7671 :
7672 0 : iov = &bio->iovs[bio->iovpos];
7673 :
7674 0 : *address = iov->iov_base;
7675 0 : *length = iov->iov_len;
7676 :
7677 0 : if (bio->iov_offset) {
7678 0 : assert(bio->iov_offset <= iov->iov_len);
7679 0 : *address += bio->iov_offset;
7680 0 : *length -= bio->iov_offset;
7681 : }
7682 :
7683 0 : bio->iov_offset += *length;
7684 0 : if (bio->iov_offset == iov->iov_len) {
7685 0 : bio->iovpos++;
7686 0 : bio->iov_offset = 0;
7687 : }
7688 :
7689 0 : return 0;
7690 : }
7691 :
7692 : static void
7693 0 : bdev_nvme_queued_reset_fused_sgl(void *ref, uint32_t sgl_offset)
7694 : {
7695 0 : struct nvme_bdev_io *bio = ref;
7696 : struct iovec *iov;
7697 :
7698 0 : bio->fused_iov_offset = sgl_offset;
7699 0 : for (bio->fused_iovpos = 0; bio->fused_iovpos < bio->fused_iovcnt; bio->fused_iovpos++) {
7700 0 : iov = &bio->fused_iovs[bio->fused_iovpos];
7701 0 : if (bio->fused_iov_offset < iov->iov_len) {
7702 0 : break;
7703 : }
7704 :
7705 0 : bio->fused_iov_offset -= iov->iov_len;
7706 : }
7707 0 : }
7708 :
7709 : static int
7710 0 : bdev_nvme_queued_next_fused_sge(void *ref, void **address, uint32_t *length)
7711 : {
7712 0 : struct nvme_bdev_io *bio = ref;
7713 : struct iovec *iov;
7714 :
7715 0 : assert(bio->fused_iovpos < bio->fused_iovcnt);
7716 :
7717 0 : iov = &bio->fused_iovs[bio->fused_iovpos];
7718 :
7719 0 : *address = iov->iov_base;
7720 0 : *length = iov->iov_len;
7721 :
7722 0 : if (bio->fused_iov_offset) {
7723 0 : assert(bio->fused_iov_offset <= iov->iov_len);
7724 0 : *address += bio->fused_iov_offset;
7725 0 : *length -= bio->fused_iov_offset;
7726 : }
7727 :
7728 0 : bio->fused_iov_offset += *length;
7729 0 : if (bio->fused_iov_offset == iov->iov_len) {
7730 0 : bio->fused_iovpos++;
7731 0 : bio->fused_iov_offset = 0;
7732 : }
7733 :
7734 0 : return 0;
7735 : }
7736 :
7737 : static int
7738 0 : bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7739 : void *md, uint64_t lba_count, uint64_t lba)
7740 : {
7741 : int rc;
7742 :
7743 0 : SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 " without PI check\n",
7744 : lba_count, lba);
7745 :
7746 0 : bio->iovs = iov;
7747 0 : bio->iovcnt = iovcnt;
7748 0 : bio->iovpos = 0;
7749 0 : bio->iov_offset = 0;
7750 :
7751 0 : rc = spdk_nvme_ns_cmd_readv_with_md(bio->io_path->nvme_ns->ns,
7752 0 : bio->io_path->qpair->qpair,
7753 : lba, lba_count,
7754 : bdev_nvme_no_pi_readv_done, bio, 0,
7755 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
7756 : md, 0, 0);
7757 :
7758 0 : if (rc != 0 && rc != -ENOMEM) {
7759 0 : SPDK_ERRLOG("no_pi_readv failed: rc = %d\n", rc);
7760 : }
7761 0 : return rc;
7762 : }
7763 :
7764 : static int
7765 3 : bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7766 : void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
7767 : struct spdk_memory_domain *domain, void *domain_ctx,
7768 : struct spdk_accel_sequence *seq)
7769 : {
7770 3 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
7771 3 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
7772 : int rc;
7773 :
7774 3 : SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 "\n",
7775 : lba_count, lba);
7776 :
7777 3 : bio->iovs = iov;
7778 3 : bio->iovcnt = iovcnt;
7779 3 : bio->iovpos = 0;
7780 3 : bio->iov_offset = 0;
7781 :
7782 3 : if (domain != NULL || seq != NULL) {
7783 1 : bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
7784 1 : bio->ext_opts.memory_domain = domain;
7785 1 : bio->ext_opts.memory_domain_ctx = domain_ctx;
7786 1 : bio->ext_opts.io_flags = flags;
7787 1 : bio->ext_opts.metadata = md;
7788 1 : bio->ext_opts.accel_sequence = seq;
7789 :
7790 1 : if (iovcnt == 1) {
7791 1 : rc = spdk_nvme_ns_cmd_read_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_readv_done,
7792 : bio, &bio->ext_opts);
7793 : } else {
7794 0 : rc = spdk_nvme_ns_cmd_readv_ext(ns, qpair, lba, lba_count,
7795 : bdev_nvme_readv_done, bio,
7796 : bdev_nvme_queued_reset_sgl,
7797 : bdev_nvme_queued_next_sge,
7798 : &bio->ext_opts);
7799 : }
7800 2 : } else if (iovcnt == 1) {
7801 2 : rc = spdk_nvme_ns_cmd_read_with_md(ns, qpair, iov[0].iov_base,
7802 : md, lba, lba_count, bdev_nvme_readv_done,
7803 : bio, flags, 0, 0);
7804 : } else {
7805 0 : rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count,
7806 : bdev_nvme_readv_done, bio, flags,
7807 : bdev_nvme_queued_reset_sgl,
7808 : bdev_nvme_queued_next_sge, md, 0, 0);
7809 : }
7810 :
7811 3 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
7812 0 : SPDK_ERRLOG("readv failed: rc = %d\n", rc);
7813 : }
7814 3 : return rc;
7815 : }
7816 :
7817 : static int
7818 25 : bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7819 : void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
7820 : struct spdk_memory_domain *domain, void *domain_ctx,
7821 : struct spdk_accel_sequence *seq,
7822 : union spdk_bdev_nvme_cdw12 cdw12, union spdk_bdev_nvme_cdw13 cdw13)
7823 : {
7824 25 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
7825 25 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
7826 : int rc;
7827 :
7828 25 : SPDK_DEBUGLOG(bdev_nvme, "write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
7829 : lba_count, lba);
7830 :
7831 25 : bio->iovs = iov;
7832 25 : bio->iovcnt = iovcnt;
7833 25 : bio->iovpos = 0;
7834 25 : bio->iov_offset = 0;
7835 :
7836 25 : if (domain != NULL || seq != NULL) {
7837 0 : bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
7838 0 : bio->ext_opts.memory_domain = domain;
7839 0 : bio->ext_opts.memory_domain_ctx = domain_ctx;
7840 0 : bio->ext_opts.io_flags = flags | SPDK_NVME_IO_FLAGS_DIRECTIVE(cdw12.write.dtype);
7841 0 : bio->ext_opts.cdw13 = cdw13.raw;
7842 0 : bio->ext_opts.metadata = md;
7843 0 : bio->ext_opts.accel_sequence = seq;
7844 :
7845 0 : if (iovcnt == 1) {
7846 0 : rc = spdk_nvme_ns_cmd_write_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_writev_done,
7847 : bio, &bio->ext_opts);
7848 : } else {
7849 0 : rc = spdk_nvme_ns_cmd_writev_ext(ns, qpair, lba, lba_count,
7850 : bdev_nvme_writev_done, bio,
7851 : bdev_nvme_queued_reset_sgl,
7852 : bdev_nvme_queued_next_sge,
7853 : &bio->ext_opts);
7854 : }
7855 25 : } else if (iovcnt == 1) {
7856 25 : rc = spdk_nvme_ns_cmd_write_with_md(ns, qpair, iov[0].iov_base,
7857 : md, lba, lba_count, bdev_nvme_writev_done,
7858 : bio, flags, 0, 0);
7859 : } else {
7860 0 : rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
7861 : bdev_nvme_writev_done, bio, flags,
7862 : bdev_nvme_queued_reset_sgl,
7863 : bdev_nvme_queued_next_sge, md, 0, 0);
7864 : }
7865 :
7866 25 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
7867 0 : SPDK_ERRLOG("writev failed: rc = %d\n", rc);
7868 : }
7869 25 : return rc;
7870 : }
7871 :
7872 : static int
7873 0 : bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7874 : void *md, uint64_t lba_count, uint64_t zslba,
7875 : uint32_t flags)
7876 : {
7877 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
7878 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
7879 : int rc;
7880 :
7881 0 : SPDK_DEBUGLOG(bdev_nvme, "zone append %" PRIu64 " blocks to zone start lba %#" PRIx64 "\n",
7882 : lba_count, zslba);
7883 :
7884 0 : bio->iovs = iov;
7885 0 : bio->iovcnt = iovcnt;
7886 0 : bio->iovpos = 0;
7887 0 : bio->iov_offset = 0;
7888 :
7889 0 : if (iovcnt == 1) {
7890 0 : rc = spdk_nvme_zns_zone_append_with_md(ns, qpair, iov[0].iov_base, md, zslba,
7891 : lba_count,
7892 : bdev_nvme_zone_appendv_done, bio,
7893 : flags,
7894 : 0, 0);
7895 : } else {
7896 0 : rc = spdk_nvme_zns_zone_appendv_with_md(ns, qpair, zslba, lba_count,
7897 : bdev_nvme_zone_appendv_done, bio, flags,
7898 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
7899 : md, 0, 0);
7900 : }
7901 :
7902 0 : if (rc != 0 && rc != -ENOMEM) {
7903 0 : SPDK_ERRLOG("zone append failed: rc = %d\n", rc);
7904 : }
7905 0 : return rc;
7906 : }
7907 :
7908 : static int
7909 1 : bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7910 : void *md, uint64_t lba_count, uint64_t lba,
7911 : uint32_t flags)
7912 : {
7913 : int rc;
7914 :
7915 1 : SPDK_DEBUGLOG(bdev_nvme, "compare %" PRIu64 " blocks with offset %#" PRIx64 "\n",
7916 : lba_count, lba);
7917 :
7918 1 : bio->iovs = iov;
7919 1 : bio->iovcnt = iovcnt;
7920 1 : bio->iovpos = 0;
7921 1 : bio->iov_offset = 0;
7922 :
7923 1 : rc = spdk_nvme_ns_cmd_comparev_with_md(bio->io_path->nvme_ns->ns,
7924 1 : bio->io_path->qpair->qpair,
7925 : lba, lba_count,
7926 : bdev_nvme_comparev_done, bio, flags,
7927 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
7928 : md, 0, 0);
7929 :
7930 1 : if (rc != 0 && rc != -ENOMEM) {
7931 0 : SPDK_ERRLOG("comparev failed: rc = %d\n", rc);
7932 : }
7933 1 : return rc;
7934 : }
7935 :
7936 : static int
7937 2 : bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt,
7938 : struct iovec *write_iov, int write_iovcnt,
7939 : void *md, uint64_t lba_count, uint64_t lba, uint32_t flags)
7940 : {
7941 2 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
7942 2 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
7943 2 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7944 : int rc;
7945 :
7946 2 : SPDK_DEBUGLOG(bdev_nvme, "compare and write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
7947 : lba_count, lba);
7948 :
7949 2 : bio->iovs = cmp_iov;
7950 2 : bio->iovcnt = cmp_iovcnt;
7951 2 : bio->iovpos = 0;
7952 2 : bio->iov_offset = 0;
7953 2 : bio->fused_iovs = write_iov;
7954 2 : bio->fused_iovcnt = write_iovcnt;
7955 2 : bio->fused_iovpos = 0;
7956 2 : bio->fused_iov_offset = 0;
7957 :
7958 2 : if (bdev_io->num_retries == 0) {
7959 2 : bio->first_fused_submitted = false;
7960 2 : bio->first_fused_completed = false;
7961 : }
7962 :
7963 2 : if (!bio->first_fused_submitted) {
7964 2 : flags |= SPDK_NVME_IO_FLAGS_FUSE_FIRST;
7965 2 : memset(&bio->cpl, 0, sizeof(bio->cpl));
7966 :
7967 2 : rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count,
7968 : bdev_nvme_comparev_and_writev_done, bio, flags,
7969 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, md, 0, 0);
7970 2 : if (rc == 0) {
7971 2 : bio->first_fused_submitted = true;
7972 2 : flags &= ~SPDK_NVME_IO_FLAGS_FUSE_FIRST;
7973 : } else {
7974 0 : if (rc != -ENOMEM) {
7975 0 : SPDK_ERRLOG("compare failed: rc = %d\n", rc);
7976 : }
7977 0 : return rc;
7978 : }
7979 : }
7980 :
7981 2 : flags |= SPDK_NVME_IO_FLAGS_FUSE_SECOND;
7982 :
7983 2 : rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
7984 : bdev_nvme_comparev_and_writev_done, bio, flags,
7985 : bdev_nvme_queued_reset_fused_sgl, bdev_nvme_queued_next_fused_sge, md, 0, 0);
7986 2 : if (rc != 0 && rc != -ENOMEM) {
7987 0 : SPDK_ERRLOG("write failed: rc = %d\n", rc);
7988 0 : rc = 0;
7989 : }
7990 :
7991 2 : return rc;
7992 : }
7993 :
7994 : static int
7995 1 : bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
7996 : {
7997 1 : struct spdk_nvme_dsm_range dsm_ranges[SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES];
7998 : struct spdk_nvme_dsm_range *range;
7999 : uint64_t offset, remaining;
8000 : uint64_t num_ranges_u64;
8001 : uint16_t num_ranges;
8002 : int rc;
8003 :
8004 1 : num_ranges_u64 = (num_blocks + SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS - 1) /
8005 : SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8006 1 : if (num_ranges_u64 > SPDK_COUNTOF(dsm_ranges)) {
8007 0 : SPDK_ERRLOG("Unmap request for %" PRIu64 " blocks is too large\n", num_blocks);
8008 0 : return -EINVAL;
8009 : }
8010 1 : num_ranges = (uint16_t)num_ranges_u64;
8011 :
8012 1 : offset = offset_blocks;
8013 1 : remaining = num_blocks;
8014 1 : range = &dsm_ranges[0];
8015 :
8016 : /* Fill max-size ranges until the remaining blocks fit into one range */
8017 1 : while (remaining > SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS) {
8018 0 : range->attributes.raw = 0;
8019 0 : range->length = SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8020 0 : range->starting_lba = offset;
8021 :
8022 0 : offset += SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8023 0 : remaining -= SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8024 0 : range++;
8025 : }
8026 :
8027 : /* Final range describes the remaining blocks */
8028 1 : range->attributes.raw = 0;
8029 1 : range->length = remaining;
8030 1 : range->starting_lba = offset;
8031 :
8032 1 : rc = spdk_nvme_ns_cmd_dataset_management(bio->io_path->nvme_ns->ns,
8033 1 : bio->io_path->qpair->qpair,
8034 : SPDK_NVME_DSM_ATTR_DEALLOCATE,
8035 : dsm_ranges, num_ranges,
8036 : bdev_nvme_queued_done, bio);
8037 :
8038 1 : return rc;
8039 : }
8040 :
8041 : static int
8042 0 : bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
8043 : {
8044 0 : if (num_blocks > UINT16_MAX + 1) {
8045 0 : SPDK_ERRLOG("NVMe write zeroes is limited to 16-bit block count\n");
8046 0 : return -EINVAL;
8047 : }
8048 :
8049 0 : return spdk_nvme_ns_cmd_write_zeroes(bio->io_path->nvme_ns->ns,
8050 0 : bio->io_path->qpair->qpair,
8051 : offset_blocks, num_blocks,
8052 : bdev_nvme_queued_done, bio,
8053 : 0);
8054 : }
8055 :
8056 : static int
8057 0 : bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones,
8058 : struct spdk_bdev_zone_info *info)
8059 : {
8060 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8061 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8062 0 : uint32_t zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
8063 0 : uint64_t zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
8064 0 : uint64_t total_zones = spdk_nvme_zns_ns_get_num_zones(ns);
8065 :
8066 0 : if (zone_id % zone_size != 0) {
8067 0 : return -EINVAL;
8068 : }
8069 :
8070 0 : if (num_zones > total_zones || !num_zones) {
8071 0 : return -EINVAL;
8072 : }
8073 :
8074 0 : assert(!bio->zone_report_buf);
8075 0 : bio->zone_report_buf = calloc(1, zone_report_bufsize);
8076 0 : if (!bio->zone_report_buf) {
8077 0 : return -ENOMEM;
8078 : }
8079 :
8080 0 : bio->handled_zones = 0;
8081 :
8082 0 : return spdk_nvme_zns_report_zones(ns, qpair, bio->zone_report_buf, zone_report_bufsize,
8083 : zone_id, SPDK_NVME_ZRA_LIST_ALL, true,
8084 : bdev_nvme_get_zone_info_done, bio);
8085 : }
8086 :
8087 : static int
8088 0 : bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
8089 : enum spdk_bdev_zone_action action)
8090 : {
8091 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8092 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8093 :
8094 0 : switch (action) {
8095 0 : case SPDK_BDEV_ZONE_CLOSE:
8096 0 : return spdk_nvme_zns_close_zone(ns, qpair, zone_id, false,
8097 : bdev_nvme_zone_management_done, bio);
8098 0 : case SPDK_BDEV_ZONE_FINISH:
8099 0 : return spdk_nvme_zns_finish_zone(ns, qpair, zone_id, false,
8100 : bdev_nvme_zone_management_done, bio);
8101 0 : case SPDK_BDEV_ZONE_OPEN:
8102 0 : return spdk_nvme_zns_open_zone(ns, qpair, zone_id, false,
8103 : bdev_nvme_zone_management_done, bio);
8104 0 : case SPDK_BDEV_ZONE_RESET:
8105 0 : return spdk_nvme_zns_reset_zone(ns, qpair, zone_id, false,
8106 : bdev_nvme_zone_management_done, bio);
8107 0 : case SPDK_BDEV_ZONE_OFFLINE:
8108 0 : return spdk_nvme_zns_offline_zone(ns, qpair, zone_id, false,
8109 : bdev_nvme_zone_management_done, bio);
8110 0 : default:
8111 0 : return -EINVAL;
8112 : }
8113 : }
8114 :
8115 : static void
8116 5 : bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
8117 : struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes)
8118 : {
8119 : struct nvme_io_path *io_path;
8120 : struct nvme_ctrlr *nvme_ctrlr;
8121 : uint32_t max_xfer_size;
8122 5 : int rc = -ENXIO;
8123 :
8124 : /* Choose the first ctrlr which is not failed. */
8125 8 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
8126 7 : nvme_ctrlr = io_path->qpair->ctrlr;
8127 :
8128 : /* We should skip any unavailable nvme_ctrlr rather than checking
8129 : * if the return value of spdk_nvme_ctrlr_cmd_admin_raw() is -ENXIO.
8130 : */
8131 7 : if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
8132 3 : continue;
8133 : }
8134 :
8135 4 : max_xfer_size = spdk_nvme_ctrlr_get_max_xfer_size(nvme_ctrlr->ctrlr);
8136 :
8137 4 : if (nbytes > max_xfer_size) {
8138 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8139 0 : rc = -EINVAL;
8140 0 : goto err;
8141 : }
8142 :
8143 4 : rc = spdk_nvme_ctrlr_cmd_admin_raw(nvme_ctrlr->ctrlr, cmd, buf, (uint32_t)nbytes,
8144 : bdev_nvme_admin_passthru_done, bio);
8145 4 : if (rc == 0) {
8146 4 : return;
8147 : }
8148 : }
8149 :
8150 1 : err:
8151 1 : bdev_nvme_admin_complete(bio, rc);
8152 : }
8153 :
8154 : static int
8155 0 : bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
8156 : void *buf, size_t nbytes)
8157 : {
8158 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8159 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8160 0 : uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8161 0 : struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8162 :
8163 0 : if (nbytes > max_xfer_size) {
8164 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8165 0 : return -EINVAL;
8166 : }
8167 :
8168 : /*
8169 : * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
8170 : * so fill it out automatically.
8171 : */
8172 0 : cmd->nsid = spdk_nvme_ns_get_id(ns);
8173 :
8174 0 : return spdk_nvme_ctrlr_cmd_io_raw(ctrlr, qpair, cmd, buf,
8175 : (uint32_t)nbytes, bdev_nvme_queued_done, bio);
8176 : }
8177 :
8178 : static int
8179 0 : bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
8180 : void *buf, size_t nbytes, void *md_buf, size_t md_len)
8181 : {
8182 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8183 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8184 0 : size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
8185 0 : uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8186 0 : struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8187 :
8188 0 : if (nbytes > max_xfer_size) {
8189 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8190 0 : return -EINVAL;
8191 : }
8192 :
8193 0 : if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
8194 0 : SPDK_ERRLOG("invalid meta data buffer size\n");
8195 0 : return -EINVAL;
8196 : }
8197 :
8198 : /*
8199 : * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
8200 : * so fill it out automatically.
8201 : */
8202 0 : cmd->nsid = spdk_nvme_ns_get_id(ns);
8203 :
8204 0 : return spdk_nvme_ctrlr_cmd_io_raw_with_md(ctrlr, qpair, cmd, buf,
8205 : (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio);
8206 : }
8207 :
8208 : static int
8209 0 : bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio,
8210 : struct spdk_nvme_cmd *cmd, struct iovec *iov, int iovcnt,
8211 : size_t nbytes, void *md_buf, size_t md_len)
8212 : {
8213 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8214 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8215 0 : size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
8216 0 : uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8217 0 : struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8218 :
8219 0 : bio->iovs = iov;
8220 0 : bio->iovcnt = iovcnt;
8221 0 : bio->iovpos = 0;
8222 0 : bio->iov_offset = 0;
8223 :
8224 0 : if (nbytes > max_xfer_size) {
8225 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8226 0 : return -EINVAL;
8227 : }
8228 :
8229 0 : if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
8230 0 : SPDK_ERRLOG("invalid meta data buffer size\n");
8231 0 : return -EINVAL;
8232 : }
8233 :
8234 : /*
8235 : * Each NVMe bdev is a specific namespace, and all NVMe I/O commands
8236 : * require a nsid, so fill it out automatically.
8237 : */
8238 0 : cmd->nsid = spdk_nvme_ns_get_id(ns);
8239 :
8240 0 : return spdk_nvme_ctrlr_cmd_iov_raw_with_md(
8241 : ctrlr, qpair, cmd, (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio,
8242 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge);
8243 : }
8244 :
8245 : static void
8246 6 : bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
8247 : struct nvme_bdev_io *bio_to_abort)
8248 : {
8249 : struct nvme_io_path *io_path;
8250 6 : int rc = 0;
8251 :
8252 6 : rc = bdev_nvme_abort_retry_io(nbdev_ch, bio_to_abort);
8253 6 : if (rc == 0) {
8254 1 : bdev_nvme_admin_complete(bio, 0);
8255 1 : return;
8256 : }
8257 :
8258 5 : io_path = bio_to_abort->io_path;
8259 5 : if (io_path != NULL) {
8260 3 : rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
8261 3 : io_path->qpair->qpair,
8262 : bio_to_abort,
8263 : bdev_nvme_abort_done, bio);
8264 : } else {
8265 3 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
8266 2 : rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
8267 : NULL,
8268 : bio_to_abort,
8269 : bdev_nvme_abort_done, bio);
8270 :
8271 2 : if (rc != -ENOENT) {
8272 1 : break;
8273 : }
8274 : }
8275 : }
8276 :
8277 5 : if (rc != 0) {
8278 : /* If no command was found or there was any error, complete the abort
8279 : * request with failure.
8280 : */
8281 2 : bdev_nvme_admin_complete(bio, rc);
8282 : }
8283 : }
8284 :
8285 : static int
8286 0 : bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks, uint64_t src_offset_blocks,
8287 : uint64_t num_blocks)
8288 : {
8289 0 : struct spdk_nvme_scc_source_range range = {
8290 : .slba = src_offset_blocks,
8291 0 : .nlb = num_blocks - 1
8292 : };
8293 :
8294 0 : return spdk_nvme_ns_cmd_copy(bio->io_path->nvme_ns->ns,
8295 0 : bio->io_path->qpair->qpair,
8296 : &range, 1, dst_offset_blocks,
8297 : bdev_nvme_queued_done, bio);
8298 : }
8299 :
8300 : static void
8301 0 : bdev_nvme_opts_config_json(struct spdk_json_write_ctx *w)
8302 : {
8303 : const char *action;
8304 : uint32_t i;
8305 :
8306 0 : if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET) {
8307 0 : action = "reset";
8308 0 : } else if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT) {
8309 0 : action = "abort";
8310 : } else {
8311 0 : action = "none";
8312 : }
8313 :
8314 0 : spdk_json_write_object_begin(w);
8315 :
8316 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_set_options");
8317 :
8318 0 : spdk_json_write_named_object_begin(w, "params");
8319 0 : spdk_json_write_named_string(w, "action_on_timeout", action);
8320 0 : spdk_json_write_named_uint64(w, "timeout_us", g_opts.timeout_us);
8321 0 : spdk_json_write_named_uint64(w, "timeout_admin_us", g_opts.timeout_admin_us);
8322 0 : spdk_json_write_named_uint32(w, "keep_alive_timeout_ms", g_opts.keep_alive_timeout_ms);
8323 0 : spdk_json_write_named_uint32(w, "arbitration_burst", g_opts.arbitration_burst);
8324 0 : spdk_json_write_named_uint32(w, "low_priority_weight", g_opts.low_priority_weight);
8325 0 : spdk_json_write_named_uint32(w, "medium_priority_weight", g_opts.medium_priority_weight);
8326 0 : spdk_json_write_named_uint32(w, "high_priority_weight", g_opts.high_priority_weight);
8327 0 : spdk_json_write_named_uint64(w, "nvme_adminq_poll_period_us", g_opts.nvme_adminq_poll_period_us);
8328 0 : spdk_json_write_named_uint64(w, "nvme_ioq_poll_period_us", g_opts.nvme_ioq_poll_period_us);
8329 0 : spdk_json_write_named_uint32(w, "io_queue_requests", g_opts.io_queue_requests);
8330 0 : spdk_json_write_named_bool(w, "delay_cmd_submit", g_opts.delay_cmd_submit);
8331 0 : spdk_json_write_named_uint32(w, "transport_retry_count", g_opts.transport_retry_count);
8332 0 : spdk_json_write_named_int32(w, "bdev_retry_count", g_opts.bdev_retry_count);
8333 0 : spdk_json_write_named_uint8(w, "transport_ack_timeout", g_opts.transport_ack_timeout);
8334 0 : spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", g_opts.ctrlr_loss_timeout_sec);
8335 0 : spdk_json_write_named_uint32(w, "reconnect_delay_sec", g_opts.reconnect_delay_sec);
8336 0 : spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec", g_opts.fast_io_fail_timeout_sec);
8337 0 : spdk_json_write_named_bool(w, "disable_auto_failback", g_opts.disable_auto_failback);
8338 0 : spdk_json_write_named_bool(w, "generate_uuids", g_opts.generate_uuids);
8339 0 : spdk_json_write_named_uint8(w, "transport_tos", g_opts.transport_tos);
8340 0 : spdk_json_write_named_bool(w, "nvme_error_stat", g_opts.nvme_error_stat);
8341 0 : spdk_json_write_named_uint32(w, "rdma_srq_size", g_opts.rdma_srq_size);
8342 0 : spdk_json_write_named_bool(w, "io_path_stat", g_opts.io_path_stat);
8343 0 : spdk_json_write_named_bool(w, "allow_accel_sequence", g_opts.allow_accel_sequence);
8344 0 : spdk_json_write_named_uint32(w, "rdma_max_cq_size", g_opts.rdma_max_cq_size);
8345 0 : spdk_json_write_named_uint16(w, "rdma_cm_event_timeout_ms", g_opts.rdma_cm_event_timeout_ms);
8346 0 : spdk_json_write_named_array_begin(w, "dhchap_digests");
8347 0 : for (i = 0; i < 32; ++i) {
8348 0 : if (g_opts.dhchap_digests & SPDK_BIT(i)) {
8349 0 : spdk_json_write_string(w, spdk_nvme_dhchap_get_digest_name(i));
8350 : }
8351 : }
8352 0 : spdk_json_write_array_end(w);
8353 0 : spdk_json_write_named_array_begin(w, "dhchap_dhgroups");
8354 0 : for (i = 0; i < 32; ++i) {
8355 0 : if (g_opts.dhchap_dhgroups & SPDK_BIT(i)) {
8356 0 : spdk_json_write_string(w, spdk_nvme_dhchap_get_dhgroup_name(i));
8357 : }
8358 : }
8359 :
8360 0 : spdk_json_write_array_end(w);
8361 0 : spdk_json_write_object_end(w);
8362 :
8363 0 : spdk_json_write_object_end(w);
8364 0 : }
8365 :
8366 : static void
8367 0 : bdev_nvme_discovery_config_json(struct spdk_json_write_ctx *w, struct discovery_ctx *ctx)
8368 : {
8369 0 : struct spdk_nvme_transport_id trid;
8370 :
8371 0 : spdk_json_write_object_begin(w);
8372 :
8373 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_start_discovery");
8374 :
8375 0 : spdk_json_write_named_object_begin(w, "params");
8376 0 : spdk_json_write_named_string(w, "name", ctx->name);
8377 0 : spdk_json_write_named_string(w, "hostnqn", ctx->hostnqn);
8378 :
8379 0 : trid = ctx->trid;
8380 0 : memset(trid.subnqn, 0, sizeof(trid.subnqn));
8381 0 : nvme_bdev_dump_trid_json(&trid, w);
8382 :
8383 0 : spdk_json_write_named_bool(w, "wait_for_attach", ctx->wait_for_attach);
8384 0 : spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", ctx->bdev_opts.ctrlr_loss_timeout_sec);
8385 0 : spdk_json_write_named_uint32(w, "reconnect_delay_sec", ctx->bdev_opts.reconnect_delay_sec);
8386 0 : spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
8387 : ctx->bdev_opts.fast_io_fail_timeout_sec);
8388 0 : spdk_json_write_object_end(w);
8389 :
8390 0 : spdk_json_write_object_end(w);
8391 0 : }
8392 :
8393 : #ifdef SPDK_CONFIG_NVME_CUSE
8394 : static void
8395 0 : nvme_ctrlr_cuse_config_json(struct spdk_json_write_ctx *w,
8396 : struct nvme_ctrlr *nvme_ctrlr)
8397 0 : {
8398 0 : size_t cuse_name_size = 128;
8399 0 : char cuse_name[cuse_name_size];
8400 :
8401 0 : if (spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr,
8402 : cuse_name, &cuse_name_size) != 0) {
8403 0 : return;
8404 : }
8405 :
8406 0 : spdk_json_write_object_begin(w);
8407 :
8408 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_cuse_register");
8409 :
8410 0 : spdk_json_write_named_object_begin(w, "params");
8411 0 : spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
8412 0 : spdk_json_write_object_end(w);
8413 :
8414 0 : spdk_json_write_object_end(w);
8415 : }
8416 : #endif
8417 :
8418 : static void
8419 0 : nvme_ctrlr_config_json(struct spdk_json_write_ctx *w,
8420 : struct nvme_ctrlr *nvme_ctrlr)
8421 : {
8422 : struct spdk_nvme_transport_id *trid;
8423 : const struct spdk_nvme_ctrlr_opts *opts;
8424 :
8425 0 : if (nvme_ctrlr->opts.from_discovery_service) {
8426 : /* Do not emit an RPC for this - it will be implicitly
8427 : * covered by a separate bdev_nvme_start_discovery or
8428 : * bdev_nvme_start_mdns_discovery RPC.
8429 : */
8430 0 : return;
8431 : }
8432 :
8433 0 : trid = &nvme_ctrlr->active_path_id->trid;
8434 :
8435 0 : spdk_json_write_object_begin(w);
8436 :
8437 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_attach_controller");
8438 :
8439 0 : spdk_json_write_named_object_begin(w, "params");
8440 0 : spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
8441 0 : nvme_bdev_dump_trid_json(trid, w);
8442 0 : spdk_json_write_named_bool(w, "prchk_reftag",
8443 0 : (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_REFTAG) != 0);
8444 0 : spdk_json_write_named_bool(w, "prchk_guard",
8445 0 : (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_GUARD) != 0);
8446 0 : spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", nvme_ctrlr->opts.ctrlr_loss_timeout_sec);
8447 0 : spdk_json_write_named_uint32(w, "reconnect_delay_sec", nvme_ctrlr->opts.reconnect_delay_sec);
8448 0 : spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
8449 : nvme_ctrlr->opts.fast_io_fail_timeout_sec);
8450 0 : if (nvme_ctrlr->psk != NULL) {
8451 0 : spdk_json_write_named_string(w, "psk", spdk_key_get_name(nvme_ctrlr->psk));
8452 0 : } else if (nvme_ctrlr->opts.psk[0] != '\0') {
8453 0 : spdk_json_write_named_string(w, "psk", nvme_ctrlr->opts.psk);
8454 : }
8455 :
8456 0 : opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
8457 0 : spdk_json_write_named_string(w, "hostnqn", opts->hostnqn);
8458 0 : spdk_json_write_named_bool(w, "hdgst", opts->header_digest);
8459 0 : spdk_json_write_named_bool(w, "ddgst", opts->data_digest);
8460 0 : if (opts->src_addr[0] != '\0') {
8461 0 : spdk_json_write_named_string(w, "hostaddr", opts->src_addr);
8462 : }
8463 0 : if (opts->src_svcid[0] != '\0') {
8464 0 : spdk_json_write_named_string(w, "hostsvcid", opts->src_svcid);
8465 : }
8466 :
8467 0 : spdk_json_write_object_end(w);
8468 :
8469 0 : spdk_json_write_object_end(w);
8470 : }
8471 :
8472 : static void
8473 0 : bdev_nvme_hotplug_config_json(struct spdk_json_write_ctx *w)
8474 : {
8475 0 : spdk_json_write_object_begin(w);
8476 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_set_hotplug");
8477 :
8478 0 : spdk_json_write_named_object_begin(w, "params");
8479 0 : spdk_json_write_named_uint64(w, "period_us", g_nvme_hotplug_poll_period_us);
8480 0 : spdk_json_write_named_bool(w, "enable", g_nvme_hotplug_enabled);
8481 0 : spdk_json_write_object_end(w);
8482 :
8483 0 : spdk_json_write_object_end(w);
8484 0 : }
8485 :
8486 : static int
8487 0 : bdev_nvme_config_json(struct spdk_json_write_ctx *w)
8488 : {
8489 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
8490 : struct nvme_ctrlr *nvme_ctrlr;
8491 : struct discovery_ctx *ctx;
8492 :
8493 0 : bdev_nvme_opts_config_json(w);
8494 :
8495 0 : pthread_mutex_lock(&g_bdev_nvme_mutex);
8496 :
8497 0 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
8498 0 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
8499 0 : nvme_ctrlr_config_json(w, nvme_ctrlr);
8500 :
8501 : #ifdef SPDK_CONFIG_NVME_CUSE
8502 0 : nvme_ctrlr_cuse_config_json(w, nvme_ctrlr);
8503 : #endif
8504 : }
8505 : }
8506 :
8507 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
8508 0 : if (!ctx->from_mdns_discovery_service) {
8509 0 : bdev_nvme_discovery_config_json(w, ctx);
8510 : }
8511 : }
8512 :
8513 0 : bdev_nvme_mdns_discovery_config_json(w);
8514 :
8515 : /* Dump as last parameter to give all NVMe bdevs chance to be constructed
8516 : * before enabling hotplug poller.
8517 : */
8518 0 : bdev_nvme_hotplug_config_json(w);
8519 :
8520 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
8521 0 : return 0;
8522 : }
8523 :
8524 : struct spdk_nvme_ctrlr *
8525 1 : bdev_nvme_get_ctrlr(struct spdk_bdev *bdev)
8526 : {
8527 : struct nvme_bdev *nbdev;
8528 : struct nvme_ns *nvme_ns;
8529 :
8530 1 : if (!bdev || bdev->module != &nvme_if) {
8531 0 : return NULL;
8532 : }
8533 :
8534 1 : nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
8535 1 : nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
8536 1 : assert(nvme_ns != NULL);
8537 :
8538 1 : return nvme_ns->ctrlr->ctrlr;
8539 : }
8540 :
8541 : void
8542 0 : nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path)
8543 : {
8544 0 : struct nvme_ns *nvme_ns = io_path->nvme_ns;
8545 0 : struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
8546 : const struct spdk_nvme_ctrlr_data *cdata;
8547 : const struct spdk_nvme_transport_id *trid;
8548 : const struct nvme_bdev_channel *nbdev_ch;
8549 : const char *adrfam_str;
8550 : bool current;
8551 :
8552 0 : spdk_json_write_object_begin(w);
8553 :
8554 0 : spdk_json_write_named_string(w, "bdev_name", nvme_ns->bdev->disk.name);
8555 :
8556 0 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
8557 0 : trid = spdk_nvme_ctrlr_get_transport_id(nvme_ctrlr->ctrlr);
8558 :
8559 0 : spdk_json_write_named_uint32(w, "cntlid", cdata->cntlid);
8560 0 : nbdev_ch = io_path->nbdev_ch;
8561 0 : if (nbdev_ch == NULL) {
8562 0 : current = false;
8563 0 : } else if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
8564 0 : struct nvme_io_path *optimized_io_path = NULL;
8565 :
8566 0 : STAILQ_FOREACH(optimized_io_path, &nbdev_ch->io_path_list, stailq) {
8567 0 : if (optimized_io_path->nvme_ns->ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE) {
8568 0 : break;
8569 : }
8570 : }
8571 :
8572 0 : current = nvme_io_path_is_available(io_path);
8573 0 : if (io_path->nvme_ns->ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE) {
8574 : /* A non-optimized path is only current if there are no optimized paths. */
8575 0 : current = current && (optimized_io_path == NULL);
8576 : }
8577 : } else {
8578 0 : if (nbdev_ch->current_io_path) {
8579 0 : current = (io_path == nbdev_ch->current_io_path);
8580 : } else {
8581 : struct nvme_io_path *first_path;
8582 :
8583 : /* We arrived here as there are no optimized paths for active-passive
8584 : * mode. Check if this io_path is the first one available on the list.
8585 : */
8586 0 : current = false;
8587 0 : STAILQ_FOREACH(first_path, &nbdev_ch->io_path_list, stailq) {
8588 0 : if (nvme_io_path_is_available(first_path)) {
8589 0 : current = (io_path == first_path);
8590 0 : break;
8591 : }
8592 : }
8593 : }
8594 : }
8595 0 : spdk_json_write_named_bool(w, "current", current);
8596 0 : spdk_json_write_named_bool(w, "connected", nvme_qpair_is_connected(io_path->qpair));
8597 0 : spdk_json_write_named_bool(w, "accessible", nvme_ns_is_accessible(nvme_ns));
8598 :
8599 0 : spdk_json_write_named_object_begin(w, "transport");
8600 0 : spdk_json_write_named_string(w, "trtype", trid->trstring);
8601 0 : spdk_json_write_named_string(w, "traddr", trid->traddr);
8602 0 : if (trid->trsvcid[0] != '\0') {
8603 0 : spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
8604 : }
8605 0 : adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
8606 0 : if (adrfam_str) {
8607 0 : spdk_json_write_named_string(w, "adrfam", adrfam_str);
8608 : }
8609 0 : spdk_json_write_object_end(w);
8610 :
8611 0 : spdk_json_write_object_end(w);
8612 0 : }
8613 :
8614 : void
8615 0 : bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w)
8616 : {
8617 : struct discovery_ctx *ctx;
8618 : struct discovery_entry_ctx *entry_ctx;
8619 :
8620 0 : spdk_json_write_array_begin(w);
8621 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
8622 0 : spdk_json_write_object_begin(w);
8623 0 : spdk_json_write_named_string(w, "name", ctx->name);
8624 :
8625 0 : spdk_json_write_named_object_begin(w, "trid");
8626 0 : nvme_bdev_dump_trid_json(&ctx->trid, w);
8627 0 : spdk_json_write_object_end(w);
8628 :
8629 0 : spdk_json_write_named_array_begin(w, "referrals");
8630 0 : TAILQ_FOREACH(entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
8631 0 : spdk_json_write_object_begin(w);
8632 0 : spdk_json_write_named_object_begin(w, "trid");
8633 0 : nvme_bdev_dump_trid_json(&entry_ctx->trid, w);
8634 0 : spdk_json_write_object_end(w);
8635 0 : spdk_json_write_object_end(w);
8636 : }
8637 0 : spdk_json_write_array_end(w);
8638 :
8639 0 : spdk_json_write_object_end(w);
8640 : }
8641 0 : spdk_json_write_array_end(w);
8642 0 : }
8643 :
8644 1 : SPDK_LOG_REGISTER_COMPONENT(bdev_nvme)
8645 :
8646 1 : SPDK_TRACE_REGISTER_FN(bdev_nvme_trace, "bdev_nvme", TRACE_GROUP_BDEV_NVME)
8647 : {
8648 0 : struct spdk_trace_tpoint_opts opts[] = {
8649 : {
8650 : "BDEV_NVME_IO_START", TRACE_BDEV_NVME_IO_START,
8651 : OWNER_TYPE_NONE, OBJECT_BDEV_NVME_IO, 1,
8652 : {{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
8653 : },
8654 : {
8655 : "BDEV_NVME_IO_DONE", TRACE_BDEV_NVME_IO_DONE,
8656 : OWNER_TYPE_NONE, OBJECT_BDEV_NVME_IO, 0,
8657 : {{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
8658 : }
8659 : };
8660 :
8661 :
8662 0 : spdk_trace_register_object(OBJECT_BDEV_NVME_IO, 'N');
8663 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
8664 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
8665 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
8666 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
8667 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
8668 0 : }
|