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 CTRLR_STRING(nvme_ctrlr) \
36 : (spdk_nvme_trtype_is_fabrics(nvme_ctrlr->active_path_id->trid.trtype) ? \
37 : nvme_ctrlr->active_path_id->trid.subnqn : nvme_ctrlr->active_path_id->trid.traddr)
38 :
39 : #define CTRLR_ID(nvme_ctrlr) (spdk_nvme_ctrlr_get_id(nvme_ctrlr->ctrlr))
40 :
41 : #define NVME_CTRLR_ERRLOG(ctrlr, format, ...) \
42 : SPDK_ERRLOG("[%s, %u] " format, CTRLR_STRING(ctrlr), CTRLR_ID(ctrlr), ##__VA_ARGS__);
43 :
44 : #define NVME_CTRLR_WARNLOG(ctrlr, format, ...) \
45 : SPDK_WARNLOG("[%s, %u] " format, CTRLR_STRING(ctrlr), CTRLR_ID(ctrlr), ##__VA_ARGS__);
46 :
47 : #define NVME_CTRLR_NOTICELOG(ctrlr, format, ...) \
48 : SPDK_NOTICELOG("[%s, %u] " format, CTRLR_STRING(ctrlr), CTRLR_ID(ctrlr), ##__VA_ARGS__);
49 :
50 : #define NVME_CTRLR_INFOLOG(ctrlr, format, ...) \
51 : SPDK_INFOLOG(bdev_nvme, "[%s, %u] " format, CTRLR_STRING(ctrlr), CTRLR_ID(ctrlr), ##__VA_ARGS__);
52 :
53 : #ifdef DEBUG
54 : #define NVME_CTRLR_DEBUGLOG(ctrlr, format, ...) \
55 : SPDK_DEBUGLOG(bdev_nvme, "[%s, %u] " format, CTRLR_STRING(ctrlr), CTRLR_ID(ctrlr), ##__VA_ARGS__);
56 : #else
57 : #define NVME_CTRLR_DEBUGLOG(ctrlr, ...) do { } while (0)
58 : #endif
59 :
60 : #define BDEV_STRING(nbdev) (nbdev->disk.name)
61 :
62 : #define NVME_BDEV_ERRLOG(nbdev, format, ...) \
63 : SPDK_ERRLOG("[%s] " format, BDEV_STRING(nbdev), ##__VA_ARGS__);
64 :
65 : #define NVME_BDEV_WARNLOG(nbdev, format, ...) \
66 : SPDK_WARNLOG("[%s] " format, BDEV_STRING(nbdev), ##__VA_ARGS__);
67 :
68 : #define NVME_BDEV_NOTICELOG(nbdev, format, ...) \
69 : SPDK_NOTICELOG("[%s] " format, BDEV_STRING(nbdev), ##__VA_ARGS__);
70 :
71 : #define NVME_BDEV_INFOLOG(nbdev, format, ...) \
72 : SPDK_INFOLOG(bdev_nvme, "[%s] " format, BDEV_STRING(nbdev), ##__VA_ARGS__);
73 :
74 : #define SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT true
75 : #define SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS (10000)
76 :
77 : #define NSID_STR_LEN 10
78 :
79 : #define SPDK_CONTROLLER_NAME_MAX 512
80 :
81 : static int bdev_nvme_config_json(struct spdk_json_write_ctx *w);
82 :
83 : struct nvme_bdev_io {
84 : /** array of iovecs to transfer. */
85 : struct iovec *iovs;
86 :
87 : /** Number of iovecs in iovs array. */
88 : int iovcnt;
89 :
90 : /** Current iovec position. */
91 : int iovpos;
92 :
93 : /** Offset in current iovec. */
94 : uint32_t iov_offset;
95 :
96 : /** Offset in current iovec. */
97 : uint32_t fused_iov_offset;
98 :
99 : /** array of iovecs to transfer. */
100 : struct iovec *fused_iovs;
101 :
102 : /** Number of iovecs in iovs array. */
103 : int fused_iovcnt;
104 :
105 : /** Current iovec position. */
106 : int fused_iovpos;
107 :
108 : /** I/O path the current I/O or admin passthrough is submitted on, or the I/O path
109 : * being reset in a reset I/O.
110 : */
111 : struct nvme_io_path *io_path;
112 :
113 : /** Saved status for admin passthru completion event, PI error verification, or intermediate compare-and-write status */
114 : struct spdk_nvme_cpl cpl;
115 :
116 : /** Extended IO opts passed by the user to bdev layer and mapped to NVME format */
117 : struct spdk_nvme_ns_cmd_ext_io_opts ext_opts;
118 :
119 : /** Keeps track if first of fused commands was submitted */
120 : bool first_fused_submitted;
121 :
122 : /** Keeps track if first of fused commands was completed */
123 : bool first_fused_completed;
124 :
125 : /* How many times the current I/O was retried. */
126 : int32_t retry_count;
127 :
128 : /** Expiration value in ticks to retry the current I/O. */
129 : uint64_t retry_ticks;
130 :
131 : /** Temporary pointer to zone report buffer */
132 : struct spdk_nvme_zns_zone_report *zone_report_buf;
133 :
134 : /** Keep track of how many zones that have been copied to the spdk_bdev_zone_info struct */
135 : uint64_t handled_zones;
136 :
137 : /* Current tsc at submit time. */
138 : uint64_t submit_tsc;
139 :
140 : /* Used to put nvme_bdev_io into the list */
141 : TAILQ_ENTRY(nvme_bdev_io) retry_link;
142 : };
143 :
144 : struct nvme_probe_skip_entry {
145 : struct spdk_nvme_transport_id trid;
146 : TAILQ_ENTRY(nvme_probe_skip_entry) tailq;
147 : };
148 : /* All the controllers deleted by users via RPC are skipped by hotplug monitor */
149 : static TAILQ_HEAD(, nvme_probe_skip_entry) g_skipped_nvme_ctrlrs = TAILQ_HEAD_INITIALIZER(
150 : g_skipped_nvme_ctrlrs);
151 :
152 : #define BDEV_NVME_DEFAULT_DIGESTS (SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA256) | \
153 : SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA384) | \
154 : SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA512))
155 :
156 : #define BDEV_NVME_DEFAULT_DHGROUPS (SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_NULL) | \
157 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_2048) | \
158 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_3072) | \
159 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_4096) | \
160 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_6144) | \
161 : SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_8192))
162 :
163 : static struct spdk_bdev_nvme_opts g_opts = {
164 : .action_on_timeout = SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE,
165 : .timeout_us = 0,
166 : .timeout_admin_us = 0,
167 : .keep_alive_timeout_ms = SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS,
168 : .transport_retry_count = 4,
169 : .arbitration_burst = 0,
170 : .low_priority_weight = 0,
171 : .medium_priority_weight = 0,
172 : .high_priority_weight = 0,
173 : .nvme_adminq_poll_period_us = 10000ULL,
174 : .nvme_ioq_poll_period_us = 0,
175 : .io_queue_requests = 0,
176 : .delay_cmd_submit = SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT,
177 : .bdev_retry_count = 3,
178 : .transport_ack_timeout = 0,
179 : .ctrlr_loss_timeout_sec = 0,
180 : .reconnect_delay_sec = 0,
181 : .fast_io_fail_timeout_sec = 0,
182 : .disable_auto_failback = false,
183 : .generate_uuids = false,
184 : .transport_tos = 0,
185 : .nvme_error_stat = false,
186 : .io_path_stat = false,
187 : .allow_accel_sequence = false,
188 : .dhchap_digests = BDEV_NVME_DEFAULT_DIGESTS,
189 : .dhchap_dhgroups = BDEV_NVME_DEFAULT_DHGROUPS,
190 : };
191 :
192 : #define NVME_HOTPLUG_POLL_PERIOD_MAX 10000000ULL
193 : #define NVME_HOTPLUG_POLL_PERIOD_DEFAULT 100000ULL
194 :
195 : static int g_hot_insert_nvme_controller_index = 0;
196 : static uint64_t g_nvme_hotplug_poll_period_us = NVME_HOTPLUG_POLL_PERIOD_DEFAULT;
197 : static bool g_nvme_hotplug_enabled = false;
198 : struct spdk_thread *g_bdev_nvme_init_thread;
199 : static struct spdk_poller *g_hotplug_poller;
200 : static struct spdk_poller *g_hotplug_probe_poller;
201 : static struct spdk_nvme_probe_ctx *g_hotplug_probe_ctx;
202 :
203 : static void nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
204 : struct nvme_async_probe_ctx *ctx);
205 : static void nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
206 : struct nvme_async_probe_ctx *ctx);
207 : static int bdev_nvme_library_init(void);
208 : static void bdev_nvme_library_fini(void);
209 : static void _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch,
210 : struct spdk_bdev_io *bdev_io);
211 : static void bdev_nvme_submit_request(struct spdk_io_channel *ch,
212 : struct spdk_bdev_io *bdev_io);
213 : static int bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
214 : void *md, uint64_t lba_count, uint64_t lba,
215 : uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
216 : struct spdk_accel_sequence *seq);
217 : static int bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
218 : void *md, uint64_t lba_count, uint64_t lba);
219 : static int bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
220 : void *md, uint64_t lba_count, uint64_t lba,
221 : uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
222 : struct spdk_accel_sequence *seq,
223 : union spdk_bdev_nvme_cdw12 cdw12, union spdk_bdev_nvme_cdw13 cdw13);
224 : static int bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
225 : void *md, uint64_t lba_count,
226 : uint64_t zslba, uint32_t flags);
227 : static int bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
228 : void *md, uint64_t lba_count, uint64_t lba,
229 : uint32_t flags);
230 : static int bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio,
231 : struct iovec *cmp_iov, int cmp_iovcnt, struct iovec *write_iov,
232 : int write_iovcnt, void *md, uint64_t lba_count, uint64_t lba,
233 : uint32_t flags);
234 : static int bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id,
235 : uint32_t num_zones, struct spdk_bdev_zone_info *info);
236 : static int bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
237 : enum spdk_bdev_zone_action action);
238 : static void bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch,
239 : struct nvme_bdev_io *bio,
240 : struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes);
241 : static int bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
242 : void *buf, size_t nbytes);
243 : static int bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
244 : void *buf, size_t nbytes, void *md_buf, size_t md_len);
245 : static int bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
246 : struct iovec *iov, int iovcnt, size_t nbytes,
247 : void *md_buf, size_t md_len);
248 : static void bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch,
249 : struct nvme_bdev_io *bio, struct nvme_bdev_io *bio_to_abort);
250 : static void bdev_nvme_reset_io(struct nvme_bdev *nbdev, struct nvme_bdev_io *bio);
251 : static int bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
252 : static int bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
253 : static void remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr);
254 : static int nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr);
255 :
256 : static struct nvme_ns *nvme_ns_alloc(void);
257 : static void nvme_ns_free(struct nvme_ns *ns);
258 :
259 : static int
260 176 : nvme_ns_cmp(struct nvme_ns *ns1, struct nvme_ns *ns2)
261 : {
262 176 : return ns1->id < ns2->id ? -1 : ns1->id > ns2->id;
263 : }
264 :
265 958 : RB_GENERATE_STATIC(nvme_ns_tree, nvme_ns, node, nvme_ns_cmp);
266 :
267 : struct spdk_nvme_qpair *
268 1 : bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch)
269 : {
270 : struct nvme_ctrlr_channel *ctrlr_ch;
271 :
272 1 : assert(ctrlr_io_ch != NULL);
273 :
274 1 : ctrlr_ch = spdk_io_channel_get_ctx(ctrlr_io_ch);
275 :
276 1 : return ctrlr_ch->qpair->qpair;
277 : }
278 :
279 : static int
280 0 : bdev_nvme_get_ctx_size(void)
281 : {
282 0 : return sizeof(struct nvme_bdev_io);
283 : }
284 :
285 : static struct spdk_bdev_module nvme_if = {
286 : .name = "nvme",
287 : .async_fini = true,
288 : .module_init = bdev_nvme_library_init,
289 : .module_fini = bdev_nvme_library_fini,
290 : .config_json = bdev_nvme_config_json,
291 : .get_ctx_size = bdev_nvme_get_ctx_size,
292 :
293 : };
294 1 : SPDK_BDEV_MODULE_REGISTER(nvme, &nvme_if)
295 :
296 : struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs = TAILQ_HEAD_INITIALIZER(g_nvme_bdev_ctrlrs);
297 : pthread_mutex_t g_bdev_nvme_mutex = PTHREAD_MUTEX_INITIALIZER;
298 : bool g_bdev_nvme_module_finish;
299 :
300 : struct nvme_bdev_ctrlr *
301 333 : nvme_bdev_ctrlr_get_by_name(const char *name)
302 : {
303 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
304 :
305 333 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
306 171 : if (strcmp(name, nbdev_ctrlr->name) == 0) {
307 171 : break;
308 : }
309 : }
310 :
311 333 : return nbdev_ctrlr;
312 : }
313 :
314 : static struct nvme_ctrlr *
315 59 : nvme_bdev_ctrlr_get_ctrlr(struct nvme_bdev_ctrlr *nbdev_ctrlr,
316 : const struct spdk_nvme_transport_id *trid, const char *hostnqn)
317 : {
318 : const struct spdk_nvme_ctrlr_opts *opts;
319 : struct nvme_ctrlr *nvme_ctrlr;
320 :
321 100 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
322 75 : opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
323 75 : if (spdk_nvme_transport_id_compare(trid, &nvme_ctrlr->active_path_id->trid) == 0 &&
324 34 : strcmp(hostnqn, opts->hostnqn) == 0) {
325 34 : break;
326 : }
327 : }
328 :
329 59 : return nvme_ctrlr;
330 : }
331 :
332 : struct nvme_ctrlr *
333 0 : nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr,
334 : uint16_t cntlid)
335 : {
336 : struct nvme_ctrlr *nvme_ctrlr;
337 : const struct spdk_nvme_ctrlr_data *cdata;
338 :
339 0 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
340 0 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
341 0 : if (cdata->cntlid == cntlid) {
342 0 : break;
343 : }
344 : }
345 :
346 0 : return nvme_ctrlr;
347 : }
348 :
349 : static struct nvme_bdev *
350 75 : nvme_bdev_ctrlr_get_bdev(struct nvme_bdev_ctrlr *nbdev_ctrlr, uint32_t nsid)
351 : {
352 : struct nvme_bdev *bdev;
353 :
354 75 : pthread_mutex_lock(&g_bdev_nvme_mutex);
355 109 : TAILQ_FOREACH(bdev, &nbdev_ctrlr->bdevs, tailq) {
356 69 : if (bdev->nsid == nsid) {
357 35 : break;
358 : }
359 : }
360 75 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
361 :
362 75 : return bdev;
363 : }
364 :
365 : struct nvme_ns *
366 145 : nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid)
367 : {
368 145 : struct nvme_ns ns;
369 :
370 145 : assert(nsid > 0);
371 :
372 145 : ns.id = nsid;
373 145 : return RB_FIND(nvme_ns_tree, &nvme_ctrlr->namespaces, &ns);
374 : }
375 :
376 : struct nvme_ns *
377 165 : nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr)
378 : {
379 165 : return RB_MIN(nvme_ns_tree, &nvme_ctrlr->namespaces);
380 : }
381 :
382 : struct nvme_ns *
383 74 : nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns)
384 : {
385 74 : if (ns == NULL) {
386 0 : return NULL;
387 : }
388 :
389 74 : return RB_NEXT(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
390 : }
391 :
392 : static struct nvme_ctrlr *
393 53 : nvme_ctrlr_get(const struct spdk_nvme_transport_id *trid, const char *hostnqn)
394 : {
395 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
396 53 : struct nvme_ctrlr *nvme_ctrlr = NULL;
397 :
398 53 : pthread_mutex_lock(&g_bdev_nvme_mutex);
399 72 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
400 19 : nvme_ctrlr = nvme_bdev_ctrlr_get_ctrlr(nbdev_ctrlr, trid, hostnqn);
401 19 : if (nvme_ctrlr != NULL) {
402 0 : break;
403 : }
404 : }
405 53 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
406 :
407 53 : return nvme_ctrlr;
408 : }
409 :
410 : struct nvme_ctrlr *
411 126 : nvme_ctrlr_get_by_name(const char *name)
412 : {
413 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
414 126 : struct nvme_ctrlr *nvme_ctrlr = NULL;
415 :
416 126 : if (name == NULL) {
417 0 : return NULL;
418 : }
419 :
420 126 : pthread_mutex_lock(&g_bdev_nvme_mutex);
421 126 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
422 126 : if (nbdev_ctrlr != NULL) {
423 60 : nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
424 : }
425 126 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
426 :
427 126 : return nvme_ctrlr;
428 : }
429 :
430 : void
431 0 : nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx)
432 : {
433 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
434 :
435 0 : pthread_mutex_lock(&g_bdev_nvme_mutex);
436 0 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
437 0 : fn(nbdev_ctrlr, ctx);
438 : }
439 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
440 0 : }
441 :
442 : struct nvme_ctrlr_channel_iter {
443 : nvme_ctrlr_for_each_channel_msg fn;
444 : nvme_ctrlr_for_each_channel_done cpl;
445 : struct spdk_io_channel_iter *i;
446 : void *ctx;
447 : };
448 :
449 : void
450 166 : nvme_ctrlr_for_each_channel_continue(struct nvme_ctrlr_channel_iter *iter, int status)
451 : {
452 166 : spdk_for_each_channel_continue(iter->i, status);
453 166 : }
454 :
455 : static void
456 166 : nvme_ctrlr_each_channel_msg(struct spdk_io_channel_iter *i)
457 : {
458 166 : struct nvme_ctrlr_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
459 166 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
460 166 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
461 166 : struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(ch);
462 :
463 166 : iter->i = i;
464 166 : iter->fn(iter, nvme_ctrlr, ctrlr_ch, iter->ctx);
465 166 : }
466 :
467 : static void
468 97 : nvme_ctrlr_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
469 : {
470 97 : struct nvme_ctrlr_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
471 97 : struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
472 :
473 97 : iter->i = i;
474 97 : iter->cpl(nvme_ctrlr, iter->ctx, status);
475 :
476 97 : free(iter);
477 97 : }
478 :
479 : void
480 97 : nvme_ctrlr_for_each_channel(struct nvme_ctrlr *nvme_ctrlr,
481 : nvme_ctrlr_for_each_channel_msg fn, void *ctx,
482 : nvme_ctrlr_for_each_channel_done cpl)
483 : {
484 : struct nvme_ctrlr_channel_iter *iter;
485 :
486 97 : assert(nvme_ctrlr != NULL && fn != NULL);
487 :
488 97 : iter = calloc(1, sizeof(struct nvme_ctrlr_channel_iter));
489 97 : if (iter == NULL) {
490 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
491 0 : assert(false);
492 : return;
493 : }
494 :
495 97 : iter->fn = fn;
496 97 : iter->cpl = cpl;
497 97 : iter->ctx = ctx;
498 :
499 97 : spdk_for_each_channel(nvme_ctrlr, nvme_ctrlr_each_channel_msg,
500 : iter, nvme_ctrlr_each_channel_cpl);
501 : }
502 :
503 : struct nvme_bdev_channel_iter {
504 : nvme_bdev_for_each_channel_msg fn;
505 : nvme_bdev_for_each_channel_done cpl;
506 : struct spdk_io_channel_iter *i;
507 : void *ctx;
508 : };
509 :
510 : void
511 69 : nvme_bdev_for_each_channel_continue(struct nvme_bdev_channel_iter *iter, int status)
512 : {
513 69 : spdk_for_each_channel_continue(iter->i, status);
514 69 : }
515 :
516 : static void
517 69 : nvme_bdev_each_channel_msg(struct spdk_io_channel_iter *i)
518 : {
519 69 : struct nvme_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
520 69 : struct nvme_bdev *nbdev = spdk_io_channel_iter_get_io_device(i);
521 69 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
522 69 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
523 :
524 69 : iter->i = i;
525 69 : iter->fn(iter, nbdev, nbdev_ch, iter->ctx);
526 69 : }
527 :
528 : static void
529 60 : nvme_bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
530 : {
531 60 : struct nvme_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
532 60 : struct nvme_bdev *nbdev = spdk_io_channel_iter_get_io_device(i);
533 :
534 60 : iter->i = i;
535 60 : iter->cpl(nbdev, iter->ctx, status);
536 :
537 60 : free(iter);
538 60 : }
539 :
540 : void
541 60 : nvme_bdev_for_each_channel(struct nvme_bdev *nbdev,
542 : nvme_bdev_for_each_channel_msg fn, void *ctx,
543 : nvme_bdev_for_each_channel_done cpl)
544 : {
545 : struct nvme_bdev_channel_iter *iter;
546 :
547 60 : assert(nbdev != NULL && fn != NULL);
548 :
549 60 : iter = calloc(1, sizeof(struct nvme_bdev_channel_iter));
550 60 : if (iter == NULL) {
551 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
552 0 : assert(false);
553 : return;
554 : }
555 :
556 60 : iter->fn = fn;
557 60 : iter->cpl = cpl;
558 60 : iter->ctx = ctx;
559 :
560 60 : spdk_for_each_channel(nbdev, nvme_bdev_each_channel_msg, iter,
561 : nvme_bdev_each_channel_cpl);
562 : }
563 :
564 : void
565 0 : nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid, struct spdk_json_write_ctx *w)
566 : {
567 : const char *trtype_str;
568 : const char *adrfam_str;
569 :
570 0 : trtype_str = spdk_nvme_transport_id_trtype_str(trid->trtype);
571 0 : if (trtype_str) {
572 0 : spdk_json_write_named_string(w, "trtype", trtype_str);
573 : }
574 :
575 0 : adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
576 0 : if (adrfam_str) {
577 0 : spdk_json_write_named_string(w, "adrfam", adrfam_str);
578 : }
579 :
580 0 : if (trid->traddr[0] != '\0') {
581 0 : spdk_json_write_named_string(w, "traddr", trid->traddr);
582 : }
583 :
584 0 : if (trid->trsvcid[0] != '\0') {
585 0 : spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
586 : }
587 :
588 0 : if (trid->subnqn[0] != '\0') {
589 0 : spdk_json_write_named_string(w, "subnqn", trid->subnqn);
590 : }
591 0 : }
592 :
593 : static void
594 61 : nvme_bdev_ctrlr_delete(struct nvme_bdev_ctrlr *nbdev_ctrlr,
595 : struct nvme_ctrlr *nvme_ctrlr)
596 : {
597 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_delete, nvme_ctrlr->nbdev_ctrlr->name);
598 61 : pthread_mutex_lock(&g_bdev_nvme_mutex);
599 :
600 61 : TAILQ_REMOVE(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
601 61 : if (!TAILQ_EMPTY(&nbdev_ctrlr->ctrlrs)) {
602 15 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
603 :
604 15 : return;
605 : }
606 46 : TAILQ_REMOVE(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
607 :
608 46 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
609 :
610 46 : assert(TAILQ_EMPTY(&nbdev_ctrlr->bdevs));
611 :
612 46 : free(nbdev_ctrlr->name);
613 46 : free(nbdev_ctrlr);
614 : }
615 :
616 : static void
617 62 : _nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
618 : {
619 : struct nvme_path_id *path_id, *tmp_path;
620 : struct nvme_ns *ns, *tmp_ns;
621 :
622 62 : free(nvme_ctrlr->copied_ana_desc);
623 62 : spdk_free(nvme_ctrlr->ana_log_page);
624 :
625 62 : if (nvme_ctrlr->opal_dev) {
626 0 : spdk_opal_dev_destruct(nvme_ctrlr->opal_dev);
627 0 : nvme_ctrlr->opal_dev = NULL;
628 : }
629 :
630 62 : if (nvme_ctrlr->nbdev_ctrlr) {
631 61 : nvme_bdev_ctrlr_delete(nvme_ctrlr->nbdev_ctrlr, nvme_ctrlr);
632 : }
633 :
634 62 : RB_FOREACH_SAFE(ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp_ns) {
635 0 : RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
636 0 : nvme_ns_free(ns);
637 : }
638 :
639 124 : TAILQ_FOREACH_SAFE(path_id, &nvme_ctrlr->trids, link, tmp_path) {
640 62 : TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
641 62 : free(path_id);
642 : }
643 :
644 62 : pthread_mutex_destroy(&nvme_ctrlr->mutex);
645 62 : spdk_keyring_put_key(nvme_ctrlr->psk);
646 62 : spdk_keyring_put_key(nvme_ctrlr->dhchap_key);
647 62 : spdk_keyring_put_key(nvme_ctrlr->dhchap_ctrlr_key);
648 62 : free(nvme_ctrlr);
649 :
650 62 : pthread_mutex_lock(&g_bdev_nvme_mutex);
651 62 : if (g_bdev_nvme_module_finish && TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
652 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
653 0 : spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
654 0 : spdk_bdev_module_fini_done();
655 0 : return;
656 : }
657 62 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
658 : }
659 :
660 : static int
661 62 : nvme_detach_poller(void *arg)
662 : {
663 62 : struct nvme_ctrlr *nvme_ctrlr = arg;
664 : int rc;
665 :
666 62 : rc = spdk_nvme_detach_poll_async(nvme_ctrlr->detach_ctx);
667 62 : if (rc != -EAGAIN) {
668 62 : spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
669 62 : _nvme_ctrlr_delete(nvme_ctrlr);
670 : }
671 :
672 62 : return SPDK_POLLER_BUSY;
673 : }
674 :
675 : static void
676 62 : nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
677 : {
678 : int rc;
679 :
680 62 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
681 :
682 62 : if (spdk_interrupt_mode_is_enabled()) {
683 0 : spdk_interrupt_unregister(&nvme_ctrlr->intr);
684 : }
685 :
686 : /* First, unregister the adminq poller, as the driver will poll adminq if necessary */
687 62 : spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);
688 :
689 : /* If we got here, the reset/detach poller cannot be active */
690 62 : assert(nvme_ctrlr->reset_detach_poller == NULL);
691 62 : nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(nvme_detach_poller,
692 : nvme_ctrlr, 1000);
693 62 : if (nvme_ctrlr->reset_detach_poller == NULL) {
694 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to register detach poller\n");
695 0 : goto error;
696 : }
697 :
698 62 : rc = spdk_nvme_detach_async(nvme_ctrlr->ctrlr, &nvme_ctrlr->detach_ctx);
699 62 : if (rc != 0) {
700 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to detach the NVMe controller\n");
701 0 : goto error;
702 : }
703 :
704 62 : return;
705 0 : error:
706 : /* We don't have a good way to handle errors here, so just do what we can and delete the
707 : * controller without detaching the underlying NVMe device.
708 : */
709 0 : spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
710 0 : _nvme_ctrlr_delete(nvme_ctrlr);
711 : }
712 :
713 : static void
714 61 : nvme_ctrlr_unregister_cb(void *io_device)
715 : {
716 61 : struct nvme_ctrlr *nvme_ctrlr = io_device;
717 :
718 61 : nvme_ctrlr_delete(nvme_ctrlr);
719 61 : }
720 :
721 : static void
722 61 : nvme_ctrlr_unregister(void *ctx)
723 : {
724 61 : struct nvme_ctrlr *nvme_ctrlr = ctx;
725 :
726 61 : spdk_io_device_unregister(nvme_ctrlr, nvme_ctrlr_unregister_cb);
727 61 : }
728 :
729 : static bool
730 249 : nvme_ctrlr_can_be_unregistered(struct nvme_ctrlr *nvme_ctrlr)
731 : {
732 249 : if (!nvme_ctrlr->destruct) {
733 131 : return false;
734 : }
735 :
736 118 : if (nvme_ctrlr->ref > 0) {
737 57 : return false;
738 : }
739 :
740 61 : if (nvme_ctrlr->resetting) {
741 0 : return false;
742 : }
743 :
744 61 : if (nvme_ctrlr->ana_log_page_updating) {
745 0 : return false;
746 : }
747 :
748 61 : if (nvme_ctrlr->io_path_cache_clearing) {
749 0 : return false;
750 : }
751 :
752 61 : return true;
753 : }
754 :
755 : static void
756 172 : nvme_ctrlr_put_ref(struct nvme_ctrlr *nvme_ctrlr)
757 : {
758 172 : pthread_mutex_lock(&nvme_ctrlr->mutex);
759 : SPDK_DTRACE_PROBE2(bdev_nvme_ctrlr_release, nvme_ctrlr->nbdev_ctrlr->name, nvme_ctrlr->ref);
760 :
761 172 : assert(nvme_ctrlr->ref > 0);
762 172 : nvme_ctrlr->ref--;
763 :
764 172 : if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
765 111 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
766 111 : return;
767 : }
768 :
769 61 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
770 :
771 61 : spdk_thread_exec_msg(nvme_ctrlr->thread, nvme_ctrlr_unregister, nvme_ctrlr);
772 : }
773 :
774 : static void
775 111 : nvme_ctrlr_get_ref(struct nvme_ctrlr *nvme_ctrlr)
776 : {
777 111 : pthread_mutex_lock(&nvme_ctrlr->mutex);
778 111 : nvme_ctrlr->ref++;
779 111 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
780 111 : }
781 :
782 : static void
783 259 : bdev_nvme_clear_current_io_path(struct nvme_bdev_channel *nbdev_ch)
784 : {
785 259 : nbdev_ch->current_io_path = NULL;
786 259 : nbdev_ch->rr_counter = 0;
787 259 : }
788 :
789 : static struct nvme_io_path *
790 8 : _bdev_nvme_get_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
791 : {
792 : struct nvme_io_path *io_path;
793 :
794 16 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
795 15 : if (io_path->nvme_ns == nvme_ns) {
796 7 : break;
797 : }
798 : }
799 :
800 8 : return io_path;
801 : }
802 :
803 : static struct nvme_io_path *
804 39 : nvme_io_path_alloc(void)
805 : {
806 : struct nvme_io_path *io_path;
807 :
808 39 : io_path = calloc(1, sizeof(*io_path));
809 39 : if (io_path == NULL) {
810 0 : SPDK_ERRLOG("Failed to alloc io_path.\n");
811 0 : return NULL;
812 : }
813 :
814 39 : if (g_opts.io_path_stat) {
815 0 : io_path->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
816 0 : if (io_path->stat == NULL) {
817 0 : free(io_path);
818 0 : SPDK_ERRLOG("Failed to alloc io_path stat.\n");
819 0 : return NULL;
820 : }
821 0 : spdk_bdev_reset_io_stat(io_path->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
822 : }
823 :
824 39 : return io_path;
825 : }
826 :
827 : static void
828 39 : nvme_io_path_free(struct nvme_io_path *io_path)
829 : {
830 39 : free(io_path->stat);
831 39 : free(io_path);
832 39 : }
833 :
834 : static int
835 39 : _bdev_nvme_add_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
836 : {
837 : struct nvme_io_path *io_path;
838 : struct spdk_io_channel *ch;
839 : struct nvme_ctrlr_channel *ctrlr_ch;
840 : struct nvme_qpair *nvme_qpair;
841 :
842 39 : io_path = nvme_io_path_alloc();
843 39 : if (io_path == NULL) {
844 0 : return -ENOMEM;
845 : }
846 :
847 39 : io_path->nvme_ns = nvme_ns;
848 :
849 39 : ch = spdk_get_io_channel(nvme_ns->ctrlr);
850 39 : if (ch == NULL) {
851 0 : nvme_io_path_free(io_path);
852 0 : SPDK_ERRLOG("Failed to alloc io_channel.\n");
853 0 : return -ENOMEM;
854 : }
855 :
856 39 : ctrlr_ch = spdk_io_channel_get_ctx(ch);
857 :
858 39 : nvme_qpair = ctrlr_ch->qpair;
859 39 : assert(nvme_qpair != NULL);
860 :
861 39 : io_path->qpair = nvme_qpair;
862 39 : TAILQ_INSERT_TAIL(&nvme_qpair->io_path_list, io_path, tailq);
863 :
864 39 : io_path->nbdev_ch = nbdev_ch;
865 39 : STAILQ_INSERT_TAIL(&nbdev_ch->io_path_list, io_path, stailq);
866 :
867 39 : bdev_nvme_clear_current_io_path(nbdev_ch);
868 :
869 39 : return 0;
870 : }
871 :
872 : static void
873 39 : bdev_nvme_clear_retry_io_path(struct nvme_bdev_channel *nbdev_ch,
874 : struct nvme_io_path *io_path)
875 : {
876 : struct nvme_bdev_io *bio;
877 :
878 40 : TAILQ_FOREACH(bio, &nbdev_ch->retry_io_list, retry_link) {
879 1 : if (bio->io_path == io_path) {
880 1 : bio->io_path = NULL;
881 : }
882 : }
883 39 : }
884 :
885 : static void
886 39 : _bdev_nvme_delete_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *io_path)
887 : {
888 : struct spdk_io_channel *ch;
889 : struct nvme_qpair *nvme_qpair;
890 : struct nvme_ctrlr_channel *ctrlr_ch;
891 : struct nvme_bdev *nbdev;
892 :
893 39 : nbdev = spdk_io_channel_get_io_device(spdk_io_channel_from_ctx(nbdev_ch));
894 :
895 : /* Add the statistics to nvme_ns before this path is destroyed. */
896 39 : pthread_mutex_lock(&nbdev->mutex);
897 39 : if (nbdev->ref != 0 && io_path->nvme_ns->stat != NULL && io_path->stat != NULL) {
898 0 : spdk_bdev_add_io_stat(io_path->nvme_ns->stat, io_path->stat);
899 : }
900 39 : pthread_mutex_unlock(&nbdev->mutex);
901 :
902 39 : bdev_nvme_clear_current_io_path(nbdev_ch);
903 39 : bdev_nvme_clear_retry_io_path(nbdev_ch, io_path);
904 :
905 39 : STAILQ_REMOVE(&nbdev_ch->io_path_list, io_path, nvme_io_path, stailq);
906 39 : io_path->nbdev_ch = NULL;
907 :
908 39 : nvme_qpair = io_path->qpair;
909 39 : assert(nvme_qpair != NULL);
910 :
911 39 : ctrlr_ch = nvme_qpair->ctrlr_ch;
912 39 : assert(ctrlr_ch != NULL);
913 :
914 39 : ch = spdk_io_channel_from_ctx(ctrlr_ch);
915 39 : spdk_put_io_channel(ch);
916 :
917 : /* After an io_path is removed, I/Os submitted to it may complete and update statistics
918 : * of the io_path. To avoid heap-use-after-free error from this case, do not free the
919 : * io_path here but free the io_path when the associated qpair is freed. It is ensured
920 : * that all I/Os submitted to the io_path are completed when the associated qpair is freed.
921 : */
922 39 : }
923 :
924 : static void
925 26 : _bdev_nvme_delete_io_paths(struct nvme_bdev_channel *nbdev_ch)
926 : {
927 : struct nvme_io_path *io_path, *tmp_io_path;
928 :
929 63 : STAILQ_FOREACH_SAFE(io_path, &nbdev_ch->io_path_list, stailq, tmp_io_path) {
930 37 : _bdev_nvme_delete_io_path(nbdev_ch, io_path);
931 : }
932 26 : }
933 :
934 : static int
935 26 : bdev_nvme_create_bdev_channel_cb(void *io_device, void *ctx_buf)
936 : {
937 26 : struct nvme_bdev_channel *nbdev_ch = ctx_buf;
938 26 : struct nvme_bdev *nbdev = io_device;
939 : struct nvme_ns *nvme_ns;
940 : int rc;
941 :
942 26 : STAILQ_INIT(&nbdev_ch->io_path_list);
943 26 : TAILQ_INIT(&nbdev_ch->retry_io_list);
944 :
945 26 : pthread_mutex_lock(&nbdev->mutex);
946 :
947 26 : nbdev_ch->mp_policy = nbdev->mp_policy;
948 26 : nbdev_ch->mp_selector = nbdev->mp_selector;
949 26 : nbdev_ch->rr_min_io = nbdev->rr_min_io;
950 :
951 63 : TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
952 37 : rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
953 37 : if (rc != 0) {
954 0 : pthread_mutex_unlock(&nbdev->mutex);
955 :
956 0 : _bdev_nvme_delete_io_paths(nbdev_ch);
957 0 : return rc;
958 : }
959 : }
960 26 : pthread_mutex_unlock(&nbdev->mutex);
961 :
962 26 : return 0;
963 : }
964 :
965 : /* If cpl != NULL, complete the bdev_io with nvme status based on 'cpl'.
966 : * If cpl == NULL, complete the bdev_io with bdev status based on 'status'.
967 : */
968 : static inline void
969 58 : __bdev_nvme_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status,
970 : const struct spdk_nvme_cpl *cpl)
971 : {
972 58 : spdk_trace_record(TRACE_BDEV_NVME_IO_DONE, 0, 0, (uintptr_t)bdev_io->driver_ctx,
973 : (uintptr_t)bdev_io);
974 58 : if (cpl) {
975 29 : spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc);
976 : } else {
977 29 : spdk_bdev_io_complete(bdev_io, status);
978 : }
979 58 : }
980 :
981 : static void bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch);
982 :
983 : static void
984 26 : bdev_nvme_destroy_bdev_channel_cb(void *io_device, void *ctx_buf)
985 : {
986 26 : struct nvme_bdev_channel *nbdev_ch = ctx_buf;
987 :
988 26 : bdev_nvme_abort_retry_ios(nbdev_ch);
989 26 : _bdev_nvme_delete_io_paths(nbdev_ch);
990 26 : }
991 :
992 : static inline bool
993 62 : bdev_nvme_io_type_is_admin(enum spdk_bdev_io_type io_type)
994 : {
995 62 : switch (io_type) {
996 5 : case SPDK_BDEV_IO_TYPE_RESET:
997 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
998 : case SPDK_BDEV_IO_TYPE_ABORT:
999 5 : return true;
1000 57 : default:
1001 57 : break;
1002 : }
1003 :
1004 57 : return false;
1005 : }
1006 :
1007 : static inline bool
1008 98 : nvme_ns_is_active(struct nvme_ns *nvme_ns)
1009 : {
1010 98 : if (spdk_unlikely(nvme_ns->ana_state_updating)) {
1011 1 : return false;
1012 : }
1013 :
1014 97 : if (spdk_unlikely(nvme_ns->ns == NULL)) {
1015 0 : return false;
1016 : }
1017 :
1018 97 : return true;
1019 : }
1020 :
1021 : static inline bool
1022 86 : nvme_ns_is_accessible(struct nvme_ns *nvme_ns)
1023 : {
1024 86 : if (spdk_unlikely(!nvme_ns_is_active(nvme_ns))) {
1025 1 : return false;
1026 : }
1027 :
1028 85 : switch (nvme_ns->ana_state) {
1029 76 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
1030 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1031 76 : return true;
1032 9 : default:
1033 9 : break;
1034 : }
1035 :
1036 9 : return false;
1037 : }
1038 :
1039 : static inline bool
1040 128 : nvme_qpair_is_connected(struct nvme_qpair *nvme_qpair)
1041 : {
1042 128 : if (spdk_unlikely(nvme_qpair->qpair == NULL)) {
1043 23 : return false;
1044 : }
1045 :
1046 105 : if (spdk_unlikely(spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
1047 : SPDK_NVME_QPAIR_FAILURE_NONE)) {
1048 2 : return false;
1049 : }
1050 :
1051 103 : if (spdk_unlikely(nvme_qpair->ctrlr_ch->reset_iter != NULL)) {
1052 0 : return false;
1053 : }
1054 :
1055 103 : return true;
1056 : }
1057 :
1058 : static inline bool
1059 102 : nvme_io_path_is_available(struct nvme_io_path *io_path)
1060 : {
1061 102 : if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
1062 16 : return false;
1063 : }
1064 :
1065 86 : if (spdk_unlikely(!nvme_ns_is_accessible(io_path->nvme_ns))) {
1066 10 : return false;
1067 : }
1068 :
1069 76 : return true;
1070 : }
1071 :
1072 : static inline bool
1073 9 : nvme_ctrlr_is_failed(struct nvme_ctrlr *nvme_ctrlr)
1074 : {
1075 9 : if (nvme_ctrlr->destruct) {
1076 0 : return true;
1077 : }
1078 :
1079 9 : if (nvme_ctrlr->fast_io_fail_timedout) {
1080 2 : return true;
1081 : }
1082 :
1083 7 : if (nvme_ctrlr->resetting) {
1084 5 : if (nvme_ctrlr->opts.reconnect_delay_sec != 0) {
1085 5 : return false;
1086 : } else {
1087 0 : return true;
1088 : }
1089 : }
1090 :
1091 2 : if (nvme_ctrlr->reconnect_is_delayed) {
1092 2 : return false;
1093 : }
1094 :
1095 0 : if (nvme_ctrlr->disabled) {
1096 0 : return true;
1097 : }
1098 :
1099 0 : if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
1100 0 : return true;
1101 : } else {
1102 0 : return false;
1103 : }
1104 : }
1105 :
1106 : static bool
1107 20 : nvme_ctrlr_is_available(struct nvme_ctrlr *nvme_ctrlr)
1108 : {
1109 20 : if (nvme_ctrlr->destruct) {
1110 0 : return false;
1111 : }
1112 :
1113 20 : if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
1114 3 : return false;
1115 : }
1116 :
1117 17 : if (nvme_ctrlr->resetting || nvme_ctrlr->reconnect_is_delayed) {
1118 1 : return false;
1119 : }
1120 :
1121 16 : if (nvme_ctrlr->disabled) {
1122 0 : return false;
1123 : }
1124 :
1125 16 : return true;
1126 : }
1127 :
1128 : /* Simulate circular linked list. */
1129 : static inline struct nvme_io_path *
1130 99 : nvme_io_path_get_next(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *prev_path)
1131 : {
1132 : struct nvme_io_path *next_path;
1133 :
1134 99 : if (prev_path != NULL) {
1135 39 : next_path = STAILQ_NEXT(prev_path, stailq);
1136 39 : if (next_path != NULL) {
1137 14 : return next_path;
1138 : }
1139 : }
1140 :
1141 85 : return STAILQ_FIRST(&nbdev_ch->io_path_list);
1142 : }
1143 :
1144 : static struct nvme_io_path *
1145 67 : _bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
1146 : {
1147 67 : struct nvme_io_path *io_path, *start, *non_optimized = NULL;
1148 :
1149 67 : start = nvme_io_path_get_next(nbdev_ch, nbdev_ch->current_io_path);
1150 :
1151 67 : io_path = start;
1152 : do {
1153 79 : if (spdk_likely(nvme_io_path_is_available(io_path))) {
1154 57 : switch (io_path->nvme_ns->ana_state) {
1155 47 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
1156 47 : nbdev_ch->current_io_path = io_path;
1157 47 : return io_path;
1158 10 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1159 10 : if (non_optimized == NULL) {
1160 7 : non_optimized = io_path;
1161 : }
1162 10 : break;
1163 0 : default:
1164 0 : assert(false);
1165 : break;
1166 : }
1167 : }
1168 32 : io_path = nvme_io_path_get_next(nbdev_ch, io_path);
1169 32 : } while (io_path != start);
1170 :
1171 20 : if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
1172 : /* We come here only if there is no optimized path. Cache even non_optimized
1173 : * path for load balance across multiple non_optimized paths.
1174 : */
1175 1 : nbdev_ch->current_io_path = non_optimized;
1176 : }
1177 :
1178 20 : return non_optimized;
1179 : }
1180 :
1181 : static struct nvme_io_path *
1182 4 : _bdev_nvme_find_io_path_min_qd(struct nvme_bdev_channel *nbdev_ch)
1183 : {
1184 : struct nvme_io_path *io_path;
1185 4 : struct nvme_io_path *optimized = NULL, *non_optimized = NULL;
1186 4 : uint32_t opt_min_qd = UINT32_MAX, non_opt_min_qd = UINT32_MAX;
1187 : uint32_t num_outstanding_reqs;
1188 :
1189 16 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
1190 12 : if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
1191 : /* The device is currently resetting. */
1192 0 : continue;
1193 : }
1194 :
1195 12 : if (spdk_unlikely(!nvme_ns_is_active(io_path->nvme_ns))) {
1196 0 : continue;
1197 : }
1198 :
1199 12 : num_outstanding_reqs = spdk_nvme_qpair_get_num_outstanding_reqs(io_path->qpair->qpair);
1200 12 : switch (io_path->nvme_ns->ana_state) {
1201 6 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
1202 6 : if (num_outstanding_reqs < opt_min_qd) {
1203 5 : opt_min_qd = num_outstanding_reqs;
1204 5 : optimized = io_path;
1205 : }
1206 6 : break;
1207 3 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1208 3 : if (num_outstanding_reqs < non_opt_min_qd) {
1209 3 : non_opt_min_qd = num_outstanding_reqs;
1210 3 : non_optimized = io_path;
1211 : }
1212 3 : break;
1213 3 : default:
1214 3 : break;
1215 : }
1216 : }
1217 :
1218 : /* don't cache io path for BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH selector */
1219 4 : if (optimized != NULL) {
1220 3 : return optimized;
1221 : }
1222 :
1223 1 : return non_optimized;
1224 : }
1225 :
1226 : static inline struct nvme_io_path *
1227 105 : bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
1228 : {
1229 105 : if (spdk_likely(nbdev_ch->current_io_path != NULL)) {
1230 41 : if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE) {
1231 31 : return nbdev_ch->current_io_path;
1232 10 : } else if (nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
1233 10 : if (++nbdev_ch->rr_counter < nbdev_ch->rr_min_io) {
1234 3 : return nbdev_ch->current_io_path;
1235 : }
1236 7 : nbdev_ch->rr_counter = 0;
1237 : }
1238 : }
1239 :
1240 71 : if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE ||
1241 14 : nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
1242 67 : return _bdev_nvme_find_io_path(nbdev_ch);
1243 : } else {
1244 4 : return _bdev_nvme_find_io_path_min_qd(nbdev_ch);
1245 : }
1246 : }
1247 :
1248 : /* Return true if there is any io_path whose qpair is active or ctrlr is not failed,
1249 : * or false otherwise.
1250 : *
1251 : * If any io_path has an active qpair but find_io_path() returned NULL, its namespace
1252 : * is likely to be non-accessible now but may become accessible.
1253 : *
1254 : * If any io_path has an unfailed ctrlr but find_io_path() returned NULL, the ctrlr
1255 : * is likely to be resetting now but the reset may succeed. A ctrlr is set to unfailed
1256 : * when starting to reset it but it is set to failed when the reset failed. Hence, if
1257 : * a ctrlr is unfailed, it is likely that it works fine or is resetting.
1258 : */
1259 : static bool
1260 15 : any_io_path_may_become_available(struct nvme_bdev_channel *nbdev_ch)
1261 : {
1262 : struct nvme_io_path *io_path;
1263 :
1264 15 : if (nbdev_ch->resetting) {
1265 1 : return false;
1266 : }
1267 :
1268 16 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
1269 14 : if (io_path->nvme_ns->ana_transition_timedout) {
1270 0 : continue;
1271 : }
1272 :
1273 14 : if (nvme_qpair_is_connected(io_path->qpair) ||
1274 9 : !nvme_ctrlr_is_failed(io_path->qpair->ctrlr)) {
1275 12 : return true;
1276 : }
1277 : }
1278 :
1279 2 : return false;
1280 : }
1281 :
1282 : static void
1283 14 : bdev_nvme_retry_io(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
1284 : {
1285 14 : struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
1286 : struct spdk_io_channel *ch;
1287 :
1288 14 : if (nbdev_io->io_path != NULL && nvme_io_path_is_available(nbdev_io->io_path)) {
1289 3 : _bdev_nvme_submit_request(nbdev_ch, bdev_io);
1290 : } else {
1291 11 : ch = spdk_io_channel_from_ctx(nbdev_ch);
1292 11 : bdev_nvme_submit_request(ch, bdev_io);
1293 : }
1294 14 : }
1295 :
1296 : static int
1297 14 : bdev_nvme_retry_ios(void *arg)
1298 : {
1299 14 : struct nvme_bdev_channel *nbdev_ch = arg;
1300 : struct nvme_bdev_io *bio, *tmp_bio;
1301 : uint64_t now, delay_us;
1302 :
1303 14 : now = spdk_get_ticks();
1304 :
1305 28 : TAILQ_FOREACH_SAFE(bio, &nbdev_ch->retry_io_list, retry_link, tmp_bio) {
1306 15 : if (bio->retry_ticks > now) {
1307 1 : break;
1308 : }
1309 :
1310 14 : TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
1311 :
1312 14 : bdev_nvme_retry_io(nbdev_ch, spdk_bdev_io_from_ctx(bio));
1313 : }
1314 :
1315 14 : spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1316 :
1317 14 : bio = TAILQ_FIRST(&nbdev_ch->retry_io_list);
1318 14 : if (bio != NULL) {
1319 4 : delay_us = (bio->retry_ticks - now) * SPDK_SEC_TO_USEC / spdk_get_ticks_hz();
1320 :
1321 4 : nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
1322 : delay_us);
1323 : }
1324 :
1325 14 : return SPDK_POLLER_BUSY;
1326 : }
1327 :
1328 : static void
1329 16 : bdev_nvme_queue_retry_io(struct nvme_bdev_channel *nbdev_ch,
1330 : struct nvme_bdev_io *bio, uint64_t delay_ms)
1331 : {
1332 : struct nvme_bdev_io *tmp_bio;
1333 :
1334 16 : bio->retry_ticks = spdk_get_ticks() + delay_ms * spdk_get_ticks_hz() / 1000ULL;
1335 :
1336 16 : TAILQ_FOREACH_REVERSE(tmp_bio, &nbdev_ch->retry_io_list, retry_io_head, retry_link) {
1337 1 : if (tmp_bio->retry_ticks <= bio->retry_ticks) {
1338 1 : TAILQ_INSERT_AFTER(&nbdev_ch->retry_io_list, tmp_bio, bio,
1339 : retry_link);
1340 1 : return;
1341 : }
1342 : }
1343 :
1344 : /* No earlier I/Os were found. This I/O must be the new head. */
1345 15 : TAILQ_INSERT_HEAD(&nbdev_ch->retry_io_list, bio, retry_link);
1346 :
1347 15 : spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1348 :
1349 15 : nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
1350 : delay_ms * 1000ULL);
1351 : }
1352 :
1353 : static void
1354 58 : bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch)
1355 : {
1356 : struct nvme_bdev_io *bio, *tmp_bio;
1357 :
1358 59 : TAILQ_FOREACH_SAFE(bio, &nbdev_ch->retry_io_list, retry_link, tmp_bio) {
1359 1 : TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
1360 1 : __bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_ABORTED, NULL);
1361 : }
1362 :
1363 58 : spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1364 58 : }
1365 :
1366 : static int
1367 6 : bdev_nvme_abort_retry_io(struct nvme_bdev_channel *nbdev_ch,
1368 : struct nvme_bdev_io *bio_to_abort)
1369 : {
1370 : struct nvme_bdev_io *bio;
1371 :
1372 6 : TAILQ_FOREACH(bio, &nbdev_ch->retry_io_list, retry_link) {
1373 1 : if (bio == bio_to_abort) {
1374 1 : TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
1375 1 : __bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_ABORTED, NULL);
1376 1 : return 0;
1377 : }
1378 : }
1379 :
1380 5 : return -ENOENT;
1381 : }
1382 :
1383 : static void
1384 12 : bdev_nvme_update_nvme_error_stat(struct spdk_bdev_io *bdev_io, const struct spdk_nvme_cpl *cpl)
1385 : {
1386 : struct nvme_bdev *nbdev;
1387 : uint16_t sct, sc;
1388 :
1389 12 : assert(spdk_nvme_cpl_is_error(cpl));
1390 :
1391 12 : nbdev = bdev_io->bdev->ctxt;
1392 :
1393 12 : if (nbdev->err_stat == NULL) {
1394 12 : return;
1395 : }
1396 :
1397 0 : sct = cpl->status.sct;
1398 0 : sc = cpl->status.sc;
1399 :
1400 0 : pthread_mutex_lock(&nbdev->mutex);
1401 :
1402 0 : nbdev->err_stat->status_type[sct]++;
1403 0 : switch (sct) {
1404 0 : case SPDK_NVME_SCT_GENERIC:
1405 : case SPDK_NVME_SCT_COMMAND_SPECIFIC:
1406 : case SPDK_NVME_SCT_MEDIA_ERROR:
1407 : case SPDK_NVME_SCT_PATH:
1408 0 : nbdev->err_stat->status[sct][sc]++;
1409 0 : break;
1410 0 : default:
1411 0 : break;
1412 : }
1413 :
1414 0 : pthread_mutex_unlock(&nbdev->mutex);
1415 : }
1416 :
1417 : static inline void
1418 20 : bdev_nvme_update_io_path_stat(struct nvme_bdev_io *bio)
1419 : {
1420 20 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1421 20 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
1422 20 : uint32_t blocklen = bdev_io->bdev->blocklen;
1423 : struct spdk_bdev_io_stat *stat;
1424 : uint64_t tsc_diff;
1425 :
1426 20 : if (bio->io_path->stat == NULL) {
1427 20 : return;
1428 : }
1429 :
1430 0 : tsc_diff = spdk_get_ticks() - bio->submit_tsc;
1431 0 : stat = bio->io_path->stat;
1432 :
1433 0 : switch (bdev_io->type) {
1434 0 : case SPDK_BDEV_IO_TYPE_READ:
1435 0 : stat->bytes_read += num_blocks * blocklen;
1436 0 : stat->num_read_ops++;
1437 0 : stat->read_latency_ticks += tsc_diff;
1438 0 : if (stat->max_read_latency_ticks < tsc_diff) {
1439 0 : stat->max_read_latency_ticks = tsc_diff;
1440 : }
1441 0 : if (stat->min_read_latency_ticks > tsc_diff) {
1442 0 : stat->min_read_latency_ticks = tsc_diff;
1443 : }
1444 0 : break;
1445 0 : case SPDK_BDEV_IO_TYPE_WRITE:
1446 0 : stat->bytes_written += num_blocks * blocklen;
1447 0 : stat->num_write_ops++;
1448 0 : stat->write_latency_ticks += tsc_diff;
1449 0 : if (stat->max_write_latency_ticks < tsc_diff) {
1450 0 : stat->max_write_latency_ticks = tsc_diff;
1451 : }
1452 0 : if (stat->min_write_latency_ticks > tsc_diff) {
1453 0 : stat->min_write_latency_ticks = tsc_diff;
1454 : }
1455 0 : break;
1456 0 : case SPDK_BDEV_IO_TYPE_UNMAP:
1457 0 : stat->bytes_unmapped += num_blocks * blocklen;
1458 0 : stat->num_unmap_ops++;
1459 0 : stat->unmap_latency_ticks += tsc_diff;
1460 0 : if (stat->max_unmap_latency_ticks < tsc_diff) {
1461 0 : stat->max_unmap_latency_ticks = tsc_diff;
1462 : }
1463 0 : if (stat->min_unmap_latency_ticks > tsc_diff) {
1464 0 : stat->min_unmap_latency_ticks = tsc_diff;
1465 : }
1466 0 : break;
1467 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
1468 : /* Track the data in the start phase only */
1469 0 : if (!bdev_io->u.bdev.zcopy.start) {
1470 0 : break;
1471 : }
1472 0 : if (bdev_io->u.bdev.zcopy.populate) {
1473 0 : stat->bytes_read += num_blocks * blocklen;
1474 0 : stat->num_read_ops++;
1475 0 : stat->read_latency_ticks += tsc_diff;
1476 0 : if (stat->max_read_latency_ticks < tsc_diff) {
1477 0 : stat->max_read_latency_ticks = tsc_diff;
1478 : }
1479 0 : if (stat->min_read_latency_ticks > tsc_diff) {
1480 0 : stat->min_read_latency_ticks = tsc_diff;
1481 : }
1482 : } else {
1483 0 : stat->bytes_written += num_blocks * blocklen;
1484 0 : stat->num_write_ops++;
1485 0 : stat->write_latency_ticks += tsc_diff;
1486 0 : if (stat->max_write_latency_ticks < tsc_diff) {
1487 0 : stat->max_write_latency_ticks = tsc_diff;
1488 : }
1489 0 : if (stat->min_write_latency_ticks > tsc_diff) {
1490 0 : stat->min_write_latency_ticks = tsc_diff;
1491 : }
1492 : }
1493 0 : break;
1494 0 : case SPDK_BDEV_IO_TYPE_COPY:
1495 0 : stat->bytes_copied += num_blocks * blocklen;
1496 0 : stat->num_copy_ops++;
1497 0 : stat->copy_latency_ticks += tsc_diff;
1498 0 : if (stat->max_copy_latency_ticks < tsc_diff) {
1499 0 : stat->max_copy_latency_ticks = tsc_diff;
1500 : }
1501 0 : if (stat->min_copy_latency_ticks > tsc_diff) {
1502 0 : stat->min_copy_latency_ticks = tsc_diff;
1503 : }
1504 0 : break;
1505 0 : default:
1506 0 : break;
1507 : }
1508 : }
1509 :
1510 : static bool
1511 7 : bdev_nvme_check_retry_io(struct nvme_bdev_io *bio,
1512 : const struct spdk_nvme_cpl *cpl,
1513 : struct nvme_bdev_channel *nbdev_ch,
1514 : uint64_t *_delay_ms)
1515 : {
1516 7 : struct nvme_io_path *io_path = bio->io_path;
1517 7 : struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
1518 : const struct spdk_nvme_ctrlr_data *cdata;
1519 :
1520 7 : if (spdk_nvme_cpl_is_path_error(cpl) ||
1521 5 : spdk_nvme_cpl_is_aborted_sq_deletion(cpl) ||
1522 4 : !nvme_io_path_is_available(io_path) ||
1523 4 : !nvme_ctrlr_is_available(nvme_ctrlr)) {
1524 3 : bdev_nvme_clear_current_io_path(nbdev_ch);
1525 3 : bio->io_path = NULL;
1526 3 : if (spdk_nvme_cpl_is_ana_error(cpl)) {
1527 1 : if (nvme_ctrlr_read_ana_log_page(nvme_ctrlr) == 0) {
1528 1 : io_path->nvme_ns->ana_state_updating = true;
1529 : }
1530 : }
1531 3 : if (!any_io_path_may_become_available(nbdev_ch)) {
1532 0 : return false;
1533 : }
1534 3 : *_delay_ms = 0;
1535 : } else {
1536 4 : bio->retry_count++;
1537 :
1538 4 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
1539 :
1540 4 : if (cpl->status.crd != 0) {
1541 1 : *_delay_ms = cdata->crdt[cpl->status.crd] * 100;
1542 : } else {
1543 3 : *_delay_ms = 0;
1544 : }
1545 : }
1546 :
1547 7 : return true;
1548 : }
1549 :
1550 : static inline void
1551 32 : bdev_nvme_io_complete_nvme_status(struct nvme_bdev_io *bio,
1552 : const struct spdk_nvme_cpl *cpl)
1553 : {
1554 32 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1555 : struct nvme_bdev_channel *nbdev_ch;
1556 32 : uint64_t delay_ms;
1557 :
1558 32 : assert(!bdev_nvme_io_type_is_admin(bdev_io->type));
1559 :
1560 32 : if (spdk_likely(spdk_nvme_cpl_is_success(cpl))) {
1561 20 : bdev_nvme_update_io_path_stat(bio);
1562 20 : goto complete;
1563 : }
1564 :
1565 : /* Update error counts before deciding if retry is needed.
1566 : * Hence, error counts may be more than the number of I/O errors.
1567 : */
1568 12 : bdev_nvme_update_nvme_error_stat(bdev_io, cpl);
1569 :
1570 12 : if (cpl->status.dnr != 0 || spdk_nvme_cpl_is_aborted_by_request(cpl) ||
1571 8 : (g_opts.bdev_retry_count != -1 && bio->retry_count >= g_opts.bdev_retry_count)) {
1572 5 : goto complete;
1573 : }
1574 :
1575 : /* At this point we don't know whether the sequence was successfully executed or not, so we
1576 : * cannot retry the IO */
1577 7 : if (bdev_io->u.bdev.accel_sequence != NULL) {
1578 0 : goto complete;
1579 : }
1580 :
1581 7 : nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
1582 :
1583 7 : if (bdev_nvme_check_retry_io(bio, cpl, nbdev_ch, &delay_ms)) {
1584 7 : bdev_nvme_queue_retry_io(nbdev_ch, bio, delay_ms);
1585 7 : return;
1586 : }
1587 :
1588 25 : complete:
1589 25 : bio->retry_count = 0;
1590 25 : bio->submit_tsc = 0;
1591 25 : bdev_io->u.bdev.accel_sequence = NULL;
1592 25 : __bdev_nvme_io_complete(bdev_io, 0, cpl);
1593 : }
1594 :
1595 : static inline void
1596 13 : bdev_nvme_io_complete(struct nvme_bdev_io *bio, int rc)
1597 : {
1598 13 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1599 : struct nvme_bdev_channel *nbdev_ch;
1600 : enum spdk_bdev_io_status io_status;
1601 :
1602 13 : assert(!bdev_nvme_io_type_is_admin(bdev_io->type));
1603 :
1604 13 : switch (rc) {
1605 1 : case 0:
1606 1 : io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1607 1 : break;
1608 0 : case -ENOMEM:
1609 0 : io_status = SPDK_BDEV_IO_STATUS_NOMEM;
1610 0 : break;
1611 12 : case -ENXIO:
1612 12 : if (g_opts.bdev_retry_count == -1 || bio->retry_count < g_opts.bdev_retry_count) {
1613 12 : nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
1614 :
1615 12 : bdev_nvme_clear_current_io_path(nbdev_ch);
1616 12 : bio->io_path = NULL;
1617 :
1618 12 : if (any_io_path_may_become_available(nbdev_ch)) {
1619 9 : bdev_nvme_queue_retry_io(nbdev_ch, bio, 1000ULL);
1620 9 : return;
1621 : }
1622 : }
1623 :
1624 : /* fallthrough */
1625 : default:
1626 3 : spdk_accel_sequence_abort(bdev_io->u.bdev.accel_sequence);
1627 3 : bdev_io->u.bdev.accel_sequence = NULL;
1628 3 : io_status = SPDK_BDEV_IO_STATUS_FAILED;
1629 3 : break;
1630 : }
1631 :
1632 4 : bio->retry_count = 0;
1633 4 : bio->submit_tsc = 0;
1634 4 : __bdev_nvme_io_complete(bdev_io, io_status, NULL);
1635 : }
1636 :
1637 : static inline void
1638 4 : bdev_nvme_admin_complete(struct nvme_bdev_io *bio, int rc)
1639 : {
1640 4 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1641 : enum spdk_bdev_io_status io_status;
1642 :
1643 4 : switch (rc) {
1644 1 : case 0:
1645 1 : io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1646 1 : break;
1647 0 : case -ENOMEM:
1648 0 : io_status = SPDK_BDEV_IO_STATUS_NOMEM;
1649 0 : break;
1650 3 : case -ENXIO:
1651 : /* fallthrough */
1652 : default:
1653 3 : io_status = SPDK_BDEV_IO_STATUS_FAILED;
1654 3 : break;
1655 : }
1656 :
1657 4 : __bdev_nvme_io_complete(bdev_io, io_status, NULL);
1658 4 : }
1659 :
1660 : static void
1661 3 : bdev_nvme_clear_io_path_caches_done(struct nvme_ctrlr *nvme_ctrlr,
1662 : void *ctx, int status)
1663 : {
1664 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
1665 :
1666 3 : assert(nvme_ctrlr->io_path_cache_clearing == true);
1667 3 : nvme_ctrlr->io_path_cache_clearing = false;
1668 :
1669 3 : if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
1670 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1671 3 : return;
1672 : }
1673 :
1674 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1675 :
1676 0 : nvme_ctrlr_unregister(nvme_ctrlr);
1677 : }
1678 :
1679 : static void
1680 416 : _bdev_nvme_clear_io_path_cache(struct nvme_qpair *nvme_qpair)
1681 : {
1682 : struct nvme_io_path *io_path;
1683 :
1684 651 : TAILQ_FOREACH(io_path, &nvme_qpair->io_path_list, tailq) {
1685 235 : if (io_path->nbdev_ch == NULL) {
1686 72 : continue;
1687 : }
1688 163 : bdev_nvme_clear_current_io_path(io_path->nbdev_ch);
1689 : }
1690 416 : }
1691 :
1692 : static void
1693 1 : bdev_nvme_clear_io_path_cache(struct nvme_ctrlr_channel_iter *i,
1694 : struct nvme_ctrlr *nvme_ctrlr,
1695 : struct nvme_ctrlr_channel *ctrlr_ch,
1696 : void *ctx)
1697 : {
1698 1 : assert(ctrlr_ch->qpair != NULL);
1699 :
1700 1 : _bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);
1701 :
1702 1 : nvme_ctrlr_for_each_channel_continue(i, 0);
1703 1 : }
1704 :
1705 : static void
1706 3 : bdev_nvme_clear_io_path_caches(struct nvme_ctrlr *nvme_ctrlr)
1707 : {
1708 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
1709 3 : if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
1710 : nvme_ctrlr->io_path_cache_clearing) {
1711 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1712 0 : return;
1713 : }
1714 :
1715 3 : nvme_ctrlr->io_path_cache_clearing = true;
1716 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
1717 :
1718 3 : nvme_ctrlr_for_each_channel(nvme_ctrlr,
1719 : bdev_nvme_clear_io_path_cache,
1720 : NULL,
1721 : bdev_nvme_clear_io_path_caches_done);
1722 : }
1723 :
1724 : static struct nvme_qpair *
1725 121 : nvme_poll_group_get_qpair(struct nvme_poll_group *group, struct spdk_nvme_qpair *qpair)
1726 : {
1727 : struct nvme_qpair *nvme_qpair;
1728 :
1729 138 : TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
1730 138 : if (nvme_qpair->qpair == qpair) {
1731 121 : break;
1732 : }
1733 : }
1734 :
1735 121 : return nvme_qpair;
1736 : }
1737 :
1738 : static void nvme_qpair_delete(struct nvme_qpair *nvme_qpair);
1739 :
1740 : static void
1741 121 : bdev_nvme_disconnected_qpair_cb(struct spdk_nvme_qpair *qpair, void *poll_group_ctx)
1742 : {
1743 121 : struct nvme_poll_group *group = poll_group_ctx;
1744 : struct nvme_qpair *nvme_qpair;
1745 : struct nvme_ctrlr *nvme_ctrlr;
1746 : struct nvme_ctrlr_channel *ctrlr_ch;
1747 : int status;
1748 :
1749 121 : nvme_qpair = nvme_poll_group_get_qpair(group, qpair);
1750 121 : if (nvme_qpair == NULL) {
1751 0 : return;
1752 : }
1753 :
1754 121 : if (nvme_qpair->qpair != NULL) {
1755 121 : spdk_nvme_ctrlr_free_io_qpair(nvme_qpair->qpair);
1756 121 : nvme_qpair->qpair = NULL;
1757 : }
1758 :
1759 121 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
1760 :
1761 121 : nvme_ctrlr = nvme_qpair->ctrlr;
1762 121 : ctrlr_ch = nvme_qpair->ctrlr_ch;
1763 :
1764 121 : if (ctrlr_ch != NULL) {
1765 74 : if (ctrlr_ch->reset_iter != NULL) {
1766 : /* We are in a full reset sequence. */
1767 69 : if (ctrlr_ch->connect_poller != NULL) {
1768 : /* qpair was failed to connect. Abort the reset sequence. */
1769 0 : NVME_CTRLR_INFOLOG(nvme_ctrlr,
1770 : "qpair %p was failed to connect. abort the reset ctrlr sequence.\n",
1771 : qpair);
1772 0 : spdk_poller_unregister(&ctrlr_ch->connect_poller);
1773 0 : status = -1;
1774 : } else {
1775 : /* qpair was completed to disconnect. Just move to the next ctrlr_channel. */
1776 69 : NVME_CTRLR_INFOLOG(nvme_ctrlr,
1777 : "qpair %p was disconnected and freed in a reset ctrlr sequence.\n",
1778 : qpair);
1779 69 : status = 0;
1780 : }
1781 69 : nvme_ctrlr_for_each_channel_continue(ctrlr_ch->reset_iter, status);
1782 69 : ctrlr_ch->reset_iter = NULL;
1783 : } else {
1784 : /* qpair was disconnected unexpectedly. Reset controller for recovery. */
1785 5 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "qpair %p was disconnected and freed. reset controller.\n",
1786 : qpair);
1787 5 : bdev_nvme_failover_ctrlr(nvme_ctrlr);
1788 : }
1789 : } else {
1790 : /* In this case, ctrlr_channel is already deleted. */
1791 47 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "qpair %p was disconnected and freed. delete nvme_qpair.\n",
1792 : qpair);
1793 47 : nvme_qpair_delete(nvme_qpair);
1794 : }
1795 : }
1796 :
1797 : static void
1798 0 : bdev_nvme_check_io_qpairs(struct nvme_poll_group *group)
1799 : {
1800 : struct nvme_qpair *nvme_qpair;
1801 :
1802 0 : TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
1803 0 : if (nvme_qpair->qpair == NULL || nvme_qpair->ctrlr_ch == NULL) {
1804 0 : continue;
1805 : }
1806 :
1807 0 : if (spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
1808 : SPDK_NVME_QPAIR_FAILURE_NONE) {
1809 0 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
1810 : }
1811 : }
1812 0 : }
1813 :
1814 : static int
1815 1209 : bdev_nvme_poll(void *arg)
1816 : {
1817 1209 : struct nvme_poll_group *group = arg;
1818 : int64_t num_completions;
1819 :
1820 1209 : if (group->collect_spin_stat && group->start_ticks == 0) {
1821 0 : group->start_ticks = spdk_get_ticks();
1822 : }
1823 :
1824 1209 : num_completions = spdk_nvme_poll_group_process_completions(group->group, 0,
1825 : bdev_nvme_disconnected_qpair_cb);
1826 1209 : if (group->collect_spin_stat) {
1827 0 : if (num_completions > 0) {
1828 0 : if (group->end_ticks != 0) {
1829 0 : group->spin_ticks += (group->end_ticks - group->start_ticks);
1830 0 : group->end_ticks = 0;
1831 : }
1832 0 : group->start_ticks = 0;
1833 : } else {
1834 0 : group->end_ticks = spdk_get_ticks();
1835 : }
1836 : }
1837 :
1838 1209 : if (spdk_unlikely(num_completions < 0)) {
1839 0 : bdev_nvme_check_io_qpairs(group);
1840 : }
1841 :
1842 1209 : return num_completions > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
1843 : }
1844 :
1845 : static int bdev_nvme_poll_adminq(void *arg);
1846 :
1847 : static void
1848 142 : bdev_nvme_change_adminq_poll_period(struct nvme_ctrlr *nvme_ctrlr, uint64_t new_period_us)
1849 : {
1850 142 : if (spdk_interrupt_mode_is_enabled()) {
1851 0 : return;
1852 : }
1853 :
1854 142 : spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);
1855 :
1856 142 : nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq,
1857 : nvme_ctrlr, new_period_us);
1858 : }
1859 :
1860 : static int
1861 191 : bdev_nvme_poll_adminq(void *arg)
1862 : {
1863 : int32_t rc;
1864 191 : struct nvme_ctrlr *nvme_ctrlr = arg;
1865 : nvme_ctrlr_disconnected_cb disconnected_cb;
1866 :
1867 191 : assert(nvme_ctrlr != NULL);
1868 :
1869 191 : rc = spdk_nvme_ctrlr_process_admin_completions(nvme_ctrlr->ctrlr);
1870 191 : if (rc < 0) {
1871 86 : disconnected_cb = nvme_ctrlr->disconnected_cb;
1872 86 : nvme_ctrlr->disconnected_cb = NULL;
1873 :
1874 86 : if (disconnected_cb != NULL) {
1875 71 : bdev_nvme_change_adminq_poll_period(nvme_ctrlr,
1876 : g_opts.nvme_adminq_poll_period_us);
1877 71 : disconnected_cb(nvme_ctrlr);
1878 : } else {
1879 15 : bdev_nvme_failover_ctrlr(nvme_ctrlr);
1880 : }
1881 105 : } else if (spdk_nvme_ctrlr_get_admin_qp_failure_reason(nvme_ctrlr->ctrlr) !=
1882 : SPDK_NVME_QPAIR_FAILURE_NONE) {
1883 0 : bdev_nvme_clear_io_path_caches(nvme_ctrlr);
1884 : }
1885 :
1886 191 : return rc == 0 ? SPDK_POLLER_IDLE : SPDK_POLLER_BUSY;
1887 : }
1888 :
1889 : static void
1890 39 : nvme_bdev_free(void *io_device)
1891 : {
1892 39 : struct nvme_bdev *nvme_disk = io_device;
1893 :
1894 39 : pthread_mutex_destroy(&nvme_disk->mutex);
1895 39 : free(nvme_disk->disk.name);
1896 39 : free(nvme_disk->err_stat);
1897 39 : free(nvme_disk);
1898 39 : }
1899 :
1900 : static int
1901 38 : bdev_nvme_destruct(void *ctx)
1902 : {
1903 38 : struct nvme_bdev *nvme_disk = ctx;
1904 : struct nvme_ns *nvme_ns, *tmp_nvme_ns;
1905 :
1906 : SPDK_DTRACE_PROBE2(bdev_nvme_destruct, nvme_disk->nbdev_ctrlr->name, nvme_disk->nsid);
1907 :
1908 77 : TAILQ_FOREACH_SAFE(nvme_ns, &nvme_disk->nvme_ns_list, tailq, tmp_nvme_ns) {
1909 39 : pthread_mutex_lock(&nvme_ns->ctrlr->mutex);
1910 :
1911 39 : nvme_ns->bdev = NULL;
1912 :
1913 39 : assert(nvme_ns->id > 0);
1914 :
1915 39 : if (nvme_ctrlr_get_ns(nvme_ns->ctrlr, nvme_ns->id) == NULL) {
1916 0 : pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
1917 :
1918 0 : nvme_ctrlr_put_ref(nvme_ns->ctrlr);
1919 0 : nvme_ns_free(nvme_ns);
1920 : } else {
1921 39 : pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
1922 : }
1923 : }
1924 :
1925 38 : pthread_mutex_lock(&g_bdev_nvme_mutex);
1926 38 : TAILQ_REMOVE(&nvme_disk->nbdev_ctrlr->bdevs, nvme_disk, tailq);
1927 38 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
1928 :
1929 38 : spdk_io_device_unregister(nvme_disk, nvme_bdev_free);
1930 :
1931 38 : return 0;
1932 : }
1933 :
1934 : static int
1935 122 : bdev_nvme_create_qpair(struct nvme_qpair *nvme_qpair)
1936 : {
1937 : struct nvme_ctrlr *nvme_ctrlr;
1938 122 : struct spdk_nvme_io_qpair_opts opts;
1939 : struct spdk_nvme_qpair *qpair;
1940 : int rc;
1941 :
1942 122 : nvme_ctrlr = nvme_qpair->ctrlr;
1943 :
1944 122 : spdk_nvme_ctrlr_get_default_io_qpair_opts(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
1945 122 : opts.create_only = true;
1946 : /* In interrupt mode qpairs must be created in sync mode, else it will never be connected.
1947 : * delay_cmd_submit must be false as in interrupt mode requests cannot be submitted in
1948 : * completion context.
1949 : */
1950 122 : if (!spdk_interrupt_mode_is_enabled()) {
1951 122 : opts.async_mode = true;
1952 122 : opts.delay_cmd_submit = g_opts.delay_cmd_submit;
1953 : }
1954 122 : opts.io_queue_requests = spdk_max(g_opts.io_queue_requests, opts.io_queue_requests);
1955 122 : g_opts.io_queue_requests = opts.io_queue_requests;
1956 :
1957 122 : qpair = spdk_nvme_ctrlr_alloc_io_qpair(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
1958 122 : if (qpair == NULL) {
1959 0 : return -1;
1960 : }
1961 :
1962 : SPDK_DTRACE_PROBE3(bdev_nvme_create_qpair, nvme_ctrlr->nbdev_ctrlr->name,
1963 : spdk_nvme_qpair_get_id(qpair), spdk_thread_get_id(nvme_ctrlr->thread));
1964 :
1965 122 : assert(nvme_qpair->group != NULL);
1966 :
1967 122 : rc = spdk_nvme_poll_group_add(nvme_qpair->group->group, qpair);
1968 122 : if (rc != 0) {
1969 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Unable to begin polling on NVMe Channel.\n");
1970 0 : goto err;
1971 : }
1972 :
1973 122 : rc = spdk_nvme_ctrlr_connect_io_qpair(nvme_ctrlr->ctrlr, qpair);
1974 122 : if (rc != 0) {
1975 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Unable to connect I/O qpair.\n");
1976 0 : goto err;
1977 : }
1978 :
1979 122 : nvme_qpair->qpair = qpair;
1980 :
1981 122 : if (!g_opts.disable_auto_failback) {
1982 85 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
1983 : }
1984 :
1985 122 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Connecting qpair %p:%u started.\n",
1986 : qpair, spdk_nvme_qpair_get_id(qpair));
1987 :
1988 122 : return 0;
1989 :
1990 0 : err:
1991 0 : spdk_nvme_ctrlr_free_io_qpair(qpair);
1992 :
1993 0 : return rc;
1994 : }
1995 :
1996 : static void bdev_nvme_reset_io_continue(void *cb_arg, int rc);
1997 :
1998 : static void
1999 71 : bdev_nvme_complete_pending_resets(struct nvme_ctrlr *nvme_ctrlr, bool success)
2000 : {
2001 71 : int rc = 0;
2002 : struct nvme_bdev_io *bio;
2003 :
2004 71 : if (!success) {
2005 33 : rc = -1;
2006 : }
2007 :
2008 83 : while (!TAILQ_EMPTY(&nvme_ctrlr->pending_resets)) {
2009 12 : bio = TAILQ_FIRST(&nvme_ctrlr->pending_resets);
2010 12 : TAILQ_REMOVE(&nvme_ctrlr->pending_resets, bio, retry_link);
2011 :
2012 12 : bdev_nvme_reset_io_continue(bio, rc);
2013 : }
2014 71 : }
2015 :
2016 : /* This function marks the current trid as failed by storing the current ticks
2017 : * and then sets the next trid to the active trid within a controller if exists.
2018 : *
2019 : * The purpose of the boolean return value is to request the caller to disconnect
2020 : * the current trid now to try connecting the next trid.
2021 : */
2022 : static bool
2023 62 : bdev_nvme_failover_trid(struct nvme_ctrlr *nvme_ctrlr, bool remove, bool start)
2024 : {
2025 : struct nvme_path_id *path_id, *next_path;
2026 : int rc __attribute__((unused));
2027 :
2028 62 : path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
2029 62 : assert(path_id);
2030 62 : assert(path_id == nvme_ctrlr->active_path_id);
2031 62 : next_path = TAILQ_NEXT(path_id, link);
2032 :
2033 : /* Update the last failed time. It means the trid is failed if its last
2034 : * failed time is non-zero.
2035 : */
2036 62 : path_id->last_failed_tsc = spdk_get_ticks();
2037 :
2038 62 : if (next_path == NULL) {
2039 : /* There is no alternate trid within a controller. */
2040 51 : return false;
2041 : }
2042 :
2043 11 : if (!start && nvme_ctrlr->opts.reconnect_delay_sec == 0) {
2044 : /* Connect is not retried in a controller reset sequence. Connecting
2045 : * the next trid will be done by the next bdev_nvme_failover_ctrlr() call.
2046 : */
2047 3 : return false;
2048 : }
2049 :
2050 8 : assert(path_id->trid.trtype != SPDK_NVME_TRANSPORT_PCIE);
2051 :
2052 8 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Start failover from %s:%s to %s:%s\n",
2053 : path_id->trid.traddr, path_id->trid.trsvcid,
2054 : next_path->trid.traddr, next_path->trid.trsvcid);
2055 :
2056 8 : spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
2057 8 : nvme_ctrlr->active_path_id = next_path;
2058 8 : rc = spdk_nvme_ctrlr_set_trid(nvme_ctrlr->ctrlr, &next_path->trid);
2059 8 : assert(rc == 0);
2060 8 : TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
2061 8 : if (!remove) {
2062 : /** Shuffle the old trid to the end of the list and use the new one.
2063 : * Allows for round robin through multiple connections.
2064 : */
2065 6 : TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, path_id, link);
2066 : } else {
2067 2 : free(path_id);
2068 : }
2069 :
2070 8 : if (start || next_path->last_failed_tsc == 0) {
2071 : /* bdev_nvme_failover_ctrlr() is just called or the next trid is not failed
2072 : * or used yet. Try the next trid now.
2073 : */
2074 7 : return true;
2075 : }
2076 :
2077 1 : if (spdk_get_ticks() > next_path->last_failed_tsc + spdk_get_ticks_hz() *
2078 1 : nvme_ctrlr->opts.reconnect_delay_sec) {
2079 : /* Enough backoff passed since the next trid failed. Try the next trid now. */
2080 0 : return true;
2081 : }
2082 :
2083 : /* The next trid will be tried after reconnect_delay_sec seconds. */
2084 1 : return false;
2085 : }
2086 :
2087 : static bool
2088 89 : bdev_nvme_check_ctrlr_loss_timeout(struct nvme_ctrlr *nvme_ctrlr)
2089 : {
2090 : int32_t elapsed;
2091 :
2092 89 : if (nvme_ctrlr->opts.ctrlr_loss_timeout_sec == 0 ||
2093 37 : nvme_ctrlr->opts.ctrlr_loss_timeout_sec == -1) {
2094 63 : return false;
2095 : }
2096 :
2097 26 : elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
2098 26 : if (elapsed >= nvme_ctrlr->opts.ctrlr_loss_timeout_sec) {
2099 6 : return true;
2100 : } else {
2101 20 : return false;
2102 : }
2103 : }
2104 :
2105 : static bool
2106 12 : bdev_nvme_check_fast_io_fail_timeout(struct nvme_ctrlr *nvme_ctrlr)
2107 : {
2108 : uint32_t elapsed;
2109 :
2110 12 : if (nvme_ctrlr->opts.fast_io_fail_timeout_sec == 0) {
2111 8 : return false;
2112 : }
2113 :
2114 4 : elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
2115 4 : if (elapsed >= nvme_ctrlr->opts.fast_io_fail_timeout_sec) {
2116 2 : return true;
2117 : } else {
2118 2 : return false;
2119 : }
2120 : }
2121 :
2122 : static void bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success);
2123 :
2124 : static void
2125 72 : nvme_ctrlr_disconnect(struct nvme_ctrlr *nvme_ctrlr, nvme_ctrlr_disconnected_cb cb_fn)
2126 : {
2127 : int rc;
2128 :
2129 72 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Start disconnecting ctrlr.\n");
2130 :
2131 72 : rc = spdk_nvme_ctrlr_disconnect(nvme_ctrlr->ctrlr);
2132 72 : if (rc != 0) {
2133 1 : NVME_CTRLR_WARNLOG(nvme_ctrlr, "disconnecting ctrlr failed.\n");
2134 :
2135 : /* Disconnect fails if ctrlr is already resetting or removed. In this case,
2136 : * fail the reset sequence immediately.
2137 : */
2138 1 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2139 1 : return;
2140 : }
2141 :
2142 : /* spdk_nvme_ctrlr_disconnect() may complete asynchronously later by polling adminq.
2143 : * Set callback here to execute the specified operation after ctrlr is really disconnected.
2144 : */
2145 71 : assert(nvme_ctrlr->disconnected_cb == NULL);
2146 71 : nvme_ctrlr->disconnected_cb = cb_fn;
2147 :
2148 : /* During disconnection, reduce the period to poll adminq more often. */
2149 71 : bdev_nvme_change_adminq_poll_period(nvme_ctrlr, 0);
2150 : }
2151 :
2152 : enum bdev_nvme_op_after_reset {
2153 : OP_NONE,
2154 : OP_COMPLETE_PENDING_DESTRUCT,
2155 : OP_DESTRUCT,
2156 : OP_DELAYED_RECONNECT,
2157 : OP_FAILOVER,
2158 : };
2159 :
2160 : typedef enum bdev_nvme_op_after_reset _bdev_nvme_op_after_reset;
2161 :
2162 : static _bdev_nvme_op_after_reset
2163 71 : bdev_nvme_check_op_after_reset(struct nvme_ctrlr *nvme_ctrlr, bool success)
2164 : {
2165 71 : if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
2166 : /* Complete pending destruct after reset completes. */
2167 0 : return OP_COMPLETE_PENDING_DESTRUCT;
2168 71 : } else if (nvme_ctrlr->pending_failover) {
2169 3 : nvme_ctrlr->pending_failover = false;
2170 3 : nvme_ctrlr->reset_start_tsc = 0;
2171 3 : return OP_FAILOVER;
2172 68 : } else if (success || nvme_ctrlr->opts.reconnect_delay_sec == 0) {
2173 54 : nvme_ctrlr->reset_start_tsc = 0;
2174 54 : return OP_NONE;
2175 14 : } else if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
2176 2 : return OP_DESTRUCT;
2177 : } else {
2178 12 : if (bdev_nvme_check_fast_io_fail_timeout(nvme_ctrlr)) {
2179 2 : nvme_ctrlr->fast_io_fail_timedout = true;
2180 : }
2181 12 : return OP_DELAYED_RECONNECT;
2182 : }
2183 : }
2184 :
2185 : static int bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug);
2186 : static void bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
2187 :
2188 : static int
2189 9 : bdev_nvme_reconnect_delay_timer_expired(void *ctx)
2190 : {
2191 9 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2192 :
2193 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect_delay, nvme_ctrlr->nbdev_ctrlr->name);
2194 9 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2195 :
2196 9 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2197 :
2198 9 : if (!nvme_ctrlr->reconnect_is_delayed) {
2199 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2200 0 : return SPDK_POLLER_BUSY;
2201 : }
2202 :
2203 9 : nvme_ctrlr->reconnect_is_delayed = false;
2204 :
2205 9 : if (nvme_ctrlr->destruct) {
2206 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2207 0 : return SPDK_POLLER_BUSY;
2208 : }
2209 :
2210 9 : assert(nvme_ctrlr->resetting == false);
2211 9 : nvme_ctrlr->resetting = true;
2212 :
2213 9 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2214 :
2215 9 : spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);
2216 :
2217 9 : bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2218 9 : return SPDK_POLLER_BUSY;
2219 : }
2220 :
2221 : static void
2222 12 : bdev_nvme_start_reconnect_delay_timer(struct nvme_ctrlr *nvme_ctrlr)
2223 : {
2224 12 : spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);
2225 :
2226 12 : assert(nvme_ctrlr->reconnect_is_delayed == false);
2227 12 : nvme_ctrlr->reconnect_is_delayed = true;
2228 :
2229 12 : assert(nvme_ctrlr->reconnect_delay_timer == NULL);
2230 12 : nvme_ctrlr->reconnect_delay_timer = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_delay_timer_expired,
2231 : nvme_ctrlr,
2232 : nvme_ctrlr->opts.reconnect_delay_sec * SPDK_SEC_TO_USEC);
2233 12 : }
2234 :
2235 : static void remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr);
2236 :
2237 : static void
2238 71 : bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success)
2239 : {
2240 71 : bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
2241 71 : void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
2242 : enum bdev_nvme_op_after_reset op_after_reset;
2243 :
2244 71 : assert(nvme_ctrlr->thread == spdk_get_thread());
2245 :
2246 71 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2247 71 : if (!success) {
2248 : /* Connecting the active trid failed. Set the next alternate trid to the
2249 : * active trid if it exists.
2250 : */
2251 35 : if (bdev_nvme_failover_trid(nvme_ctrlr, false, false)) {
2252 : /* The next alternate trid exists and is ready to try. Try it now. */
2253 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2254 :
2255 2 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Try the next alternate trid %s:%s now.\n",
2256 : nvme_ctrlr->active_path_id->trid.traddr,
2257 : nvme_ctrlr->active_path_id->trid.trsvcid);
2258 :
2259 2 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
2260 2 : return;
2261 : }
2262 :
2263 : /* We came here if there is no alternate trid or if the next trid exists but
2264 : * is not ready to try. We will try the active trid after reconnect_delay_sec
2265 : * seconds if it is non-zero or at the next reset call otherwise.
2266 : */
2267 : } else {
2268 : /* Connecting the active trid succeeded. Clear the last failed time because it
2269 : * means the trid is failed if its last failed time is non-zero.
2270 : */
2271 36 : nvme_ctrlr->active_path_id->last_failed_tsc = 0;
2272 : }
2273 :
2274 69 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Clear pending resets.\n");
2275 :
2276 : /* Make sure we clear any pending resets before returning. */
2277 69 : bdev_nvme_complete_pending_resets(nvme_ctrlr, success);
2278 :
2279 69 : if (!success) {
2280 33 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Resetting controller failed.\n");
2281 : } else {
2282 36 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Resetting controller successful.\n");
2283 : }
2284 :
2285 69 : nvme_ctrlr->resetting = false;
2286 69 : nvme_ctrlr->dont_retry = false;
2287 69 : nvme_ctrlr->in_failover = false;
2288 :
2289 69 : nvme_ctrlr->ctrlr_op_cb_fn = NULL;
2290 69 : nvme_ctrlr->ctrlr_op_cb_arg = NULL;
2291 :
2292 69 : op_after_reset = bdev_nvme_check_op_after_reset(nvme_ctrlr, success);
2293 69 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2294 :
2295 : /* Delay callbacks when the next operation is a failover. */
2296 69 : if (ctrlr_op_cb_fn && op_after_reset != OP_FAILOVER) {
2297 17 : ctrlr_op_cb_fn(ctrlr_op_cb_arg, success ? 0 : -1);
2298 : }
2299 :
2300 69 : switch (op_after_reset) {
2301 0 : case OP_COMPLETE_PENDING_DESTRUCT:
2302 0 : nvme_ctrlr_unregister(nvme_ctrlr);
2303 0 : break;
2304 2 : case OP_DESTRUCT:
2305 2 : bdev_nvme_delete_ctrlr(nvme_ctrlr, false);
2306 2 : remove_discovery_entry(nvme_ctrlr);
2307 2 : break;
2308 12 : case OP_DELAYED_RECONNECT:
2309 12 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_start_reconnect_delay_timer);
2310 12 : break;
2311 3 : case OP_FAILOVER:
2312 3 : nvme_ctrlr->ctrlr_op_cb_fn = ctrlr_op_cb_fn;
2313 3 : nvme_ctrlr->ctrlr_op_cb_arg = ctrlr_op_cb_arg;
2314 3 : bdev_nvme_failover_ctrlr(nvme_ctrlr);
2315 3 : break;
2316 52 : default:
2317 52 : break;
2318 : }
2319 : }
2320 :
2321 : static void
2322 0 : bdev_nvme_reset_create_qpairs_failed(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2323 : {
2324 0 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2325 0 : }
2326 :
2327 : static void
2328 104 : bdev_nvme_reset_destroy_qpair(struct nvme_ctrlr_channel_iter *i,
2329 : struct nvme_ctrlr *nvme_ctrlr,
2330 : struct nvme_ctrlr_channel *ctrlr_ch, void *ctx)
2331 : {
2332 : struct nvme_qpair *nvme_qpair;
2333 : struct spdk_nvme_qpair *qpair;
2334 :
2335 104 : nvme_qpair = ctrlr_ch->qpair;
2336 104 : assert(nvme_qpair != NULL);
2337 :
2338 104 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
2339 :
2340 104 : qpair = nvme_qpair->qpair;
2341 104 : if (qpair != NULL) {
2342 69 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Start disconnecting qpair %p:%u.\n",
2343 : qpair, spdk_nvme_qpair_get_id(qpair));
2344 :
2345 69 : if (nvme_qpair->ctrlr->dont_retry) {
2346 53 : spdk_nvme_qpair_set_abort_dnr(qpair, true);
2347 : }
2348 69 : spdk_nvme_ctrlr_disconnect_io_qpair(qpair);
2349 :
2350 : /* The current full reset sequence will move to the next
2351 : * ctrlr_channel after the qpair is actually disconnected.
2352 : */
2353 69 : assert(ctrlr_ch->reset_iter == NULL);
2354 69 : ctrlr_ch->reset_iter = i;
2355 : } else {
2356 35 : nvme_ctrlr_for_each_channel_continue(i, 0);
2357 : }
2358 104 : }
2359 :
2360 : static void
2361 36 : bdev_nvme_reset_create_qpairs_done(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2362 : {
2363 36 : if (status == 0) {
2364 36 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "qpairs were created after ctrlr reset.\n");
2365 :
2366 36 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, true);
2367 : } else {
2368 0 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "qpairs were failed to create after ctrlr reset.\n");
2369 :
2370 : /* Delete the added qpairs and quiesce ctrlr to make the states clean. */
2371 0 : nvme_ctrlr_for_each_channel(nvme_ctrlr,
2372 : bdev_nvme_reset_destroy_qpair,
2373 : NULL,
2374 : bdev_nvme_reset_create_qpairs_failed);
2375 : }
2376 36 : }
2377 :
2378 : static int
2379 61 : bdev_nvme_reset_check_qpair_connected(void *ctx)
2380 : {
2381 61 : struct nvme_ctrlr_channel *ctrlr_ch = ctx;
2382 61 : struct nvme_qpair *nvme_qpair = ctrlr_ch->qpair;
2383 : struct spdk_nvme_qpair *qpair;
2384 :
2385 61 : if (ctrlr_ch->reset_iter == NULL) {
2386 : /* qpair was already failed to connect and the reset sequence is being aborted. */
2387 0 : assert(ctrlr_ch->connect_poller == NULL);
2388 0 : assert(nvme_qpair->qpair == NULL);
2389 :
2390 0 : NVME_CTRLR_INFOLOG(nvme_qpair->ctrlr,
2391 : "qpair was already failed to connect. reset is being aborted.\n");
2392 0 : return SPDK_POLLER_BUSY;
2393 : }
2394 :
2395 61 : qpair = nvme_qpair->qpair;
2396 61 : assert(qpair != NULL);
2397 :
2398 61 : if (!spdk_nvme_qpair_is_connected(qpair)) {
2399 0 : return SPDK_POLLER_BUSY;
2400 : }
2401 :
2402 61 : NVME_CTRLR_INFOLOG(nvme_qpair->ctrlr, "qpair %p:%u was connected.\n",
2403 : qpair, spdk_nvme_qpair_get_id(qpair));
2404 :
2405 61 : spdk_poller_unregister(&ctrlr_ch->connect_poller);
2406 :
2407 : /* qpair was completed to connect. Move to the next ctrlr_channel */
2408 61 : nvme_ctrlr_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
2409 61 : ctrlr_ch->reset_iter = NULL;
2410 :
2411 61 : if (!g_opts.disable_auto_failback) {
2412 44 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
2413 : }
2414 :
2415 61 : return SPDK_POLLER_BUSY;
2416 : }
2417 :
2418 : static void
2419 61 : bdev_nvme_reset_create_qpair(struct nvme_ctrlr_channel_iter *i,
2420 : struct nvme_ctrlr *nvme_ctrlr,
2421 : struct nvme_ctrlr_channel *ctrlr_ch,
2422 : void *ctx)
2423 : {
2424 61 : struct nvme_qpair *nvme_qpair = ctrlr_ch->qpair;
2425 : struct spdk_nvme_qpair *qpair;
2426 : int rc;
2427 :
2428 61 : rc = bdev_nvme_create_qpair(nvme_qpair);
2429 61 : if (rc == 0) {
2430 61 : ctrlr_ch->connect_poller = SPDK_POLLER_REGISTER(bdev_nvme_reset_check_qpair_connected,
2431 : ctrlr_ch, 0);
2432 :
2433 61 : qpair = nvme_qpair->qpair;
2434 :
2435 61 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Start checking qpair %p:%u to be connected.\n",
2436 : qpair, spdk_nvme_qpair_get_id(qpair));
2437 :
2438 : /* The current full reset sequence will move to the next
2439 : * ctrlr_channel after the qpair is actually connected.
2440 : */
2441 61 : assert(ctrlr_ch->reset_iter == NULL);
2442 61 : ctrlr_ch->reset_iter = i;
2443 : } else {
2444 0 : nvme_ctrlr_for_each_channel_continue(i, rc);
2445 : }
2446 61 : }
2447 :
2448 : static void
2449 36 : nvme_ctrlr_check_namespaces(struct nvme_ctrlr *nvme_ctrlr)
2450 : {
2451 36 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
2452 : struct nvme_ns *nvme_ns;
2453 :
2454 36 : for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
2455 57 : nvme_ns != NULL;
2456 21 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
2457 21 : if (!spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
2458 1 : SPDK_DEBUGLOG(bdev_nvme, "NSID %u was removed during reset.\n", nvme_ns->id);
2459 : /* NS can be added again. Just nullify nvme_ns->ns. */
2460 1 : nvme_ns->ns = NULL;
2461 : }
2462 : }
2463 36 : }
2464 :
2465 :
2466 : static int
2467 70 : bdev_nvme_reconnect_ctrlr_poll(void *arg)
2468 : {
2469 70 : struct nvme_ctrlr *nvme_ctrlr = arg;
2470 : struct spdk_nvme_transport_id *trid;
2471 70 : int rc = -ETIMEDOUT;
2472 :
2473 70 : if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
2474 : /* Mark the ctrlr as failed. The next call to
2475 : * spdk_nvme_ctrlr_reconnect_poll_async() will then
2476 : * do the necessary cleanup and return failure.
2477 : */
2478 2 : spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
2479 : }
2480 :
2481 70 : rc = spdk_nvme_ctrlr_reconnect_poll_async(nvme_ctrlr->ctrlr);
2482 70 : if (rc == -EAGAIN) {
2483 0 : return SPDK_POLLER_BUSY;
2484 : }
2485 :
2486 70 : spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
2487 70 : if (rc == 0) {
2488 36 : trid = &nvme_ctrlr->active_path_id->trid;
2489 :
2490 36 : if (spdk_nvme_trtype_is_fabrics(trid->trtype)) {
2491 36 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "ctrlr was connected to %s:%s. Create qpairs.\n",
2492 : trid->traddr, trid->trsvcid);
2493 : } else {
2494 0 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "ctrlr was connected. Create qpairs.\n");
2495 : }
2496 :
2497 36 : nvme_ctrlr_check_namespaces(nvme_ctrlr);
2498 :
2499 : /* Recreate all of the I/O queue pairs */
2500 36 : nvme_ctrlr_for_each_channel(nvme_ctrlr,
2501 : bdev_nvme_reset_create_qpair,
2502 : NULL,
2503 : bdev_nvme_reset_create_qpairs_done);
2504 : } else {
2505 34 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "ctrlr could not be connected.\n");
2506 :
2507 34 : bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2508 : }
2509 70 : return SPDK_POLLER_BUSY;
2510 : }
2511 :
2512 : static void
2513 70 : bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2514 : {
2515 70 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Start reconnecting ctrlr.\n");
2516 :
2517 70 : spdk_nvme_ctrlr_reconnect_async(nvme_ctrlr->ctrlr);
2518 :
2519 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect, nvme_ctrlr->nbdev_ctrlr->name);
2520 70 : assert(nvme_ctrlr->reset_detach_poller == NULL);
2521 70 : nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_ctrlr_poll,
2522 : nvme_ctrlr, 0);
2523 70 : }
2524 :
2525 : static void
2526 57 : bdev_nvme_reset_destroy_qpair_done(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2527 : {
2528 : SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reset, nvme_ctrlr->nbdev_ctrlr->name);
2529 57 : assert(status == 0);
2530 :
2531 57 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "qpairs were deleted.\n");
2532 :
2533 57 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2534 0 : bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2535 : } else {
2536 57 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
2537 : }
2538 57 : }
2539 :
2540 : static void
2541 57 : bdev_nvme_reset_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
2542 : {
2543 57 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Delete qpairs for reset.\n");
2544 :
2545 57 : nvme_ctrlr_for_each_channel(nvme_ctrlr,
2546 : bdev_nvme_reset_destroy_qpair,
2547 : NULL,
2548 : bdev_nvme_reset_destroy_qpair_done);
2549 57 : }
2550 :
2551 : static void
2552 3 : bdev_nvme_reconnect_ctrlr_now(void *ctx)
2553 : {
2554 3 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2555 :
2556 3 : assert(nvme_ctrlr->resetting == true);
2557 3 : assert(nvme_ctrlr->thread == spdk_get_thread());
2558 :
2559 3 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2560 :
2561 3 : spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);
2562 :
2563 3 : bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2564 3 : }
2565 :
2566 : static void
2567 57 : _bdev_nvme_reset_ctrlr(void *ctx)
2568 : {
2569 57 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2570 :
2571 57 : assert(nvme_ctrlr->resetting == true);
2572 57 : assert(nvme_ctrlr->thread == spdk_get_thread());
2573 :
2574 57 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2575 0 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reset_destroy_qpairs);
2576 : } else {
2577 57 : bdev_nvme_reset_destroy_qpairs(nvme_ctrlr);
2578 : }
2579 57 : }
2580 :
2581 : static int
2582 50 : bdev_nvme_reset_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, spdk_msg_fn *msg_fn)
2583 : {
2584 50 : if (nvme_ctrlr->destruct) {
2585 3 : return -ENXIO;
2586 : }
2587 :
2588 47 : if (nvme_ctrlr->resetting) {
2589 14 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Unable to perform reset, already in progress.\n");
2590 14 : return -EBUSY;
2591 : }
2592 :
2593 33 : if (nvme_ctrlr->disabled) {
2594 1 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Unable to perform reset. Controller is disabled.\n");
2595 1 : return -EALREADY;
2596 : }
2597 :
2598 32 : nvme_ctrlr->resetting = true;
2599 32 : nvme_ctrlr->dont_retry = true;
2600 :
2601 32 : if (nvme_ctrlr->reconnect_is_delayed) {
2602 1 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "Reconnect is already scheduled.\n");
2603 1 : *msg_fn = bdev_nvme_reconnect_ctrlr_now;
2604 1 : nvme_ctrlr->reconnect_is_delayed = false;
2605 : } else {
2606 31 : *msg_fn = _bdev_nvme_reset_ctrlr;
2607 31 : assert(nvme_ctrlr->reset_start_tsc == 0);
2608 : }
2609 :
2610 32 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2611 :
2612 32 : return 0;
2613 : }
2614 :
2615 : static int
2616 24 : bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2617 : {
2618 24 : spdk_msg_fn msg_fn;
2619 : int rc;
2620 :
2621 24 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2622 24 : rc = bdev_nvme_reset_ctrlr_unsafe(nvme_ctrlr, &msg_fn);
2623 24 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2624 :
2625 24 : if (rc == 0) {
2626 19 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
2627 : }
2628 :
2629 24 : return rc;
2630 : }
2631 :
2632 : static int
2633 3 : bdev_nvme_enable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2634 : {
2635 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2636 3 : if (nvme_ctrlr->destruct) {
2637 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2638 0 : return -ENXIO;
2639 : }
2640 :
2641 3 : if (nvme_ctrlr->resetting) {
2642 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2643 0 : return -EBUSY;
2644 : }
2645 :
2646 3 : if (!nvme_ctrlr->disabled) {
2647 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2648 1 : return -EALREADY;
2649 : }
2650 :
2651 2 : nvme_ctrlr->disabled = false;
2652 2 : nvme_ctrlr->resetting = true;
2653 :
2654 2 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2655 :
2656 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2657 :
2658 2 : spdk_thread_send_msg(nvme_ctrlr->thread, bdev_nvme_reconnect_ctrlr_now, nvme_ctrlr);
2659 2 : return 0;
2660 : }
2661 :
2662 : static void
2663 2 : bdev_nvme_disable_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr)
2664 : {
2665 2 : bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
2666 2 : void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
2667 : enum bdev_nvme_op_after_reset op_after_disable;
2668 :
2669 2 : assert(nvme_ctrlr->thread == spdk_get_thread());
2670 :
2671 2 : nvme_ctrlr->ctrlr_op_cb_fn = NULL;
2672 2 : nvme_ctrlr->ctrlr_op_cb_arg = NULL;
2673 :
2674 2 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2675 :
2676 2 : nvme_ctrlr->resetting = false;
2677 2 : nvme_ctrlr->dont_retry = false;
2678 :
2679 2 : op_after_disable = bdev_nvme_check_op_after_reset(nvme_ctrlr, true);
2680 :
2681 2 : nvme_ctrlr->disabled = true;
2682 2 : spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);
2683 :
2684 : /* Make sure we clear any pending resets before returning. */
2685 2 : bdev_nvme_complete_pending_resets(nvme_ctrlr, true);
2686 :
2687 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2688 :
2689 2 : if (ctrlr_op_cb_fn) {
2690 0 : ctrlr_op_cb_fn(ctrlr_op_cb_arg, 0);
2691 : }
2692 :
2693 2 : switch (op_after_disable) {
2694 0 : case OP_COMPLETE_PENDING_DESTRUCT:
2695 0 : nvme_ctrlr_unregister(nvme_ctrlr);
2696 0 : break;
2697 2 : default:
2698 2 : break;
2699 : }
2700 2 : }
2701 :
2702 : static void
2703 1 : bdev_nvme_disable_destroy_qpairs_done(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2704 : {
2705 1 : assert(status == 0);
2706 :
2707 1 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2708 0 : bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
2709 : } else {
2710 1 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_ctrlr_complete);
2711 : }
2712 1 : }
2713 :
2714 : static void
2715 1 : bdev_nvme_disable_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
2716 : {
2717 1 : nvme_ctrlr_for_each_channel(nvme_ctrlr,
2718 : bdev_nvme_reset_destroy_qpair,
2719 : NULL,
2720 : bdev_nvme_disable_destroy_qpairs_done);
2721 1 : }
2722 :
2723 : static void
2724 1 : _bdev_nvme_cancel_reconnect_and_disable_ctrlr(void *ctx)
2725 : {
2726 1 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2727 :
2728 1 : assert(nvme_ctrlr->resetting == true);
2729 1 : assert(nvme_ctrlr->thread == spdk_get_thread());
2730 :
2731 1 : spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2732 :
2733 1 : bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
2734 1 : }
2735 :
2736 : static void
2737 1 : _bdev_nvme_disconnect_and_disable_ctrlr(void *ctx)
2738 : {
2739 1 : struct nvme_ctrlr *nvme_ctrlr = ctx;
2740 :
2741 1 : assert(nvme_ctrlr->resetting == true);
2742 1 : assert(nvme_ctrlr->thread == spdk_get_thread());
2743 :
2744 1 : if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2745 0 : nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_destroy_qpairs);
2746 : } else {
2747 1 : bdev_nvme_disable_destroy_qpairs(nvme_ctrlr);
2748 : }
2749 1 : }
2750 :
2751 : static int
2752 5 : bdev_nvme_disable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2753 : {
2754 : spdk_msg_fn msg_fn;
2755 :
2756 5 : pthread_mutex_lock(&nvme_ctrlr->mutex);
2757 5 : if (nvme_ctrlr->destruct) {
2758 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2759 1 : return -ENXIO;
2760 : }
2761 :
2762 4 : if (nvme_ctrlr->resetting) {
2763 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2764 1 : return -EBUSY;
2765 : }
2766 :
2767 3 : if (nvme_ctrlr->disabled) {
2768 1 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2769 1 : return -EALREADY;
2770 : }
2771 :
2772 2 : nvme_ctrlr->resetting = true;
2773 2 : nvme_ctrlr->dont_retry = true;
2774 :
2775 2 : if (nvme_ctrlr->reconnect_is_delayed) {
2776 1 : msg_fn = _bdev_nvme_cancel_reconnect_and_disable_ctrlr;
2777 1 : nvme_ctrlr->reconnect_is_delayed = false;
2778 : } else {
2779 1 : msg_fn = _bdev_nvme_disconnect_and_disable_ctrlr;
2780 : }
2781 :
2782 2 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2783 :
2784 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
2785 :
2786 2 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
2787 2 : return 0;
2788 : }
2789 :
2790 : static int
2791 6 : nvme_ctrlr_op(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
2792 : bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2793 : {
2794 : int rc;
2795 :
2796 6 : switch (op) {
2797 5 : case NVME_CTRLR_OP_RESET:
2798 5 : rc = bdev_nvme_reset_ctrlr(nvme_ctrlr);
2799 5 : break;
2800 0 : case NVME_CTRLR_OP_ENABLE:
2801 0 : rc = bdev_nvme_enable_ctrlr(nvme_ctrlr);
2802 0 : break;
2803 0 : case NVME_CTRLR_OP_DISABLE:
2804 0 : rc = bdev_nvme_disable_ctrlr(nvme_ctrlr);
2805 0 : break;
2806 1 : default:
2807 1 : rc = -EINVAL;
2808 1 : break;
2809 : }
2810 :
2811 6 : if (rc == 0) {
2812 3 : assert(nvme_ctrlr->ctrlr_op_cb_fn == NULL);
2813 3 : assert(nvme_ctrlr->ctrlr_op_cb_arg == NULL);
2814 3 : nvme_ctrlr->ctrlr_op_cb_fn = cb_fn;
2815 3 : nvme_ctrlr->ctrlr_op_cb_arg = cb_arg;
2816 : }
2817 6 : return rc;
2818 : }
2819 :
2820 : struct nvme_ctrlr_op_rpc_ctx {
2821 : struct nvme_ctrlr *nvme_ctrlr;
2822 : struct spdk_thread *orig_thread;
2823 : enum nvme_ctrlr_op op;
2824 : int rc;
2825 : bdev_nvme_ctrlr_op_cb cb_fn;
2826 : void *cb_arg;
2827 : };
2828 :
2829 : static void
2830 4 : _nvme_ctrlr_op_rpc_complete(void *_ctx)
2831 : {
2832 4 : struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
2833 :
2834 4 : assert(ctx != NULL);
2835 4 : assert(ctx->cb_fn != NULL);
2836 :
2837 4 : ctx->cb_fn(ctx->cb_arg, ctx->rc);
2838 :
2839 4 : free(ctx);
2840 4 : }
2841 :
2842 : static void
2843 4 : nvme_ctrlr_op_rpc_complete(void *cb_arg, int rc)
2844 : {
2845 4 : struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;
2846 :
2847 4 : ctx->rc = rc;
2848 :
2849 4 : spdk_thread_send_msg(ctx->orig_thread, _nvme_ctrlr_op_rpc_complete, ctx);
2850 4 : }
2851 :
2852 : void
2853 4 : nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
2854 : bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2855 : {
2856 : struct nvme_ctrlr_op_rpc_ctx *ctx;
2857 : int rc;
2858 :
2859 4 : assert(cb_fn != NULL);
2860 :
2861 4 : ctx = calloc(1, sizeof(*ctx));
2862 4 : if (ctx == NULL) {
2863 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
2864 0 : cb_fn(cb_arg, -ENOMEM);
2865 0 : return;
2866 : }
2867 :
2868 4 : ctx->orig_thread = spdk_get_thread();
2869 4 : ctx->cb_fn = cb_fn;
2870 4 : ctx->cb_arg = cb_arg;
2871 :
2872 4 : rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_ctrlr_op_rpc_complete, ctx);
2873 4 : if (rc == 0) {
2874 1 : return;
2875 3 : } else if (rc == -EALREADY) {
2876 0 : rc = 0;
2877 : }
2878 :
2879 3 : nvme_ctrlr_op_rpc_complete(ctx, rc);
2880 : }
2881 :
2882 : static void nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc);
2883 :
2884 : static void
2885 2 : _nvme_bdev_ctrlr_op_rpc_continue(void *_ctx)
2886 : {
2887 2 : struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
2888 : struct nvme_ctrlr *prev_nvme_ctrlr, *next_nvme_ctrlr;
2889 : int rc;
2890 :
2891 2 : prev_nvme_ctrlr = ctx->nvme_ctrlr;
2892 2 : ctx->nvme_ctrlr = NULL;
2893 :
2894 2 : if (ctx->rc != 0) {
2895 0 : goto complete;
2896 : }
2897 :
2898 2 : next_nvme_ctrlr = TAILQ_NEXT(prev_nvme_ctrlr, tailq);
2899 2 : if (next_nvme_ctrlr == NULL) {
2900 1 : goto complete;
2901 : }
2902 :
2903 1 : rc = nvme_ctrlr_op(next_nvme_ctrlr, ctx->op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
2904 1 : if (rc == 0) {
2905 1 : ctx->nvme_ctrlr = next_nvme_ctrlr;
2906 1 : return;
2907 0 : } else if (rc == -EALREADY) {
2908 0 : ctx->nvme_ctrlr = next_nvme_ctrlr;
2909 0 : rc = 0;
2910 : }
2911 :
2912 0 : ctx->rc = rc;
2913 :
2914 1 : complete:
2915 1 : ctx->cb_fn(ctx->cb_arg, ctx->rc);
2916 1 : free(ctx);
2917 : }
2918 :
2919 : static void
2920 2 : nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc)
2921 : {
2922 2 : struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;
2923 :
2924 2 : ctx->rc = rc;
2925 :
2926 2 : spdk_thread_send_msg(ctx->orig_thread, _nvme_bdev_ctrlr_op_rpc_continue, ctx);
2927 2 : }
2928 :
2929 : void
2930 1 : nvme_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op,
2931 : bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2932 : {
2933 : struct nvme_ctrlr_op_rpc_ctx *ctx;
2934 : struct nvme_ctrlr *nvme_ctrlr;
2935 : int rc;
2936 :
2937 1 : assert(cb_fn != NULL);
2938 :
2939 1 : ctx = calloc(1, sizeof(*ctx));
2940 1 : if (ctx == NULL) {
2941 0 : SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
2942 0 : cb_fn(cb_arg, -ENOMEM);
2943 0 : return;
2944 : }
2945 :
2946 1 : ctx->orig_thread = spdk_get_thread();
2947 1 : ctx->op = op;
2948 1 : ctx->cb_fn = cb_fn;
2949 1 : ctx->cb_arg = cb_arg;
2950 :
2951 1 : nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
2952 1 : assert(nvme_ctrlr != NULL);
2953 :
2954 1 : rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
2955 1 : if (rc == 0) {
2956 1 : ctx->nvme_ctrlr = nvme_ctrlr;
2957 1 : return;
2958 0 : } else if (rc == -EALREADY) {
2959 0 : ctx->nvme_ctrlr = nvme_ctrlr;
2960 0 : rc = 0;
2961 : }
2962 :
2963 0 : nvme_bdev_ctrlr_op_rpc_continue(ctx, rc);
2964 : }
2965 :
2966 : static int _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio);
2967 :
2968 : static void
2969 16 : bdev_nvme_unfreeze_bdev_channel_done(struct nvme_bdev *nbdev, void *ctx, int status)
2970 : {
2971 16 : struct nvme_bdev_io *bio = ctx;
2972 : enum spdk_bdev_io_status io_status;
2973 :
2974 16 : if (bio->cpl.cdw0 == 0) {
2975 12 : io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2976 : } else {
2977 4 : io_status = SPDK_BDEV_IO_STATUS_FAILED;
2978 : }
2979 :
2980 16 : NVME_BDEV_INFOLOG(nbdev, "reset_io %p completed, status:%d\n", bio, io_status);
2981 :
2982 16 : __bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), io_status, NULL);
2983 16 : }
2984 :
2985 : static void
2986 32 : bdev_nvme_unfreeze_bdev_channel(struct nvme_bdev_channel_iter *i,
2987 : struct nvme_bdev *nbdev,
2988 : struct nvme_bdev_channel *nbdev_ch, void *ctx)
2989 : {
2990 32 : bdev_nvme_abort_retry_ios(nbdev_ch);
2991 32 : nbdev_ch->resetting = false;
2992 :
2993 32 : nvme_bdev_for_each_channel_continue(i, 0);
2994 32 : }
2995 :
2996 : static void
2997 16 : bdev_nvme_reset_io_complete(struct nvme_bdev_io *bio)
2998 : {
2999 16 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
3000 16 : struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
3001 :
3002 : /* Abort all queued I/Os for retry. */
3003 16 : nvme_bdev_for_each_channel(nbdev,
3004 : bdev_nvme_unfreeze_bdev_channel,
3005 : bio,
3006 : bdev_nvme_unfreeze_bdev_channel_done);
3007 16 : }
3008 :
3009 : static void
3010 26 : _bdev_nvme_reset_io_continue(void *ctx)
3011 : {
3012 26 : struct nvme_bdev_io *bio = ctx;
3013 : struct nvme_io_path *prev_io_path, *next_io_path;
3014 : int rc;
3015 :
3016 26 : prev_io_path = bio->io_path;
3017 26 : bio->io_path = NULL;
3018 :
3019 26 : next_io_path = STAILQ_NEXT(prev_io_path, stailq);
3020 26 : if (next_io_path == NULL) {
3021 16 : goto complete;
3022 : }
3023 :
3024 10 : rc = _bdev_nvme_reset_io(next_io_path, bio);
3025 10 : if (rc == 0) {
3026 10 : return;
3027 : }
3028 :
3029 0 : complete:
3030 16 : bdev_nvme_reset_io_complete(bio);
3031 : }
3032 :
3033 : static void
3034 26 : bdev_nvme_reset_io_continue(void *cb_arg, int rc)
3035 : {
3036 26 : struct nvme_bdev_io *bio = cb_arg;
3037 26 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
3038 26 : struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
3039 :
3040 26 : NVME_BDEV_INFOLOG(nbdev, "continue reset_io %p, rc:%d\n", bio, rc);
3041 :
3042 : /* Reset status is initialized as "failed". Set to "success" once we have at least one
3043 : * successfully reset nvme_ctrlr.
3044 : */
3045 26 : if (rc == 0) {
3046 16 : bio->cpl.cdw0 = 0;
3047 : }
3048 :
3049 26 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), _bdev_nvme_reset_io_continue, bio);
3050 26 : }
3051 :
3052 : static int
3053 26 : _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio)
3054 : {
3055 26 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
3056 26 : struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
3057 26 : struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
3058 26 : spdk_msg_fn msg_fn;
3059 : int rc;
3060 :
3061 26 : assert(bio->io_path == NULL);
3062 26 : bio->io_path = io_path;
3063 :
3064 26 : pthread_mutex_lock(&nvme_ctrlr->mutex);
3065 26 : rc = bdev_nvme_reset_ctrlr_unsafe(nvme_ctrlr, &msg_fn);
3066 26 : if (rc == -EBUSY) {
3067 : /*
3068 : * Reset call is queued only if it is from the app framework. This is on purpose so that
3069 : * we don't interfere with the app framework reset strategy. i.e. we are deferring to the
3070 : * upper level. If they are in the middle of a reset, we won't try to schedule another one.
3071 : */
3072 12 : TAILQ_INSERT_TAIL(&nvme_ctrlr->pending_resets, bio, retry_link);
3073 : }
3074 26 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
3075 :
3076 26 : if (rc == 0) {
3077 13 : assert(nvme_ctrlr->ctrlr_op_cb_fn == NULL);
3078 13 : assert(nvme_ctrlr->ctrlr_op_cb_arg == NULL);
3079 13 : nvme_ctrlr->ctrlr_op_cb_fn = bdev_nvme_reset_io_continue;
3080 13 : nvme_ctrlr->ctrlr_op_cb_arg = bio;
3081 :
3082 13 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
3083 :
3084 13 : NVME_BDEV_INFOLOG(nbdev, "reset_io %p started resetting ctrlr [%s, %u].\n",
3085 : bio, CTRLR_STRING(nvme_ctrlr), CTRLR_ID(nvme_ctrlr));
3086 13 : } else if (rc == -EBUSY) {
3087 12 : rc = 0;
3088 :
3089 12 : NVME_BDEV_INFOLOG(nbdev, "reset_io %p was queued to ctrlr [%s, %u].\n",
3090 : bio, CTRLR_STRING(nvme_ctrlr), CTRLR_ID(nvme_ctrlr));
3091 : } else {
3092 1 : NVME_BDEV_INFOLOG(nbdev, "reset_io %p could not reset ctrlr [%s, %u], rc:%d\n",
3093 : bio, CTRLR_STRING(nvme_ctrlr), CTRLR_ID(nvme_ctrlr), rc);
3094 : }
3095 :
3096 26 : return rc;
3097 : }
3098 :
3099 : static void
3100 16 : bdev_nvme_freeze_bdev_channel_done(struct nvme_bdev *nbdev, void *ctx, int status)
3101 : {
3102 16 : struct nvme_bdev_io *bio = ctx;
3103 16 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
3104 : struct nvme_bdev_channel *nbdev_ch;
3105 : struct nvme_io_path *io_path;
3106 : int rc;
3107 :
3108 16 : nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
3109 :
3110 : /* Initialize with failed status. With multipath it is enough to have at least one successful
3111 : * nvme_ctrlr reset. If there is none, reset status will remain failed.
3112 : */
3113 16 : bio->cpl.cdw0 = 1;
3114 :
3115 : /* Reset all nvme_ctrlrs of a bdev controller sequentially. */
3116 16 : io_path = STAILQ_FIRST(&nbdev_ch->io_path_list);
3117 16 : assert(io_path != NULL);
3118 :
3119 16 : rc = _bdev_nvme_reset_io(io_path, bio);
3120 16 : if (rc != 0) {
3121 : /* If the current nvme_ctrlr is disabled, skip it and move to the next nvme_ctrlr. */
3122 1 : rc = (rc == -EALREADY) ? 0 : rc;
3123 :
3124 1 : bdev_nvme_reset_io_continue(bio, rc);
3125 : }
3126 16 : }
3127 :
3128 : static void
3129 30 : bdev_nvme_freeze_bdev_channel(struct nvme_bdev_channel_iter *i,
3130 : struct nvme_bdev *nbdev,
3131 : struct nvme_bdev_channel *nbdev_ch, void *ctx)
3132 : {
3133 30 : nbdev_ch->resetting = true;
3134 :
3135 30 : nvme_bdev_for_each_channel_continue(i, 0);
3136 30 : }
3137 :
3138 : static void
3139 15 : bdev_nvme_reset_io(struct nvme_bdev *nbdev, struct nvme_bdev_io *bio)
3140 : {
3141 15 : NVME_BDEV_INFOLOG(nbdev, "reset_io %p started.\n", bio);
3142 :
3143 15 : nvme_bdev_for_each_channel(nbdev,
3144 : bdev_nvme_freeze_bdev_channel,
3145 : bio,
3146 : bdev_nvme_freeze_bdev_channel_done);
3147 15 : }
3148 :
3149 : static int
3150 32 : bdev_nvme_failover_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool remove)
3151 : {
3152 32 : if (nvme_ctrlr->destruct) {
3153 : /* Don't bother resetting if the controller is in the process of being destructed. */
3154 2 : return -ENXIO;
3155 : }
3156 :
3157 30 : if (nvme_ctrlr->resetting) {
3158 3 : if (!nvme_ctrlr->in_failover) {
3159 3 : NVME_CTRLR_NOTICELOG(nvme_ctrlr,
3160 : "Reset is already in progress. Defer failover until reset completes.\n");
3161 :
3162 : /* Defer failover until reset completes. */
3163 3 : nvme_ctrlr->pending_failover = true;
3164 3 : return -EINPROGRESS;
3165 : } else {
3166 0 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Unable to perform failover, already in progress.\n");
3167 0 : return -EBUSY;
3168 : }
3169 : }
3170 :
3171 27 : bdev_nvme_failover_trid(nvme_ctrlr, remove, true);
3172 :
3173 27 : if (nvme_ctrlr->reconnect_is_delayed) {
3174 1 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Reconnect is already scheduled.\n");
3175 :
3176 : /* We rely on the next reconnect for the failover. */
3177 1 : return -EALREADY;
3178 : }
3179 :
3180 26 : if (nvme_ctrlr->disabled) {
3181 0 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Controller is disabled.\n");
3182 :
3183 : /* We rely on the enablement for the failover. */
3184 0 : return -EALREADY;
3185 : }
3186 :
3187 26 : nvme_ctrlr->resetting = true;
3188 26 : nvme_ctrlr->in_failover = true;
3189 :
3190 26 : assert(nvme_ctrlr->reset_start_tsc == 0);
3191 26 : nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
3192 :
3193 26 : return 0;
3194 : }
3195 :
3196 : static int
3197 30 : bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
3198 : {
3199 : int rc;
3200 :
3201 30 : pthread_mutex_lock(&nvme_ctrlr->mutex);
3202 30 : rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, false);
3203 30 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
3204 :
3205 30 : if (rc == 0) {
3206 25 : spdk_thread_send_msg(nvme_ctrlr->thread, _bdev_nvme_reset_ctrlr, nvme_ctrlr);
3207 5 : } else if (rc == -EALREADY) {
3208 0 : rc = 0;
3209 : }
3210 :
3211 30 : return rc;
3212 : }
3213 :
3214 : static int bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks,
3215 : uint64_t num_blocks);
3216 :
3217 : static int bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks,
3218 : uint64_t num_blocks);
3219 :
3220 : static int bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks,
3221 : uint64_t src_offset_blocks,
3222 : uint64_t num_blocks);
3223 :
3224 : static void
3225 1 : bdev_nvme_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3226 : bool success)
3227 : {
3228 1 : struct nvme_bdev_io *bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3229 : int ret;
3230 :
3231 1 : if (!success) {
3232 0 : ret = -EINVAL;
3233 0 : goto exit;
3234 : }
3235 :
3236 1 : if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
3237 0 : ret = -ENXIO;
3238 0 : goto exit;
3239 : }
3240 :
3241 1 : ret = bdev_nvme_readv(bio,
3242 : bdev_io->u.bdev.iovs,
3243 : bdev_io->u.bdev.iovcnt,
3244 : bdev_io->u.bdev.md_buf,
3245 : bdev_io->u.bdev.num_blocks,
3246 : bdev_io->u.bdev.offset_blocks,
3247 : bdev_io->u.bdev.dif_check_flags,
3248 : bdev_io->u.bdev.memory_domain,
3249 : bdev_io->u.bdev.memory_domain_ctx,
3250 : bdev_io->u.bdev.accel_sequence);
3251 :
3252 1 : exit:
3253 1 : if (spdk_unlikely(ret != 0)) {
3254 0 : bdev_nvme_io_complete(bio, ret);
3255 : }
3256 1 : }
3257 :
3258 : static inline void
3259 59 : _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
3260 : {
3261 59 : struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3262 59 : struct spdk_bdev *bdev = bdev_io->bdev;
3263 : struct nvme_bdev_io *nbdev_io_to_abort;
3264 59 : int rc = 0;
3265 :
3266 59 : switch (bdev_io->type) {
3267 3 : case SPDK_BDEV_IO_TYPE_READ:
3268 3 : if (bdev_io->u.bdev.iovs && bdev_io->u.bdev.iovs[0].iov_base) {
3269 :
3270 2 : rc = bdev_nvme_readv(nbdev_io,
3271 : bdev_io->u.bdev.iovs,
3272 : bdev_io->u.bdev.iovcnt,
3273 : bdev_io->u.bdev.md_buf,
3274 : bdev_io->u.bdev.num_blocks,
3275 : bdev_io->u.bdev.offset_blocks,
3276 : bdev_io->u.bdev.dif_check_flags,
3277 : bdev_io->u.bdev.memory_domain,
3278 : bdev_io->u.bdev.memory_domain_ctx,
3279 : bdev_io->u.bdev.accel_sequence);
3280 : } else {
3281 1 : spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb,
3282 1 : bdev_io->u.bdev.num_blocks * bdev->blocklen);
3283 1 : rc = 0;
3284 : }
3285 3 : break;
3286 25 : case SPDK_BDEV_IO_TYPE_WRITE:
3287 25 : rc = bdev_nvme_writev(nbdev_io,
3288 : bdev_io->u.bdev.iovs,
3289 : bdev_io->u.bdev.iovcnt,
3290 : bdev_io->u.bdev.md_buf,
3291 : bdev_io->u.bdev.num_blocks,
3292 : bdev_io->u.bdev.offset_blocks,
3293 : bdev_io->u.bdev.dif_check_flags,
3294 : bdev_io->u.bdev.memory_domain,
3295 : bdev_io->u.bdev.memory_domain_ctx,
3296 : bdev_io->u.bdev.accel_sequence,
3297 : bdev_io->u.bdev.nvme_cdw12,
3298 : bdev_io->u.bdev.nvme_cdw13);
3299 25 : break;
3300 1 : case SPDK_BDEV_IO_TYPE_COMPARE:
3301 1 : rc = bdev_nvme_comparev(nbdev_io,
3302 : bdev_io->u.bdev.iovs,
3303 : bdev_io->u.bdev.iovcnt,
3304 : bdev_io->u.bdev.md_buf,
3305 : bdev_io->u.bdev.num_blocks,
3306 : bdev_io->u.bdev.offset_blocks,
3307 : bdev_io->u.bdev.dif_check_flags);
3308 1 : break;
3309 2 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
3310 2 : rc = bdev_nvme_comparev_and_writev(nbdev_io,
3311 : bdev_io->u.bdev.iovs,
3312 : bdev_io->u.bdev.iovcnt,
3313 : bdev_io->u.bdev.fused_iovs,
3314 : bdev_io->u.bdev.fused_iovcnt,
3315 : bdev_io->u.bdev.md_buf,
3316 : bdev_io->u.bdev.num_blocks,
3317 : bdev_io->u.bdev.offset_blocks,
3318 : bdev_io->u.bdev.dif_check_flags);
3319 2 : break;
3320 1 : case SPDK_BDEV_IO_TYPE_UNMAP:
3321 1 : rc = bdev_nvme_unmap(nbdev_io,
3322 : bdev_io->u.bdev.offset_blocks,
3323 : bdev_io->u.bdev.num_blocks);
3324 1 : break;
3325 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3326 0 : rc = bdev_nvme_write_zeroes(nbdev_io,
3327 : bdev_io->u.bdev.offset_blocks,
3328 : bdev_io->u.bdev.num_blocks);
3329 0 : break;
3330 15 : case SPDK_BDEV_IO_TYPE_RESET:
3331 15 : nbdev_io->io_path = NULL;
3332 15 : bdev_nvme_reset_io(bdev->ctxt, nbdev_io);
3333 15 : return;
3334 :
3335 1 : case SPDK_BDEV_IO_TYPE_FLUSH:
3336 1 : bdev_nvme_io_complete(nbdev_io, 0);
3337 1 : return;
3338 :
3339 0 : case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
3340 0 : rc = bdev_nvme_zone_appendv(nbdev_io,
3341 : bdev_io->u.bdev.iovs,
3342 : bdev_io->u.bdev.iovcnt,
3343 : bdev_io->u.bdev.md_buf,
3344 : bdev_io->u.bdev.num_blocks,
3345 : bdev_io->u.bdev.offset_blocks,
3346 : bdev_io->u.bdev.dif_check_flags);
3347 0 : break;
3348 0 : case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
3349 0 : rc = bdev_nvme_get_zone_info(nbdev_io,
3350 : bdev_io->u.zone_mgmt.zone_id,
3351 : bdev_io->u.zone_mgmt.num_zones,
3352 0 : bdev_io->u.zone_mgmt.buf);
3353 0 : break;
3354 0 : case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
3355 0 : rc = bdev_nvme_zone_management(nbdev_io,
3356 : bdev_io->u.zone_mgmt.zone_id,
3357 : bdev_io->u.zone_mgmt.zone_action);
3358 0 : break;
3359 5 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3360 5 : nbdev_io->io_path = NULL;
3361 5 : bdev_nvme_admin_passthru(nbdev_ch,
3362 : nbdev_io,
3363 : &bdev_io->u.nvme_passthru.cmd,
3364 : bdev_io->u.nvme_passthru.buf,
3365 : bdev_io->u.nvme_passthru.nbytes);
3366 5 : return;
3367 :
3368 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3369 0 : rc = bdev_nvme_io_passthru(nbdev_io,
3370 : &bdev_io->u.nvme_passthru.cmd,
3371 : bdev_io->u.nvme_passthru.buf,
3372 : bdev_io->u.nvme_passthru.nbytes);
3373 0 : break;
3374 0 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3375 0 : rc = bdev_nvme_io_passthru_md(nbdev_io,
3376 : &bdev_io->u.nvme_passthru.cmd,
3377 : bdev_io->u.nvme_passthru.buf,
3378 : bdev_io->u.nvme_passthru.nbytes,
3379 : bdev_io->u.nvme_passthru.md_buf,
3380 : bdev_io->u.nvme_passthru.md_len);
3381 0 : break;
3382 0 : case SPDK_BDEV_IO_TYPE_NVME_IOV_MD:
3383 0 : rc = bdev_nvme_iov_passthru_md(nbdev_io,
3384 : &bdev_io->u.nvme_passthru.cmd,
3385 : bdev_io->u.nvme_passthru.iovs,
3386 : bdev_io->u.nvme_passthru.iovcnt,
3387 : bdev_io->u.nvme_passthru.nbytes,
3388 : bdev_io->u.nvme_passthru.md_buf,
3389 : bdev_io->u.nvme_passthru.md_len);
3390 0 : break;
3391 6 : case SPDK_BDEV_IO_TYPE_ABORT:
3392 6 : nbdev_io->io_path = NULL;
3393 6 : nbdev_io_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx;
3394 6 : bdev_nvme_abort(nbdev_ch,
3395 : nbdev_io,
3396 : nbdev_io_to_abort);
3397 6 : return;
3398 :
3399 0 : case SPDK_BDEV_IO_TYPE_COPY:
3400 0 : rc = bdev_nvme_copy(nbdev_io,
3401 : bdev_io->u.bdev.offset_blocks,
3402 : bdev_io->u.bdev.copy.src_offset_blocks,
3403 : bdev_io->u.bdev.num_blocks);
3404 0 : break;
3405 0 : default:
3406 0 : rc = -EINVAL;
3407 0 : break;
3408 : }
3409 :
3410 32 : if (spdk_unlikely(rc != 0)) {
3411 0 : bdev_nvme_io_complete(nbdev_io, rc);
3412 : }
3413 : }
3414 :
3415 : static void
3416 68 : bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
3417 : {
3418 68 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
3419 68 : struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3420 :
3421 68 : if (spdk_likely(nbdev_io->submit_tsc == 0)) {
3422 68 : nbdev_io->submit_tsc = spdk_bdev_io_get_submit_tsc(bdev_io);
3423 : } else {
3424 : /* There are cases where submit_tsc != 0, i.e. retry I/O.
3425 : * We need to update submit_tsc here.
3426 : */
3427 0 : nbdev_io->submit_tsc = spdk_get_ticks();
3428 : }
3429 :
3430 68 : spdk_trace_record(TRACE_BDEV_NVME_IO_START, 0, 0, (uintptr_t)nbdev_io, (uintptr_t)bdev_io);
3431 68 : nbdev_io->io_path = bdev_nvme_find_io_path(nbdev_ch);
3432 68 : if (spdk_unlikely(!nbdev_io->io_path)) {
3433 13 : if (!bdev_nvme_io_type_is_admin(bdev_io->type)) {
3434 12 : bdev_nvme_io_complete(nbdev_io, -ENXIO);
3435 12 : return;
3436 : }
3437 :
3438 : /* Admin commands do not use the optimal I/O path.
3439 : * Simply fall through even if it is not found.
3440 : */
3441 : }
3442 :
3443 56 : _bdev_nvme_submit_request(nbdev_ch, bdev_io);
3444 : }
3445 :
3446 : static bool
3447 0 : bdev_nvme_is_supported_csi(enum spdk_nvme_csi csi)
3448 : {
3449 0 : switch (csi) {
3450 0 : case SPDK_NVME_CSI_NVM:
3451 0 : return true;
3452 0 : case SPDK_NVME_CSI_ZNS:
3453 0 : return true;
3454 0 : default:
3455 0 : return false;
3456 : }
3457 : }
3458 :
3459 : static bool
3460 0 : bdev_nvme_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
3461 : {
3462 0 : struct nvme_bdev *nbdev = ctx;
3463 : struct nvme_ns *nvme_ns;
3464 : struct spdk_nvme_ns *ns;
3465 : struct spdk_nvme_ctrlr *ctrlr;
3466 : const struct spdk_nvme_ctrlr_data *cdata;
3467 :
3468 0 : nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
3469 0 : assert(nvme_ns != NULL);
3470 0 : ns = nvme_ns->ns;
3471 0 : if (ns == NULL) {
3472 0 : return false;
3473 : }
3474 :
3475 0 : if (!bdev_nvme_is_supported_csi(spdk_nvme_ns_get_csi(ns))) {
3476 0 : switch (io_type) {
3477 0 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3478 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3479 0 : return true;
3480 :
3481 0 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3482 0 : return spdk_nvme_ns_get_md_size(ns) ? true : false;
3483 :
3484 0 : default:
3485 0 : return false;
3486 : }
3487 : }
3488 :
3489 0 : ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3490 :
3491 0 : switch (io_type) {
3492 0 : case SPDK_BDEV_IO_TYPE_READ:
3493 : case SPDK_BDEV_IO_TYPE_WRITE:
3494 : case SPDK_BDEV_IO_TYPE_RESET:
3495 : case SPDK_BDEV_IO_TYPE_FLUSH:
3496 : case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3497 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3498 : case SPDK_BDEV_IO_TYPE_ABORT:
3499 0 : return true;
3500 :
3501 0 : case SPDK_BDEV_IO_TYPE_COMPARE:
3502 0 : return spdk_nvme_ns_supports_compare(ns);
3503 :
3504 0 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3505 0 : return spdk_nvme_ns_get_md_size(ns) ? true : false;
3506 :
3507 0 : case SPDK_BDEV_IO_TYPE_UNMAP:
3508 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3509 0 : return cdata->oncs.dsm;
3510 :
3511 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3512 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3513 0 : return cdata->oncs.write_zeroes;
3514 :
3515 0 : case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
3516 0 : if (spdk_nvme_ctrlr_get_flags(ctrlr) &
3517 : SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED) {
3518 0 : return true;
3519 : }
3520 0 : return false;
3521 :
3522 0 : case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
3523 : case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
3524 0 : return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS;
3525 :
3526 0 : case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
3527 0 : return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS &&
3528 0 : spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ZONE_APPEND_SUPPORTED;
3529 :
3530 0 : case SPDK_BDEV_IO_TYPE_COPY:
3531 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3532 0 : return cdata->oncs.copy;
3533 :
3534 0 : default:
3535 0 : return false;
3536 : }
3537 : }
3538 :
3539 : static int
3540 61 : nvme_qpair_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ctrlr_channel *ctrlr_ch)
3541 : {
3542 : struct nvme_qpair *nvme_qpair;
3543 : struct spdk_io_channel *pg_ch;
3544 : int rc;
3545 :
3546 61 : nvme_qpair = calloc(1, sizeof(*nvme_qpair));
3547 61 : if (!nvme_qpair) {
3548 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to alloc nvme_qpair.\n");
3549 0 : return -1;
3550 : }
3551 :
3552 61 : TAILQ_INIT(&nvme_qpair->io_path_list);
3553 :
3554 61 : nvme_qpair->ctrlr = nvme_ctrlr;
3555 61 : nvme_qpair->ctrlr_ch = ctrlr_ch;
3556 :
3557 61 : pg_ch = spdk_get_io_channel(&g_nvme_bdev_ctrlrs);
3558 61 : if (!pg_ch) {
3559 0 : free(nvme_qpair);
3560 0 : return -1;
3561 : }
3562 :
3563 61 : nvme_qpair->group = spdk_io_channel_get_ctx(pg_ch);
3564 :
3565 : #ifdef SPDK_CONFIG_VTUNE
3566 : nvme_qpair->group->collect_spin_stat = true;
3567 : #else
3568 61 : nvme_qpair->group->collect_spin_stat = false;
3569 : #endif
3570 :
3571 61 : if (!nvme_ctrlr->disabled) {
3572 : /* If a nvme_ctrlr is disabled, don't try to create qpair for it. Qpair will
3573 : * be created when it's enabled.
3574 : */
3575 61 : rc = bdev_nvme_create_qpair(nvme_qpair);
3576 61 : if (rc != 0) {
3577 : /* nvme_ctrlr can't create IO qpair if connection is down.
3578 : * If reconnect_delay_sec is non-zero, creating IO qpair is retried
3579 : * after reconnect_delay_sec seconds. If bdev_retry_count is non-zero,
3580 : * submitted IO will be queued until IO qpair is successfully created.
3581 : *
3582 : * Hence, if both are satisfied, ignore the failure.
3583 : */
3584 0 : if (nvme_ctrlr->opts.reconnect_delay_sec == 0 || g_opts.bdev_retry_count == 0) {
3585 0 : spdk_put_io_channel(pg_ch);
3586 0 : free(nvme_qpair);
3587 0 : return rc;
3588 : }
3589 : }
3590 : }
3591 :
3592 61 : TAILQ_INSERT_TAIL(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);
3593 :
3594 61 : ctrlr_ch->qpair = nvme_qpair;
3595 :
3596 61 : nvme_ctrlr_get_ref(nvme_ctrlr);
3597 :
3598 61 : return 0;
3599 : }
3600 :
3601 : static int
3602 61 : bdev_nvme_create_ctrlr_channel_cb(void *io_device, void *ctx_buf)
3603 : {
3604 61 : struct nvme_ctrlr *nvme_ctrlr = io_device;
3605 61 : struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
3606 :
3607 61 : return nvme_qpair_create(nvme_ctrlr, ctrlr_ch);
3608 : }
3609 :
3610 : static void
3611 61 : nvme_qpair_delete(struct nvme_qpair *nvme_qpair)
3612 : {
3613 : struct nvme_io_path *io_path, *next;
3614 :
3615 61 : assert(nvme_qpair->group != NULL);
3616 :
3617 100 : TAILQ_FOREACH_SAFE(io_path, &nvme_qpair->io_path_list, tailq, next) {
3618 39 : TAILQ_REMOVE(&nvme_qpair->io_path_list, io_path, tailq);
3619 39 : nvme_io_path_free(io_path);
3620 : }
3621 :
3622 61 : TAILQ_REMOVE(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);
3623 :
3624 61 : spdk_put_io_channel(spdk_io_channel_from_ctx(nvme_qpair->group));
3625 :
3626 61 : nvme_ctrlr_put_ref(nvme_qpair->ctrlr);
3627 :
3628 61 : free(nvme_qpair);
3629 61 : }
3630 :
3631 : static void
3632 61 : bdev_nvme_destroy_ctrlr_channel_cb(void *io_device, void *ctx_buf)
3633 : {
3634 61 : struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
3635 : struct nvme_qpair *nvme_qpair;
3636 :
3637 61 : nvme_qpair = ctrlr_ch->qpair;
3638 61 : assert(nvme_qpair != NULL);
3639 :
3640 61 : _bdev_nvme_clear_io_path_cache(nvme_qpair);
3641 :
3642 61 : if (nvme_qpair->qpair != NULL) {
3643 47 : if (ctrlr_ch->reset_iter == NULL) {
3644 47 : spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);
3645 : } else {
3646 : /* Skip current ctrlr_channel in a full reset sequence because
3647 : * it is being deleted now. The qpair is already being disconnected.
3648 : * We do not have to restart disconnecting it.
3649 : */
3650 0 : nvme_ctrlr_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
3651 : }
3652 :
3653 : /* We cannot release a reference to the poll group now.
3654 : * The qpair may be disconnected asynchronously later.
3655 : * We need to poll it until it is actually disconnected.
3656 : * Just detach the qpair from the deleting ctrlr_channel.
3657 : */
3658 47 : nvme_qpair->ctrlr_ch = NULL;
3659 : } else {
3660 14 : assert(ctrlr_ch->reset_iter == NULL);
3661 :
3662 14 : nvme_qpair_delete(nvme_qpair);
3663 : }
3664 61 : }
3665 :
3666 : static inline struct spdk_io_channel *
3667 0 : bdev_nvme_get_accel_channel(struct nvme_poll_group *group)
3668 : {
3669 0 : if (spdk_unlikely(!group->accel_channel)) {
3670 0 : group->accel_channel = spdk_accel_get_io_channel();
3671 0 : if (!group->accel_channel) {
3672 0 : SPDK_ERRLOG("Cannot get the accel_channel for bdev nvme polling group=%p\n",
3673 : group);
3674 0 : return NULL;
3675 : }
3676 : }
3677 :
3678 0 : return group->accel_channel;
3679 : }
3680 :
3681 : static void
3682 0 : bdev_nvme_finish_sequence(void *seq, spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
3683 : {
3684 0 : spdk_accel_sequence_finish(seq, cb_fn, cb_arg);
3685 0 : }
3686 :
3687 : static void
3688 0 : bdev_nvme_abort_sequence(void *seq)
3689 : {
3690 0 : spdk_accel_sequence_abort(seq);
3691 0 : }
3692 :
3693 : static void
3694 0 : bdev_nvme_reverse_sequence(void *seq)
3695 : {
3696 0 : spdk_accel_sequence_reverse(seq);
3697 0 : }
3698 :
3699 : static int
3700 0 : bdev_nvme_append_crc32c(void *ctx, void **seq, uint32_t *dst, struct iovec *iovs, uint32_t iovcnt,
3701 : struct spdk_memory_domain *domain, void *domain_ctx, uint32_t seed,
3702 : spdk_nvme_accel_step_cb cb_fn, void *cb_arg)
3703 : {
3704 : struct spdk_io_channel *ch;
3705 0 : struct nvme_poll_group *group = ctx;
3706 :
3707 0 : ch = bdev_nvme_get_accel_channel(group);
3708 0 : if (spdk_unlikely(ch == NULL)) {
3709 0 : return -ENOMEM;
3710 : }
3711 :
3712 0 : return spdk_accel_append_crc32c((struct spdk_accel_sequence **)seq, ch, dst, iovs, iovcnt,
3713 : domain, domain_ctx, seed, cb_fn, cb_arg);
3714 : }
3715 :
3716 : static int
3717 0 : bdev_nvme_append_copy(void *ctx, void **seq, struct iovec *dst_iovs, uint32_t dst_iovcnt,
3718 : struct spdk_memory_domain *dst_domain, void *dst_domain_ctx,
3719 : struct iovec *src_iovs, uint32_t src_iovcnt,
3720 : struct spdk_memory_domain *src_domain, void *src_domain_ctx,
3721 : spdk_nvme_accel_step_cb cb_fn, void *cb_arg)
3722 : {
3723 : struct spdk_io_channel *ch;
3724 0 : struct nvme_poll_group *group = ctx;
3725 :
3726 0 : ch = bdev_nvme_get_accel_channel(group);
3727 0 : if (spdk_unlikely(ch == NULL)) {
3728 0 : return -ENOMEM;
3729 : }
3730 :
3731 0 : return spdk_accel_append_copy((struct spdk_accel_sequence **)seq, ch,
3732 : dst_iovs, dst_iovcnt, dst_domain, dst_domain_ctx,
3733 : src_iovs, src_iovcnt, src_domain, src_domain_ctx,
3734 : cb_fn, cb_arg);
3735 : }
3736 :
3737 : static struct spdk_nvme_accel_fn_table g_bdev_nvme_accel_fn_table = {
3738 : .table_size = sizeof(struct spdk_nvme_accel_fn_table),
3739 : .append_crc32c = bdev_nvme_append_crc32c,
3740 : .append_copy = bdev_nvme_append_copy,
3741 : .finish_sequence = bdev_nvme_finish_sequence,
3742 : .reverse_sequence = bdev_nvme_reverse_sequence,
3743 : .abort_sequence = bdev_nvme_abort_sequence,
3744 : };
3745 :
3746 : static int
3747 0 : bdev_nvme_interrupt_wrapper(void *ctx)
3748 : {
3749 : int num_events;
3750 0 : struct nvme_poll_group *group = ctx;
3751 :
3752 0 : num_events = spdk_nvme_poll_group_wait(group->group, bdev_nvme_disconnected_qpair_cb);
3753 0 : if (spdk_unlikely(num_events < 0)) {
3754 0 : bdev_nvme_check_io_qpairs(group);
3755 : }
3756 :
3757 0 : return num_events;
3758 : }
3759 :
3760 : static int
3761 46 : bdev_nvme_create_poll_group_cb(void *io_device, void *ctx_buf)
3762 : {
3763 46 : struct nvme_poll_group *group = ctx_buf;
3764 : uint64_t period;
3765 : int fd;
3766 :
3767 46 : TAILQ_INIT(&group->qpair_list);
3768 :
3769 46 : group->group = spdk_nvme_poll_group_create(group, &g_bdev_nvme_accel_fn_table);
3770 46 : if (group->group == NULL) {
3771 0 : return -1;
3772 : }
3773 :
3774 46 : period = spdk_interrupt_mode_is_enabled() ? 0 : g_opts.nvme_ioq_poll_period_us;
3775 46 : group->poller = SPDK_POLLER_REGISTER(bdev_nvme_poll, group, period);
3776 :
3777 46 : if (group->poller == NULL) {
3778 0 : spdk_nvme_poll_group_destroy(group->group);
3779 0 : return -1;
3780 : }
3781 :
3782 46 : if (spdk_interrupt_mode_is_enabled()) {
3783 0 : spdk_poller_register_interrupt(group->poller, NULL, NULL);
3784 :
3785 0 : fd = spdk_nvme_poll_group_get_fd(group->group);
3786 0 : if (fd < 0) {
3787 0 : spdk_nvme_poll_group_destroy(group->group);
3788 0 : return -1;
3789 : }
3790 :
3791 0 : group->intr = SPDK_INTERRUPT_REGISTER(fd, bdev_nvme_interrupt_wrapper, group);
3792 0 : if (!group->intr) {
3793 0 : spdk_nvme_poll_group_destroy(group->group);
3794 0 : return -1;
3795 : }
3796 : }
3797 :
3798 46 : return 0;
3799 : }
3800 :
3801 : static void
3802 46 : bdev_nvme_destroy_poll_group_cb(void *io_device, void *ctx_buf)
3803 : {
3804 46 : struct nvme_poll_group *group = ctx_buf;
3805 :
3806 46 : assert(TAILQ_EMPTY(&group->qpair_list));
3807 :
3808 46 : if (group->accel_channel) {
3809 0 : spdk_put_io_channel(group->accel_channel);
3810 : }
3811 :
3812 46 : if (spdk_interrupt_mode_is_enabled()) {
3813 0 : spdk_interrupt_unregister(&group->intr);
3814 : }
3815 :
3816 46 : spdk_poller_unregister(&group->poller);
3817 46 : if (spdk_nvme_poll_group_destroy(group->group)) {
3818 0 : SPDK_ERRLOG("Unable to destroy a poll group for the NVMe bdev module.\n");
3819 0 : assert(false);
3820 : }
3821 46 : }
3822 :
3823 : static struct spdk_io_channel *
3824 0 : bdev_nvme_get_io_channel(void *ctx)
3825 : {
3826 0 : struct nvme_bdev *nvme_bdev = ctx;
3827 :
3828 0 : return spdk_get_io_channel(nvme_bdev);
3829 : }
3830 :
3831 : static void *
3832 0 : bdev_nvme_get_module_ctx(void *ctx)
3833 : {
3834 0 : struct nvme_bdev *nvme_bdev = ctx;
3835 : struct nvme_ns *nvme_ns;
3836 :
3837 0 : if (!nvme_bdev || nvme_bdev->disk.module != &nvme_if) {
3838 0 : return NULL;
3839 : }
3840 :
3841 0 : nvme_ns = TAILQ_FIRST(&nvme_bdev->nvme_ns_list);
3842 0 : if (!nvme_ns) {
3843 0 : return NULL;
3844 : }
3845 :
3846 0 : return nvme_ns->ns;
3847 : }
3848 :
3849 : static const char *
3850 0 : _nvme_ana_state_str(enum spdk_nvme_ana_state ana_state)
3851 : {
3852 0 : switch (ana_state) {
3853 0 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
3854 0 : return "optimized";
3855 0 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
3856 0 : return "non_optimized";
3857 0 : case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3858 0 : return "inaccessible";
3859 0 : case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3860 0 : return "persistent_loss";
3861 0 : case SPDK_NVME_ANA_CHANGE_STATE:
3862 0 : return "change";
3863 0 : default:
3864 0 : return NULL;
3865 : }
3866 : }
3867 :
3868 : static int
3869 8 : bdev_nvme_get_memory_domains(void *ctx, struct spdk_memory_domain **domains, int array_size)
3870 : {
3871 8 : struct spdk_memory_domain **_domains = NULL;
3872 8 : struct nvme_bdev *nbdev = ctx;
3873 : struct nvme_ns *nvme_ns;
3874 8 : int i = 0, _array_size = array_size;
3875 8 : int rc = 0;
3876 :
3877 22 : TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
3878 14 : if (domains && array_size >= i) {
3879 11 : _domains = &domains[i];
3880 : } else {
3881 3 : _domains = NULL;
3882 : }
3883 14 : rc = spdk_nvme_ctrlr_get_memory_domains(nvme_ns->ctrlr->ctrlr, _domains, _array_size);
3884 14 : if (rc > 0) {
3885 13 : i += rc;
3886 13 : if (_array_size >= rc) {
3887 9 : _array_size -= rc;
3888 : } else {
3889 4 : _array_size = 0;
3890 : }
3891 1 : } else if (rc < 0) {
3892 0 : return rc;
3893 : }
3894 : }
3895 :
3896 8 : return i;
3897 : }
3898 :
3899 : static const char *
3900 0 : nvme_ctrlr_get_state_str(struct nvme_ctrlr *nvme_ctrlr)
3901 : {
3902 0 : if (nvme_ctrlr->destruct) {
3903 0 : return "deleting";
3904 0 : } else if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
3905 0 : return "failed";
3906 0 : } else if (nvme_ctrlr->resetting) {
3907 0 : return "resetting";
3908 0 : } else if (nvme_ctrlr->reconnect_is_delayed > 0) {
3909 0 : return "reconnect_is_delayed";
3910 0 : } else if (nvme_ctrlr->disabled) {
3911 0 : return "disabled";
3912 : } else {
3913 0 : return "enabled";
3914 : }
3915 : }
3916 :
3917 : void
3918 0 : nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr)
3919 0 : {
3920 : struct spdk_nvme_transport_id *trid;
3921 : const struct spdk_nvme_ctrlr_opts *opts;
3922 : const struct spdk_nvme_ctrlr_data *cdata;
3923 : struct nvme_path_id *path_id;
3924 : int32_t numa_id;
3925 :
3926 0 : spdk_json_write_object_begin(w);
3927 :
3928 0 : spdk_json_write_named_string(w, "state", nvme_ctrlr_get_state_str(nvme_ctrlr));
3929 :
3930 : #ifdef SPDK_CONFIG_NVME_CUSE
3931 0 : size_t cuse_name_size = 128;
3932 0 : char cuse_name[cuse_name_size];
3933 :
3934 0 : int rc = spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr, cuse_name, &cuse_name_size);
3935 0 : if (rc == 0) {
3936 0 : spdk_json_write_named_string(w, "cuse_device", cuse_name);
3937 : }
3938 : #endif
3939 0 : trid = &nvme_ctrlr->active_path_id->trid;
3940 0 : spdk_json_write_named_object_begin(w, "trid");
3941 0 : nvme_bdev_dump_trid_json(trid, w);
3942 0 : spdk_json_write_object_end(w);
3943 :
3944 0 : path_id = TAILQ_NEXT(nvme_ctrlr->active_path_id, link);
3945 0 : if (path_id != NULL) {
3946 0 : spdk_json_write_named_array_begin(w, "alternate_trids");
3947 : do {
3948 0 : trid = &path_id->trid;
3949 0 : spdk_json_write_object_begin(w);
3950 0 : nvme_bdev_dump_trid_json(trid, w);
3951 0 : spdk_json_write_object_end(w);
3952 :
3953 0 : path_id = TAILQ_NEXT(path_id, link);
3954 0 : } while (path_id != NULL);
3955 0 : spdk_json_write_array_end(w);
3956 : }
3957 :
3958 0 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
3959 0 : spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);
3960 :
3961 0 : opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
3962 0 : spdk_json_write_named_object_begin(w, "host");
3963 0 : spdk_json_write_named_string(w, "nqn", opts->hostnqn);
3964 0 : spdk_json_write_named_string(w, "addr", opts->src_addr);
3965 0 : spdk_json_write_named_string(w, "svcid", opts->src_svcid);
3966 0 : spdk_json_write_object_end(w);
3967 :
3968 0 : numa_id = spdk_nvme_ctrlr_get_numa_id(nvme_ctrlr->ctrlr);
3969 0 : if (numa_id != SPDK_ENV_NUMA_ID_ANY) {
3970 0 : spdk_json_write_named_uint32(w, "numa_id", numa_id);
3971 : }
3972 0 : spdk_json_write_object_end(w);
3973 0 : }
3974 :
3975 : static void
3976 0 : nvme_namespace_info_json(struct spdk_json_write_ctx *w,
3977 : struct nvme_ns *nvme_ns)
3978 0 : {
3979 : struct spdk_nvme_ns *ns;
3980 : struct spdk_nvme_ctrlr *ctrlr;
3981 : const struct spdk_nvme_ctrlr_data *cdata;
3982 : const struct spdk_nvme_transport_id *trid;
3983 : union spdk_nvme_vs_register vs;
3984 : const struct spdk_nvme_ns_data *nsdata;
3985 0 : char buf[128];
3986 :
3987 0 : ns = nvme_ns->ns;
3988 0 : if (ns == NULL) {
3989 0 : return;
3990 : }
3991 :
3992 0 : ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3993 :
3994 0 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3995 0 : trid = spdk_nvme_ctrlr_get_transport_id(ctrlr);
3996 0 : vs = spdk_nvme_ctrlr_get_regs_vs(ctrlr);
3997 :
3998 0 : spdk_json_write_object_begin(w);
3999 :
4000 0 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
4001 0 : spdk_json_write_named_string(w, "pci_address", trid->traddr);
4002 : }
4003 :
4004 0 : spdk_json_write_named_object_begin(w, "trid");
4005 :
4006 0 : nvme_bdev_dump_trid_json(trid, w);
4007 :
4008 0 : spdk_json_write_object_end(w);
4009 :
4010 : #ifdef SPDK_CONFIG_NVME_CUSE
4011 0 : size_t cuse_name_size = 128;
4012 0 : char cuse_name[cuse_name_size];
4013 :
4014 0 : int rc = spdk_nvme_cuse_get_ns_name(ctrlr, spdk_nvme_ns_get_id(ns),
4015 : cuse_name, &cuse_name_size);
4016 0 : if (rc == 0) {
4017 0 : spdk_json_write_named_string(w, "cuse_device", cuse_name);
4018 : }
4019 : #endif
4020 :
4021 0 : spdk_json_write_named_object_begin(w, "ctrlr_data");
4022 :
4023 0 : spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);
4024 :
4025 0 : spdk_json_write_named_string_fmt(w, "vendor_id", "0x%04x", cdata->vid);
4026 :
4027 0 : snprintf(buf, sizeof(cdata->mn) + 1, "%s", cdata->mn);
4028 0 : spdk_str_trim(buf);
4029 0 : spdk_json_write_named_string(w, "model_number", buf);
4030 :
4031 0 : snprintf(buf, sizeof(cdata->sn) + 1, "%s", cdata->sn);
4032 0 : spdk_str_trim(buf);
4033 0 : spdk_json_write_named_string(w, "serial_number", buf);
4034 :
4035 0 : snprintf(buf, sizeof(cdata->fr) + 1, "%s", cdata->fr);
4036 0 : spdk_str_trim(buf);
4037 0 : spdk_json_write_named_string(w, "firmware_revision", buf);
4038 :
4039 0 : if (cdata->subnqn[0] != '\0') {
4040 0 : spdk_json_write_named_string(w, "subnqn", cdata->subnqn);
4041 : }
4042 :
4043 0 : spdk_json_write_named_object_begin(w, "oacs");
4044 :
4045 0 : spdk_json_write_named_uint32(w, "security", cdata->oacs.security);
4046 0 : spdk_json_write_named_uint32(w, "format", cdata->oacs.format);
4047 0 : spdk_json_write_named_uint32(w, "firmware", cdata->oacs.firmware);
4048 0 : spdk_json_write_named_uint32(w, "ns_manage", cdata->oacs.ns_manage);
4049 :
4050 0 : spdk_json_write_object_end(w);
4051 :
4052 0 : spdk_json_write_named_bool(w, "multi_ctrlr", cdata->cmic.multi_ctrlr);
4053 0 : spdk_json_write_named_bool(w, "ana_reporting", cdata->cmic.ana_reporting);
4054 :
4055 0 : spdk_json_write_object_end(w);
4056 :
4057 0 : spdk_json_write_named_object_begin(w, "vs");
4058 :
4059 0 : spdk_json_write_name(w, "nvme_version");
4060 0 : if (vs.bits.ter) {
4061 0 : spdk_json_write_string_fmt(w, "%u.%u.%u", vs.bits.mjr, vs.bits.mnr, vs.bits.ter);
4062 : } else {
4063 0 : spdk_json_write_string_fmt(w, "%u.%u", vs.bits.mjr, vs.bits.mnr);
4064 : }
4065 :
4066 0 : spdk_json_write_object_end(w);
4067 :
4068 0 : nsdata = spdk_nvme_ns_get_data(ns);
4069 :
4070 0 : spdk_json_write_named_object_begin(w, "ns_data");
4071 :
4072 0 : spdk_json_write_named_uint32(w, "id", spdk_nvme_ns_get_id(ns));
4073 :
4074 0 : if (cdata->cmic.ana_reporting) {
4075 0 : spdk_json_write_named_string(w, "ana_state",
4076 : _nvme_ana_state_str(nvme_ns->ana_state));
4077 : }
4078 :
4079 0 : spdk_json_write_named_bool(w, "can_share", nsdata->nmic.can_share);
4080 :
4081 0 : spdk_json_write_object_end(w);
4082 :
4083 0 : if (cdata->oacs.security) {
4084 0 : spdk_json_write_named_object_begin(w, "security");
4085 :
4086 0 : spdk_json_write_named_bool(w, "opal", nvme_ns->bdev->opal);
4087 :
4088 0 : spdk_json_write_object_end(w);
4089 : }
4090 :
4091 0 : spdk_json_write_object_end(w);
4092 : }
4093 :
4094 : static const char *
4095 0 : nvme_bdev_get_mp_policy_str(struct nvme_bdev *nbdev)
4096 : {
4097 0 : switch (nbdev->mp_policy) {
4098 0 : case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
4099 0 : return "active_passive";
4100 0 : case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
4101 0 : return "active_active";
4102 0 : default:
4103 0 : assert(false);
4104 : return "invalid";
4105 : }
4106 : }
4107 :
4108 : static const char *
4109 0 : nvme_bdev_get_mp_selector_str(struct nvme_bdev *nbdev)
4110 : {
4111 0 : switch (nbdev->mp_selector) {
4112 0 : case BDEV_NVME_MP_SELECTOR_ROUND_ROBIN:
4113 0 : return "round_robin";
4114 0 : case BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH:
4115 0 : return "queue_depth";
4116 0 : default:
4117 0 : assert(false);
4118 : return "invalid";
4119 : }
4120 : }
4121 :
4122 : static int
4123 0 : bdev_nvme_dump_info_json(void *ctx, struct spdk_json_write_ctx *w)
4124 : {
4125 0 : struct nvme_bdev *nvme_bdev = ctx;
4126 : struct nvme_ns *nvme_ns;
4127 :
4128 0 : pthread_mutex_lock(&nvme_bdev->mutex);
4129 0 : spdk_json_write_named_array_begin(w, "nvme");
4130 0 : TAILQ_FOREACH(nvme_ns, &nvme_bdev->nvme_ns_list, tailq) {
4131 0 : nvme_namespace_info_json(w, nvme_ns);
4132 : }
4133 0 : spdk_json_write_array_end(w);
4134 0 : spdk_json_write_named_string(w, "mp_policy", nvme_bdev_get_mp_policy_str(nvme_bdev));
4135 0 : if (nvme_bdev->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
4136 0 : spdk_json_write_named_string(w, "selector", nvme_bdev_get_mp_selector_str(nvme_bdev));
4137 0 : if (nvme_bdev->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
4138 0 : spdk_json_write_named_uint32(w, "rr_min_io", nvme_bdev->rr_min_io);
4139 : }
4140 : }
4141 0 : pthread_mutex_unlock(&nvme_bdev->mutex);
4142 :
4143 0 : return 0;
4144 : }
4145 :
4146 : static void
4147 0 : bdev_nvme_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
4148 : {
4149 : /* No config per bdev needed */
4150 0 : }
4151 :
4152 : static uint64_t
4153 0 : bdev_nvme_get_spin_time(struct spdk_io_channel *ch)
4154 : {
4155 0 : struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
4156 : struct nvme_io_path *io_path;
4157 : struct nvme_poll_group *group;
4158 0 : uint64_t spin_time = 0;
4159 :
4160 0 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
4161 0 : group = io_path->qpair->group;
4162 :
4163 0 : if (!group || !group->collect_spin_stat) {
4164 0 : continue;
4165 : }
4166 :
4167 0 : if (group->end_ticks != 0) {
4168 0 : group->spin_ticks += (group->end_ticks - group->start_ticks);
4169 0 : group->end_ticks = 0;
4170 : }
4171 :
4172 0 : spin_time += group->spin_ticks;
4173 0 : group->start_ticks = 0;
4174 0 : group->spin_ticks = 0;
4175 : }
4176 :
4177 0 : return (spin_time * 1000000ULL) / spdk_get_ticks_hz();
4178 : }
4179 :
4180 : static void
4181 0 : bdev_nvme_reset_device_stat(void *ctx)
4182 : {
4183 0 : struct nvme_bdev *nbdev = ctx;
4184 :
4185 0 : if (nbdev->err_stat != NULL) {
4186 0 : memset(nbdev->err_stat, 0, sizeof(struct nvme_error_stat));
4187 : }
4188 0 : }
4189 :
4190 : /* JSON string should be lowercases and underscore delimited string. */
4191 : static void
4192 0 : bdev_nvme_format_nvme_status(char *dst, const char *src)
4193 : {
4194 0 : char tmp[256];
4195 :
4196 0 : spdk_strcpy_replace(dst, 256, src, " - ", "_");
4197 0 : spdk_strcpy_replace(tmp, 256, dst, "-", "_");
4198 0 : spdk_strcpy_replace(dst, 256, tmp, " ", "_");
4199 0 : spdk_strlwr(dst);
4200 0 : }
4201 :
4202 : static void
4203 0 : bdev_nvme_dump_device_stat_json(void *ctx, struct spdk_json_write_ctx *w)
4204 : {
4205 0 : struct nvme_bdev *nbdev = ctx;
4206 0 : struct spdk_nvme_status status = {};
4207 : uint16_t sct, sc;
4208 0 : char status_json[256];
4209 : const char *status_str;
4210 :
4211 0 : if (nbdev->err_stat == NULL) {
4212 0 : return;
4213 : }
4214 :
4215 0 : spdk_json_write_named_object_begin(w, "nvme_error");
4216 :
4217 0 : spdk_json_write_named_object_begin(w, "status_type");
4218 0 : for (sct = 0; sct < 8; sct++) {
4219 0 : if (nbdev->err_stat->status_type[sct] == 0) {
4220 0 : continue;
4221 : }
4222 0 : status.sct = sct;
4223 :
4224 0 : status_str = spdk_nvme_cpl_get_status_type_string(&status);
4225 0 : assert(status_str != NULL);
4226 0 : bdev_nvme_format_nvme_status(status_json, status_str);
4227 :
4228 0 : spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status_type[sct]);
4229 : }
4230 0 : spdk_json_write_object_end(w);
4231 :
4232 0 : spdk_json_write_named_object_begin(w, "status_code");
4233 0 : for (sct = 0; sct < 4; sct++) {
4234 0 : status.sct = sct;
4235 0 : for (sc = 0; sc < 256; sc++) {
4236 0 : if (nbdev->err_stat->status[sct][sc] == 0) {
4237 0 : continue;
4238 : }
4239 0 : status.sc = sc;
4240 :
4241 0 : status_str = spdk_nvme_cpl_get_status_string(&status);
4242 0 : assert(status_str != NULL);
4243 0 : bdev_nvme_format_nvme_status(status_json, status_str);
4244 :
4245 0 : spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status[sct][sc]);
4246 : }
4247 : }
4248 0 : spdk_json_write_object_end(w);
4249 :
4250 0 : spdk_json_write_object_end(w);
4251 : }
4252 :
4253 : static bool
4254 0 : bdev_nvme_accel_sequence_supported(void *ctx, enum spdk_bdev_io_type type)
4255 : {
4256 0 : struct nvme_bdev *nbdev = ctx;
4257 : struct nvme_ns *nvme_ns;
4258 : struct spdk_nvme_ctrlr *ctrlr;
4259 :
4260 0 : if (!g_opts.allow_accel_sequence) {
4261 0 : return false;
4262 : }
4263 :
4264 0 : switch (type) {
4265 0 : case SPDK_BDEV_IO_TYPE_WRITE:
4266 : case SPDK_BDEV_IO_TYPE_READ:
4267 0 : break;
4268 0 : default:
4269 0 : return false;
4270 : }
4271 :
4272 0 : nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
4273 0 : assert(nvme_ns != NULL);
4274 :
4275 0 : ctrlr = nvme_ns->ctrlr->ctrlr;
4276 0 : assert(ctrlr != NULL);
4277 :
4278 0 : return spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ACCEL_SEQUENCE_SUPPORTED;
4279 : }
4280 :
4281 : static const struct spdk_bdev_fn_table nvmelib_fn_table = {
4282 : .destruct = bdev_nvme_destruct,
4283 : .submit_request = bdev_nvme_submit_request,
4284 : .io_type_supported = bdev_nvme_io_type_supported,
4285 : .get_io_channel = bdev_nvme_get_io_channel,
4286 : .dump_info_json = bdev_nvme_dump_info_json,
4287 : .write_config_json = bdev_nvme_write_config_json,
4288 : .get_spin_time = bdev_nvme_get_spin_time,
4289 : .get_module_ctx = bdev_nvme_get_module_ctx,
4290 : .get_memory_domains = bdev_nvme_get_memory_domains,
4291 : .accel_sequence_supported = bdev_nvme_accel_sequence_supported,
4292 : .reset_device_stat = bdev_nvme_reset_device_stat,
4293 : .dump_device_stat_json = bdev_nvme_dump_device_stat_json,
4294 : };
4295 :
4296 : typedef int (*bdev_nvme_parse_ana_log_page_cb)(
4297 : const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg);
4298 :
4299 : static int
4300 42 : bdev_nvme_parse_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
4301 : bdev_nvme_parse_ana_log_page_cb cb_fn, void *cb_arg)
4302 : {
4303 : struct spdk_nvme_ana_group_descriptor *copied_desc;
4304 : uint8_t *orig_desc;
4305 : uint32_t i, desc_size, copy_len;
4306 42 : int rc = 0;
4307 :
4308 42 : if (nvme_ctrlr->ana_log_page == NULL) {
4309 0 : return -EINVAL;
4310 : }
4311 :
4312 42 : copied_desc = nvme_ctrlr->copied_ana_desc;
4313 :
4314 42 : orig_desc = (uint8_t *)nvme_ctrlr->ana_log_page + sizeof(struct spdk_nvme_ana_page);
4315 42 : copy_len = nvme_ctrlr->max_ana_log_page_size - sizeof(struct spdk_nvme_ana_page);
4316 :
4317 72 : for (i = 0; i < nvme_ctrlr->ana_log_page->num_ana_group_desc; i++) {
4318 67 : memcpy(copied_desc, orig_desc, copy_len);
4319 :
4320 67 : rc = cb_fn(copied_desc, cb_arg);
4321 67 : if (rc != 0) {
4322 37 : break;
4323 : }
4324 :
4325 30 : desc_size = sizeof(struct spdk_nvme_ana_group_descriptor) +
4326 30 : copied_desc->num_of_nsid * sizeof(uint32_t);
4327 30 : orig_desc += desc_size;
4328 30 : copy_len -= desc_size;
4329 : }
4330 :
4331 42 : return rc;
4332 : }
4333 :
4334 : static int
4335 5 : nvme_ns_ana_transition_timedout(void *ctx)
4336 : {
4337 5 : struct nvme_ns *nvme_ns = ctx;
4338 :
4339 5 : spdk_poller_unregister(&nvme_ns->anatt_timer);
4340 5 : nvme_ns->ana_transition_timedout = true;
4341 :
4342 5 : return SPDK_POLLER_BUSY;
4343 : }
4344 :
4345 : static void
4346 46 : _nvme_ns_set_ana_state(struct nvme_ns *nvme_ns,
4347 : const struct spdk_nvme_ana_group_descriptor *desc)
4348 : {
4349 : const struct spdk_nvme_ctrlr_data *cdata;
4350 :
4351 46 : nvme_ns->ana_group_id = desc->ana_group_id;
4352 46 : nvme_ns->ana_state = desc->ana_state;
4353 46 : nvme_ns->ana_state_updating = false;
4354 :
4355 46 : switch (nvme_ns->ana_state) {
4356 39 : case SPDK_NVME_ANA_OPTIMIZED_STATE:
4357 : case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
4358 39 : nvme_ns->ana_transition_timedout = false;
4359 39 : spdk_poller_unregister(&nvme_ns->anatt_timer);
4360 39 : break;
4361 :
4362 6 : case SPDK_NVME_ANA_INACCESSIBLE_STATE:
4363 : case SPDK_NVME_ANA_CHANGE_STATE:
4364 6 : if (nvme_ns->anatt_timer != NULL) {
4365 1 : break;
4366 : }
4367 :
4368 5 : cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
4369 5 : nvme_ns->anatt_timer = SPDK_POLLER_REGISTER(nvme_ns_ana_transition_timedout,
4370 : nvme_ns,
4371 : cdata->anatt * SPDK_SEC_TO_USEC);
4372 5 : break;
4373 1 : default:
4374 1 : break;
4375 : }
4376 46 : }
4377 :
4378 : static int
4379 60 : nvme_ns_set_ana_state(const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg)
4380 : {
4381 60 : struct nvme_ns *nvme_ns = cb_arg;
4382 : uint32_t i;
4383 :
4384 60 : assert(nvme_ns->ns != NULL);
4385 :
4386 82 : for (i = 0; i < desc->num_of_nsid; i++) {
4387 59 : if (desc->nsid[i] != spdk_nvme_ns_get_id(nvme_ns->ns)) {
4388 22 : continue;
4389 : }
4390 :
4391 37 : _nvme_ns_set_ana_state(nvme_ns, desc);
4392 37 : return 1;
4393 : }
4394 :
4395 23 : return 0;
4396 : }
4397 :
4398 : static int
4399 5 : nvme_generate_uuid(const char *sn, uint32_t nsid, struct spdk_uuid *uuid)
4400 : {
4401 5 : int rc = 0;
4402 5 : struct spdk_uuid new_uuid, namespace_uuid;
4403 5 : char merged_str[SPDK_NVME_CTRLR_SN_LEN + NSID_STR_LEN + 1] = {'\0'};
4404 : /* This namespace UUID was generated using uuid_generate() method. */
4405 5 : const char *namespace_str = {"edaed2de-24bc-4b07-b559-f47ecbe730fd"};
4406 : int size;
4407 :
4408 5 : assert(strlen(sn) <= SPDK_NVME_CTRLR_SN_LEN);
4409 :
4410 5 : spdk_uuid_set_null(&new_uuid);
4411 5 : spdk_uuid_set_null(&namespace_uuid);
4412 :
4413 5 : size = snprintf(merged_str, sizeof(merged_str), "%s%"PRIu32, sn, nsid);
4414 5 : if (size <= 0 || (unsigned long)size >= sizeof(merged_str)) {
4415 0 : return -EINVAL;
4416 : }
4417 :
4418 5 : spdk_uuid_parse(&namespace_uuid, namespace_str);
4419 :
4420 5 : rc = spdk_uuid_generate_sha1(&new_uuid, &namespace_uuid, merged_str, size);
4421 5 : if (rc == 0) {
4422 5 : memcpy(uuid, &new_uuid, sizeof(struct spdk_uuid));
4423 : }
4424 :
4425 5 : return rc;
4426 : }
4427 :
4428 : static int
4429 39 : nvme_disk_create(struct spdk_bdev *disk, const char *base_name,
4430 : struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns,
4431 : struct spdk_bdev_nvme_ctrlr_opts *bdev_opts, void *ctx)
4432 : {
4433 : const struct spdk_uuid *uuid;
4434 : const uint8_t *nguid;
4435 : const struct spdk_nvme_ctrlr_data *cdata;
4436 : const struct spdk_nvme_ns_data *nsdata;
4437 : const struct spdk_nvme_ctrlr_opts *opts;
4438 : enum spdk_nvme_csi csi;
4439 : uint32_t atomic_bs, phys_bs, bs;
4440 39 : char sn_tmp[SPDK_NVME_CTRLR_SN_LEN + 1] = {'\0'};
4441 : int rc;
4442 :
4443 39 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
4444 39 : csi = spdk_nvme_ns_get_csi(ns);
4445 39 : opts = spdk_nvme_ctrlr_get_opts(ctrlr);
4446 :
4447 39 : switch (csi) {
4448 39 : case SPDK_NVME_CSI_NVM:
4449 39 : disk->product_name = "NVMe disk";
4450 39 : break;
4451 0 : case SPDK_NVME_CSI_ZNS:
4452 0 : disk->product_name = "NVMe ZNS disk";
4453 0 : disk->zoned = true;
4454 0 : disk->zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
4455 0 : disk->max_zone_append_size = spdk_nvme_zns_ctrlr_get_max_zone_append_size(ctrlr) /
4456 0 : spdk_nvme_ns_get_extended_sector_size(ns);
4457 0 : disk->max_open_zones = spdk_nvme_zns_ns_get_max_open_zones(ns);
4458 0 : disk->max_active_zones = spdk_nvme_zns_ns_get_max_active_zones(ns);
4459 0 : break;
4460 0 : default:
4461 0 : if (bdev_opts->allow_unrecognized_csi) {
4462 0 : disk->product_name = "NVMe Passthrough disk";
4463 0 : break;
4464 : }
4465 0 : SPDK_ERRLOG("unsupported CSI: %u\n", csi);
4466 0 : return -ENOTSUP;
4467 : }
4468 :
4469 39 : nguid = spdk_nvme_ns_get_nguid(ns);
4470 39 : if (!nguid) {
4471 39 : uuid = spdk_nvme_ns_get_uuid(ns);
4472 39 : if (uuid) {
4473 12 : disk->uuid = *uuid;
4474 27 : } else if (g_opts.generate_uuids) {
4475 0 : spdk_strcpy_pad(sn_tmp, cdata->sn, SPDK_NVME_CTRLR_SN_LEN, '\0');
4476 0 : rc = nvme_generate_uuid(sn_tmp, spdk_nvme_ns_get_id(ns), &disk->uuid);
4477 0 : if (rc < 0) {
4478 0 : SPDK_ERRLOG("UUID generation failed (%s)\n", spdk_strerror(-rc));
4479 0 : return rc;
4480 : }
4481 : }
4482 : } else {
4483 0 : memcpy(&disk->uuid, nguid, sizeof(disk->uuid));
4484 : }
4485 :
4486 39 : disk->name = spdk_sprintf_alloc("%sn%d", base_name, spdk_nvme_ns_get_id(ns));
4487 39 : if (!disk->name) {
4488 0 : return -ENOMEM;
4489 : }
4490 :
4491 39 : disk->write_cache = 0;
4492 39 : if (cdata->vwc.present) {
4493 : /* Enable if the Volatile Write Cache exists */
4494 0 : disk->write_cache = 1;
4495 : }
4496 39 : if (cdata->oncs.write_zeroes) {
4497 0 : disk->max_write_zeroes = UINT16_MAX + 1;
4498 : }
4499 39 : disk->blocklen = spdk_nvme_ns_get_extended_sector_size(ns);
4500 39 : disk->blockcnt = spdk_nvme_ns_get_num_sectors(ns);
4501 39 : disk->max_segment_size = spdk_nvme_ctrlr_get_max_xfer_size(ctrlr);
4502 39 : disk->ctratt.raw = cdata->ctratt.raw;
4503 39 : disk->nsid = spdk_nvme_ns_get_id(ns);
4504 : /* NVMe driver will split one request into multiple requests
4505 : * based on MDTS and stripe boundary, the bdev layer will use
4506 : * max_segment_size and max_num_segments to split one big IO
4507 : * into multiple requests, then small request can't run out
4508 : * of NVMe internal requests data structure.
4509 : */
4510 39 : if (opts && opts->io_queue_requests) {
4511 0 : disk->max_num_segments = opts->io_queue_requests / 2;
4512 : }
4513 39 : if (spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_SGL_SUPPORTED) {
4514 : /* The nvme driver will try to split I/O that have too many
4515 : * SGEs, but it doesn't work if that last SGE doesn't end on
4516 : * an aggregate total that is block aligned. The bdev layer has
4517 : * a more robust splitting framework, so use that instead for
4518 : * this case. (See issue #3269.)
4519 : */
4520 0 : uint16_t max_sges = spdk_nvme_ctrlr_get_max_sges(ctrlr);
4521 :
4522 0 : if (disk->max_num_segments == 0) {
4523 0 : disk->max_num_segments = max_sges;
4524 : } else {
4525 0 : disk->max_num_segments = spdk_min(disk->max_num_segments, max_sges);
4526 : }
4527 : }
4528 39 : disk->optimal_io_boundary = spdk_nvme_ns_get_optimal_io_boundary(ns);
4529 :
4530 39 : nsdata = spdk_nvme_ns_get_data(ns);
4531 39 : bs = spdk_nvme_ns_get_sector_size(ns);
4532 39 : atomic_bs = bs;
4533 39 : phys_bs = bs;
4534 39 : if (nsdata->nabo == 0) {
4535 39 : if (nsdata->nsfeat.ns_atomic_write_unit && nsdata->nawupf) {
4536 0 : atomic_bs = bs * (1 + nsdata->nawupf);
4537 : } else {
4538 39 : atomic_bs = bs * (1 + cdata->awupf);
4539 : }
4540 : }
4541 39 : if (nsdata->nsfeat.optperf) {
4542 0 : phys_bs = bs * (1 + nsdata->npwg);
4543 : }
4544 39 : disk->phys_blocklen = spdk_min(phys_bs, atomic_bs);
4545 :
4546 39 : disk->md_len = spdk_nvme_ns_get_md_size(ns);
4547 39 : if (disk->md_len != 0) {
4548 0 : disk->md_interleave = nsdata->flbas.extended;
4549 0 : disk->dif_type = (enum spdk_dif_type)spdk_nvme_ns_get_pi_type(ns);
4550 0 : if (disk->dif_type != SPDK_DIF_DISABLE) {
4551 0 : disk->dif_is_head_of_md = nsdata->dps.md_start;
4552 0 : disk->dif_check_flags = bdev_opts->prchk_flags;
4553 0 : disk->dif_pi_format = (enum spdk_dif_pi_format)spdk_nvme_ns_get_pi_format(ns);
4554 : }
4555 : }
4556 :
4557 39 : if (!(spdk_nvme_ctrlr_get_flags(ctrlr) &
4558 : SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED)) {
4559 39 : disk->acwu = 0;
4560 0 : } else if (nsdata->nsfeat.ns_atomic_write_unit) {
4561 0 : disk->acwu = nsdata->nacwu + 1; /* 0-based */
4562 : } else {
4563 0 : disk->acwu = cdata->acwu + 1; /* 0-based */
4564 : }
4565 :
4566 39 : if (cdata->oncs.copy) {
4567 : /* For now bdev interface allows only single segment copy */
4568 0 : disk->max_copy = nsdata->mssrl;
4569 : }
4570 :
4571 39 : disk->ctxt = ctx;
4572 39 : disk->fn_table = &nvmelib_fn_table;
4573 39 : disk->module = &nvme_if;
4574 :
4575 39 : disk->numa.id_valid = 1;
4576 39 : disk->numa.id = spdk_nvme_ctrlr_get_numa_id(ctrlr);
4577 :
4578 39 : return 0;
4579 : }
4580 :
4581 : static struct nvme_bdev *
4582 39 : nvme_bdev_alloc(void)
4583 : {
4584 : struct nvme_bdev *bdev;
4585 : int rc;
4586 :
4587 39 : bdev = calloc(1, sizeof(*bdev));
4588 39 : if (!bdev) {
4589 0 : SPDK_ERRLOG("bdev calloc() failed\n");
4590 0 : return NULL;
4591 : }
4592 :
4593 39 : if (g_opts.nvme_error_stat) {
4594 0 : bdev->err_stat = calloc(1, sizeof(struct nvme_error_stat));
4595 0 : if (!bdev->err_stat) {
4596 0 : SPDK_ERRLOG("err_stat calloc() failed\n");
4597 0 : free(bdev);
4598 0 : return NULL;
4599 : }
4600 : }
4601 :
4602 39 : rc = pthread_mutex_init(&bdev->mutex, NULL);
4603 39 : if (rc != 0) {
4604 0 : free(bdev->err_stat);
4605 0 : free(bdev);
4606 0 : return NULL;
4607 : }
4608 :
4609 39 : bdev->ref = 1;
4610 39 : bdev->mp_policy = BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE;
4611 39 : bdev->mp_selector = BDEV_NVME_MP_SELECTOR_ROUND_ROBIN;
4612 39 : bdev->rr_min_io = UINT32_MAX;
4613 39 : TAILQ_INIT(&bdev->nvme_ns_list);
4614 :
4615 39 : return bdev;
4616 : }
4617 :
4618 : static int
4619 39 : nvme_bdev_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4620 : {
4621 : struct nvme_bdev *bdev;
4622 39 : struct nvme_bdev_ctrlr *nbdev_ctrlr = nvme_ctrlr->nbdev_ctrlr;
4623 : int rc;
4624 :
4625 39 : bdev = nvme_bdev_alloc();
4626 39 : if (bdev == NULL) {
4627 0 : SPDK_ERRLOG("Failed to allocate NVMe bdev\n");
4628 0 : return -ENOMEM;
4629 : }
4630 :
4631 39 : bdev->opal = nvme_ctrlr->opal_dev != NULL;
4632 :
4633 39 : rc = nvme_disk_create(&bdev->disk, nbdev_ctrlr->name, nvme_ctrlr->ctrlr,
4634 : nvme_ns->ns, &nvme_ctrlr->opts, bdev);
4635 39 : if (rc != 0) {
4636 0 : SPDK_ERRLOG("Failed to create NVMe disk\n");
4637 0 : nvme_bdev_free(bdev);
4638 0 : return rc;
4639 : }
4640 :
4641 39 : spdk_io_device_register(bdev,
4642 : bdev_nvme_create_bdev_channel_cb,
4643 : bdev_nvme_destroy_bdev_channel_cb,
4644 : sizeof(struct nvme_bdev_channel),
4645 39 : bdev->disk.name);
4646 :
4647 39 : nvme_ns->bdev = bdev;
4648 39 : bdev->nsid = nvme_ns->id;
4649 39 : TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
4650 :
4651 39 : bdev->nbdev_ctrlr = nbdev_ctrlr;
4652 39 : TAILQ_INSERT_TAIL(&nbdev_ctrlr->bdevs, bdev, tailq);
4653 :
4654 39 : rc = spdk_bdev_register(&bdev->disk);
4655 39 : if (rc != 0) {
4656 1 : SPDK_ERRLOG("spdk_bdev_register() failed\n");
4657 1 : spdk_io_device_unregister(bdev, NULL);
4658 1 : nvme_ns->bdev = NULL;
4659 1 : TAILQ_REMOVE(&nbdev_ctrlr->bdevs, bdev, tailq);
4660 1 : nvme_bdev_free(bdev);
4661 1 : return rc;
4662 : }
4663 :
4664 38 : return 0;
4665 : }
4666 :
4667 : static bool
4668 23 : bdev_nvme_compare_ns(struct spdk_nvme_ns *ns1, struct spdk_nvme_ns *ns2)
4669 : {
4670 : const struct spdk_nvme_ns_data *nsdata1, *nsdata2;
4671 : const struct spdk_uuid *uuid1, *uuid2;
4672 :
4673 23 : nsdata1 = spdk_nvme_ns_get_data(ns1);
4674 23 : nsdata2 = spdk_nvme_ns_get_data(ns2);
4675 23 : uuid1 = spdk_nvme_ns_get_uuid(ns1);
4676 23 : uuid2 = spdk_nvme_ns_get_uuid(ns2);
4677 :
4678 23 : return memcmp(nsdata1->nguid, nsdata2->nguid, sizeof(nsdata1->nguid)) == 0 &&
4679 22 : nsdata1->eui64 == nsdata2->eui64 &&
4680 21 : ((uuid1 == NULL && uuid2 == NULL) ||
4681 59 : (uuid1 != NULL && uuid2 != NULL && spdk_uuid_compare(uuid1, uuid2) == 0)) &&
4682 18 : spdk_nvme_ns_get_csi(ns1) == spdk_nvme_ns_get_csi(ns2);
4683 : }
4684 :
4685 : static bool
4686 0 : hotplug_probe_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
4687 : struct spdk_nvme_ctrlr_opts *opts)
4688 : {
4689 : struct nvme_probe_skip_entry *entry;
4690 :
4691 0 : TAILQ_FOREACH(entry, &g_skipped_nvme_ctrlrs, tailq) {
4692 0 : if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
4693 0 : return false;
4694 : }
4695 : }
4696 :
4697 0 : opts->arbitration_burst = (uint8_t)g_opts.arbitration_burst;
4698 0 : opts->low_priority_weight = (uint8_t)g_opts.low_priority_weight;
4699 0 : opts->medium_priority_weight = (uint8_t)g_opts.medium_priority_weight;
4700 0 : opts->high_priority_weight = (uint8_t)g_opts.high_priority_weight;
4701 0 : opts->disable_read_ana_log_page = true;
4702 :
4703 0 : SPDK_DEBUGLOG(bdev_nvme, "Attaching to %s\n", trid->traddr);
4704 :
4705 0 : return true;
4706 : }
4707 :
4708 : static void
4709 0 : nvme_abort_cpl(void *ctx, const struct spdk_nvme_cpl *cpl)
4710 : {
4711 0 : struct nvme_ctrlr *nvme_ctrlr = ctx;
4712 :
4713 0 : if (spdk_nvme_cpl_is_error(cpl)) {
4714 0 : NVME_CTRLR_WARNLOG(nvme_ctrlr, "Abort failed. Resetting controller. sc is %u, sct is %u.\n",
4715 : cpl->status.sc, cpl->status.sct);
4716 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4717 0 : } else if (cpl->cdw0 & 0x1) {
4718 0 : NVME_CTRLR_WARNLOG(nvme_ctrlr, "Specified command could not be aborted.\n");
4719 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4720 : }
4721 0 : }
4722 :
4723 : static void
4724 0 : timeout_cb(void *cb_arg, struct spdk_nvme_ctrlr *ctrlr,
4725 : struct spdk_nvme_qpair *qpair, uint16_t cid)
4726 : {
4727 0 : struct nvme_ctrlr *nvme_ctrlr = cb_arg;
4728 : union spdk_nvme_csts_register csts;
4729 : int rc;
4730 :
4731 0 : assert(nvme_ctrlr->ctrlr == ctrlr);
4732 :
4733 0 : NVME_CTRLR_WARNLOG(nvme_ctrlr, "Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n",
4734 : ctrlr, qpair, cid);
4735 :
4736 : /* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O
4737 : * queue. (Note: qpair == NULL when there's an admin cmd timeout.) Otherwise we
4738 : * would submit another fabrics cmd on the admin queue to read CSTS and check for its
4739 : * completion recursively.
4740 : */
4741 0 : if (nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE || qpair != NULL) {
4742 0 : csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
4743 0 : if (csts.bits.cfs) {
4744 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Controller Fatal Status, reset required\n");
4745 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4746 0 : return;
4747 : }
4748 : }
4749 :
4750 0 : switch (g_opts.action_on_timeout) {
4751 0 : case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT:
4752 0 : if (qpair) {
4753 : /* Don't send abort to ctrlr when ctrlr is not available. */
4754 0 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4755 0 : if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
4756 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4757 0 : NVME_CTRLR_NOTICELOG(nvme_ctrlr, "Quit abort. Ctrlr is not available.\n");
4758 0 : return;
4759 : }
4760 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4761 :
4762 0 : rc = spdk_nvme_ctrlr_cmd_abort(ctrlr, qpair, cid,
4763 : nvme_abort_cpl, nvme_ctrlr);
4764 0 : if (rc == 0) {
4765 0 : return;
4766 : }
4767 :
4768 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Unable to send abort. Resetting, rc is %d.\n", rc);
4769 : }
4770 :
4771 : /* FALLTHROUGH */
4772 : case SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET:
4773 0 : bdev_nvme_reset_ctrlr(nvme_ctrlr);
4774 0 : break;
4775 0 : case SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE:
4776 0 : NVME_CTRLR_DEBUGLOG(nvme_ctrlr, "No action for nvme controller timeout.\n");
4777 0 : break;
4778 0 : default:
4779 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "An invalid timeout action value is found.\n");
4780 0 : break;
4781 : }
4782 : }
4783 :
4784 : static struct nvme_ns *
4785 52 : nvme_ns_alloc(void)
4786 : {
4787 : struct nvme_ns *nvme_ns;
4788 :
4789 52 : nvme_ns = calloc(1, sizeof(struct nvme_ns));
4790 52 : if (nvme_ns == NULL) {
4791 0 : return NULL;
4792 : }
4793 :
4794 52 : if (g_opts.io_path_stat) {
4795 0 : nvme_ns->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
4796 0 : if (nvme_ns->stat == NULL) {
4797 0 : free(nvme_ns);
4798 0 : return NULL;
4799 : }
4800 0 : spdk_bdev_reset_io_stat(nvme_ns->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
4801 : }
4802 :
4803 52 : return nvme_ns;
4804 : }
4805 :
4806 : static void
4807 52 : nvme_ns_free(struct nvme_ns *nvme_ns)
4808 : {
4809 52 : free(nvme_ns->stat);
4810 52 : free(nvme_ns);
4811 52 : }
4812 :
4813 : static void
4814 52 : nvme_ctrlr_populate_namespace_done(struct nvme_ns *nvme_ns, int rc)
4815 : {
4816 52 : struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
4817 52 : struct nvme_async_probe_ctx *ctx = nvme_ns->probe_ctx;
4818 :
4819 52 : if (rc == 0) {
4820 50 : nvme_ns->probe_ctx = NULL;
4821 50 : nvme_ctrlr_get_ref(nvme_ctrlr);
4822 : } else {
4823 2 : RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4824 2 : nvme_ns_free(nvme_ns);
4825 : }
4826 :
4827 52 : if (ctx) {
4828 51 : ctx->populates_in_progress--;
4829 51 : if (ctx->populates_in_progress == 0) {
4830 12 : nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
4831 : }
4832 : }
4833 52 : }
4834 :
4835 : static void
4836 2 : bdev_nvme_add_io_path(struct nvme_bdev_channel_iter *i,
4837 : struct nvme_bdev *nbdev,
4838 : struct nvme_bdev_channel *nbdev_ch, void *ctx)
4839 : {
4840 2 : struct nvme_ns *nvme_ns = ctx;
4841 : int rc;
4842 :
4843 2 : rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
4844 2 : if (rc != 0) {
4845 0 : SPDK_ERRLOG("Failed to add I/O path to bdev_channel dynamically.\n");
4846 : }
4847 :
4848 2 : nvme_bdev_for_each_channel_continue(i, rc);
4849 2 : }
4850 :
4851 : static void
4852 2 : bdev_nvme_delete_io_path(struct nvme_bdev_channel_iter *i,
4853 : struct nvme_bdev *nbdev,
4854 : struct nvme_bdev_channel *nbdev_ch, void *ctx)
4855 : {
4856 2 : struct nvme_ns *nvme_ns = ctx;
4857 : struct nvme_io_path *io_path;
4858 :
4859 2 : io_path = _bdev_nvme_get_io_path(nbdev_ch, nvme_ns);
4860 2 : if (io_path != NULL) {
4861 2 : _bdev_nvme_delete_io_path(nbdev_ch, io_path);
4862 : }
4863 :
4864 2 : nvme_bdev_for_each_channel_continue(i, 0);
4865 2 : }
4866 :
4867 : static void
4868 0 : bdev_nvme_add_io_path_failed(struct nvme_bdev *nbdev, void *ctx, int status)
4869 : {
4870 0 : struct nvme_ns *nvme_ns = ctx;
4871 :
4872 0 : nvme_ctrlr_populate_namespace_done(nvme_ns, -1);
4873 0 : }
4874 :
4875 : static void
4876 12 : bdev_nvme_add_io_path_done(struct nvme_bdev *nbdev, void *ctx, int status)
4877 : {
4878 12 : struct nvme_ns *nvme_ns = ctx;
4879 :
4880 12 : if (status == 0) {
4881 12 : nvme_ctrlr_populate_namespace_done(nvme_ns, 0);
4882 : } else {
4883 : /* Delete the added io_paths and fail populating the namespace. */
4884 0 : nvme_bdev_for_each_channel(nbdev,
4885 : bdev_nvme_delete_io_path,
4886 : nvme_ns,
4887 : bdev_nvme_add_io_path_failed);
4888 : }
4889 12 : }
4890 :
4891 : static int
4892 13 : nvme_bdev_add_ns(struct nvme_bdev *bdev, struct nvme_ns *nvme_ns)
4893 : {
4894 : struct nvme_ns *tmp_ns;
4895 : const struct spdk_nvme_ns_data *nsdata;
4896 :
4897 13 : nsdata = spdk_nvme_ns_get_data(nvme_ns->ns);
4898 13 : if (!nsdata->nmic.can_share) {
4899 0 : SPDK_ERRLOG("Namespace cannot be shared.\n");
4900 0 : return -EINVAL;
4901 : }
4902 :
4903 13 : pthread_mutex_lock(&bdev->mutex);
4904 :
4905 13 : tmp_ns = TAILQ_FIRST(&bdev->nvme_ns_list);
4906 13 : assert(tmp_ns != NULL);
4907 :
4908 13 : if (tmp_ns->ns != NULL && !bdev_nvme_compare_ns(nvme_ns->ns, tmp_ns->ns)) {
4909 1 : pthread_mutex_unlock(&bdev->mutex);
4910 1 : SPDK_ERRLOG("Namespaces are not identical.\n");
4911 1 : return -EINVAL;
4912 : }
4913 :
4914 12 : bdev->ref++;
4915 12 : TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
4916 12 : nvme_ns->bdev = bdev;
4917 :
4918 12 : pthread_mutex_unlock(&bdev->mutex);
4919 :
4920 : /* Add nvme_io_path to nvme_bdev_channels dynamically. */
4921 12 : nvme_bdev_for_each_channel(bdev,
4922 : bdev_nvme_add_io_path,
4923 : nvme_ns,
4924 : bdev_nvme_add_io_path_done);
4925 :
4926 12 : return 0;
4927 : }
4928 :
4929 : static void
4930 52 : nvme_ctrlr_populate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4931 : {
4932 : struct spdk_nvme_ns *ns;
4933 : struct nvme_bdev *bdev;
4934 52 : int rc = 0;
4935 :
4936 52 : ns = spdk_nvme_ctrlr_get_ns(nvme_ctrlr->ctrlr, nvme_ns->id);
4937 52 : if (!ns) {
4938 0 : NVME_CTRLR_DEBUGLOG(nvme_ctrlr, "Invalid NS %d\n", nvme_ns->id);
4939 0 : rc = -EINVAL;
4940 0 : goto done;
4941 : }
4942 :
4943 52 : nvme_ns->ns = ns;
4944 52 : nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
4945 :
4946 52 : if (nvme_ctrlr->ana_log_page != NULL) {
4947 38 : bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ns_set_ana_state, nvme_ns);
4948 : }
4949 :
4950 52 : bdev = nvme_bdev_ctrlr_get_bdev(nvme_ctrlr->nbdev_ctrlr, nvme_ns->id);
4951 52 : if (bdev == NULL) {
4952 39 : rc = nvme_bdev_create(nvme_ctrlr, nvme_ns);
4953 : } else {
4954 13 : rc = nvme_bdev_add_ns(bdev, nvme_ns);
4955 13 : if (rc == 0) {
4956 12 : return;
4957 : }
4958 : }
4959 1 : done:
4960 40 : nvme_ctrlr_populate_namespace_done(nvme_ns, rc);
4961 : }
4962 :
4963 : static void
4964 50 : nvme_ctrlr_depopulate_namespace_done(struct nvme_ns *nvme_ns)
4965 : {
4966 50 : struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
4967 :
4968 50 : assert(nvme_ctrlr != NULL);
4969 :
4970 50 : pthread_mutex_lock(&nvme_ctrlr->mutex);
4971 :
4972 50 : RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4973 :
4974 50 : if (nvme_ns->bdev != NULL) {
4975 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4976 0 : return;
4977 : }
4978 :
4979 50 : nvme_ns_free(nvme_ns);
4980 50 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
4981 :
4982 50 : nvme_ctrlr_put_ref(nvme_ctrlr);
4983 : }
4984 :
4985 : static void
4986 11 : bdev_nvme_delete_io_path_done(struct nvme_bdev *nbdev, void *ctx, int status)
4987 : {
4988 11 : struct nvme_ns *nvme_ns = ctx;
4989 :
4990 11 : nvme_ctrlr_depopulate_namespace_done(nvme_ns);
4991 11 : }
4992 :
4993 : static void
4994 50 : nvme_ctrlr_depopulate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4995 : {
4996 : struct nvme_bdev *bdev;
4997 :
4998 50 : spdk_poller_unregister(&nvme_ns->anatt_timer);
4999 :
5000 50 : bdev = nvme_ns->bdev;
5001 50 : if (bdev != NULL) {
5002 46 : pthread_mutex_lock(&bdev->mutex);
5003 :
5004 46 : assert(bdev->ref > 0);
5005 46 : bdev->ref--;
5006 46 : if (bdev->ref == 0) {
5007 35 : pthread_mutex_unlock(&bdev->mutex);
5008 :
5009 35 : spdk_bdev_unregister(&bdev->disk, NULL, NULL);
5010 : } else {
5011 : /* spdk_bdev_unregister() is not called until the last nvme_ns is
5012 : * depopulated. Hence we need to remove nvme_ns from bdev->nvme_ns_list
5013 : * and clear nvme_ns->bdev here.
5014 : */
5015 11 : TAILQ_REMOVE(&bdev->nvme_ns_list, nvme_ns, tailq);
5016 11 : nvme_ns->bdev = NULL;
5017 :
5018 11 : pthread_mutex_unlock(&bdev->mutex);
5019 :
5020 : /* Delete nvme_io_paths from nvme_bdev_channels dynamically. After that,
5021 : * we call depopulate_namespace_done() to avoid use-after-free.
5022 : */
5023 11 : nvme_bdev_for_each_channel(bdev,
5024 : bdev_nvme_delete_io_path,
5025 : nvme_ns,
5026 : bdev_nvme_delete_io_path_done);
5027 11 : return;
5028 : }
5029 : }
5030 :
5031 39 : nvme_ctrlr_depopulate_namespace_done(nvme_ns);
5032 : }
5033 :
5034 : static void
5035 63 : nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
5036 : struct nvme_async_probe_ctx *ctx)
5037 : {
5038 63 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5039 : struct nvme_ns *nvme_ns, *next;
5040 : struct spdk_nvme_ns *ns;
5041 : struct nvme_bdev *bdev;
5042 : uint32_t nsid;
5043 : int rc;
5044 : uint64_t num_sectors;
5045 :
5046 63 : if (ctx) {
5047 : /* Initialize this count to 1 to handle the populate functions
5048 : * calling nvme_ctrlr_populate_namespace_done() immediately.
5049 : */
5050 47 : ctx->populates_in_progress = 1;
5051 : }
5052 :
5053 : /* First loop over our existing namespaces and see if they have been
5054 : * removed. */
5055 63 : nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
5056 67 : while (nvme_ns != NULL) {
5057 4 : next = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
5058 :
5059 4 : if (spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
5060 : /* NS is still there or added again. Its attributes may have changed. */
5061 3 : ns = spdk_nvme_ctrlr_get_ns(ctrlr, nvme_ns->id);
5062 3 : if (nvme_ns->ns != ns) {
5063 1 : assert(nvme_ns->ns == NULL);
5064 1 : nvme_ns->ns = ns;
5065 1 : NVME_CTRLR_DEBUGLOG(nvme_ctrlr, "NSID %u was added\n", nvme_ns->id);
5066 : }
5067 :
5068 3 : num_sectors = spdk_nvme_ns_get_num_sectors(ns);
5069 3 : bdev = nvme_ns->bdev;
5070 3 : assert(bdev != NULL);
5071 3 : if (bdev->disk.blockcnt != num_sectors) {
5072 1 : NVME_CTRLR_NOTICELOG(nvme_ctrlr,
5073 : "NSID %u is resized: bdev name %s, old size %" PRIu64 ", new size %" PRIu64 "\n",
5074 : nvme_ns->id,
5075 : bdev->disk.name,
5076 : bdev->disk.blockcnt,
5077 : num_sectors);
5078 1 : rc = spdk_bdev_notify_blockcnt_change(&bdev->disk, num_sectors);
5079 1 : if (rc != 0) {
5080 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr,
5081 : "Could not change num blocks for nvme bdev: name %s, errno: %d.\n",
5082 : bdev->disk.name, rc);
5083 : }
5084 : }
5085 : } else {
5086 : /* Namespace was removed */
5087 1 : nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
5088 : }
5089 :
5090 4 : nvme_ns = next;
5091 : }
5092 :
5093 : /* Loop through all of the namespaces at the nvme level and see if any of them are new */
5094 63 : nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
5095 118 : while (nsid != 0) {
5096 55 : nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);
5097 :
5098 55 : if (nvme_ns == NULL) {
5099 : /* Found a new one */
5100 52 : nvme_ns = nvme_ns_alloc();
5101 52 : if (nvme_ns == NULL) {
5102 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to allocate namespace\n");
5103 : /* This just fails to attach the namespace. It may work on a future attempt. */
5104 0 : continue;
5105 : }
5106 :
5107 52 : nvme_ns->id = nsid;
5108 52 : nvme_ns->ctrlr = nvme_ctrlr;
5109 :
5110 52 : nvme_ns->bdev = NULL;
5111 :
5112 52 : if (ctx) {
5113 51 : ctx->populates_in_progress++;
5114 : }
5115 52 : nvme_ns->probe_ctx = ctx;
5116 :
5117 52 : RB_INSERT(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
5118 :
5119 52 : nvme_ctrlr_populate_namespace(nvme_ctrlr, nvme_ns);
5120 : }
5121 :
5122 55 : nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid);
5123 : }
5124 :
5125 63 : if (ctx) {
5126 : /* Decrement this count now that the loop is over to account
5127 : * for the one we started with. If the count is then 0, we
5128 : * know any populate_namespace functions completed immediately,
5129 : * so we'll kick the callback here.
5130 : */
5131 47 : ctx->populates_in_progress--;
5132 47 : if (ctx->populates_in_progress == 0) {
5133 35 : nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
5134 : }
5135 : }
5136 :
5137 63 : }
5138 :
5139 : static void
5140 62 : nvme_ctrlr_depopulate_namespaces(struct nvme_ctrlr *nvme_ctrlr)
5141 : {
5142 : struct nvme_ns *nvme_ns, *tmp;
5143 :
5144 111 : RB_FOREACH_SAFE(nvme_ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp) {
5145 49 : nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
5146 : }
5147 62 : }
5148 :
5149 : static uint32_t
5150 37 : nvme_ctrlr_get_ana_log_page_size(struct nvme_ctrlr *nvme_ctrlr)
5151 : {
5152 37 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5153 : const struct spdk_nvme_ctrlr_data *cdata;
5154 37 : uint32_t nsid, ns_count = 0;
5155 :
5156 37 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5157 :
5158 37 : for (nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
5159 82 : nsid != 0; nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid)) {
5160 45 : ns_count++;
5161 : }
5162 :
5163 37 : return sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
5164 37 : sizeof(struct spdk_nvme_ana_group_descriptor) + ns_count *
5165 : sizeof(uint32_t);
5166 : }
5167 :
5168 : static int
5169 7 : nvme_ctrlr_set_ana_states(const struct spdk_nvme_ana_group_descriptor *desc,
5170 : void *cb_arg)
5171 : {
5172 7 : struct nvme_ctrlr *nvme_ctrlr = cb_arg;
5173 : struct nvme_ns *nvme_ns;
5174 : uint32_t i, nsid;
5175 :
5176 13 : for (i = 0; i < desc->num_of_nsid; i++) {
5177 6 : nsid = desc->nsid[i];
5178 6 : if (nsid == 0) {
5179 0 : continue;
5180 : }
5181 :
5182 6 : nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);
5183 :
5184 6 : if (nvme_ns == NULL) {
5185 : /* Target told us that an inactive namespace had an ANA change */
5186 1 : continue;
5187 : }
5188 :
5189 5 : _nvme_ns_set_ana_state(nvme_ns, desc);
5190 : }
5191 :
5192 7 : return 0;
5193 : }
5194 :
5195 : static void
5196 0 : bdev_nvme_disable_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
5197 : {
5198 : struct nvme_ns *nvme_ns;
5199 :
5200 0 : spdk_free(nvme_ctrlr->ana_log_page);
5201 0 : nvme_ctrlr->ana_log_page = NULL;
5202 :
5203 0 : for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
5204 0 : nvme_ns != NULL;
5205 0 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
5206 0 : nvme_ns->ana_state_updating = false;
5207 0 : nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
5208 : }
5209 0 : }
5210 :
5211 : static void
5212 3 : nvme_ctrlr_read_ana_log_page_done(void *ctx, const struct spdk_nvme_cpl *cpl)
5213 : {
5214 3 : struct nvme_ctrlr *nvme_ctrlr = ctx;
5215 :
5216 3 : if (cpl != NULL && spdk_nvme_cpl_is_success(cpl)) {
5217 3 : bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ctrlr_set_ana_states,
5218 : nvme_ctrlr);
5219 : } else {
5220 0 : bdev_nvme_disable_read_ana_log_page(nvme_ctrlr);
5221 : }
5222 :
5223 3 : pthread_mutex_lock(&nvme_ctrlr->mutex);
5224 :
5225 3 : assert(nvme_ctrlr->ana_log_page_updating == true);
5226 3 : nvme_ctrlr->ana_log_page_updating = false;
5227 :
5228 3 : if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
5229 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
5230 :
5231 0 : nvme_ctrlr_unregister(nvme_ctrlr);
5232 : } else {
5233 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
5234 :
5235 3 : bdev_nvme_clear_io_path_caches(nvme_ctrlr);
5236 : }
5237 3 : }
5238 :
5239 : static int
5240 6 : nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
5241 : {
5242 : uint32_t ana_log_page_size;
5243 : int rc;
5244 :
5245 6 : if (nvme_ctrlr->ana_log_page == NULL) {
5246 0 : return -EINVAL;
5247 : }
5248 :
5249 6 : ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);
5250 :
5251 6 : if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
5252 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr,
5253 : "ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
5254 : ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
5255 0 : return -EINVAL;
5256 : }
5257 :
5258 6 : pthread_mutex_lock(&nvme_ctrlr->mutex);
5259 6 : if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
5260 : nvme_ctrlr->ana_log_page_updating) {
5261 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
5262 3 : return -EBUSY;
5263 : }
5264 :
5265 3 : nvme_ctrlr->ana_log_page_updating = true;
5266 3 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
5267 :
5268 3 : rc = spdk_nvme_ctrlr_cmd_get_log_page(nvme_ctrlr->ctrlr,
5269 : SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
5270 : SPDK_NVME_GLOBAL_NS_TAG,
5271 3 : nvme_ctrlr->ana_log_page,
5272 : ana_log_page_size, 0,
5273 : nvme_ctrlr_read_ana_log_page_done,
5274 : nvme_ctrlr);
5275 3 : if (rc != 0) {
5276 0 : nvme_ctrlr_read_ana_log_page_done(nvme_ctrlr, NULL);
5277 : }
5278 :
5279 3 : return rc;
5280 : }
5281 :
5282 : static void
5283 0 : dummy_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5284 : {
5285 0 : }
5286 :
5287 : struct bdev_nvme_set_preferred_path_ctx {
5288 : struct spdk_bdev_desc *desc;
5289 : struct nvme_ns *nvme_ns;
5290 : bdev_nvme_set_preferred_path_cb cb_fn;
5291 : void *cb_arg;
5292 : };
5293 :
5294 : static void
5295 3 : bdev_nvme_set_preferred_path_done(struct nvme_bdev *nbdev, void *_ctx, int status)
5296 : {
5297 3 : struct bdev_nvme_set_preferred_path_ctx *ctx = _ctx;
5298 :
5299 3 : assert(ctx != NULL);
5300 3 : assert(ctx->desc != NULL);
5301 3 : assert(ctx->cb_fn != NULL);
5302 :
5303 3 : spdk_bdev_close(ctx->desc);
5304 :
5305 3 : ctx->cb_fn(ctx->cb_arg, status);
5306 :
5307 3 : free(ctx);
5308 3 : }
5309 :
5310 : static void
5311 2 : _bdev_nvme_set_preferred_path(struct nvme_bdev_channel_iter *i,
5312 : struct nvme_bdev *nbdev,
5313 : struct nvme_bdev_channel *nbdev_ch, void *_ctx)
5314 : {
5315 2 : struct bdev_nvme_set_preferred_path_ctx *ctx = _ctx;
5316 : struct nvme_io_path *io_path, *prev;
5317 :
5318 2 : prev = NULL;
5319 3 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
5320 3 : if (io_path->nvme_ns == ctx->nvme_ns) {
5321 2 : break;
5322 : }
5323 1 : prev = io_path;
5324 : }
5325 :
5326 2 : if (io_path != NULL) {
5327 2 : if (prev != NULL) {
5328 1 : STAILQ_REMOVE_AFTER(&nbdev_ch->io_path_list, prev, stailq);
5329 1 : STAILQ_INSERT_HEAD(&nbdev_ch->io_path_list, io_path, stailq);
5330 : }
5331 :
5332 : /* We can set io_path to nbdev_ch->current_io_path directly here.
5333 : * However, it needs to be conditional. To simplify the code,
5334 : * just clear nbdev_ch->current_io_path and let find_io_path()
5335 : * fill it.
5336 : *
5337 : * Automatic failback may be disabled. Hence even if the io_path is
5338 : * already at the head, clear nbdev_ch->current_io_path.
5339 : */
5340 2 : bdev_nvme_clear_current_io_path(nbdev_ch);
5341 : }
5342 :
5343 2 : nvme_bdev_for_each_channel_continue(i, 0);
5344 2 : }
5345 :
5346 : static struct nvme_ns *
5347 3 : bdev_nvme_set_preferred_ns(struct nvme_bdev *nbdev, uint16_t cntlid)
5348 : {
5349 : struct nvme_ns *nvme_ns, *prev;
5350 : const struct spdk_nvme_ctrlr_data *cdata;
5351 :
5352 3 : prev = NULL;
5353 6 : TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
5354 6 : cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
5355 :
5356 6 : if (cdata->cntlid == cntlid) {
5357 3 : break;
5358 : }
5359 3 : prev = nvme_ns;
5360 : }
5361 :
5362 3 : if (nvme_ns != NULL && prev != NULL) {
5363 2 : TAILQ_REMOVE(&nbdev->nvme_ns_list, nvme_ns, tailq);
5364 2 : TAILQ_INSERT_HEAD(&nbdev->nvme_ns_list, nvme_ns, tailq);
5365 : }
5366 :
5367 3 : return nvme_ns;
5368 : }
5369 :
5370 : /* This function supports only multipath mode. There is only a single I/O path
5371 : * for each NVMe-oF controller. Hence, just move the matched I/O path to the
5372 : * head of the I/O path list for each NVMe bdev channel.
5373 : *
5374 : * NVMe bdev channel may be acquired after completing this function. move the
5375 : * matched namespace to the head of the namespace list for the NVMe bdev too.
5376 : */
5377 : void
5378 3 : bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid,
5379 : bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg)
5380 : {
5381 : struct bdev_nvme_set_preferred_path_ctx *ctx;
5382 : struct spdk_bdev *bdev;
5383 : struct nvme_bdev *nbdev;
5384 3 : int rc = 0;
5385 :
5386 3 : assert(cb_fn != NULL);
5387 :
5388 3 : ctx = calloc(1, sizeof(*ctx));
5389 3 : if (ctx == NULL) {
5390 0 : SPDK_ERRLOG("Failed to alloc context.\n");
5391 0 : rc = -ENOMEM;
5392 0 : goto err_alloc;
5393 : }
5394 :
5395 3 : ctx->cb_fn = cb_fn;
5396 3 : ctx->cb_arg = cb_arg;
5397 :
5398 3 : rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
5399 3 : if (rc != 0) {
5400 0 : SPDK_ERRLOG("Failed to open bdev %s.\n", name);
5401 0 : goto err_open;
5402 : }
5403 :
5404 3 : bdev = spdk_bdev_desc_get_bdev(ctx->desc);
5405 :
5406 3 : if (bdev->module != &nvme_if) {
5407 0 : SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
5408 0 : rc = -ENODEV;
5409 0 : goto err_bdev;
5410 : }
5411 :
5412 3 : nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
5413 :
5414 3 : pthread_mutex_lock(&nbdev->mutex);
5415 :
5416 3 : ctx->nvme_ns = bdev_nvme_set_preferred_ns(nbdev, cntlid);
5417 3 : if (ctx->nvme_ns == NULL) {
5418 0 : pthread_mutex_unlock(&nbdev->mutex);
5419 :
5420 0 : SPDK_ERRLOG("bdev %s does not have namespace to controller %u.\n", name, cntlid);
5421 0 : rc = -ENODEV;
5422 0 : goto err_bdev;
5423 : }
5424 :
5425 3 : pthread_mutex_unlock(&nbdev->mutex);
5426 :
5427 3 : nvme_bdev_for_each_channel(nbdev,
5428 : _bdev_nvme_set_preferred_path,
5429 : ctx,
5430 : bdev_nvme_set_preferred_path_done);
5431 3 : return;
5432 :
5433 0 : err_bdev:
5434 0 : spdk_bdev_close(ctx->desc);
5435 0 : err_open:
5436 0 : free(ctx);
5437 0 : err_alloc:
5438 0 : cb_fn(cb_arg, rc);
5439 : }
5440 :
5441 : struct bdev_nvme_set_multipath_policy_ctx {
5442 : struct spdk_bdev_desc *desc;
5443 : spdk_bdev_nvme_set_multipath_policy_cb cb_fn;
5444 : void *cb_arg;
5445 : };
5446 :
5447 : static void
5448 3 : bdev_nvme_set_multipath_policy_done(struct nvme_bdev *nbdev, void *_ctx, int status)
5449 : {
5450 3 : struct bdev_nvme_set_multipath_policy_ctx *ctx = _ctx;
5451 :
5452 3 : assert(ctx != NULL);
5453 3 : assert(ctx->desc != NULL);
5454 3 : assert(ctx->cb_fn != NULL);
5455 :
5456 3 : spdk_bdev_close(ctx->desc);
5457 :
5458 3 : ctx->cb_fn(ctx->cb_arg, status);
5459 :
5460 3 : free(ctx);
5461 3 : }
5462 :
5463 : static void
5464 1 : _bdev_nvme_set_multipath_policy(struct nvme_bdev_channel_iter *i,
5465 : struct nvme_bdev *nbdev,
5466 : struct nvme_bdev_channel *nbdev_ch, void *ctx)
5467 : {
5468 1 : nbdev_ch->mp_policy = nbdev->mp_policy;
5469 1 : nbdev_ch->mp_selector = nbdev->mp_selector;
5470 1 : nbdev_ch->rr_min_io = nbdev->rr_min_io;
5471 1 : bdev_nvme_clear_current_io_path(nbdev_ch);
5472 :
5473 1 : nvme_bdev_for_each_channel_continue(i, 0);
5474 1 : }
5475 :
5476 : void
5477 3 : spdk_bdev_nvme_set_multipath_policy(const char *name, enum spdk_bdev_nvme_multipath_policy policy,
5478 : enum spdk_bdev_nvme_multipath_selector selector, uint32_t rr_min_io,
5479 : spdk_bdev_nvme_set_multipath_policy_cb cb_fn, void *cb_arg)
5480 : {
5481 : struct bdev_nvme_set_multipath_policy_ctx *ctx;
5482 : struct spdk_bdev *bdev;
5483 : struct nvme_bdev *nbdev;
5484 : int rc;
5485 :
5486 3 : assert(cb_fn != NULL);
5487 :
5488 3 : switch (policy) {
5489 1 : case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
5490 1 : break;
5491 2 : case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
5492 2 : switch (selector) {
5493 1 : case BDEV_NVME_MP_SELECTOR_ROUND_ROBIN:
5494 1 : if (rr_min_io == UINT32_MAX) {
5495 0 : rr_min_io = 1;
5496 1 : } else if (rr_min_io == 0) {
5497 0 : rc = -EINVAL;
5498 0 : goto exit;
5499 : }
5500 1 : break;
5501 1 : case BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH:
5502 1 : break;
5503 0 : default:
5504 0 : rc = -EINVAL;
5505 0 : goto exit;
5506 : }
5507 2 : break;
5508 0 : default:
5509 0 : rc = -EINVAL;
5510 0 : goto exit;
5511 : }
5512 :
5513 3 : ctx = calloc(1, sizeof(*ctx));
5514 3 : if (ctx == NULL) {
5515 0 : SPDK_ERRLOG("Failed to alloc context.\n");
5516 0 : rc = -ENOMEM;
5517 0 : goto exit;
5518 : }
5519 :
5520 3 : ctx->cb_fn = cb_fn;
5521 3 : ctx->cb_arg = cb_arg;
5522 :
5523 3 : rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
5524 3 : if (rc != 0) {
5525 0 : SPDK_ERRLOG("Failed to open bdev %s.\n", name);
5526 0 : rc = -ENODEV;
5527 0 : goto err_open;
5528 : }
5529 :
5530 3 : bdev = spdk_bdev_desc_get_bdev(ctx->desc);
5531 3 : if (bdev->module != &nvme_if) {
5532 0 : SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
5533 0 : rc = -ENODEV;
5534 0 : goto err_module;
5535 : }
5536 3 : nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
5537 :
5538 3 : pthread_mutex_lock(&nbdev->mutex);
5539 3 : nbdev->mp_policy = policy;
5540 3 : nbdev->mp_selector = selector;
5541 3 : nbdev->rr_min_io = rr_min_io;
5542 3 : pthread_mutex_unlock(&nbdev->mutex);
5543 :
5544 3 : nvme_bdev_for_each_channel(nbdev,
5545 : _bdev_nvme_set_multipath_policy,
5546 : ctx,
5547 : bdev_nvme_set_multipath_policy_done);
5548 3 : return;
5549 :
5550 0 : err_module:
5551 0 : spdk_bdev_close(ctx->desc);
5552 0 : err_open:
5553 0 : free(ctx);
5554 0 : exit:
5555 0 : cb_fn(cb_arg, rc);
5556 : }
5557 :
5558 : static void
5559 3 : aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
5560 : {
5561 3 : struct nvme_ctrlr *nvme_ctrlr = arg;
5562 : union spdk_nvme_async_event_completion event;
5563 :
5564 3 : if (spdk_nvme_cpl_is_error(cpl)) {
5565 0 : SPDK_WARNLOG("AER request execute failed\n");
5566 0 : return;
5567 : }
5568 :
5569 3 : event.raw = cpl->cdw0;
5570 3 : if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
5571 3 : (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) {
5572 2 : nvme_ctrlr_populate_namespaces(nvme_ctrlr, NULL);
5573 1 : } else if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
5574 1 : (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_ANA_CHANGE)) {
5575 1 : nvme_ctrlr_read_ana_log_page(nvme_ctrlr);
5576 : }
5577 : }
5578 :
5579 : static void
5580 53 : free_nvme_async_probe_ctx(struct nvme_async_probe_ctx *ctx)
5581 : {
5582 53 : spdk_keyring_put_key(ctx->drv_opts.tls_psk);
5583 53 : spdk_keyring_put_key(ctx->drv_opts.dhchap_key);
5584 53 : spdk_keyring_put_key(ctx->drv_opts.dhchap_ctrlr_key);
5585 53 : free(ctx);
5586 53 : }
5587 :
5588 : static void
5589 53 : populate_namespaces_cb(struct nvme_async_probe_ctx *ctx, int rc)
5590 : {
5591 53 : if (ctx->cb_fn) {
5592 53 : ctx->cb_fn(ctx->cb_ctx, ctx->reported_bdevs, rc);
5593 : }
5594 :
5595 53 : ctx->namespaces_populated = true;
5596 53 : if (ctx->probe_done) {
5597 : /* The probe was already completed, so we need to free the context
5598 : * here. This can happen for cases like OCSSD, where we need to
5599 : * send additional commands to the SSD after attach.
5600 : */
5601 32 : free_nvme_async_probe_ctx(ctx);
5602 : }
5603 53 : }
5604 :
5605 : static int
5606 20 : bdev_nvme_remove_poller(void *ctx)
5607 : {
5608 20 : struct spdk_nvme_transport_id trid_pcie;
5609 :
5610 20 : if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
5611 1 : spdk_poller_unregister(&g_hotplug_poller);
5612 1 : return SPDK_POLLER_IDLE;
5613 : }
5614 :
5615 19 : memset(&trid_pcie, 0, sizeof(trid_pcie));
5616 19 : spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);
5617 :
5618 19 : if (spdk_nvme_scan_attached(&trid_pcie)) {
5619 0 : SPDK_ERRLOG_RATELIMIT("spdk_nvme_scan_attached() failed\n");
5620 : }
5621 :
5622 19 : return SPDK_POLLER_BUSY;
5623 : }
5624 :
5625 : static void
5626 61 : nvme_ctrlr_create_done(struct nvme_ctrlr *nvme_ctrlr,
5627 : struct nvme_async_probe_ctx *ctx)
5628 : {
5629 61 : struct spdk_nvme_transport_id *trid = &nvme_ctrlr->active_path_id->trid;
5630 :
5631 61 : if (spdk_nvme_trtype_is_fabrics(trid->trtype)) {
5632 61 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "ctrlr was created to %s:%s\n",
5633 : trid->traddr, trid->trsvcid);
5634 : } else {
5635 0 : NVME_CTRLR_INFOLOG(nvme_ctrlr, "ctrlr was created\n");
5636 : }
5637 :
5638 61 : spdk_io_device_register(nvme_ctrlr,
5639 : bdev_nvme_create_ctrlr_channel_cb,
5640 : bdev_nvme_destroy_ctrlr_channel_cb,
5641 : sizeof(struct nvme_ctrlr_channel),
5642 61 : nvme_ctrlr->nbdev_ctrlr->name);
5643 :
5644 61 : nvme_ctrlr_populate_namespaces(nvme_ctrlr, ctx);
5645 :
5646 61 : if (g_hotplug_poller == NULL) {
5647 2 : g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_remove_poller, NULL,
5648 : NVME_HOTPLUG_POLL_PERIOD_DEFAULT);
5649 : }
5650 61 : }
5651 :
5652 : static void
5653 31 : nvme_ctrlr_init_ana_log_page_done(void *_ctx, const struct spdk_nvme_cpl *cpl)
5654 : {
5655 31 : struct nvme_ctrlr *nvme_ctrlr = _ctx;
5656 31 : struct nvme_async_probe_ctx *ctx = nvme_ctrlr->probe_ctx;
5657 :
5658 31 : nvme_ctrlr->probe_ctx = NULL;
5659 :
5660 31 : if (spdk_nvme_cpl_is_error(cpl)) {
5661 0 : nvme_ctrlr_delete(nvme_ctrlr);
5662 :
5663 0 : if (ctx != NULL) {
5664 0 : ctx->reported_bdevs = 0;
5665 0 : populate_namespaces_cb(ctx, -1);
5666 : }
5667 0 : return;
5668 : }
5669 :
5670 31 : nvme_ctrlr_create_done(nvme_ctrlr, ctx);
5671 : }
5672 :
5673 : static int
5674 31 : nvme_ctrlr_init_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
5675 : struct nvme_async_probe_ctx *ctx)
5676 : {
5677 31 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5678 : const struct spdk_nvme_ctrlr_data *cdata;
5679 : uint32_t ana_log_page_size;
5680 :
5681 31 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5682 :
5683 : /* Set buffer size enough to include maximum number of allowed namespaces. */
5684 31 : ana_log_page_size = sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
5685 31 : sizeof(struct spdk_nvme_ana_group_descriptor) + cdata->mnan *
5686 : sizeof(uint32_t);
5687 :
5688 31 : nvme_ctrlr->ana_log_page = spdk_zmalloc(ana_log_page_size, 64, NULL,
5689 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
5690 31 : if (nvme_ctrlr->ana_log_page == NULL) {
5691 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "could not allocate ANA log page buffer\n");
5692 0 : return -ENXIO;
5693 : }
5694 :
5695 : /* Each descriptor in a ANA log page is not ensured to be 8-bytes aligned.
5696 : * Hence copy each descriptor to a temporary area when parsing it.
5697 : *
5698 : * Allocate a buffer whose size is as large as ANA log page buffer because
5699 : * we do not know the size of a descriptor until actually reading it.
5700 : */
5701 31 : nvme_ctrlr->copied_ana_desc = calloc(1, ana_log_page_size);
5702 31 : if (nvme_ctrlr->copied_ana_desc == NULL) {
5703 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "could not allocate a buffer to parse ANA descriptor\n");
5704 0 : return -ENOMEM;
5705 : }
5706 :
5707 31 : nvme_ctrlr->max_ana_log_page_size = ana_log_page_size;
5708 :
5709 31 : nvme_ctrlr->probe_ctx = ctx;
5710 :
5711 : /* Then, set the read size only to include the current active namespaces. */
5712 31 : ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);
5713 :
5714 31 : if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
5715 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
5716 : ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
5717 0 : return -EINVAL;
5718 : }
5719 :
5720 31 : return spdk_nvme_ctrlr_cmd_get_log_page(ctrlr,
5721 : SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
5722 : SPDK_NVME_GLOBAL_NS_TAG,
5723 31 : nvme_ctrlr->ana_log_page,
5724 : ana_log_page_size, 0,
5725 : nvme_ctrlr_init_ana_log_page_done,
5726 : nvme_ctrlr);
5727 : }
5728 :
5729 : /* hostnqn and subnqn were already verified before attaching a controller.
5730 : * Hence check only the multipath capability and cntlid here.
5731 : */
5732 : static bool
5733 16 : bdev_nvme_check_multipath(struct nvme_bdev_ctrlr *nbdev_ctrlr, struct spdk_nvme_ctrlr *ctrlr)
5734 : {
5735 : struct nvme_ctrlr *tmp;
5736 : const struct spdk_nvme_ctrlr_data *cdata, *tmp_cdata;
5737 :
5738 16 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5739 :
5740 16 : if (!cdata->cmic.multi_ctrlr) {
5741 0 : SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
5742 0 : return false;
5743 : }
5744 :
5745 33 : TAILQ_FOREACH(tmp, &nbdev_ctrlr->ctrlrs, tailq) {
5746 18 : tmp_cdata = spdk_nvme_ctrlr_get_data(tmp->ctrlr);
5747 :
5748 18 : if (!tmp_cdata->cmic.multi_ctrlr) {
5749 0 : NVME_CTRLR_ERRLOG(tmp, "Ctrlr%u does not support multipath.\n", cdata->cntlid);
5750 0 : return false;
5751 : }
5752 18 : if (cdata->cntlid == tmp_cdata->cntlid) {
5753 1 : NVME_CTRLR_ERRLOG(tmp, "cntlid %u are duplicated.\n", tmp_cdata->cntlid);
5754 1 : return false;
5755 : }
5756 : }
5757 :
5758 15 : return true;
5759 : }
5760 :
5761 :
5762 : static int
5763 62 : nvme_bdev_ctrlr_create(const char *name, struct nvme_ctrlr *nvme_ctrlr)
5764 : {
5765 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
5766 62 : struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5767 : struct nvme_ctrlr *nctrlr;
5768 62 : int rc = 0;
5769 :
5770 62 : pthread_mutex_lock(&g_bdev_nvme_mutex);
5771 :
5772 62 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
5773 62 : if (nbdev_ctrlr != NULL) {
5774 16 : if (!bdev_nvme_check_multipath(nbdev_ctrlr, ctrlr)) {
5775 1 : rc = -EINVAL;
5776 1 : goto exit;
5777 : }
5778 32 : TAILQ_FOREACH(nctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
5779 17 : if (nctrlr->opts.multipath != nvme_ctrlr->opts.multipath) {
5780 : /* All controllers with the same name must be configured the same
5781 : * way, either for multipath or failover. If the configuration doesn't
5782 : * match - report error.
5783 : */
5784 0 : rc = -EINVAL;
5785 0 : goto exit;
5786 : }
5787 : }
5788 : } else {
5789 46 : nbdev_ctrlr = calloc(1, sizeof(*nbdev_ctrlr));
5790 46 : if (nbdev_ctrlr == NULL) {
5791 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to allocate nvme_bdev_ctrlr.\n");
5792 0 : rc = -ENOMEM;
5793 0 : goto exit;
5794 : }
5795 46 : nbdev_ctrlr->name = strdup(name);
5796 46 : if (nbdev_ctrlr->name == NULL) {
5797 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "Failed to allocate name of nvme_bdev_ctrlr.\n");
5798 0 : free(nbdev_ctrlr);
5799 0 : goto exit;
5800 : }
5801 46 : TAILQ_INIT(&nbdev_ctrlr->ctrlrs);
5802 46 : TAILQ_INIT(&nbdev_ctrlr->bdevs);
5803 46 : TAILQ_INSERT_TAIL(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
5804 : }
5805 61 : nvme_ctrlr->nbdev_ctrlr = nbdev_ctrlr;
5806 61 : TAILQ_INSERT_TAIL(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
5807 62 : exit:
5808 62 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
5809 62 : return rc;
5810 : }
5811 :
5812 : static int
5813 62 : nvme_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr,
5814 : const char *name,
5815 : const struct spdk_nvme_transport_id *trid,
5816 : struct nvme_async_probe_ctx *ctx)
5817 : {
5818 : struct nvme_ctrlr *nvme_ctrlr;
5819 : struct nvme_path_id *path_id;
5820 : const struct spdk_nvme_ctrlr_data *cdata;
5821 62 : struct spdk_event_handler_opts opts = {
5822 : .opts_size = SPDK_SIZEOF(&opts, fd_type),
5823 : };
5824 : uint64_t period;
5825 : int fd, rc;
5826 :
5827 62 : nvme_ctrlr = calloc(1, sizeof(*nvme_ctrlr));
5828 62 : if (nvme_ctrlr == NULL) {
5829 0 : SPDK_ERRLOG("Failed to allocate device struct\n");
5830 0 : return -ENOMEM;
5831 : }
5832 :
5833 62 : rc = pthread_mutex_init(&nvme_ctrlr->mutex, NULL);
5834 62 : if (rc != 0) {
5835 0 : free(nvme_ctrlr);
5836 0 : return rc;
5837 : }
5838 :
5839 62 : TAILQ_INIT(&nvme_ctrlr->trids);
5840 62 : TAILQ_INIT(&nvme_ctrlr->pending_resets);
5841 62 : RB_INIT(&nvme_ctrlr->namespaces);
5842 :
5843 : /* Get another reference to the key, so the first one can be released from probe_ctx */
5844 62 : if (ctx != NULL) {
5845 48 : if (ctx->drv_opts.tls_psk != NULL) {
5846 0 : nvme_ctrlr->psk = spdk_keyring_get_key(
5847 : spdk_key_get_name(ctx->drv_opts.tls_psk));
5848 0 : if (nvme_ctrlr->psk == NULL) {
5849 : /* Could only happen if the key was removed in the meantime */
5850 0 : SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5851 : spdk_key_get_name(ctx->drv_opts.tls_psk));
5852 0 : rc = -ENOKEY;
5853 0 : goto err;
5854 : }
5855 : }
5856 :
5857 48 : if (ctx->drv_opts.dhchap_key != NULL) {
5858 0 : nvme_ctrlr->dhchap_key = spdk_keyring_get_key(
5859 : spdk_key_get_name(ctx->drv_opts.dhchap_key));
5860 0 : if (nvme_ctrlr->dhchap_key == NULL) {
5861 0 : SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5862 : spdk_key_get_name(ctx->drv_opts.dhchap_key));
5863 0 : rc = -ENOKEY;
5864 0 : goto err;
5865 : }
5866 : }
5867 :
5868 48 : if (ctx->drv_opts.dhchap_ctrlr_key != NULL) {
5869 0 : nvme_ctrlr->dhchap_ctrlr_key =
5870 0 : spdk_keyring_get_key(
5871 : spdk_key_get_name(ctx->drv_opts.dhchap_ctrlr_key));
5872 0 : if (nvme_ctrlr->dhchap_ctrlr_key == NULL) {
5873 0 : SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5874 : spdk_key_get_name(ctx->drv_opts.dhchap_ctrlr_key));
5875 0 : rc = -ENOKEY;
5876 0 : goto err;
5877 : }
5878 : }
5879 : }
5880 :
5881 : /* Check if we manage to enable interrupts on the controller. */
5882 62 : if (spdk_interrupt_mode_is_enabled() && ctx != NULL && !ctx->drv_opts.enable_interrupts) {
5883 0 : SPDK_ERRLOG("Failed to enable interrupts on the controller\n");
5884 0 : rc = -ENOTSUP;
5885 0 : goto err;
5886 : }
5887 :
5888 62 : path_id = calloc(1, sizeof(*path_id));
5889 62 : if (path_id == NULL) {
5890 0 : SPDK_ERRLOG("Failed to allocate trid entry pointer\n");
5891 0 : rc = -ENOMEM;
5892 0 : goto err;
5893 : }
5894 :
5895 62 : path_id->trid = *trid;
5896 62 : if (ctx != NULL) {
5897 48 : memcpy(path_id->hostid.hostaddr, ctx->drv_opts.src_addr, sizeof(path_id->hostid.hostaddr));
5898 48 : memcpy(path_id->hostid.hostsvcid, ctx->drv_opts.src_svcid, sizeof(path_id->hostid.hostsvcid));
5899 : }
5900 62 : nvme_ctrlr->active_path_id = path_id;
5901 62 : TAILQ_INSERT_HEAD(&nvme_ctrlr->trids, path_id, link);
5902 :
5903 62 : nvme_ctrlr->thread = spdk_get_thread();
5904 62 : nvme_ctrlr->ctrlr = ctrlr;
5905 62 : nvme_ctrlr->ref = 1;
5906 :
5907 62 : if (spdk_nvme_ctrlr_is_ocssd_supported(ctrlr)) {
5908 0 : SPDK_ERRLOG("OCSSDs are not supported");
5909 0 : rc = -ENOTSUP;
5910 0 : goto err;
5911 : }
5912 :
5913 62 : if (ctx != NULL) {
5914 48 : memcpy(&nvme_ctrlr->opts, &ctx->bdev_opts, sizeof(ctx->bdev_opts));
5915 : } else {
5916 14 : spdk_bdev_nvme_get_default_ctrlr_opts(&nvme_ctrlr->opts);
5917 : }
5918 :
5919 62 : period = spdk_interrupt_mode_is_enabled() ? 0 : g_opts.nvme_adminq_poll_period_us;
5920 :
5921 62 : nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq, nvme_ctrlr,
5922 : period);
5923 :
5924 62 : if (spdk_interrupt_mode_is_enabled()) {
5925 0 : spdk_poller_register_interrupt(nvme_ctrlr->adminq_timer_poller, NULL, NULL);
5926 :
5927 0 : fd = spdk_nvme_ctrlr_get_admin_qp_fd(nvme_ctrlr->ctrlr, &opts);
5928 0 : if (fd < 0) {
5929 0 : rc = fd;
5930 0 : goto err;
5931 : }
5932 :
5933 0 : nvme_ctrlr->intr = SPDK_INTERRUPT_REGISTER_EXT(fd, bdev_nvme_poll_adminq,
5934 : nvme_ctrlr, &opts);
5935 0 : if (!nvme_ctrlr->intr) {
5936 0 : rc = -EINVAL;
5937 0 : goto err;
5938 : }
5939 : }
5940 :
5941 62 : if (g_opts.timeout_us > 0) {
5942 : /* Register timeout callback. Timeout values for IO vs. admin reqs can be different. */
5943 : /* If timeout_admin_us is 0 (not specified), admin uses same timeout as IO. */
5944 0 : uint64_t adm_timeout_us = (g_opts.timeout_admin_us == 0) ?
5945 0 : g_opts.timeout_us : g_opts.timeout_admin_us;
5946 0 : spdk_nvme_ctrlr_register_timeout_callback(ctrlr, g_opts.timeout_us,
5947 : adm_timeout_us, timeout_cb, nvme_ctrlr);
5948 : }
5949 :
5950 62 : spdk_nvme_ctrlr_register_aer_callback(ctrlr, aer_cb, nvme_ctrlr);
5951 62 : spdk_nvme_ctrlr_set_remove_cb(ctrlr, remove_cb, nvme_ctrlr);
5952 :
5953 62 : if (spdk_nvme_ctrlr_get_flags(ctrlr) &
5954 : SPDK_NVME_CTRLR_SECURITY_SEND_RECV_SUPPORTED) {
5955 0 : nvme_ctrlr->opal_dev = spdk_opal_dev_construct(ctrlr);
5956 : }
5957 :
5958 62 : rc = nvme_bdev_ctrlr_create(name, nvme_ctrlr);
5959 62 : if (rc != 0) {
5960 1 : goto err;
5961 : }
5962 :
5963 61 : cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5964 :
5965 61 : if (cdata->cmic.ana_reporting) {
5966 31 : rc = nvme_ctrlr_init_ana_log_page(nvme_ctrlr, ctx);
5967 31 : if (rc == 0) {
5968 31 : return 0;
5969 : }
5970 : } else {
5971 30 : nvme_ctrlr_create_done(nvme_ctrlr, ctx);
5972 30 : return 0;
5973 : }
5974 :
5975 1 : err:
5976 1 : nvme_ctrlr_delete(nvme_ctrlr);
5977 1 : return rc;
5978 : }
5979 :
5980 : void
5981 34 : spdk_bdev_nvme_get_default_ctrlr_opts(struct spdk_bdev_nvme_ctrlr_opts *opts)
5982 : {
5983 34 : opts->prchk_flags = 0;
5984 34 : opts->ctrlr_loss_timeout_sec = g_opts.ctrlr_loss_timeout_sec;
5985 34 : opts->reconnect_delay_sec = g_opts.reconnect_delay_sec;
5986 34 : opts->fast_io_fail_timeout_sec = g_opts.fast_io_fail_timeout_sec;
5987 34 : opts->multipath = true;
5988 34 : }
5989 :
5990 : static void
5991 0 : attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
5992 : struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *drv_opts)
5993 : {
5994 : char *name;
5995 :
5996 0 : name = spdk_sprintf_alloc("HotInNvme%d", g_hot_insert_nvme_controller_index++);
5997 0 : if (!name) {
5998 0 : SPDK_ERRLOG("Failed to assign name to NVMe device\n");
5999 0 : return;
6000 : }
6001 :
6002 0 : if (nvme_ctrlr_create(ctrlr, name, trid, NULL) == 0) {
6003 0 : SPDK_DEBUGLOG(bdev_nvme, "Attached to %s (%s)\n", trid->traddr, name);
6004 : } else {
6005 0 : SPDK_ERRLOG("Failed to attach to %s (%s)\n", trid->traddr, name);
6006 : }
6007 :
6008 0 : free(name);
6009 : }
6010 :
6011 : static void
6012 61 : _nvme_ctrlr_destruct(void *ctx)
6013 : {
6014 61 : struct nvme_ctrlr *nvme_ctrlr = ctx;
6015 :
6016 61 : nvme_ctrlr_depopulate_namespaces(nvme_ctrlr);
6017 61 : nvme_ctrlr_put_ref(nvme_ctrlr);
6018 61 : }
6019 :
6020 : static int
6021 58 : bdev_nvme_delete_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
6022 : {
6023 : struct nvme_probe_skip_entry *entry;
6024 :
6025 : /* The controller's destruction was already started */
6026 58 : if (nvme_ctrlr->destruct) {
6027 0 : return -EALREADY;
6028 : }
6029 :
6030 58 : if (!hotplug &&
6031 58 : nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
6032 0 : entry = calloc(1, sizeof(*entry));
6033 0 : if (!entry) {
6034 0 : return -ENOMEM;
6035 : }
6036 0 : entry->trid = nvme_ctrlr->active_path_id->trid;
6037 0 : TAILQ_INSERT_TAIL(&g_skipped_nvme_ctrlrs, entry, tailq);
6038 : }
6039 :
6040 58 : nvme_ctrlr->destruct = true;
6041 58 : return 0;
6042 : }
6043 :
6044 : static int
6045 2 : bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
6046 : {
6047 : int rc;
6048 :
6049 2 : pthread_mutex_lock(&nvme_ctrlr->mutex);
6050 2 : rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, hotplug);
6051 2 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
6052 :
6053 2 : if (rc == 0) {
6054 2 : _nvme_ctrlr_destruct(nvme_ctrlr);
6055 0 : } else if (rc == -EALREADY) {
6056 0 : rc = 0;
6057 : }
6058 :
6059 2 : return rc;
6060 : }
6061 :
6062 : static void
6063 0 : remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr)
6064 : {
6065 0 : struct nvme_ctrlr *nvme_ctrlr = cb_ctx;
6066 :
6067 0 : bdev_nvme_delete_ctrlr(nvme_ctrlr, true);
6068 0 : }
6069 :
6070 : static int
6071 0 : bdev_nvme_hotplug_probe(void *arg)
6072 : {
6073 0 : if (g_hotplug_probe_ctx == NULL) {
6074 0 : spdk_poller_unregister(&g_hotplug_probe_poller);
6075 0 : return SPDK_POLLER_IDLE;
6076 : }
6077 :
6078 0 : if (spdk_nvme_probe_poll_async(g_hotplug_probe_ctx) != -EAGAIN) {
6079 0 : g_hotplug_probe_ctx = NULL;
6080 0 : spdk_poller_unregister(&g_hotplug_probe_poller);
6081 : }
6082 :
6083 0 : return SPDK_POLLER_BUSY;
6084 : }
6085 :
6086 : static int
6087 0 : bdev_nvme_hotplug(void *arg)
6088 : {
6089 0 : struct spdk_nvme_transport_id trid_pcie;
6090 :
6091 0 : if (g_hotplug_probe_ctx) {
6092 0 : return SPDK_POLLER_BUSY;
6093 : }
6094 :
6095 0 : memset(&trid_pcie, 0, sizeof(trid_pcie));
6096 0 : spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);
6097 :
6098 0 : g_hotplug_probe_ctx = spdk_nvme_probe_async(&trid_pcie, NULL,
6099 : hotplug_probe_cb, attach_cb, NULL);
6100 :
6101 0 : if (g_hotplug_probe_ctx) {
6102 0 : assert(g_hotplug_probe_poller == NULL);
6103 0 : g_hotplug_probe_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug_probe, NULL, 1000);
6104 : }
6105 :
6106 0 : return SPDK_POLLER_BUSY;
6107 : }
6108 :
6109 : void
6110 0 : bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts)
6111 : {
6112 0 : *opts = g_opts;
6113 0 : }
6114 :
6115 : static bool bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
6116 : uint32_t reconnect_delay_sec,
6117 : uint32_t fast_io_fail_timeout_sec);
6118 :
6119 : static int
6120 0 : bdev_nvme_validate_opts(const struct spdk_bdev_nvme_opts *opts)
6121 : {
6122 0 : if ((opts->timeout_us == 0) && (opts->timeout_admin_us != 0)) {
6123 : /* Can't set timeout_admin_us without also setting timeout_us */
6124 0 : SPDK_WARNLOG("Invalid options: Can't have (timeout_us == 0) with (timeout_admin_us > 0)\n");
6125 0 : return -EINVAL;
6126 : }
6127 :
6128 0 : if (opts->bdev_retry_count < -1) {
6129 0 : SPDK_WARNLOG("Invalid option: bdev_retry_count can't be less than -1.\n");
6130 0 : return -EINVAL;
6131 : }
6132 :
6133 0 : if (!bdev_nvme_check_io_error_resiliency_params(opts->ctrlr_loss_timeout_sec,
6134 0 : opts->reconnect_delay_sec,
6135 0 : opts->fast_io_fail_timeout_sec)) {
6136 0 : return -EINVAL;
6137 : }
6138 :
6139 0 : return 0;
6140 : }
6141 :
6142 : int
6143 0 : bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts)
6144 : {
6145 : int ret;
6146 :
6147 0 : ret = bdev_nvme_validate_opts(opts);
6148 0 : if (ret) {
6149 0 : SPDK_WARNLOG("Failed to set nvme opts.\n");
6150 0 : return ret;
6151 : }
6152 :
6153 0 : if (g_bdev_nvme_init_thread != NULL) {
6154 0 : if (!TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
6155 0 : return -EPERM;
6156 : }
6157 : }
6158 :
6159 0 : if (opts->rdma_srq_size != 0 ||
6160 0 : opts->rdma_max_cq_size != 0 ||
6161 0 : opts->rdma_cm_event_timeout_ms != 0) {
6162 0 : struct spdk_nvme_transport_opts drv_opts;
6163 :
6164 0 : spdk_nvme_transport_get_opts(&drv_opts, sizeof(drv_opts));
6165 0 : if (opts->rdma_srq_size != 0) {
6166 0 : drv_opts.rdma_srq_size = opts->rdma_srq_size;
6167 : }
6168 0 : if (opts->rdma_max_cq_size != 0) {
6169 0 : drv_opts.rdma_max_cq_size = opts->rdma_max_cq_size;
6170 : }
6171 0 : if (opts->rdma_cm_event_timeout_ms != 0) {
6172 0 : drv_opts.rdma_cm_event_timeout_ms = opts->rdma_cm_event_timeout_ms;
6173 : }
6174 :
6175 0 : ret = spdk_nvme_transport_set_opts(&drv_opts, sizeof(drv_opts));
6176 0 : if (ret) {
6177 0 : SPDK_ERRLOG("Failed to set NVMe transport opts.\n");
6178 0 : return ret;
6179 : }
6180 : }
6181 :
6182 0 : g_opts = *opts;
6183 :
6184 0 : return 0;
6185 : }
6186 :
6187 : struct set_nvme_hotplug_ctx {
6188 : uint64_t period_us;
6189 : bool enabled;
6190 : spdk_msg_fn fn;
6191 : void *fn_ctx;
6192 : };
6193 :
6194 : static void
6195 0 : set_nvme_hotplug_period_cb(void *_ctx)
6196 : {
6197 0 : struct set_nvme_hotplug_ctx *ctx = _ctx;
6198 :
6199 0 : spdk_poller_unregister(&g_hotplug_poller);
6200 0 : if (ctx->enabled) {
6201 0 : g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug, NULL, ctx->period_us);
6202 : } else {
6203 0 : g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_remove_poller, NULL,
6204 : NVME_HOTPLUG_POLL_PERIOD_DEFAULT);
6205 : }
6206 :
6207 0 : g_nvme_hotplug_poll_period_us = ctx->period_us;
6208 0 : g_nvme_hotplug_enabled = ctx->enabled;
6209 0 : if (ctx->fn) {
6210 0 : ctx->fn(ctx->fn_ctx);
6211 : }
6212 :
6213 0 : free(ctx);
6214 0 : }
6215 :
6216 : int
6217 0 : bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx)
6218 : {
6219 : struct set_nvme_hotplug_ctx *ctx;
6220 :
6221 0 : if (enabled == true && !spdk_process_is_primary()) {
6222 0 : return -EPERM;
6223 : }
6224 :
6225 0 : ctx = calloc(1, sizeof(*ctx));
6226 0 : if (ctx == NULL) {
6227 0 : return -ENOMEM;
6228 : }
6229 :
6230 0 : period_us = period_us == 0 ? NVME_HOTPLUG_POLL_PERIOD_DEFAULT : period_us;
6231 0 : ctx->period_us = spdk_min(period_us, NVME_HOTPLUG_POLL_PERIOD_MAX);
6232 0 : ctx->enabled = enabled;
6233 0 : ctx->fn = cb;
6234 0 : ctx->fn_ctx = cb_ctx;
6235 :
6236 0 : spdk_thread_send_msg(g_bdev_nvme_init_thread, set_nvme_hotplug_period_cb, ctx);
6237 0 : return 0;
6238 : }
6239 :
6240 : static void
6241 47 : nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
6242 : struct nvme_async_probe_ctx *ctx)
6243 : {
6244 : struct nvme_ns *nvme_ns;
6245 : struct nvme_bdev *nvme_bdev;
6246 : size_t j;
6247 :
6248 47 : assert(nvme_ctrlr != NULL);
6249 :
6250 47 : if (ctx->names == NULL) {
6251 0 : ctx->reported_bdevs = 0;
6252 0 : populate_namespaces_cb(ctx, 0);
6253 0 : return;
6254 : }
6255 :
6256 : /*
6257 : * Report the new bdevs that were created in this call.
6258 : * There can be more than one bdev per NVMe controller.
6259 : */
6260 47 : j = 0;
6261 47 : nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
6262 96 : while (nvme_ns != NULL) {
6263 49 : nvme_bdev = nvme_ns->bdev;
6264 49 : if (j < ctx->max_bdevs) {
6265 49 : ctx->names[j] = nvme_bdev->disk.name;
6266 49 : j++;
6267 : } else {
6268 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr,
6269 : "Maximum number of namespaces supported per NVMe controller is %du. "
6270 : "Unable to return all names of created bdevs\n",
6271 : ctx->max_bdevs);
6272 0 : ctx->reported_bdevs = 0;
6273 0 : populate_namespaces_cb(ctx, -ERANGE);
6274 0 : return;
6275 : }
6276 :
6277 49 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
6278 : }
6279 :
6280 47 : ctx->reported_bdevs = j;
6281 47 : populate_namespaces_cb(ctx, 0);
6282 : }
6283 :
6284 : static int
6285 9 : bdev_nvme_check_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
6286 : struct spdk_nvme_ctrlr *new_ctrlr,
6287 : struct spdk_nvme_transport_id *trid)
6288 : {
6289 : struct nvme_path_id *tmp_trid;
6290 :
6291 9 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
6292 0 : NVME_CTRLR_ERRLOG(nvme_ctrlr, "PCIe failover is not supported.\n");
6293 0 : return -ENOTSUP;
6294 : }
6295 :
6296 : /* Currently we only support failover to the same transport type. */
6297 9 : if (nvme_ctrlr->active_path_id->trid.trtype != trid->trtype) {
6298 0 : NVME_CTRLR_WARNLOG(nvme_ctrlr,
6299 : "Failover from trtype: %s to a different trtype: %s is not supported currently\n",
6300 : spdk_nvme_transport_id_trtype_str(nvme_ctrlr->active_path_id->trid.trtype),
6301 : spdk_nvme_transport_id_trtype_str(trid->trtype));
6302 0 : return -EINVAL;
6303 : }
6304 :
6305 :
6306 : /* Currently we only support failover to the same NQN. */
6307 9 : if (strncmp(trid->subnqn, nvme_ctrlr->active_path_id->trid.subnqn, SPDK_NVMF_NQN_MAX_LEN)) {
6308 0 : NVME_CTRLR_WARNLOG(nvme_ctrlr,
6309 : "Failover from subnqn: %s to a different subnqn: %s is not supported currently\n",
6310 : nvme_ctrlr->active_path_id->trid.subnqn, trid->subnqn);
6311 0 : return -EINVAL;
6312 : }
6313 :
6314 : /* Skip all the other checks if we've already registered this path. */
6315 21 : TAILQ_FOREACH(tmp_trid, &nvme_ctrlr->trids, link) {
6316 12 : if (!spdk_nvme_transport_id_compare(&tmp_trid->trid, trid)) {
6317 0 : NVME_CTRLR_WARNLOG(nvme_ctrlr, "This path (traddr: %s subnqn: %s) is already registered\n",
6318 : trid->traddr, trid->subnqn);
6319 0 : return -EALREADY;
6320 : }
6321 : }
6322 :
6323 9 : return 0;
6324 : }
6325 :
6326 : static int
6327 9 : bdev_nvme_check_secondary_namespace(struct nvme_ctrlr *nvme_ctrlr,
6328 : struct spdk_nvme_ctrlr *new_ctrlr)
6329 : {
6330 : struct nvme_ns *nvme_ns;
6331 : struct spdk_nvme_ns *new_ns;
6332 :
6333 9 : nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
6334 9 : while (nvme_ns != NULL) {
6335 0 : new_ns = spdk_nvme_ctrlr_get_ns(new_ctrlr, nvme_ns->id);
6336 0 : assert(new_ns != NULL);
6337 :
6338 0 : if (!bdev_nvme_compare_ns(nvme_ns->ns, new_ns)) {
6339 0 : return -EINVAL;
6340 : }
6341 :
6342 0 : nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
6343 : }
6344 :
6345 9 : return 0;
6346 : }
6347 :
6348 : static int
6349 9 : _bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
6350 : struct spdk_nvme_transport_id *trid)
6351 : {
6352 : struct nvme_path_id *active_id, *new_trid, *tmp_trid;
6353 :
6354 9 : new_trid = calloc(1, sizeof(*new_trid));
6355 9 : if (new_trid == NULL) {
6356 0 : return -ENOMEM;
6357 : }
6358 9 : new_trid->trid = *trid;
6359 :
6360 9 : active_id = nvme_ctrlr->active_path_id;
6361 9 : assert(active_id != NULL);
6362 9 : assert(active_id == TAILQ_FIRST(&nvme_ctrlr->trids));
6363 :
6364 : /* Skip the active trid not to replace it until it is failed. */
6365 9 : tmp_trid = TAILQ_NEXT(active_id, link);
6366 9 : if (tmp_trid == NULL) {
6367 6 : goto add_tail;
6368 : }
6369 :
6370 : /* It means the trid is faled if its last failed time is non-zero.
6371 : * Insert the new alternate trid before any failed trid.
6372 : */
6373 5 : TAILQ_FOREACH_FROM(tmp_trid, &nvme_ctrlr->trids, link) {
6374 3 : if (tmp_trid->last_failed_tsc != 0) {
6375 1 : TAILQ_INSERT_BEFORE(tmp_trid, new_trid, link);
6376 1 : return 0;
6377 : }
6378 : }
6379 :
6380 2 : add_tail:
6381 8 : TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, new_trid, link);
6382 8 : return 0;
6383 : }
6384 :
6385 : /* This is the case that a secondary path is added to an existing
6386 : * nvme_ctrlr for failover. After checking if it can access the same
6387 : * namespaces as the primary path, it is disconnected until failover occurs.
6388 : */
6389 : static int
6390 9 : bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
6391 : struct spdk_nvme_ctrlr *new_ctrlr,
6392 : struct spdk_nvme_transport_id *trid)
6393 : {
6394 : int rc;
6395 :
6396 9 : assert(nvme_ctrlr != NULL);
6397 :
6398 9 : pthread_mutex_lock(&nvme_ctrlr->mutex);
6399 :
6400 9 : rc = bdev_nvme_check_secondary_trid(nvme_ctrlr, new_ctrlr, trid);
6401 9 : if (rc != 0) {
6402 0 : goto exit;
6403 : }
6404 :
6405 9 : rc = bdev_nvme_check_secondary_namespace(nvme_ctrlr, new_ctrlr);
6406 9 : if (rc != 0) {
6407 0 : goto exit;
6408 : }
6409 :
6410 9 : rc = _bdev_nvme_add_secondary_trid(nvme_ctrlr, trid);
6411 :
6412 9 : exit:
6413 9 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
6414 :
6415 9 : spdk_nvme_detach(new_ctrlr);
6416 :
6417 9 : return rc;
6418 : }
6419 :
6420 : static void
6421 48 : connect_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
6422 : struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
6423 : {
6424 48 : struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6425 : struct nvme_async_probe_ctx *ctx;
6426 : int rc;
6427 :
6428 48 : ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
6429 48 : ctx->ctrlr_attached = true;
6430 :
6431 48 : rc = nvme_ctrlr_create(ctrlr, ctx->base_name, &ctx->trid, ctx);
6432 48 : if (rc != 0) {
6433 1 : ctx->reported_bdevs = 0;
6434 1 : populate_namespaces_cb(ctx, rc);
6435 : }
6436 48 : }
6437 :
6438 :
6439 : static void
6440 4 : connect_set_failover_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
6441 : struct spdk_nvme_ctrlr *ctrlr,
6442 : const struct spdk_nvme_ctrlr_opts *opts)
6443 : {
6444 4 : struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6445 : struct nvme_ctrlr *nvme_ctrlr;
6446 : struct nvme_async_probe_ctx *ctx;
6447 : int rc;
6448 :
6449 4 : ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
6450 4 : ctx->ctrlr_attached = true;
6451 :
6452 4 : nvme_ctrlr = nvme_ctrlr_get_by_name(ctx->base_name);
6453 4 : if (nvme_ctrlr) {
6454 4 : rc = bdev_nvme_add_secondary_trid(nvme_ctrlr, ctrlr, &ctx->trid);
6455 : } else {
6456 0 : rc = -ENODEV;
6457 : }
6458 :
6459 4 : ctx->reported_bdevs = 0;
6460 4 : populate_namespaces_cb(ctx, rc);
6461 4 : }
6462 :
6463 : static int
6464 53 : bdev_nvme_async_poll(void *arg)
6465 : {
6466 53 : struct nvme_async_probe_ctx *ctx = arg;
6467 : int rc;
6468 :
6469 53 : rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
6470 53 : if (spdk_unlikely(rc != -EAGAIN)) {
6471 53 : ctx->probe_done = true;
6472 53 : spdk_poller_unregister(&ctx->poller);
6473 53 : if (!ctx->ctrlr_attached) {
6474 : /* The probe is done, but no controller was attached.
6475 : * That means we had a failure, so report -EIO back to
6476 : * the caller (usually the RPC). populate_namespaces_cb()
6477 : * will take care of freeing the nvme_async_probe_ctx.
6478 : */
6479 1 : ctx->reported_bdevs = 0;
6480 1 : populate_namespaces_cb(ctx, -EIO);
6481 52 : } else if (ctx->namespaces_populated) {
6482 : /* The namespaces for the attached controller were all
6483 : * populated and the response was already sent to the
6484 : * caller (usually the RPC). So free the context here.
6485 : */
6486 21 : free_nvme_async_probe_ctx(ctx);
6487 : }
6488 : }
6489 :
6490 53 : return SPDK_POLLER_BUSY;
6491 : }
6492 :
6493 : static bool
6494 72 : bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
6495 : uint32_t reconnect_delay_sec,
6496 : uint32_t fast_io_fail_timeout_sec)
6497 : {
6498 72 : if (ctrlr_loss_timeout_sec < -1) {
6499 1 : SPDK_ERRLOG("ctrlr_loss_timeout_sec can't be less than -1.\n");
6500 1 : return false;
6501 71 : } else if (ctrlr_loss_timeout_sec == -1) {
6502 14 : if (reconnect_delay_sec == 0) {
6503 1 : SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
6504 1 : return false;
6505 13 : } else if (fast_io_fail_timeout_sec != 0 &&
6506 : fast_io_fail_timeout_sec < reconnect_delay_sec) {
6507 1 : SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io-fail_timeout_sec.\n");
6508 1 : return false;
6509 : }
6510 57 : } else if (ctrlr_loss_timeout_sec != 0) {
6511 11 : if (reconnect_delay_sec == 0) {
6512 1 : SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
6513 1 : return false;
6514 10 : } else if (reconnect_delay_sec > (uint32_t)ctrlr_loss_timeout_sec) {
6515 1 : SPDK_ERRLOG("reconnect_delay_sec can't be more than ctrlr_loss_timeout_sec.\n");
6516 1 : return false;
6517 9 : } else if (fast_io_fail_timeout_sec != 0) {
6518 6 : if (fast_io_fail_timeout_sec < reconnect_delay_sec) {
6519 1 : SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io_fail_timeout_sec.\n");
6520 1 : return false;
6521 5 : } else if (fast_io_fail_timeout_sec > (uint32_t)ctrlr_loss_timeout_sec) {
6522 1 : SPDK_ERRLOG("fast_io_fail_timeout_sec can't be more than ctrlr_loss_timeout_sec.\n");
6523 1 : return false;
6524 : }
6525 : }
6526 46 : } else if (reconnect_delay_sec != 0 || fast_io_fail_timeout_sec != 0) {
6527 2 : SPDK_ERRLOG("Both reconnect_delay_sec and fast_io_fail_timeout_sec must be 0 if ctrlr_loss_timeout_sec is 0.\n");
6528 2 : return false;
6529 : }
6530 :
6531 63 : return true;
6532 : }
6533 :
6534 : int
6535 53 : spdk_bdev_nvme_create(struct spdk_nvme_transport_id *trid,
6536 : const char *base_name,
6537 : const char **names,
6538 : uint32_t count,
6539 : spdk_bdev_nvme_create_cb cb_fn,
6540 : void *cb_ctx,
6541 : struct spdk_nvme_ctrlr_opts *drv_opts,
6542 : struct spdk_bdev_nvme_ctrlr_opts *bdev_opts)
6543 : {
6544 : struct nvme_probe_skip_entry *entry, *tmp;
6545 : struct nvme_async_probe_ctx *ctx;
6546 : spdk_nvme_attach_cb attach_cb;
6547 : struct nvme_ctrlr *nvme_ctrlr;
6548 : int len;
6549 :
6550 : /* TODO expand this check to include both the host and target TRIDs.
6551 : * Only if both are the same should we fail.
6552 : */
6553 53 : if (nvme_ctrlr_get(trid, drv_opts->hostnqn) != NULL) {
6554 0 : SPDK_ERRLOG("A controller with the provided trid (traddr: %s, hostnqn: %s) "
6555 : "already exists.\n", trid->traddr, drv_opts->hostnqn);
6556 0 : return -EEXIST;
6557 : }
6558 :
6559 53 : len = strnlen(base_name, SPDK_CONTROLLER_NAME_MAX);
6560 :
6561 53 : if (len == 0 || len == SPDK_CONTROLLER_NAME_MAX) {
6562 0 : SPDK_ERRLOG("controller name must be between 1 and %d characters\n", SPDK_CONTROLLER_NAME_MAX - 1);
6563 0 : return -EINVAL;
6564 : }
6565 :
6566 53 : if (bdev_opts != NULL &&
6567 53 : !bdev_nvme_check_io_error_resiliency_params(bdev_opts->ctrlr_loss_timeout_sec,
6568 : bdev_opts->reconnect_delay_sec,
6569 : bdev_opts->fast_io_fail_timeout_sec)) {
6570 0 : return -EINVAL;
6571 : }
6572 :
6573 53 : ctx = calloc(1, sizeof(*ctx));
6574 53 : if (!ctx) {
6575 0 : return -ENOMEM;
6576 : }
6577 53 : ctx->base_name = base_name;
6578 53 : ctx->names = names;
6579 53 : ctx->max_bdevs = count;
6580 53 : ctx->cb_fn = cb_fn;
6581 53 : ctx->cb_ctx = cb_ctx;
6582 53 : ctx->trid = *trid;
6583 :
6584 53 : if (bdev_opts) {
6585 53 : memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
6586 : } else {
6587 0 : spdk_bdev_nvme_get_default_ctrlr_opts(&ctx->bdev_opts);
6588 : }
6589 :
6590 53 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
6591 0 : TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, tmp) {
6592 0 : if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
6593 0 : TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
6594 0 : free(entry);
6595 0 : break;
6596 : }
6597 : }
6598 : }
6599 :
6600 53 : memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
6601 53 : ctx->drv_opts.transport_retry_count = g_opts.transport_retry_count;
6602 53 : ctx->drv_opts.transport_ack_timeout = g_opts.transport_ack_timeout;
6603 53 : ctx->drv_opts.keep_alive_timeout_ms = g_opts.keep_alive_timeout_ms;
6604 53 : ctx->drv_opts.disable_read_ana_log_page = true;
6605 53 : ctx->drv_opts.transport_tos = g_opts.transport_tos;
6606 :
6607 53 : if (spdk_interrupt_mode_is_enabled()) {
6608 0 : if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
6609 0 : ctx->drv_opts.enable_interrupts = true;
6610 : } else {
6611 0 : SPDK_ERRLOG("Interrupt mode is only supported with PCIe transport\n");
6612 0 : free_nvme_async_probe_ctx(ctx);
6613 0 : return -ENOTSUP;
6614 : }
6615 : }
6616 :
6617 53 : if (ctx->bdev_opts.psk != NULL) {
6618 0 : ctx->drv_opts.tls_psk = spdk_keyring_get_key(ctx->bdev_opts.psk);
6619 0 : if (ctx->drv_opts.tls_psk == NULL) {
6620 0 : SPDK_ERRLOG("Could not load PSK: %s\n", ctx->bdev_opts.psk);
6621 0 : free_nvme_async_probe_ctx(ctx);
6622 0 : return -ENOKEY;
6623 : }
6624 : }
6625 :
6626 53 : if (ctx->bdev_opts.dhchap_key != NULL) {
6627 0 : ctx->drv_opts.dhchap_key = spdk_keyring_get_key(ctx->bdev_opts.dhchap_key);
6628 0 : if (ctx->drv_opts.dhchap_key == NULL) {
6629 0 : SPDK_ERRLOG("Could not load DH-HMAC-CHAP key: %s\n",
6630 : ctx->bdev_opts.dhchap_key);
6631 0 : free_nvme_async_probe_ctx(ctx);
6632 0 : return -ENOKEY;
6633 : }
6634 :
6635 0 : ctx->drv_opts.dhchap_digests = g_opts.dhchap_digests;
6636 0 : ctx->drv_opts.dhchap_dhgroups = g_opts.dhchap_dhgroups;
6637 : }
6638 53 : if (ctx->bdev_opts.dhchap_ctrlr_key != NULL) {
6639 0 : ctx->drv_opts.dhchap_ctrlr_key =
6640 0 : spdk_keyring_get_key(ctx->bdev_opts.dhchap_ctrlr_key);
6641 0 : if (ctx->drv_opts.dhchap_ctrlr_key == NULL) {
6642 0 : SPDK_ERRLOG("Could not load DH-HMAC-CHAP controller key: %s\n",
6643 : ctx->bdev_opts.dhchap_ctrlr_key);
6644 0 : free_nvme_async_probe_ctx(ctx);
6645 0 : return -ENOKEY;
6646 : }
6647 : }
6648 :
6649 53 : if (nvme_bdev_ctrlr_get_by_name(base_name) == NULL || ctx->bdev_opts.multipath) {
6650 49 : attach_cb = connect_attach_cb;
6651 : } else {
6652 4 : attach_cb = connect_set_failover_cb;
6653 : }
6654 :
6655 53 : nvme_ctrlr = nvme_ctrlr_get_by_name(ctx->base_name);
6656 53 : if (nvme_ctrlr && nvme_ctrlr->opts.multipath != ctx->bdev_opts.multipath) {
6657 : /* All controllers with the same name must be configured the same
6658 : * way, either for multipath or failover. If the configuration doesn't
6659 : * match - report error.
6660 : */
6661 0 : free_nvme_async_probe_ctx(ctx);
6662 0 : return -EINVAL;
6663 : }
6664 :
6665 53 : ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, attach_cb);
6666 53 : if (ctx->probe_ctx == NULL) {
6667 0 : SPDK_ERRLOG("No controller was found with provided trid (traddr: %s)\n", trid->traddr);
6668 0 : free_nvme_async_probe_ctx(ctx);
6669 0 : return -ENODEV;
6670 : }
6671 53 : ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_async_poll, ctx, 1000);
6672 :
6673 53 : return 0;
6674 : }
6675 :
6676 : struct bdev_nvme_delete_ctx {
6677 : char *name;
6678 : struct nvme_path_id path_id;
6679 : bdev_nvme_delete_done_fn delete_done;
6680 : void *delete_done_ctx;
6681 : uint64_t timeout_ticks;
6682 : struct spdk_poller *poller;
6683 : };
6684 :
6685 : static void
6686 2 : free_bdev_nvme_delete_ctx(struct bdev_nvme_delete_ctx *ctx)
6687 : {
6688 2 : if (ctx != NULL) {
6689 1 : free(ctx->name);
6690 1 : free(ctx);
6691 : }
6692 2 : }
6693 :
6694 : static bool
6695 76 : nvme_path_id_compare(struct nvme_path_id *p, const struct nvme_path_id *path_id)
6696 : {
6697 76 : if (path_id->trid.trtype != 0) {
6698 21 : if (path_id->trid.trtype == SPDK_NVME_TRANSPORT_CUSTOM) {
6699 0 : if (strcasecmp(path_id->trid.trstring, p->trid.trstring) != 0) {
6700 0 : return false;
6701 : }
6702 : } else {
6703 21 : if (path_id->trid.trtype != p->trid.trtype) {
6704 0 : return false;
6705 : }
6706 : }
6707 : }
6708 :
6709 76 : if (!spdk_mem_all_zero(path_id->trid.traddr, sizeof(path_id->trid.traddr))) {
6710 21 : if (strcasecmp(path_id->trid.traddr, p->trid.traddr) != 0) {
6711 11 : return false;
6712 : }
6713 : }
6714 :
6715 65 : if (path_id->trid.adrfam != 0) {
6716 0 : if (path_id->trid.adrfam != p->trid.adrfam) {
6717 0 : return false;
6718 : }
6719 : }
6720 :
6721 65 : if (!spdk_mem_all_zero(path_id->trid.trsvcid, sizeof(path_id->trid.trsvcid))) {
6722 10 : if (strcasecmp(path_id->trid.trsvcid, p->trid.trsvcid) != 0) {
6723 0 : return false;
6724 : }
6725 : }
6726 :
6727 65 : if (!spdk_mem_all_zero(path_id->trid.subnqn, sizeof(path_id->trid.subnqn))) {
6728 10 : if (strcmp(path_id->trid.subnqn, p->trid.subnqn) != 0) {
6729 0 : return false;
6730 : }
6731 : }
6732 :
6733 65 : if (!spdk_mem_all_zero(path_id->hostid.hostaddr, sizeof(path_id->hostid.hostaddr))) {
6734 0 : if (strcmp(path_id->hostid.hostaddr, p->hostid.hostaddr) != 0) {
6735 0 : return false;
6736 : }
6737 : }
6738 :
6739 65 : if (!spdk_mem_all_zero(path_id->hostid.hostsvcid, sizeof(path_id->hostid.hostsvcid))) {
6740 0 : if (strcmp(path_id->hostid.hostsvcid, p->hostid.hostsvcid) != 0) {
6741 0 : return false;
6742 : }
6743 : }
6744 :
6745 65 : return true;
6746 : }
6747 :
6748 : static bool
6749 2 : nvme_path_id_exists(const char *name, const struct nvme_path_id *path_id)
6750 : {
6751 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
6752 : struct nvme_ctrlr *ctrlr;
6753 : struct nvme_path_id *p;
6754 :
6755 2 : pthread_mutex_lock(&g_bdev_nvme_mutex);
6756 2 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
6757 2 : if (!nbdev_ctrlr) {
6758 1 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6759 1 : return false;
6760 : }
6761 :
6762 1 : TAILQ_FOREACH(ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
6763 1 : pthread_mutex_lock(&ctrlr->mutex);
6764 1 : TAILQ_FOREACH(p, &ctrlr->trids, link) {
6765 1 : if (nvme_path_id_compare(p, path_id)) {
6766 1 : pthread_mutex_unlock(&ctrlr->mutex);
6767 1 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6768 1 : return true;
6769 : }
6770 : }
6771 0 : pthread_mutex_unlock(&ctrlr->mutex);
6772 : }
6773 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6774 :
6775 0 : return false;
6776 : }
6777 :
6778 : static int
6779 2 : bdev_nvme_delete_complete_poll(void *arg)
6780 : {
6781 2 : struct bdev_nvme_delete_ctx *ctx = arg;
6782 2 : int rc = 0;
6783 :
6784 2 : if (nvme_path_id_exists(ctx->name, &ctx->path_id)) {
6785 1 : if (ctx->timeout_ticks > spdk_get_ticks()) {
6786 1 : return SPDK_POLLER_BUSY;
6787 : }
6788 :
6789 0 : SPDK_ERRLOG("NVMe path '%s' still exists after delete\n", ctx->name);
6790 0 : rc = -ETIMEDOUT;
6791 : }
6792 :
6793 1 : spdk_poller_unregister(&ctx->poller);
6794 :
6795 1 : ctx->delete_done(ctx->delete_done_ctx, rc);
6796 1 : free_bdev_nvme_delete_ctx(ctx);
6797 :
6798 1 : return SPDK_POLLER_BUSY;
6799 : }
6800 :
6801 : static int
6802 65 : _bdev_nvme_delete(struct nvme_ctrlr *nvme_ctrlr, const struct nvme_path_id *path_id)
6803 : {
6804 : struct nvme_path_id *p, *t;
6805 : spdk_msg_fn msg_fn;
6806 65 : int rc = -ENXIO;
6807 :
6808 65 : pthread_mutex_lock(&nvme_ctrlr->mutex);
6809 :
6810 75 : TAILQ_FOREACH_REVERSE_SAFE(p, &nvme_ctrlr->trids, nvme_paths, link, t) {
6811 75 : if (p == TAILQ_FIRST(&nvme_ctrlr->trids)) {
6812 65 : break;
6813 : }
6814 :
6815 10 : if (!nvme_path_id_compare(p, path_id)) {
6816 3 : continue;
6817 : }
6818 :
6819 : /* We are not using the specified path. */
6820 7 : TAILQ_REMOVE(&nvme_ctrlr->trids, p, link);
6821 7 : free(p);
6822 7 : rc = 0;
6823 : }
6824 :
6825 65 : if (p == NULL || !nvme_path_id_compare(p, path_id)) {
6826 8 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
6827 8 : return rc;
6828 : }
6829 :
6830 : /* If we made it here, then this path is a match! Now we need to remove it. */
6831 :
6832 : /* This is the active path in use right now. The active path is always the first in the list. */
6833 57 : assert(p == nvme_ctrlr->active_path_id);
6834 :
6835 57 : if (!TAILQ_NEXT(p, link)) {
6836 : /* The current path is the only path. */
6837 56 : msg_fn = _nvme_ctrlr_destruct;
6838 56 : rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, false);
6839 : } else {
6840 : /* There is an alternative path. */
6841 1 : msg_fn = _bdev_nvme_reset_ctrlr;
6842 1 : rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, true);
6843 : }
6844 :
6845 57 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
6846 :
6847 57 : if (rc == 0) {
6848 57 : spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
6849 0 : } else if (rc == -EALREADY) {
6850 0 : rc = 0;
6851 : }
6852 :
6853 57 : return rc;
6854 : }
6855 :
6856 : int
6857 50 : bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id,
6858 : bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx)
6859 : {
6860 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
6861 : struct nvme_ctrlr *nvme_ctrlr, *tmp_nvme_ctrlr;
6862 50 : struct bdev_nvme_delete_ctx *ctx = NULL;
6863 50 : int rc = -ENXIO, _rc;
6864 :
6865 50 : if (name == NULL || path_id == NULL) {
6866 0 : rc = -EINVAL;
6867 0 : goto exit;
6868 : }
6869 :
6870 50 : pthread_mutex_lock(&g_bdev_nvme_mutex);
6871 :
6872 50 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
6873 50 : if (nbdev_ctrlr == NULL) {
6874 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6875 :
6876 0 : SPDK_ERRLOG("Failed to find NVMe bdev controller\n");
6877 0 : rc = -ENODEV;
6878 0 : goto exit;
6879 : }
6880 :
6881 115 : TAILQ_FOREACH_SAFE(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq, tmp_nvme_ctrlr) {
6882 65 : _rc = _bdev_nvme_delete(nvme_ctrlr, path_id);
6883 65 : if (_rc < 0 && _rc != -ENXIO) {
6884 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6885 0 : rc = _rc;
6886 0 : goto exit;
6887 65 : } else if (_rc == 0) {
6888 : /* We traverse all remaining nvme_ctrlrs even if one nvme_ctrlr
6889 : * was deleted successfully. To remember the successful deletion,
6890 : * overwrite rc only if _rc is zero.
6891 : */
6892 59 : rc = 0;
6893 : }
6894 : }
6895 :
6896 50 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
6897 :
6898 50 : if (rc != 0 || delete_done == NULL) {
6899 49 : goto exit;
6900 : }
6901 :
6902 1 : ctx = calloc(1, sizeof(*ctx));
6903 1 : if (ctx == NULL) {
6904 0 : SPDK_ERRLOG("Failed to allocate context for bdev_nvme_delete\n");
6905 0 : rc = -ENOMEM;
6906 0 : goto exit;
6907 : }
6908 :
6909 1 : ctx->name = strdup(name);
6910 1 : if (ctx->name == NULL) {
6911 0 : SPDK_ERRLOG("Failed to copy controller name for deletion\n");
6912 0 : rc = -ENOMEM;
6913 0 : goto exit;
6914 : }
6915 :
6916 1 : ctx->delete_done = delete_done;
6917 1 : ctx->delete_done_ctx = delete_done_ctx;
6918 1 : ctx->path_id = *path_id;
6919 1 : ctx->timeout_ticks = spdk_get_ticks() + 10 * spdk_get_ticks_hz();
6920 1 : ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_delete_complete_poll, ctx, 1000);
6921 1 : if (ctx->poller == NULL) {
6922 0 : SPDK_ERRLOG("Failed to register bdev_nvme_delete poller\n");
6923 0 : rc = -ENOMEM;
6924 0 : goto exit;
6925 : }
6926 :
6927 1 : exit:
6928 50 : if (rc != 0) {
6929 1 : free_bdev_nvme_delete_ctx(ctx);
6930 : }
6931 :
6932 50 : return rc;
6933 : }
6934 :
6935 : #define DISCOVERY_INFOLOG(ctx, format, ...) \
6936 : SPDK_INFOLOG(bdev_nvme, "Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);
6937 :
6938 : #define DISCOVERY_ERRLOG(ctx, format, ...) \
6939 : SPDK_ERRLOG("Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);
6940 :
6941 : struct discovery_entry_ctx {
6942 : char name[128];
6943 : struct spdk_nvme_transport_id trid;
6944 : struct spdk_nvme_ctrlr_opts drv_opts;
6945 : struct spdk_nvmf_discovery_log_page_entry entry;
6946 : TAILQ_ENTRY(discovery_entry_ctx) tailq;
6947 : struct discovery_ctx *ctx;
6948 : };
6949 :
6950 : struct discovery_ctx {
6951 : char *name;
6952 : spdk_bdev_nvme_start_discovery_fn start_cb_fn;
6953 : spdk_bdev_nvme_stop_discovery_fn stop_cb_fn;
6954 : void *cb_ctx;
6955 : struct spdk_nvme_probe_ctx *probe_ctx;
6956 : struct spdk_nvme_detach_ctx *detach_ctx;
6957 : struct spdk_nvme_ctrlr *ctrlr;
6958 : struct spdk_nvme_transport_id trid;
6959 : struct discovery_entry_ctx *entry_ctx_in_use;
6960 : struct spdk_poller *poller;
6961 : struct spdk_nvme_ctrlr_opts drv_opts;
6962 : struct spdk_bdev_nvme_ctrlr_opts bdev_opts;
6963 : struct spdk_nvmf_discovery_log_page *log_page;
6964 : TAILQ_ENTRY(discovery_ctx) tailq;
6965 : TAILQ_HEAD(, discovery_entry_ctx) nvm_entry_ctxs;
6966 : TAILQ_HEAD(, discovery_entry_ctx) discovery_entry_ctxs;
6967 : int rc;
6968 : bool wait_for_attach;
6969 : uint64_t timeout_ticks;
6970 : /* Denotes that the discovery service is being started. We're waiting
6971 : * for the initial connection to the discovery controller to be
6972 : * established and attach discovered NVM ctrlrs.
6973 : */
6974 : bool initializing;
6975 : /* Denotes if a discovery is currently in progress for this context.
6976 : * That includes connecting to newly discovered subsystems. Used to
6977 : * ensure we do not start a new discovery until an existing one is
6978 : * complete.
6979 : */
6980 : bool in_progress;
6981 :
6982 : /* Denotes if another discovery is needed after the one in progress
6983 : * completes. Set when we receive an AER completion while a discovery
6984 : * is already in progress.
6985 : */
6986 : bool pending;
6987 :
6988 : /* Signal to the discovery context poller that it should stop the
6989 : * discovery service, including detaching from the current discovery
6990 : * controller.
6991 : */
6992 : bool stop;
6993 :
6994 : struct spdk_thread *calling_thread;
6995 : uint32_t index;
6996 : uint32_t attach_in_progress;
6997 : char *hostnqn;
6998 :
6999 : /* Denotes if the discovery service was started by the mdns discovery.
7000 : */
7001 : bool from_mdns_discovery_service;
7002 : };
7003 :
7004 : TAILQ_HEAD(discovery_ctxs, discovery_ctx);
7005 : static struct discovery_ctxs g_discovery_ctxs = TAILQ_HEAD_INITIALIZER(g_discovery_ctxs);
7006 :
7007 : static void get_discovery_log_page(struct discovery_ctx *ctx);
7008 :
7009 : static void
7010 0 : free_discovery_ctx(struct discovery_ctx *ctx)
7011 : {
7012 0 : free(ctx->log_page);
7013 0 : free(ctx->hostnqn);
7014 0 : free(ctx->name);
7015 0 : free(ctx);
7016 0 : }
7017 :
7018 : static void
7019 0 : discovery_complete(struct discovery_ctx *ctx)
7020 : {
7021 0 : ctx->initializing = false;
7022 0 : ctx->in_progress = false;
7023 0 : if (ctx->pending) {
7024 0 : ctx->pending = false;
7025 0 : get_discovery_log_page(ctx);
7026 : }
7027 0 : }
7028 :
7029 : static void
7030 0 : build_trid_from_log_page_entry(struct spdk_nvme_transport_id *trid,
7031 : struct spdk_nvmf_discovery_log_page_entry *entry)
7032 : {
7033 : char *space;
7034 :
7035 0 : trid->trtype = entry->trtype;
7036 0 : trid->adrfam = entry->adrfam;
7037 0 : memcpy(trid->traddr, entry->traddr, sizeof(entry->traddr));
7038 0 : memcpy(trid->trsvcid, entry->trsvcid, sizeof(entry->trsvcid));
7039 : /* Because the source buffer (entry->subnqn) is longer than trid->subnqn, and
7040 : * before call to this function trid->subnqn is zeroed out, we need
7041 : * to copy sizeof(trid->subnqn) minus one byte to make sure the last character
7042 : * remains 0. Then we can shorten the string (replace ' ' with 0) if required
7043 : */
7044 0 : memcpy(trid->subnqn, entry->subnqn, sizeof(trid->subnqn) - 1);
7045 :
7046 : /* We want the traddr, trsvcid and subnqn fields to be NULL-terminated.
7047 : * But the log page entries typically pad them with spaces, not zeroes.
7048 : * So add a NULL terminator to each of these fields at the appropriate
7049 : * location.
7050 : */
7051 0 : space = strchr(trid->traddr, ' ');
7052 0 : if (space) {
7053 0 : *space = 0;
7054 : }
7055 0 : space = strchr(trid->trsvcid, ' ');
7056 0 : if (space) {
7057 0 : *space = 0;
7058 : }
7059 0 : space = strchr(trid->subnqn, ' ');
7060 0 : if (space) {
7061 0 : *space = 0;
7062 : }
7063 0 : }
7064 :
7065 : static void
7066 0 : _stop_discovery(void *_ctx)
7067 : {
7068 0 : struct discovery_ctx *ctx = _ctx;
7069 :
7070 0 : if (ctx->attach_in_progress > 0) {
7071 0 : spdk_thread_send_msg(spdk_get_thread(), _stop_discovery, ctx);
7072 0 : return;
7073 : }
7074 :
7075 0 : ctx->stop = true;
7076 :
7077 0 : while (!TAILQ_EMPTY(&ctx->nvm_entry_ctxs)) {
7078 : struct discovery_entry_ctx *entry_ctx;
7079 0 : struct nvme_path_id path = {};
7080 :
7081 0 : entry_ctx = TAILQ_FIRST(&ctx->nvm_entry_ctxs);
7082 0 : path.trid = entry_ctx->trid;
7083 0 : bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
7084 0 : TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
7085 0 : free(entry_ctx);
7086 : }
7087 :
7088 0 : while (!TAILQ_EMPTY(&ctx->discovery_entry_ctxs)) {
7089 : struct discovery_entry_ctx *entry_ctx;
7090 :
7091 0 : entry_ctx = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
7092 0 : TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
7093 0 : free(entry_ctx);
7094 : }
7095 :
7096 0 : free(ctx->entry_ctx_in_use);
7097 0 : ctx->entry_ctx_in_use = NULL;
7098 : }
7099 :
7100 : static void
7101 0 : stop_discovery(struct discovery_ctx *ctx, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
7102 : {
7103 0 : ctx->stop_cb_fn = cb_fn;
7104 0 : ctx->cb_ctx = cb_ctx;
7105 :
7106 0 : if (ctx->attach_in_progress > 0) {
7107 0 : DISCOVERY_INFOLOG(ctx, "stopping discovery with attach_in_progress: %"PRIu32"\n",
7108 : ctx->attach_in_progress);
7109 : }
7110 :
7111 0 : _stop_discovery(ctx);
7112 0 : }
7113 :
7114 : static void
7115 2 : remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr)
7116 : {
7117 : struct discovery_ctx *d_ctx;
7118 : struct nvme_path_id *path_id;
7119 2 : struct spdk_nvme_transport_id trid = {};
7120 : struct discovery_entry_ctx *entry_ctx, *tmp;
7121 :
7122 2 : path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
7123 :
7124 2 : TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
7125 0 : TAILQ_FOREACH_SAFE(entry_ctx, &d_ctx->nvm_entry_ctxs, tailq, tmp) {
7126 0 : build_trid_from_log_page_entry(&trid, &entry_ctx->entry);
7127 0 : if (spdk_nvme_transport_id_compare(&trid, &path_id->trid) != 0) {
7128 0 : continue;
7129 : }
7130 :
7131 0 : TAILQ_REMOVE(&d_ctx->nvm_entry_ctxs, entry_ctx, tailq);
7132 0 : free(entry_ctx);
7133 0 : DISCOVERY_INFOLOG(d_ctx, "Remove discovery entry: %s:%s:%s\n",
7134 : trid.subnqn, trid.traddr, trid.trsvcid);
7135 :
7136 : /* Fail discovery ctrlr to force reattach attempt */
7137 0 : spdk_nvme_ctrlr_fail(d_ctx->ctrlr);
7138 : }
7139 : }
7140 2 : }
7141 :
7142 : static void
7143 0 : discovery_remove_controllers(struct discovery_ctx *ctx)
7144 : {
7145 0 : struct spdk_nvmf_discovery_log_page *log_page = ctx->log_page;
7146 : struct discovery_entry_ctx *entry_ctx, *tmp;
7147 : struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
7148 0 : struct spdk_nvme_transport_id old_trid = {};
7149 : uint64_t numrec, i;
7150 : bool found;
7151 :
7152 0 : numrec = from_le64(&log_page->numrec);
7153 0 : TAILQ_FOREACH_SAFE(entry_ctx, &ctx->nvm_entry_ctxs, tailq, tmp) {
7154 0 : found = false;
7155 0 : old_entry = &entry_ctx->entry;
7156 0 : build_trid_from_log_page_entry(&old_trid, old_entry);
7157 0 : for (i = 0; i < numrec; i++) {
7158 0 : new_entry = &log_page->entries[i];
7159 0 : if (!memcmp(old_entry, new_entry, sizeof(*old_entry))) {
7160 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s found again\n",
7161 : old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
7162 0 : found = true;
7163 0 : break;
7164 : }
7165 : }
7166 0 : if (!found) {
7167 0 : struct nvme_path_id path = {};
7168 :
7169 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s not found\n",
7170 : old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
7171 :
7172 0 : path.trid = entry_ctx->trid;
7173 0 : bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
7174 0 : TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
7175 0 : free(entry_ctx);
7176 : }
7177 : }
7178 0 : free(log_page);
7179 0 : ctx->log_page = NULL;
7180 0 : discovery_complete(ctx);
7181 0 : }
7182 :
7183 : static void
7184 0 : complete_discovery_start(struct discovery_ctx *ctx, int status)
7185 : {
7186 0 : ctx->timeout_ticks = 0;
7187 0 : ctx->rc = status;
7188 0 : if (ctx->start_cb_fn) {
7189 0 : ctx->start_cb_fn(ctx->cb_ctx, status);
7190 0 : ctx->start_cb_fn = NULL;
7191 0 : ctx->cb_ctx = NULL;
7192 : }
7193 0 : }
7194 :
7195 : static void
7196 0 : discovery_attach_controller_done(void *cb_ctx, size_t bdev_count, int rc)
7197 : {
7198 0 : struct discovery_entry_ctx *entry_ctx = cb_ctx;
7199 0 : struct discovery_ctx *ctx = entry_ctx->ctx;
7200 :
7201 0 : DISCOVERY_INFOLOG(ctx, "attach %s done\n", entry_ctx->name);
7202 0 : ctx->attach_in_progress--;
7203 0 : if (ctx->attach_in_progress == 0) {
7204 0 : complete_discovery_start(ctx, ctx->rc);
7205 0 : if (ctx->initializing && ctx->rc != 0) {
7206 0 : DISCOVERY_ERRLOG(ctx, "stopping discovery due to errors: %d\n", ctx->rc);
7207 0 : stop_discovery(ctx, NULL, ctx->cb_ctx);
7208 : } else {
7209 0 : discovery_remove_controllers(ctx);
7210 : }
7211 : }
7212 0 : }
7213 :
7214 : static struct discovery_entry_ctx *
7215 0 : create_discovery_entry_ctx(struct discovery_ctx *ctx, struct spdk_nvme_transport_id *trid)
7216 : {
7217 : struct discovery_entry_ctx *new_ctx;
7218 :
7219 0 : new_ctx = calloc(1, sizeof(*new_ctx));
7220 0 : if (new_ctx == NULL) {
7221 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7222 0 : return NULL;
7223 : }
7224 :
7225 0 : new_ctx->ctx = ctx;
7226 0 : memcpy(&new_ctx->trid, trid, sizeof(*trid));
7227 0 : spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
7228 0 : snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
7229 0 : return new_ctx;
7230 : }
7231 :
7232 : static void
7233 0 : discovery_log_page_cb(void *cb_arg, int rc, const struct spdk_nvme_cpl *cpl,
7234 : struct spdk_nvmf_discovery_log_page *log_page)
7235 : {
7236 0 : struct discovery_ctx *ctx = cb_arg;
7237 : struct discovery_entry_ctx *entry_ctx, *tmp;
7238 : struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
7239 : uint64_t numrec, i;
7240 : bool found;
7241 :
7242 0 : if (rc || spdk_nvme_cpl_is_error(cpl)) {
7243 0 : DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
7244 0 : return;
7245 : }
7246 :
7247 0 : ctx->log_page = log_page;
7248 0 : assert(ctx->attach_in_progress == 0);
7249 0 : numrec = from_le64(&log_page->numrec);
7250 0 : TAILQ_FOREACH_SAFE(entry_ctx, &ctx->discovery_entry_ctxs, tailq, tmp) {
7251 0 : TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
7252 0 : free(entry_ctx);
7253 : }
7254 0 : for (i = 0; i < numrec; i++) {
7255 0 : found = false;
7256 0 : new_entry = &log_page->entries[i];
7257 0 : if (new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
7258 0 : new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
7259 : struct discovery_entry_ctx *new_ctx;
7260 0 : struct spdk_nvme_transport_id trid = {};
7261 :
7262 0 : build_trid_from_log_page_entry(&trid, new_entry);
7263 0 : new_ctx = create_discovery_entry_ctx(ctx, &trid);
7264 0 : if (new_ctx == NULL) {
7265 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7266 0 : break;
7267 : }
7268 :
7269 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, new_ctx, tailq);
7270 0 : continue;
7271 : }
7272 0 : TAILQ_FOREACH(entry_ctx, &ctx->nvm_entry_ctxs, tailq) {
7273 0 : old_entry = &entry_ctx->entry;
7274 0 : if (!memcmp(new_entry, old_entry, sizeof(*new_entry))) {
7275 0 : found = true;
7276 0 : break;
7277 : }
7278 : }
7279 0 : if (!found) {
7280 0 : struct discovery_entry_ctx *subnqn_ctx = NULL, *new_ctx;
7281 : struct discovery_ctx *d_ctx;
7282 :
7283 0 : TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
7284 0 : TAILQ_FOREACH(subnqn_ctx, &d_ctx->nvm_entry_ctxs, tailq) {
7285 0 : if (!memcmp(subnqn_ctx->entry.subnqn, new_entry->subnqn,
7286 : sizeof(new_entry->subnqn))) {
7287 0 : break;
7288 : }
7289 : }
7290 0 : if (subnqn_ctx) {
7291 0 : break;
7292 : }
7293 : }
7294 :
7295 0 : new_ctx = calloc(1, sizeof(*new_ctx));
7296 0 : if (new_ctx == NULL) {
7297 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7298 0 : break;
7299 : }
7300 :
7301 0 : new_ctx->ctx = ctx;
7302 0 : memcpy(&new_ctx->entry, new_entry, sizeof(*new_entry));
7303 0 : build_trid_from_log_page_entry(&new_ctx->trid, new_entry);
7304 0 : if (subnqn_ctx) {
7305 0 : snprintf(new_ctx->name, sizeof(new_ctx->name), "%s", subnqn_ctx->name);
7306 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new path for %s\n",
7307 : new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
7308 : new_ctx->name);
7309 : } else {
7310 0 : snprintf(new_ctx->name, sizeof(new_ctx->name), "%s%d", ctx->name, ctx->index++);
7311 0 : DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new subsystem %s\n",
7312 : new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
7313 : new_ctx->name);
7314 : }
7315 0 : spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
7316 0 : snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
7317 0 : rc = spdk_bdev_nvme_create(&new_ctx->trid, new_ctx->name, NULL, 0,
7318 : discovery_attach_controller_done, new_ctx,
7319 : &new_ctx->drv_opts, &ctx->bdev_opts);
7320 0 : if (rc == 0) {
7321 0 : TAILQ_INSERT_TAIL(&ctx->nvm_entry_ctxs, new_ctx, tailq);
7322 0 : ctx->attach_in_progress++;
7323 : } else {
7324 0 : DISCOVERY_ERRLOG(ctx, "spdk_bdev_nvme_create failed (%s)\n", spdk_strerror(-rc));
7325 : }
7326 : }
7327 : }
7328 :
7329 0 : if (ctx->attach_in_progress == 0) {
7330 0 : discovery_remove_controllers(ctx);
7331 : }
7332 : }
7333 :
7334 : static void
7335 0 : get_discovery_log_page(struct discovery_ctx *ctx)
7336 : {
7337 : int rc;
7338 :
7339 0 : assert(ctx->in_progress == false);
7340 0 : ctx->in_progress = true;
7341 0 : rc = spdk_nvme_ctrlr_get_discovery_log_page(ctx->ctrlr, discovery_log_page_cb, ctx);
7342 0 : if (rc != 0) {
7343 0 : DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
7344 : }
7345 0 : DISCOVERY_INFOLOG(ctx, "sent discovery log page command\n");
7346 0 : }
7347 :
7348 : static void
7349 0 : discovery_aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
7350 : {
7351 0 : struct discovery_ctx *ctx = arg;
7352 0 : uint32_t log_page_id = (cpl->cdw0 & 0xFF0000) >> 16;
7353 :
7354 0 : if (spdk_nvme_cpl_is_error(cpl)) {
7355 0 : DISCOVERY_ERRLOG(ctx, "aer failed\n");
7356 0 : return;
7357 : }
7358 :
7359 0 : if (log_page_id != SPDK_NVME_LOG_DISCOVERY) {
7360 0 : DISCOVERY_ERRLOG(ctx, "unexpected log page 0x%x\n", log_page_id);
7361 0 : return;
7362 : }
7363 :
7364 0 : DISCOVERY_INFOLOG(ctx, "got aer\n");
7365 0 : if (ctx->in_progress) {
7366 0 : ctx->pending = true;
7367 0 : return;
7368 : }
7369 :
7370 0 : get_discovery_log_page(ctx);
7371 : }
7372 :
7373 : static void
7374 0 : discovery_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
7375 : struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
7376 : {
7377 0 : struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
7378 : struct discovery_ctx *ctx;
7379 :
7380 0 : ctx = SPDK_CONTAINEROF(user_opts, struct discovery_ctx, drv_opts);
7381 :
7382 0 : DISCOVERY_INFOLOG(ctx, "discovery ctrlr attached\n");
7383 0 : ctx->probe_ctx = NULL;
7384 0 : ctx->ctrlr = ctrlr;
7385 :
7386 0 : if (ctx->rc != 0) {
7387 0 : DISCOVERY_ERRLOG(ctx, "encountered error while attaching discovery ctrlr: %d\n",
7388 : ctx->rc);
7389 0 : return;
7390 : }
7391 :
7392 0 : spdk_nvme_ctrlr_register_aer_callback(ctx->ctrlr, discovery_aer_cb, ctx);
7393 : }
7394 :
7395 : static int
7396 0 : discovery_poller(void *arg)
7397 : {
7398 0 : struct discovery_ctx *ctx = arg;
7399 : struct spdk_nvme_transport_id *trid;
7400 : int rc;
7401 :
7402 0 : if (ctx->detach_ctx) {
7403 0 : rc = spdk_nvme_detach_poll_async(ctx->detach_ctx);
7404 0 : if (rc != -EAGAIN) {
7405 0 : ctx->detach_ctx = NULL;
7406 0 : ctx->ctrlr = NULL;
7407 : }
7408 0 : } else if (ctx->stop) {
7409 0 : if (ctx->ctrlr != NULL) {
7410 0 : rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
7411 0 : if (rc == 0) {
7412 0 : return SPDK_POLLER_BUSY;
7413 : }
7414 0 : DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
7415 : }
7416 0 : spdk_poller_unregister(&ctx->poller);
7417 0 : TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
7418 0 : assert(ctx->start_cb_fn == NULL);
7419 0 : if (ctx->stop_cb_fn != NULL) {
7420 0 : ctx->stop_cb_fn(ctx->cb_ctx);
7421 : }
7422 0 : free_discovery_ctx(ctx);
7423 0 : } else if (ctx->probe_ctx == NULL && ctx->ctrlr == NULL) {
7424 0 : if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7425 0 : DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
7426 0 : assert(ctx->initializing);
7427 0 : spdk_poller_unregister(&ctx->poller);
7428 0 : TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
7429 0 : complete_discovery_start(ctx, -ETIMEDOUT);
7430 0 : stop_discovery(ctx, NULL, NULL);
7431 0 : free_discovery_ctx(ctx);
7432 0 : return SPDK_POLLER_BUSY;
7433 : }
7434 :
7435 0 : assert(ctx->entry_ctx_in_use == NULL);
7436 0 : ctx->entry_ctx_in_use = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
7437 0 : TAILQ_REMOVE(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7438 0 : trid = &ctx->entry_ctx_in_use->trid;
7439 :
7440 : /* All controllers must be configured explicitely either for multipath or failover.
7441 : * While discovery use multipath mode, we need to set this in bdev options as well.
7442 : */
7443 0 : ctx->bdev_opts.multipath = true;
7444 :
7445 0 : ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, discovery_attach_cb);
7446 0 : if (ctx->probe_ctx) {
7447 0 : spdk_poller_unregister(&ctx->poller);
7448 0 : ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000);
7449 : } else {
7450 0 : DISCOVERY_ERRLOG(ctx, "could not start discovery connect\n");
7451 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7452 0 : ctx->entry_ctx_in_use = NULL;
7453 : }
7454 0 : } else if (ctx->probe_ctx) {
7455 0 : if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7456 0 : DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
7457 0 : complete_discovery_start(ctx, -ETIMEDOUT);
7458 0 : return SPDK_POLLER_BUSY;
7459 : }
7460 :
7461 0 : rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
7462 0 : if (rc != -EAGAIN) {
7463 0 : if (ctx->rc != 0) {
7464 0 : assert(ctx->initializing);
7465 0 : stop_discovery(ctx, NULL, ctx->cb_ctx);
7466 : } else {
7467 0 : assert(rc == 0);
7468 0 : DISCOVERY_INFOLOG(ctx, "discovery ctrlr connected\n");
7469 0 : ctx->rc = rc;
7470 0 : get_discovery_log_page(ctx);
7471 : }
7472 : }
7473 : } else {
7474 0 : if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7475 0 : DISCOVERY_ERRLOG(ctx, "timed out while attaching NVM ctrlrs\n");
7476 0 : complete_discovery_start(ctx, -ETIMEDOUT);
7477 : /* We need to wait until all NVM ctrlrs are attached before we stop the
7478 : * discovery service to make sure we don't detach a ctrlr that is still
7479 : * being attached.
7480 : */
7481 0 : if (ctx->attach_in_progress == 0) {
7482 0 : stop_discovery(ctx, NULL, ctx->cb_ctx);
7483 0 : return SPDK_POLLER_BUSY;
7484 : }
7485 : }
7486 :
7487 0 : rc = spdk_nvme_ctrlr_process_admin_completions(ctx->ctrlr);
7488 0 : if (rc < 0) {
7489 0 : spdk_poller_unregister(&ctx->poller);
7490 0 : ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
7491 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7492 0 : ctx->entry_ctx_in_use = NULL;
7493 :
7494 0 : rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
7495 0 : if (rc != 0) {
7496 0 : DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
7497 0 : ctx->ctrlr = NULL;
7498 : }
7499 : }
7500 : }
7501 :
7502 0 : return SPDK_POLLER_BUSY;
7503 : }
7504 :
7505 : static void
7506 0 : start_discovery_poller(void *arg)
7507 : {
7508 0 : struct discovery_ctx *ctx = arg;
7509 :
7510 0 : TAILQ_INSERT_TAIL(&g_discovery_ctxs, ctx, tailq);
7511 0 : ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
7512 0 : }
7513 :
7514 : int
7515 0 : bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid,
7516 : const char *base_name,
7517 : struct spdk_nvme_ctrlr_opts *drv_opts,
7518 : struct spdk_bdev_nvme_ctrlr_opts *bdev_opts,
7519 : uint64_t attach_timeout,
7520 : bool from_mdns,
7521 : spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx)
7522 : {
7523 : struct discovery_ctx *ctx;
7524 : struct discovery_entry_ctx *discovery_entry_ctx;
7525 :
7526 0 : snprintf(trid->subnqn, sizeof(trid->subnqn), "%s", SPDK_NVMF_DISCOVERY_NQN);
7527 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7528 0 : if (strcmp(ctx->name, base_name) == 0) {
7529 0 : return -EEXIST;
7530 : }
7531 :
7532 0 : if (ctx->entry_ctx_in_use != NULL) {
7533 0 : if (!spdk_nvme_transport_id_compare(trid, &ctx->entry_ctx_in_use->trid)) {
7534 0 : return -EEXIST;
7535 : }
7536 : }
7537 :
7538 0 : TAILQ_FOREACH(discovery_entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
7539 0 : if (!spdk_nvme_transport_id_compare(trid, &discovery_entry_ctx->trid)) {
7540 0 : return -EEXIST;
7541 : }
7542 : }
7543 : }
7544 :
7545 0 : ctx = calloc(1, sizeof(*ctx));
7546 0 : if (ctx == NULL) {
7547 0 : return -ENOMEM;
7548 : }
7549 :
7550 0 : ctx->name = strdup(base_name);
7551 0 : if (ctx->name == NULL) {
7552 0 : free_discovery_ctx(ctx);
7553 0 : return -ENOMEM;
7554 : }
7555 0 : memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
7556 0 : memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
7557 0 : ctx->from_mdns_discovery_service = from_mdns;
7558 0 : ctx->bdev_opts.from_discovery_service = true;
7559 0 : ctx->calling_thread = spdk_get_thread();
7560 0 : ctx->start_cb_fn = cb_fn;
7561 0 : ctx->cb_ctx = cb_ctx;
7562 0 : ctx->initializing = true;
7563 0 : if (ctx->start_cb_fn) {
7564 : /* We can use this when dumping json to denote if this RPC parameter
7565 : * was specified or not.
7566 : */
7567 0 : ctx->wait_for_attach = true;
7568 : }
7569 0 : if (attach_timeout != 0) {
7570 0 : ctx->timeout_ticks = spdk_get_ticks() + attach_timeout *
7571 0 : spdk_get_ticks_hz() / 1000ull;
7572 : }
7573 0 : TAILQ_INIT(&ctx->nvm_entry_ctxs);
7574 0 : TAILQ_INIT(&ctx->discovery_entry_ctxs);
7575 0 : memcpy(&ctx->trid, trid, sizeof(*trid));
7576 : /* Even if user did not specify hostnqn, we can still strdup("\0"); */
7577 0 : ctx->hostnqn = strdup(ctx->drv_opts.hostnqn);
7578 0 : if (ctx->hostnqn == NULL) {
7579 0 : free_discovery_ctx(ctx);
7580 0 : return -ENOMEM;
7581 : }
7582 0 : discovery_entry_ctx = create_discovery_entry_ctx(ctx, trid);
7583 0 : if (discovery_entry_ctx == NULL) {
7584 0 : DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7585 0 : free_discovery_ctx(ctx);
7586 0 : return -ENOMEM;
7587 : }
7588 :
7589 0 : TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, discovery_entry_ctx, tailq);
7590 0 : spdk_thread_send_msg(g_bdev_nvme_init_thread, start_discovery_poller, ctx);
7591 0 : return 0;
7592 : }
7593 :
7594 : int
7595 0 : bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
7596 : {
7597 : struct discovery_ctx *ctx;
7598 :
7599 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7600 0 : if (strcmp(name, ctx->name) == 0) {
7601 0 : if (ctx->stop) {
7602 0 : return -EALREADY;
7603 : }
7604 : /* If we're still starting the discovery service and ->rc is non-zero, we're
7605 : * going to stop it as soon as we can
7606 : */
7607 0 : if (ctx->initializing && ctx->rc != 0) {
7608 0 : return -EALREADY;
7609 : }
7610 0 : stop_discovery(ctx, cb_fn, cb_ctx);
7611 0 : return 0;
7612 : }
7613 : }
7614 :
7615 0 : return -ENOENT;
7616 : }
7617 :
7618 : static int
7619 1 : bdev_nvme_library_init(void)
7620 : {
7621 1 : g_bdev_nvme_init_thread = spdk_get_thread();
7622 :
7623 1 : spdk_io_device_register(&g_nvme_bdev_ctrlrs, bdev_nvme_create_poll_group_cb,
7624 : bdev_nvme_destroy_poll_group_cb,
7625 : sizeof(struct nvme_poll_group), "nvme_poll_groups");
7626 :
7627 1 : return 0;
7628 : }
7629 :
7630 : static void
7631 1 : bdev_nvme_fini_destruct_ctrlrs(void)
7632 : {
7633 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
7634 : struct nvme_ctrlr *nvme_ctrlr;
7635 :
7636 1 : pthread_mutex_lock(&g_bdev_nvme_mutex);
7637 1 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
7638 0 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
7639 0 : pthread_mutex_lock(&nvme_ctrlr->mutex);
7640 0 : if (nvme_ctrlr->destruct) {
7641 : /* This controller's destruction was already started
7642 : * before the application started shutting down
7643 : */
7644 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
7645 0 : continue;
7646 : }
7647 0 : nvme_ctrlr->destruct = true;
7648 0 : pthread_mutex_unlock(&nvme_ctrlr->mutex);
7649 :
7650 0 : spdk_thread_send_msg(nvme_ctrlr->thread, _nvme_ctrlr_destruct,
7651 : nvme_ctrlr);
7652 : }
7653 : }
7654 :
7655 1 : g_bdev_nvme_module_finish = true;
7656 1 : if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
7657 1 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
7658 1 : spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
7659 1 : spdk_bdev_module_fini_done();
7660 1 : return;
7661 : }
7662 :
7663 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
7664 : }
7665 :
7666 : static void
7667 0 : check_discovery_fini(void *arg)
7668 : {
7669 0 : if (TAILQ_EMPTY(&g_discovery_ctxs)) {
7670 0 : bdev_nvme_fini_destruct_ctrlrs();
7671 : }
7672 0 : }
7673 :
7674 : static void
7675 1 : bdev_nvme_library_fini(void)
7676 : {
7677 : struct nvme_probe_skip_entry *entry, *entry_tmp;
7678 : struct discovery_ctx *ctx;
7679 :
7680 1 : spdk_poller_unregister(&g_hotplug_poller);
7681 1 : free(g_hotplug_probe_ctx);
7682 1 : g_hotplug_probe_ctx = NULL;
7683 :
7684 1 : TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, entry_tmp) {
7685 0 : TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
7686 0 : free(entry);
7687 : }
7688 :
7689 1 : assert(spdk_get_thread() == g_bdev_nvme_init_thread);
7690 1 : if (TAILQ_EMPTY(&g_discovery_ctxs)) {
7691 1 : bdev_nvme_fini_destruct_ctrlrs();
7692 : } else {
7693 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7694 0 : stop_discovery(ctx, check_discovery_fini, NULL);
7695 : }
7696 : }
7697 1 : }
7698 :
7699 : static void
7700 0 : bdev_nvme_verify_pi_error(struct nvme_bdev_io *bio)
7701 : {
7702 0 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7703 0 : struct spdk_bdev *bdev = bdev_io->bdev;
7704 0 : struct spdk_dif_ctx dif_ctx;
7705 0 : struct spdk_dif_error err_blk = {};
7706 : int rc;
7707 0 : struct spdk_dif_ctx_init_ext_opts dif_opts;
7708 :
7709 0 : dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
7710 0 : dif_opts.dif_pi_format = bdev->dif_pi_format;
7711 0 : rc = spdk_dif_ctx_init(&dif_ctx,
7712 0 : bdev->blocklen, bdev->md_len, bdev->md_interleave,
7713 0 : bdev->dif_is_head_of_md, bdev->dif_type,
7714 : bdev_io->u.bdev.dif_check_flags,
7715 0 : bdev_io->u.bdev.offset_blocks, 0, 0, 0, 0, &dif_opts);
7716 0 : if (rc != 0) {
7717 0 : SPDK_ERRLOG("Initialization of DIF context failed\n");
7718 0 : return;
7719 : }
7720 :
7721 0 : if (bdev->md_interleave) {
7722 0 : rc = spdk_dif_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
7723 0 : bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
7724 : } else {
7725 0 : struct iovec md_iov = {
7726 0 : .iov_base = bdev_io->u.bdev.md_buf,
7727 0 : .iov_len = bdev_io->u.bdev.num_blocks * bdev->md_len,
7728 : };
7729 :
7730 0 : rc = spdk_dix_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
7731 0 : &md_iov, bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
7732 : }
7733 :
7734 0 : if (rc != 0) {
7735 0 : SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
7736 : err_blk.err_type, err_blk.err_offset);
7737 : } else {
7738 0 : SPDK_ERRLOG("Hardware reported PI error but SPDK could not find any.\n");
7739 : }
7740 : }
7741 :
7742 : static void
7743 0 : bdev_nvme_no_pi_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7744 : {
7745 0 : struct nvme_bdev_io *bio = ref;
7746 :
7747 0 : if (spdk_nvme_cpl_is_success(cpl)) {
7748 : /* Run PI verification for read data buffer. */
7749 0 : bdev_nvme_verify_pi_error(bio);
7750 : }
7751 :
7752 : /* Return original completion status */
7753 0 : bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
7754 0 : }
7755 :
7756 : static void
7757 3 : bdev_nvme_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7758 : {
7759 3 : struct nvme_bdev_io *bio = ref;
7760 3 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7761 : int ret;
7762 :
7763 3 : if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
7764 0 : SPDK_ERRLOG("readv completed with PI error (sct=%d, sc=%d)\n",
7765 : cpl->status.sct, cpl->status.sc);
7766 :
7767 : /* Save completion status to use after verifying PI error. */
7768 0 : bio->cpl = *cpl;
7769 :
7770 0 : if (spdk_likely(nvme_io_path_is_available(bio->io_path))) {
7771 : /* Read without PI checking to verify PI error. */
7772 0 : ret = bdev_nvme_no_pi_readv(bio,
7773 : bdev_io->u.bdev.iovs,
7774 : bdev_io->u.bdev.iovcnt,
7775 : bdev_io->u.bdev.md_buf,
7776 : bdev_io->u.bdev.num_blocks,
7777 : bdev_io->u.bdev.offset_blocks);
7778 0 : if (ret == 0) {
7779 0 : return;
7780 : }
7781 : }
7782 : }
7783 :
7784 3 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7785 : }
7786 :
7787 : static void
7788 25 : bdev_nvme_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7789 : {
7790 25 : struct nvme_bdev_io *bio = ref;
7791 :
7792 25 : if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
7793 0 : SPDK_ERRLOG("writev completed with PI error (sct=%d, sc=%d)\n",
7794 : cpl->status.sct, cpl->status.sc);
7795 : /* Run PI verification for write data buffer if PI error is detected. */
7796 0 : bdev_nvme_verify_pi_error(bio);
7797 : }
7798 :
7799 25 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7800 25 : }
7801 :
7802 : static void
7803 0 : bdev_nvme_zone_appendv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7804 : {
7805 0 : struct nvme_bdev_io *bio = ref;
7806 0 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7807 :
7808 : /* spdk_bdev_io_get_append_location() requires that the ALBA is stored in offset_blocks.
7809 : * Additionally, offset_blocks has to be set before calling bdev_nvme_verify_pi_error().
7810 : */
7811 0 : bdev_io->u.bdev.offset_blocks = *(uint64_t *)&cpl->cdw0;
7812 :
7813 0 : if (spdk_nvme_cpl_is_pi_error(cpl)) {
7814 0 : SPDK_ERRLOG("zone append completed with PI error (sct=%d, sc=%d)\n",
7815 : cpl->status.sct, cpl->status.sc);
7816 : /* Run PI verification for zone append data buffer if PI error is detected. */
7817 0 : bdev_nvme_verify_pi_error(bio);
7818 : }
7819 :
7820 0 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7821 0 : }
7822 :
7823 : static void
7824 1 : bdev_nvme_comparev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7825 : {
7826 1 : struct nvme_bdev_io *bio = ref;
7827 :
7828 1 : if (spdk_nvme_cpl_is_pi_error(cpl)) {
7829 0 : SPDK_ERRLOG("comparev completed with PI error (sct=%d, sc=%d)\n",
7830 : cpl->status.sct, cpl->status.sc);
7831 : /* Run PI verification for compare data buffer if PI error is detected. */
7832 0 : bdev_nvme_verify_pi_error(bio);
7833 : }
7834 :
7835 1 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7836 1 : }
7837 :
7838 : static void
7839 4 : bdev_nvme_comparev_and_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7840 : {
7841 4 : struct nvme_bdev_io *bio = ref;
7842 :
7843 : /* Compare operation completion */
7844 4 : if (!bio->first_fused_completed) {
7845 : /* Save compare result for write callback */
7846 2 : bio->cpl = *cpl;
7847 2 : bio->first_fused_completed = true;
7848 2 : return;
7849 : }
7850 :
7851 : /* Write operation completion */
7852 2 : if (spdk_nvme_cpl_is_error(&bio->cpl)) {
7853 : /* If bio->cpl is already an error, it means the compare operation failed. In that case,
7854 : * complete the IO with the compare operation's status.
7855 : */
7856 1 : if (!spdk_nvme_cpl_is_error(cpl)) {
7857 1 : SPDK_ERRLOG("Unexpected write success after compare failure.\n");
7858 : }
7859 :
7860 1 : bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
7861 : } else {
7862 1 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7863 : }
7864 : }
7865 :
7866 : static void
7867 1 : bdev_nvme_queued_done(void *ref, const struct spdk_nvme_cpl *cpl)
7868 : {
7869 1 : struct nvme_bdev_io *bio = ref;
7870 :
7871 1 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7872 1 : }
7873 :
7874 : static int
7875 0 : fill_zone_from_report(struct spdk_bdev_zone_info *info, struct spdk_nvme_zns_zone_desc *desc)
7876 : {
7877 0 : switch (desc->zt) {
7878 0 : case SPDK_NVME_ZONE_TYPE_SEQWR:
7879 0 : info->type = SPDK_BDEV_ZONE_TYPE_SEQWR;
7880 0 : break;
7881 0 : default:
7882 0 : SPDK_ERRLOG("Invalid zone type: %#x in zone report\n", desc->zt);
7883 0 : return -EIO;
7884 : }
7885 :
7886 0 : switch (desc->zs) {
7887 0 : case SPDK_NVME_ZONE_STATE_EMPTY:
7888 0 : info->state = SPDK_BDEV_ZONE_STATE_EMPTY;
7889 0 : break;
7890 0 : case SPDK_NVME_ZONE_STATE_IOPEN:
7891 0 : info->state = SPDK_BDEV_ZONE_STATE_IMP_OPEN;
7892 0 : break;
7893 0 : case SPDK_NVME_ZONE_STATE_EOPEN:
7894 0 : info->state = SPDK_BDEV_ZONE_STATE_EXP_OPEN;
7895 0 : break;
7896 0 : case SPDK_NVME_ZONE_STATE_CLOSED:
7897 0 : info->state = SPDK_BDEV_ZONE_STATE_CLOSED;
7898 0 : break;
7899 0 : case SPDK_NVME_ZONE_STATE_RONLY:
7900 0 : info->state = SPDK_BDEV_ZONE_STATE_READ_ONLY;
7901 0 : break;
7902 0 : case SPDK_NVME_ZONE_STATE_FULL:
7903 0 : info->state = SPDK_BDEV_ZONE_STATE_FULL;
7904 0 : break;
7905 0 : case SPDK_NVME_ZONE_STATE_OFFLINE:
7906 0 : info->state = SPDK_BDEV_ZONE_STATE_OFFLINE;
7907 0 : break;
7908 0 : default:
7909 0 : SPDK_ERRLOG("Invalid zone state: %#x in zone report\n", desc->zs);
7910 0 : return -EIO;
7911 : }
7912 :
7913 0 : info->zone_id = desc->zslba;
7914 0 : info->write_pointer = desc->wp;
7915 0 : info->capacity = desc->zcap;
7916 :
7917 0 : return 0;
7918 : }
7919 :
7920 : static void
7921 0 : bdev_nvme_get_zone_info_done(void *ref, const struct spdk_nvme_cpl *cpl)
7922 : {
7923 0 : struct nvme_bdev_io *bio = ref;
7924 0 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7925 0 : uint64_t zone_id = bdev_io->u.zone_mgmt.zone_id;
7926 0 : uint32_t zones_to_copy = bdev_io->u.zone_mgmt.num_zones;
7927 0 : struct spdk_bdev_zone_info *info = bdev_io->u.zone_mgmt.buf;
7928 : uint64_t max_zones_per_buf, i;
7929 : uint32_t zone_report_bufsize;
7930 : struct spdk_nvme_ns *ns;
7931 : struct spdk_nvme_qpair *qpair;
7932 : int ret;
7933 :
7934 0 : if (spdk_nvme_cpl_is_error(cpl)) {
7935 0 : goto out_complete_io_nvme_cpl;
7936 : }
7937 :
7938 0 : if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
7939 0 : ret = -ENXIO;
7940 0 : goto out_complete_io_ret;
7941 : }
7942 :
7943 0 : ns = bio->io_path->nvme_ns->ns;
7944 0 : qpair = bio->io_path->qpair->qpair;
7945 :
7946 0 : zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
7947 0 : max_zones_per_buf = (zone_report_bufsize - sizeof(*bio->zone_report_buf)) /
7948 : sizeof(bio->zone_report_buf->descs[0]);
7949 :
7950 0 : if (bio->zone_report_buf->nr_zones > max_zones_per_buf) {
7951 0 : ret = -EINVAL;
7952 0 : goto out_complete_io_ret;
7953 : }
7954 :
7955 0 : if (!bio->zone_report_buf->nr_zones) {
7956 0 : ret = -EINVAL;
7957 0 : goto out_complete_io_ret;
7958 : }
7959 :
7960 0 : for (i = 0; i < bio->zone_report_buf->nr_zones && bio->handled_zones < zones_to_copy; i++) {
7961 0 : ret = fill_zone_from_report(&info[bio->handled_zones],
7962 0 : &bio->zone_report_buf->descs[i]);
7963 0 : if (ret) {
7964 0 : goto out_complete_io_ret;
7965 : }
7966 0 : bio->handled_zones++;
7967 : }
7968 :
7969 0 : if (bio->handled_zones < zones_to_copy) {
7970 0 : uint64_t zone_size_lba = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
7971 0 : uint64_t slba = zone_id + (zone_size_lba * bio->handled_zones);
7972 :
7973 0 : memset(bio->zone_report_buf, 0, zone_report_bufsize);
7974 0 : ret = spdk_nvme_zns_report_zones(ns, qpair,
7975 0 : bio->zone_report_buf, zone_report_bufsize,
7976 : slba, SPDK_NVME_ZRA_LIST_ALL, true,
7977 : bdev_nvme_get_zone_info_done, bio);
7978 0 : if (!ret) {
7979 0 : return;
7980 : } else {
7981 0 : goto out_complete_io_ret;
7982 : }
7983 : }
7984 :
7985 0 : out_complete_io_nvme_cpl:
7986 0 : free(bio->zone_report_buf);
7987 0 : bio->zone_report_buf = NULL;
7988 0 : bdev_nvme_io_complete_nvme_status(bio, cpl);
7989 0 : return;
7990 :
7991 0 : out_complete_io_ret:
7992 0 : free(bio->zone_report_buf);
7993 0 : bio->zone_report_buf = NULL;
7994 0 : bdev_nvme_io_complete(bio, ret);
7995 : }
7996 :
7997 : static void
7998 0 : bdev_nvme_zone_management_done(void *ref, const struct spdk_nvme_cpl *cpl)
7999 : {
8000 0 : struct nvme_bdev_io *bio = ref;
8001 :
8002 0 : bdev_nvme_io_complete_nvme_status(bio, cpl);
8003 0 : }
8004 :
8005 : static void
8006 4 : bdev_nvme_admin_passthru_complete_nvme_status(void *ctx)
8007 : {
8008 4 : struct nvme_bdev_io *bio = ctx;
8009 4 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
8010 4 : const struct spdk_nvme_cpl *cpl = &bio->cpl;
8011 :
8012 4 : assert(bdev_nvme_io_type_is_admin(bdev_io->type));
8013 :
8014 4 : __bdev_nvme_io_complete(bdev_io, 0, cpl);
8015 4 : }
8016 :
8017 : static void
8018 3 : bdev_nvme_abort_complete(void *ctx)
8019 : {
8020 3 : struct nvme_bdev_io *bio = ctx;
8021 3 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
8022 :
8023 3 : if (spdk_nvme_cpl_is_abort_success(&bio->cpl)) {
8024 3 : __bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS, NULL);
8025 : } else {
8026 0 : __bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED, NULL);
8027 : }
8028 3 : }
8029 :
8030 : static void
8031 3 : bdev_nvme_abort_done(void *ref, const struct spdk_nvme_cpl *cpl)
8032 : {
8033 3 : struct nvme_bdev_io *bio = ref;
8034 3 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
8035 :
8036 3 : bio->cpl = *cpl;
8037 3 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), bdev_nvme_abort_complete, bio);
8038 3 : }
8039 :
8040 : static void
8041 4 : bdev_nvme_admin_passthru_done(void *ref, const struct spdk_nvme_cpl *cpl)
8042 : {
8043 4 : struct nvme_bdev_io *bio = ref;
8044 4 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
8045 :
8046 4 : bio->cpl = *cpl;
8047 4 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
8048 : bdev_nvme_admin_passthru_complete_nvme_status, bio);
8049 4 : }
8050 :
8051 : static void
8052 0 : bdev_nvme_queued_reset_sgl(void *ref, uint32_t sgl_offset)
8053 : {
8054 0 : struct nvme_bdev_io *bio = ref;
8055 : struct iovec *iov;
8056 :
8057 0 : bio->iov_offset = sgl_offset;
8058 0 : for (bio->iovpos = 0; bio->iovpos < bio->iovcnt; bio->iovpos++) {
8059 0 : iov = &bio->iovs[bio->iovpos];
8060 0 : if (bio->iov_offset < iov->iov_len) {
8061 0 : break;
8062 : }
8063 :
8064 0 : bio->iov_offset -= iov->iov_len;
8065 : }
8066 0 : }
8067 :
8068 : static int
8069 0 : bdev_nvme_queued_next_sge(void *ref, void **address, uint32_t *length)
8070 : {
8071 0 : struct nvme_bdev_io *bio = ref;
8072 : struct iovec *iov;
8073 :
8074 0 : assert(bio->iovpos < bio->iovcnt);
8075 :
8076 0 : iov = &bio->iovs[bio->iovpos];
8077 :
8078 0 : *address = iov->iov_base;
8079 0 : *length = iov->iov_len;
8080 :
8081 0 : if (bio->iov_offset) {
8082 0 : assert(bio->iov_offset <= iov->iov_len);
8083 0 : *address += bio->iov_offset;
8084 0 : *length -= bio->iov_offset;
8085 : }
8086 :
8087 0 : bio->iov_offset += *length;
8088 0 : if (bio->iov_offset == iov->iov_len) {
8089 0 : bio->iovpos++;
8090 0 : bio->iov_offset = 0;
8091 : }
8092 :
8093 0 : return 0;
8094 : }
8095 :
8096 : static void
8097 0 : bdev_nvme_queued_reset_fused_sgl(void *ref, uint32_t sgl_offset)
8098 : {
8099 0 : struct nvme_bdev_io *bio = ref;
8100 : struct iovec *iov;
8101 :
8102 0 : bio->fused_iov_offset = sgl_offset;
8103 0 : for (bio->fused_iovpos = 0; bio->fused_iovpos < bio->fused_iovcnt; bio->fused_iovpos++) {
8104 0 : iov = &bio->fused_iovs[bio->fused_iovpos];
8105 0 : if (bio->fused_iov_offset < iov->iov_len) {
8106 0 : break;
8107 : }
8108 :
8109 0 : bio->fused_iov_offset -= iov->iov_len;
8110 : }
8111 0 : }
8112 :
8113 : static int
8114 0 : bdev_nvme_queued_next_fused_sge(void *ref, void **address, uint32_t *length)
8115 : {
8116 0 : struct nvme_bdev_io *bio = ref;
8117 : struct iovec *iov;
8118 :
8119 0 : assert(bio->fused_iovpos < bio->fused_iovcnt);
8120 :
8121 0 : iov = &bio->fused_iovs[bio->fused_iovpos];
8122 :
8123 0 : *address = iov->iov_base;
8124 0 : *length = iov->iov_len;
8125 :
8126 0 : if (bio->fused_iov_offset) {
8127 0 : assert(bio->fused_iov_offset <= iov->iov_len);
8128 0 : *address += bio->fused_iov_offset;
8129 0 : *length -= bio->fused_iov_offset;
8130 : }
8131 :
8132 0 : bio->fused_iov_offset += *length;
8133 0 : if (bio->fused_iov_offset == iov->iov_len) {
8134 0 : bio->fused_iovpos++;
8135 0 : bio->fused_iov_offset = 0;
8136 : }
8137 :
8138 0 : return 0;
8139 : }
8140 :
8141 : static int
8142 0 : bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8143 : void *md, uint64_t lba_count, uint64_t lba)
8144 : {
8145 : int rc;
8146 :
8147 0 : SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 " without PI check\n",
8148 : lba_count, lba);
8149 :
8150 0 : bio->iovs = iov;
8151 0 : bio->iovcnt = iovcnt;
8152 0 : bio->iovpos = 0;
8153 0 : bio->iov_offset = 0;
8154 :
8155 0 : rc = spdk_nvme_ns_cmd_readv_with_md(bio->io_path->nvme_ns->ns,
8156 0 : bio->io_path->qpair->qpair,
8157 : lba, lba_count,
8158 : bdev_nvme_no_pi_readv_done, bio, 0,
8159 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
8160 : md, 0, 0);
8161 :
8162 0 : if (rc != 0 && rc != -ENOMEM) {
8163 0 : SPDK_ERRLOG("no_pi_readv failed: rc = %d\n", rc);
8164 : }
8165 0 : return rc;
8166 : }
8167 :
8168 : static int
8169 3 : bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8170 : void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
8171 : struct spdk_memory_domain *domain, void *domain_ctx,
8172 : struct spdk_accel_sequence *seq)
8173 : {
8174 3 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8175 3 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8176 : int rc;
8177 :
8178 3 : SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8179 : lba_count, lba);
8180 :
8181 3 : bio->iovs = iov;
8182 3 : bio->iovcnt = iovcnt;
8183 3 : bio->iovpos = 0;
8184 3 : bio->iov_offset = 0;
8185 :
8186 3 : if (domain != NULL || seq != NULL) {
8187 1 : bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
8188 1 : bio->ext_opts.memory_domain = domain;
8189 1 : bio->ext_opts.memory_domain_ctx = domain_ctx;
8190 1 : bio->ext_opts.io_flags = flags;
8191 1 : bio->ext_opts.metadata = md;
8192 1 : bio->ext_opts.accel_sequence = seq;
8193 :
8194 1 : if (iovcnt == 1) {
8195 1 : rc = spdk_nvme_ns_cmd_read_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_readv_done,
8196 : bio, &bio->ext_opts);
8197 : } else {
8198 0 : rc = spdk_nvme_ns_cmd_readv_ext(ns, qpair, lba, lba_count,
8199 : bdev_nvme_readv_done, bio,
8200 : bdev_nvme_queued_reset_sgl,
8201 : bdev_nvme_queued_next_sge,
8202 : &bio->ext_opts);
8203 : }
8204 2 : } else if (iovcnt == 1) {
8205 2 : rc = spdk_nvme_ns_cmd_read_with_md(ns, qpair, iov[0].iov_base,
8206 : md, lba, lba_count, bdev_nvme_readv_done,
8207 : bio, flags, 0, 0);
8208 : } else {
8209 0 : rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count,
8210 : bdev_nvme_readv_done, bio, flags,
8211 : bdev_nvme_queued_reset_sgl,
8212 : bdev_nvme_queued_next_sge, md, 0, 0);
8213 : }
8214 :
8215 3 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
8216 0 : SPDK_ERRLOG("readv failed: rc = %d\n", rc);
8217 : }
8218 3 : return rc;
8219 : }
8220 :
8221 : static int
8222 25 : bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8223 : void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
8224 : struct spdk_memory_domain *domain, void *domain_ctx,
8225 : struct spdk_accel_sequence *seq,
8226 : union spdk_bdev_nvme_cdw12 cdw12, union spdk_bdev_nvme_cdw13 cdw13)
8227 : {
8228 25 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8229 25 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8230 : int rc;
8231 :
8232 25 : SPDK_DEBUGLOG(bdev_nvme, "write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8233 : lba_count, lba);
8234 :
8235 25 : bio->iovs = iov;
8236 25 : bio->iovcnt = iovcnt;
8237 25 : bio->iovpos = 0;
8238 25 : bio->iov_offset = 0;
8239 :
8240 25 : if (domain != NULL || seq != NULL) {
8241 0 : bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
8242 0 : bio->ext_opts.memory_domain = domain;
8243 0 : bio->ext_opts.memory_domain_ctx = domain_ctx;
8244 0 : bio->ext_opts.io_flags = flags | SPDK_NVME_IO_FLAGS_DIRECTIVE(cdw12.write.dtype);
8245 0 : bio->ext_opts.cdw13 = cdw13.raw;
8246 0 : bio->ext_opts.metadata = md;
8247 0 : bio->ext_opts.accel_sequence = seq;
8248 :
8249 0 : if (iovcnt == 1) {
8250 0 : rc = spdk_nvme_ns_cmd_write_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_writev_done,
8251 : bio, &bio->ext_opts);
8252 : } else {
8253 0 : rc = spdk_nvme_ns_cmd_writev_ext(ns, qpair, lba, lba_count,
8254 : bdev_nvme_writev_done, bio,
8255 : bdev_nvme_queued_reset_sgl,
8256 : bdev_nvme_queued_next_sge,
8257 : &bio->ext_opts);
8258 : }
8259 25 : } else if (iovcnt == 1) {
8260 25 : rc = spdk_nvme_ns_cmd_write_with_md(ns, qpair, iov[0].iov_base,
8261 : md, lba, lba_count, bdev_nvme_writev_done,
8262 : bio, flags, 0, 0);
8263 : } else {
8264 0 : rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
8265 : bdev_nvme_writev_done, bio, flags,
8266 : bdev_nvme_queued_reset_sgl,
8267 : bdev_nvme_queued_next_sge, md, 0, 0);
8268 : }
8269 :
8270 25 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
8271 0 : SPDK_ERRLOG("writev failed: rc = %d\n", rc);
8272 : }
8273 25 : return rc;
8274 : }
8275 :
8276 : static int
8277 0 : bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8278 : void *md, uint64_t lba_count, uint64_t zslba,
8279 : uint32_t flags)
8280 : {
8281 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8282 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8283 : int rc;
8284 :
8285 0 : SPDK_DEBUGLOG(bdev_nvme, "zone append %" PRIu64 " blocks to zone start lba %#" PRIx64 "\n",
8286 : lba_count, zslba);
8287 :
8288 0 : bio->iovs = iov;
8289 0 : bio->iovcnt = iovcnt;
8290 0 : bio->iovpos = 0;
8291 0 : bio->iov_offset = 0;
8292 :
8293 0 : if (iovcnt == 1) {
8294 0 : rc = spdk_nvme_zns_zone_append_with_md(ns, qpair, iov[0].iov_base, md, zslba,
8295 : lba_count,
8296 : bdev_nvme_zone_appendv_done, bio,
8297 : flags,
8298 : 0, 0);
8299 : } else {
8300 0 : rc = spdk_nvme_zns_zone_appendv_with_md(ns, qpair, zslba, lba_count,
8301 : bdev_nvme_zone_appendv_done, bio, flags,
8302 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
8303 : md, 0, 0);
8304 : }
8305 :
8306 0 : if (rc != 0 && rc != -ENOMEM) {
8307 0 : SPDK_ERRLOG("zone append failed: rc = %d\n", rc);
8308 : }
8309 0 : return rc;
8310 : }
8311 :
8312 : static int
8313 1 : bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8314 : void *md, uint64_t lba_count, uint64_t lba,
8315 : uint32_t flags)
8316 : {
8317 : int rc;
8318 :
8319 1 : SPDK_DEBUGLOG(bdev_nvme, "compare %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8320 : lba_count, lba);
8321 :
8322 1 : bio->iovs = iov;
8323 1 : bio->iovcnt = iovcnt;
8324 1 : bio->iovpos = 0;
8325 1 : bio->iov_offset = 0;
8326 :
8327 1 : rc = spdk_nvme_ns_cmd_comparev_with_md(bio->io_path->nvme_ns->ns,
8328 1 : bio->io_path->qpair->qpair,
8329 : lba, lba_count,
8330 : bdev_nvme_comparev_done, bio, flags,
8331 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
8332 : md, 0, 0);
8333 :
8334 1 : if (rc != 0 && rc != -ENOMEM) {
8335 0 : SPDK_ERRLOG("comparev failed: rc = %d\n", rc);
8336 : }
8337 1 : return rc;
8338 : }
8339 :
8340 : static int
8341 2 : bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt,
8342 : struct iovec *write_iov, int write_iovcnt,
8343 : void *md, uint64_t lba_count, uint64_t lba, uint32_t flags)
8344 : {
8345 2 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8346 2 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8347 2 : struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
8348 : int rc;
8349 :
8350 2 : SPDK_DEBUGLOG(bdev_nvme, "compare and write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8351 : lba_count, lba);
8352 :
8353 2 : bio->iovs = cmp_iov;
8354 2 : bio->iovcnt = cmp_iovcnt;
8355 2 : bio->iovpos = 0;
8356 2 : bio->iov_offset = 0;
8357 2 : bio->fused_iovs = write_iov;
8358 2 : bio->fused_iovcnt = write_iovcnt;
8359 2 : bio->fused_iovpos = 0;
8360 2 : bio->fused_iov_offset = 0;
8361 :
8362 2 : if (bdev_io->num_retries == 0) {
8363 2 : bio->first_fused_submitted = false;
8364 2 : bio->first_fused_completed = false;
8365 : }
8366 :
8367 2 : if (!bio->first_fused_submitted) {
8368 2 : flags |= SPDK_NVME_IO_FLAGS_FUSE_FIRST;
8369 2 : memset(&bio->cpl, 0, sizeof(bio->cpl));
8370 :
8371 2 : rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count,
8372 : bdev_nvme_comparev_and_writev_done, bio, flags,
8373 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, md, 0, 0);
8374 2 : if (rc == 0) {
8375 2 : bio->first_fused_submitted = true;
8376 2 : flags &= ~SPDK_NVME_IO_FLAGS_FUSE_FIRST;
8377 : } else {
8378 0 : if (rc != -ENOMEM) {
8379 0 : SPDK_ERRLOG("compare failed: rc = %d\n", rc);
8380 : }
8381 0 : return rc;
8382 : }
8383 : }
8384 :
8385 2 : flags |= SPDK_NVME_IO_FLAGS_FUSE_SECOND;
8386 :
8387 2 : rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
8388 : bdev_nvme_comparev_and_writev_done, bio, flags,
8389 : bdev_nvme_queued_reset_fused_sgl, bdev_nvme_queued_next_fused_sge, md, 0, 0);
8390 2 : if (rc != 0 && rc != -ENOMEM) {
8391 0 : SPDK_ERRLOG("write failed: rc = %d\n", rc);
8392 0 : rc = 0;
8393 : }
8394 :
8395 2 : return rc;
8396 : }
8397 :
8398 : static int
8399 1 : bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
8400 : {
8401 1 : struct spdk_nvme_dsm_range dsm_ranges[SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES];
8402 : struct spdk_nvme_dsm_range *range;
8403 : uint64_t offset, remaining;
8404 : uint64_t num_ranges_u64;
8405 : uint16_t num_ranges;
8406 : int rc;
8407 :
8408 1 : num_ranges_u64 = (num_blocks + SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS - 1) /
8409 : SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8410 1 : if (num_ranges_u64 > SPDK_COUNTOF(dsm_ranges)) {
8411 0 : SPDK_ERRLOG("Unmap request for %" PRIu64 " blocks is too large\n", num_blocks);
8412 0 : return -EINVAL;
8413 : }
8414 1 : num_ranges = (uint16_t)num_ranges_u64;
8415 :
8416 1 : offset = offset_blocks;
8417 1 : remaining = num_blocks;
8418 1 : range = &dsm_ranges[0];
8419 :
8420 : /* Fill max-size ranges until the remaining blocks fit into one range */
8421 1 : while (remaining > SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS) {
8422 0 : range->attributes.raw = 0;
8423 0 : range->length = SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8424 0 : range->starting_lba = offset;
8425 :
8426 0 : offset += SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8427 0 : remaining -= SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8428 0 : range++;
8429 : }
8430 :
8431 : /* Final range describes the remaining blocks */
8432 1 : range->attributes.raw = 0;
8433 1 : range->length = remaining;
8434 1 : range->starting_lba = offset;
8435 :
8436 1 : rc = spdk_nvme_ns_cmd_dataset_management(bio->io_path->nvme_ns->ns,
8437 1 : bio->io_path->qpair->qpair,
8438 : SPDK_NVME_DSM_ATTR_DEALLOCATE,
8439 : dsm_ranges, num_ranges,
8440 : bdev_nvme_queued_done, bio);
8441 :
8442 1 : return rc;
8443 : }
8444 :
8445 : static int
8446 0 : bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
8447 : {
8448 0 : if (num_blocks > UINT16_MAX + 1) {
8449 0 : SPDK_ERRLOG("NVMe write zeroes is limited to 16-bit block count\n");
8450 0 : return -EINVAL;
8451 : }
8452 :
8453 0 : return spdk_nvme_ns_cmd_write_zeroes(bio->io_path->nvme_ns->ns,
8454 0 : bio->io_path->qpair->qpair,
8455 : offset_blocks, num_blocks,
8456 : bdev_nvme_queued_done, bio,
8457 : 0);
8458 : }
8459 :
8460 : static int
8461 0 : bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones,
8462 : struct spdk_bdev_zone_info *info)
8463 : {
8464 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8465 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8466 0 : uint32_t zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
8467 0 : uint64_t zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
8468 0 : uint64_t total_zones = spdk_nvme_zns_ns_get_num_zones(ns);
8469 :
8470 0 : if (zone_id % zone_size != 0) {
8471 0 : return -EINVAL;
8472 : }
8473 :
8474 0 : if (num_zones > total_zones || !num_zones) {
8475 0 : return -EINVAL;
8476 : }
8477 :
8478 0 : assert(!bio->zone_report_buf);
8479 0 : bio->zone_report_buf = calloc(1, zone_report_bufsize);
8480 0 : if (!bio->zone_report_buf) {
8481 0 : return -ENOMEM;
8482 : }
8483 :
8484 0 : bio->handled_zones = 0;
8485 :
8486 0 : return spdk_nvme_zns_report_zones(ns, qpair, bio->zone_report_buf, zone_report_bufsize,
8487 : zone_id, SPDK_NVME_ZRA_LIST_ALL, true,
8488 : bdev_nvme_get_zone_info_done, bio);
8489 : }
8490 :
8491 : static int
8492 0 : bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
8493 : enum spdk_bdev_zone_action action)
8494 : {
8495 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8496 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8497 :
8498 0 : switch (action) {
8499 0 : case SPDK_BDEV_ZONE_CLOSE:
8500 0 : return spdk_nvme_zns_close_zone(ns, qpair, zone_id, false,
8501 : bdev_nvme_zone_management_done, bio);
8502 0 : case SPDK_BDEV_ZONE_FINISH:
8503 0 : return spdk_nvme_zns_finish_zone(ns, qpair, zone_id, false,
8504 : bdev_nvme_zone_management_done, bio);
8505 0 : case SPDK_BDEV_ZONE_OPEN:
8506 0 : return spdk_nvme_zns_open_zone(ns, qpair, zone_id, false,
8507 : bdev_nvme_zone_management_done, bio);
8508 0 : case SPDK_BDEV_ZONE_RESET:
8509 0 : return spdk_nvme_zns_reset_zone(ns, qpair, zone_id, false,
8510 : bdev_nvme_zone_management_done, bio);
8511 0 : case SPDK_BDEV_ZONE_OFFLINE:
8512 0 : return spdk_nvme_zns_offline_zone(ns, qpair, zone_id, false,
8513 : bdev_nvme_zone_management_done, bio);
8514 0 : default:
8515 0 : return -EINVAL;
8516 : }
8517 : }
8518 :
8519 : static void
8520 5 : bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
8521 : struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes)
8522 : {
8523 : struct nvme_io_path *io_path;
8524 : struct nvme_ctrlr *nvme_ctrlr;
8525 : uint32_t max_xfer_size;
8526 5 : int rc = -ENXIO;
8527 :
8528 : /* Choose the first ctrlr which is not failed. */
8529 8 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
8530 7 : nvme_ctrlr = io_path->qpair->ctrlr;
8531 :
8532 : /* We should skip any unavailable nvme_ctrlr rather than checking
8533 : * if the return value of spdk_nvme_ctrlr_cmd_admin_raw() is -ENXIO.
8534 : */
8535 7 : if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
8536 3 : continue;
8537 : }
8538 :
8539 4 : max_xfer_size = spdk_nvme_ctrlr_get_max_xfer_size(nvme_ctrlr->ctrlr);
8540 :
8541 4 : if (nbytes > max_xfer_size) {
8542 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8543 0 : rc = -EINVAL;
8544 0 : goto err;
8545 : }
8546 :
8547 4 : rc = spdk_nvme_ctrlr_cmd_admin_raw(nvme_ctrlr->ctrlr, cmd, buf, (uint32_t)nbytes,
8548 : bdev_nvme_admin_passthru_done, bio);
8549 4 : if (rc == 0) {
8550 4 : return;
8551 : }
8552 : }
8553 :
8554 1 : err:
8555 1 : bdev_nvme_admin_complete(bio, rc);
8556 : }
8557 :
8558 : static int
8559 0 : bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
8560 : void *buf, size_t nbytes)
8561 : {
8562 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8563 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8564 0 : uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8565 0 : struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8566 :
8567 0 : if (nbytes > max_xfer_size) {
8568 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8569 0 : return -EINVAL;
8570 : }
8571 :
8572 : /*
8573 : * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
8574 : * so fill it out automatically.
8575 : */
8576 0 : cmd->nsid = spdk_nvme_ns_get_id(ns);
8577 :
8578 0 : return spdk_nvme_ctrlr_cmd_io_raw(ctrlr, qpair, cmd, buf,
8579 : (uint32_t)nbytes, bdev_nvme_queued_done, bio);
8580 : }
8581 :
8582 : static int
8583 0 : bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
8584 : void *buf, size_t nbytes, void *md_buf, size_t md_len)
8585 : {
8586 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8587 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8588 0 : size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
8589 0 : uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8590 0 : struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8591 :
8592 0 : if (nbytes > max_xfer_size) {
8593 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8594 0 : return -EINVAL;
8595 : }
8596 :
8597 0 : if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
8598 0 : SPDK_ERRLOG("invalid meta data buffer size\n");
8599 0 : return -EINVAL;
8600 : }
8601 :
8602 : /*
8603 : * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
8604 : * so fill it out automatically.
8605 : */
8606 0 : cmd->nsid = spdk_nvme_ns_get_id(ns);
8607 :
8608 0 : return spdk_nvme_ctrlr_cmd_io_raw_with_md(ctrlr, qpair, cmd, buf,
8609 : (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio);
8610 : }
8611 :
8612 : static int
8613 0 : bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio,
8614 : struct spdk_nvme_cmd *cmd, struct iovec *iov, int iovcnt,
8615 : size_t nbytes, void *md_buf, size_t md_len)
8616 : {
8617 0 : struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8618 0 : struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8619 0 : size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
8620 0 : uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8621 0 : struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8622 :
8623 0 : bio->iovs = iov;
8624 0 : bio->iovcnt = iovcnt;
8625 0 : bio->iovpos = 0;
8626 0 : bio->iov_offset = 0;
8627 :
8628 0 : if (nbytes > max_xfer_size) {
8629 0 : SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8630 0 : return -EINVAL;
8631 : }
8632 :
8633 0 : if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
8634 0 : SPDK_ERRLOG("invalid meta data buffer size\n");
8635 0 : return -EINVAL;
8636 : }
8637 :
8638 : /*
8639 : * Each NVMe bdev is a specific namespace, and all NVMe I/O commands
8640 : * require a nsid, so fill it out automatically.
8641 : */
8642 0 : cmd->nsid = spdk_nvme_ns_get_id(ns);
8643 :
8644 0 : return spdk_nvme_ctrlr_cmd_iov_raw_with_md(
8645 : ctrlr, qpair, cmd, (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio,
8646 : bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge);
8647 : }
8648 :
8649 : static void
8650 6 : bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
8651 : struct nvme_bdev_io *bio_to_abort)
8652 : {
8653 : struct nvme_io_path *io_path;
8654 6 : int rc = 0;
8655 :
8656 6 : rc = bdev_nvme_abort_retry_io(nbdev_ch, bio_to_abort);
8657 6 : if (rc == 0) {
8658 1 : bdev_nvme_admin_complete(bio, 0);
8659 1 : return;
8660 : }
8661 :
8662 5 : io_path = bio_to_abort->io_path;
8663 5 : if (io_path != NULL) {
8664 3 : rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
8665 3 : io_path->qpair->qpair,
8666 : bio_to_abort,
8667 : bdev_nvme_abort_done, bio);
8668 : } else {
8669 3 : STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
8670 2 : rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
8671 : NULL,
8672 : bio_to_abort,
8673 : bdev_nvme_abort_done, bio);
8674 :
8675 2 : if (rc != -ENOENT) {
8676 1 : break;
8677 : }
8678 : }
8679 : }
8680 :
8681 5 : if (rc != 0) {
8682 : /* If no command was found or there was any error, complete the abort
8683 : * request with failure.
8684 : */
8685 2 : bdev_nvme_admin_complete(bio, rc);
8686 : }
8687 : }
8688 :
8689 : static int
8690 0 : bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks, uint64_t src_offset_blocks,
8691 : uint64_t num_blocks)
8692 : {
8693 0 : struct spdk_nvme_scc_source_range range = {
8694 : .slba = src_offset_blocks,
8695 0 : .nlb = num_blocks - 1
8696 : };
8697 :
8698 0 : return spdk_nvme_ns_cmd_copy(bio->io_path->nvme_ns->ns,
8699 0 : bio->io_path->qpair->qpair,
8700 : &range, 1, dst_offset_blocks,
8701 : bdev_nvme_queued_done, bio);
8702 : }
8703 :
8704 : static void
8705 0 : bdev_nvme_opts_config_json(struct spdk_json_write_ctx *w)
8706 : {
8707 : const char *action;
8708 : uint32_t i;
8709 :
8710 0 : if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET) {
8711 0 : action = "reset";
8712 0 : } else if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT) {
8713 0 : action = "abort";
8714 : } else {
8715 0 : action = "none";
8716 : }
8717 :
8718 0 : spdk_json_write_object_begin(w);
8719 :
8720 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_set_options");
8721 :
8722 0 : spdk_json_write_named_object_begin(w, "params");
8723 0 : spdk_json_write_named_string(w, "action_on_timeout", action);
8724 0 : spdk_json_write_named_uint64(w, "timeout_us", g_opts.timeout_us);
8725 0 : spdk_json_write_named_uint64(w, "timeout_admin_us", g_opts.timeout_admin_us);
8726 0 : spdk_json_write_named_uint32(w, "keep_alive_timeout_ms", g_opts.keep_alive_timeout_ms);
8727 0 : spdk_json_write_named_uint32(w, "arbitration_burst", g_opts.arbitration_burst);
8728 0 : spdk_json_write_named_uint32(w, "low_priority_weight", g_opts.low_priority_weight);
8729 0 : spdk_json_write_named_uint32(w, "medium_priority_weight", g_opts.medium_priority_weight);
8730 0 : spdk_json_write_named_uint32(w, "high_priority_weight", g_opts.high_priority_weight);
8731 0 : spdk_json_write_named_uint64(w, "nvme_adminq_poll_period_us", g_opts.nvme_adminq_poll_period_us);
8732 0 : spdk_json_write_named_uint64(w, "nvme_ioq_poll_period_us", g_opts.nvme_ioq_poll_period_us);
8733 0 : spdk_json_write_named_uint32(w, "io_queue_requests", g_opts.io_queue_requests);
8734 0 : spdk_json_write_named_bool(w, "delay_cmd_submit", g_opts.delay_cmd_submit);
8735 0 : spdk_json_write_named_uint32(w, "transport_retry_count", g_opts.transport_retry_count);
8736 0 : spdk_json_write_named_int32(w, "bdev_retry_count", g_opts.bdev_retry_count);
8737 0 : spdk_json_write_named_uint8(w, "transport_ack_timeout", g_opts.transport_ack_timeout);
8738 0 : spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", g_opts.ctrlr_loss_timeout_sec);
8739 0 : spdk_json_write_named_uint32(w, "reconnect_delay_sec", g_opts.reconnect_delay_sec);
8740 0 : spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec", g_opts.fast_io_fail_timeout_sec);
8741 0 : spdk_json_write_named_bool(w, "disable_auto_failback", g_opts.disable_auto_failback);
8742 0 : spdk_json_write_named_bool(w, "generate_uuids", g_opts.generate_uuids);
8743 0 : spdk_json_write_named_uint8(w, "transport_tos", g_opts.transport_tos);
8744 0 : spdk_json_write_named_bool(w, "nvme_error_stat", g_opts.nvme_error_stat);
8745 0 : spdk_json_write_named_uint32(w, "rdma_srq_size", g_opts.rdma_srq_size);
8746 0 : spdk_json_write_named_bool(w, "io_path_stat", g_opts.io_path_stat);
8747 0 : spdk_json_write_named_bool(w, "allow_accel_sequence", g_opts.allow_accel_sequence);
8748 0 : spdk_json_write_named_uint32(w, "rdma_max_cq_size", g_opts.rdma_max_cq_size);
8749 0 : spdk_json_write_named_uint16(w, "rdma_cm_event_timeout_ms", g_opts.rdma_cm_event_timeout_ms);
8750 0 : spdk_json_write_named_array_begin(w, "dhchap_digests");
8751 0 : for (i = 0; i < 32; ++i) {
8752 0 : if (g_opts.dhchap_digests & SPDK_BIT(i)) {
8753 0 : spdk_json_write_string(w, spdk_nvme_dhchap_get_digest_name(i));
8754 : }
8755 : }
8756 0 : spdk_json_write_array_end(w);
8757 0 : spdk_json_write_named_array_begin(w, "dhchap_dhgroups");
8758 0 : for (i = 0; i < 32; ++i) {
8759 0 : if (g_opts.dhchap_dhgroups & SPDK_BIT(i)) {
8760 0 : spdk_json_write_string(w, spdk_nvme_dhchap_get_dhgroup_name(i));
8761 : }
8762 : }
8763 :
8764 0 : spdk_json_write_array_end(w);
8765 0 : spdk_json_write_object_end(w);
8766 :
8767 0 : spdk_json_write_object_end(w);
8768 0 : }
8769 :
8770 : static void
8771 0 : bdev_nvme_discovery_config_json(struct spdk_json_write_ctx *w, struct discovery_ctx *ctx)
8772 : {
8773 0 : struct spdk_nvme_transport_id trid;
8774 :
8775 0 : spdk_json_write_object_begin(w);
8776 :
8777 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_start_discovery");
8778 :
8779 0 : spdk_json_write_named_object_begin(w, "params");
8780 0 : spdk_json_write_named_string(w, "name", ctx->name);
8781 0 : spdk_json_write_named_string(w, "hostnqn", ctx->hostnqn);
8782 :
8783 0 : trid = ctx->trid;
8784 0 : memset(trid.subnqn, 0, sizeof(trid.subnqn));
8785 0 : nvme_bdev_dump_trid_json(&trid, w);
8786 :
8787 0 : spdk_json_write_named_bool(w, "wait_for_attach", ctx->wait_for_attach);
8788 0 : spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", ctx->bdev_opts.ctrlr_loss_timeout_sec);
8789 0 : spdk_json_write_named_uint32(w, "reconnect_delay_sec", ctx->bdev_opts.reconnect_delay_sec);
8790 0 : spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
8791 : ctx->bdev_opts.fast_io_fail_timeout_sec);
8792 0 : spdk_json_write_object_end(w);
8793 :
8794 0 : spdk_json_write_object_end(w);
8795 0 : }
8796 :
8797 : #ifdef SPDK_CONFIG_NVME_CUSE
8798 : static void
8799 0 : nvme_ctrlr_cuse_config_json(struct spdk_json_write_ctx *w,
8800 : struct nvme_ctrlr *nvme_ctrlr)
8801 0 : {
8802 0 : size_t cuse_name_size = 128;
8803 0 : char cuse_name[cuse_name_size];
8804 :
8805 0 : if (spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr,
8806 : cuse_name, &cuse_name_size) != 0) {
8807 0 : return;
8808 : }
8809 :
8810 0 : spdk_json_write_object_begin(w);
8811 :
8812 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_cuse_register");
8813 :
8814 0 : spdk_json_write_named_object_begin(w, "params");
8815 0 : spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
8816 0 : spdk_json_write_object_end(w);
8817 :
8818 0 : spdk_json_write_object_end(w);
8819 : }
8820 : #endif
8821 :
8822 : static void
8823 0 : nvme_ctrlr_config_json(struct spdk_json_write_ctx *w,
8824 : struct nvme_ctrlr *nvme_ctrlr,
8825 : struct nvme_path_id *path_id)
8826 : {
8827 : struct spdk_nvme_transport_id *trid;
8828 : const struct spdk_nvme_ctrlr_opts *opts;
8829 :
8830 0 : if (nvme_ctrlr->opts.from_discovery_service) {
8831 : /* Do not emit an RPC for this - it will be implicitly
8832 : * covered by a separate bdev_nvme_start_discovery or
8833 : * bdev_nvme_start_mdns_discovery RPC.
8834 : */
8835 0 : return;
8836 : }
8837 :
8838 0 : trid = &path_id->trid;
8839 :
8840 0 : spdk_json_write_object_begin(w);
8841 :
8842 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_attach_controller");
8843 :
8844 0 : spdk_json_write_named_object_begin(w, "params");
8845 0 : spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
8846 0 : nvme_bdev_dump_trid_json(trid, w);
8847 0 : spdk_json_write_named_bool(w, "prchk_reftag",
8848 0 : (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_REFTAG) != 0);
8849 0 : spdk_json_write_named_bool(w, "prchk_guard",
8850 0 : (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_GUARD) != 0);
8851 0 : spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", nvme_ctrlr->opts.ctrlr_loss_timeout_sec);
8852 0 : spdk_json_write_named_uint32(w, "reconnect_delay_sec", nvme_ctrlr->opts.reconnect_delay_sec);
8853 0 : spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
8854 : nvme_ctrlr->opts.fast_io_fail_timeout_sec);
8855 0 : if (nvme_ctrlr->psk != NULL) {
8856 0 : spdk_json_write_named_string(w, "psk", spdk_key_get_name(nvme_ctrlr->psk));
8857 : }
8858 0 : if (nvme_ctrlr->dhchap_key != NULL) {
8859 0 : spdk_json_write_named_string(w, "dhchap_key",
8860 : spdk_key_get_name(nvme_ctrlr->dhchap_key));
8861 : }
8862 0 : if (nvme_ctrlr->dhchap_ctrlr_key != NULL) {
8863 0 : spdk_json_write_named_string(w, "dhchap_ctrlr_key",
8864 : spdk_key_get_name(nvme_ctrlr->dhchap_ctrlr_key));
8865 : }
8866 0 : opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
8867 0 : spdk_json_write_named_string(w, "hostnqn", opts->hostnqn);
8868 0 : spdk_json_write_named_bool(w, "hdgst", opts->header_digest);
8869 0 : spdk_json_write_named_bool(w, "ddgst", opts->data_digest);
8870 0 : if (opts->src_addr[0] != '\0') {
8871 0 : spdk_json_write_named_string(w, "hostaddr", opts->src_addr);
8872 : }
8873 0 : if (opts->src_svcid[0] != '\0') {
8874 0 : spdk_json_write_named_string(w, "hostsvcid", opts->src_svcid);
8875 : }
8876 :
8877 0 : if (nvme_ctrlr->opts.multipath) {
8878 0 : spdk_json_write_named_string(w, "multipath", "multipath");
8879 : }
8880 0 : spdk_json_write_object_end(w);
8881 :
8882 0 : spdk_json_write_object_end(w);
8883 : }
8884 :
8885 : static void
8886 0 : bdev_nvme_hotplug_config_json(struct spdk_json_write_ctx *w)
8887 : {
8888 0 : spdk_json_write_object_begin(w);
8889 0 : spdk_json_write_named_string(w, "method", "bdev_nvme_set_hotplug");
8890 :
8891 0 : spdk_json_write_named_object_begin(w, "params");
8892 0 : spdk_json_write_named_uint64(w, "period_us", g_nvme_hotplug_poll_period_us);
8893 0 : spdk_json_write_named_bool(w, "enable", g_nvme_hotplug_enabled);
8894 0 : spdk_json_write_object_end(w);
8895 :
8896 0 : spdk_json_write_object_end(w);
8897 0 : }
8898 :
8899 : static int
8900 0 : bdev_nvme_config_json(struct spdk_json_write_ctx *w)
8901 : {
8902 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
8903 : struct nvme_ctrlr *nvme_ctrlr;
8904 : struct discovery_ctx *ctx;
8905 : struct nvme_path_id *path_id;
8906 :
8907 0 : bdev_nvme_opts_config_json(w);
8908 :
8909 0 : pthread_mutex_lock(&g_bdev_nvme_mutex);
8910 :
8911 0 : TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
8912 0 : TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
8913 0 : path_id = nvme_ctrlr->active_path_id;
8914 0 : assert(path_id == TAILQ_FIRST(&nvme_ctrlr->trids));
8915 0 : nvme_ctrlr_config_json(w, nvme_ctrlr, path_id);
8916 :
8917 0 : path_id = TAILQ_NEXT(path_id, link);
8918 0 : while (path_id != NULL) {
8919 0 : nvme_ctrlr_config_json(w, nvme_ctrlr, path_id);
8920 0 : path_id = TAILQ_NEXT(path_id, link);
8921 : }
8922 :
8923 : #ifdef SPDK_CONFIG_NVME_CUSE
8924 0 : nvme_ctrlr_cuse_config_json(w, nvme_ctrlr);
8925 : #endif
8926 : }
8927 : }
8928 :
8929 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
8930 0 : if (!ctx->from_mdns_discovery_service) {
8931 0 : bdev_nvme_discovery_config_json(w, ctx);
8932 : }
8933 : }
8934 :
8935 0 : bdev_nvme_mdns_discovery_config_json(w);
8936 :
8937 : /* Dump as last parameter to give all NVMe bdevs chance to be constructed
8938 : * before enabling hotplug poller.
8939 : */
8940 0 : bdev_nvme_hotplug_config_json(w);
8941 :
8942 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
8943 0 : return 0;
8944 : }
8945 :
8946 : struct spdk_nvme_ctrlr *
8947 1 : bdev_nvme_get_ctrlr(struct spdk_bdev *bdev)
8948 : {
8949 : struct nvme_bdev *nbdev;
8950 : struct nvme_ns *nvme_ns;
8951 :
8952 1 : if (!bdev || bdev->module != &nvme_if) {
8953 0 : return NULL;
8954 : }
8955 :
8956 1 : nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
8957 1 : nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
8958 1 : assert(nvme_ns != NULL);
8959 :
8960 1 : return nvme_ns->ctrlr->ctrlr;
8961 : }
8962 :
8963 : static bool
8964 12 : nvme_io_path_is_current(struct nvme_io_path *io_path)
8965 : {
8966 : const struct nvme_bdev_channel *nbdev_ch;
8967 : bool current;
8968 :
8969 12 : if (!nvme_io_path_is_available(io_path)) {
8970 4 : return false;
8971 : }
8972 :
8973 8 : nbdev_ch = io_path->nbdev_ch;
8974 8 : if (nbdev_ch == NULL) {
8975 1 : current = false;
8976 7 : } else if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
8977 3 : struct nvme_io_path *optimized_io_path = NULL;
8978 :
8979 6 : STAILQ_FOREACH(optimized_io_path, &nbdev_ch->io_path_list, stailq) {
8980 5 : if (optimized_io_path->nvme_ns->ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE) {
8981 2 : break;
8982 : }
8983 : }
8984 :
8985 : /* A non-optimized path is only current if there are no optimized paths. */
8986 3 : current = (io_path->nvme_ns->ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE) ||
8987 : (optimized_io_path == NULL);
8988 : } else {
8989 4 : if (nbdev_ch->current_io_path) {
8990 1 : current = (io_path == nbdev_ch->current_io_path);
8991 : } else {
8992 : struct nvme_io_path *first_path;
8993 :
8994 : /* We arrived here as there are no optimized paths for active-passive
8995 : * mode. Check if this io_path is the first one available on the list.
8996 : */
8997 3 : current = false;
8998 3 : STAILQ_FOREACH(first_path, &nbdev_ch->io_path_list, stailq) {
8999 3 : if (nvme_io_path_is_available(first_path)) {
9000 3 : current = (io_path == first_path);
9001 3 : break;
9002 : }
9003 : }
9004 : }
9005 : }
9006 :
9007 8 : return current;
9008 : }
9009 :
9010 : static struct nvme_ctrlr *
9011 0 : bdev_nvme_next_ctrlr_unsafe(struct nvme_bdev_ctrlr *nbdev_ctrlr, struct nvme_ctrlr *prev)
9012 : {
9013 : struct nvme_ctrlr *next;
9014 :
9015 : /* Must be called under g_bdev_nvme_mutex */
9016 0 : next = prev != NULL ? TAILQ_NEXT(prev, tailq) : TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
9017 0 : while (next != NULL) {
9018 : /* ref can be 0 when the ctrlr was released, but hasn't been detached yet */
9019 0 : pthread_mutex_lock(&next->mutex);
9020 0 : if (next->ref > 0) {
9021 0 : next->ref++;
9022 0 : pthread_mutex_unlock(&next->mutex);
9023 0 : return next;
9024 : }
9025 :
9026 0 : pthread_mutex_unlock(&next->mutex);
9027 0 : next = TAILQ_NEXT(next, tailq);
9028 : }
9029 :
9030 0 : return NULL;
9031 : }
9032 :
9033 : struct bdev_nvme_set_keys_ctx {
9034 : struct nvme_ctrlr *nctrlr;
9035 : struct spdk_key *dhchap_key;
9036 : struct spdk_key *dhchap_ctrlr_key;
9037 : struct spdk_thread *thread;
9038 : bdev_nvme_set_keys_cb cb_fn;
9039 : void *cb_ctx;
9040 : int status;
9041 : };
9042 :
9043 : static void
9044 0 : bdev_nvme_free_set_keys_ctx(struct bdev_nvme_set_keys_ctx *ctx)
9045 : {
9046 0 : if (ctx == NULL) {
9047 0 : return;
9048 : }
9049 :
9050 0 : spdk_keyring_put_key(ctx->dhchap_key);
9051 0 : spdk_keyring_put_key(ctx->dhchap_ctrlr_key);
9052 0 : free(ctx);
9053 : }
9054 :
9055 : static void
9056 0 : _bdev_nvme_set_keys_done(void *_ctx)
9057 : {
9058 0 : struct bdev_nvme_set_keys_ctx *ctx = _ctx;
9059 :
9060 0 : ctx->cb_fn(ctx->cb_ctx, ctx->status);
9061 :
9062 0 : if (ctx->nctrlr != NULL) {
9063 0 : nvme_ctrlr_put_ref(ctx->nctrlr);
9064 : }
9065 0 : bdev_nvme_free_set_keys_ctx(ctx);
9066 0 : }
9067 :
9068 : static void
9069 0 : bdev_nvme_set_keys_done(struct bdev_nvme_set_keys_ctx *ctx, int status)
9070 : {
9071 0 : ctx->status = status;
9072 0 : spdk_thread_exec_msg(ctx->thread, _bdev_nvme_set_keys_done, ctx);
9073 0 : }
9074 :
9075 : static void bdev_nvme_authenticate_ctrlr(struct bdev_nvme_set_keys_ctx *ctx);
9076 :
9077 : static void
9078 0 : bdev_nvme_authenticate_ctrlr_continue(struct bdev_nvme_set_keys_ctx *ctx)
9079 : {
9080 : struct nvme_ctrlr *next;
9081 :
9082 0 : pthread_mutex_lock(&g_bdev_nvme_mutex);
9083 0 : next = bdev_nvme_next_ctrlr_unsafe(NULL, ctx->nctrlr);
9084 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
9085 :
9086 0 : nvme_ctrlr_put_ref(ctx->nctrlr);
9087 0 : ctx->nctrlr = next;
9088 :
9089 0 : if (next == NULL) {
9090 0 : bdev_nvme_set_keys_done(ctx, 0);
9091 : } else {
9092 0 : bdev_nvme_authenticate_ctrlr(ctx);
9093 : }
9094 0 : }
9095 :
9096 : static void
9097 0 : bdev_nvme_authenticate_qpairs_done(struct spdk_io_channel_iter *i, int status)
9098 : {
9099 0 : struct bdev_nvme_set_keys_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
9100 :
9101 0 : if (status != 0) {
9102 0 : bdev_nvme_set_keys_done(ctx, status);
9103 0 : return;
9104 : }
9105 0 : bdev_nvme_authenticate_ctrlr_continue(ctx);
9106 : }
9107 :
9108 : static void
9109 0 : bdev_nvme_authenticate_qpair_done(void *ctx, int status)
9110 : {
9111 0 : spdk_for_each_channel_continue(ctx, status);
9112 0 : }
9113 :
9114 : static void
9115 0 : bdev_nvme_authenticate_qpair(struct spdk_io_channel_iter *i)
9116 : {
9117 0 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
9118 0 : struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(ch);
9119 0 : struct nvme_qpair *qpair = ctrlr_ch->qpair;
9120 : int rc;
9121 :
9122 0 : if (!nvme_qpair_is_connected(qpair)) {
9123 0 : spdk_for_each_channel_continue(i, 0);
9124 0 : return;
9125 : }
9126 :
9127 0 : rc = spdk_nvme_qpair_authenticate(qpair->qpair, bdev_nvme_authenticate_qpair_done, i);
9128 0 : if (rc != 0) {
9129 0 : spdk_for_each_channel_continue(i, rc);
9130 : }
9131 : }
9132 :
9133 : static void
9134 0 : bdev_nvme_authenticate_ctrlr_done(void *_ctx, int status)
9135 : {
9136 0 : struct bdev_nvme_set_keys_ctx *ctx = _ctx;
9137 :
9138 0 : if (status != 0) {
9139 0 : bdev_nvme_set_keys_done(ctx, status);
9140 0 : return;
9141 : }
9142 :
9143 0 : spdk_for_each_channel(ctx->nctrlr, bdev_nvme_authenticate_qpair, ctx,
9144 : bdev_nvme_authenticate_qpairs_done);
9145 : }
9146 :
9147 : static void
9148 0 : bdev_nvme_authenticate_ctrlr(struct bdev_nvme_set_keys_ctx *ctx)
9149 : {
9150 0 : struct spdk_nvme_ctrlr_key_opts opts = {};
9151 0 : struct nvme_ctrlr *nctrlr = ctx->nctrlr;
9152 : int rc;
9153 :
9154 0 : opts.size = SPDK_SIZEOF(&opts, dhchap_ctrlr_key);
9155 0 : opts.dhchap_key = ctx->dhchap_key;
9156 0 : opts.dhchap_ctrlr_key = ctx->dhchap_ctrlr_key;
9157 0 : rc = spdk_nvme_ctrlr_set_keys(nctrlr->ctrlr, &opts);
9158 0 : if (rc != 0) {
9159 0 : bdev_nvme_set_keys_done(ctx, rc);
9160 0 : return;
9161 : }
9162 :
9163 0 : if (ctx->dhchap_key != NULL) {
9164 0 : rc = spdk_nvme_ctrlr_authenticate(nctrlr->ctrlr,
9165 : bdev_nvme_authenticate_ctrlr_done, ctx);
9166 0 : if (rc != 0) {
9167 0 : bdev_nvme_set_keys_done(ctx, rc);
9168 : }
9169 : } else {
9170 0 : bdev_nvme_authenticate_ctrlr_continue(ctx);
9171 : }
9172 : }
9173 :
9174 : int
9175 0 : bdev_nvme_set_keys(const char *name, const char *dhchap_key, const char *dhchap_ctrlr_key,
9176 : bdev_nvme_set_keys_cb cb_fn, void *cb_ctx)
9177 : {
9178 : struct bdev_nvme_set_keys_ctx *ctx;
9179 : struct nvme_bdev_ctrlr *nbdev_ctrlr;
9180 : struct nvme_ctrlr *nctrlr;
9181 :
9182 0 : ctx = calloc(1, sizeof(*ctx));
9183 0 : if (ctx == NULL) {
9184 0 : return -ENOMEM;
9185 : }
9186 :
9187 0 : if (dhchap_key != NULL) {
9188 0 : ctx->dhchap_key = spdk_keyring_get_key(dhchap_key);
9189 0 : if (ctx->dhchap_key == NULL) {
9190 0 : SPDK_ERRLOG("Could not find key %s for bdev %s\n", dhchap_key, name);
9191 0 : bdev_nvme_free_set_keys_ctx(ctx);
9192 0 : return -ENOKEY;
9193 : }
9194 : }
9195 0 : if (dhchap_ctrlr_key != NULL) {
9196 0 : ctx->dhchap_ctrlr_key = spdk_keyring_get_key(dhchap_ctrlr_key);
9197 0 : if (ctx->dhchap_ctrlr_key == NULL) {
9198 0 : SPDK_ERRLOG("Could not find key %s for bdev %s\n", dhchap_ctrlr_key, name);
9199 0 : bdev_nvme_free_set_keys_ctx(ctx);
9200 0 : return -ENOKEY;
9201 : }
9202 : }
9203 :
9204 0 : pthread_mutex_lock(&g_bdev_nvme_mutex);
9205 0 : nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
9206 0 : if (nbdev_ctrlr == NULL) {
9207 0 : SPDK_ERRLOG("Could not find bdev_ctrlr %s\n", name);
9208 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
9209 0 : bdev_nvme_free_set_keys_ctx(ctx);
9210 0 : return -ENODEV;
9211 : }
9212 0 : nctrlr = bdev_nvme_next_ctrlr_unsafe(nbdev_ctrlr, NULL);
9213 0 : if (nctrlr == NULL) {
9214 0 : SPDK_ERRLOG("Could not find any nvme_ctrlrs on bdev_ctrlr %s\n", name);
9215 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
9216 0 : bdev_nvme_free_set_keys_ctx(ctx);
9217 0 : return -ENODEV;
9218 : }
9219 0 : pthread_mutex_unlock(&g_bdev_nvme_mutex);
9220 :
9221 0 : ctx->nctrlr = nctrlr;
9222 0 : ctx->cb_fn = cb_fn;
9223 0 : ctx->cb_ctx = cb_ctx;
9224 0 : ctx->thread = spdk_get_thread();
9225 :
9226 0 : bdev_nvme_authenticate_ctrlr(ctx);
9227 :
9228 0 : return 0;
9229 : }
9230 :
9231 : void
9232 0 : nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path)
9233 : {
9234 0 : struct nvme_ns *nvme_ns = io_path->nvme_ns;
9235 0 : struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
9236 : const struct spdk_nvme_ctrlr_data *cdata;
9237 : const struct spdk_nvme_transport_id *trid;
9238 : const char *adrfam_str;
9239 :
9240 0 : spdk_json_write_object_begin(w);
9241 :
9242 0 : spdk_json_write_named_string(w, "bdev_name", nvme_ns->bdev->disk.name);
9243 :
9244 0 : cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
9245 0 : trid = spdk_nvme_ctrlr_get_transport_id(nvme_ctrlr->ctrlr);
9246 :
9247 0 : spdk_json_write_named_uint32(w, "cntlid", cdata->cntlid);
9248 0 : spdk_json_write_named_bool(w, "current", nvme_io_path_is_current(io_path));
9249 0 : spdk_json_write_named_bool(w, "connected", nvme_qpair_is_connected(io_path->qpair));
9250 0 : spdk_json_write_named_bool(w, "accessible", nvme_ns_is_accessible(nvme_ns));
9251 :
9252 0 : spdk_json_write_named_object_begin(w, "transport");
9253 0 : spdk_json_write_named_string(w, "trtype", trid->trstring);
9254 0 : spdk_json_write_named_string(w, "traddr", trid->traddr);
9255 0 : if (trid->trsvcid[0] != '\0') {
9256 0 : spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
9257 : }
9258 0 : adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
9259 0 : if (adrfam_str) {
9260 0 : spdk_json_write_named_string(w, "adrfam", adrfam_str);
9261 : }
9262 0 : spdk_json_write_object_end(w);
9263 :
9264 0 : spdk_json_write_object_end(w);
9265 0 : }
9266 :
9267 : void
9268 0 : bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w)
9269 : {
9270 : struct discovery_ctx *ctx;
9271 : struct discovery_entry_ctx *entry_ctx;
9272 :
9273 0 : spdk_json_write_array_begin(w);
9274 0 : TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
9275 0 : spdk_json_write_object_begin(w);
9276 0 : spdk_json_write_named_string(w, "name", ctx->name);
9277 :
9278 0 : spdk_json_write_named_object_begin(w, "trid");
9279 0 : nvme_bdev_dump_trid_json(&ctx->trid, w);
9280 0 : spdk_json_write_object_end(w);
9281 :
9282 0 : spdk_json_write_named_array_begin(w, "referrals");
9283 0 : TAILQ_FOREACH(entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
9284 0 : spdk_json_write_object_begin(w);
9285 0 : spdk_json_write_named_object_begin(w, "trid");
9286 0 : nvme_bdev_dump_trid_json(&entry_ctx->trid, w);
9287 0 : spdk_json_write_object_end(w);
9288 0 : spdk_json_write_object_end(w);
9289 : }
9290 0 : spdk_json_write_array_end(w);
9291 :
9292 0 : spdk_json_write_object_end(w);
9293 : }
9294 0 : spdk_json_write_array_end(w);
9295 0 : }
9296 :
9297 1 : SPDK_LOG_REGISTER_COMPONENT(bdev_nvme)
9298 :
9299 : static void
9300 0 : bdev_nvme_trace(void)
9301 : {
9302 0 : struct spdk_trace_tpoint_opts opts[] = {
9303 : {
9304 : "BDEV_NVME_IO_START", TRACE_BDEV_NVME_IO_START,
9305 : OWNER_TYPE_NONE, OBJECT_BDEV_NVME_IO, 1,
9306 : {{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
9307 : },
9308 : {
9309 : "BDEV_NVME_IO_DONE", TRACE_BDEV_NVME_IO_DONE,
9310 : OWNER_TYPE_NONE, OBJECT_BDEV_NVME_IO, 0,
9311 : {{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
9312 : }
9313 : };
9314 :
9315 :
9316 0 : spdk_trace_register_object(OBJECT_BDEV_NVME_IO, 'N');
9317 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
9318 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
9319 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
9320 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
9321 0 : spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
9322 0 : }
9323 1 : SPDK_TRACE_REGISTER_FN(bdev_nvme_trace, "bdev_nvme", TRACE_GROUP_BDEV_NVME)
|
