xref: /spdk/module/bdev/nvme/bdev_nvme.h (revision 60241941e6cfa4fc04cfcf6840c79f941ccf85d0)
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) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5  *   Copyright (c) 2022 Dell Inc, or its subsidiaries. All rights reserved.
6  */
7 
8 #ifndef SPDK_BDEV_NVME_H
9 #define SPDK_BDEV_NVME_H
10 
11 #include "spdk/stdinc.h"
12 
13 #include "spdk/queue.h"
14 #include "spdk/nvme.h"
15 #include "spdk/bdev_module.h"
16 #include "spdk/module/bdev/nvme.h"
17 #include "spdk/jsonrpc.h"
18 
19 TAILQ_HEAD(nvme_bdev_ctrlrs, nvme_bdev_ctrlr);
20 extern struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs;
21 extern pthread_mutex_t g_bdev_nvme_mutex;
22 extern bool g_bdev_nvme_module_finish;
23 extern struct spdk_thread *g_bdev_nvme_init_thread;
24 
25 #define NVME_MAX_CONTROLLERS 1024
26 
27 typedef void (*spdk_bdev_nvme_start_discovery_fn)(void *ctx, int status);
28 typedef void (*spdk_bdev_nvme_stop_discovery_fn)(void *ctx);
29 
30 struct nvme_async_probe_ctx {
31 	struct spdk_nvme_probe_ctx *probe_ctx;
32 	const char *base_name;
33 	const char **names;
34 	uint32_t max_bdevs;
35 	uint32_t reported_bdevs;
36 	struct spdk_poller *poller;
37 	struct spdk_nvme_transport_id trid;
38 	struct spdk_bdev_nvme_ctrlr_opts bdev_opts;
39 	struct spdk_nvme_ctrlr_opts drv_opts;
40 	spdk_bdev_nvme_create_cb cb_fn;
41 	void *cb_ctx;
42 	uint32_t populates_in_progress;
43 	bool ctrlr_attached;
44 	bool probe_done;
45 	bool namespaces_populated;
46 };
47 
48 struct nvme_ns {
49 	uint32_t			id;
50 	struct spdk_nvme_ns		*ns;
51 	struct nvme_ctrlr		*ctrlr;
52 	struct nvme_bdev		*bdev;
53 	uint32_t			ana_group_id;
54 	enum spdk_nvme_ana_state	ana_state;
55 	bool				ana_state_updating;
56 	bool				ana_transition_timedout;
57 	struct spdk_poller		*anatt_timer;
58 	struct nvme_async_probe_ctx	*probe_ctx;
59 	TAILQ_ENTRY(nvme_ns)		tailq;
60 	RB_ENTRY(nvme_ns)		node;
61 
62 	/**
63 	 * record io path stat before destroyed. Allocation of stat is
64 	 * decided by option io_path_stat of RPC
65 	 * bdev_nvme_set_options
66 	 */
67 	struct spdk_bdev_io_stat	*stat;
68 };
69 
70 struct nvme_bdev_io;
71 struct nvme_bdev_ctrlr;
72 struct nvme_bdev;
73 struct nvme_io_path;
74 
75 struct nvme_path_id {
76 	struct spdk_nvme_transport_id		trid;
77 	struct spdk_nvme_host_id		hostid;
78 	TAILQ_ENTRY(nvme_path_id)		link;
79 	uint64_t				last_failed_tsc;
80 };
81 
82 typedef void (*bdev_nvme_ctrlr_op_cb)(void *cb_arg, int rc);
83 typedef void (*nvme_ctrlr_disconnected_cb)(struct nvme_ctrlr *nvme_ctrlr);
84 
85 struct nvme_ctrlr {
86 	/**
87 	 * points to pinned, physically contiguous memory region;
88 	 * contains 4KB IDENTIFY structure for controller which is
89 	 *  target for CONTROLLER IDENTIFY command during initialization
90 	 */
91 	struct spdk_nvme_ctrlr			*ctrlr;
92 	struct nvme_path_id			*active_path_id;
93 	int					ref;
94 
95 	uint32_t				resetting : 1;
96 	uint32_t				reconnect_is_delayed : 1;
97 	uint32_t				in_failover : 1;
98 	uint32_t				pending_failover : 1;
99 	uint32_t				fast_io_fail_timedout : 1;
100 	uint32_t				destruct : 1;
101 	uint32_t				ana_log_page_updating : 1;
102 	uint32_t				io_path_cache_clearing : 1;
103 	uint32_t				dont_retry : 1;
104 	uint32_t				disabled : 1;
105 
106 	struct spdk_bdev_nvme_ctrlr_opts	opts;
107 
108 	RB_HEAD(nvme_ns_tree, nvme_ns)		namespaces;
109 
110 	struct spdk_opal_dev			*opal_dev;
111 
112 	struct spdk_poller			*adminq_timer_poller;
113 	struct spdk_thread			*thread;
114 
115 	bdev_nvme_ctrlr_op_cb			ctrlr_op_cb_fn;
116 	void					*ctrlr_op_cb_arg;
117 	/* Poller used to check for reset/detach completion */
118 	struct spdk_poller			*reset_detach_poller;
119 	struct spdk_nvme_detach_ctx		*detach_ctx;
120 
121 	uint64_t				reset_start_tsc;
122 	struct spdk_poller			*reconnect_delay_timer;
123 
124 	nvme_ctrlr_disconnected_cb		disconnected_cb;
125 
126 	/** linked list pointer for device list */
127 	TAILQ_ENTRY(nvme_ctrlr)			tailq;
128 	struct nvme_bdev_ctrlr			*nbdev_ctrlr;
129 
130 	TAILQ_HEAD(nvme_paths, nvme_path_id)	trids;
131 
132 	uint32_t				max_ana_log_page_size;
133 	struct spdk_nvme_ana_page		*ana_log_page;
134 	struct spdk_nvme_ana_group_descriptor	*copied_ana_desc;
135 
136 	struct nvme_async_probe_ctx		*probe_ctx;
137 	struct spdk_key				*psk;
138 	struct spdk_key				*dhchap_key;
139 	struct spdk_key				*dhchap_ctrlr_key;
140 
141 	pthread_mutex_t				mutex;
142 };
143 
144 struct nvme_bdev_ctrlr {
145 	char				*name;
146 	TAILQ_HEAD(, nvme_ctrlr)	ctrlrs;
147 	TAILQ_HEAD(, nvme_bdev)		bdevs;
148 	TAILQ_ENTRY(nvme_bdev_ctrlr)	tailq;
149 };
150 
151 struct nvme_error_stat {
152 	uint32_t status_type[8];
153 	uint32_t status[4][256];
154 };
155 
156 struct nvme_bdev {
157 	struct spdk_bdev			disk;
158 	uint32_t				nsid;
159 	struct nvme_bdev_ctrlr			*nbdev_ctrlr;
160 	pthread_mutex_t				mutex;
161 	int					ref;
162 	enum spdk_bdev_nvme_multipath_policy	mp_policy;
163 	enum spdk_bdev_nvme_multipath_selector	mp_selector;
164 	uint32_t				rr_min_io;
165 	TAILQ_HEAD(, nvme_ns)			nvme_ns_list;
166 	bool					opal;
167 	TAILQ_ENTRY(nvme_bdev)			tailq;
168 	struct nvme_error_stat			*err_stat;
169 };
170 
171 struct nvme_qpair {
172 	struct nvme_ctrlr		*ctrlr;
173 	struct spdk_nvme_qpair		*qpair;
174 	struct nvme_poll_group		*group;
175 	struct nvme_ctrlr_channel	*ctrlr_ch;
176 
177 	/* The following is used to update io_path cache of nvme_bdev_channels. */
178 	TAILQ_HEAD(, nvme_io_path)	io_path_list;
179 
180 	TAILQ_ENTRY(nvme_qpair)		tailq;
181 };
182 
183 struct nvme_ctrlr_channel {
184 	struct nvme_qpair		*qpair;
185 	TAILQ_HEAD(, nvme_bdev_io)	pending_resets;
186 
187 	struct spdk_io_channel_iter	*reset_iter;
188 	struct spdk_poller		*connect_poller;
189 };
190 
191 struct nvme_io_path {
192 	struct nvme_ns			*nvme_ns;
193 	struct nvme_qpair		*qpair;
194 	STAILQ_ENTRY(nvme_io_path)	stailq;
195 
196 	/* The following are used to update io_path cache of the nvme_bdev_channel. */
197 	struct nvme_bdev_channel	*nbdev_ch;
198 	TAILQ_ENTRY(nvme_io_path)	tailq;
199 
200 	/* allocation of stat is decided by option io_path_stat of RPC bdev_nvme_set_options */
201 	struct spdk_bdev_io_stat	*stat;
202 };
203 
204 struct nvme_bdev_channel {
205 	struct nvme_io_path			*current_io_path;
206 	enum spdk_bdev_nvme_multipath_policy	mp_policy;
207 	enum spdk_bdev_nvme_multipath_selector	mp_selector;
208 	uint32_t				rr_min_io;
209 	uint32_t				rr_counter;
210 	STAILQ_HEAD(, nvme_io_path)		io_path_list;
211 	TAILQ_HEAD(retry_io_head, nvme_bdev_io)	retry_io_list;
212 	struct spdk_poller			*retry_io_poller;
213 };
214 
215 struct nvme_poll_group {
216 	struct spdk_nvme_poll_group		*group;
217 	struct spdk_io_channel			*accel_channel;
218 	struct spdk_poller			*poller;
219 	bool					collect_spin_stat;
220 	uint64_t				spin_ticks;
221 	uint64_t				start_ticks;
222 	uint64_t				end_ticks;
223 	TAILQ_HEAD(, nvme_qpair)		qpair_list;
224 };
225 
226 void nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path);
227 
228 struct nvme_ctrlr *nvme_ctrlr_get_by_name(const char *name);
229 
230 struct nvme_ctrlr *nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr,
231 		uint16_t cntlid);
232 
233 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr_get_by_name(const char *name);
234 
235 typedef void (*nvme_bdev_ctrlr_for_each_fn)(struct nvme_bdev_ctrlr *nbdev_ctrlr, void *ctx);
236 
237 void nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx);
238 
239 void nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid,
240 			      struct spdk_json_write_ctx *w);
241 
242 void nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr);
243 
244 struct nvme_ns *nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid);
245 struct nvme_ns *nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr);
246 struct nvme_ns *nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns);
247 
248 enum spdk_bdev_timeout_action {
249 	SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE = 0,
250 	SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET,
251 	SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT,
252 };
253 
254 struct spdk_bdev_nvme_opts {
255 	enum spdk_bdev_timeout_action action_on_timeout;
256 	uint64_t timeout_us;
257 	uint64_t timeout_admin_us;
258 	uint32_t keep_alive_timeout_ms;
259 	/* The number of attempts per I/O in the transport layer before an I/O fails. */
260 	uint32_t transport_retry_count;
261 	uint32_t arbitration_burst;
262 	uint32_t low_priority_weight;
263 	uint32_t medium_priority_weight;
264 	uint32_t high_priority_weight;
265 	uint64_t nvme_adminq_poll_period_us;
266 	uint64_t nvme_ioq_poll_period_us;
267 	uint32_t io_queue_requests;
268 	bool delay_cmd_submit;
269 	/* The number of attempts per I/O in the bdev layer before an I/O fails. */
270 	int32_t bdev_retry_count;
271 	uint8_t transport_ack_timeout;
272 	int32_t ctrlr_loss_timeout_sec;
273 	uint32_t reconnect_delay_sec;
274 	uint32_t fast_io_fail_timeout_sec;
275 	bool disable_auto_failback;
276 	bool generate_uuids;
277 	/* Type of Service - RDMA only */
278 	uint8_t transport_tos;
279 	bool nvme_error_stat;
280 	uint32_t rdma_srq_size;
281 	bool io_path_stat;
282 	bool allow_accel_sequence;
283 	uint32_t rdma_max_cq_size;
284 	uint16_t rdma_cm_event_timeout_ms;
285 	uint32_t dhchap_digests;
286 	uint32_t dhchap_dhgroups;
287 };
288 
289 struct spdk_nvme_qpair *bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch);
290 void bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts);
291 int bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts);
292 int bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx);
293 
294 int bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid, const char *base_name,
295 			      struct spdk_nvme_ctrlr_opts *drv_opts, struct spdk_bdev_nvme_ctrlr_opts *bdev_opts,
296 			      uint64_t timeout, bool from_mdns,
297 			      spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx);
298 int bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn,
299 			     void *cb_ctx);
300 void bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w);
301 
302 int bdev_nvme_start_mdns_discovery(const char *base_name,
303 				   const char *svcname,
304 				   struct spdk_nvme_ctrlr_opts *drv_opts,
305 				   struct spdk_bdev_nvme_ctrlr_opts *bdev_opts);
306 int bdev_nvme_stop_mdns_discovery(const char *name);
307 void bdev_nvme_get_mdns_discovery_info(struct spdk_jsonrpc_request *request);
308 void bdev_nvme_mdns_discovery_config_json(struct spdk_json_write_ctx *w);
309 
310 struct spdk_nvme_ctrlr *bdev_nvme_get_ctrlr(struct spdk_bdev *bdev);
311 
312 typedef void (*bdev_nvme_delete_done_fn)(void *ctx, int rc);
313 
314 /**
315  * Delete NVMe controller with all bdevs on top of it, or delete the specified path
316  * if there is any alternative path. Requires to pass name of NVMe controller.
317  *
318  * \param name NVMe controller name
319  * \param path_id The specified path to remove (optional)
320  * \param delete_done Callback function on delete complete (optional)
321  * \param delete_done_ctx Context passed to callback (optional)
322  * \return zero on success,
323  *		-EINVAL on wrong parameters or
324  *		-ENODEV if controller is not found or
325  *		-ENOMEM on no memory
326  */
327 int bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id,
328 		     bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx);
329 
330 enum nvme_ctrlr_op {
331 	NVME_CTRLR_OP_RESET = 1,
332 	NVME_CTRLR_OP_ENABLE,
333 	NVME_CTRLR_OP_DISABLE,
334 };
335 
336 /**
337  * Perform specified operation on an NVMe controller.
338  *
339  * NOTE: The callback function is always called after this function returns except for
340  * out of memory cases.
341  *
342  * \param nvme_ctrlr The specified NVMe controller to operate
343  * \param op Operation code
344  * \param cb_fn Function to be called back after operation completes
345  * \param cb_arg Argument for callback function
346  */
347 void nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
348 		       bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg);
349 
350 /**
351  * Perform specified operation on all NVMe controllers in an NVMe bdev controller.
352  *
353  * NOTE: The callback function is always called after this function returns except for
354  * out of memory cases.
355  *
356  * \param nbdev_ctrlr The specified NVMe bdev controller to operate
357  * \param op Operation code
358  * \param cb_fn Function to be called back after operation completes
359  * \param cb_arg Argument for callback function
360  */
361 void nvme_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op,
362 			    bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg);
363 
364 typedef void (*bdev_nvme_set_preferred_path_cb)(void *cb_arg, int rc);
365 
366 /**
367  * Set the preferred I/O path for an NVMe bdev in multipath mode.
368  *
369  * NOTE: This function does not support NVMe bdevs in failover mode.
370  *
371  * \param name NVMe bdev name
372  * \param cntlid NVMe-oF controller ID
373  * \param cb_fn Function to be called back after completion.
374  * \param cb_arg Argument for callback function.
375  */
376 void bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid,
377 				  bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg);
378 
379 #endif /* SPDK_BDEV_NVME_H */
380