xref: /spdk/module/bdev/raid/bdev_raid.h (revision 8b5c4be8b4096f1af6ece9df5cd8322550fc67fd)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (C) 2018 Intel Corporation.
3  *   All rights reserved.
4  */
5 
6 #ifndef SPDK_BDEV_RAID_INTERNAL_H
7 #define SPDK_BDEV_RAID_INTERNAL_H
8 
9 #include "spdk/bdev_module.h"
10 #include "spdk/uuid.h"
11 
12 #define RAID_BDEV_MIN_DATA_OFFSET_SIZE	(1024*1024) /* 1 MiB */
13 
14 enum raid_level {
15 	INVALID_RAID_LEVEL	= -1,
16 	RAID0			= 0,
17 	RAID1			= 1,
18 	RAID5F			= 95, /* 0x5f */
19 	CONCAT			= 99,
20 };
21 
22 /*
23  * Raid state describes the state of the raid. This raid bdev can be either in
24  * configured list or configuring list
25  */
26 enum raid_bdev_state {
27 	/* raid bdev is ready and is seen by upper layers */
28 	RAID_BDEV_STATE_ONLINE,
29 
30 	/*
31 	 * raid bdev is configuring, not all underlying bdevs are present.
32 	 * And can't be seen by upper layers.
33 	 */
34 	RAID_BDEV_STATE_CONFIGURING,
35 
36 	/*
37 	 * In offline state, raid bdev layer will complete all incoming commands without
38 	 * submitting to underlying base nvme bdevs
39 	 */
40 	RAID_BDEV_STATE_OFFLINE,
41 
42 	/* raid bdev state max, new states should be added before this */
43 	RAID_BDEV_STATE_MAX
44 };
45 
46 enum raid_process_type {
47 	RAID_PROCESS_NONE,
48 	RAID_PROCESS_REBUILD,
49 	RAID_PROCESS_MAX
50 };
51 
52 typedef void (*raid_base_bdev_cb)(void *ctx, int status);
53 
54 /*
55  * raid_base_bdev_info contains information for the base bdevs which are part of some
56  * raid. This structure contains the per base bdev information. Whatever is
57  * required per base device for raid bdev will be kept here
58  */
59 struct raid_base_bdev_info {
60 	/* The raid bdev that this base bdev belongs to */
61 	struct raid_bdev	*raid_bdev;
62 
63 	/* name of the bdev */
64 	char			*name;
65 
66 	/* uuid of the bdev */
67 	struct spdk_uuid	uuid;
68 
69 	/*
70 	 * Pointer to base bdev descriptor opened by raid bdev. This is NULL when the bdev for
71 	 * this slot is missing.
72 	 */
73 	struct spdk_bdev_desc	*desc;
74 
75 	/* offset in blocks from the start of the base bdev to the start of the data region */
76 	uint64_t		data_offset;
77 
78 	/* size in blocks of the base bdev's data region */
79 	uint64_t		data_size;
80 
81 	/*
82 	 * When underlying base device calls the hot plug function on drive removal,
83 	 * this flag will be set and later after doing some processing, base device
84 	 * descriptor will be closed
85 	 */
86 	bool			remove_scheduled;
87 
88 	/* callback for base bdev removal */
89 	raid_base_bdev_cb	remove_cb;
90 
91 	/* context of the callback */
92 	void			*remove_cb_ctx;
93 
94 	/* Hold the number of blocks to know how large the base bdev is resized. */
95 	uint64_t		blockcnt;
96 
97 	/* io channel for the app thread */
98 	struct spdk_io_channel	*app_thread_ch;
99 
100 	/* Set to true when base bdev has completed the configuration process */
101 	bool			is_configured;
102 
103 	/* Set to true if this base bdev is the target of a background process */
104 	bool			is_process_target;
105 
106 	/* Set to true to indicate that the base bdev is being removed because of a failure */
107 	bool			is_failed;
108 
109 	/* callback for base bdev configuration */
110 	raid_base_bdev_cb	configure_cb;
111 
112 	/* context of the callback */
113 	void			*configure_cb_ctx;
114 };
115 
116 struct raid_bdev_io;
117 typedef void (*raid_bdev_io_completion_cb)(struct raid_bdev_io *raid_io,
118 		enum spdk_bdev_io_status status);
119 
120 /*
121  * raid_bdev_io is the context part of bdev_io. It contains the information
122  * related to bdev_io for a raid bdev
123  */
124 struct raid_bdev_io {
125 	/* The raid bdev associated with this IO */
126 	struct raid_bdev *raid_bdev;
127 
128 	uint64_t offset_blocks;
129 	uint64_t num_blocks;
130 	struct iovec *iovs;
131 	int iovcnt;
132 	enum spdk_bdev_io_type type;
133 	struct spdk_memory_domain *memory_domain;
134 	void *memory_domain_ctx;
135 	void *md_buf;
136 
137 	/* WaitQ entry, used only in waitq logic */
138 	struct spdk_bdev_io_wait_entry	waitq_entry;
139 
140 	/* Context of the original channel for this IO */
141 	struct raid_bdev_io_channel	*raid_ch;
142 
143 	/* Used for tracking progress on io requests sent to member disks. */
144 	uint64_t			base_bdev_io_remaining;
145 	uint8_t				base_bdev_io_submitted;
146 	enum spdk_bdev_io_status	base_bdev_io_status;
147 	/* This will be the raid_io completion status unless any base io's status is different. */
148 	enum spdk_bdev_io_status	base_bdev_io_status_default;
149 
150 	/* Private data for the raid module */
151 	void				*module_private;
152 
153 	/* Custom completion callback. Overrides bdev_io completion if set. */
154 	raid_bdev_io_completion_cb	completion_cb;
155 
156 	struct {
157 		uint64_t		offset;
158 		struct iovec		*iov;
159 		struct iovec		iov_copy;
160 	} split;
161 };
162 
163 struct raid_bdev_process_request {
164 	struct raid_bdev_process *process;
165 	struct raid_base_bdev_info *target;
166 	struct spdk_io_channel *target_ch;
167 	uint64_t offset_blocks;
168 	uint32_t num_blocks;
169 	struct iovec iov;
170 	void *md_buf;
171 	/* bdev_io is raid_io's driver_ctx - don't reorder them!
172 	 * These are needed for re-using raid module I/O functions for process I/O. */
173 	struct spdk_bdev_io bdev_io;
174 	struct raid_bdev_io raid_io;
175 	TAILQ_ENTRY(raid_bdev_process_request) link;
176 };
177 
178 /*
179  * raid_bdev is the single entity structure which contains SPDK block device
180  * and the information related to any raid bdev either configured or
181  * in configuring list. io device is created on this.
182  */
183 struct raid_bdev {
184 	/* raid bdev device, this will get registered in bdev layer */
185 	struct spdk_bdev		bdev;
186 
187 	/* the raid bdev descriptor, opened for internal use */
188 	struct spdk_bdev_desc		*self_desc;
189 
190 	/* link of raid bdev to link it to global raid bdev list */
191 	TAILQ_ENTRY(raid_bdev)		global_link;
192 
193 	/* array of base bdev info */
194 	struct raid_base_bdev_info	*base_bdev_info;
195 
196 	/* strip size of raid bdev in blocks */
197 	uint32_t			strip_size;
198 
199 	/* strip size of raid bdev in KB */
200 	uint32_t			strip_size_kb;
201 
202 	/* strip size bit shift for optimized calculation */
203 	uint32_t			strip_size_shift;
204 
205 	/* state of raid bdev */
206 	enum raid_bdev_state		state;
207 
208 	/* number of base bdevs comprising raid bdev  */
209 	uint8_t				num_base_bdevs;
210 
211 	/* number of base bdevs discovered */
212 	uint8_t				num_base_bdevs_discovered;
213 
214 	/*
215 	 * Number of operational base bdevs, i.e. how many we know/expect to be working. This
216 	 * will be less than num_base_bdevs when starting a degraded array.
217 	 */
218 	uint8_t				num_base_bdevs_operational;
219 
220 	/* minimum number of viable base bdevs that are required by array to operate */
221 	uint8_t				min_base_bdevs_operational;
222 
223 	/* Raid Level of this raid bdev */
224 	enum raid_level			level;
225 
226 	/* Set to true if destroy of this raid bdev is started. */
227 	bool				destroy_started;
228 
229 	/* Module for RAID-level specific operations */
230 	struct raid_bdev_module		*module;
231 
232 	/* Private data for the raid module */
233 	void				*module_private;
234 
235 	/* Superblock */
236 	bool				superblock_enabled;
237 	struct raid_bdev_superblock	*sb;
238 
239 	/* Superblock buffer used for I/O */
240 	void				*sb_io_buf;
241 	uint32_t			sb_io_buf_size;
242 
243 	/* Raid bdev background process, e.g. rebuild */
244 	struct raid_bdev_process	*process;
245 };
246 
247 #define RAID_FOR_EACH_BASE_BDEV(r, i) \
248 	for (i = r->base_bdev_info; i < r->base_bdev_info + r->num_base_bdevs; i++)
249 
250 struct raid_bdev_io_channel;
251 
252 /* TAIL head for raid bdev list */
253 TAILQ_HEAD(raid_all_tailq, raid_bdev);
254 
255 extern struct raid_all_tailq		g_raid_bdev_list;
256 
257 typedef void (*raid_bdev_destruct_cb)(void *cb_ctx, int rc);
258 
259 int raid_bdev_create(const char *name, uint32_t strip_size, uint8_t num_base_bdevs,
260 		     enum raid_level level, bool superblock, const struct spdk_uuid *uuid,
261 		     struct raid_bdev **raid_bdev_out);
262 void raid_bdev_delete(struct raid_bdev *raid_bdev, raid_bdev_destruct_cb cb_fn, void *cb_ctx);
263 int raid_bdev_add_base_bdev(struct raid_bdev *raid_bdev, const char *name,
264 			    raid_base_bdev_cb cb_fn, void *cb_ctx);
265 struct raid_bdev *raid_bdev_find_by_name(const char *name);
266 enum raid_level raid_bdev_str_to_level(const char *str);
267 const char *raid_bdev_level_to_str(enum raid_level level);
268 enum raid_bdev_state raid_bdev_str_to_state(const char *str);
269 const char *raid_bdev_state_to_str(enum raid_bdev_state state);
270 const char *raid_bdev_process_to_str(enum raid_process_type value);
271 void raid_bdev_write_info_json(struct raid_bdev *raid_bdev, struct spdk_json_write_ctx *w);
272 int raid_bdev_remove_base_bdev(struct spdk_bdev *base_bdev, raid_base_bdev_cb cb_fn, void *cb_ctx);
273 
274 /*
275  * RAID module descriptor
276  */
277 struct raid_bdev_module {
278 	/* RAID level implemented by this module */
279 	enum raid_level level;
280 
281 	/* Minimum required number of base bdevs. Must be > 0. */
282 	uint8_t base_bdevs_min;
283 
284 	/*
285 	 * RAID constraint. Determines number of base bdevs that can be removed
286 	 * without failing the array.
287 	 */
288 	struct {
289 		enum {
290 			CONSTRAINT_UNSET = 0,
291 			CONSTRAINT_MAX_BASE_BDEVS_REMOVED,
292 			CONSTRAINT_MIN_BASE_BDEVS_OPERATIONAL,
293 		} type;
294 		uint8_t value;
295 	} base_bdevs_constraint;
296 
297 	/* Set to true if this module supports memory domains. */
298 	bool memory_domains_supported;
299 
300 	/* Set to true if this module supports DIF/DIX */
301 	bool dif_supported;
302 
303 	/*
304 	 * Called when the raid is starting, right before changing the state to
305 	 * online and registering the bdev. Parameters of the bdev like blockcnt
306 	 * should be set here.
307 	 *
308 	 * Non-zero return value will abort the startup process.
309 	 */
310 	int (*start)(struct raid_bdev *raid_bdev);
311 
312 	/*
313 	 * Called when the raid is stopping, right before changing the state to
314 	 * offline and unregistering the bdev. Optional.
315 	 *
316 	 * The function should return false if it is asynchronous. Then, after
317 	 * the async operation has completed and the module is fully stopped
318 	 * raid_bdev_module_stop_done() must be called.
319 	 */
320 	bool (*stop)(struct raid_bdev *raid_bdev);
321 
322 	/* Handler for R/W requests */
323 	void (*submit_rw_request)(struct raid_bdev_io *raid_io);
324 
325 	/* Handler for requests without payload (flush, unmap). Optional. */
326 	void (*submit_null_payload_request)(struct raid_bdev_io *raid_io);
327 
328 	/*
329 	 * Called when the bdev's IO channel is created to get the module's private IO channel.
330 	 * Optional.
331 	 */
332 	struct spdk_io_channel *(*get_io_channel)(struct raid_bdev *raid_bdev);
333 
334 	/*
335 	 * Called when a base_bdev is resized to resize the raid if the condition
336 	 * is satisfied. Optional.
337 	 *
338 	 * Returns true if the resize was performed.
339 	 */
340 	bool (*resize)(struct raid_bdev *raid_bdev);
341 
342 	/* Handler for raid process requests. Required for raid modules with redundancy. */
343 	int (*submit_process_request)(struct raid_bdev_process_request *process_req,
344 				      struct raid_bdev_io_channel *raid_ch);
345 
346 	TAILQ_ENTRY(raid_bdev_module) link;
347 };
348 
349 void raid_bdev_module_list_add(struct raid_bdev_module *raid_module);
350 
351 #define __RAID_MODULE_REGISTER(line) __RAID_MODULE_REGISTER_(line)
352 #define __RAID_MODULE_REGISTER_(line) raid_module_register_##line
353 
354 #define RAID_MODULE_REGISTER(_module)					\
355 __attribute__((constructor)) static void				\
356 __RAID_MODULE_REGISTER(__LINE__)(void)					\
357 {									\
358     raid_bdev_module_list_add(_module);					\
359 }
360 
361 bool raid_bdev_io_complete_part(struct raid_bdev_io *raid_io, uint64_t completed,
362 				enum spdk_bdev_io_status status);
363 void raid_bdev_queue_io_wait(struct raid_bdev_io *raid_io, struct spdk_bdev *bdev,
364 			     struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn);
365 void raid_bdev_io_complete(struct raid_bdev_io *raid_io, enum spdk_bdev_io_status status);
366 void raid_bdev_module_stop_done(struct raid_bdev *raid_bdev);
367 struct spdk_io_channel *raid_bdev_channel_get_base_channel(struct raid_bdev_io_channel *raid_ch,
368 		uint8_t idx);
369 void *raid_bdev_channel_get_module_ctx(struct raid_bdev_io_channel *raid_ch);
370 struct raid_base_bdev_info *raid_bdev_channel_get_base_info(struct raid_bdev_io_channel *raid_ch,
371 		struct spdk_bdev *base_bdev);
372 void raid_bdev_process_request_complete(struct raid_bdev_process_request *process_req, int status);
373 void raid_bdev_io_init(struct raid_bdev_io *raid_io, struct raid_bdev_io_channel *raid_ch,
374 		       enum spdk_bdev_io_type type, uint64_t offset_blocks,
375 		       uint64_t num_blocks, struct iovec *iovs, int iovcnt, void *md_buf,
376 		       struct spdk_memory_domain *memory_domain, void *memory_domain_ctx);
377 void raid_bdev_fail_base_bdev(struct raid_base_bdev_info *base_info);
378 
379 static inline uint8_t
380 raid_bdev_base_bdev_slot(struct raid_base_bdev_info *base_info)
381 {
382 	return base_info - base_info->raid_bdev->base_bdev_info;
383 }
384 
385 static inline void
386 raid_bdev_io_set_default_status(struct raid_bdev_io *raid_io, enum spdk_bdev_io_status status)
387 {
388 	assert(raid_io->base_bdev_io_submitted == 0);
389 	raid_io->base_bdev_io_status = status;
390 	raid_io->base_bdev_io_status_default = status;
391 }
392 
393 int raid_bdev_remap_dix_reftag(void *md_buf, uint64_t num_blocks,
394 			       struct spdk_bdev *bdev, uint32_t remapped_offset);
395 int raid_bdev_verify_dix_reftag(struct iovec *iovs, int iovcnt, void *md_buf,
396 				uint64_t num_blocks, struct spdk_bdev *bdev, uint32_t offset_blocks);
397 
398 /**
399  * Raid bdev I/O read/write wrapper for spdk_bdev_readv_blocks_ext function.
400  */
401 static inline int
402 raid_bdev_readv_blocks_ext(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
403 			   struct iovec *iov, int iovcnt, uint64_t offset_blocks,
404 			   uint64_t num_blocks, spdk_bdev_io_completion_cb cb, void *cb_arg,
405 			   struct spdk_bdev_ext_io_opts *opts)
406 {
407 	return spdk_bdev_readv_blocks_ext(base_info->desc, ch, iov, iovcnt,
408 					  base_info->data_offset + offset_blocks, num_blocks, cb, cb_arg, opts);
409 }
410 
411 /**
412  * Raid bdev I/O read/write wrapper for spdk_bdev_writev_blocks_ext function.
413  */
414 static inline int
415 raid_bdev_writev_blocks_ext(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
416 			    struct iovec *iov, int iovcnt, uint64_t offset_blocks,
417 			    uint64_t num_blocks, spdk_bdev_io_completion_cb cb, void *cb_arg,
418 			    struct spdk_bdev_ext_io_opts *opts)
419 {
420 	int rc;
421 	uint64_t remapped_offset_blocks = base_info->data_offset + offset_blocks;
422 
423 	if (spdk_unlikely(spdk_bdev_get_dif_type(&base_info->raid_bdev->bdev) != SPDK_DIF_DISABLE &&
424 			  (base_info->raid_bdev->bdev.dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK))) {
425 		rc = raid_bdev_remap_dix_reftag(opts->metadata, num_blocks, &base_info->raid_bdev->bdev,
426 						remapped_offset_blocks);
427 		if (rc != 0) {
428 			return rc;
429 		}
430 	}
431 
432 	return spdk_bdev_writev_blocks_ext(base_info->desc, ch, iov, iovcnt,
433 					   remapped_offset_blocks, num_blocks, cb, cb_arg, opts);
434 }
435 
436 /**
437  * Raid bdev I/O read/write wrapper for spdk_bdev_unmap_blocks function.
438  */
439 static inline int
440 raid_bdev_unmap_blocks(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
441 		       uint64_t offset_blocks, uint64_t num_blocks,
442 		       spdk_bdev_io_completion_cb cb, void *cb_arg)
443 {
444 	return spdk_bdev_unmap_blocks(base_info->desc, ch, base_info->data_offset + offset_blocks,
445 				      num_blocks, cb, cb_arg);
446 }
447 
448 /**
449  * Raid bdev I/O read/write wrapper for spdk_bdev_flush_blocks function.
450  */
451 static inline int
452 raid_bdev_flush_blocks(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
453 		       uint64_t offset_blocks, uint64_t num_blocks,
454 		       spdk_bdev_io_completion_cb cb, void *cb_arg)
455 {
456 	return spdk_bdev_flush_blocks(base_info->desc, ch, base_info->data_offset + offset_blocks,
457 				      num_blocks, cb, cb_arg);
458 }
459 
460 /*
461  * Definitions related to raid bdev superblock
462  */
463 
464 #define RAID_BDEV_SB_VERSION_MAJOR	1
465 #define RAID_BDEV_SB_VERSION_MINOR	0
466 
467 #define RAID_BDEV_SB_NAME_SIZE		64
468 
469 enum raid_bdev_sb_base_bdev_state {
470 	RAID_SB_BASE_BDEV_MISSING	= 0,
471 	RAID_SB_BASE_BDEV_CONFIGURED	= 1,
472 	RAID_SB_BASE_BDEV_FAILED	= 2,
473 	RAID_SB_BASE_BDEV_SPARE		= 3,
474 };
475 
476 struct raid_bdev_sb_base_bdev {
477 	/* uuid of the base bdev */
478 	struct spdk_uuid	uuid;
479 	/* offset in blocks from base device start to the start of raid data area */
480 	uint64_t		data_offset;
481 	/* size in blocks of the base device raid data area */
482 	uint64_t		data_size;
483 	/* state of the base bdev */
484 	uint32_t		state;
485 	/* feature/status flags */
486 	uint32_t		flags;
487 	/* slot number of this base bdev in the raid */
488 	uint8_t			slot;
489 
490 	uint8_t			reserved[23];
491 };
492 SPDK_STATIC_ASSERT(sizeof(struct raid_bdev_sb_base_bdev) == 64, "incorrect size");
493 
494 struct raid_bdev_superblock {
495 #define RAID_BDEV_SB_SIG "SPDKRAID"
496 	uint8_t			signature[8];
497 	struct {
498 		/* incremented when a breaking change in the superblock structure is made */
499 		uint16_t	major;
500 		/* incremented for changes in the superblock that are backward compatible */
501 		uint16_t	minor;
502 	} version;
503 	/* length in bytes of the entire superblock */
504 	uint32_t		length;
505 	/* crc32c checksum of the entire superblock */
506 	uint32_t		crc;
507 	/* feature/status flags */
508 	uint32_t		flags;
509 	/* unique id of the raid bdev */
510 	struct spdk_uuid	uuid;
511 	/* name of the raid bdev */
512 	uint8_t			name[RAID_BDEV_SB_NAME_SIZE];
513 	/* size of the raid bdev in blocks */
514 	uint64_t		raid_size;
515 	/* the raid bdev block size - must be the same for all base bdevs */
516 	uint32_t		block_size;
517 	/* the raid level */
518 	uint32_t		level;
519 	/* strip (chunk) size in blocks */
520 	uint32_t		strip_size;
521 	/* state of the raid */
522 	uint32_t		state;
523 	/* sequence number, incremented on every superblock update */
524 	uint64_t		seq_number;
525 	/* number of raid base devices */
526 	uint8_t			num_base_bdevs;
527 
528 	uint8_t			reserved[118];
529 
530 	/* size of the base bdevs array */
531 	uint8_t			base_bdevs_size;
532 	/* array of base bdev descriptors */
533 	struct raid_bdev_sb_base_bdev base_bdevs[];
534 };
535 SPDK_STATIC_ASSERT(sizeof(struct raid_bdev_superblock) == 256, "incorrect size");
536 
537 #define RAID_BDEV_SB_MAX_LENGTH (sizeof(struct raid_bdev_superblock) + UINT8_MAX * sizeof(struct raid_bdev_sb_base_bdev))
538 
539 SPDK_STATIC_ASSERT(RAID_BDEV_SB_MAX_LENGTH < RAID_BDEV_MIN_DATA_OFFSET_SIZE,
540 		   "Incorrect min data offset");
541 
542 typedef void (*raid_bdev_write_sb_cb)(int status, struct raid_bdev *raid_bdev, void *ctx);
543 typedef void (*raid_bdev_load_sb_cb)(const struct raid_bdev_superblock *sb, int status, void *ctx);
544 
545 int raid_bdev_alloc_superblock(struct raid_bdev *raid_bdev, uint32_t block_size);
546 void raid_bdev_free_superblock(struct raid_bdev *raid_bdev);
547 void raid_bdev_init_superblock(struct raid_bdev *raid_bdev);
548 void raid_bdev_write_superblock(struct raid_bdev *raid_bdev, raid_bdev_write_sb_cb cb,
549 				void *cb_ctx);
550 int raid_bdev_load_base_bdev_superblock(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
551 					raid_bdev_load_sb_cb cb, void *cb_ctx);
552 
553 struct spdk_raid_bdev_opts {
554 	/* Size of the background process window in KiB */
555 	uint32_t process_window_size_kb;
556 	/* Maximum bandwidth in MiB to process per second */
557 	uint32_t process_max_bandwidth_mb_sec;
558 };
559 
560 void raid_bdev_get_opts(struct spdk_raid_bdev_opts *opts);
561 int raid_bdev_set_opts(const struct spdk_raid_bdev_opts *opts);
562 
563 #endif /* SPDK_BDEV_RAID_INTERNAL_H */
564