xref: /spdk/module/bdev/raid/bdev_raid.h (revision 45a053c5777494f4e8ce4bc1191c9de3920377f7)
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 	/* callback for base bdev configuration */
104 	raid_base_bdev_cb	configure_cb;
105 
106 	/* context of the callback */
107 	void			*configure_cb_ctx;
108 };
109 
110 struct raid_bdev_io;
111 typedef void (*raid_bdev_io_completion_cb)(struct raid_bdev_io *raid_io,
112 		enum spdk_bdev_io_status status);
113 
114 /*
115  * raid_bdev_io is the context part of bdev_io. It contains the information
116  * related to bdev_io for a raid bdev
117  */
118 struct raid_bdev_io {
119 	/* The raid bdev associated with this IO */
120 	struct raid_bdev *raid_bdev;
121 
122 	uint64_t offset_blocks;
123 	uint64_t num_blocks;
124 	struct iovec *iovs;
125 	int iovcnt;
126 	enum spdk_bdev_io_type type;
127 	struct spdk_memory_domain *memory_domain;
128 	void *memory_domain_ctx;
129 	void *md_buf;
130 
131 	/* WaitQ entry, used only in waitq logic */
132 	struct spdk_bdev_io_wait_entry	waitq_entry;
133 
134 	/* Context of the original channel for this IO */
135 	struct raid_bdev_io_channel	*raid_ch;
136 
137 	/* Used for tracking progress on io requests sent to member disks. */
138 	uint64_t			base_bdev_io_remaining;
139 	uint8_t				base_bdev_io_submitted;
140 	enum spdk_bdev_io_status	base_bdev_io_status;
141 
142 	/* Private data for the raid module */
143 	void				*module_private;
144 
145 	/* Custom completion callback. Overrides bdev_io completion if set. */
146 	raid_bdev_io_completion_cb	completion_cb;
147 
148 	struct {
149 		uint64_t		offset;
150 		struct iovec		*iov;
151 		struct iovec		iov_copy;
152 	} split;
153 };
154 
155 struct raid_bdev_process_request {
156 	struct raid_bdev_process *process;
157 	struct raid_base_bdev_info *target;
158 	struct spdk_io_channel *target_ch;
159 	uint64_t offset_blocks;
160 	uint32_t num_blocks;
161 	struct iovec iov;
162 	void *md_buf;
163 	/* bdev_io is raid_io's driver_ctx - don't reorder them!
164 	 * These are needed for re-using raid module I/O functions for process I/O. */
165 	struct spdk_bdev_io bdev_io;
166 	struct raid_bdev_io raid_io;
167 	TAILQ_ENTRY(raid_bdev_process_request) link;
168 };
169 
170 /*
171  * raid_bdev is the single entity structure which contains SPDK block device
172  * and the information related to any raid bdev either configured or
173  * in configuring list. io device is created on this.
174  */
175 struct raid_bdev {
176 	/* raid bdev device, this will get registered in bdev layer */
177 	struct spdk_bdev		bdev;
178 
179 	/* link of raid bdev to link it to global raid bdev list */
180 	TAILQ_ENTRY(raid_bdev)		global_link;
181 
182 	/* array of base bdev info */
183 	struct raid_base_bdev_info	*base_bdev_info;
184 
185 	/* lock to protect the base bdev array */
186 	struct spdk_spinlock		base_bdev_lock;
187 
188 	/* strip size of raid bdev in blocks */
189 	uint32_t			strip_size;
190 
191 	/* strip size of raid bdev in KB */
192 	uint32_t			strip_size_kb;
193 
194 	/* strip size bit shift for optimized calculation */
195 	uint32_t			strip_size_shift;
196 
197 	/* block length bit shift for optimized calculation */
198 	uint32_t			blocklen_shift;
199 
200 	/* state of raid bdev */
201 	enum raid_bdev_state		state;
202 
203 	/* number of base bdevs comprising raid bdev  */
204 	uint8_t				num_base_bdevs;
205 
206 	/* number of base bdevs discovered */
207 	uint8_t				num_base_bdevs_discovered;
208 
209 	/*
210 	 * Number of operational base bdevs, i.e. how many we know/expect to be working. This
211 	 * will be less than num_base_bdevs when starting a degraded array.
212 	 */
213 	uint8_t				num_base_bdevs_operational;
214 
215 	/* minimum number of viable base bdevs that are required by array to operate */
216 	uint8_t				min_base_bdevs_operational;
217 
218 	/* Raid Level of this raid bdev */
219 	enum raid_level			level;
220 
221 	/* Set to true if destroy of this raid bdev is started. */
222 	bool				destroy_started;
223 
224 	/* Module for RAID-level specific operations */
225 	struct raid_bdev_module		*module;
226 
227 	/* Private data for the raid module */
228 	void				*module_private;
229 
230 	/* Superblock */
231 	struct raid_bdev_superblock	*sb;
232 
233 	/* Raid bdev background process, e.g. rebuild */
234 	struct raid_bdev_process	*process;
235 };
236 
237 #define RAID_FOR_EACH_BASE_BDEV(r, i) \
238 	for (i = r->base_bdev_info; i < r->base_bdev_info + r->num_base_bdevs; i++)
239 
240 struct raid_bdev_io_channel;
241 
242 /* TAIL head for raid bdev list */
243 TAILQ_HEAD(raid_all_tailq, raid_bdev);
244 
245 extern struct raid_all_tailq		g_raid_bdev_list;
246 
247 typedef void (*raid_bdev_destruct_cb)(void *cb_ctx, int rc);
248 
249 int raid_bdev_create(const char *name, uint32_t strip_size, uint8_t num_base_bdevs,
250 		     enum raid_level level, bool superblock, const struct spdk_uuid *uuid,
251 		     struct raid_bdev **raid_bdev_out);
252 void raid_bdev_delete(struct raid_bdev *raid_bdev, raid_bdev_destruct_cb cb_fn, void *cb_ctx);
253 int raid_bdev_add_base_device(struct raid_bdev *raid_bdev, const char *name, uint8_t slot);
254 struct raid_bdev *raid_bdev_find_by_name(const char *name);
255 enum raid_level raid_bdev_str_to_level(const char *str);
256 const char *raid_bdev_level_to_str(enum raid_level level);
257 enum raid_bdev_state raid_bdev_str_to_state(const char *str);
258 const char *raid_bdev_state_to_str(enum raid_bdev_state state);
259 const char *raid_bdev_process_to_str(enum raid_process_type value);
260 void raid_bdev_write_info_json(struct raid_bdev *raid_bdev, struct spdk_json_write_ctx *w);
261 int raid_bdev_remove_base_bdev(struct spdk_bdev *base_bdev, raid_base_bdev_cb cb_fn, void *cb_ctx);
262 int raid_bdev_attach_base_bdev(struct raid_bdev *raid_bdev, struct spdk_bdev *base_bdev,
263 			       raid_base_bdev_cb cb_fn, void *cb_ctx);
264 
265 /*
266  * RAID module descriptor
267  */
268 struct raid_bdev_module {
269 	/* RAID level implemented by this module */
270 	enum raid_level level;
271 
272 	/* Minimum required number of base bdevs. Must be > 0. */
273 	uint8_t base_bdevs_min;
274 
275 	/*
276 	 * RAID constraint. Determines number of base bdevs that can be removed
277 	 * without failing the array.
278 	 */
279 	struct {
280 		enum {
281 			CONSTRAINT_UNSET = 0,
282 			CONSTRAINT_MAX_BASE_BDEVS_REMOVED,
283 			CONSTRAINT_MIN_BASE_BDEVS_OPERATIONAL,
284 		} type;
285 		uint8_t value;
286 	} base_bdevs_constraint;
287 
288 	/* Set to true if this module supports memory domains. */
289 	bool memory_domains_supported;
290 
291 	/*
292 	 * Called when the raid is starting, right before changing the state to
293 	 * online and registering the bdev. Parameters of the bdev like blockcnt
294 	 * should be set here.
295 	 *
296 	 * Non-zero return value will abort the startup process.
297 	 */
298 	int (*start)(struct raid_bdev *raid_bdev);
299 
300 	/*
301 	 * Called when the raid is stopping, right before changing the state to
302 	 * offline and unregistering the bdev. Optional.
303 	 *
304 	 * The function should return false if it is asynchronous. Then, after
305 	 * the async operation has completed and the module is fully stopped
306 	 * raid_bdev_module_stop_done() must be called.
307 	 */
308 	bool (*stop)(struct raid_bdev *raid_bdev);
309 
310 	/* Handler for R/W requests */
311 	void (*submit_rw_request)(struct raid_bdev_io *raid_io);
312 
313 	/* Handler for requests without payload (flush, unmap). Optional. */
314 	void (*submit_null_payload_request)(struct raid_bdev_io *raid_io);
315 
316 	/*
317 	 * Called when the bdev's IO channel is created to get the module's private IO channel.
318 	 * Optional.
319 	 */
320 	struct spdk_io_channel *(*get_io_channel)(struct raid_bdev *raid_bdev);
321 
322 	/*
323 	 * Called when a base_bdev is resized to resize the raid if the condition
324 	 * is satisfied.
325 	 */
326 	void (*resize)(struct raid_bdev *raid_bdev);
327 
328 	/* Handler for raid process requests. Required for raid modules with redundancy. */
329 	int (*submit_process_request)(struct raid_bdev_process_request *process_req,
330 				      struct raid_bdev_io_channel *raid_ch);
331 
332 	TAILQ_ENTRY(raid_bdev_module) link;
333 };
334 
335 void raid_bdev_module_list_add(struct raid_bdev_module *raid_module);
336 
337 #define __RAID_MODULE_REGISTER(line) __RAID_MODULE_REGISTER_(line)
338 #define __RAID_MODULE_REGISTER_(line) raid_module_register_##line
339 
340 #define RAID_MODULE_REGISTER(_module)					\
341 __attribute__((constructor)) static void				\
342 __RAID_MODULE_REGISTER(__LINE__)(void)					\
343 {									\
344     raid_bdev_module_list_add(_module);					\
345 }
346 
347 bool raid_bdev_io_complete_part(struct raid_bdev_io *raid_io, uint64_t completed,
348 				enum spdk_bdev_io_status status);
349 void raid_bdev_queue_io_wait(struct raid_bdev_io *raid_io, struct spdk_bdev *bdev,
350 			     struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn);
351 void raid_bdev_io_complete(struct raid_bdev_io *raid_io, enum spdk_bdev_io_status status);
352 void raid_bdev_module_stop_done(struct raid_bdev *raid_bdev);
353 struct spdk_io_channel *raid_bdev_channel_get_base_channel(struct raid_bdev_io_channel *raid_ch,
354 		uint8_t idx);
355 void *raid_bdev_channel_get_module_ctx(struct raid_bdev_io_channel *raid_ch);
356 void raid_bdev_process_request_complete(struct raid_bdev_process_request *process_req, int status);
357 void raid_bdev_io_init(struct raid_bdev_io *raid_io, struct raid_bdev_io_channel *raid_ch,
358 		       enum spdk_bdev_io_type type, uint64_t offset_blocks,
359 		       uint64_t num_blocks, struct iovec *iovs, int iovcnt, void *md_buf,
360 		       struct spdk_memory_domain *memory_domain, void *memory_domain_ctx);
361 
362 static inline uint8_t
363 raid_bdev_base_bdev_slot(struct raid_base_bdev_info *base_info)
364 {
365 	return base_info - base_info->raid_bdev->base_bdev_info;
366 }
367 
368 /**
369  * Raid bdev I/O read/write wrapper for spdk_bdev_readv_blocks_ext function.
370  */
371 static inline int
372 raid_bdev_readv_blocks_ext(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
373 			   struct iovec *iov, int iovcnt, uint64_t offset_blocks,
374 			   uint64_t num_blocks, spdk_bdev_io_completion_cb cb, void *cb_arg,
375 			   struct spdk_bdev_ext_io_opts *opts)
376 {
377 	return spdk_bdev_readv_blocks_ext(base_info->desc, ch, iov, iovcnt,
378 					  base_info->data_offset + offset_blocks, num_blocks, cb, cb_arg, opts);
379 }
380 
381 /**
382  * Raid bdev I/O read/write wrapper for spdk_bdev_writev_blocks_ext function.
383  */
384 static inline int
385 raid_bdev_writev_blocks_ext(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
386 			    struct iovec *iov, int iovcnt, uint64_t offset_blocks,
387 			    uint64_t num_blocks, spdk_bdev_io_completion_cb cb, void *cb_arg,
388 			    struct spdk_bdev_ext_io_opts *opts)
389 {
390 	return spdk_bdev_writev_blocks_ext(base_info->desc, ch, iov, iovcnt,
391 					   base_info->data_offset + offset_blocks, num_blocks, cb, cb_arg, opts);
392 }
393 
394 /**
395  * Raid bdev I/O read/write wrapper for spdk_bdev_unmap_blocks function.
396  */
397 static inline int
398 raid_bdev_unmap_blocks(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
399 		       uint64_t offset_blocks, uint64_t num_blocks,
400 		       spdk_bdev_io_completion_cb cb, void *cb_arg)
401 {
402 	return spdk_bdev_unmap_blocks(base_info->desc, ch, base_info->data_offset + offset_blocks,
403 				      num_blocks, cb, cb_arg);
404 }
405 
406 /**
407  * Raid bdev I/O read/write wrapper for spdk_bdev_flush_blocks function.
408  */
409 static inline int
410 raid_bdev_flush_blocks(struct raid_base_bdev_info *base_info, struct spdk_io_channel *ch,
411 		       uint64_t offset_blocks, uint64_t num_blocks,
412 		       spdk_bdev_io_completion_cb cb, void *cb_arg)
413 {
414 	return spdk_bdev_flush_blocks(base_info->desc, ch, base_info->data_offset + offset_blocks,
415 				      num_blocks, cb, cb_arg);
416 }
417 
418 /*
419  * Definitions related to raid bdev superblock
420  */
421 
422 #define RAID_BDEV_SB_VERSION_MAJOR	1
423 #define RAID_BDEV_SB_VERSION_MINOR	0
424 
425 #define RAID_BDEV_SB_NAME_SIZE		64
426 
427 enum raid_bdev_sb_base_bdev_state {
428 	RAID_SB_BASE_BDEV_MISSING	= 0,
429 	RAID_SB_BASE_BDEV_CONFIGURED	= 1,
430 	RAID_SB_BASE_BDEV_FAILED	= 2,
431 	RAID_SB_BASE_BDEV_SPARE		= 3,
432 };
433 
434 struct raid_bdev_sb_base_bdev {
435 	/* uuid of the base bdev */
436 	struct spdk_uuid	uuid;
437 	/* offset in blocks from base device start to the start of raid data area */
438 	uint64_t		data_offset;
439 	/* size in blocks of the base device raid data area */
440 	uint64_t		data_size;
441 	/* state of the base bdev */
442 	uint32_t		state;
443 	/* feature/status flags */
444 	uint32_t		flags;
445 	/* slot number of this base bdev in the raid */
446 	uint8_t			slot;
447 
448 	uint8_t			reserved[23];
449 };
450 SPDK_STATIC_ASSERT(sizeof(struct raid_bdev_sb_base_bdev) == 64, "incorrect size");
451 
452 struct raid_bdev_superblock {
453 #define RAID_BDEV_SB_SIG "SPDKRAID"
454 	uint8_t			signature[8];
455 	struct {
456 		/* incremented when a breaking change in the superblock structure is made */
457 		uint16_t	major;
458 		/* incremented for changes in the superblock that are backward compatible */
459 		uint16_t	minor;
460 	} version;
461 	/* length in bytes of the entire superblock */
462 	uint32_t		length;
463 	/* crc32c checksum of the entire superblock */
464 	uint32_t		crc;
465 	/* feature/status flags */
466 	uint32_t		flags;
467 	/* unique id of the raid bdev */
468 	struct spdk_uuid	uuid;
469 	/* name of the raid bdev */
470 	uint8_t			name[RAID_BDEV_SB_NAME_SIZE];
471 	/* size of the raid bdev in blocks */
472 	uint64_t		raid_size;
473 	/* the raid bdev block size - must be the same for all base bdevs */
474 	uint32_t		block_size;
475 	/* the raid level */
476 	uint32_t		level;
477 	/* strip (chunk) size in blocks */
478 	uint32_t		strip_size;
479 	/* state of the raid */
480 	uint32_t		state;
481 	/* sequence number, incremented on every superblock update */
482 	uint64_t		seq_number;
483 	/* number of raid base devices */
484 	uint8_t			num_base_bdevs;
485 
486 	uint8_t			reserved[118];
487 
488 	/* size of the base bdevs array */
489 	uint8_t			base_bdevs_size;
490 	/* array of base bdev descriptors */
491 	struct raid_bdev_sb_base_bdev base_bdevs[];
492 };
493 SPDK_STATIC_ASSERT(sizeof(struct raid_bdev_superblock) == 256, "incorrect size");
494 
495 #define RAID_BDEV_SB_MAX_LENGTH \
496 	SPDK_ALIGN_CEIL((sizeof(struct raid_bdev_superblock) + UINT8_MAX * sizeof(struct raid_bdev_sb_base_bdev)), 0x1000)
497 
498 SPDK_STATIC_ASSERT(RAID_BDEV_SB_MAX_LENGTH < RAID_BDEV_MIN_DATA_OFFSET_SIZE,
499 		   "Incorrect min data offset");
500 
501 typedef void (*raid_bdev_write_sb_cb)(int status, struct raid_bdev *raid_bdev, void *ctx);
502 typedef void (*raid_bdev_load_sb_cb)(const struct raid_bdev_superblock *sb, int status, void *ctx);
503 
504 void raid_bdev_init_superblock(struct raid_bdev *raid_bdev);
505 void raid_bdev_write_superblock(struct raid_bdev *raid_bdev, raid_bdev_write_sb_cb cb,
506 				void *cb_ctx);
507 int raid_bdev_load_base_bdev_superblock(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
508 					raid_bdev_load_sb_cb cb, void *cb_ctx);
509 
510 #endif /* SPDK_BDEV_RAID_INTERNAL_H */
511