xref: /linux/drivers/md/dm-vdo/vio.c (revision 1553a1c48281243359a9529a10ddb551f3b967ab)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2023 Red Hat
4  */
5 
6 #include "vio.h"
7 
8 #include <linux/bio.h>
9 #include <linux/blkdev.h>
10 #include <linux/kernel.h>
11 #include <linux/ratelimit.h>
12 
13 #include "logger.h"
14 #include "memory-alloc.h"
15 #include "permassert.h"
16 
17 #include "constants.h"
18 #include "io-submitter.h"
19 #include "vdo.h"
20 
21 /* A vio_pool is a collection of preallocated vios. */
22 struct vio_pool {
23 	/* The number of objects managed by the pool */
24 	size_t size;
25 	/* The list of objects which are available */
26 	struct list_head available;
27 	/* The queue of requestors waiting for objects from the pool */
28 	struct vdo_wait_queue waiting;
29 	/* The number of objects currently in use */
30 	size_t busy_count;
31 	/* The list of objects which are in use */
32 	struct list_head busy;
33 	/* The ID of the thread on which this pool may be used */
34 	thread_id_t thread_id;
35 	/* The buffer backing the pool's vios */
36 	char *buffer;
37 	/* The pool entries */
38 	struct pooled_vio vios[];
39 };
40 
41 physical_block_number_t pbn_from_vio_bio(struct bio *bio)
42 {
43 	struct vio *vio = bio->bi_private;
44 	struct vdo *vdo = vio->completion.vdo;
45 	physical_block_number_t pbn = bio->bi_iter.bi_sector / VDO_SECTORS_PER_BLOCK;
46 
47 	return ((pbn == VDO_GEOMETRY_BLOCK_LOCATION) ? pbn : pbn + vdo->geometry.bio_offset);
48 }
49 
50 static int create_multi_block_bio(block_count_t size, struct bio **bio_ptr)
51 {
52 	struct bio *bio = NULL;
53 	int result;
54 
55 	result = vdo_allocate_extended(struct bio, size + 1, struct bio_vec,
56 				       "bio", &bio);
57 	if (result != VDO_SUCCESS)
58 		return result;
59 
60 	*bio_ptr = bio;
61 	return VDO_SUCCESS;
62 }
63 
64 int vdo_create_bio(struct bio **bio_ptr)
65 {
66 	return create_multi_block_bio(1, bio_ptr);
67 }
68 
69 void vdo_free_bio(struct bio *bio)
70 {
71 	if (bio == NULL)
72 		return;
73 
74 	bio_uninit(bio);
75 	vdo_free(vdo_forget(bio));
76 }
77 
78 int allocate_vio_components(struct vdo *vdo, enum vio_type vio_type,
79 			    enum vio_priority priority, void *parent,
80 			    unsigned int block_count, char *data, struct vio *vio)
81 {
82 	struct bio *bio;
83 	int result;
84 
85 	result = VDO_ASSERT(block_count <= MAX_BLOCKS_PER_VIO,
86 			    "block count %u does not exceed maximum %u", block_count,
87 			    MAX_BLOCKS_PER_VIO);
88 	if (result != VDO_SUCCESS)
89 		return result;
90 
91 	result = VDO_ASSERT(((vio_type != VIO_TYPE_UNINITIALIZED) && (vio_type != VIO_TYPE_DATA)),
92 			    "%d is a metadata type", vio_type);
93 	if (result != VDO_SUCCESS)
94 		return result;
95 
96 	result = create_multi_block_bio(block_count, &bio);
97 	if (result != VDO_SUCCESS)
98 		return result;
99 
100 	initialize_vio(vio, bio, block_count, vio_type, priority, vdo);
101 	vio->completion.parent = parent;
102 	vio->data = data;
103 	return VDO_SUCCESS;
104 }
105 
106 /**
107  * create_multi_block_metadata_vio() - Create a vio.
108  * @vdo: The vdo on which the vio will operate.
109  * @vio_type: The type of vio to create.
110  * @priority: The relative priority to assign to the vio.
111  * @parent: The parent of the vio.
112  * @block_count: The size of the vio in blocks.
113  * @data: The buffer.
114  * @vio_ptr: A pointer to hold the new vio.
115  *
116  * Return: VDO_SUCCESS or an error.
117  */
118 int create_multi_block_metadata_vio(struct vdo *vdo, enum vio_type vio_type,
119 				    enum vio_priority priority, void *parent,
120 				    unsigned int block_count, char *data,
121 				    struct vio **vio_ptr)
122 {
123 	struct vio *vio;
124 	int result;
125 
126 	BUILD_BUG_ON(sizeof(struct vio) > 256);
127 
128 	/*
129 	 * Metadata vios should use direct allocation and not use the buffer pool, which is
130 	 * reserved for submissions from the linux block layer.
131 	 */
132 	result = vdo_allocate(1, struct vio, __func__, &vio);
133 	if (result != VDO_SUCCESS) {
134 		vdo_log_error("metadata vio allocation failure %d", result);
135 		return result;
136 	}
137 
138 	result = allocate_vio_components(vdo, vio_type, priority, parent, block_count,
139 					 data, vio);
140 	if (result != VDO_SUCCESS) {
141 		vdo_free(vio);
142 		return result;
143 	}
144 
145 	*vio_ptr  = vio;
146 	return VDO_SUCCESS;
147 }
148 
149 /**
150  * free_vio_components() - Free the components of a vio embedded in a larger structure.
151  * @vio: The vio to destroy
152  */
153 void free_vio_components(struct vio *vio)
154 {
155 	if (vio == NULL)
156 		return;
157 
158 	BUG_ON(is_data_vio(vio));
159 	vdo_free_bio(vdo_forget(vio->bio));
160 }
161 
162 /**
163  * free_vio() - Destroy a vio.
164  * @vio: The vio to destroy.
165  */
166 void free_vio(struct vio *vio)
167 {
168 	free_vio_components(vio);
169 	vdo_free(vio);
170 }
171 
172 /* Set bio properties for a VDO read or write. */
173 void vdo_set_bio_properties(struct bio *bio, struct vio *vio, bio_end_io_t callback,
174 			    blk_opf_t bi_opf, physical_block_number_t pbn)
175 {
176 	struct vdo *vdo = vio->completion.vdo;
177 	struct device_config *config = vdo->device_config;
178 
179 	pbn -= vdo->geometry.bio_offset;
180 	vio->bio_zone = ((pbn / config->thread_counts.bio_rotation_interval) %
181 			 config->thread_counts.bio_threads);
182 
183 	bio->bi_private = vio;
184 	bio->bi_end_io = callback;
185 	bio->bi_opf = bi_opf;
186 	bio->bi_iter.bi_sector = pbn * VDO_SECTORS_PER_BLOCK;
187 }
188 
189 /*
190  * Prepares the bio to perform IO with the specified buffer. May only be used on a VDO-allocated
191  * bio, as it assumes the bio wraps a 4k buffer that is 4k aligned, but there does not have to be a
192  * vio associated with the bio.
193  */
194 int vio_reset_bio(struct vio *vio, char *data, bio_end_io_t callback,
195 		  blk_opf_t bi_opf, physical_block_number_t pbn)
196 {
197 	int bvec_count, offset, len, i;
198 	struct bio *bio = vio->bio;
199 
200 	bio_reset(bio, bio->bi_bdev, bi_opf);
201 	vdo_set_bio_properties(bio, vio, callback, bi_opf, pbn);
202 	if (data == NULL)
203 		return VDO_SUCCESS;
204 
205 	bio->bi_io_vec = bio->bi_inline_vecs;
206 	bio->bi_max_vecs = vio->block_count + 1;
207 	len = VDO_BLOCK_SIZE * vio->block_count;
208 	offset = offset_in_page(data);
209 	bvec_count = DIV_ROUND_UP(offset + len, PAGE_SIZE);
210 
211 	/*
212 	 * If we knew that data was always on one page, or contiguous pages, we wouldn't need the
213 	 * loop. But if we're using vmalloc, it's not impossible that the data is in different
214 	 * pages that can't be merged in bio_add_page...
215 	 */
216 	for (i = 0; (i < bvec_count) && (len > 0); i++) {
217 		struct page *page;
218 		int bytes_added;
219 		int bytes = PAGE_SIZE - offset;
220 
221 		if (bytes > len)
222 			bytes = len;
223 
224 		page = is_vmalloc_addr(data) ? vmalloc_to_page(data) : virt_to_page(data);
225 		bytes_added = bio_add_page(bio, page, bytes, offset);
226 
227 		if (bytes_added != bytes) {
228 			return vdo_log_error_strerror(VDO_BIO_CREATION_FAILED,
229 						      "Could only add %i bytes to bio",
230 						      bytes_added);
231 		}
232 
233 		data += bytes;
234 		len -= bytes;
235 		offset = 0;
236 	}
237 
238 	return VDO_SUCCESS;
239 }
240 
241 /**
242  * update_vio_error_stats() - Update per-vio error stats and log the error.
243  * @vio: The vio which got an error.
244  * @format: The format of the message to log (a printf style format).
245  */
246 void update_vio_error_stats(struct vio *vio, const char *format, ...)
247 {
248 	static DEFINE_RATELIMIT_STATE(error_limiter, DEFAULT_RATELIMIT_INTERVAL,
249 				      DEFAULT_RATELIMIT_BURST);
250 	va_list args;
251 	int priority;
252 	struct vdo *vdo = vio->completion.vdo;
253 
254 	switch (vio->completion.result) {
255 	case VDO_READ_ONLY:
256 		atomic64_inc(&vdo->stats.read_only_error_count);
257 		return;
258 
259 	case VDO_NO_SPACE:
260 		atomic64_inc(&vdo->stats.no_space_error_count);
261 		priority = VDO_LOG_DEBUG;
262 		break;
263 
264 	default:
265 		priority = VDO_LOG_ERR;
266 	}
267 
268 	if (!__ratelimit(&error_limiter))
269 		return;
270 
271 	va_start(args, format);
272 	vdo_vlog_strerror(priority, vio->completion.result, VDO_LOGGING_MODULE_NAME,
273 			  format, args);
274 	va_end(args);
275 }
276 
277 void vio_record_metadata_io_error(struct vio *vio)
278 {
279 	const char *description;
280 	physical_block_number_t pbn = pbn_from_vio_bio(vio->bio);
281 
282 	if (bio_op(vio->bio) == REQ_OP_READ) {
283 		description = "read";
284 	} else if ((vio->bio->bi_opf & REQ_PREFLUSH) == REQ_PREFLUSH) {
285 		description = (((vio->bio->bi_opf & REQ_FUA) == REQ_FUA) ?
286 			       "write+preflush+fua" :
287 			       "write+preflush");
288 	} else if ((vio->bio->bi_opf & REQ_FUA) == REQ_FUA) {
289 		description = "write+fua";
290 	} else {
291 		description = "write";
292 	}
293 
294 	update_vio_error_stats(vio,
295 			       "Completing %s vio of type %u for physical block %llu with error",
296 			       description, vio->type, (unsigned long long) pbn);
297 }
298 
299 /**
300  * make_vio_pool() - Create a new vio pool.
301  * @vdo: The vdo.
302  * @pool_size: The number of vios in the pool.
303  * @thread_id: The ID of the thread using this pool.
304  * @vio_type: The type of vios in the pool.
305  * @priority: The priority with which vios from the pool should be enqueued.
306  * @context: The context that each entry will have.
307  * @pool_ptr: The resulting pool.
308  *
309  * Return: A success or error code.
310  */
311 int make_vio_pool(struct vdo *vdo, size_t pool_size, thread_id_t thread_id,
312 		  enum vio_type vio_type, enum vio_priority priority, void *context,
313 		  struct vio_pool **pool_ptr)
314 {
315 	struct vio_pool *pool;
316 	char *ptr;
317 	int result;
318 
319 	result = vdo_allocate_extended(struct vio_pool, pool_size, struct pooled_vio,
320 				       __func__, &pool);
321 	if (result != VDO_SUCCESS)
322 		return result;
323 
324 	pool->thread_id = thread_id;
325 	INIT_LIST_HEAD(&pool->available);
326 	INIT_LIST_HEAD(&pool->busy);
327 
328 	result = vdo_allocate(pool_size * VDO_BLOCK_SIZE, char,
329 			      "VIO pool buffer", &pool->buffer);
330 	if (result != VDO_SUCCESS) {
331 		free_vio_pool(pool);
332 		return result;
333 	}
334 
335 	ptr = pool->buffer;
336 	for (pool->size = 0; pool->size < pool_size; pool->size++, ptr += VDO_BLOCK_SIZE) {
337 		struct pooled_vio *pooled = &pool->vios[pool->size];
338 
339 		result = allocate_vio_components(vdo, vio_type, priority, NULL, 1, ptr,
340 						 &pooled->vio);
341 		if (result != VDO_SUCCESS) {
342 			free_vio_pool(pool);
343 			return result;
344 		}
345 
346 		pooled->context = context;
347 		list_add_tail(&pooled->pool_entry, &pool->available);
348 	}
349 
350 	*pool_ptr = pool;
351 	return VDO_SUCCESS;
352 }
353 
354 /**
355  * free_vio_pool() - Destroy a vio pool.
356  * @pool: The pool to free.
357  */
358 void free_vio_pool(struct vio_pool *pool)
359 {
360 	struct pooled_vio *pooled, *tmp;
361 
362 	if (pool == NULL)
363 		return;
364 
365 	/* Remove all available vios from the object pool. */
366 	VDO_ASSERT_LOG_ONLY(!vdo_waitq_has_waiters(&pool->waiting),
367 			    "VIO pool must not have any waiters when being freed");
368 	VDO_ASSERT_LOG_ONLY((pool->busy_count == 0),
369 			    "VIO pool must not have %zu busy entries when being freed",
370 			    pool->busy_count);
371 	VDO_ASSERT_LOG_ONLY(list_empty(&pool->busy),
372 			    "VIO pool must not have busy entries when being freed");
373 
374 	list_for_each_entry_safe(pooled, tmp, &pool->available, pool_entry) {
375 		list_del(&pooled->pool_entry);
376 		free_vio_components(&pooled->vio);
377 		pool->size--;
378 	}
379 
380 	VDO_ASSERT_LOG_ONLY(pool->size == 0,
381 			    "VIO pool must not have missing entries when being freed");
382 
383 	vdo_free(vdo_forget(pool->buffer));
384 	vdo_free(pool);
385 }
386 
387 /**
388  * is_vio_pool_busy() - Check whether an vio pool has outstanding entries.
389  *
390  * Return: true if the pool is busy.
391  */
392 bool is_vio_pool_busy(struct vio_pool *pool)
393 {
394 	return (pool->busy_count != 0);
395 }
396 
397 /**
398  * acquire_vio_from_pool() - Acquire a vio and buffer from the pool (asynchronous).
399  * @pool: The vio pool.
400  * @waiter: Object that is requesting a vio.
401  */
402 void acquire_vio_from_pool(struct vio_pool *pool, struct vdo_waiter *waiter)
403 {
404 	struct pooled_vio *pooled;
405 
406 	VDO_ASSERT_LOG_ONLY((pool->thread_id == vdo_get_callback_thread_id()),
407 			    "acquire from active vio_pool called from correct thread");
408 
409 	if (list_empty(&pool->available)) {
410 		vdo_waitq_enqueue_waiter(&pool->waiting, waiter);
411 		return;
412 	}
413 
414 	pooled = list_first_entry(&pool->available, struct pooled_vio, pool_entry);
415 	pool->busy_count++;
416 	list_move_tail(&pooled->pool_entry, &pool->busy);
417 	(*waiter->callback)(waiter, pooled);
418 }
419 
420 /**
421  * return_vio_to_pool() - Return a vio to the pool
422  * @pool: The vio pool.
423  * @vio: The pooled vio to return.
424  */
425 void return_vio_to_pool(struct vio_pool *pool, struct pooled_vio *vio)
426 {
427 	VDO_ASSERT_LOG_ONLY((pool->thread_id == vdo_get_callback_thread_id()),
428 			    "vio pool entry returned on same thread as it was acquired");
429 
430 	vio->vio.completion.error_handler = NULL;
431 	vio->vio.completion.parent = NULL;
432 	if (vdo_waitq_has_waiters(&pool->waiting)) {
433 		vdo_waitq_notify_next_waiter(&pool->waiting, NULL, vio);
434 		return;
435 	}
436 
437 	list_move_tail(&vio->pool_entry, &pool->available);
438 	--pool->busy_count;
439 }
440 
441 /*
442  * Various counting functions for statistics.
443  * These are used for bios coming into VDO, as well as bios generated by VDO.
444  */
445 void vdo_count_bios(struct atomic_bio_stats *bio_stats, struct bio *bio)
446 {
447 	if (((bio->bi_opf & REQ_PREFLUSH) != 0) && (bio->bi_iter.bi_size == 0)) {
448 		atomic64_inc(&bio_stats->empty_flush);
449 		atomic64_inc(&bio_stats->flush);
450 		return;
451 	}
452 
453 	switch (bio_op(bio)) {
454 	case REQ_OP_WRITE:
455 		atomic64_inc(&bio_stats->write);
456 		break;
457 	case REQ_OP_READ:
458 		atomic64_inc(&bio_stats->read);
459 		break;
460 	case REQ_OP_DISCARD:
461 		atomic64_inc(&bio_stats->discard);
462 		break;
463 		/*
464 		 * All other operations are filtered out in dmvdo.c, or not created by VDO, so
465 		 * shouldn't exist.
466 		 */
467 	default:
468 		VDO_ASSERT_LOG_ONLY(0, "Bio operation %d not a write, read, discard, or empty flush",
469 				    bio_op(bio));
470 	}
471 
472 	if ((bio->bi_opf & REQ_PREFLUSH) != 0)
473 		atomic64_inc(&bio_stats->flush);
474 	if (bio->bi_opf & REQ_FUA)
475 		atomic64_inc(&bio_stats->fua);
476 }
477 
478 static void count_all_bios_completed(struct vio *vio, struct bio *bio)
479 {
480 	struct atomic_statistics *stats = &vio->completion.vdo->stats;
481 
482 	if (is_data_vio(vio)) {
483 		vdo_count_bios(&stats->bios_out_completed, bio);
484 		return;
485 	}
486 
487 	vdo_count_bios(&stats->bios_meta_completed, bio);
488 	if (vio->type == VIO_TYPE_RECOVERY_JOURNAL)
489 		vdo_count_bios(&stats->bios_journal_completed, bio);
490 	else if (vio->type == VIO_TYPE_BLOCK_MAP)
491 		vdo_count_bios(&stats->bios_page_cache_completed, bio);
492 }
493 
494 void vdo_count_completed_bios(struct bio *bio)
495 {
496 	struct vio *vio = (struct vio *) bio->bi_private;
497 
498 	atomic64_inc(&vio->completion.vdo->stats.bios_completed);
499 	count_all_bios_completed(vio, bio);
500 }
501