io-submitter.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright 2023 Red Hat
  4 */
  5
  6#include "io-submitter.h"
  7
  8#include <linux/bio.h>
  9#include <linux/kernel.h>
 10#include <linux/mutex.h>
 11
 12#include "memory-alloc.h"
 13#include "permassert.h"
 14
 15#include "data-vio.h"
 16#include "logger.h"
 17#include "types.h"
 18#include "vdo.h"
 19#include "vio.h"
 20
 21/*
 22 * Submission of bio operations to the underlying storage device will go through a separate work
 23 * queue thread (or more than one) to prevent blocking in other threads if the storage device has a
 24 * full queue. The plug structure allows that thread to do better batching of requests to make the
 25 * I/O more efficient.
 26 *
 27 * When multiple worker threads are used, a thread is chosen for a I/O operation submission based
 28 * on the PBN, so a given PBN will consistently wind up on the same thread. Flush operations are
 29 * assigned round-robin.
 30 *
 31 * The map (protected by the mutex) collects pending I/O operations so that the worker thread can
 32 * reorder them to try to encourage I/O request merging in the request queue underneath.
 33 */
 34struct bio_queue_data {
 35	struct vdo_work_queue *queue;
 36	struct blk_plug plug;
 37	struct int_map *map;
 38	struct mutex lock;
 39	unsigned int queue_number;
 40};
 41
 42struct io_submitter {
 43	unsigned int num_bio_queues_used;
 44	unsigned int bio_queue_rotation_interval;
 45	struct bio_queue_data bio_queue_data[];
 46};
 47
 48static void start_bio_queue(void *ptr)
 49{
 50	struct bio_queue_data *bio_queue_data = ptr;
 51
 52	blk_start_plug(&bio_queue_data->plug);
 53}
 54
 55static void finish_bio_queue(void *ptr)
 56{
 57	struct bio_queue_data *bio_queue_data = ptr;
 58
 59	blk_finish_plug(&bio_queue_data->plug);
 60}
 61
 62static const struct vdo_work_queue_type bio_queue_type = {
 63	.start = start_bio_queue,
 64	.finish = finish_bio_queue,
 65	.max_priority = BIO_Q_MAX_PRIORITY,
 66	.default_priority = BIO_Q_DATA_PRIORITY,
 67};
 68
 69/**
 70 * count_all_bios() - Determine which bio counter to use.
 71 * @vio: The vio associated with the bio.
 72 * @bio: The bio to count.
 73 */
 74static void count_all_bios(struct vio *vio, struct bio *bio)
 75{
 76	struct atomic_statistics *stats = &vio->completion.vdo->stats;
 77
 78	if (is_data_vio(vio)) {
 79		vdo_count_bios(&stats->bios_out, bio);
 80		return;
 81	}
 82
 83	vdo_count_bios(&stats->bios_meta, bio);
 84	if (vio->type == VIO_TYPE_RECOVERY_JOURNAL)
 85		vdo_count_bios(&stats->bios_journal, bio);
 86	else if (vio->type == VIO_TYPE_BLOCK_MAP)
 87		vdo_count_bios(&stats->bios_page_cache, bio);
 88}
 89
 90/**
 91 * assert_in_bio_zone() - Assert that a vio is in the correct bio zone and not in interrupt
 92 *                        context.
 93 * @vio: The vio to check.
 94 */
 95static void assert_in_bio_zone(struct vio *vio)
 96{
 97	VDO_ASSERT_LOG_ONLY(!in_interrupt(), "not in interrupt context");
 98	assert_vio_in_bio_zone(vio);
 99}
100
101/**
102 * send_bio_to_device() - Update stats and tracing info, then submit the supplied bio to the OS for
103 *                        processing.
104 * @vio: The vio associated with the bio.
105 * @bio: The bio to submit to the OS.
106 */
107static void send_bio_to_device(struct vio *vio, struct bio *bio)
108{
109	struct vdo *vdo = vio->completion.vdo;
110
111	assert_in_bio_zone(vio);
112	atomic64_inc(&vdo->stats.bios_submitted);
113	count_all_bios(vio, bio);
114	bio_set_dev(bio, vdo_get_backing_device(vdo));
115	submit_bio_noacct(bio);
116}
117
118/**
119 * vdo_submit_vio() - Submits a vio's bio to the underlying block device. May block if the device
120 *		      is busy. This callback should be used by vios which did not attempt to merge.
121 * @completion: The vio to submit.
122 */
123void vdo_submit_vio(struct vdo_completion *completion)
124{
125	struct vio *vio = as_vio(completion);
126
127	send_bio_to_device(vio, vio->bio);
128}
129
130/**
131 * get_bio_list() - Extract the list of bios to submit from a vio.
132 * @vio: The vio submitting I/O.
133 *
134 * The list will always contain at least one entry (the bio for the vio on which it is called), but
135 * other bios may have been merged with it as well.
136 *
137 * Return: The head of the bio list to submit.
138 */
139static struct bio *get_bio_list(struct vio *vio)
140{
141	struct bio *bio;
142	struct io_submitter *submitter = vio->completion.vdo->io_submitter;
143	struct bio_queue_data *bio_queue_data = &(submitter->bio_queue_data[vio->bio_zone]);
144
145	assert_in_bio_zone(vio);
146
147	mutex_lock(&bio_queue_data->lock);
148	vdo_int_map_remove(bio_queue_data->map,
149			   vio->bios_merged.head->bi_iter.bi_sector);
150	vdo_int_map_remove(bio_queue_data->map,
151			   vio->bios_merged.tail->bi_iter.bi_sector);
152	bio = vio->bios_merged.head;
153	bio_list_init(&vio->bios_merged);
154	mutex_unlock(&bio_queue_data->lock);
155
156	return bio;
157}
158
159/**
160 * submit_data_vio() - Submit a data_vio's bio to the storage below along with
161 *		       any bios that have been merged with it.
162 * @completion: The vio to submit.
163 *
164 * Context: This call may block and so should only be called from a bio thread.
165 */
166static void submit_data_vio(struct vdo_completion *completion)
167{
168	struct bio *bio, *next;
169	struct vio *vio = as_vio(completion);
170
171	assert_in_bio_zone(vio);
172	for (bio = get_bio_list(vio); bio != NULL; bio = next) {
173		next = bio->bi_next;
174		bio->bi_next = NULL;
175		send_bio_to_device((struct vio *) bio->bi_private, bio);
176	}
177}
178
179/**
180 * get_mergeable_locked() - Attempt to find an already queued bio that the current bio can be
181 *                          merged with.
182 * @map: The bio map to use for merging.
183 * @vio: The vio we want to merge.
184 * @back_merge: Set to true for a back merge, false for a front merge.
185 *
186 * There are two types of merging possible, forward and backward, which are distinguished by a flag
187 * that uses kernel elevator terminology.
188 *
189 * Return: The vio to merge to, NULL if no merging is possible.
190 */
191static struct vio *get_mergeable_locked(struct int_map *map, struct vio *vio,
192					bool back_merge)
193{
194	struct bio *bio = vio->bio;
195	sector_t merge_sector = bio->bi_iter.bi_sector;
196	struct vio *vio_merge;
197
198	if (back_merge)
199		merge_sector -= VDO_SECTORS_PER_BLOCK;
200	else
201		merge_sector += VDO_SECTORS_PER_BLOCK;
202
203	vio_merge = vdo_int_map_get(map, merge_sector);
204
205	if (vio_merge == NULL)
206		return NULL;
207
208	if (vio->completion.priority != vio_merge->completion.priority)
209		return NULL;
210
211	if (bio_data_dir(bio) != bio_data_dir(vio_merge->bio))
212		return NULL;
213
214	if (bio_list_empty(&vio_merge->bios_merged))
215		return NULL;
216
217	if (back_merge) {
218		return (vio_merge->bios_merged.tail->bi_iter.bi_sector == merge_sector ?
219			vio_merge : NULL);
220	}
221
222	return (vio_merge->bios_merged.head->bi_iter.bi_sector == merge_sector ?
223		vio_merge : NULL);
224}
225
226static int map_merged_vio(struct int_map *bio_map, struct vio *vio)
227{
228	int result;
229	sector_t bio_sector;
230
231	bio_sector = vio->bios_merged.head->bi_iter.bi_sector;
232	result = vdo_int_map_put(bio_map, bio_sector, vio, true, NULL);
233	if (result != VDO_SUCCESS)
234		return result;
235
236	bio_sector = vio->bios_merged.tail->bi_iter.bi_sector;
237	return vdo_int_map_put(bio_map, bio_sector, vio, true, NULL);
238}
239
240static int merge_to_prev_tail(struct int_map *bio_map, struct vio *vio,
241			      struct vio *prev_vio)
242{
243	vdo_int_map_remove(bio_map, prev_vio->bios_merged.tail->bi_iter.bi_sector);
244	bio_list_merge(&prev_vio->bios_merged, &vio->bios_merged);
245	return map_merged_vio(bio_map, prev_vio);
246}
247
248static int merge_to_next_head(struct int_map *bio_map, struct vio *vio,
249			      struct vio *next_vio)
250{
251	/*
252	 * Handle "next merge" and "gap fill" cases the same way so as to reorder bios in a way
253	 * that's compatible with using funnel queues in work queues. This avoids removing an
254	 * existing completion.
255	 */
256	vdo_int_map_remove(bio_map, next_vio->bios_merged.head->bi_iter.bi_sector);
257	bio_list_merge_head(&next_vio->bios_merged, &vio->bios_merged);
258	return map_merged_vio(bio_map, next_vio);
259}
260
261/**
262 * try_bio_map_merge() - Attempt to merge a vio's bio with other pending I/Os.
263 * @vio: The vio to merge.
264 *
265 * Currently this is only used for data_vios, but is broken out for future use with metadata vios.
266 *
267 * Return: Whether or not the vio was merged.
268 */
269static bool try_bio_map_merge(struct vio *vio)
270{
271	int result;
272	bool merged = true;
273	struct bio *bio = vio->bio;
274	struct vio *prev_vio, *next_vio;
275	struct vdo *vdo = vio->completion.vdo;
276	struct bio_queue_data *bio_queue_data =
277		&vdo->io_submitter->bio_queue_data[vio->bio_zone];
278
279	bio->bi_next = NULL;
280	bio_list_init(&vio->bios_merged);
281	bio_list_add(&vio->bios_merged, bio);
282
283	mutex_lock(&bio_queue_data->lock);
284	prev_vio = get_mergeable_locked(bio_queue_data->map, vio, true);
285	next_vio = get_mergeable_locked(bio_queue_data->map, vio, false);
286	if (prev_vio == next_vio)
287		next_vio = NULL;
288
289	if ((prev_vio == NULL) && (next_vio == NULL)) {
290		/* no merge. just add to bio_queue */
291		merged = false;
292		result = vdo_int_map_put(bio_queue_data->map,
293					 bio->bi_iter.bi_sector,
294					 vio, true, NULL);
295	} else if (next_vio == NULL) {
296		/* Only prev. merge to prev's tail */
297		result = merge_to_prev_tail(bio_queue_data->map, vio, prev_vio);
298	} else {
299		/* Only next. merge to next's head */
300		result = merge_to_next_head(bio_queue_data->map, vio, next_vio);
301	}
302	mutex_unlock(&bio_queue_data->lock);
303
304	/* We don't care about failure of int_map_put in this case. */
305	VDO_ASSERT_LOG_ONLY(result == VDO_SUCCESS, "bio map insertion succeeds");
306	return merged;
307}
308
309/**
310 * vdo_submit_data_vio() - Submit I/O for a data_vio.
311 * @data_vio: The data_vio for which to issue I/O.
312 *
313 * If possible, this I/O will be merged other pending I/Os. Otherwise, the data_vio will be sent to
314 * the appropriate bio zone directly.
315 */
316void vdo_submit_data_vio(struct data_vio *data_vio)
317{
318	if (try_bio_map_merge(&data_vio->vio))
319		return;
320
321	launch_data_vio_bio_zone_callback(data_vio, submit_data_vio);
322}
323
324/**
325 * __submit_metadata_vio() - Submit I/O for a metadata vio.
326 * @vio: The vio for which to issue I/O.
327 * @physical: The physical block number to read or write.
328 * @callback: The bio endio function which will be called after the I/O completes.
329 * @error_handler: The handler for submission or I/O errors; may be NULL.
330 * @operation: The type of I/O to perform.
331 * @data: The buffer to read or write; may be NULL.
332 * @size: The I/O amount in bytes.
333 *
334 * The vio is enqueued on a vdo bio queue so that bio submission (which may block) does not block
335 * other vdo threads.
336 *
337 * That the error handler will run on the correct thread is only true so long as the thread calling
338 * this function, and the thread set in the endio callback are the same, as well as the fact that
339 * no error can occur on the bio queue. Currently this is true for all callers, but additional care
340 * will be needed if this ever changes.
341 */
342void __submit_metadata_vio(struct vio *vio, physical_block_number_t physical,
343			   bio_end_io_t callback, vdo_action_fn error_handler,
344			   blk_opf_t operation, char *data, int size)
345{
346	int result;
347	struct vdo_completion *completion = &vio->completion;
348	const struct admin_state_code *code = vdo_get_admin_state(completion->vdo);
349
350
351	VDO_ASSERT_LOG_ONLY(!code->quiescent, "I/O not allowed in state %s", code->name);
352
353	vdo_reset_completion(completion);
354	completion->error_handler = error_handler;
355	result = vio_reset_bio_with_size(vio, data, size, callback, operation | REQ_META,
356					 physical);
357	if (result != VDO_SUCCESS) {
358		continue_vio(vio, result);
359		return;
360	}
361
362	vdo_set_completion_callback(completion, vdo_submit_vio,
363				    get_vio_bio_zone_thread_id(vio));
364	vdo_launch_completion_with_priority(completion, get_metadata_priority(vio));
365}
366
367/**
368 * vdo_make_io_submitter() - Create an io_submitter structure.
369 * @thread_count: Number of bio-submission threads to set up.
370 * @rotation_interval: Interval to use when rotating between bio-submission threads when enqueuing
371 *                     completions.
372 * @max_requests_active: Number of bios for merge tracking.
373 * @vdo: The vdo which will use this submitter.
374 * @io_submitter_ptr: pointer to the new data structure.
375 *
376 * Return: VDO_SUCCESS or an error.
377 */
378int vdo_make_io_submitter(unsigned int thread_count, unsigned int rotation_interval,
379			  unsigned int max_requests_active, struct vdo *vdo,
380			  struct io_submitter **io_submitter_ptr)
381{
382	unsigned int i;
383	struct io_submitter *io_submitter;
384	int result;
385
386	result = vdo_allocate_extended(struct io_submitter, thread_count,
387				       struct bio_queue_data, "bio submission data",
388				       &io_submitter);
389	if (result != VDO_SUCCESS)
390		return result;
391
392	io_submitter->bio_queue_rotation_interval = rotation_interval;
393
394	/* Setup for each bio-submission work queue */
395	for (i = 0; i < thread_count; i++) {
396		struct bio_queue_data *bio_queue_data = &io_submitter->bio_queue_data[i];
397
398		mutex_init(&bio_queue_data->lock);
399		/*
400		 * One I/O operation per request, but both first & last sector numbers.
401		 *
402		 * If requests are assigned to threads round-robin, they should be distributed
403		 * quite evenly. But if they're assigned based on PBN, things can sometimes be very
404		 * uneven. So for now, we'll assume that all requests *may* wind up on one thread,
405		 * and thus all in the same map.
406		 */
407		result = vdo_int_map_create(max_requests_active * 2,
408					    &bio_queue_data->map);
409		if (result != VDO_SUCCESS) {
410			/*
411			 * Clean up the partially initialized bio-queue entirely and indicate that
412			 * initialization failed.
413			 */
414			vdo_log_error("bio map initialization failed %d", result);
415			vdo_cleanup_io_submitter(io_submitter);
416			vdo_free_io_submitter(io_submitter);
417			return result;
418		}
419
420		bio_queue_data->queue_number = i;
421		result = vdo_make_thread(vdo, vdo->thread_config.bio_threads[i],
422					 &bio_queue_type, 1, (void **) &bio_queue_data);
423		if (result != VDO_SUCCESS) {
424			/*
425			 * Clean up the partially initialized bio-queue entirely and indicate that
426			 * initialization failed.
427			 */
428			vdo_int_map_free(vdo_forget(bio_queue_data->map));
429			vdo_log_error("bio queue initialization failed %d", result);
430			vdo_cleanup_io_submitter(io_submitter);
431			vdo_free_io_submitter(io_submitter);
432			return result;
433		}
434
435		bio_queue_data->queue = vdo->threads[vdo->thread_config.bio_threads[i]].queue;
436		io_submitter->num_bio_queues_used++;
437	}
438
439	*io_submitter_ptr = io_submitter;
440
441	return VDO_SUCCESS;
442}
443
444/**
445 * vdo_cleanup_io_submitter() - Tear down the io_submitter fields as needed for a physical layer.
446 * @io_submitter: The I/O submitter data to tear down; may be NULL.
447 */
448void vdo_cleanup_io_submitter(struct io_submitter *io_submitter)
449{
450	int i;
451
452	if (io_submitter == NULL)
453		return;
454
455	for (i = io_submitter->num_bio_queues_used - 1; i >= 0; i--)
456		vdo_finish_work_queue(io_submitter->bio_queue_data[i].queue);
457}
458
459/**
460 * vdo_free_io_submitter() - Free the io_submitter fields and structure as needed.
461 * @io_submitter: The I/O submitter data to destroy.
462 *
463 * This must be called after vdo_cleanup_io_submitter(). It is used to release resources late in
464 * the shutdown process to avoid or reduce the chance of race conditions.
465 */
466void vdo_free_io_submitter(struct io_submitter *io_submitter)
467{
468	int i;
469
470	if (io_submitter == NULL)
471		return;
472
473	for (i = io_submitter->num_bio_queues_used - 1; i >= 0; i--) {
474		io_submitter->num_bio_queues_used--;
475		/* vdo_destroy() will free the work queue, so just give up our reference to it. */
476		vdo_forget(io_submitter->bio_queue_data[i].queue);
477		vdo_int_map_free(vdo_forget(io_submitter->bio_queue_data[i].map));
478	}
479	vdo_free(io_submitter);
480}