1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 2023 Red Hat
4 */
5
6#include "logical-zone.h"
7
8#include "logger.h"
9#include "memory-alloc.h"
10#include "permassert.h"
11#include "string-utils.h"
12
13#include "action-manager.h"
14#include "admin-state.h"
15#include "block-map.h"
16#include "completion.h"
17#include "constants.h"
18#include "data-vio.h"
19#include "flush.h"
20#include "int-map.h"
21#include "physical-zone.h"
22#include "vdo.h"
23
24#define ALLOCATIONS_PER_ZONE 128
25
26/**
27 * as_logical_zone() - Convert a generic vdo_completion to a logical_zone.
28 * @completion: The completion to convert.
29 *
30 * Return: The completion as a logical_zone.
31 */
32static struct logical_zone *as_logical_zone(struct vdo_completion *completion)
33{
34	vdo_assert_completion_type(completion, VDO_GENERATION_FLUSHED_COMPLETION);
35	return container_of(completion, struct logical_zone, completion);
36}
37
38/* get_thread_id_for_zone() - Implements vdo_zone_thread_getter_fn. */
39static thread_id_t get_thread_id_for_zone(void *context, zone_count_t zone_number)
40{
41	struct logical_zones *zones = context;
42
43	return zones->zones[zone_number].thread_id;
44}
45
46/**
47 * initialize_zone() - Initialize a logical zone.
48 * @zones: The logical_zones to which this zone belongs.
49 * @zone_number: The logical_zone's index.
50 */
51static int initialize_zone(struct logical_zones *zones, zone_count_t zone_number)
52{
53	int result;
54	struct vdo *vdo = zones->vdo;
55	struct logical_zone *zone = &zones->zones[zone_number];
56	zone_count_t allocation_zone_number;
57
58	result = vdo_int_map_create(VDO_LOCK_MAP_CAPACITY, &zone->lbn_operations);
59	if (result != VDO_SUCCESS)
60		return result;
61
62	if (zone_number < vdo->thread_config.logical_zone_count - 1)
63		zone->next = &zones->zones[zone_number + 1];
64
65	vdo_initialize_completion(&zone->completion, vdo,
66				  VDO_GENERATION_FLUSHED_COMPLETION);
67	zone->zones = zones;
68	zone->zone_number = zone_number;
69	zone->thread_id = vdo->thread_config.logical_threads[zone_number];
70	zone->block_map_zone = &vdo->block_map->zones[zone_number];
71	INIT_LIST_HEAD(&zone->write_vios);
72	vdo_set_admin_state_code(&zone->state, VDO_ADMIN_STATE_NORMAL_OPERATION);
73
74	allocation_zone_number = zone->thread_id % vdo->thread_config.physical_zone_count;
75	zone->allocation_zone = &vdo->physical_zones->zones[allocation_zone_number];
76
77	return vdo_make_default_thread(vdo, zone->thread_id);
78}
79
80/**
81 * vdo_make_logical_zones() - Create a set of logical zones.
82 * @vdo: The vdo to which the zones will belong.
83 * @zones_ptr: A pointer to hold the new zones.
84 *
85 * Return: VDO_SUCCESS or an error code.
86 */
87int vdo_make_logical_zones(struct vdo *vdo, struct logical_zones **zones_ptr)
88{
89	struct logical_zones *zones;
90	int result;
91	zone_count_t zone;
92	zone_count_t zone_count = vdo->thread_config.logical_zone_count;
93
94	if (zone_count == 0)
95		return VDO_SUCCESS;
96
97	result = vdo_allocate_extended(struct logical_zones, zone_count,
98				       struct logical_zone, __func__, &zones);
99	if (result != VDO_SUCCESS)
100		return result;
101
102	zones->vdo = vdo;
103	zones->zone_count = zone_count;
104	for (zone = 0; zone < zone_count; zone++) {
105		result = initialize_zone(zones, zone);
106		if (result != VDO_SUCCESS) {
107			vdo_free_logical_zones(zones);
108			return result;
109		}
110	}
111
112	result = vdo_make_action_manager(zones->zone_count, get_thread_id_for_zone,
113					 vdo->thread_config.admin_thread, zones, NULL,
114					 vdo, &zones->manager);
115	if (result != VDO_SUCCESS) {
116		vdo_free_logical_zones(zones);
117		return result;
118	}
119
120	*zones_ptr = zones;
121	return VDO_SUCCESS;
122}
123
124/**
125 * vdo_free_logical_zones() - Free a set of logical zones.
126 * @zones: The set of zones to free.
127 */
128void vdo_free_logical_zones(struct logical_zones *zones)
129{
130	zone_count_t index;
131
132	if (zones == NULL)
133		return;
134
135	vdo_free(vdo_forget(zones->manager));
136
137	for (index = 0; index < zones->zone_count; index++)
138		vdo_int_map_free(vdo_forget(zones->zones[index].lbn_operations));
139
140	vdo_free(zones);
141}
142
143static inline void assert_on_zone_thread(struct logical_zone *zone, const char *what)
144{
145	VDO_ASSERT_LOG_ONLY((vdo_get_callback_thread_id() == zone->thread_id),
146			    "%s() called on correct thread", what);
147}
148
149/**
150 * check_for_drain_complete() - Check whether this zone has drained.
151 * @zone: The zone to check.
152 */
153static void check_for_drain_complete(struct logical_zone *zone)
154{
155	if (!vdo_is_state_draining(&zone->state) || zone->notifying ||
156	    !list_empty(&zone->write_vios))
157		return;
158
159	vdo_finish_draining(&zone->state);
160}
161
162/**
163 * initiate_drain() - Initiate a drain.
164 *
165 * Implements vdo_admin_initiator_fn.
166 */
167static void initiate_drain(struct admin_state *state)
168{
169	check_for_drain_complete(container_of(state, struct logical_zone, state));
170}
171
172/**
173 * drain_logical_zone() - Drain a logical zone.
174 *
175 * Implements vdo_zone_action_fn.
176 */
177static void drain_logical_zone(void *context, zone_count_t zone_number,
178			       struct vdo_completion *parent)
179{
180	struct logical_zones *zones = context;
181
182	vdo_start_draining(&zones->zones[zone_number].state,
183			   vdo_get_current_manager_operation(zones->manager), parent,
184			   initiate_drain);
185}
186
187void vdo_drain_logical_zones(struct logical_zones *zones,
188			     const struct admin_state_code *operation,
189			     struct vdo_completion *parent)
190{
191	vdo_schedule_operation(zones->manager, operation, NULL, drain_logical_zone, NULL,
192			       parent);
193}
194
195/**
196 * resume_logical_zone() - Resume a logical zone.
197 *
198 * Implements vdo_zone_action_fn.
199 */
200static void resume_logical_zone(void *context, zone_count_t zone_number,
201				struct vdo_completion *parent)
202{
203	struct logical_zone *zone = &(((struct logical_zones *) context)->zones[zone_number]);
204
205	vdo_fail_completion(parent, vdo_resume_if_quiescent(&zone->state));
206}
207
208/**
209 * vdo_resume_logical_zones() - Resume a set of logical zones.
210 * @zones: The logical zones to resume.
211 * @parent: The object to notify when the zones have resumed.
212 */
213void vdo_resume_logical_zones(struct logical_zones *zones, struct vdo_completion *parent)
214{
215	vdo_schedule_operation(zones->manager, VDO_ADMIN_STATE_RESUMING, NULL,
216			       resume_logical_zone, NULL, parent);
217}
218
219/**
220 * update_oldest_active_generation() - Update the oldest active generation.
221 * @zone: The zone.
222 *
223 * Return: true if the oldest active generation has changed.
224 */
225static bool update_oldest_active_generation(struct logical_zone *zone)
226{
227	struct data_vio *data_vio =
228		list_first_entry_or_null(&zone->write_vios, struct data_vio,
229					 write_entry);
230	sequence_number_t oldest =
231		(data_vio == NULL) ? zone->flush_generation : data_vio->flush_generation;
232
233	if (oldest == zone->oldest_active_generation)
234		return false;
235
236	WRITE_ONCE(zone->oldest_active_generation, oldest);
237	return true;
238}
239
240/**
241 * vdo_increment_logical_zone_flush_generation() - Increment the flush generation in a logical
242 *                                                 zone.
243 * @zone: The logical zone.
244 * @expected_generation: The expected value of the flush generation before the increment.
245 */
246void vdo_increment_logical_zone_flush_generation(struct logical_zone *zone,
247						 sequence_number_t expected_generation)
248{
249	assert_on_zone_thread(zone, __func__);
250	VDO_ASSERT_LOG_ONLY((zone->flush_generation == expected_generation),
251			    "logical zone %u flush generation %llu should be %llu before increment",
252			    zone->zone_number, (unsigned long long) zone->flush_generation,
253			    (unsigned long long) expected_generation);
254
255	zone->flush_generation++;
256	zone->ios_in_flush_generation = 0;
257	update_oldest_active_generation(zone);
258}
259
260/**
261 * vdo_acquire_flush_generation_lock() - Acquire the shared lock on a flush generation by a write
262 *                                       data_vio.
263 * @data_vio: The data_vio.
264 */
265void vdo_acquire_flush_generation_lock(struct data_vio *data_vio)
266{
267	struct logical_zone *zone = data_vio->logical.zone;
268
269	assert_on_zone_thread(zone, __func__);
270	VDO_ASSERT_LOG_ONLY(vdo_is_state_normal(&zone->state), "vdo state is normal");
271
272	data_vio->flush_generation = zone->flush_generation;
273	list_add_tail(&data_vio->write_entry, &zone->write_vios);
274	zone->ios_in_flush_generation++;
275}
276
277static void attempt_generation_complete_notification(struct vdo_completion *completion);
278
279/**
280 * notify_flusher() - Notify the flush that at least one generation no longer has active VIOs.
281 * @completion: The zone completion.
282 *
283 * This callback is registered in attempt_generation_complete_notification().
284 */
285static void notify_flusher(struct vdo_completion *completion)
286{
287	struct logical_zone *zone = as_logical_zone(completion);
288
289	vdo_complete_flushes(zone->zones->vdo->flusher);
290	vdo_launch_completion_callback(completion,
291				       attempt_generation_complete_notification,
292				       zone->thread_id);
293}
294
295/**
296 * attempt_generation_complete_notification() - Notify the flusher if some generation no
297 *                                              longer has active VIOs.
298 * @completion: The zone completion.
299 */
300static void attempt_generation_complete_notification(struct vdo_completion *completion)
301{
302	struct logical_zone *zone = as_logical_zone(completion);
303
304	assert_on_zone_thread(zone, __func__);
305	if (zone->oldest_active_generation <= zone->notification_generation) {
306		zone->notifying = false;
307		check_for_drain_complete(zone);
308		return;
309	}
310
311	zone->notifying = true;
312	zone->notification_generation = zone->oldest_active_generation;
313	vdo_launch_completion_callback(&zone->completion, notify_flusher,
314				       vdo_get_flusher_thread_id(zone->zones->vdo->flusher));
315}
316
317/**
318 * vdo_release_flush_generation_lock() - Release the shared lock on a flush generation held by a
319 *                                       write data_vio.
320 * @data_vio: The data_vio whose lock is to be released.
321 *
322 * If there are pending flushes, and this data_vio completes the oldest generation active in this
323 * zone, an attempt will be made to finish any flushes which may now be complete.
324 */
325void vdo_release_flush_generation_lock(struct data_vio *data_vio)
326{
327	struct logical_zone *zone = data_vio->logical.zone;
328
329	assert_on_zone_thread(zone, __func__);
330
331	if (!data_vio_has_flush_generation_lock(data_vio))
332		return;
333
334	list_del_init(&data_vio->write_entry);
335	VDO_ASSERT_LOG_ONLY((zone->oldest_active_generation <= data_vio->flush_generation),
336			    "data_vio releasing lock on generation %llu is not older than oldest active generation %llu",
337			    (unsigned long long) data_vio->flush_generation,
338			    (unsigned long long) zone->oldest_active_generation);
339
340	if (!update_oldest_active_generation(zone) || zone->notifying)
341		return;
342
343	attempt_generation_complete_notification(&zone->completion);
344}
345
346struct physical_zone *vdo_get_next_allocation_zone(struct logical_zone *zone)
347{
348	if (zone->allocation_count == ALLOCATIONS_PER_ZONE) {
349		zone->allocation_count = 0;
350		zone->allocation_zone = zone->allocation_zone->next;
351	}
352
353	zone->allocation_count++;
354	return zone->allocation_zone;
355}
356
357/**
358 * vdo_dump_logical_zone() - Dump information about a logical zone to the log for debugging.
359 * @zone: The zone to dump
360 *
361 * Context: the information is dumped in a thread-unsafe fashion.
362 *
363 */
364void vdo_dump_logical_zone(const struct logical_zone *zone)
365{
366	vdo_log_info("logical_zone %u", zone->zone_number);
367	vdo_log_info("  flush_generation=%llu oldest_active_generation=%llu notification_generation=%llu notifying=%s ios_in_flush_generation=%llu",
368		     (unsigned long long) READ_ONCE(zone->flush_generation),
369		     (unsigned long long) READ_ONCE(zone->oldest_active_generation),
370		     (unsigned long long) READ_ONCE(zone->notification_generation),
371		     vdo_bool_to_string(READ_ONCE(zone->notifying)),
372		     (unsigned long long) READ_ONCE(zone->ios_in_flush_generation));
373}
374