xref: /linux/drivers/md/dm-vdo/slab-depot.c (revision a5f998094fa344cdd1342164948abb4d7c6101ce)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2023 Red Hat
4  */
5 
6 #include "slab-depot.h"
7 
8 #include <linux/atomic.h>
9 #include <linux/bio.h>
10 #include <linux/err.h>
11 #include <linux/log2.h>
12 #include <linux/min_heap.h>
13 #include <linux/minmax.h>
14 
15 #include "logger.h"
16 #include "memory-alloc.h"
17 #include "numeric.h"
18 #include "permassert.h"
19 #include "string-utils.h"
20 
21 #include "action-manager.h"
22 #include "admin-state.h"
23 #include "completion.h"
24 #include "constants.h"
25 #include "data-vio.h"
26 #include "encodings.h"
27 #include "io-submitter.h"
28 #include "physical-zone.h"
29 #include "priority-table.h"
30 #include "recovery-journal.h"
31 #include "repair.h"
32 #include "status-codes.h"
33 #include "types.h"
34 #include "vdo.h"
35 #include "vio.h"
36 #include "wait-queue.h"
37 
38 static const u64 BYTES_PER_WORD = sizeof(u64);
39 static const bool NORMAL_OPERATION = true;
40 
41 /**
42  * get_lock() - Get the lock object for a slab journal block by sequence number.
43  * @journal: The vdo_slab journal to retrieve from.
44  * @sequence_number: Sequence number of the block.
45  *
46  * Return: The lock object for the given sequence number.
47  */
get_lock(struct slab_journal * journal,sequence_number_t sequence_number)48 static inline struct journal_lock * __must_check get_lock(struct slab_journal *journal,
49 							  sequence_number_t sequence_number)
50 {
51 	return &journal->locks[sequence_number % journal->size];
52 }
53 
is_slab_open(struct vdo_slab * slab)54 static bool is_slab_open(struct vdo_slab *slab)
55 {
56 	return (!vdo_is_state_quiescing(&slab->state) &&
57 		!vdo_is_state_quiescent(&slab->state));
58 }
59 
60 /**
61  * must_make_entries_to_flush() - Check whether there are entry waiters which should delay a flush.
62  * @journal: The journal to check.
63  *
64  * Return: true if there are no entry waiters, or if the slab is unrecovered.
65  */
must_make_entries_to_flush(struct slab_journal * journal)66 static inline bool __must_check must_make_entries_to_flush(struct slab_journal *journal)
67 {
68 	return ((journal->slab->status != VDO_SLAB_REBUILDING) &&
69 		vdo_waitq_has_waiters(&journal->entry_waiters));
70 }
71 
72 /**
73  * is_reaping() - Check whether a reap is currently in progress.
74  * @journal: The journal which may be reaping.
75  *
76  * Return: true if the journal is reaping.
77  */
is_reaping(struct slab_journal * journal)78 static inline bool __must_check is_reaping(struct slab_journal *journal)
79 {
80 	return (journal->head != journal->unreapable);
81 }
82 
83 /**
84  * initialize_tail_block() - Initialize tail block as a new block.
85  * @journal: The journal whose tail block is being initialized.
86  */
initialize_tail_block(struct slab_journal * journal)87 static void initialize_tail_block(struct slab_journal *journal)
88 {
89 	struct slab_journal_block_header *header = &journal->tail_header;
90 
91 	header->sequence_number = journal->tail;
92 	header->entry_count = 0;
93 	header->has_block_map_increments = false;
94 }
95 
96 /**
97  * initialize_journal_state() - Set all journal fields appropriately to start journaling.
98  * @journal: The journal to be reset, based on its tail sequence number.
99  */
initialize_journal_state(struct slab_journal * journal)100 static void initialize_journal_state(struct slab_journal *journal)
101 {
102 	journal->unreapable = journal->head;
103 	journal->reap_lock = get_lock(journal, journal->unreapable);
104 	journal->next_commit = journal->tail;
105 	journal->summarized = journal->last_summarized = journal->tail;
106 	initialize_tail_block(journal);
107 }
108 
109 /**
110  * block_is_full() - Check whether a journal block is full.
111  * @journal: The slab journal for the block.
112  *
113  * Return: True if the tail block is full.
114  */
block_is_full(struct slab_journal * journal)115 static bool __must_check block_is_full(struct slab_journal *journal)
116 {
117 	journal_entry_count_t count = journal->tail_header.entry_count;
118 
119 	return (journal->tail_header.has_block_map_increments ?
120 		(journal->full_entries_per_block == count) :
121 		(journal->entries_per_block == count));
122 }
123 
124 static void add_entries(struct slab_journal *journal);
125 static void update_tail_block_location(struct slab_journal *journal);
126 static void release_journal_locks(struct vdo_waiter *waiter, void *context);
127 
128 /**
129  * is_slab_journal_blank() - Check whether a slab's journal is blank.
130  * @slab: The slab to check.
131  *
132  * A slab journal is blank if it has never had any entries recorded in it.
133  *
134  * Return: True if the slab's journal has never been modified.
135  */
is_slab_journal_blank(const struct vdo_slab * slab)136 static bool is_slab_journal_blank(const struct vdo_slab *slab)
137 {
138 	return ((slab->journal.tail == 1) &&
139 		(slab->journal.tail_header.entry_count == 0));
140 }
141 
142 /**
143  * mark_slab_journal_dirty() - Put a slab journal on the dirty list of its allocator in the correct
144  *                             order.
145  * @journal: The journal to be marked dirty.
146  * @lock: The recovery journal lock held by the slab journal.
147  */
mark_slab_journal_dirty(struct slab_journal * journal,sequence_number_t lock)148 static void mark_slab_journal_dirty(struct slab_journal *journal, sequence_number_t lock)
149 {
150 	struct slab_journal *dirty_journal;
151 	struct list_head *dirty_list = &journal->slab->allocator->dirty_slab_journals;
152 
153 	VDO_ASSERT_LOG_ONLY(journal->recovery_lock == 0, "slab journal was clean");
154 
155 	journal->recovery_lock = lock;
156 	list_for_each_entry_reverse(dirty_journal, dirty_list, dirty_entry) {
157 		if (dirty_journal->recovery_lock <= journal->recovery_lock)
158 			break;
159 	}
160 
161 	list_move_tail(&journal->dirty_entry, dirty_journal->dirty_entry.next);
162 }
163 
mark_slab_journal_clean(struct slab_journal * journal)164 static void mark_slab_journal_clean(struct slab_journal *journal)
165 {
166 	journal->recovery_lock = 0;
167 	list_del_init(&journal->dirty_entry);
168 }
169 
check_if_slab_drained(struct vdo_slab * slab)170 static void check_if_slab_drained(struct vdo_slab *slab)
171 {
172 	bool read_only;
173 	struct slab_journal *journal = &slab->journal;
174 	const struct admin_state_code *code;
175 
176 	if (!vdo_is_state_draining(&slab->state) ||
177 	    must_make_entries_to_flush(journal) ||
178 	    is_reaping(journal) ||
179 	    journal->waiting_to_commit ||
180 	    !list_empty(&journal->uncommitted_blocks) ||
181 	    journal->updating_slab_summary ||
182 	    (slab->active_count > 0))
183 		return;
184 
185 	/* When not suspending or recovering, the slab must be clean. */
186 	code = vdo_get_admin_state_code(&slab->state);
187 	read_only = vdo_is_read_only(slab->allocator->depot->vdo);
188 	if (!read_only &&
189 	    vdo_waitq_has_waiters(&slab->dirty_blocks) &&
190 	    (code != VDO_ADMIN_STATE_SUSPENDING) &&
191 	    (code != VDO_ADMIN_STATE_RECOVERING))
192 		return;
193 
194 	vdo_finish_draining_with_result(&slab->state,
195 					(read_only ? VDO_READ_ONLY : VDO_SUCCESS));
196 }
197 
198 /* FULLNESS HINT COMPUTATION */
199 
200 /**
201  * compute_fullness_hint() - Translate a slab's free block count into a 'fullness hint' that can be
202  *                           stored in a slab_summary_entry's 7 bits that are dedicated to its free
203  *                           count.
204  * @depot: The depot whose summary being updated.
205  * @free_blocks: The number of free blocks.
206  *
207  * Note: the number of free blocks must be strictly less than 2^23 blocks, even though
208  * theoretically slabs could contain precisely 2^23 blocks; there is an assumption that at least
209  * one block is used by metadata. This assumption is necessary; otherwise, the fullness hint might
210  * overflow. The fullness hint formula is roughly (fullness >> 16) & 0x7f, but (2^23 >> 16) & 0x7f
211  * is 0, which would make it impossible to distinguish completely full from completely empty.
212  *
213  * Return: A fullness hint, which can be stored in 7 bits.
214  */
compute_fullness_hint(struct slab_depot * depot,block_count_t free_blocks)215 static u8 __must_check compute_fullness_hint(struct slab_depot *depot,
216 					     block_count_t free_blocks)
217 {
218 	block_count_t hint;
219 
220 	VDO_ASSERT_LOG_ONLY((free_blocks < (1 << 23)), "free blocks must be less than 2^23");
221 
222 	if (free_blocks == 0)
223 		return 0;
224 
225 	hint = free_blocks >> depot->hint_shift;
226 	return ((hint == 0) ? 1 : hint);
227 }
228 
229 /**
230  * check_summary_drain_complete() - Check whether an allocators summary has finished draining.
231  * @allocator: The allocator to check.
232  */
check_summary_drain_complete(struct block_allocator * allocator)233 static void check_summary_drain_complete(struct block_allocator *allocator)
234 {
235 	if (!vdo_is_state_draining(&allocator->summary_state) ||
236 	    (allocator->summary_write_count > 0))
237 		return;
238 
239 	vdo_finish_operation(&allocator->summary_state,
240 			     (vdo_is_read_only(allocator->depot->vdo) ?
241 			      VDO_READ_ONLY : VDO_SUCCESS));
242 }
243 
244 /**
245  * notify_summary_waiters() - Wake all the waiters in a given queue.
246  * @allocator: The block allocator summary which owns the queue.
247  * @queue: The queue to notify.
248  */
notify_summary_waiters(struct block_allocator * allocator,struct vdo_wait_queue * queue)249 static void notify_summary_waiters(struct block_allocator *allocator,
250 				   struct vdo_wait_queue *queue)
251 {
252 	int result = (vdo_is_read_only(allocator->depot->vdo) ?
253 		      VDO_READ_ONLY : VDO_SUCCESS);
254 
255 	vdo_waitq_notify_all_waiters(queue, NULL, &result);
256 }
257 
258 static void launch_write(struct slab_summary_block *summary_block);
259 
260 /**
261  * finish_updating_slab_summary_block() - Finish processing a block which attempted to write,
262  *                                        whether or not the attempt succeeded.
263  * @block: The block.
264  */
finish_updating_slab_summary_block(struct slab_summary_block * block)265 static void finish_updating_slab_summary_block(struct slab_summary_block *block)
266 {
267 	notify_summary_waiters(block->allocator, &block->current_update_waiters);
268 	block->writing = false;
269 	block->allocator->summary_write_count--;
270 	if (vdo_waitq_has_waiters(&block->next_update_waiters))
271 		launch_write(block);
272 	else
273 		check_summary_drain_complete(block->allocator);
274 }
275 
276 /**
277  * finish_update() - This is the callback for a successful summary block write.
278  * @completion: The write vio.
279  */
finish_update(struct vdo_completion * completion)280 static void finish_update(struct vdo_completion *completion)
281 {
282 	struct slab_summary_block *block =
283 		container_of(as_vio(completion), struct slab_summary_block, vio);
284 
285 	atomic64_inc(&block->allocator->depot->summary_statistics.blocks_written);
286 	finish_updating_slab_summary_block(block);
287 }
288 
289 /**
290  * handle_write_error() - Handle an error writing a slab summary block.
291  * @completion: The write VIO.
292  */
handle_write_error(struct vdo_completion * completion)293 static void handle_write_error(struct vdo_completion *completion)
294 {
295 	struct slab_summary_block *block =
296 		container_of(as_vio(completion), struct slab_summary_block, vio);
297 
298 	vio_record_metadata_io_error(as_vio(completion));
299 	vdo_enter_read_only_mode(completion->vdo, completion->result);
300 	finish_updating_slab_summary_block(block);
301 }
302 
write_slab_summary_endio(struct bio * bio)303 static void write_slab_summary_endio(struct bio *bio)
304 {
305 	struct vio *vio = bio->bi_private;
306 	struct slab_summary_block *block =
307 		container_of(vio, struct slab_summary_block, vio);
308 
309 	continue_vio_after_io(vio, finish_update, block->allocator->thread_id);
310 }
311 
312 /**
313  * launch_write() - Write a slab summary block unless it is currently out for writing.
314  * @block: The block that needs to be committed.
315  */
launch_write(struct slab_summary_block * block)316 static void launch_write(struct slab_summary_block *block)
317 {
318 	struct block_allocator *allocator = block->allocator;
319 	struct slab_depot *depot = allocator->depot;
320 	physical_block_number_t pbn;
321 
322 	if (block->writing)
323 		return;
324 
325 	allocator->summary_write_count++;
326 	vdo_waitq_transfer_all_waiters(&block->next_update_waiters,
327 				       &block->current_update_waiters);
328 	block->writing = true;
329 
330 	if (vdo_is_read_only(depot->vdo)) {
331 		finish_updating_slab_summary_block(block);
332 		return;
333 	}
334 
335 	memcpy(block->outgoing_entries, block->entries, VDO_BLOCK_SIZE);
336 
337 	/*
338 	 * Flush before writing to ensure that the slab journal tail blocks and reference updates
339 	 * covered by this summary update are stable. Otherwise, a subsequent recovery could
340 	 * encounter a slab summary update that refers to a slab journal tail block that has not
341 	 * actually been written. In such cases, the slab journal referenced will be treated as
342 	 * empty, causing any data within the slab which predates the existing recovery journal
343 	 * entries to be lost.
344 	 */
345 	pbn = (depot->summary_origin +
346 	       (VDO_SLAB_SUMMARY_BLOCKS_PER_ZONE * allocator->zone_number) +
347 	       block->index);
348 	vdo_submit_metadata_vio(&block->vio, pbn, write_slab_summary_endio,
349 				handle_write_error, REQ_OP_WRITE | REQ_PREFLUSH);
350 }
351 
352 /**
353  * update_slab_summary_entry() - Update the entry for a slab.
354  * @slab: The slab whose entry is to be updated.
355  * @waiter: The waiter that is updating the summary.
356  * @tail_block_offset: The offset of the slab journal's tail block.
357  * @load_ref_counts: Whether the reference counts must be loaded from disk on the vdo load.
358  * @is_clean: Whether the slab is clean.
359  * @free_blocks: The number of free blocks.
360  */
update_slab_summary_entry(struct vdo_slab * slab,struct vdo_waiter * waiter,tail_block_offset_t tail_block_offset,bool load_ref_counts,bool is_clean,block_count_t free_blocks)361 static void update_slab_summary_entry(struct vdo_slab *slab, struct vdo_waiter *waiter,
362 				      tail_block_offset_t tail_block_offset,
363 				      bool load_ref_counts, bool is_clean,
364 				      block_count_t free_blocks)
365 {
366 	u8 index = slab->slab_number / VDO_SLAB_SUMMARY_ENTRIES_PER_BLOCK;
367 	struct block_allocator *allocator = slab->allocator;
368 	struct slab_summary_block *block = &allocator->summary_blocks[index];
369 	int result;
370 	struct slab_summary_entry *entry;
371 
372 	if (vdo_is_read_only(block->vio.completion.vdo)) {
373 		result = VDO_READ_ONLY;
374 		waiter->callback(waiter, &result);
375 		return;
376 	}
377 
378 	if (vdo_is_state_draining(&allocator->summary_state) ||
379 	    vdo_is_state_quiescent(&allocator->summary_state)) {
380 		result = VDO_INVALID_ADMIN_STATE;
381 		waiter->callback(waiter, &result);
382 		return;
383 	}
384 
385 	entry = &allocator->summary_entries[slab->slab_number];
386 	*entry = (struct slab_summary_entry) {
387 		.tail_block_offset = tail_block_offset,
388 		.load_ref_counts = (entry->load_ref_counts || load_ref_counts),
389 		.is_dirty = !is_clean,
390 		.fullness_hint = compute_fullness_hint(allocator->depot, free_blocks),
391 	};
392 	vdo_waitq_enqueue_waiter(&block->next_update_waiters, waiter);
393 	launch_write(block);
394 }
395 
396 /**
397  * finish_reaping() - Actually advance the head of the journal now that any necessary flushes are
398  *                    complete.
399  * @journal: The journal to be reaped.
400  */
finish_reaping(struct slab_journal * journal)401 static void finish_reaping(struct slab_journal *journal)
402 {
403 	journal->head = journal->unreapable;
404 	add_entries(journal);
405 	check_if_slab_drained(journal->slab);
406 }
407 
408 static void reap_slab_journal(struct slab_journal *journal);
409 
410 /**
411  * complete_reaping() - Finish reaping now that we have flushed the lower layer and then try
412  *                      reaping again in case we deferred reaping due to an outstanding vio.
413  * @completion: The flush vio.
414  */
complete_reaping(struct vdo_completion * completion)415 static void complete_reaping(struct vdo_completion *completion)
416 {
417 	struct slab_journal *journal = completion->parent;
418 
419 	return_vio_to_pool(vio_as_pooled_vio(as_vio(completion)));
420 	finish_reaping(journal);
421 	reap_slab_journal(journal);
422 }
423 
424 /**
425  * handle_flush_error() - Handle an error flushing the lower layer.
426  * @completion: The flush vio.
427  */
handle_flush_error(struct vdo_completion * completion)428 static void handle_flush_error(struct vdo_completion *completion)
429 {
430 	vio_record_metadata_io_error(as_vio(completion));
431 	vdo_enter_read_only_mode(completion->vdo, completion->result);
432 	complete_reaping(completion);
433 }
434 
flush_endio(struct bio * bio)435 static void flush_endio(struct bio *bio)
436 {
437 	struct vio *vio = bio->bi_private;
438 	struct slab_journal *journal = vio->completion.parent;
439 
440 	continue_vio_after_io(vio, complete_reaping,
441 			      journal->slab->allocator->thread_id);
442 }
443 
444 /**
445  * flush_for_reaping() - A waiter callback for getting a vio with which to flush the lower layer
446  *                       prior to reaping.
447  * @waiter: The journal as a flush waiter.
448  * @context: The newly acquired flush vio.
449  */
flush_for_reaping(struct vdo_waiter * waiter,void * context)450 static void flush_for_reaping(struct vdo_waiter *waiter, void *context)
451 {
452 	struct slab_journal *journal =
453 		container_of(waiter, struct slab_journal, flush_waiter);
454 	struct pooled_vio *pooled = context;
455 	struct vio *vio = &pooled->vio;
456 
457 	vio->completion.parent = journal;
458 	vdo_submit_flush_vio(vio, flush_endio, handle_flush_error);
459 }
460 
461 /**
462  * reap_slab_journal() - Conduct a reap on a slab journal to reclaim unreferenced blocks.
463  * @journal: The slab journal.
464  */
reap_slab_journal(struct slab_journal * journal)465 static void reap_slab_journal(struct slab_journal *journal)
466 {
467 	bool reaped = false;
468 
469 	if (is_reaping(journal)) {
470 		/* We already have a reap in progress so wait for it to finish. */
471 		return;
472 	}
473 
474 	if ((journal->slab->status != VDO_SLAB_REBUILT) ||
475 	    !vdo_is_state_normal(&journal->slab->state) ||
476 	    vdo_is_read_only(journal->slab->allocator->depot->vdo)) {
477 		/*
478 		 * We must not reap in the first two cases, and there's no point in read-only mode.
479 		 */
480 		return;
481 	}
482 
483 	/*
484 	 * Start reclaiming blocks only when the journal head has no references. Then stop when a
485 	 * block is referenced or reap reaches the most recently written block, referenced by the
486 	 * slab summary, which has the sequence number just before the tail.
487 	 */
488 	while ((journal->unreapable < journal->tail) && (journal->reap_lock->count == 0)) {
489 		reaped = true;
490 		journal->unreapable++;
491 		journal->reap_lock++;
492 		if (journal->reap_lock == &journal->locks[journal->size])
493 			journal->reap_lock = &journal->locks[0];
494 	}
495 
496 	if (!reaped)
497 		return;
498 
499 	/*
500 	 * It is never safe to reap a slab journal block without first issuing a flush, regardless
501 	 * of whether a user flush has been received or not. In the absence of the flush, the
502 	 * reference block write which released the locks allowing the slab journal to reap may not
503 	 * be persisted. Although slab summary writes will eventually issue flushes, multiple slab
504 	 * journal block writes can be issued while previous slab summary updates have not yet been
505 	 * made. Even though those slab journal block writes will be ignored if the slab summary
506 	 * update is not persisted, they may still overwrite the to-be-reaped slab journal block
507 	 * resulting in a loss of reference count updates.
508 	 */
509 	journal->flush_waiter.callback = flush_for_reaping;
510 	acquire_vio_from_pool(journal->slab->allocator->vio_pool,
511 			      &journal->flush_waiter);
512 }
513 
514 /**
515  * adjust_slab_journal_block_reference() - Adjust the reference count for a slab journal block.
516  * @journal: The slab journal.
517  * @sequence_number: The journal sequence number of the referenced block.
518  * @adjustment: Amount to adjust the reference counter.
519  *
520  * Note that when the adjustment is negative, the slab journal will be reaped.
521  */
adjust_slab_journal_block_reference(struct slab_journal * journal,sequence_number_t sequence_number,int adjustment)522 static void adjust_slab_journal_block_reference(struct slab_journal *journal,
523 						sequence_number_t sequence_number,
524 						int adjustment)
525 {
526 	struct journal_lock *lock;
527 
528 	if (sequence_number == 0)
529 		return;
530 
531 	if (journal->slab->status == VDO_SLAB_REPLAYING) {
532 		/* Locks should not be used during offline replay. */
533 		return;
534 	}
535 
536 	VDO_ASSERT_LOG_ONLY((adjustment != 0), "adjustment must be non-zero");
537 	lock = get_lock(journal, sequence_number);
538 	if (adjustment < 0) {
539 		VDO_ASSERT_LOG_ONLY((-adjustment <= lock->count),
540 				    "adjustment %d of lock count %u for slab journal block %llu must not underflow",
541 				    adjustment, lock->count,
542 				    (unsigned long long) sequence_number);
543 	}
544 
545 	lock->count += adjustment;
546 	if (lock->count == 0)
547 		reap_slab_journal(journal);
548 }
549 
550 /**
551  * release_journal_locks() - Callback invoked after a slab summary update completes.
552  * @waiter: The slab summary waiter that has just been notified.
553  * @context: The result code of the update.
554  *
555  * Registered in the constructor on behalf of update_tail_block_location().
556  *
557  * Implements waiter_callback_fn.
558  */
release_journal_locks(struct vdo_waiter * waiter,void * context)559 static void release_journal_locks(struct vdo_waiter *waiter, void *context)
560 {
561 	sequence_number_t first, i;
562 	struct slab_journal *journal =
563 		container_of(waiter, struct slab_journal, slab_summary_waiter);
564 	int result = *((int *) context);
565 
566 	if (result != VDO_SUCCESS) {
567 		if (result != VDO_READ_ONLY) {
568 			/*
569 			 * Don't bother logging what might be lots of errors if we are already in
570 			 * read-only mode.
571 			 */
572 			vdo_log_error_strerror(result, "failed slab summary update %llu",
573 					       (unsigned long long) journal->summarized);
574 		}
575 
576 		journal->updating_slab_summary = false;
577 		vdo_enter_read_only_mode(journal->slab->allocator->depot->vdo, result);
578 		check_if_slab_drained(journal->slab);
579 		return;
580 	}
581 
582 	if (journal->partial_write_in_progress && (journal->summarized == journal->tail)) {
583 		journal->partial_write_in_progress = false;
584 		add_entries(journal);
585 	}
586 
587 	first = journal->last_summarized;
588 	journal->last_summarized = journal->summarized;
589 	for (i = journal->summarized - 1; i >= first; i--) {
590 		/*
591 		 * Release the lock the summarized block held on the recovery journal. (During
592 		 * replay, recovery_start will always be 0.)
593 		 */
594 		if (journal->recovery_journal != NULL) {
595 			zone_count_t zone_number = journal->slab->allocator->zone_number;
596 			struct journal_lock *lock = get_lock(journal, i);
597 
598 			vdo_release_recovery_journal_block_reference(journal->recovery_journal,
599 								     lock->recovery_start,
600 								     VDO_ZONE_TYPE_PHYSICAL,
601 								     zone_number);
602 		}
603 
604 		/*
605 		 * Release our own lock against reaping for blocks that are committed. (This
606 		 * function will not change locks during replay.)
607 		 */
608 		adjust_slab_journal_block_reference(journal, i, -1);
609 	}
610 
611 	journal->updating_slab_summary = false;
612 
613 	reap_slab_journal(journal);
614 
615 	/* Check if the slab summary needs to be updated again. */
616 	update_tail_block_location(journal);
617 }
618 
619 /**
620  * update_tail_block_location() - Update the tail block location in the slab summary, if necessary.
621  * @journal: The slab journal that is updating its tail block location.
622  */
update_tail_block_location(struct slab_journal * journal)623 static void update_tail_block_location(struct slab_journal *journal)
624 {
625 	block_count_t free_block_count;
626 	struct vdo_slab *slab = journal->slab;
627 
628 	if (journal->updating_slab_summary ||
629 	    vdo_is_read_only(journal->slab->allocator->depot->vdo) ||
630 	    (journal->last_summarized >= journal->next_commit)) {
631 		check_if_slab_drained(slab);
632 		return;
633 	}
634 
635 	if (slab->status != VDO_SLAB_REBUILT) {
636 		u8 hint = slab->allocator->summary_entries[slab->slab_number].fullness_hint;
637 
638 		free_block_count = ((block_count_t) hint) << slab->allocator->depot->hint_shift;
639 	} else {
640 		free_block_count = slab->free_blocks;
641 	}
642 
643 	journal->summarized = journal->next_commit;
644 	journal->updating_slab_summary = true;
645 
646 	/*
647 	 * Update slab summary as dirty.
648 	 * vdo_slab journal can only reap past sequence number 1 when all the ref counts for this
649 	 * slab have been written to the layer. Therefore, indicate that the ref counts must be
650 	 * loaded when the journal head has reaped past sequence number 1.
651 	 */
652 	update_slab_summary_entry(slab, &journal->slab_summary_waiter,
653 				  journal->summarized % journal->size,
654 				  (journal->head > 1), false, free_block_count);
655 }
656 
657 /**
658  * reopen_slab_journal() - Reopen a slab's journal by emptying it and then adding pending entries.
659  * @slab: The slab to reopen.
660  */
reopen_slab_journal(struct vdo_slab * slab)661 static void reopen_slab_journal(struct vdo_slab *slab)
662 {
663 	struct slab_journal *journal = &slab->journal;
664 	sequence_number_t block;
665 
666 	VDO_ASSERT_LOG_ONLY(journal->tail_header.entry_count == 0,
667 			    "vdo_slab journal's active block empty before reopening");
668 	journal->head = journal->tail;
669 	initialize_journal_state(journal);
670 
671 	/* Ensure no locks are spuriously held on an empty journal. */
672 	for (block = 1; block <= journal->size; block++) {
673 		VDO_ASSERT_LOG_ONLY((get_lock(journal, block)->count == 0),
674 				    "Scrubbed journal's block %llu is not locked",
675 				    (unsigned long long) block);
676 	}
677 
678 	add_entries(journal);
679 }
680 
get_committing_sequence_number(const struct pooled_vio * vio)681 static sequence_number_t get_committing_sequence_number(const struct pooled_vio *vio)
682 {
683 	const struct packed_slab_journal_block *block =
684 		(const struct packed_slab_journal_block *) vio->vio.data;
685 
686 	return __le64_to_cpu(block->header.sequence_number);
687 }
688 
689 /**
690  * complete_write() - Handle post-commit processing.
691  * @completion: The write vio as a completion.
692  *
693  * This is the callback registered by write_slab_journal_block().
694  */
complete_write(struct vdo_completion * completion)695 static void complete_write(struct vdo_completion *completion)
696 {
697 	int result = completion->result;
698 	struct pooled_vio *pooled = vio_as_pooled_vio(as_vio(completion));
699 	struct slab_journal *journal = completion->parent;
700 	sequence_number_t committed = get_committing_sequence_number(pooled);
701 
702 	list_del_init(&pooled->list_entry);
703 	return_vio_to_pool(pooled);
704 
705 	if (result != VDO_SUCCESS) {
706 		vio_record_metadata_io_error(as_vio(completion));
707 		vdo_log_error_strerror(result, "cannot write slab journal block %llu",
708 				       (unsigned long long) committed);
709 		vdo_enter_read_only_mode(journal->slab->allocator->depot->vdo, result);
710 		check_if_slab_drained(journal->slab);
711 		return;
712 	}
713 
714 	WRITE_ONCE(journal->events->blocks_written, journal->events->blocks_written + 1);
715 
716 	if (list_empty(&journal->uncommitted_blocks)) {
717 		/* If no blocks are outstanding, then the commit point is at the tail. */
718 		journal->next_commit = journal->tail;
719 	} else {
720 		/* The commit point is always the beginning of the oldest incomplete block. */
721 		pooled = container_of(journal->uncommitted_blocks.next,
722 				      struct pooled_vio, list_entry);
723 		journal->next_commit = get_committing_sequence_number(pooled);
724 	}
725 
726 	update_tail_block_location(journal);
727 }
728 
write_slab_journal_endio(struct bio * bio)729 static void write_slab_journal_endio(struct bio *bio)
730 {
731 	struct vio *vio = bio->bi_private;
732 	struct slab_journal *journal = vio->completion.parent;
733 
734 	continue_vio_after_io(vio, complete_write, journal->slab->allocator->thread_id);
735 }
736 
737 /**
738  * write_slab_journal_block() - Write a slab journal block.
739  * @waiter: The vio pool waiter which was just notified.
740  * @context: The vio pool entry for the write.
741  *
742  * Callback from acquire_vio_from_pool() registered in commit_tail().
743  */
write_slab_journal_block(struct vdo_waiter * waiter,void * context)744 static void write_slab_journal_block(struct vdo_waiter *waiter, void *context)
745 {
746 	struct pooled_vio *pooled = context;
747 	struct vio *vio = &pooled->vio;
748 	struct slab_journal *journal =
749 		container_of(waiter, struct slab_journal, resource_waiter);
750 	struct slab_journal_block_header *header = &journal->tail_header;
751 	int unused_entries = journal->entries_per_block - header->entry_count;
752 	physical_block_number_t block_number;
753 	const struct admin_state_code *operation;
754 
755 	header->head = journal->head;
756 	list_add_tail(&pooled->list_entry, &journal->uncommitted_blocks);
757 	vdo_pack_slab_journal_block_header(header, &journal->block->header);
758 
759 	/* Copy the tail block into the vio. */
760 	memcpy(pooled->vio.data, journal->block, VDO_BLOCK_SIZE);
761 
762 	VDO_ASSERT_LOG_ONLY(unused_entries >= 0, "vdo_slab journal block is not overfull");
763 	if (unused_entries > 0) {
764 		/*
765 		 * Release the per-entry locks for any unused entries in the block we are about to
766 		 * write.
767 		 */
768 		adjust_slab_journal_block_reference(journal, header->sequence_number,
769 						    -unused_entries);
770 		journal->partial_write_in_progress = !block_is_full(journal);
771 	}
772 
773 	block_number = journal->slab->journal_origin +
774 		(header->sequence_number % journal->size);
775 	vio->completion.parent = journal;
776 
777 	/*
778 	 * This block won't be read in recovery until the slab summary is updated to refer to it.
779 	 * The slab summary update does a flush which is sufficient to protect us from corruption
780 	 * due to out of order slab journal, reference block, or block map writes.
781 	 */
782 	vdo_submit_metadata_vio(vdo_forget(vio), block_number, write_slab_journal_endio,
783 				complete_write, REQ_OP_WRITE);
784 
785 	/* Since the write is submitted, the tail block structure can be reused. */
786 	journal->tail++;
787 	initialize_tail_block(journal);
788 	journal->waiting_to_commit = false;
789 
790 	operation = vdo_get_admin_state_code(&journal->slab->state);
791 	if (operation == VDO_ADMIN_STATE_WAITING_FOR_RECOVERY) {
792 		vdo_finish_operation(&journal->slab->state,
793 				     (vdo_is_read_only(journal->slab->allocator->depot->vdo) ?
794 				      VDO_READ_ONLY : VDO_SUCCESS));
795 		return;
796 	}
797 
798 	add_entries(journal);
799 }
800 
801 /**
802  * commit_tail() - Commit the tail block of the slab journal.
803  * @journal: The journal whose tail block should be committed.
804  */
commit_tail(struct slab_journal * journal)805 static void commit_tail(struct slab_journal *journal)
806 {
807 	if ((journal->tail_header.entry_count == 0) && must_make_entries_to_flush(journal)) {
808 		/*
809 		 * There are no entries at the moment, but there are some waiters, so defer
810 		 * initiating the flush until those entries are ready to write.
811 		 */
812 		return;
813 	}
814 
815 	if (vdo_is_read_only(journal->slab->allocator->depot->vdo) ||
816 	    journal->waiting_to_commit ||
817 	    (journal->tail_header.entry_count == 0)) {
818 		/*
819 		 * There is nothing to do since the tail block is empty, or writing, or the journal
820 		 * is in read-only mode.
821 		 */
822 		return;
823 	}
824 
825 	/*
826 	 * Since we are about to commit the tail block, this journal no longer needs to be on the
827 	 * list of journals which the recovery journal might ask to commit.
828 	 */
829 	mark_slab_journal_clean(journal);
830 
831 	journal->waiting_to_commit = true;
832 
833 	journal->resource_waiter.callback = write_slab_journal_block;
834 	acquire_vio_from_pool(journal->slab->allocator->vio_pool,
835 			      &journal->resource_waiter);
836 }
837 
838 /**
839  * encode_slab_journal_entry() - Encode a slab journal entry.
840  * @tail_header: The unpacked header for the block.
841  * @payload: The journal block payload to hold the entry.
842  * @sbn: The slab block number of the entry to encode.
843  * @operation: The type of the entry.
844  * @increment: True if this is an increment.
845  */
encode_slab_journal_entry(struct slab_journal_block_header * tail_header,slab_journal_payload * payload,slab_block_number sbn,enum journal_operation operation,bool increment)846 static void encode_slab_journal_entry(struct slab_journal_block_header *tail_header,
847 				      slab_journal_payload *payload,
848 				      slab_block_number sbn,
849 				      enum journal_operation operation,
850 				      bool increment)
851 {
852 	journal_entry_count_t entry_number = tail_header->entry_count++;
853 
854 	if (operation == VDO_JOURNAL_BLOCK_MAP_REMAPPING) {
855 		if (!tail_header->has_block_map_increments) {
856 			memset(payload->full_entries.entry_types, 0,
857 			       VDO_SLAB_JOURNAL_ENTRY_TYPES_SIZE);
858 			tail_header->has_block_map_increments = true;
859 		}
860 
861 		payload->full_entries.entry_types[entry_number / 8] |=
862 			((u8)1 << (entry_number % 8));
863 	}
864 
865 	vdo_pack_slab_journal_entry(&payload->entries[entry_number], sbn, increment);
866 }
867 
868 /**
869  * expand_journal_point() - Convert a recovery journal journal_point which refers to both an
870  *                          increment and a decrement to a single point which refers to one or the
871  *                          other.
872  * @recovery_point: The journal point to convert.
873  * @increment: Whether the current entry is an increment.
874  *
875  * Return: The expanded journal point
876  *
877  * Because each data_vio has but a single recovery journal point, but may need to make both
878  * increment and decrement entries in the same slab journal. In order to distinguish the two
879  * entries, the entry count of the expanded journal point is twice the actual recovery journal
880  * entry count for increments, and one more than that for decrements.
881  */
expand_journal_point(struct journal_point recovery_point,bool increment)882 static struct journal_point expand_journal_point(struct journal_point recovery_point,
883 						 bool increment)
884 {
885 	recovery_point.entry_count *= 2;
886 	if (!increment)
887 		recovery_point.entry_count++;
888 
889 	return recovery_point;
890 }
891 
892 /**
893  * add_entry() - Actually add an entry to the slab journal, potentially firing off a write if a
894  *               block becomes full.
895  * @journal: The slab journal to append to.
896  * @pbn: The pbn being adjusted.
897  * @operation: The type of entry to make.
898  * @increment: True if this is an increment.
899  * @recovery_point: The expanded recovery point.
900  *
901  * This function is synchronous.
902  */
add_entry(struct slab_journal * journal,physical_block_number_t pbn,enum journal_operation operation,bool increment,struct journal_point recovery_point)903 static void add_entry(struct slab_journal *journal, physical_block_number_t pbn,
904 		      enum journal_operation operation, bool increment,
905 		      struct journal_point recovery_point)
906 {
907 	struct packed_slab_journal_block *block = journal->block;
908 	int result;
909 
910 	result = VDO_ASSERT(vdo_before_journal_point(&journal->tail_header.recovery_point,
911 						     &recovery_point),
912 			    "recovery journal point is monotonically increasing, recovery point: %llu.%u, block recovery point: %llu.%u",
913 			    (unsigned long long) recovery_point.sequence_number,
914 			    recovery_point.entry_count,
915 			    (unsigned long long) journal->tail_header.recovery_point.sequence_number,
916 			    journal->tail_header.recovery_point.entry_count);
917 	if (result != VDO_SUCCESS) {
918 		vdo_enter_read_only_mode(journal->slab->allocator->depot->vdo, result);
919 		return;
920 	}
921 
922 	if (operation == VDO_JOURNAL_BLOCK_MAP_REMAPPING) {
923 		result = VDO_ASSERT((journal->tail_header.entry_count <
924 				     journal->full_entries_per_block),
925 				    "block has room for full entries");
926 		if (result != VDO_SUCCESS) {
927 			vdo_enter_read_only_mode(journal->slab->allocator->depot->vdo,
928 						 result);
929 			return;
930 		}
931 	}
932 
933 	encode_slab_journal_entry(&journal->tail_header, &block->payload,
934 				  pbn - journal->slab->start, operation, increment);
935 	journal->tail_header.recovery_point = recovery_point;
936 	if (block_is_full(journal))
937 		commit_tail(journal);
938 }
939 
journal_length(const struct slab_journal * journal)940 static inline block_count_t journal_length(const struct slab_journal *journal)
941 {
942 	return journal->tail - journal->head;
943 }
944 
945 /**
946  * vdo_attempt_replay_into_slab() - Replay a recovery journal entry into a slab's journal.
947  * @slab: The slab to play into.
948  * @pbn: The PBN for the entry.
949  * @operation: The type of entry to add.
950  * @increment: True if this entry is an increment.
951  * @recovery_point: The recovery journal point corresponding to this entry.
952  * @parent: The completion to notify when there is space to add the entry if the entry could not be
953  *          added immediately.
954  *
955  * Return: True if the entry was added immediately.
956  */
vdo_attempt_replay_into_slab(struct vdo_slab * slab,physical_block_number_t pbn,enum journal_operation operation,bool increment,struct journal_point * recovery_point,struct vdo_completion * parent)957 bool vdo_attempt_replay_into_slab(struct vdo_slab *slab, physical_block_number_t pbn,
958 				  enum journal_operation operation, bool increment,
959 				  struct journal_point *recovery_point,
960 				  struct vdo_completion *parent)
961 {
962 	struct slab_journal *journal = &slab->journal;
963 	struct slab_journal_block_header *header = &journal->tail_header;
964 	struct journal_point expanded = expand_journal_point(*recovery_point, increment);
965 
966 	/* Only accept entries after the current recovery point. */
967 	if (!vdo_before_journal_point(&journal->tail_header.recovery_point, &expanded))
968 		return true;
969 
970 	if ((header->entry_count >= journal->full_entries_per_block) &&
971 	    (header->has_block_map_increments || (operation == VDO_JOURNAL_BLOCK_MAP_REMAPPING))) {
972 		/*
973 		 * The tail block does not have room for the entry we are attempting to add so
974 		 * commit the tail block now.
975 		 */
976 		commit_tail(journal);
977 	}
978 
979 	if (journal->waiting_to_commit) {
980 		vdo_start_operation_with_waiter(&journal->slab->state,
981 						VDO_ADMIN_STATE_WAITING_FOR_RECOVERY,
982 						parent, NULL);
983 		return false;
984 	}
985 
986 	if (journal_length(journal) >= journal->size) {
987 		/*
988 		 * We must have reaped the current head before the crash, since the blocked
989 		 * threshold keeps us from having more entries than fit in a slab journal; hence we
990 		 * can just advance the head (and unreapable block), as needed.
991 		 */
992 		journal->head++;
993 		journal->unreapable++;
994 	}
995 
996 	if (journal->slab->status == VDO_SLAB_REBUILT)
997 		journal->slab->status = VDO_SLAB_REPLAYING;
998 
999 	add_entry(journal, pbn, operation, increment, expanded);
1000 	return true;
1001 }
1002 
1003 /**
1004  * requires_reaping() - Check whether the journal must be reaped before adding new entries.
1005  * @journal: The journal to check.
1006  *
1007  * Return: True if the journal must be reaped.
1008  */
requires_reaping(const struct slab_journal * journal)1009 static bool requires_reaping(const struct slab_journal *journal)
1010 {
1011 	return (journal_length(journal) >= journal->blocking_threshold);
1012 }
1013 
1014 /** finish_summary_update() - A waiter callback that resets the writing state of a slab. */
finish_summary_update(struct vdo_waiter * waiter,void * context)1015 static void finish_summary_update(struct vdo_waiter *waiter, void *context)
1016 {
1017 	struct vdo_slab *slab = container_of(waiter, struct vdo_slab, summary_waiter);
1018 	int result = *((int *) context);
1019 
1020 	slab->active_count--;
1021 
1022 	if ((result != VDO_SUCCESS) && (result != VDO_READ_ONLY)) {
1023 		vdo_log_error_strerror(result, "failed to update slab summary");
1024 		vdo_enter_read_only_mode(slab->allocator->depot->vdo, result);
1025 	}
1026 
1027 	check_if_slab_drained(slab);
1028 }
1029 
1030 static void write_reference_block(struct vdo_waiter *waiter, void *context);
1031 
1032 /**
1033  * launch_reference_block_write() - Launch the write of a dirty reference block by first acquiring
1034  *                                  a VIO for it from the pool.
1035  * @waiter: The waiter of the block which is starting to write.
1036  * @context: The parent slab of the block.
1037  *
1038  * This can be asynchronous since the writer will have to wait if all VIOs in the pool are
1039  * currently in use.
1040  */
launch_reference_block_write(struct vdo_waiter * waiter,void * context)1041 static void launch_reference_block_write(struct vdo_waiter *waiter, void *context)
1042 {
1043 	struct vdo_slab *slab = context;
1044 
1045 	if (vdo_is_read_only(slab->allocator->depot->vdo))
1046 		return;
1047 
1048 	slab->active_count++;
1049 	container_of(waiter, struct reference_block, waiter)->is_writing = true;
1050 	waiter->callback = write_reference_block;
1051 	acquire_vio_from_pool(slab->allocator->vio_pool, waiter);
1052 }
1053 
save_dirty_reference_blocks(struct vdo_slab * slab)1054 static void save_dirty_reference_blocks(struct vdo_slab *slab)
1055 {
1056 	vdo_waitq_notify_all_waiters(&slab->dirty_blocks,
1057 				     launch_reference_block_write, slab);
1058 	check_if_slab_drained(slab);
1059 }
1060 
1061 /**
1062  * finish_reference_block_write() - After a reference block has written, clean it, release its
1063  *                                  locks, and return its VIO to the pool.
1064  * @completion: The VIO that just finished writing.
1065  */
finish_reference_block_write(struct vdo_completion * completion)1066 static void finish_reference_block_write(struct vdo_completion *completion)
1067 {
1068 	struct vio *vio = as_vio(completion);
1069 	struct pooled_vio *pooled = vio_as_pooled_vio(vio);
1070 	struct reference_block *block = completion->parent;
1071 	struct vdo_slab *slab = block->slab;
1072 	tail_block_offset_t offset;
1073 
1074 	slab->active_count--;
1075 
1076 	/* Release the slab journal lock. */
1077 	adjust_slab_journal_block_reference(&slab->journal,
1078 					    block->slab_journal_lock_to_release, -1);
1079 	return_vio_to_pool(pooled);
1080 
1081 	/*
1082 	 * We can't clear the is_writing flag earlier as releasing the slab journal lock may cause
1083 	 * us to be dirtied again, but we don't want to double enqueue.
1084 	 */
1085 	block->is_writing = false;
1086 
1087 	if (vdo_is_read_only(completion->vdo)) {
1088 		check_if_slab_drained(slab);
1089 		return;
1090 	}
1091 
1092 	/* Re-queue the block if it was re-dirtied while it was writing. */
1093 	if (block->is_dirty) {
1094 		vdo_waitq_enqueue_waiter(&block->slab->dirty_blocks, &block->waiter);
1095 		if (vdo_is_state_draining(&slab->state)) {
1096 			/* We must be saving, and this block will otherwise not be relaunched. */
1097 			save_dirty_reference_blocks(slab);
1098 		}
1099 
1100 		return;
1101 	}
1102 
1103 	/*
1104 	 * Mark the slab as clean in the slab summary if there are no dirty or writing blocks
1105 	 * and no summary update in progress.
1106 	 */
1107 	if ((slab->active_count > 0) || vdo_waitq_has_waiters(&slab->dirty_blocks)) {
1108 		check_if_slab_drained(slab);
1109 		return;
1110 	}
1111 
1112 	offset = slab->allocator->summary_entries[slab->slab_number].tail_block_offset;
1113 	slab->active_count++;
1114 	slab->summary_waiter.callback = finish_summary_update;
1115 	update_slab_summary_entry(slab, &slab->summary_waiter, offset,
1116 				  true, true, slab->free_blocks);
1117 }
1118 
1119 /**
1120  * get_reference_counters_for_block() - Find the reference counters for a given block.
1121  * @block: The reference_block in question.
1122  *
1123  * Return: A pointer to the reference counters for this block.
1124  */
get_reference_counters_for_block(struct reference_block * block)1125 static vdo_refcount_t * __must_check get_reference_counters_for_block(struct reference_block *block)
1126 {
1127 	size_t block_index = block - block->slab->reference_blocks;
1128 
1129 	return &block->slab->counters[block_index * COUNTS_PER_BLOCK];
1130 }
1131 
1132 /**
1133  * pack_reference_block() - Copy data from a reference block to a buffer ready to be written out.
1134  * @block: The block to copy.
1135  * @buffer: The char buffer to fill with the packed block.
1136  */
pack_reference_block(struct reference_block * block,void * buffer)1137 static void pack_reference_block(struct reference_block *block, void *buffer)
1138 {
1139 	struct packed_reference_block *packed = buffer;
1140 	vdo_refcount_t *counters = get_reference_counters_for_block(block);
1141 	sector_count_t i;
1142 	struct packed_journal_point commit_point;
1143 
1144 	vdo_pack_journal_point(&block->slab->slab_journal_point, &commit_point);
1145 
1146 	for (i = 0; i < VDO_SECTORS_PER_BLOCK; i++) {
1147 		packed->sectors[i].commit_point = commit_point;
1148 		memcpy(packed->sectors[i].counts, counters + (i * COUNTS_PER_SECTOR),
1149 		       (sizeof(vdo_refcount_t) * COUNTS_PER_SECTOR));
1150 	}
1151 }
1152 
write_reference_block_endio(struct bio * bio)1153 static void write_reference_block_endio(struct bio *bio)
1154 {
1155 	struct vio *vio = bio->bi_private;
1156 	struct reference_block *block = vio->completion.parent;
1157 	thread_id_t thread_id = block->slab->allocator->thread_id;
1158 
1159 	continue_vio_after_io(vio, finish_reference_block_write, thread_id);
1160 }
1161 
1162 /**
1163  * handle_io_error() - Handle an I/O error reading or writing a reference count block.
1164  * @completion: The VIO doing the I/O as a completion.
1165  */
handle_io_error(struct vdo_completion * completion)1166 static void handle_io_error(struct vdo_completion *completion)
1167 {
1168 	int result = completion->result;
1169 	struct vio *vio = as_vio(completion);
1170 	struct vdo_slab *slab = ((struct reference_block *) completion->parent)->slab;
1171 
1172 	vio_record_metadata_io_error(vio);
1173 	return_vio_to_pool(vio_as_pooled_vio(vio));
1174 	slab->active_count -= vio->io_size / VDO_BLOCK_SIZE;
1175 	vdo_enter_read_only_mode(slab->allocator->depot->vdo, result);
1176 	check_if_slab_drained(slab);
1177 }
1178 
1179 /**
1180  * write_reference_block() - After a dirty block waiter has gotten a VIO from the VIO pool, copy
1181  *                           its counters and associated data into the VIO, and launch the write.
1182  * @waiter: The waiter of the dirty block.
1183  * @context: The VIO returned by the pool.
1184  */
write_reference_block(struct vdo_waiter * waiter,void * context)1185 static void write_reference_block(struct vdo_waiter *waiter, void *context)
1186 {
1187 	size_t block_offset;
1188 	physical_block_number_t pbn;
1189 	struct pooled_vio *pooled = context;
1190 	struct vdo_completion *completion = &pooled->vio.completion;
1191 	struct reference_block *block = container_of(waiter, struct reference_block,
1192 						     waiter);
1193 
1194 	pack_reference_block(block, pooled->vio.data);
1195 	block_offset = (block - block->slab->reference_blocks);
1196 	pbn = (block->slab->ref_counts_origin + block_offset);
1197 	block->slab_journal_lock_to_release = block->slab_journal_lock;
1198 	completion->parent = block;
1199 
1200 	/*
1201 	 * Mark the block as clean, since we won't be committing any updates that happen after this
1202 	 * moment. As long as VIO order is preserved, two VIOs updating this block at once will not
1203 	 * cause complications.
1204 	 */
1205 	block->is_dirty = false;
1206 
1207 	/*
1208 	 * Flush before writing to ensure that the recovery journal and slab journal entries which
1209 	 * cover this reference update are stable. This prevents data corruption that can be caused
1210 	 * by out of order writes.
1211 	 */
1212 	WRITE_ONCE(block->slab->allocator->ref_counts_statistics.blocks_written,
1213 		   block->slab->allocator->ref_counts_statistics.blocks_written + 1);
1214 
1215 	completion->callback_thread_id = ((struct block_allocator *) pooled->context)->thread_id;
1216 	vdo_submit_metadata_vio(&pooled->vio, pbn, write_reference_block_endio,
1217 				handle_io_error, REQ_OP_WRITE | REQ_PREFLUSH);
1218 }
1219 
reclaim_journal_space(struct slab_journal * journal)1220 static void reclaim_journal_space(struct slab_journal *journal)
1221 {
1222 	block_count_t length = journal_length(journal);
1223 	struct vdo_slab *slab = journal->slab;
1224 	block_count_t write_count = vdo_waitq_num_waiters(&slab->dirty_blocks);
1225 	block_count_t written;
1226 
1227 	if ((length < journal->flushing_threshold) || (write_count == 0))
1228 		return;
1229 
1230 	/* The slab journal is over the first threshold, schedule some reference block writes. */
1231 	WRITE_ONCE(journal->events->flush_count, journal->events->flush_count + 1);
1232 	if (length < journal->flushing_deadline) {
1233 		/* Schedule more writes the closer to the deadline we get. */
1234 		write_count /= journal->flushing_deadline - length + 1;
1235 		write_count = max_t(block_count_t, write_count, 1);
1236 	}
1237 
1238 	for (written = 0; written < write_count; written++) {
1239 		vdo_waitq_notify_next_waiter(&slab->dirty_blocks,
1240 					     launch_reference_block_write, slab);
1241 	}
1242 }
1243 
1244 /**
1245  * reference_count_to_status() - Convert a reference count to a reference status.
1246  * @count: The count to convert.
1247  *
1248  * Return: The appropriate reference status.
1249  */
reference_count_to_status(vdo_refcount_t count)1250 static enum reference_status __must_check reference_count_to_status(vdo_refcount_t count)
1251 {
1252 	if (count == EMPTY_REFERENCE_COUNT)
1253 		return RS_FREE;
1254 	else if (count == 1)
1255 		return RS_SINGLE;
1256 	else if (count == PROVISIONAL_REFERENCE_COUNT)
1257 		return RS_PROVISIONAL;
1258 	else
1259 		return RS_SHARED;
1260 }
1261 
1262 /**
1263  * dirty_block() - Mark a reference count block as dirty, potentially adding it to the dirty queue
1264  *                 if it wasn't already dirty.
1265  * @block: The reference block to mark as dirty.
1266  */
dirty_block(struct reference_block * block)1267 static void dirty_block(struct reference_block *block)
1268 {
1269 	if (block->is_dirty)
1270 		return;
1271 
1272 	block->is_dirty = true;
1273 	if (!block->is_writing)
1274 		vdo_waitq_enqueue_waiter(&block->slab->dirty_blocks, &block->waiter);
1275 }
1276 
1277 /**
1278  * get_reference_block() - Get the reference block that covers the given block index.
1279  * @slab: The slab containing the references.
1280  * @index: The index of the physical block.
1281  */
get_reference_block(struct vdo_slab * slab,slab_block_number index)1282 static struct reference_block * __must_check get_reference_block(struct vdo_slab *slab,
1283 								 slab_block_number index)
1284 {
1285 	return &slab->reference_blocks[index / COUNTS_PER_BLOCK];
1286 }
1287 
1288 /**
1289  * slab_block_number_from_pbn() - Determine the index within the slab of a particular physical
1290  *                                block number.
1291  * @slab: The slab.
1292  * @pbn: The physical block number.
1293  * @slab_block_number_ptr: A pointer to the slab block number.
1294  *
1295  * Return: VDO_SUCCESS or an error code.
1296  */
slab_block_number_from_pbn(struct vdo_slab * slab,physical_block_number_t pbn,slab_block_number * slab_block_number_ptr)1297 static int __must_check slab_block_number_from_pbn(struct vdo_slab *slab,
1298 						   physical_block_number_t pbn,
1299 						   slab_block_number *slab_block_number_ptr)
1300 {
1301 	u64 slab_block_number;
1302 
1303 	if (pbn < slab->start)
1304 		return VDO_OUT_OF_RANGE;
1305 
1306 	slab_block_number = pbn - slab->start;
1307 	if (slab_block_number >= slab->allocator->depot->slab_config.data_blocks)
1308 		return VDO_OUT_OF_RANGE;
1309 
1310 	*slab_block_number_ptr = slab_block_number;
1311 	return VDO_SUCCESS;
1312 }
1313 
1314 /**
1315  * get_reference_counter() - Get the reference counter that covers the given physical block number.
1316  * @slab: The slab to query.
1317  * @pbn: The physical block number.
1318  * @counter_ptr: A pointer to the reference counter.
1319  */
get_reference_counter(struct vdo_slab * slab,physical_block_number_t pbn,vdo_refcount_t ** counter_ptr)1320 static int __must_check get_reference_counter(struct vdo_slab *slab,
1321 					      physical_block_number_t pbn,
1322 					      vdo_refcount_t **counter_ptr)
1323 {
1324 	slab_block_number index;
1325 	int result = slab_block_number_from_pbn(slab, pbn, &index);
1326 
1327 	if (result != VDO_SUCCESS)
1328 		return result;
1329 
1330 	*counter_ptr = &slab->counters[index];
1331 
1332 	return VDO_SUCCESS;
1333 }
1334 
calculate_slab_priority(struct vdo_slab * slab)1335 static unsigned int calculate_slab_priority(struct vdo_slab *slab)
1336 {
1337 	block_count_t free_blocks = slab->free_blocks;
1338 	unsigned int unopened_slab_priority = slab->allocator->unopened_slab_priority;
1339 	unsigned int priority;
1340 
1341 	/*
1342 	 * Wholly full slabs must be the only ones with lowest priority, 0.
1343 	 *
1344 	 * Slabs that have never been opened (empty, newly initialized, and never been written to)
1345 	 * have lower priority than previously opened slabs that have a significant number of free
1346 	 * blocks. This ranking causes VDO to avoid writing physical blocks for the first time
1347 	 * unless there are very few free blocks that have been previously written to.
1348 	 *
1349 	 * Since VDO doesn't discard blocks currently, reusing previously written blocks makes VDO
1350 	 * a better client of any underlying storage that is thinly-provisioned (though discarding
1351 	 * would be better).
1352 	 *
1353 	 * For all other slabs, the priority is derived from the logarithm of the number of free
1354 	 * blocks. Slabs with the same order of magnitude of free blocks have the same priority.
1355 	 * With 2^23 blocks, the priority will range from 1 to 25. The reserved
1356 	 * unopened_slab_priority divides the range and is skipped by the logarithmic mapping.
1357 	 */
1358 
1359 	if (free_blocks == 0)
1360 		return 0;
1361 
1362 	if (is_slab_journal_blank(slab))
1363 		return unopened_slab_priority;
1364 
1365 	priority = (1 + ilog2(free_blocks));
1366 	return ((priority < unopened_slab_priority) ? priority : priority + 1);
1367 }
1368 
1369 /*
1370  * Slabs are essentially prioritized by an approximation of the number of free blocks in the slab
1371  * so slabs with lots of free blocks will be opened for allocation before slabs that have few free
1372  * blocks.
1373  */
prioritize_slab(struct vdo_slab * slab)1374 static void prioritize_slab(struct vdo_slab *slab)
1375 {
1376 	VDO_ASSERT_LOG_ONLY(list_empty(&slab->allocq_entry),
1377 			    "a slab must not already be on a list when prioritizing");
1378 	slab->priority = calculate_slab_priority(slab);
1379 	vdo_priority_table_enqueue(slab->allocator->prioritized_slabs,
1380 				   slab->priority, &slab->allocq_entry);
1381 }
1382 
1383 /**
1384  * adjust_free_block_count() - Adjust the free block count and (if needed) reprioritize the slab.
1385  * @slab: The slab.
1386  * @incremented: True if the free block count went up.
1387  */
adjust_free_block_count(struct vdo_slab * slab,bool incremented)1388 static void adjust_free_block_count(struct vdo_slab *slab, bool incremented)
1389 {
1390 	struct block_allocator *allocator = slab->allocator;
1391 
1392 	WRITE_ONCE(allocator->allocated_blocks,
1393 		   allocator->allocated_blocks + (incremented ? -1 : 1));
1394 
1395 	/* The open slab doesn't need to be reprioritized until it is closed. */
1396 	if (slab == allocator->open_slab)
1397 		return;
1398 
1399 	/* Don't bother adjusting the priority table if unneeded. */
1400 	if (slab->priority == calculate_slab_priority(slab))
1401 		return;
1402 
1403 	/*
1404 	 * Reprioritize the slab to reflect the new free block count by removing it from the table
1405 	 * and re-enqueuing it with the new priority.
1406 	 */
1407 	vdo_priority_table_remove(allocator->prioritized_slabs, &slab->allocq_entry);
1408 	prioritize_slab(slab);
1409 }
1410 
1411 /**
1412  * increment_for_data() - Increment the reference count for a data block.
1413  * @slab: The slab which owns the block.
1414  * @block: The reference block which contains the block being updated.
1415  * @block_number: The block to update.
1416  * @old_status: The reference status of the data block before this increment.
1417  * @lock: The pbn_lock associated with this increment (may be NULL).
1418  * @counter_ptr: A pointer to the count for the data block (in, out).
1419  * @adjust_block_count: Whether to update the allocator's free block count.
1420  *
1421  * Return: VDO_SUCCESS or an error.
1422  */
increment_for_data(struct vdo_slab * slab,struct reference_block * block,slab_block_number block_number,enum reference_status old_status,struct pbn_lock * lock,vdo_refcount_t * counter_ptr,bool adjust_block_count)1423 static int increment_for_data(struct vdo_slab *slab, struct reference_block *block,
1424 			      slab_block_number block_number,
1425 			      enum reference_status old_status,
1426 			      struct pbn_lock *lock, vdo_refcount_t *counter_ptr,
1427 			      bool adjust_block_count)
1428 {
1429 	switch (old_status) {
1430 	case RS_FREE:
1431 		*counter_ptr = 1;
1432 		block->allocated_count++;
1433 		slab->free_blocks--;
1434 		if (adjust_block_count)
1435 			adjust_free_block_count(slab, false);
1436 
1437 		break;
1438 
1439 	case RS_PROVISIONAL:
1440 		*counter_ptr = 1;
1441 		break;
1442 
1443 	default:
1444 		/* Single or shared */
1445 		if (*counter_ptr >= MAXIMUM_REFERENCE_COUNT) {
1446 			return vdo_log_error_strerror(VDO_REF_COUNT_INVALID,
1447 						      "Incrementing a block already having 254 references (slab %u, offset %u)",
1448 						      slab->slab_number, block_number);
1449 		}
1450 		(*counter_ptr)++;
1451 	}
1452 
1453 	if (lock != NULL)
1454 		vdo_unassign_pbn_lock_provisional_reference(lock);
1455 	return VDO_SUCCESS;
1456 }
1457 
1458 /**
1459  * decrement_for_data() - Decrement the reference count for a data block.
1460  * @slab: The slab which owns the block.
1461  * @block: The reference block which contains the block being updated.
1462  * @block_number: The block to update.
1463  * @old_status: The reference status of the data block before this decrement.
1464  * @updater: The reference updater doing this operation in case we need to look up the pbn lock.
1465  * @counter_ptr: A pointer to the count for the data block (in, out).
1466  * @adjust_block_count: Whether to update the allocator's free block count.
1467  *
1468  * Return: VDO_SUCCESS or an error.
1469  */
decrement_for_data(struct vdo_slab * slab,struct reference_block * block,slab_block_number block_number,enum reference_status old_status,struct reference_updater * updater,vdo_refcount_t * counter_ptr,bool adjust_block_count)1470 static int decrement_for_data(struct vdo_slab *slab, struct reference_block *block,
1471 			      slab_block_number block_number,
1472 			      enum reference_status old_status,
1473 			      struct reference_updater *updater,
1474 			      vdo_refcount_t *counter_ptr, bool adjust_block_count)
1475 {
1476 	switch (old_status) {
1477 	case RS_FREE:
1478 		return vdo_log_error_strerror(VDO_REF_COUNT_INVALID,
1479 					      "Decrementing free block at offset %u in slab %u",
1480 					      block_number, slab->slab_number);
1481 
1482 	case RS_PROVISIONAL:
1483 	case RS_SINGLE:
1484 		if (updater->zpbn.zone != NULL) {
1485 			struct pbn_lock *lock = vdo_get_physical_zone_pbn_lock(updater->zpbn.zone,
1486 									       updater->zpbn.pbn);
1487 
1488 			if (lock != NULL) {
1489 				/*
1490 				 * There is a read lock on this block, so the block must not become
1491 				 * unreferenced.
1492 				 */
1493 				*counter_ptr = PROVISIONAL_REFERENCE_COUNT;
1494 				vdo_assign_pbn_lock_provisional_reference(lock);
1495 				break;
1496 			}
1497 		}
1498 
1499 		*counter_ptr = EMPTY_REFERENCE_COUNT;
1500 		block->allocated_count--;
1501 		slab->free_blocks++;
1502 		if (adjust_block_count)
1503 			adjust_free_block_count(slab, true);
1504 
1505 		break;
1506 
1507 	default:
1508 		/* Shared */
1509 		(*counter_ptr)--;
1510 	}
1511 
1512 	return VDO_SUCCESS;
1513 }
1514 
1515 /**
1516  * increment_for_block_map() - Increment the reference count for a block map page.
1517  * @slab: The slab which owns the block.
1518  * @block: The reference block which contains the block being updated.
1519  * @block_number: The block to update.
1520  * @old_status: The reference status of the block before this increment.
1521  * @lock: The pbn_lock associated with this increment (may be NULL).
1522  * @normal_operation: Whether we are in normal operation vs. recovery or rebuild.
1523  * @counter_ptr: A pointer to the count for the block (in, out).
1524  * @adjust_block_count: Whether to update the allocator's free block count.
1525  *
1526  * All block map increments should be from provisional to MAXIMUM_REFERENCE_COUNT. Since block map
1527  * blocks never dedupe they should never be adjusted from any other state. The adjustment always
1528  * results in MAXIMUM_REFERENCE_COUNT as this value is used to prevent dedupe against block map
1529  * blocks.
1530  *
1531  * Return: VDO_SUCCESS or an error.
1532  */
increment_for_block_map(struct vdo_slab * slab,struct reference_block * block,slab_block_number block_number,enum reference_status old_status,struct pbn_lock * lock,bool normal_operation,vdo_refcount_t * counter_ptr,bool adjust_block_count)1533 static int increment_for_block_map(struct vdo_slab *slab, struct reference_block *block,
1534 				   slab_block_number block_number,
1535 				   enum reference_status old_status,
1536 				   struct pbn_lock *lock, bool normal_operation,
1537 				   vdo_refcount_t *counter_ptr, bool adjust_block_count)
1538 {
1539 	switch (old_status) {
1540 	case RS_FREE:
1541 		if (normal_operation) {
1542 			return vdo_log_error_strerror(VDO_REF_COUNT_INVALID,
1543 						      "Incrementing unallocated block map block (slab %u, offset %u)",
1544 						      slab->slab_number, block_number);
1545 		}
1546 
1547 		*counter_ptr = MAXIMUM_REFERENCE_COUNT;
1548 		block->allocated_count++;
1549 		slab->free_blocks--;
1550 		if (adjust_block_count)
1551 			adjust_free_block_count(slab, false);
1552 
1553 		return VDO_SUCCESS;
1554 
1555 	case RS_PROVISIONAL:
1556 		if (!normal_operation)
1557 			return vdo_log_error_strerror(VDO_REF_COUNT_INVALID,
1558 						      "Block map block had provisional reference during replay (slab %u, offset %u)",
1559 						      slab->slab_number, block_number);
1560 
1561 		*counter_ptr = MAXIMUM_REFERENCE_COUNT;
1562 		if (lock != NULL)
1563 			vdo_unassign_pbn_lock_provisional_reference(lock);
1564 		return VDO_SUCCESS;
1565 
1566 	default:
1567 		return vdo_log_error_strerror(VDO_REF_COUNT_INVALID,
1568 					      "Incrementing a block map block which is already referenced %u times (slab %u, offset %u)",
1569 					      *counter_ptr, slab->slab_number,
1570 					      block_number);
1571 	}
1572 }
1573 
is_valid_journal_point(const struct journal_point * point)1574 static bool __must_check is_valid_journal_point(const struct journal_point *point)
1575 {
1576 	return ((point != NULL) && (point->sequence_number > 0));
1577 }
1578 
1579 /**
1580  * update_reference_count() - Update the reference count of a block.
1581  * @slab: The slab which owns the block.
1582  * @block: The reference block which contains the block being updated.
1583  * @block_number: The block to update.
1584  * @slab_journal_point: The slab journal point at which this update is journaled.
1585  * @updater: The reference updater.
1586  * @normal_operation: Whether we are in normal operation vs. recovery or rebuild.
1587  * @adjust_block_count: Whether to update the slab's free block count.
1588  * @provisional_decrement_ptr: A pointer which will be set to true if this update was a decrement
1589  *                             of a provisional reference.
1590  *
1591  * Return: VDO_SUCCESS or an error.
1592  */
update_reference_count(struct vdo_slab * slab,struct reference_block * block,slab_block_number block_number,const struct journal_point * slab_journal_point,struct reference_updater * updater,bool normal_operation,bool adjust_block_count,bool * provisional_decrement_ptr)1593 static int update_reference_count(struct vdo_slab *slab, struct reference_block *block,
1594 				  slab_block_number block_number,
1595 				  const struct journal_point *slab_journal_point,
1596 				  struct reference_updater *updater,
1597 				  bool normal_operation, bool adjust_block_count,
1598 				  bool *provisional_decrement_ptr)
1599 {
1600 	vdo_refcount_t *counter_ptr = &slab->counters[block_number];
1601 	enum reference_status old_status = reference_count_to_status(*counter_ptr);
1602 	int result;
1603 
1604 	if (!updater->increment) {
1605 		result = decrement_for_data(slab, block, block_number, old_status,
1606 					    updater, counter_ptr, adjust_block_count);
1607 		if ((result == VDO_SUCCESS) && (old_status == RS_PROVISIONAL)) {
1608 			if (provisional_decrement_ptr != NULL)
1609 				*provisional_decrement_ptr = true;
1610 			return VDO_SUCCESS;
1611 		}
1612 	} else if (updater->operation == VDO_JOURNAL_DATA_REMAPPING) {
1613 		result = increment_for_data(slab, block, block_number, old_status,
1614 					    updater->lock, counter_ptr, adjust_block_count);
1615 	} else {
1616 		result = increment_for_block_map(slab, block, block_number, old_status,
1617 						 updater->lock, normal_operation,
1618 						 counter_ptr, adjust_block_count);
1619 	}
1620 
1621 	if (result != VDO_SUCCESS)
1622 		return result;
1623 
1624 	if (is_valid_journal_point(slab_journal_point))
1625 		slab->slab_journal_point = *slab_journal_point;
1626 
1627 	return VDO_SUCCESS;
1628 }
1629 
adjust_reference_count(struct vdo_slab * slab,struct reference_updater * updater,const struct journal_point * slab_journal_point)1630 static int __must_check adjust_reference_count(struct vdo_slab *slab,
1631 					       struct reference_updater *updater,
1632 					       const struct journal_point *slab_journal_point)
1633 {
1634 	slab_block_number block_number;
1635 	int result;
1636 	struct reference_block *block;
1637 	bool provisional_decrement = false;
1638 
1639 	if (!is_slab_open(slab))
1640 		return VDO_INVALID_ADMIN_STATE;
1641 
1642 	result = slab_block_number_from_pbn(slab, updater->zpbn.pbn, &block_number);
1643 	if (result != VDO_SUCCESS)
1644 		return result;
1645 
1646 	block = get_reference_block(slab, block_number);
1647 	result = update_reference_count(slab, block, block_number, slab_journal_point,
1648 					updater, NORMAL_OPERATION, true,
1649 					&provisional_decrement);
1650 	if ((result != VDO_SUCCESS) || provisional_decrement)
1651 		return result;
1652 
1653 	if (block->is_dirty && (block->slab_journal_lock > 0)) {
1654 		sequence_number_t entry_lock = slab_journal_point->sequence_number;
1655 		/*
1656 		 * This block is already dirty and a slab journal entry has been made for it since
1657 		 * the last time it was clean. We must release the per-entry slab journal lock for
1658 		 * the entry associated with the update we are now doing.
1659 		 */
1660 		result = VDO_ASSERT(is_valid_journal_point(slab_journal_point),
1661 				    "Reference count adjustments need slab journal points.");
1662 		if (result != VDO_SUCCESS)
1663 			return result;
1664 
1665 		adjust_slab_journal_block_reference(&slab->journal, entry_lock, -1);
1666 		return VDO_SUCCESS;
1667 	}
1668 
1669 	/*
1670 	 * This may be the first time we are applying an update for which there is a slab journal
1671 	 * entry to this block since the block was cleaned. Therefore, we convert the per-entry
1672 	 * slab journal lock to an uncommitted reference block lock, if there is a per-entry lock.
1673 	 */
1674 	if (is_valid_journal_point(slab_journal_point))
1675 		block->slab_journal_lock = slab_journal_point->sequence_number;
1676 	else
1677 		block->slab_journal_lock = 0;
1678 
1679 	dirty_block(block);
1680 	return VDO_SUCCESS;
1681 }
1682 
1683 /**
1684  * add_entry_from_waiter() - Add an entry to the slab journal.
1685  * @waiter: The vio which should make an entry now.
1686  * @context: The slab journal to make an entry in.
1687  *
1688  * This callback is invoked by add_entries() once it has determined that we are ready to make
1689  * another entry in the slab journal. Implements waiter_callback_fn.
1690  */
add_entry_from_waiter(struct vdo_waiter * waiter,void * context)1691 static void add_entry_from_waiter(struct vdo_waiter *waiter, void *context)
1692 {
1693 	int result;
1694 	struct reference_updater *updater =
1695 		container_of(waiter, struct reference_updater, waiter);
1696 	struct data_vio *data_vio = data_vio_from_reference_updater(updater);
1697 	struct slab_journal *journal = context;
1698 	struct slab_journal_block_header *header = &journal->tail_header;
1699 	struct journal_point slab_journal_point = {
1700 		.sequence_number = header->sequence_number,
1701 		.entry_count = header->entry_count,
1702 	};
1703 	sequence_number_t recovery_block = data_vio->recovery_journal_point.sequence_number;
1704 
1705 	if (header->entry_count == 0) {
1706 		/*
1707 		 * This is the first entry in the current tail block, so get a lock on the recovery
1708 		 * journal which we will hold until this tail block is committed.
1709 		 */
1710 		get_lock(journal, header->sequence_number)->recovery_start = recovery_block;
1711 		if (journal->recovery_journal != NULL) {
1712 			zone_count_t zone_number = journal->slab->allocator->zone_number;
1713 
1714 			vdo_acquire_recovery_journal_block_reference(journal->recovery_journal,
1715 								     recovery_block,
1716 								     VDO_ZONE_TYPE_PHYSICAL,
1717 								     zone_number);
1718 		}
1719 
1720 		mark_slab_journal_dirty(journal, recovery_block);
1721 		reclaim_journal_space(journal);
1722 	}
1723 
1724 	add_entry(journal, updater->zpbn.pbn, updater->operation, updater->increment,
1725 		  expand_journal_point(data_vio->recovery_journal_point,
1726 				       updater->increment));
1727 
1728 	if (journal->slab->status != VDO_SLAB_REBUILT) {
1729 		/*
1730 		 * If the slab is unrecovered, scrubbing will take care of the count since the
1731 		 * update is now recorded in the journal.
1732 		 */
1733 		adjust_slab_journal_block_reference(journal,
1734 						    slab_journal_point.sequence_number, -1);
1735 		result = VDO_SUCCESS;
1736 	} else {
1737 		/* Now that an entry has been made in the slab journal, update the counter. */
1738 		result = adjust_reference_count(journal->slab, updater,
1739 						&slab_journal_point);
1740 	}
1741 
1742 	if (updater->increment)
1743 		continue_data_vio_with_error(data_vio, result);
1744 	else
1745 		vdo_continue_completion(&data_vio->decrement_completion, result);
1746 }
1747 
1748 /**
1749  * is_next_entry_a_block_map_increment() - Check whether the next entry to be made is a block map
1750  *                                         increment.
1751  * @journal: The journal.
1752  *
1753  * Return: true if the first entry waiter's operation is a block map increment.
1754  */
is_next_entry_a_block_map_increment(struct slab_journal * journal)1755 static inline bool is_next_entry_a_block_map_increment(struct slab_journal *journal)
1756 {
1757 	struct vdo_waiter *waiter = vdo_waitq_get_first_waiter(&journal->entry_waiters);
1758 	struct reference_updater *updater =
1759 		container_of(waiter, struct reference_updater, waiter);
1760 
1761 	return (updater->operation == VDO_JOURNAL_BLOCK_MAP_REMAPPING);
1762 }
1763 
1764 /**
1765  * add_entries() - Add as many entries as possible from the queue of vios waiting to make entries.
1766  * @journal: The journal to which entries may be added.
1767  *
1768  * By processing the queue in order, we ensure that slab journal entries are made in the same order
1769  * as recovery journal entries for the same increment or decrement.
1770  */
add_entries(struct slab_journal * journal)1771 static void add_entries(struct slab_journal *journal)
1772 {
1773 	if (journal->adding_entries) {
1774 		/* Protect against re-entrancy. */
1775 		return;
1776 	}
1777 
1778 	journal->adding_entries = true;
1779 	while (vdo_waitq_has_waiters(&journal->entry_waiters)) {
1780 		struct slab_journal_block_header *header = &journal->tail_header;
1781 
1782 		if (journal->partial_write_in_progress ||
1783 		    (journal->slab->status == VDO_SLAB_REBUILDING)) {
1784 			/*
1785 			 * Don't add entries while rebuilding or while a partial write is
1786 			 * outstanding, as it could result in reference count corruption.
1787 			 */
1788 			break;
1789 		}
1790 
1791 		if (journal->waiting_to_commit) {
1792 			/*
1793 			 * If we are waiting for resources to write the tail block, and the tail
1794 			 * block is full, we can't make another entry.
1795 			 */
1796 			WRITE_ONCE(journal->events->tail_busy_count,
1797 				   journal->events->tail_busy_count + 1);
1798 			break;
1799 		} else if (is_next_entry_a_block_map_increment(journal) &&
1800 			   (header->entry_count >= journal->full_entries_per_block)) {
1801 			/*
1802 			 * The tail block does not have room for a block map increment, so commit
1803 			 * it now.
1804 			 */
1805 			commit_tail(journal);
1806 			if (journal->waiting_to_commit) {
1807 				WRITE_ONCE(journal->events->tail_busy_count,
1808 					   journal->events->tail_busy_count + 1);
1809 				break;
1810 			}
1811 		}
1812 
1813 		/* If the slab is over the blocking threshold, make the vio wait. */
1814 		if (requires_reaping(journal)) {
1815 			WRITE_ONCE(journal->events->blocked_count,
1816 				   journal->events->blocked_count + 1);
1817 			save_dirty_reference_blocks(journal->slab);
1818 			break;
1819 		}
1820 
1821 		if (header->entry_count == 0) {
1822 			struct journal_lock *lock =
1823 				get_lock(journal, header->sequence_number);
1824 
1825 			/*
1826 			 * Check if the on disk slab journal is full. Because of the blocking and
1827 			 * scrubbing thresholds, this should never happen.
1828 			 */
1829 			if (lock->count > 0) {
1830 				VDO_ASSERT_LOG_ONLY((journal->head + journal->size) == journal->tail,
1831 						    "New block has locks, but journal is not full");
1832 
1833 				/*
1834 				 * The blocking threshold must let the journal fill up if the new
1835 				 * block has locks; if the blocking threshold is smaller than the
1836 				 * journal size, the new block cannot possibly have locks already.
1837 				 */
1838 				VDO_ASSERT_LOG_ONLY((journal->blocking_threshold >= journal->size),
1839 						    "New block can have locks already iff blocking threshold is at the end of the journal");
1840 
1841 				WRITE_ONCE(journal->events->disk_full_count,
1842 					   journal->events->disk_full_count + 1);
1843 				save_dirty_reference_blocks(journal->slab);
1844 				break;
1845 			}
1846 
1847 			/*
1848 			 * Don't allow the new block to be reaped until all of the reference count
1849 			 * blocks are written and the journal block has been fully committed as
1850 			 * well.
1851 			 */
1852 			lock->count = journal->entries_per_block + 1;
1853 
1854 			if (header->sequence_number == 1) {
1855 				struct vdo_slab *slab = journal->slab;
1856 				block_count_t i;
1857 
1858 				/*
1859 				 * This is the first entry in this slab journal, ever. Dirty all of
1860 				 * the reference count blocks. Each will acquire a lock on the tail
1861 				 * block so that the journal won't be reaped until the reference
1862 				 * counts are initialized. The lock acquisition must be done by the
1863 				 * ref_counts since here we don't know how many reference blocks
1864 				 * the ref_counts has.
1865 				 */
1866 				for (i = 0; i < slab->reference_block_count; i++) {
1867 					slab->reference_blocks[i].slab_journal_lock = 1;
1868 					dirty_block(&slab->reference_blocks[i]);
1869 				}
1870 
1871 				adjust_slab_journal_block_reference(journal, 1,
1872 								    slab->reference_block_count);
1873 			}
1874 		}
1875 
1876 		vdo_waitq_notify_next_waiter(&journal->entry_waiters,
1877 					     add_entry_from_waiter, journal);
1878 	}
1879 
1880 	journal->adding_entries = false;
1881 
1882 	/* If there are no waiters, and we are flushing or saving, commit the tail block. */
1883 	if (vdo_is_state_draining(&journal->slab->state) &&
1884 	    !vdo_is_state_suspending(&journal->slab->state) &&
1885 	    !vdo_waitq_has_waiters(&journal->entry_waiters))
1886 		commit_tail(journal);
1887 }
1888 
1889 /**
1890  * reset_search_cursor() - Reset the free block search back to the first reference counter in the
1891  *                         first reference block of a slab.
1892  * @slab: The slab.
1893  */
reset_search_cursor(struct vdo_slab * slab)1894 static void reset_search_cursor(struct vdo_slab *slab)
1895 {
1896 	struct search_cursor *cursor = &slab->search_cursor;
1897 
1898 	cursor->block = cursor->first_block;
1899 	cursor->index = 0;
1900 	cursor->end_index = min_t(u32, COUNTS_PER_BLOCK, slab->block_count);
1901 }
1902 
1903 /**
1904  * advance_search_cursor() - Advance the search cursor to the start of the next reference block in
1905  *                           a slab.
1906  * @slab: The slab.
1907  *
1908  * Wraps around to the first reference block if the current block is the last reference block.
1909  *
1910  * Return: True unless the cursor was at the last reference block.
1911  */
advance_search_cursor(struct vdo_slab * slab)1912 static bool advance_search_cursor(struct vdo_slab *slab)
1913 {
1914 	struct search_cursor *cursor = &slab->search_cursor;
1915 
1916 	/*
1917 	 * If we just finished searching the last reference block, then wrap back around to the
1918 	 * start of the array.
1919 	 */
1920 	if (cursor->block == cursor->last_block) {
1921 		reset_search_cursor(slab);
1922 		return false;
1923 	}
1924 
1925 	/* We're not already at the end, so advance to cursor to the next block. */
1926 	cursor->block++;
1927 	cursor->index = cursor->end_index;
1928 
1929 	if (cursor->block == cursor->last_block) {
1930 		/* The last reference block will usually be a runt. */
1931 		cursor->end_index = slab->block_count;
1932 	} else {
1933 		cursor->end_index += COUNTS_PER_BLOCK;
1934 	}
1935 
1936 	return true;
1937 }
1938 
1939 /**
1940  * vdo_adjust_reference_count_for_rebuild() - Adjust the reference count of a block during rebuild.
1941  * @depot: The slab depot.
1942  * @pbn: The physical block number to adjust.
1943  * @operation: The type opf operation.
1944  *
1945  * Return: VDO_SUCCESS or an error.
1946  */
vdo_adjust_reference_count_for_rebuild(struct slab_depot * depot,physical_block_number_t pbn,enum journal_operation operation)1947 int vdo_adjust_reference_count_for_rebuild(struct slab_depot *depot,
1948 					   physical_block_number_t pbn,
1949 					   enum journal_operation operation)
1950 {
1951 	int result;
1952 	slab_block_number block_number;
1953 	struct reference_block *block;
1954 	struct vdo_slab *slab = vdo_get_slab(depot, pbn);
1955 	struct reference_updater updater = {
1956 		.operation = operation,
1957 		.increment = true,
1958 	};
1959 
1960 	result = slab_block_number_from_pbn(slab, pbn, &block_number);
1961 	if (result != VDO_SUCCESS)
1962 		return result;
1963 
1964 	block = get_reference_block(slab, block_number);
1965 	result = update_reference_count(slab, block, block_number, NULL,
1966 					&updater, !NORMAL_OPERATION, false, NULL);
1967 	if (result != VDO_SUCCESS)
1968 		return result;
1969 
1970 	dirty_block(block);
1971 	return VDO_SUCCESS;
1972 }
1973 
1974 /**
1975  * replay_reference_count_change() - Replay the reference count adjustment from a slab journal
1976  *                                   entry into the reference count for a block.
1977  * @slab: The slab.
1978  * @entry_point: The slab journal point for the entry.
1979  * @entry: The slab journal entry being replayed.
1980  *
1981  * The adjustment will be ignored if it was already recorded in the reference count.
1982  *
1983  * Return: VDO_SUCCESS or an error code.
1984  */
replay_reference_count_change(struct vdo_slab * slab,const struct journal_point * entry_point,struct slab_journal_entry entry)1985 static int replay_reference_count_change(struct vdo_slab *slab,
1986 					 const struct journal_point *entry_point,
1987 					 struct slab_journal_entry entry)
1988 {
1989 	int result;
1990 	struct reference_block *block = get_reference_block(slab, entry.sbn);
1991 	sector_count_t sector = (entry.sbn % COUNTS_PER_BLOCK) / COUNTS_PER_SECTOR;
1992 	struct reference_updater updater = {
1993 		.operation = entry.operation,
1994 		.increment = entry.increment,
1995 	};
1996 
1997 	if (!vdo_before_journal_point(&block->commit_points[sector], entry_point)) {
1998 		/* This entry is already reflected in the existing counts, so do nothing. */
1999 		return VDO_SUCCESS;
2000 	}
2001 
2002 	/* This entry is not yet counted in the reference counts. */
2003 	result = update_reference_count(slab, block, entry.sbn, entry_point,
2004 					&updater, !NORMAL_OPERATION, false, NULL);
2005 	if (result != VDO_SUCCESS)
2006 		return result;
2007 
2008 	dirty_block(block);
2009 	return VDO_SUCCESS;
2010 }
2011 
2012 /**
2013  * find_zero_byte_in_word() - Find the array index of the first zero byte in word-sized range of
2014  *                            reference counters.
2015  * @word_ptr: A pointer to the eight counter bytes to check.
2016  * @start_index: The array index corresponding to word_ptr[0].
2017  * @fail_index: The array index to return if no zero byte is found.
2018  *
2019  * The search does no bounds checking; the function relies on the array being sufficiently padded.
2020  *
2021  * Return: The array index of the first zero byte in the word, or the value passed as fail_index if
2022  *         no zero byte was found.
2023  */
find_zero_byte_in_word(const u8 * word_ptr,slab_block_number start_index,slab_block_number fail_index)2024 static inline slab_block_number find_zero_byte_in_word(const u8 *word_ptr,
2025 						       slab_block_number start_index,
2026 						       slab_block_number fail_index)
2027 {
2028 	u64 word = get_unaligned_le64(word_ptr);
2029 
2030 	/* This looks like a loop, but GCC will unroll the eight iterations for us. */
2031 	unsigned int offset;
2032 
2033 	for (offset = 0; offset < BYTES_PER_WORD; offset++) {
2034 		/* Assumes little-endian byte order, which we have on X86. */
2035 		if ((word & 0xFF) == 0)
2036 			return (start_index + offset);
2037 		word >>= 8;
2038 	}
2039 
2040 	return fail_index;
2041 }
2042 
2043 /**
2044  * find_free_block() - Find the first block with a reference count of zero in the specified
2045  *                     range of reference counter indexes.
2046  * @slab: The slab counters to scan.
2047  * @index_ptr: A pointer to hold the array index of the free block.
2048  *
2049  * Return: True if a free block was found in the specified range.
2050  */
find_free_block(const struct vdo_slab * slab,slab_block_number * index_ptr)2051 static bool find_free_block(const struct vdo_slab *slab, slab_block_number *index_ptr)
2052 {
2053 	slab_block_number zero_index;
2054 	slab_block_number next_index = slab->search_cursor.index;
2055 	slab_block_number end_index = slab->search_cursor.end_index;
2056 	u8 *next_counter = &slab->counters[next_index];
2057 	u8 *end_counter = &slab->counters[end_index];
2058 
2059 	/*
2060 	 * Search every byte of the first unaligned word. (Array is padded so reading past end is
2061 	 * safe.)
2062 	 */
2063 	zero_index = find_zero_byte_in_word(next_counter, next_index, end_index);
2064 	if (zero_index < end_index) {
2065 		*index_ptr = zero_index;
2066 		return true;
2067 	}
2068 
2069 	/*
2070 	 * On architectures where unaligned word access is expensive, this would be a good place to
2071 	 * advance to an alignment boundary.
2072 	 */
2073 	next_index += BYTES_PER_WORD;
2074 	next_counter += BYTES_PER_WORD;
2075 
2076 	/*
2077 	 * Now we're word-aligned; check an word at a time until we find a word containing a zero.
2078 	 * (Array is padded so reading past end is safe.)
2079 	 */
2080 	while (next_counter < end_counter) {
2081 		/*
2082 		 * The following code is currently an exact copy of the code preceding the loop,
2083 		 * but if you try to merge them by using a do loop, it runs slower because a jump
2084 		 * instruction gets added at the start of the iteration.
2085 		 */
2086 		zero_index = find_zero_byte_in_word(next_counter, next_index, end_index);
2087 		if (zero_index < end_index) {
2088 			*index_ptr = zero_index;
2089 			return true;
2090 		}
2091 
2092 		next_index += BYTES_PER_WORD;
2093 		next_counter += BYTES_PER_WORD;
2094 	}
2095 
2096 	return false;
2097 }
2098 
2099 /**
2100  * search_current_reference_block() - Search the reference block currently saved in the search
2101  *                                    cursor for a reference count of zero, starting at the saved
2102  *                                    counter index.
2103  * @slab: The slab to search.
2104  * @free_index_ptr: A pointer to receive the array index of the zero reference count.
2105  *
2106  * Return: True if an unreferenced counter was found.
2107  */
search_current_reference_block(const struct vdo_slab * slab,slab_block_number * free_index_ptr)2108 static bool search_current_reference_block(const struct vdo_slab *slab,
2109 					   slab_block_number *free_index_ptr)
2110 {
2111 	/* Don't bother searching if the current block is known to be full. */
2112 	return ((slab->search_cursor.block->allocated_count < COUNTS_PER_BLOCK) &&
2113 		find_free_block(slab, free_index_ptr));
2114 }
2115 
2116 /**
2117  * search_reference_blocks() - Search each reference block for a reference count of zero.
2118  * @slab: The slab to search.
2119  * @free_index_ptr: A pointer to receive the array index of the zero reference count.
2120  *
2121  * Searches each reference block for a reference count of zero, starting at the reference block and
2122  * counter index saved in the search cursor and searching up to the end of the last reference
2123  * block. The search does not wrap.
2124  *
2125  * Return: True if an unreferenced counter was found.
2126  */
search_reference_blocks(struct vdo_slab * slab,slab_block_number * free_index_ptr)2127 static bool search_reference_blocks(struct vdo_slab *slab,
2128 				    slab_block_number *free_index_ptr)
2129 {
2130 	/* Start searching at the saved search position in the current block. */
2131 	if (search_current_reference_block(slab, free_index_ptr))
2132 		return true;
2133 
2134 	/* Search each reference block up to the end of the slab. */
2135 	while (advance_search_cursor(slab)) {
2136 		if (search_current_reference_block(slab, free_index_ptr))
2137 			return true;
2138 	}
2139 
2140 	return false;
2141 }
2142 
2143 /**
2144  * make_provisional_reference() - Do the bookkeeping for making a provisional reference.
2145  * @slab: The slab.
2146  * @block_number: The index for the physical block to reference.
2147  */
make_provisional_reference(struct vdo_slab * slab,slab_block_number block_number)2148 static void make_provisional_reference(struct vdo_slab *slab,
2149 				       slab_block_number block_number)
2150 {
2151 	struct reference_block *block = get_reference_block(slab, block_number);
2152 
2153 	/*
2154 	 * Make the initial transition from an unreferenced block to a
2155 	 * provisionally allocated block.
2156 	 */
2157 	slab->counters[block_number] = PROVISIONAL_REFERENCE_COUNT;
2158 
2159 	/* Account for the allocation. */
2160 	block->allocated_count++;
2161 	slab->free_blocks--;
2162 }
2163 
2164 /**
2165  * dirty_all_reference_blocks() - Mark all reference count blocks in a slab as dirty.
2166  * @slab: The slab.
2167  */
dirty_all_reference_blocks(struct vdo_slab * slab)2168 static void dirty_all_reference_blocks(struct vdo_slab *slab)
2169 {
2170 	block_count_t i;
2171 
2172 	for (i = 0; i < slab->reference_block_count; i++)
2173 		dirty_block(&slab->reference_blocks[i]);
2174 }
2175 
journal_points_equal(struct journal_point first,struct journal_point second)2176 static inline bool journal_points_equal(struct journal_point first,
2177 					struct journal_point second)
2178 {
2179 	return ((first.sequence_number == second.sequence_number) &&
2180 		(first.entry_count == second.entry_count));
2181 }
2182 
2183 /**
2184  * match_bytes() - Check an 8-byte word for bytes matching the value specified
2185  * @input: A word to examine the bytes of.
2186  * @match: The byte value sought.
2187  *
2188  * Return: 1 in each byte when the corresponding input byte matched, 0 otherwise.
2189  */
match_bytes(u64 input,u8 match)2190 static inline u64 match_bytes(u64 input, u8 match)
2191 {
2192 	u64 temp = input ^ (match * 0x0101010101010101ULL);
2193 	/* top bit of each byte is set iff top bit of temp byte is clear; rest are 0 */
2194 	u64 test_top_bits = ~temp & 0x8080808080808080ULL;
2195 	/* top bit of each byte is set iff low 7 bits of temp byte are clear; rest are useless */
2196 	u64 test_low_bits = 0x8080808080808080ULL - (temp & 0x7f7f7f7f7f7f7f7fULL);
2197 	/* return 1 when both tests indicate temp byte is 0 */
2198 	return (test_top_bits & test_low_bits) >> 7;
2199 }
2200 
2201 /**
2202  * count_valid_references() - Process a newly loaded refcount array
2203  * @counters: The array of counters from a metadata block.
2204  *
2205  * Scan an 8-byte-aligned array of counters, fixing up any provisional values that
2206  * weren't cleaned up at shutdown, changing them internally to zero.
2207  *
2208  * Return: The number of blocks with a non-zero reference count.
2209  */
count_valid_references(vdo_refcount_t * counters)2210 static unsigned int count_valid_references(vdo_refcount_t *counters)
2211 {
2212 	u64 *words = (u64 *)counters;
2213 	/* It's easier to count occurrences of a specific byte than its absences. */
2214 	unsigned int empty_count = 0;
2215 	/* For speed, we process 8 bytes at once. */
2216 	unsigned int words_left = COUNTS_PER_BLOCK / sizeof(u64);
2217 
2218 	/*
2219 	 * Sanity check assumptions used for optimizing this code: Counters are bytes. The counter
2220 	 * array is a multiple of the word size.
2221 	 */
2222 	BUILD_BUG_ON(sizeof(vdo_refcount_t) != 1);
2223 	BUILD_BUG_ON((COUNTS_PER_BLOCK % sizeof(u64)) != 0);
2224 
2225 	while (words_left > 0) {
2226 		/*
2227 		 * This is used effectively as 8 byte-size counters. Byte 0 counts how many words
2228 		 * had the target value found in byte 0, etc. We just have to avoid overflow.
2229 		 */
2230 		u64 split_count = 0;
2231 		/*
2232 		 * The counter "% 255" trick used below to fold split_count into empty_count
2233 		 * imposes a limit of 254 bytes examined each iteration of the outer loop. We
2234 		 * process a word at a time, so that limit gets rounded down to 31 u64 words.
2235 		 */
2236 		const unsigned int max_words_per_iteration = 254 / sizeof(u64);
2237 		unsigned int iter_words_left = min_t(unsigned int, words_left,
2238 						     max_words_per_iteration);
2239 
2240 		words_left -= iter_words_left;
2241 
2242 		while (iter_words_left--) {
2243 			u64 word = *words;
2244 			u64 temp;
2245 
2246 			/* First, if we have any provisional refcount values, clear them. */
2247 			temp = match_bytes(word, PROVISIONAL_REFERENCE_COUNT);
2248 			if (temp) {
2249 				/*
2250 				 * 'temp' has 0x01 bytes where 'word' has PROVISIONAL; this xor
2251 				 * will alter just those bytes, changing PROVISIONAL to EMPTY.
2252 				 */
2253 				word ^= temp * (PROVISIONAL_REFERENCE_COUNT ^ EMPTY_REFERENCE_COUNT);
2254 				*words = word;
2255 			}
2256 
2257 			/* Now count the EMPTY_REFERENCE_COUNT bytes, updating the 8 counters. */
2258 			split_count += match_bytes(word, EMPTY_REFERENCE_COUNT);
2259 			words++;
2260 		}
2261 		empty_count += split_count % 255;
2262 	}
2263 
2264 	return COUNTS_PER_BLOCK - empty_count;
2265 }
2266 
2267 /**
2268  * unpack_reference_block() - Unpack reference counts blocks into the internal memory structure.
2269  * @packed: The written reference block to be unpacked.
2270  * @block: The internal reference block to be loaded.
2271  */
unpack_reference_block(struct packed_reference_block * packed,struct reference_block * block)2272 static void unpack_reference_block(struct packed_reference_block *packed,
2273 				   struct reference_block *block)
2274 {
2275 	sector_count_t i;
2276 	struct vdo_slab *slab = block->slab;
2277 	vdo_refcount_t *counters = get_reference_counters_for_block(block);
2278 
2279 	for (i = 0; i < VDO_SECTORS_PER_BLOCK; i++) {
2280 		struct packed_reference_sector *sector = &packed->sectors[i];
2281 
2282 		vdo_unpack_journal_point(&sector->commit_point, &block->commit_points[i]);
2283 		memcpy(counters + (i * COUNTS_PER_SECTOR), sector->counts,
2284 		       (sizeof(vdo_refcount_t) * COUNTS_PER_SECTOR));
2285 		/* The slab_journal_point must be the latest point found in any sector. */
2286 		if (vdo_before_journal_point(&slab->slab_journal_point,
2287 					     &block->commit_points[i]))
2288 			slab->slab_journal_point = block->commit_points[i];
2289 
2290 		if ((i > 0) &&
2291 		    !journal_points_equal(block->commit_points[0],
2292 					  block->commit_points[i])) {
2293 			size_t block_index = block - block->slab->reference_blocks;
2294 
2295 			vdo_log_warning("Torn write detected in sector %u of reference block %zu of slab %u",
2296 					i, block_index, block->slab->slab_number);
2297 		}
2298 	}
2299 
2300 	block->allocated_count = count_valid_references(counters);
2301 }
2302 
2303 /**
2304  * finish_reference_block_load() - After a reference block has been read, unpack it.
2305  * @completion: The VIO that just finished reading.
2306  */
finish_reference_block_load(struct vdo_completion * completion)2307 static void finish_reference_block_load(struct vdo_completion *completion)
2308 {
2309 	struct vio *vio = as_vio(completion);
2310 	struct pooled_vio *pooled = vio_as_pooled_vio(vio);
2311 	struct reference_block *block = completion->parent;
2312 	struct vdo_slab *slab = block->slab;
2313 	unsigned int block_count = vio->io_size / VDO_BLOCK_SIZE;
2314 	unsigned int i;
2315 	char *data = vio->data;
2316 
2317 	for (i = 0; i < block_count; i++, block++, data += VDO_BLOCK_SIZE) {
2318 		struct packed_reference_block *packed = (struct packed_reference_block *) data;
2319 
2320 		unpack_reference_block(packed, block);
2321 		slab->free_blocks -= block->allocated_count;
2322 	}
2323 	return_vio_to_pool(pooled);
2324 	slab->active_count -= block_count;
2325 
2326 	check_if_slab_drained(slab);
2327 }
2328 
load_reference_block_endio(struct bio * bio)2329 static void load_reference_block_endio(struct bio *bio)
2330 {
2331 	struct vio *vio = bio->bi_private;
2332 	struct reference_block *block = vio->completion.parent;
2333 
2334 	continue_vio_after_io(vio, finish_reference_block_load,
2335 			      block->slab->allocator->thread_id);
2336 }
2337 
2338 /**
2339  * load_reference_block_group() - After a block waiter has gotten a VIO from the VIO pool, load
2340  *                                a set of blocks.
2341  * @waiter: The waiter of the first block to load.
2342  * @context: The VIO returned by the pool.
2343  */
load_reference_block_group(struct vdo_waiter * waiter,void * context)2344 static void load_reference_block_group(struct vdo_waiter *waiter, void *context)
2345 {
2346 	struct pooled_vio *pooled = context;
2347 	struct vio *vio = &pooled->vio;
2348 	struct reference_block *block =
2349 		container_of(waiter, struct reference_block, waiter);
2350 	u32 block_offset = block - block->slab->reference_blocks;
2351 	u32 max_block_count = block->slab->reference_block_count - block_offset;
2352 	u32 block_count = min_t(int, vio->block_count, max_block_count);
2353 
2354 	vio->completion.parent = block;
2355 	vdo_submit_metadata_vio_with_size(vio, block->slab->ref_counts_origin + block_offset,
2356 					  load_reference_block_endio, handle_io_error,
2357 					  REQ_OP_READ, block_count * VDO_BLOCK_SIZE);
2358 }
2359 
2360 /**
2361  * load_reference_blocks() - Load a slab's reference blocks from the underlying storage into a
2362  *                           pre-allocated reference counter.
2363  * @slab: The slab.
2364  */
load_reference_blocks(struct vdo_slab * slab)2365 static void load_reference_blocks(struct vdo_slab *slab)
2366 {
2367 	block_count_t i;
2368 	u64 blocks_per_vio = slab->allocator->refcount_blocks_per_big_vio;
2369 	struct vio_pool *pool = slab->allocator->refcount_big_vio_pool;
2370 
2371 	if (!pool) {
2372 		pool = slab->allocator->vio_pool;
2373 		blocks_per_vio = 1;
2374 	}
2375 
2376 	slab->free_blocks = slab->block_count;
2377 	slab->active_count = slab->reference_block_count;
2378 	for (i = 0; i < slab->reference_block_count; i += blocks_per_vio) {
2379 		struct vdo_waiter *waiter = &slab->reference_blocks[i].waiter;
2380 
2381 		waiter->callback = load_reference_block_group;
2382 		acquire_vio_from_pool(pool, waiter);
2383 	}
2384 }
2385 
2386 /**
2387  * drain_slab() - Drain all reference count I/O.
2388  * @slab: The slab.
2389  *
2390  * Depending upon the type of drain being performed (as recorded in the ref_count's vdo_slab), the
2391  * reference blocks may be loaded from disk or dirty reference blocks may be written out.
2392  */
drain_slab(struct vdo_slab * slab)2393 static void drain_slab(struct vdo_slab *slab)
2394 {
2395 	bool save;
2396 	bool load;
2397 	const struct admin_state_code *state = vdo_get_admin_state_code(&slab->state);
2398 
2399 	if (state == VDO_ADMIN_STATE_SUSPENDING)
2400 		return;
2401 
2402 	if ((state != VDO_ADMIN_STATE_REBUILDING) &&
2403 	    (state != VDO_ADMIN_STATE_SAVE_FOR_SCRUBBING))
2404 		commit_tail(&slab->journal);
2405 
2406 	if ((state == VDO_ADMIN_STATE_RECOVERING) || (slab->counters == NULL))
2407 		return;
2408 
2409 	save = false;
2410 	load = slab->allocator->summary_entries[slab->slab_number].load_ref_counts;
2411 	if (state == VDO_ADMIN_STATE_SCRUBBING) {
2412 		if (load) {
2413 			load_reference_blocks(slab);
2414 			return;
2415 		}
2416 	} else if (state == VDO_ADMIN_STATE_SAVE_FOR_SCRUBBING) {
2417 		if (!load) {
2418 			/* These reference counts were never written, so mark them all dirty. */
2419 			dirty_all_reference_blocks(slab);
2420 		}
2421 		save = true;
2422 	} else if (state == VDO_ADMIN_STATE_REBUILDING) {
2423 		/*
2424 		 * Write out the counters if the slab has written them before, or it has any
2425 		 * non-zero reference counts, or there are any slab journal blocks.
2426 		 */
2427 		block_count_t data_blocks = slab->allocator->depot->slab_config.data_blocks;
2428 
2429 		if (load || (slab->free_blocks != data_blocks) ||
2430 		    !is_slab_journal_blank(slab)) {
2431 			dirty_all_reference_blocks(slab);
2432 			save = true;
2433 		}
2434 	} else if (state == VDO_ADMIN_STATE_SAVING) {
2435 		save = (slab->status == VDO_SLAB_REBUILT);
2436 	} else {
2437 		vdo_finish_draining_with_result(&slab->state, VDO_SUCCESS);
2438 		return;
2439 	}
2440 
2441 	if (save)
2442 		save_dirty_reference_blocks(slab);
2443 }
2444 
allocate_slab_counters(struct vdo_slab * slab)2445 static int allocate_slab_counters(struct vdo_slab *slab)
2446 {
2447 	int result;
2448 	size_t index, bytes;
2449 
2450 	result = VDO_ASSERT(slab->reference_blocks == NULL,
2451 			    "vdo_slab %u doesn't allocate refcounts twice",
2452 			    slab->slab_number);
2453 	if (result != VDO_SUCCESS)
2454 		return result;
2455 
2456 	result = vdo_allocate(slab->reference_block_count, __func__, &slab->reference_blocks);
2457 	if (result != VDO_SUCCESS)
2458 		return result;
2459 
2460 	/*
2461 	 * Allocate such that the runt slab has a full-length memory array, plus a little padding
2462 	 * so we can word-search even at the very end.
2463 	 */
2464 	bytes = (slab->reference_block_count * COUNTS_PER_BLOCK) + (2 * BYTES_PER_WORD);
2465 	result = vdo_allocate(bytes, "ref counts array", &slab->counters);
2466 	if (result != VDO_SUCCESS) {
2467 		vdo_free(vdo_forget(slab->reference_blocks));
2468 		return result;
2469 	}
2470 
2471 	slab->search_cursor.first_block = slab->reference_blocks;
2472 	slab->search_cursor.last_block = &slab->reference_blocks[slab->reference_block_count - 1];
2473 	reset_search_cursor(slab);
2474 
2475 	for (index = 0; index < slab->reference_block_count; index++) {
2476 		slab->reference_blocks[index] = (struct reference_block) {
2477 			.slab = slab,
2478 		};
2479 	}
2480 
2481 	return VDO_SUCCESS;
2482 }
2483 
allocate_counters_if_clean(struct vdo_slab * slab)2484 static int allocate_counters_if_clean(struct vdo_slab *slab)
2485 {
2486 	if (vdo_is_state_clean_load(&slab->state))
2487 		return allocate_slab_counters(slab);
2488 
2489 	return VDO_SUCCESS;
2490 }
2491 
finish_loading_journal(struct vdo_completion * completion)2492 static void finish_loading_journal(struct vdo_completion *completion)
2493 {
2494 	struct vio *vio = as_vio(completion);
2495 	struct slab_journal *journal = completion->parent;
2496 	struct vdo_slab *slab = journal->slab;
2497 	struct packed_slab_journal_block *block = (struct packed_slab_journal_block *) vio->data;
2498 	struct slab_journal_block_header header;
2499 
2500 	vdo_unpack_slab_journal_block_header(&block->header, &header);
2501 
2502 	/* FIXME: should it be an error if the following conditional fails? */
2503 	if ((header.metadata_type == VDO_METADATA_SLAB_JOURNAL) &&
2504 	    (header.nonce == slab->allocator->nonce)) {
2505 		journal->tail = header.sequence_number + 1;
2506 
2507 		/*
2508 		 * If the slab is clean, this implies the slab journal is empty, so advance the
2509 		 * head appropriately.
2510 		 */
2511 		journal->head = (slab->allocator->summary_entries[slab->slab_number].is_dirty ?
2512 				 header.head : journal->tail);
2513 		journal->tail_header = header;
2514 		initialize_journal_state(journal);
2515 	}
2516 
2517 	return_vio_to_pool(vio_as_pooled_vio(vio));
2518 	vdo_finish_loading_with_result(&slab->state, allocate_counters_if_clean(slab));
2519 }
2520 
read_slab_journal_tail_endio(struct bio * bio)2521 static void read_slab_journal_tail_endio(struct bio *bio)
2522 {
2523 	struct vio *vio = bio->bi_private;
2524 	struct slab_journal *journal = vio->completion.parent;
2525 
2526 	continue_vio_after_io(vio, finish_loading_journal,
2527 			      journal->slab->allocator->thread_id);
2528 }
2529 
handle_load_error(struct vdo_completion * completion)2530 static void handle_load_error(struct vdo_completion *completion)
2531 {
2532 	int result = completion->result;
2533 	struct slab_journal *journal = completion->parent;
2534 	struct vio *vio = as_vio(completion);
2535 
2536 	vio_record_metadata_io_error(vio);
2537 	return_vio_to_pool(vio_as_pooled_vio(vio));
2538 	vdo_finish_loading_with_result(&journal->slab->state, result);
2539 }
2540 
2541 /**
2542  * read_slab_journal_tail() - Read the slab journal tail block by using a vio acquired from the vio
2543  *                            pool.
2544  * @waiter: The vio pool waiter which has just been notified.
2545  * @context: The vio pool entry given to the waiter.
2546  *
2547  * This is the success callback from acquire_vio_from_pool() when loading a slab journal.
2548  */
read_slab_journal_tail(struct vdo_waiter * waiter,void * context)2549 static void read_slab_journal_tail(struct vdo_waiter *waiter, void *context)
2550 {
2551 	struct slab_journal *journal =
2552 		container_of(waiter, struct slab_journal, resource_waiter);
2553 	struct vdo_slab *slab = journal->slab;
2554 	struct pooled_vio *pooled = context;
2555 	struct vio *vio = &pooled->vio;
2556 	tail_block_offset_t last_commit_point =
2557 		slab->allocator->summary_entries[slab->slab_number].tail_block_offset;
2558 
2559 	/*
2560 	 * Slab summary keeps the commit point offset, so the tail block is the block before that.
2561 	 * Calculation supports small journals in unit tests.
2562 	 */
2563 	tail_block_offset_t tail_block = ((last_commit_point == 0) ?
2564 					  (tail_block_offset_t)(journal->size - 1) :
2565 					  (last_commit_point - 1));
2566 
2567 	vio->completion.parent = journal;
2568 	vio->completion.callback_thread_id = slab->allocator->thread_id;
2569 	vdo_submit_metadata_vio(vio, slab->journal_origin + tail_block,
2570 				read_slab_journal_tail_endio, handle_load_error,
2571 				REQ_OP_READ);
2572 }
2573 
2574 /**
2575  * load_slab_journal() - Load a slab's journal by reading the journal's tail.
2576  * @slab: The slab.
2577  */
load_slab_journal(struct vdo_slab * slab)2578 static void load_slab_journal(struct vdo_slab *slab)
2579 {
2580 	struct slab_journal *journal = &slab->journal;
2581 	tail_block_offset_t last_commit_point;
2582 
2583 	last_commit_point = slab->allocator->summary_entries[slab->slab_number].tail_block_offset;
2584 	if ((last_commit_point == 0) &&
2585 	    !slab->allocator->summary_entries[slab->slab_number].load_ref_counts) {
2586 		/*
2587 		 * This slab claims that it has a tail block at (journal->size - 1), but a head of
2588 		 * 1. This is impossible, due to the scrubbing threshold, on a real system, so
2589 		 * don't bother reading the (bogus) data off disk.
2590 		 */
2591 		VDO_ASSERT_LOG_ONLY(((journal->size < 16) ||
2592 				     (journal->scrubbing_threshold < (journal->size - 1))),
2593 				    "Scrubbing threshold protects against reads of unwritten slab journal blocks");
2594 		vdo_finish_loading_with_result(&slab->state,
2595 					       allocate_counters_if_clean(slab));
2596 		return;
2597 	}
2598 
2599 	journal->resource_waiter.callback = read_slab_journal_tail;
2600 	acquire_vio_from_pool(slab->allocator->vio_pool, &journal->resource_waiter);
2601 }
2602 
register_slab_for_scrubbing(struct vdo_slab * slab,bool high_priority)2603 static void register_slab_for_scrubbing(struct vdo_slab *slab, bool high_priority)
2604 {
2605 	struct slab_scrubber *scrubber = &slab->allocator->scrubber;
2606 
2607 	VDO_ASSERT_LOG_ONLY((slab->status != VDO_SLAB_REBUILT),
2608 			    "slab to be scrubbed is unrecovered");
2609 
2610 	if (slab->status != VDO_SLAB_REQUIRES_SCRUBBING)
2611 		return;
2612 
2613 	list_del_init(&slab->allocq_entry);
2614 	if (!slab->was_queued_for_scrubbing) {
2615 		WRITE_ONCE(scrubber->slab_count, scrubber->slab_count + 1);
2616 		slab->was_queued_for_scrubbing = true;
2617 	}
2618 
2619 	if (high_priority) {
2620 		slab->status = VDO_SLAB_REQUIRES_HIGH_PRIORITY_SCRUBBING;
2621 		list_add_tail(&slab->allocq_entry, &scrubber->high_priority_slabs);
2622 		return;
2623 	}
2624 
2625 	list_add_tail(&slab->allocq_entry, &scrubber->slabs);
2626 }
2627 
2628 /* Queue a slab for allocation or scrubbing. */
queue_slab(struct vdo_slab * slab)2629 static void queue_slab(struct vdo_slab *slab)
2630 {
2631 	struct block_allocator *allocator = slab->allocator;
2632 	block_count_t free_blocks;
2633 	int result;
2634 
2635 	VDO_ASSERT_LOG_ONLY(list_empty(&slab->allocq_entry),
2636 			"a requeued slab must not already be on a list");
2637 
2638 	if (vdo_is_read_only(allocator->depot->vdo))
2639 		return;
2640 
2641 	free_blocks = slab->free_blocks;
2642 	result = VDO_ASSERT((free_blocks <= allocator->depot->slab_config.data_blocks),
2643 			    "rebuilt slab %u must have a valid free block count (has %llu, expected maximum %llu)",
2644 			    slab->slab_number, (unsigned long long) free_blocks,
2645 			    (unsigned long long) allocator->depot->slab_config.data_blocks);
2646 	if (result != VDO_SUCCESS) {
2647 		vdo_enter_read_only_mode(allocator->depot->vdo, result);
2648 		return;
2649 	}
2650 
2651 	if (slab->status != VDO_SLAB_REBUILT) {
2652 		register_slab_for_scrubbing(slab, false);
2653 		return;
2654 	}
2655 
2656 	if (!vdo_is_state_resuming(&slab->state)) {
2657 		/*
2658 		 * If the slab is resuming, we've already accounted for it here, so don't do it
2659 		 * again.
2660 		 * FIXME: under what situation would the slab be resuming here?
2661 		 */
2662 		WRITE_ONCE(allocator->allocated_blocks,
2663 			   allocator->allocated_blocks - free_blocks);
2664 		if (!is_slab_journal_blank(slab)) {
2665 			WRITE_ONCE(allocator->statistics.slabs_opened,
2666 				   allocator->statistics.slabs_opened + 1);
2667 		}
2668 	}
2669 
2670 	if (allocator->depot->vdo->suspend_type == VDO_ADMIN_STATE_SAVING)
2671 		reopen_slab_journal(slab);
2672 
2673 	prioritize_slab(slab);
2674 }
2675 
2676 /** Implements vdo_admin_initiator_fn. */
initiate_slab_action(struct admin_state * state)2677 static void initiate_slab_action(struct admin_state *state)
2678 {
2679 	struct vdo_slab *slab = container_of(state, struct vdo_slab, state);
2680 
2681 	if (vdo_is_state_draining(state)) {
2682 		const struct admin_state_code *operation = vdo_get_admin_state_code(state);
2683 
2684 		if (operation == VDO_ADMIN_STATE_SCRUBBING)
2685 			slab->status = VDO_SLAB_REBUILDING;
2686 
2687 		drain_slab(slab);
2688 		check_if_slab_drained(slab);
2689 		return;
2690 	}
2691 
2692 	if (vdo_is_state_loading(state)) {
2693 		load_slab_journal(slab);
2694 		return;
2695 	}
2696 
2697 	if (vdo_is_state_resuming(state)) {
2698 		queue_slab(slab);
2699 		vdo_finish_resuming(state);
2700 		return;
2701 	}
2702 
2703 	vdo_finish_operation(state, VDO_INVALID_ADMIN_STATE);
2704 }
2705 
2706 /**
2707  * get_next_slab() - Get the next slab to scrub.
2708  * @scrubber: The slab scrubber.
2709  *
2710  * Return: The next slab to scrub or NULL if there are none.
2711  */
get_next_slab(struct slab_scrubber * scrubber)2712 static struct vdo_slab *get_next_slab(struct slab_scrubber *scrubber)
2713 {
2714 	struct vdo_slab *slab;
2715 
2716 	slab = list_first_entry_or_null(&scrubber->high_priority_slabs,
2717 					struct vdo_slab, allocq_entry);
2718 	if (slab != NULL)
2719 		return slab;
2720 
2721 	return list_first_entry_or_null(&scrubber->slabs, struct vdo_slab,
2722 					allocq_entry);
2723 }
2724 
2725 /**
2726  * has_slabs_to_scrub() - Check whether a scrubber has slabs to scrub.
2727  * @scrubber: The scrubber to check.
2728  *
2729  * Return: True if the scrubber has slabs to scrub.
2730  */
has_slabs_to_scrub(struct slab_scrubber * scrubber)2731 static inline bool __must_check has_slabs_to_scrub(struct slab_scrubber *scrubber)
2732 {
2733 	return (get_next_slab(scrubber) != NULL);
2734 }
2735 
2736 /**
2737  * uninitialize_scrubber_vio() - Clean up the slab_scrubber's vio.
2738  * @scrubber: The scrubber.
2739  */
uninitialize_scrubber_vio(struct slab_scrubber * scrubber)2740 static void uninitialize_scrubber_vio(struct slab_scrubber *scrubber)
2741 {
2742 	vdo_free(vdo_forget(scrubber->vio.data));
2743 	free_vio_components(&scrubber->vio);
2744 }
2745 
2746 /**
2747  * finish_scrubbing() - Stop scrubbing, either because there are no more slabs to scrub or because
2748  *                      there's been an error.
2749  * @scrubber: The scrubber.
2750  * @result: The result of the scrubbing operation.
2751  */
finish_scrubbing(struct slab_scrubber * scrubber,int result)2752 static void finish_scrubbing(struct slab_scrubber *scrubber, int result)
2753 {
2754 	bool notify = vdo_waitq_has_waiters(&scrubber->waiters);
2755 	bool done = !has_slabs_to_scrub(scrubber);
2756 	struct block_allocator *allocator =
2757 		container_of(scrubber, struct block_allocator, scrubber);
2758 
2759 	if (done)
2760 		uninitialize_scrubber_vio(scrubber);
2761 
2762 	if (scrubber->high_priority_only) {
2763 		scrubber->high_priority_only = false;
2764 		vdo_fail_completion(vdo_forget(scrubber->vio.completion.parent), result);
2765 	} else if (done && (atomic_add_return(-1, &allocator->depot->zones_to_scrub) == 0)) {
2766 		/* All of our slabs were scrubbed, and we're the last allocator to finish. */
2767 		enum vdo_state prior_state =
2768 			atomic_cmpxchg(&allocator->depot->vdo->state, VDO_RECOVERING,
2769 				       VDO_DIRTY);
2770 
2771 		/*
2772 		 * To be safe, even if the CAS failed, ensure anything that follows is ordered with
2773 		 * respect to whatever state change did happen.
2774 		 */
2775 		smp_mb__after_atomic();
2776 
2777 		/*
2778 		 * We must check the VDO state here and not the depot's read_only_notifier since
2779 		 * the compare-swap-above could have failed due to a read-only entry which our own
2780 		 * thread does not yet know about.
2781 		 */
2782 		if (prior_state == VDO_DIRTY)
2783 			vdo_log_info("VDO commencing normal operation");
2784 		else if (prior_state == VDO_RECOVERING)
2785 			vdo_log_info("Exiting recovery mode");
2786 		free_vio_pool(vdo_forget(allocator->refcount_big_vio_pool));
2787 	}
2788 
2789 	/*
2790 	 * Note that the scrubber has stopped, and inform anyone who might be waiting for that to
2791 	 * happen.
2792 	 */
2793 	if (!vdo_finish_draining(&scrubber->admin_state))
2794 		WRITE_ONCE(scrubber->admin_state.current_state,
2795 			   VDO_ADMIN_STATE_SUSPENDED);
2796 
2797 	/*
2798 	 * We can't notify waiters until after we've finished draining or they'll just requeue.
2799 	 * Fortunately if there were waiters, we can't have been freed yet.
2800 	 */
2801 	if (notify)
2802 		vdo_waitq_notify_all_waiters(&scrubber->waiters, NULL, NULL);
2803 }
2804 
2805 static void scrub_next_slab(struct slab_scrubber *scrubber);
2806 
2807 /**
2808  * slab_scrubbed() - Notify the scrubber that a slab has been scrubbed.
2809  * @completion: The slab rebuild completion.
2810  *
2811  * This callback is registered in apply_journal_entries().
2812  */
slab_scrubbed(struct vdo_completion * completion)2813 static void slab_scrubbed(struct vdo_completion *completion)
2814 {
2815 	struct slab_scrubber *scrubber =
2816 		container_of(as_vio(completion), struct slab_scrubber, vio);
2817 	struct vdo_slab *slab = scrubber->slab;
2818 
2819 	slab->status = VDO_SLAB_REBUILT;
2820 	queue_slab(slab);
2821 	reopen_slab_journal(slab);
2822 	WRITE_ONCE(scrubber->slab_count, scrubber->slab_count - 1);
2823 	scrub_next_slab(scrubber);
2824 }
2825 
2826 /**
2827  * abort_scrubbing() - Abort scrubbing due to an error.
2828  * @scrubber: The slab scrubber.
2829  * @result: The error.
2830  */
abort_scrubbing(struct slab_scrubber * scrubber,int result)2831 static void abort_scrubbing(struct slab_scrubber *scrubber, int result)
2832 {
2833 	vdo_enter_read_only_mode(scrubber->vio.completion.vdo, result);
2834 	finish_scrubbing(scrubber, result);
2835 }
2836 
2837 /**
2838  * handle_scrubber_error() - Handle errors while rebuilding a slab.
2839  * @completion: The slab rebuild completion.
2840  */
handle_scrubber_error(struct vdo_completion * completion)2841 static void handle_scrubber_error(struct vdo_completion *completion)
2842 {
2843 	struct vio *vio = as_vio(completion);
2844 
2845 	vio_record_metadata_io_error(vio);
2846 	abort_scrubbing(container_of(vio, struct slab_scrubber, vio),
2847 			completion->result);
2848 }
2849 
2850 /**
2851  * apply_block_entries() - Apply all the entries in a block to the reference counts.
2852  * @block: A block with entries to apply.
2853  * @entry_count: The number of entries to apply.
2854  * @block_number: The sequence number of the block.
2855  * @slab: The slab to apply the entries to.
2856  *
2857  * Return: VDO_SUCCESS or an error code.
2858  */
apply_block_entries(struct packed_slab_journal_block * block,journal_entry_count_t entry_count,sequence_number_t block_number,struct vdo_slab * slab)2859 static int apply_block_entries(struct packed_slab_journal_block *block,
2860 			       journal_entry_count_t entry_count,
2861 			       sequence_number_t block_number, struct vdo_slab *slab)
2862 {
2863 	struct journal_point entry_point = {
2864 		.sequence_number = block_number,
2865 		.entry_count = 0,
2866 	};
2867 	int result;
2868 	slab_block_number max_sbn = slab->end - slab->start;
2869 
2870 	while (entry_point.entry_count < entry_count) {
2871 		struct slab_journal_entry entry =
2872 			vdo_decode_slab_journal_entry(block, entry_point.entry_count);
2873 
2874 		if (entry.sbn > max_sbn) {
2875 			/* This entry is out of bounds. */
2876 			return vdo_log_error_strerror(VDO_CORRUPT_JOURNAL,
2877 						      "vdo_slab journal entry (%llu, %u) had invalid offset %u in slab (size %u blocks)",
2878 						      (unsigned long long) block_number,
2879 						      entry_point.entry_count,
2880 						      entry.sbn, max_sbn);
2881 		}
2882 
2883 		result = replay_reference_count_change(slab, &entry_point, entry);
2884 		if (result != VDO_SUCCESS) {
2885 			vdo_log_error_strerror(result,
2886 					       "vdo_slab journal entry (%llu, %u) (%s of offset %u) could not be applied in slab %u",
2887 					       (unsigned long long) block_number,
2888 					       entry_point.entry_count,
2889 					       vdo_get_journal_operation_name(entry.operation),
2890 					       entry.sbn, slab->slab_number);
2891 			return result;
2892 		}
2893 		entry_point.entry_count++;
2894 	}
2895 
2896 	return VDO_SUCCESS;
2897 }
2898 
2899 /**
2900  * apply_journal_entries() - Find the relevant vio of the slab journal and apply all valid entries.
2901  * @completion: The metadata read vio completion.
2902  *
2903  * This is a callback registered in start_scrubbing().
2904  */
apply_journal_entries(struct vdo_completion * completion)2905 static void apply_journal_entries(struct vdo_completion *completion)
2906 {
2907 	int result;
2908 	struct slab_scrubber *scrubber =
2909 		container_of(as_vio(completion), struct slab_scrubber, vio);
2910 	struct vdo_slab *slab = scrubber->slab;
2911 	struct slab_journal *journal = &slab->journal;
2912 
2913 	/* Find the boundaries of the useful part of the journal. */
2914 	sequence_number_t tail = journal->tail;
2915 	tail_block_offset_t end_index = (tail - 1) % journal->size;
2916 	char *end_data = scrubber->vio.data + (end_index * VDO_BLOCK_SIZE);
2917 	struct packed_slab_journal_block *end_block =
2918 		(struct packed_slab_journal_block *) end_data;
2919 
2920 	sequence_number_t head = __le64_to_cpu(end_block->header.head);
2921 	tail_block_offset_t head_index = head % journal->size;
2922 	block_count_t index = head_index;
2923 
2924 	struct journal_point ref_counts_point = slab->slab_journal_point;
2925 	struct journal_point last_entry_applied = ref_counts_point;
2926 	sequence_number_t sequence;
2927 
2928 	for (sequence = head; sequence < tail; sequence++) {
2929 		char *block_data = scrubber->vio.data + (index * VDO_BLOCK_SIZE);
2930 		struct packed_slab_journal_block *block =
2931 			(struct packed_slab_journal_block *) block_data;
2932 		struct slab_journal_block_header header;
2933 
2934 		vdo_unpack_slab_journal_block_header(&block->header, &header);
2935 
2936 		if ((header.nonce != slab->allocator->nonce) ||
2937 		    (header.metadata_type != VDO_METADATA_SLAB_JOURNAL) ||
2938 		    (header.sequence_number != sequence) ||
2939 		    (header.entry_count > journal->entries_per_block) ||
2940 		    (header.has_block_map_increments &&
2941 		     (header.entry_count > journal->full_entries_per_block))) {
2942 			/* The block is not what we expect it to be. */
2943 			vdo_log_error("vdo_slab journal block for slab %u was invalid",
2944 				      slab->slab_number);
2945 			abort_scrubbing(scrubber, VDO_CORRUPT_JOURNAL);
2946 			return;
2947 		}
2948 
2949 		result = apply_block_entries(block, header.entry_count, sequence, slab);
2950 		if (result != VDO_SUCCESS) {
2951 			abort_scrubbing(scrubber, result);
2952 			return;
2953 		}
2954 
2955 		last_entry_applied.sequence_number = sequence;
2956 		last_entry_applied.entry_count = header.entry_count - 1;
2957 		index++;
2958 		if (index == journal->size)
2959 			index = 0;
2960 	}
2961 
2962 	/*
2963 	 * At the end of rebuild, the reference counters should be accurate to the end of the
2964 	 * journal we just applied.
2965 	 */
2966 	result = VDO_ASSERT(!vdo_before_journal_point(&last_entry_applied,
2967 						      &ref_counts_point),
2968 			    "Refcounts are not more accurate than the slab journal");
2969 	if (result != VDO_SUCCESS) {
2970 		abort_scrubbing(scrubber, result);
2971 		return;
2972 	}
2973 
2974 	/* Save out the rebuilt reference blocks. */
2975 	vdo_prepare_completion(completion, slab_scrubbed, handle_scrubber_error,
2976 			       slab->allocator->thread_id, completion->parent);
2977 	vdo_start_operation_with_waiter(&slab->state,
2978 					VDO_ADMIN_STATE_SAVE_FOR_SCRUBBING,
2979 					completion, initiate_slab_action);
2980 }
2981 
read_slab_journal_endio(struct bio * bio)2982 static void read_slab_journal_endio(struct bio *bio)
2983 {
2984 	struct vio *vio = bio->bi_private;
2985 	struct slab_scrubber *scrubber = container_of(vio, struct slab_scrubber, vio);
2986 
2987 	continue_vio_after_io(bio->bi_private, apply_journal_entries,
2988 			      scrubber->slab->allocator->thread_id);
2989 }
2990 
2991 /**
2992  * start_scrubbing() - Read the current slab's journal from disk now that it has been flushed.
2993  * @completion: The scrubber's vio completion.
2994  *
2995  * This callback is registered in scrub_next_slab().
2996  */
start_scrubbing(struct vdo_completion * completion)2997 static void start_scrubbing(struct vdo_completion *completion)
2998 {
2999 	struct slab_scrubber *scrubber =
3000 		container_of(as_vio(completion), struct slab_scrubber, vio);
3001 	struct vdo_slab *slab = scrubber->slab;
3002 
3003 	if (!slab->allocator->summary_entries[slab->slab_number].is_dirty) {
3004 		slab_scrubbed(completion);
3005 		return;
3006 	}
3007 
3008 	vdo_submit_metadata_vio(&scrubber->vio, slab->journal_origin,
3009 				read_slab_journal_endio, handle_scrubber_error,
3010 				REQ_OP_READ);
3011 }
3012 
3013 /**
3014  * scrub_next_slab() - Scrub the next slab if there is one.
3015  * @scrubber: The scrubber.
3016  */
scrub_next_slab(struct slab_scrubber * scrubber)3017 static void scrub_next_slab(struct slab_scrubber *scrubber)
3018 {
3019 	struct vdo_completion *completion = &scrubber->vio.completion;
3020 	struct vdo_slab *slab;
3021 
3022 	/*
3023 	 * Note: this notify call is always safe only because scrubbing can only be started when
3024 	 * the VDO is quiescent.
3025 	 */
3026 	vdo_waitq_notify_all_waiters(&scrubber->waiters, NULL, NULL);
3027 
3028 	if (vdo_is_read_only(completion->vdo)) {
3029 		finish_scrubbing(scrubber, VDO_READ_ONLY);
3030 		return;
3031 	}
3032 
3033 	slab = get_next_slab(scrubber);
3034 	if ((slab == NULL) ||
3035 	    (scrubber->high_priority_only && list_empty(&scrubber->high_priority_slabs))) {
3036 		finish_scrubbing(scrubber, VDO_SUCCESS);
3037 		return;
3038 	}
3039 
3040 	if (vdo_finish_draining(&scrubber->admin_state))
3041 		return;
3042 
3043 	list_del_init(&slab->allocq_entry);
3044 	scrubber->slab = slab;
3045 	vdo_prepare_completion(completion, start_scrubbing, handle_scrubber_error,
3046 			       slab->allocator->thread_id, completion->parent);
3047 	vdo_start_operation_with_waiter(&slab->state, VDO_ADMIN_STATE_SCRUBBING,
3048 					completion, initiate_slab_action);
3049 }
3050 
3051 /**
3052  * scrub_slabs() - Scrub all of an allocator's slabs that are eligible for scrubbing.
3053  * @allocator: The block_allocator to scrub.
3054  * @parent: The completion to notify when scrubbing is done, implies high_priority, may be NULL.
3055  */
scrub_slabs(struct block_allocator * allocator,struct vdo_completion * parent)3056 static void scrub_slabs(struct block_allocator *allocator, struct vdo_completion *parent)
3057 {
3058 	struct slab_scrubber *scrubber = &allocator->scrubber;
3059 
3060 	scrubber->vio.completion.parent = parent;
3061 	scrubber->high_priority_only = (parent != NULL);
3062 	if (!has_slabs_to_scrub(scrubber)) {
3063 		finish_scrubbing(scrubber, VDO_SUCCESS);
3064 		return;
3065 	}
3066 
3067 	if (scrubber->high_priority_only &&
3068 	    vdo_is_priority_table_empty(allocator->prioritized_slabs) &&
3069 	    list_empty(&scrubber->high_priority_slabs))
3070 		register_slab_for_scrubbing(get_next_slab(scrubber), true);
3071 
3072 	vdo_resume_if_quiescent(&scrubber->admin_state);
3073 	scrub_next_slab(scrubber);
3074 }
3075 
assert_on_allocator_thread(thread_id_t thread_id,const char * function_name)3076 static inline void assert_on_allocator_thread(thread_id_t thread_id,
3077 					      const char *function_name)
3078 {
3079 	VDO_ASSERT_LOG_ONLY((vdo_get_callback_thread_id() == thread_id),
3080 			    "%s called on correct thread", function_name);
3081 }
3082 
register_slab_with_allocator(struct block_allocator * allocator,struct vdo_slab * slab)3083 static void register_slab_with_allocator(struct block_allocator *allocator,
3084 					 struct vdo_slab *slab)
3085 {
3086 	allocator->slab_count++;
3087 	allocator->last_slab = slab->slab_number;
3088 }
3089 
3090 /**
3091  * get_depot_slab_iterator() - Return a slab_iterator over the slabs in a slab_depot.
3092  * @depot: The depot over which to iterate.
3093  * @start: The number of the slab to start iterating from.
3094  * @end: The number of the last slab which may be returned.
3095  * @stride: The difference in slab number between successive slabs.
3096  *
3097  * Iteration always occurs from higher to lower numbered slabs.
3098  *
3099  * Return: An initialized iterator structure.
3100  */
get_depot_slab_iterator(struct slab_depot * depot,slab_count_t start,slab_count_t end,slab_count_t stride)3101 static struct slab_iterator get_depot_slab_iterator(struct slab_depot *depot,
3102 						    slab_count_t start, slab_count_t end,
3103 						    slab_count_t stride)
3104 {
3105 	struct vdo_slab **slabs = depot->slabs;
3106 
3107 	return (struct slab_iterator) {
3108 		.slabs = slabs,
3109 		.next = (((slabs == NULL) || (start < end)) ? NULL : slabs[start]),
3110 		.end = end,
3111 		.stride = stride,
3112 	};
3113 }
3114 
get_slab_iterator(const struct block_allocator * allocator)3115 static struct slab_iterator get_slab_iterator(const struct block_allocator *allocator)
3116 {
3117 	return get_depot_slab_iterator(allocator->depot, allocator->last_slab,
3118 				       allocator->zone_number,
3119 				       allocator->depot->zone_count);
3120 }
3121 
3122 /**
3123  * next_slab() - Get the next slab from a slab_iterator and advance the iterator
3124  * @iterator: The slab_iterator.
3125  *
3126  * Return: The next slab or NULL if the iterator is exhausted.
3127  */
next_slab(struct slab_iterator * iterator)3128 static struct vdo_slab *next_slab(struct slab_iterator *iterator)
3129 {
3130 	struct vdo_slab *slab = iterator->next;
3131 
3132 	if ((slab == NULL) || (slab->slab_number < iterator->end + iterator->stride))
3133 		iterator->next = NULL;
3134 	else
3135 		iterator->next = iterator->slabs[slab->slab_number - iterator->stride];
3136 
3137 	return slab;
3138 }
3139 
3140 /**
3141  * abort_waiter() - Abort vios waiting to make journal entries when read-only.
3142  * @waiter: A waiting data_vio.
3143  * @context: Not used.
3144  *
3145  * This callback is invoked on all vios waiting to make slab journal entries after the VDO has gone
3146  * into read-only mode. Implements waiter_callback_fn.
3147  */
abort_waiter(struct vdo_waiter * waiter,void __always_unused * context)3148 static void abort_waiter(struct vdo_waiter *waiter, void __always_unused *context)
3149 {
3150 	struct reference_updater *updater =
3151 		container_of(waiter, struct reference_updater, waiter);
3152 	struct data_vio *data_vio = data_vio_from_reference_updater(updater);
3153 
3154 	if (updater->increment) {
3155 		continue_data_vio_with_error(data_vio, VDO_READ_ONLY);
3156 		return;
3157 	}
3158 
3159 	vdo_continue_completion(&data_vio->decrement_completion, VDO_READ_ONLY);
3160 }
3161 
3162 /* Implements vdo_read_only_notification_fn. */
notify_block_allocator_of_read_only_mode(void * listener,struct vdo_completion * parent)3163 static void notify_block_allocator_of_read_only_mode(void *listener,
3164 						     struct vdo_completion *parent)
3165 {
3166 	struct block_allocator *allocator = listener;
3167 	struct slab_iterator iterator;
3168 
3169 	assert_on_allocator_thread(allocator->thread_id, __func__);
3170 	iterator = get_slab_iterator(allocator);
3171 	while (iterator.next != NULL) {
3172 		struct vdo_slab *slab = next_slab(&iterator);
3173 
3174 		vdo_waitq_notify_all_waiters(&slab->journal.entry_waiters,
3175 					     abort_waiter, &slab->journal);
3176 		check_if_slab_drained(slab);
3177 	}
3178 
3179 	vdo_finish_completion(parent);
3180 }
3181 
3182 /**
3183  * vdo_acquire_provisional_reference() - Acquire a provisional reference on behalf of a PBN lock if
3184  *                                       the block it locks is unreferenced.
3185  * @slab: The slab which contains the block.
3186  * @pbn: The physical block to reference.
3187  * @lock: The lock.
3188  *
3189  * Return: VDO_SUCCESS or an error.
3190  */
vdo_acquire_provisional_reference(struct vdo_slab * slab,physical_block_number_t pbn,struct pbn_lock * lock)3191 int vdo_acquire_provisional_reference(struct vdo_slab *slab, physical_block_number_t pbn,
3192 				      struct pbn_lock *lock)
3193 {
3194 	slab_block_number block_number;
3195 	int result;
3196 
3197 	if (vdo_pbn_lock_has_provisional_reference(lock))
3198 		return VDO_SUCCESS;
3199 
3200 	if (!is_slab_open(slab))
3201 		return VDO_INVALID_ADMIN_STATE;
3202 
3203 	result = slab_block_number_from_pbn(slab, pbn, &block_number);
3204 	if (result != VDO_SUCCESS)
3205 		return result;
3206 
3207 	if (slab->counters[block_number] == EMPTY_REFERENCE_COUNT) {
3208 		make_provisional_reference(slab, block_number);
3209 		if (lock != NULL)
3210 			vdo_assign_pbn_lock_provisional_reference(lock);
3211 	}
3212 
3213 	if (vdo_pbn_lock_has_provisional_reference(lock))
3214 		adjust_free_block_count(slab, false);
3215 
3216 	return VDO_SUCCESS;
3217 }
3218 
allocate_slab_block(struct vdo_slab * slab,physical_block_number_t * block_number_ptr)3219 static int __must_check allocate_slab_block(struct vdo_slab *slab,
3220 					    physical_block_number_t *block_number_ptr)
3221 {
3222 	slab_block_number free_index;
3223 
3224 	if (!is_slab_open(slab))
3225 		return VDO_INVALID_ADMIN_STATE;
3226 
3227 	if (!search_reference_blocks(slab, &free_index))
3228 		return VDO_NO_SPACE;
3229 
3230 	VDO_ASSERT_LOG_ONLY((slab->counters[free_index] == EMPTY_REFERENCE_COUNT),
3231 			    "free block must have ref count of zero");
3232 	make_provisional_reference(slab, free_index);
3233 	adjust_free_block_count(slab, false);
3234 
3235 	/*
3236 	 * Update the search hint so the next search will start at the array index just past the
3237 	 * free block we just found.
3238 	 */
3239 	slab->search_cursor.index = (free_index + 1);
3240 
3241 	*block_number_ptr = slab->start + free_index;
3242 	return VDO_SUCCESS;
3243 }
3244 
3245 /**
3246  * open_slab() - Prepare a slab to be allocated from.
3247  * @slab: The slab.
3248  */
open_slab(struct vdo_slab * slab)3249 static void open_slab(struct vdo_slab *slab)
3250 {
3251 	reset_search_cursor(slab);
3252 	if (is_slab_journal_blank(slab)) {
3253 		WRITE_ONCE(slab->allocator->statistics.slabs_opened,
3254 			   slab->allocator->statistics.slabs_opened + 1);
3255 		dirty_all_reference_blocks(slab);
3256 	} else {
3257 		WRITE_ONCE(slab->allocator->statistics.slabs_reopened,
3258 			   slab->allocator->statistics.slabs_reopened + 1);
3259 	}
3260 
3261 	slab->allocator->open_slab = slab;
3262 }
3263 
3264 
3265 /*
3266  * The block allocated will have a provisional reference and the reference must be either confirmed
3267  * with a subsequent increment or vacated with a subsequent decrement via
3268  * vdo_release_block_reference().
3269  */
vdo_allocate_block(struct block_allocator * allocator,physical_block_number_t * block_number_ptr)3270 int vdo_allocate_block(struct block_allocator *allocator,
3271 		       physical_block_number_t *block_number_ptr)
3272 {
3273 	int result;
3274 
3275 	if (allocator->open_slab != NULL) {
3276 		/* Try to allocate the next block in the currently open slab. */
3277 		result = allocate_slab_block(allocator->open_slab, block_number_ptr);
3278 		if ((result == VDO_SUCCESS) || (result != VDO_NO_SPACE))
3279 			return result;
3280 
3281 		/* Put the exhausted open slab back into the priority table. */
3282 		prioritize_slab(allocator->open_slab);
3283 	}
3284 
3285 	/* Remove the highest priority slab from the priority table and make it the open slab. */
3286 	open_slab(list_entry(vdo_priority_table_dequeue(allocator->prioritized_slabs),
3287 			     struct vdo_slab, allocq_entry));
3288 
3289 	/*
3290 	 * Try allocating again. If we're out of space immediately after opening a slab, then every
3291 	 * slab must be fully allocated.
3292 	 */
3293 	return allocate_slab_block(allocator->open_slab, block_number_ptr);
3294 }
3295 
3296 /**
3297  * vdo_enqueue_clean_slab_waiter() - Wait for a clean slab.
3298  * @allocator: The block_allocator on which to wait.
3299  * @waiter: The waiter.
3300  *
3301  * Return: VDO_SUCCESS if the waiter was queued, VDO_NO_SPACE if there are no slabs to scrub, and
3302  *         some other error otherwise.
3303  */
vdo_enqueue_clean_slab_waiter(struct block_allocator * allocator,struct vdo_waiter * waiter)3304 int vdo_enqueue_clean_slab_waiter(struct block_allocator *allocator,
3305 				  struct vdo_waiter *waiter)
3306 {
3307 	if (vdo_is_read_only(allocator->depot->vdo))
3308 		return VDO_READ_ONLY;
3309 
3310 	if (vdo_is_state_quiescent(&allocator->scrubber.admin_state))
3311 		return VDO_NO_SPACE;
3312 
3313 	vdo_waitq_enqueue_waiter(&allocator->scrubber.waiters, waiter);
3314 	return VDO_SUCCESS;
3315 }
3316 
3317 /**
3318  * vdo_modify_reference_count() - Modify the reference count of a block by first making a slab
3319  *                                journal entry and then updating the reference counter.
3320  * @completion: The data_vio completion for which to add the entry.
3321  * @updater: Which of the data_vio's reference updaters is being submitted.
3322  */
vdo_modify_reference_count(struct vdo_completion * completion,struct reference_updater * updater)3323 void vdo_modify_reference_count(struct vdo_completion *completion,
3324 				struct reference_updater *updater)
3325 {
3326 	struct vdo_slab *slab = vdo_get_slab(completion->vdo->depot, updater->zpbn.pbn);
3327 
3328 	if (!is_slab_open(slab)) {
3329 		vdo_continue_completion(completion, VDO_INVALID_ADMIN_STATE);
3330 		return;
3331 	}
3332 
3333 	if (vdo_is_read_only(completion->vdo)) {
3334 		vdo_continue_completion(completion, VDO_READ_ONLY);
3335 		return;
3336 	}
3337 
3338 	vdo_waitq_enqueue_waiter(&slab->journal.entry_waiters, &updater->waiter);
3339 	if ((slab->status != VDO_SLAB_REBUILT) && requires_reaping(&slab->journal))
3340 		register_slab_for_scrubbing(slab, true);
3341 
3342 	add_entries(&slab->journal);
3343 }
3344 
3345 /* Release an unused provisional reference. */
vdo_release_block_reference(struct block_allocator * allocator,physical_block_number_t pbn)3346 int vdo_release_block_reference(struct block_allocator *allocator,
3347 				physical_block_number_t pbn)
3348 {
3349 	struct reference_updater updater;
3350 
3351 	if (pbn == VDO_ZERO_BLOCK)
3352 		return VDO_SUCCESS;
3353 
3354 	updater = (struct reference_updater) {
3355 		.operation = VDO_JOURNAL_DATA_REMAPPING,
3356 		.increment = false,
3357 		.zpbn = {
3358 			.pbn = pbn,
3359 		},
3360 	};
3361 
3362 	return adjust_reference_count(vdo_get_slab(allocator->depot, pbn),
3363 				      &updater, NULL);
3364 }
3365 
3366 /*
3367  * This is a min_heap callback function orders slab_status structures using the 'is_clean' field as
3368  * the primary key and the 'emptiness' field as the secondary key.
3369  *
3370  * Slabs need to be pushed onto the lists in the same order they are to be popped off. Popping
3371  * should always get the most empty first, so pushing should be from most empty to least empty.
3372  * Thus, the ordering is reversed from the usual sense since min_heap returns smaller elements
3373  * before larger ones.
3374  */
slab_status_is_less_than(const void * item1,const void * item2,void __always_unused * args)3375 static bool slab_status_is_less_than(const void *item1, const void *item2,
3376 					void __always_unused *args)
3377 {
3378 	const struct slab_status *info1 = item1;
3379 	const struct slab_status *info2 = item2;
3380 
3381 	if (info1->is_clean != info2->is_clean)
3382 		return info1->is_clean;
3383 	if (info1->emptiness != info2->emptiness)
3384 		return info1->emptiness > info2->emptiness;
3385 	return info1->slab_number < info2->slab_number;
3386 }
3387 
3388 static const struct min_heap_callbacks slab_status_min_heap = {
3389 	.less = slab_status_is_less_than,
3390 	.swp = NULL,
3391 };
3392 
3393 /* Inform the slab actor that a action has finished on some slab; used by apply_to_slabs(). */
slab_action_callback(struct vdo_completion * completion)3394 static void slab_action_callback(struct vdo_completion *completion)
3395 {
3396 	struct block_allocator *allocator = vdo_as_block_allocator(completion);
3397 	struct slab_actor *actor = &allocator->slab_actor;
3398 
3399 	if (--actor->slab_action_count == 0) {
3400 		actor->callback(completion);
3401 		return;
3402 	}
3403 
3404 	vdo_reset_completion(completion);
3405 }
3406 
3407 /* Preserve the error from part of an action and continue. */
handle_operation_error(struct vdo_completion * completion)3408 static void handle_operation_error(struct vdo_completion *completion)
3409 {
3410 	struct block_allocator *allocator = vdo_as_block_allocator(completion);
3411 
3412 	if (allocator->state.waiter != NULL)
3413 		vdo_set_completion_result(allocator->state.waiter, completion->result);
3414 	completion->callback(completion);
3415 }
3416 
3417 /* Perform an action on each of an allocator's slabs in parallel. */
apply_to_slabs(struct block_allocator * allocator,vdo_action_fn callback)3418 static void apply_to_slabs(struct block_allocator *allocator, vdo_action_fn callback)
3419 {
3420 	struct slab_iterator iterator;
3421 
3422 	vdo_prepare_completion(&allocator->completion, slab_action_callback,
3423 			       handle_operation_error, allocator->thread_id, NULL);
3424 	allocator->completion.requeue = false;
3425 
3426 	/*
3427 	 * Since we are going to dequeue all of the slabs, the open slab will become invalid, so
3428 	 * clear it.
3429 	 */
3430 	allocator->open_slab = NULL;
3431 
3432 	/* Ensure that we don't finish before we're done starting. */
3433 	allocator->slab_actor = (struct slab_actor) {
3434 		.slab_action_count = 1,
3435 		.callback = callback,
3436 	};
3437 
3438 	iterator = get_slab_iterator(allocator);
3439 	while (iterator.next != NULL) {
3440 		const struct admin_state_code *operation =
3441 			vdo_get_admin_state_code(&allocator->state);
3442 		struct vdo_slab *slab = next_slab(&iterator);
3443 
3444 		list_del_init(&slab->allocq_entry);
3445 		allocator->slab_actor.slab_action_count++;
3446 		vdo_start_operation_with_waiter(&slab->state, operation,
3447 						&allocator->completion,
3448 						initiate_slab_action);
3449 	}
3450 
3451 	slab_action_callback(&allocator->completion);
3452 }
3453 
finish_loading_allocator(struct vdo_completion * completion)3454 static void finish_loading_allocator(struct vdo_completion *completion)
3455 {
3456 	struct block_allocator *allocator = vdo_as_block_allocator(completion);
3457 	const struct admin_state_code *operation =
3458 		vdo_get_admin_state_code(&allocator->state);
3459 
3460 	if (allocator->eraser != NULL)
3461 		dm_kcopyd_client_destroy(vdo_forget(allocator->eraser));
3462 
3463 	if (operation == VDO_ADMIN_STATE_LOADING_FOR_RECOVERY) {
3464 		void *context =
3465 			vdo_get_current_action_context(allocator->depot->action_manager);
3466 
3467 		vdo_replay_into_slab_journals(allocator, context);
3468 		return;
3469 	}
3470 
3471 	vdo_finish_loading(&allocator->state);
3472 }
3473 
3474 static void erase_next_slab_journal(struct block_allocator *allocator);
3475 
copy_callback(int read_err,unsigned long write_err,void * context)3476 static void copy_callback(int read_err, unsigned long write_err, void *context)
3477 {
3478 	struct block_allocator *allocator = context;
3479 	int result = (((read_err == 0) && (write_err == 0)) ? VDO_SUCCESS : -EIO);
3480 
3481 	if (result != VDO_SUCCESS) {
3482 		vdo_fail_completion(&allocator->completion, result);
3483 		return;
3484 	}
3485 
3486 	erase_next_slab_journal(allocator);
3487 }
3488 
3489 /* erase_next_slab_journal() - Erase the next slab journal. */
erase_next_slab_journal(struct block_allocator * allocator)3490 static void erase_next_slab_journal(struct block_allocator *allocator)
3491 {
3492 	struct vdo_slab *slab;
3493 	physical_block_number_t pbn;
3494 	struct dm_io_region regions[1];
3495 	struct slab_depot *depot = allocator->depot;
3496 	block_count_t blocks = depot->slab_config.slab_journal_blocks;
3497 
3498 	if (allocator->slabs_to_erase.next == NULL) {
3499 		vdo_finish_completion(&allocator->completion);
3500 		return;
3501 	}
3502 
3503 	slab = next_slab(&allocator->slabs_to_erase);
3504 	pbn = slab->journal_origin - depot->vdo->geometry.bio_offset;
3505 	regions[0] = (struct dm_io_region) {
3506 		.bdev = vdo_get_backing_device(depot->vdo),
3507 		.sector = pbn * VDO_SECTORS_PER_BLOCK,
3508 		.count = blocks * VDO_SECTORS_PER_BLOCK,
3509 	};
3510 	dm_kcopyd_zero(allocator->eraser, 1, regions, 0, copy_callback, allocator);
3511 }
3512 
3513 /* Implements vdo_admin_initiator_fn. */
initiate_load(struct admin_state * state)3514 static void initiate_load(struct admin_state *state)
3515 {
3516 	struct block_allocator *allocator =
3517 		container_of(state, struct block_allocator, state);
3518 	const struct admin_state_code *operation = vdo_get_admin_state_code(state);
3519 
3520 	if (operation == VDO_ADMIN_STATE_LOADING_FOR_REBUILD) {
3521 		/*
3522 		 * Must requeue because the kcopyd client cannot be freed in the same stack frame
3523 		 * as the kcopyd callback, lest it deadlock.
3524 		 */
3525 		vdo_prepare_completion_for_requeue(&allocator->completion,
3526 						   finish_loading_allocator,
3527 						   handle_operation_error,
3528 						   allocator->thread_id, NULL);
3529 		allocator->eraser = dm_kcopyd_client_create(NULL);
3530 		if (IS_ERR(allocator->eraser)) {
3531 			vdo_fail_completion(&allocator->completion,
3532 					    PTR_ERR(allocator->eraser));
3533 			allocator->eraser = NULL;
3534 			return;
3535 		}
3536 		allocator->slabs_to_erase = get_slab_iterator(allocator);
3537 
3538 		erase_next_slab_journal(allocator);
3539 		return;
3540 	}
3541 
3542 	apply_to_slabs(allocator, finish_loading_allocator);
3543 }
3544 
3545 /**
3546  * vdo_notify_slab_journals_are_recovered() - Inform a block allocator that its slab journals have
3547  *                                            been recovered from the recovery journal.
3548  * @completion: The allocator completion.
3549  */
vdo_notify_slab_journals_are_recovered(struct vdo_completion * completion)3550 void vdo_notify_slab_journals_are_recovered(struct vdo_completion *completion)
3551 {
3552 	struct block_allocator *allocator = vdo_as_block_allocator(completion);
3553 
3554 	vdo_finish_loading_with_result(&allocator->state, completion->result);
3555 }
3556 
get_slab_statuses(struct block_allocator * allocator,struct slab_status ** statuses_ptr)3557 static int get_slab_statuses(struct block_allocator *allocator,
3558 			     struct slab_status **statuses_ptr)
3559 {
3560 	int result;
3561 	struct slab_status *statuses;
3562 	struct slab_iterator iterator = get_slab_iterator(allocator);
3563 
3564 	result = vdo_allocate(allocator->slab_count, __func__, &statuses);
3565 	if (result != VDO_SUCCESS)
3566 		return result;
3567 
3568 	*statuses_ptr = statuses;
3569 
3570 	while (iterator.next != NULL)  {
3571 		slab_count_t slab_number = next_slab(&iterator)->slab_number;
3572 
3573 		*statuses++ = (struct slab_status) {
3574 			.slab_number = slab_number,
3575 			.is_clean = !allocator->summary_entries[slab_number].is_dirty,
3576 			.emptiness = allocator->summary_entries[slab_number].fullness_hint,
3577 		};
3578 	}
3579 
3580 	return VDO_SUCCESS;
3581 }
3582 
3583 /* Prepare slabs for allocation or scrubbing. */
vdo_prepare_slabs_for_allocation(struct block_allocator * allocator)3584 static int __must_check vdo_prepare_slabs_for_allocation(struct block_allocator *allocator)
3585 {
3586 	struct slab_status current_slab_status;
3587 	DEFINE_MIN_HEAP(struct slab_status, heap) heap;
3588 	int result;
3589 	struct slab_status *slab_statuses;
3590 	struct slab_depot *depot = allocator->depot;
3591 
3592 	WRITE_ONCE(allocator->allocated_blocks,
3593 		   allocator->slab_count * depot->slab_config.data_blocks);
3594 	result = get_slab_statuses(allocator, &slab_statuses);
3595 	if (result != VDO_SUCCESS)
3596 		return result;
3597 
3598 	/* Sort the slabs by cleanliness, then by emptiness hint. */
3599 	heap = (struct heap) {
3600 		.data = slab_statuses,
3601 		.nr = allocator->slab_count,
3602 		.size = allocator->slab_count,
3603 	};
3604 	min_heapify_all(&heap, &slab_status_min_heap, NULL);
3605 
3606 	while (heap.nr > 0) {
3607 		bool high_priority;
3608 		struct vdo_slab *slab;
3609 		struct slab_journal *journal;
3610 
3611 		current_slab_status = slab_statuses[0];
3612 		min_heap_pop(&heap, &slab_status_min_heap, NULL);
3613 		slab = depot->slabs[current_slab_status.slab_number];
3614 
3615 		if ((depot->load_type == VDO_SLAB_DEPOT_REBUILD_LOAD) ||
3616 		    (!allocator->summary_entries[slab->slab_number].load_ref_counts &&
3617 		     current_slab_status.is_clean)) {
3618 			queue_slab(slab);
3619 			continue;
3620 		}
3621 
3622 		slab->status = VDO_SLAB_REQUIRES_SCRUBBING;
3623 		journal = &slab->journal;
3624 		high_priority = ((current_slab_status.is_clean &&
3625 				 (depot->load_type == VDO_SLAB_DEPOT_NORMAL_LOAD)) ||
3626 				 (journal_length(journal) >= journal->scrubbing_threshold));
3627 		register_slab_for_scrubbing(slab, high_priority);
3628 	}
3629 
3630 	vdo_free(slab_statuses);
3631 	return VDO_SUCCESS;
3632 }
3633 
status_to_string(enum slab_rebuild_status status)3634 static const char *status_to_string(enum slab_rebuild_status status)
3635 {
3636 	switch (status) {
3637 	case VDO_SLAB_REBUILT:
3638 		return "REBUILT";
3639 	case VDO_SLAB_REQUIRES_SCRUBBING:
3640 		return "SCRUBBING";
3641 	case VDO_SLAB_REQUIRES_HIGH_PRIORITY_SCRUBBING:
3642 		return "PRIORITY_SCRUBBING";
3643 	case VDO_SLAB_REBUILDING:
3644 		return "REBUILDING";
3645 	case VDO_SLAB_REPLAYING:
3646 		return "REPLAYING";
3647 	default:
3648 		return "UNKNOWN";
3649 	}
3650 }
3651 
vdo_dump_block_allocator(const struct block_allocator * allocator)3652 void vdo_dump_block_allocator(const struct block_allocator *allocator)
3653 {
3654 	unsigned int pause_counter = 0;
3655 	struct slab_iterator iterator = get_slab_iterator(allocator);
3656 	const struct slab_scrubber *scrubber = &allocator->scrubber;
3657 
3658 	vdo_log_info("block_allocator zone %u", allocator->zone_number);
3659 	while (iterator.next != NULL) {
3660 		struct vdo_slab *slab = next_slab(&iterator);
3661 		struct slab_journal *journal = &slab->journal;
3662 
3663 		if (slab->reference_blocks != NULL) {
3664 			/* Terse because there are a lot of slabs to dump and syslog is lossy. */
3665 			vdo_log_info("slab %u: P%u, %llu free", slab->slab_number,
3666 				     slab->priority,
3667 				     (unsigned long long) slab->free_blocks);
3668 		} else {
3669 			vdo_log_info("slab %u: status %s", slab->slab_number,
3670 				     status_to_string(slab->status));
3671 		}
3672 
3673 		vdo_log_info("  slab journal: entry_waiters=%zu waiting_to_commit=%s updating_slab_summary=%s head=%llu unreapable=%llu tail=%llu next_commit=%llu summarized=%llu last_summarized=%llu recovery_lock=%llu dirty=%s",
3674 			     vdo_waitq_num_waiters(&journal->entry_waiters),
3675 			     vdo_bool_to_string(journal->waiting_to_commit),
3676 			     vdo_bool_to_string(journal->updating_slab_summary),
3677 			     (unsigned long long) journal->head,
3678 			     (unsigned long long) journal->unreapable,
3679 			     (unsigned long long) journal->tail,
3680 			     (unsigned long long) journal->next_commit,
3681 			     (unsigned long long) journal->summarized,
3682 			     (unsigned long long) journal->last_summarized,
3683 			     (unsigned long long) journal->recovery_lock,
3684 			     vdo_bool_to_string(journal->recovery_lock != 0));
3685 		/*
3686 		 * Given the frequency with which the locks are just a tiny bit off, it might be
3687 		 * worth dumping all the locks, but that might be too much logging.
3688 		 */
3689 
3690 		if (slab->counters != NULL) {
3691 			/* Terse because there are a lot of slabs to dump and syslog is lossy. */
3692 			vdo_log_info("  slab: free=%u/%u blocks=%u dirty=%zu active=%zu journal@(%llu,%u)",
3693 				     slab->free_blocks, slab->block_count,
3694 				     slab->reference_block_count,
3695 				     vdo_waitq_num_waiters(&slab->dirty_blocks),
3696 				     slab->active_count,
3697 				     (unsigned long long) slab->slab_journal_point.sequence_number,
3698 				     slab->slab_journal_point.entry_count);
3699 		} else {
3700 			vdo_log_info("  no counters");
3701 		}
3702 
3703 		/*
3704 		 * Wait for a while after each batch of 32 slabs dumped, an arbitrary number,
3705 		 * allowing the kernel log a chance to be flushed instead of being overrun.
3706 		 */
3707 		if (pause_counter++ == 31) {
3708 			pause_counter = 0;
3709 			vdo_pause_for_logger();
3710 		}
3711 	}
3712 
3713 	vdo_log_info("slab_scrubber slab_count %u waiters %zu %s%s",
3714 		     READ_ONCE(scrubber->slab_count),
3715 		     vdo_waitq_num_waiters(&scrubber->waiters),
3716 		     vdo_get_admin_state_code(&scrubber->admin_state)->name,
3717 		     scrubber->high_priority_only ? ", high_priority_only " : "");
3718 }
3719 
free_slab(struct vdo_slab * slab)3720 static void free_slab(struct vdo_slab *slab)
3721 {
3722 	if (slab == NULL)
3723 		return;
3724 
3725 	list_del(&slab->allocq_entry);
3726 	vdo_free(vdo_forget(slab->journal.block));
3727 	vdo_free(vdo_forget(slab->journal.locks));
3728 	vdo_free(vdo_forget(slab->counters));
3729 	vdo_free(vdo_forget(slab->reference_blocks));
3730 	vdo_free(slab);
3731 }
3732 
initialize_slab_journal(struct vdo_slab * slab)3733 static int initialize_slab_journal(struct vdo_slab *slab)
3734 {
3735 	struct slab_journal *journal = &slab->journal;
3736 	const struct slab_config *slab_config = &slab->allocator->depot->slab_config;
3737 	int result;
3738 
3739 	result = vdo_allocate(slab_config->slab_journal_blocks, __func__, &journal->locks);
3740 	if (result != VDO_SUCCESS)
3741 		return result;
3742 
3743 	BUILD_BUG_ON(sizeof(*journal->block) != VDO_BLOCK_SIZE);
3744 	result = vdo_allocate(1, "struct packed_slab_journal_block", &journal->block);
3745 	if (result != VDO_SUCCESS)
3746 		return result;
3747 
3748 	journal->slab = slab;
3749 	journal->size = slab_config->slab_journal_blocks;
3750 	journal->flushing_threshold = slab_config->slab_journal_flushing_threshold;
3751 	journal->blocking_threshold = slab_config->slab_journal_blocking_threshold;
3752 	journal->scrubbing_threshold = slab_config->slab_journal_scrubbing_threshold;
3753 	journal->entries_per_block = VDO_SLAB_JOURNAL_ENTRIES_PER_BLOCK;
3754 	journal->full_entries_per_block = VDO_SLAB_JOURNAL_FULL_ENTRIES_PER_BLOCK;
3755 	journal->events = &slab->allocator->slab_journal_statistics;
3756 	journal->recovery_journal = slab->allocator->depot->vdo->recovery_journal;
3757 	journal->tail = 1;
3758 	journal->head = 1;
3759 
3760 	journal->flushing_deadline = journal->flushing_threshold;
3761 	/*
3762 	 * Set there to be some time between the deadline and the blocking threshold, so that
3763 	 * hopefully all are done before blocking.
3764 	 */
3765 	if ((journal->blocking_threshold - journal->flushing_threshold) > 5)
3766 		journal->flushing_deadline = journal->blocking_threshold - 5;
3767 
3768 	journal->slab_summary_waiter.callback = release_journal_locks;
3769 
3770 	INIT_LIST_HEAD(&journal->dirty_entry);
3771 	INIT_LIST_HEAD(&journal->uncommitted_blocks);
3772 
3773 	journal->tail_header.nonce = slab->allocator->nonce;
3774 	journal->tail_header.metadata_type = VDO_METADATA_SLAB_JOURNAL;
3775 	initialize_journal_state(journal);
3776 	return VDO_SUCCESS;
3777 }
3778 
3779 /**
3780  * make_slab() - Construct a new, empty slab.
3781  * @slab_origin: The physical block number within the block allocator partition of the first block
3782  *               in the slab.
3783  * @allocator: The block allocator to which the slab belongs.
3784  * @slab_number: The slab number of the slab.
3785  * @is_new: True if this slab is being allocated as part of a resize.
3786  * @slab_ptr: A pointer to receive the new slab.
3787  *
3788  * Return: VDO_SUCCESS or an error code.
3789  */
make_slab(physical_block_number_t slab_origin,struct block_allocator * allocator,slab_count_t slab_number,bool is_new,struct vdo_slab ** slab_ptr)3790 static int __must_check make_slab(physical_block_number_t slab_origin,
3791 				  struct block_allocator *allocator,
3792 				  slab_count_t slab_number, bool is_new,
3793 				  struct vdo_slab **slab_ptr)
3794 {
3795 	const struct slab_config *slab_config = &allocator->depot->slab_config;
3796 	struct vdo_slab *slab;
3797 	int result;
3798 
3799 	result = vdo_allocate(1, __func__, &slab);
3800 	if (result != VDO_SUCCESS)
3801 		return result;
3802 
3803 	*slab = (struct vdo_slab) {
3804 		.allocator = allocator,
3805 		.start = slab_origin,
3806 		.end = slab_origin + slab_config->slab_blocks,
3807 		.slab_number = slab_number,
3808 		.ref_counts_origin = slab_origin + slab_config->data_blocks,
3809 		.journal_origin =
3810 			vdo_get_slab_journal_start_block(slab_config, slab_origin),
3811 		.block_count = slab_config->data_blocks,
3812 		.free_blocks = slab_config->data_blocks,
3813 		.reference_block_count =
3814 			vdo_get_saved_reference_count_size(slab_config->data_blocks),
3815 	};
3816 	INIT_LIST_HEAD(&slab->allocq_entry);
3817 
3818 	result = initialize_slab_journal(slab);
3819 	if (result != VDO_SUCCESS) {
3820 		free_slab(slab);
3821 		return result;
3822 	}
3823 
3824 	if (is_new) {
3825 		vdo_set_admin_state_code(&slab->state, VDO_ADMIN_STATE_NEW);
3826 		result = allocate_slab_counters(slab);
3827 		if (result != VDO_SUCCESS) {
3828 			free_slab(slab);
3829 			return result;
3830 		}
3831 	} else {
3832 		vdo_set_admin_state_code(&slab->state, VDO_ADMIN_STATE_NORMAL_OPERATION);
3833 	}
3834 
3835 	*slab_ptr = slab;
3836 	return VDO_SUCCESS;
3837 }
3838 
3839 /**
3840  * allocate_slabs() - Allocate a new slab pointer array.
3841  * @depot: The depot.
3842  * @slab_count: The number of slabs the depot should have in the new array.
3843  *
3844  * Any existing slab pointers will be copied into the new array, and slabs will be allocated as
3845  * needed. The newly allocated slabs will not be distributed for use by the block allocators.
3846  *
3847  * Return: VDO_SUCCESS or an error code.
3848  */
allocate_slabs(struct slab_depot * depot,slab_count_t slab_count)3849 static int allocate_slabs(struct slab_depot *depot, slab_count_t slab_count)
3850 {
3851 	block_count_t slab_size;
3852 	bool resizing = false;
3853 	physical_block_number_t slab_origin;
3854 	int result;
3855 
3856 	result = vdo_allocate(slab_count, "slab pointer array", &depot->new_slabs);
3857 	if (result != VDO_SUCCESS)
3858 		return result;
3859 
3860 	if (depot->slabs != NULL) {
3861 		memcpy(depot->new_slabs, depot->slabs,
3862 		       depot->slab_count * sizeof(struct vdo_slab *));
3863 		resizing = true;
3864 	}
3865 
3866 	slab_size = depot->slab_config.slab_blocks;
3867 	slab_origin = depot->first_block + (depot->slab_count * slab_size);
3868 
3869 	for (depot->new_slab_count = depot->slab_count;
3870 	     depot->new_slab_count < slab_count;
3871 	     depot->new_slab_count++, slab_origin += slab_size) {
3872 		struct block_allocator *allocator =
3873 			&depot->allocators[depot->new_slab_count % depot->zone_count];
3874 		struct vdo_slab **slab_ptr = &depot->new_slabs[depot->new_slab_count];
3875 
3876 		result = make_slab(slab_origin, allocator, depot->new_slab_count,
3877 				   resizing, slab_ptr);
3878 		if (result != VDO_SUCCESS)
3879 			return result;
3880 	}
3881 
3882 	return VDO_SUCCESS;
3883 }
3884 
3885 /**
3886  * vdo_abandon_new_slabs() - Abandon any new slabs in this depot, freeing them as needed.
3887  * @depot: The depot.
3888  */
vdo_abandon_new_slabs(struct slab_depot * depot)3889 void vdo_abandon_new_slabs(struct slab_depot *depot)
3890 {
3891 	slab_count_t i;
3892 
3893 	if (depot->new_slabs == NULL)
3894 		return;
3895 
3896 	for (i = depot->slab_count; i < depot->new_slab_count; i++)
3897 		free_slab(vdo_forget(depot->new_slabs[i]));
3898 	depot->new_slab_count = 0;
3899 	depot->new_size = 0;
3900 	vdo_free(vdo_forget(depot->new_slabs));
3901 }
3902 
3903 /** Implements vdo_zone_thread_getter_fn. */
get_allocator_thread_id(void * context,zone_count_t zone_number)3904 static thread_id_t get_allocator_thread_id(void *context, zone_count_t zone_number)
3905 {
3906 	return ((struct slab_depot *) context)->allocators[zone_number].thread_id;
3907 }
3908 
3909 /**
3910  * release_recovery_journal_lock() - Request the slab journal to release the recovery journal lock
3911  *                                   it may hold on a specified recovery journal block.
3912  * @journal: The slab journal.
3913  * @recovery_lock: The sequence number of the recovery journal block whose locks should be
3914  *                 released.
3915  *
3916  * Return: True if the journal released a lock on the specified block.
3917  */
release_recovery_journal_lock(struct slab_journal * journal,sequence_number_t recovery_lock)3918 static bool __must_check release_recovery_journal_lock(struct slab_journal *journal,
3919 						       sequence_number_t recovery_lock)
3920 {
3921 	if (recovery_lock > journal->recovery_lock) {
3922 		VDO_ASSERT_LOG_ONLY((recovery_lock < journal->recovery_lock),
3923 				    "slab journal recovery lock is not older than the recovery journal head");
3924 		return false;
3925 	}
3926 
3927 	if ((recovery_lock < journal->recovery_lock) ||
3928 	    vdo_is_read_only(journal->slab->allocator->depot->vdo))
3929 		return false;
3930 
3931 	/* All locks are held by the block which is in progress; write it. */
3932 	commit_tail(journal);
3933 	return true;
3934 }
3935 
3936 /*
3937  * Request a commit of all dirty tail blocks which are locking the recovery journal block the depot
3938  * is seeking to release.
3939  *
3940  * Implements vdo_zone_action_fn.
3941  */
release_tail_block_locks(void * context,zone_count_t zone_number,struct vdo_completion * parent)3942 static void release_tail_block_locks(void *context, zone_count_t zone_number,
3943 				     struct vdo_completion *parent)
3944 {
3945 	struct slab_journal *journal, *tmp;
3946 	struct slab_depot *depot = context;
3947 	struct list_head *list = &depot->allocators[zone_number].dirty_slab_journals;
3948 
3949 	list_for_each_entry_safe(journal, tmp, list, dirty_entry) {
3950 		if (!release_recovery_journal_lock(journal,
3951 						   depot->active_release_request))
3952 			break;
3953 	}
3954 
3955 	vdo_finish_completion(parent);
3956 }
3957 
3958 /**
3959  * prepare_for_tail_block_commit() - Prepare to commit oldest tail blocks.
3960  * @context: The slab depot.
3961  * @parent: The parent operation.
3962  *
3963  * Implements vdo_action_preamble_fn.
3964  */
prepare_for_tail_block_commit(void * context,struct vdo_completion * parent)3965 static void prepare_for_tail_block_commit(void *context, struct vdo_completion *parent)
3966 {
3967 	struct slab_depot *depot = context;
3968 
3969 	depot->active_release_request = depot->new_release_request;
3970 	vdo_finish_completion(parent);
3971 }
3972 
3973 /**
3974  * schedule_tail_block_commit() - Schedule a tail block commit if necessary.
3975  * @context: The slab depot.
3976  *
3977  * This method should not be called directly. Rather, call vdo_schedule_default_action() on the
3978  * depot's action manager.
3979  *
3980  * Implements vdo_action_scheduler_fn.
3981  */
schedule_tail_block_commit(void * context)3982 static bool schedule_tail_block_commit(void *context)
3983 {
3984 	struct slab_depot *depot = context;
3985 
3986 	if (depot->new_release_request == depot->active_release_request)
3987 		return false;
3988 
3989 	return vdo_schedule_action(depot->action_manager,
3990 				   prepare_for_tail_block_commit,
3991 				   release_tail_block_locks,
3992 				   NULL, NULL);
3993 }
3994 
3995 /**
3996  * initialize_slab_scrubber() - Initialize an allocator's slab scrubber.
3997  * @allocator: The allocator being initialized
3998  *
3999  * Return: VDO_SUCCESS or an error.
4000  */
initialize_slab_scrubber(struct block_allocator * allocator)4001 static int initialize_slab_scrubber(struct block_allocator *allocator)
4002 {
4003 	struct slab_scrubber *scrubber = &allocator->scrubber;
4004 	block_count_t slab_journal_size =
4005 		allocator->depot->slab_config.slab_journal_blocks;
4006 	char *journal_data;
4007 	int result;
4008 
4009 	result = vdo_allocate(VDO_BLOCK_SIZE * slab_journal_size, __func__, &journal_data);
4010 	if (result != VDO_SUCCESS)
4011 		return result;
4012 
4013 	result = allocate_vio_components(allocator->completion.vdo,
4014 					 VIO_TYPE_SLAB_JOURNAL,
4015 					 VIO_PRIORITY_METADATA,
4016 					 allocator, slab_journal_size,
4017 					 journal_data, &scrubber->vio);
4018 	if (result != VDO_SUCCESS) {
4019 		vdo_free(journal_data);
4020 		return result;
4021 	}
4022 
4023 	INIT_LIST_HEAD(&scrubber->high_priority_slabs);
4024 	INIT_LIST_HEAD(&scrubber->slabs);
4025 	vdo_set_admin_state_code(&scrubber->admin_state, VDO_ADMIN_STATE_SUSPENDED);
4026 	return VDO_SUCCESS;
4027 }
4028 
4029 /**
4030  * initialize_slab_summary_block() - Initialize a slab_summary_block.
4031  * @allocator: The allocator which owns the block.
4032  * @index: The index of this block in its zone's summary.
4033  *
4034  * Return: VDO_SUCCESS or an error.
4035  */
initialize_slab_summary_block(struct block_allocator * allocator,block_count_t index)4036 static int __must_check initialize_slab_summary_block(struct block_allocator *allocator,
4037 						      block_count_t index)
4038 {
4039 	struct slab_summary_block *block = &allocator->summary_blocks[index];
4040 	int result;
4041 
4042 	result = vdo_allocate(VDO_BLOCK_SIZE, __func__, &block->outgoing_entries);
4043 	if (result != VDO_SUCCESS)
4044 		return result;
4045 
4046 	result = allocate_vio_components(allocator->depot->vdo, VIO_TYPE_SLAB_SUMMARY,
4047 					 VIO_PRIORITY_METADATA, NULL, 1,
4048 					 block->outgoing_entries, &block->vio);
4049 	if (result != VDO_SUCCESS)
4050 		return result;
4051 
4052 	block->allocator = allocator;
4053 	block->entries = &allocator->summary_entries[VDO_SLAB_SUMMARY_ENTRIES_PER_BLOCK * index];
4054 	block->index = index;
4055 	return VDO_SUCCESS;
4056 }
4057 
initialize_block_allocator(struct slab_depot * depot,zone_count_t zone)4058 static int __must_check initialize_block_allocator(struct slab_depot *depot,
4059 						   zone_count_t zone)
4060 {
4061 	int result;
4062 	block_count_t i;
4063 	struct block_allocator *allocator = &depot->allocators[zone];
4064 	struct vdo *vdo = depot->vdo;
4065 	block_count_t max_free_blocks = depot->slab_config.data_blocks;
4066 	unsigned int max_priority = (2 + ilog2(max_free_blocks));
4067 	u32 reference_block_count, refcount_reads_needed, refcount_blocks_per_vio;
4068 
4069 	*allocator = (struct block_allocator) {
4070 		.depot = depot,
4071 		.zone_number = zone,
4072 		.thread_id = vdo->thread_config.physical_threads[zone],
4073 		.nonce = vdo->states.vdo.nonce,
4074 	};
4075 
4076 	INIT_LIST_HEAD(&allocator->dirty_slab_journals);
4077 	vdo_set_admin_state_code(&allocator->state, VDO_ADMIN_STATE_NORMAL_OPERATION);
4078 	result = vdo_register_read_only_listener(vdo, allocator,
4079 						 notify_block_allocator_of_read_only_mode,
4080 						 allocator->thread_id);
4081 	if (result != VDO_SUCCESS)
4082 		return result;
4083 
4084 	vdo_initialize_completion(&allocator->completion, vdo, VDO_BLOCK_ALLOCATOR_COMPLETION);
4085 	result = make_vio_pool(vdo, BLOCK_ALLOCATOR_VIO_POOL_SIZE, 1, allocator->thread_id,
4086 			       VIO_TYPE_SLAB_JOURNAL, VIO_PRIORITY_METADATA,
4087 			       allocator, &allocator->vio_pool);
4088 	if (result != VDO_SUCCESS)
4089 		return result;
4090 
4091 	/* Initialize the refcount-reading vio pool. */
4092 	reference_block_count = vdo_get_saved_reference_count_size(depot->slab_config.slab_blocks);
4093 	refcount_reads_needed = DIV_ROUND_UP(reference_block_count, MAX_BLOCKS_PER_VIO);
4094 	refcount_blocks_per_vio = DIV_ROUND_UP(reference_block_count, refcount_reads_needed);
4095 	allocator->refcount_blocks_per_big_vio = refcount_blocks_per_vio;
4096 	result = make_vio_pool(vdo, BLOCK_ALLOCATOR_REFCOUNT_VIO_POOL_SIZE,
4097 			       allocator->refcount_blocks_per_big_vio, allocator->thread_id,
4098 			       VIO_TYPE_SLAB_JOURNAL, VIO_PRIORITY_METADATA,
4099 			       NULL, &allocator->refcount_big_vio_pool);
4100 	if (result != VDO_SUCCESS)
4101 		return result;
4102 
4103 	result = initialize_slab_scrubber(allocator);
4104 	if (result != VDO_SUCCESS)
4105 		return result;
4106 
4107 	result = vdo_make_priority_table(max_priority, &allocator->prioritized_slabs);
4108 	if (result != VDO_SUCCESS)
4109 		return result;
4110 
4111 	result = vdo_allocate(VDO_SLAB_SUMMARY_BLOCKS_PER_ZONE, __func__,
4112 			      &allocator->summary_blocks);
4113 	if (result != VDO_SUCCESS)
4114 		return result;
4115 
4116 	vdo_set_admin_state_code(&allocator->summary_state,
4117 				 VDO_ADMIN_STATE_NORMAL_OPERATION);
4118 	allocator->summary_entries = depot->summary_entries + (MAX_VDO_SLABS * zone);
4119 
4120 	/* Initialize each summary block. */
4121 	for (i = 0; i < VDO_SLAB_SUMMARY_BLOCKS_PER_ZONE; i++) {
4122 		result = initialize_slab_summary_block(allocator, i);
4123 		if (result != VDO_SUCCESS)
4124 			return result;
4125 	}
4126 
4127 	/*
4128 	 * Performing well atop thin provisioned storage requires either that VDO discards freed
4129 	 * blocks, or that the block allocator try to use slabs that already have allocated blocks
4130 	 * in preference to slabs that have never been opened. For reasons we have not been able to
4131 	 * fully understand, some SSD machines have been have been very sensitive (50% reduction in
4132 	 * test throughput) to very slight differences in the timing and locality of block
4133 	 * allocation. Assigning a low priority to unopened slabs (max_priority/2, say) would be
4134 	 * ideal for the story, but anything less than a very high threshold (max_priority - 1)
4135 	 * hurts on these machines.
4136 	 *
4137 	 * This sets the free block threshold for preferring to open an unopened slab to the binary
4138 	 * floor of 3/4ths the total number of data blocks in a slab, which will generally evaluate
4139 	 * to about half the slab size.
4140 	 */
4141 	allocator->unopened_slab_priority = (1 + ilog2((max_free_blocks * 3) / 4));
4142 
4143 	return VDO_SUCCESS;
4144 }
4145 
allocate_components(struct slab_depot * depot,struct partition * summary_partition)4146 static int allocate_components(struct slab_depot *depot,
4147 			       struct partition *summary_partition)
4148 {
4149 	int result;
4150 	zone_count_t zone;
4151 	slab_count_t slab_count;
4152 	u8 hint;
4153 	u32 i;
4154 	const struct thread_config *thread_config = &depot->vdo->thread_config;
4155 
4156 	result = vdo_make_action_manager(depot->zone_count, get_allocator_thread_id,
4157 					 thread_config->journal_thread, depot,
4158 					 schedule_tail_block_commit,
4159 					 depot->vdo, &depot->action_manager);
4160 	if (result != VDO_SUCCESS)
4161 		return result;
4162 
4163 	depot->origin = depot->first_block;
4164 
4165 	/* block size must be a multiple of entry size */
4166 	BUILD_BUG_ON((VDO_BLOCK_SIZE % sizeof(struct slab_summary_entry)) != 0);
4167 
4168 	depot->summary_origin = summary_partition->offset;
4169 	depot->hint_shift = vdo_get_slab_summary_hint_shift(depot->slab_size_shift);
4170 	result = vdo_allocate(MAXIMUM_VDO_SLAB_SUMMARY_ENTRIES, __func__,
4171 			      &depot->summary_entries);
4172 	if (result != VDO_SUCCESS)
4173 		return result;
4174 
4175 
4176 	/* Initialize all the entries. */
4177 	hint = compute_fullness_hint(depot, depot->slab_config.data_blocks);
4178 	for (i = 0; i < MAXIMUM_VDO_SLAB_SUMMARY_ENTRIES; i++) {
4179 		/*
4180 		 * This default tail block offset must be reflected in
4181 		 * slabJournal.c::read_slab_journal_tail().
4182 		 */
4183 		depot->summary_entries[i] = (struct slab_summary_entry) {
4184 			.tail_block_offset = 0,
4185 			.fullness_hint = hint,
4186 			.load_ref_counts = false,
4187 			.is_dirty = false,
4188 		};
4189 	}
4190 
4191 	slab_count = vdo_compute_slab_count(depot->first_block, depot->last_block,
4192 					    depot->slab_size_shift);
4193 	if (thread_config->physical_zone_count > slab_count) {
4194 		return vdo_log_error_strerror(VDO_BAD_CONFIGURATION,
4195 					      "%u physical zones exceeds slab count %u",
4196 					      thread_config->physical_zone_count,
4197 					      slab_count);
4198 	}
4199 
4200 	/* Initialize the block allocators. */
4201 	for (zone = 0; zone < depot->zone_count; zone++) {
4202 		result = initialize_block_allocator(depot, zone);
4203 		if (result != VDO_SUCCESS)
4204 			return result;
4205 	}
4206 
4207 	/* Allocate slabs. */
4208 	result = allocate_slabs(depot, slab_count);
4209 	if (result != VDO_SUCCESS)
4210 		return result;
4211 
4212 	/* Use the new slabs. */
4213 	for (i = depot->slab_count; i < depot->new_slab_count; i++) {
4214 		struct vdo_slab *slab = depot->new_slabs[i];
4215 
4216 		register_slab_with_allocator(slab->allocator, slab);
4217 		WRITE_ONCE(depot->slab_count, depot->slab_count + 1);
4218 	}
4219 
4220 	depot->slabs = depot->new_slabs;
4221 	depot->new_slabs = NULL;
4222 	depot->new_slab_count = 0;
4223 
4224 	return VDO_SUCCESS;
4225 }
4226 
4227 /**
4228  * vdo_decode_slab_depot() - Make a slab depot and configure it with the state read from the super
4229  *                           block.
4230  * @state: The slab depot state from the super block.
4231  * @vdo: The VDO which will own the depot.
4232  * @summary_partition: The partition which holds the slab summary.
4233  * @depot_ptr: A pointer to hold the depot.
4234  *
4235  * Return: A success or error code.
4236  */
vdo_decode_slab_depot(struct slab_depot_state_2_0 state,struct vdo * vdo,struct partition * summary_partition,struct slab_depot ** depot_ptr)4237 int vdo_decode_slab_depot(struct slab_depot_state_2_0 state, struct vdo *vdo,
4238 			  struct partition *summary_partition,
4239 			  struct slab_depot **depot_ptr)
4240 {
4241 	unsigned int slab_size_shift;
4242 	struct slab_depot *depot;
4243 	int result;
4244 
4245 	/*
4246 	 * Calculate the bit shift for efficiently mapping block numbers to slabs. Using a shift
4247 	 * requires that the slab size be a power of two.
4248 	 */
4249 	block_count_t slab_size = state.slab_config.slab_blocks;
4250 
4251 	if (!is_power_of_2(slab_size)) {
4252 		return vdo_log_error_strerror(UDS_INVALID_ARGUMENT,
4253 					      "slab size must be a power of two");
4254 	}
4255 	slab_size_shift = ilog2(slab_size);
4256 
4257 	if (state.zone_count > MAX_VDO_PHYSICAL_ZONES)
4258 		return vdo_log_error_strerror(UDS_CORRUPT_DATA,
4259 					      "invalid zone count");
4260 
4261 	result = vdo_allocate_extended(vdo->thread_config.physical_zone_count,
4262 				       allocators, __func__, &depot);
4263 	if (result != VDO_SUCCESS)
4264 		return result;
4265 
4266 	depot->vdo = vdo;
4267 	depot->old_zone_count = state.zone_count;
4268 	depot->zone_count = vdo->thread_config.physical_zone_count;
4269 	depot->slab_config = state.slab_config;
4270 	depot->first_block = state.first_block;
4271 	depot->last_block = state.last_block;
4272 	depot->slab_size_shift = slab_size_shift;
4273 
4274 	result = allocate_components(depot, summary_partition);
4275 	if (result != VDO_SUCCESS) {
4276 		vdo_free_slab_depot(depot);
4277 		return result;
4278 	}
4279 
4280 	*depot_ptr = depot;
4281 	return VDO_SUCCESS;
4282 }
4283 
uninitialize_allocator_summary(struct block_allocator * allocator)4284 static void uninitialize_allocator_summary(struct block_allocator *allocator)
4285 {
4286 	block_count_t i;
4287 
4288 	if (allocator->summary_blocks == NULL)
4289 		return;
4290 
4291 	for (i = 0; i < VDO_SLAB_SUMMARY_BLOCKS_PER_ZONE; i++) {
4292 		free_vio_components(&allocator->summary_blocks[i].vio);
4293 		vdo_free(vdo_forget(allocator->summary_blocks[i].outgoing_entries));
4294 	}
4295 
4296 	vdo_free(vdo_forget(allocator->summary_blocks));
4297 }
4298 
4299 /**
4300  * vdo_free_slab_depot() - Destroy a slab depot.
4301  * @depot: The depot to destroy.
4302  */
vdo_free_slab_depot(struct slab_depot * depot)4303 void vdo_free_slab_depot(struct slab_depot *depot)
4304 {
4305 	zone_count_t zone = 0;
4306 
4307 	if (depot == NULL)
4308 		return;
4309 
4310 	vdo_abandon_new_slabs(depot);
4311 
4312 	for (zone = 0; zone < depot->zone_count; zone++) {
4313 		struct block_allocator *allocator = &depot->allocators[zone];
4314 
4315 		if (allocator->eraser != NULL)
4316 			dm_kcopyd_client_destroy(vdo_forget(allocator->eraser));
4317 
4318 		uninitialize_allocator_summary(allocator);
4319 		uninitialize_scrubber_vio(&allocator->scrubber);
4320 		free_vio_pool(vdo_forget(allocator->vio_pool));
4321 		free_vio_pool(vdo_forget(allocator->refcount_big_vio_pool));
4322 		vdo_free_priority_table(vdo_forget(allocator->prioritized_slabs));
4323 	}
4324 
4325 	if (depot->slabs != NULL) {
4326 		slab_count_t i;
4327 
4328 		for (i = 0; i < depot->slab_count; i++)
4329 			free_slab(vdo_forget(depot->slabs[i]));
4330 	}
4331 
4332 	vdo_free(vdo_forget(depot->slabs));
4333 	vdo_free(vdo_forget(depot->action_manager));
4334 	vdo_free(vdo_forget(depot->summary_entries));
4335 	vdo_free(depot);
4336 }
4337 
4338 /**
4339  * vdo_record_slab_depot() - Record the state of a slab depot for encoding into the super block.
4340  * @depot: The depot to encode.
4341  *
4342  * Return: The depot state.
4343  */
vdo_record_slab_depot(const struct slab_depot * depot)4344 struct slab_depot_state_2_0 vdo_record_slab_depot(const struct slab_depot *depot)
4345 {
4346 	/*
4347 	 * If this depot is currently using 0 zones, it must have been synchronously loaded by a
4348 	 * tool and is now being saved. We did not load and combine the slab summary, so we still
4349 	 * need to do that next time we load with the old zone count rather than 0.
4350 	 */
4351 	struct slab_depot_state_2_0 state;
4352 	zone_count_t zones_to_record = depot->zone_count;
4353 
4354 	if (depot->zone_count == 0)
4355 		zones_to_record = depot->old_zone_count;
4356 
4357 	state = (struct slab_depot_state_2_0) {
4358 		.slab_config = depot->slab_config,
4359 		.first_block = depot->first_block,
4360 		.last_block = depot->last_block,
4361 		.zone_count = zones_to_record,
4362 	};
4363 
4364 	return state;
4365 }
4366 
4367 /**
4368  * vdo_allocate_reference_counters() - Allocate the reference counters for all slabs in the depot.
4369  * @depot: The slab depot.
4370  *
4371  * Context: This method may be called only before entering normal operation from the load thread.
4372  *
4373  * Return: VDO_SUCCESS or an error.
4374  */
vdo_allocate_reference_counters(struct slab_depot * depot)4375 int vdo_allocate_reference_counters(struct slab_depot *depot)
4376 {
4377 	struct slab_iterator iterator =
4378 		get_depot_slab_iterator(depot, depot->slab_count - 1, 0, 1);
4379 
4380 	while (iterator.next != NULL) {
4381 		int result = allocate_slab_counters(next_slab(&iterator));
4382 
4383 		if (result != VDO_SUCCESS)
4384 			return result;
4385 	}
4386 
4387 	return VDO_SUCCESS;
4388 }
4389 
4390 /**
4391  * get_slab_number() - Get the number of the slab that contains a specified block.
4392  * @depot: The slab depot.
4393  * @pbn: The physical block number.
4394  * @slab_number_ptr: A pointer to hold the slab number.
4395  *
4396  * Return: VDO_SUCCESS or an error.
4397  */
get_slab_number(const struct slab_depot * depot,physical_block_number_t pbn,slab_count_t * slab_number_ptr)4398 static int __must_check get_slab_number(const struct slab_depot *depot,
4399 					physical_block_number_t pbn,
4400 					slab_count_t *slab_number_ptr)
4401 {
4402 	slab_count_t slab_number;
4403 
4404 	if (pbn < depot->first_block)
4405 		return VDO_OUT_OF_RANGE;
4406 
4407 	slab_number = (pbn - depot->first_block) >> depot->slab_size_shift;
4408 	if (slab_number >= depot->slab_count)
4409 		return VDO_OUT_OF_RANGE;
4410 
4411 	*slab_number_ptr = slab_number;
4412 	return VDO_SUCCESS;
4413 }
4414 
4415 /**
4416  * vdo_get_slab() - Get the slab object for the slab that contains a specified block.
4417  * @depot: The slab depot.
4418  * @pbn: The physical block number.
4419  *
4420  * Will put the VDO in read-only mode if the PBN is not a valid data block nor the zero block.
4421  *
4422  * Return: The slab containing the block, or NULL if the block number is the zero block or
4423  * otherwise out of range.
4424  */
vdo_get_slab(const struct slab_depot * depot,physical_block_number_t pbn)4425 struct vdo_slab *vdo_get_slab(const struct slab_depot *depot,
4426 			      physical_block_number_t pbn)
4427 {
4428 	slab_count_t slab_number;
4429 	int result;
4430 
4431 	if (pbn == VDO_ZERO_BLOCK)
4432 		return NULL;
4433 
4434 	result = get_slab_number(depot, pbn, &slab_number);
4435 	if (result != VDO_SUCCESS) {
4436 		vdo_enter_read_only_mode(depot->vdo, result);
4437 		return NULL;
4438 	}
4439 
4440 	return depot->slabs[slab_number];
4441 }
4442 
4443 /**
4444  * vdo_get_increment_limit() - Determine how many new references a block can acquire.
4445  * @depot: The slab depot.
4446  * @pbn: The physical block number that is being queried.
4447  *
4448  * Context: This method must be called from the physical zone thread of the PBN.
4449  *
4450  * Return: The number of available references.
4451  */
vdo_get_increment_limit(struct slab_depot * depot,physical_block_number_t pbn)4452 u8 vdo_get_increment_limit(struct slab_depot *depot, physical_block_number_t pbn)
4453 {
4454 	struct vdo_slab *slab = vdo_get_slab(depot, pbn);
4455 	vdo_refcount_t *counter_ptr = NULL;
4456 	int result;
4457 
4458 	if ((slab == NULL) || (slab->status != VDO_SLAB_REBUILT))
4459 		return 0;
4460 
4461 	result = get_reference_counter(slab, pbn, &counter_ptr);
4462 	if (result != VDO_SUCCESS)
4463 		return 0;
4464 
4465 	if (*counter_ptr == PROVISIONAL_REFERENCE_COUNT)
4466 		return (MAXIMUM_REFERENCE_COUNT - 1);
4467 
4468 	return (MAXIMUM_REFERENCE_COUNT - *counter_ptr);
4469 }
4470 
4471 /**
4472  * vdo_is_physical_data_block() - Determine whether the given PBN refers to a data block.
4473  * @depot: The depot.
4474  * @pbn: The physical block number to ask about.
4475  *
4476  * Return: True if the PBN corresponds to a data block.
4477  */
vdo_is_physical_data_block(const struct slab_depot * depot,physical_block_number_t pbn)4478 bool vdo_is_physical_data_block(const struct slab_depot *depot,
4479 				physical_block_number_t pbn)
4480 {
4481 	slab_count_t slab_number;
4482 	slab_block_number sbn;
4483 
4484 	return ((pbn == VDO_ZERO_BLOCK) ||
4485 		((get_slab_number(depot, pbn, &slab_number) == VDO_SUCCESS) &&
4486 		 (slab_block_number_from_pbn(depot->slabs[slab_number], pbn, &sbn) ==
4487 		  VDO_SUCCESS)));
4488 }
4489 
4490 /**
4491  * vdo_get_slab_depot_allocated_blocks() - Get the total number of data blocks allocated across all
4492  * the slabs in the depot.
4493  * @depot: The slab depot.
4494  *
4495  * This is the total number of blocks with a non-zero reference count.
4496  *
4497  * Context: This may be called from any thread.
4498  *
4499  * Return: The total number of blocks with a non-zero reference count.
4500  */
vdo_get_slab_depot_allocated_blocks(const struct slab_depot * depot)4501 block_count_t vdo_get_slab_depot_allocated_blocks(const struct slab_depot *depot)
4502 {
4503 	block_count_t total = 0;
4504 	zone_count_t zone;
4505 
4506 	for (zone = 0; zone < depot->zone_count; zone++) {
4507 		/* The allocators are responsible for thread safety. */
4508 		total += READ_ONCE(depot->allocators[zone].allocated_blocks);
4509 	}
4510 
4511 	return total;
4512 }
4513 
4514 /**
4515  * vdo_get_slab_depot_data_blocks() - Get the total number of data blocks in all the slabs in the
4516  *                                    depot.
4517  * @depot: The slab depot.
4518  *
4519  * Context: This may be called from any thread.
4520  *
4521  * Return: The total number of data blocks in all slabs.
4522  */
vdo_get_slab_depot_data_blocks(const struct slab_depot * depot)4523 block_count_t vdo_get_slab_depot_data_blocks(const struct slab_depot *depot)
4524 {
4525 	return (READ_ONCE(depot->slab_count) * depot->slab_config.data_blocks);
4526 }
4527 
4528 /**
4529  * finish_combining_zones() - Clean up after saving out the combined slab summary.
4530  * @completion: The vio which was used to write the summary data.
4531  */
finish_combining_zones(struct vdo_completion * completion)4532 static void finish_combining_zones(struct vdo_completion *completion)
4533 {
4534 	int result = completion->result;
4535 	struct vdo_completion *parent = completion->parent;
4536 
4537 	free_vio(as_vio(vdo_forget(completion)));
4538 	vdo_fail_completion(parent, result);
4539 }
4540 
handle_combining_error(struct vdo_completion * completion)4541 static void handle_combining_error(struct vdo_completion *completion)
4542 {
4543 	vio_record_metadata_io_error(as_vio(completion));
4544 	finish_combining_zones(completion);
4545 }
4546 
write_summary_endio(struct bio * bio)4547 static void write_summary_endio(struct bio *bio)
4548 {
4549 	struct vio *vio = bio->bi_private;
4550 	struct vdo *vdo = vio->completion.vdo;
4551 
4552 	continue_vio_after_io(vio, finish_combining_zones,
4553 			      vdo->thread_config.admin_thread);
4554 }
4555 
4556 /**
4557  * combine_summaries() - Treating the current entries buffer as the on-disk value of all zones,
4558  *                       update every zone to the correct values for every slab.
4559  * @depot: The depot whose summary entries should be combined.
4560  */
combine_summaries(struct slab_depot * depot)4561 static void combine_summaries(struct slab_depot *depot)
4562 {
4563 	/*
4564 	 * Combine all the old summary data into the portion of the buffer corresponding to the
4565 	 * first zone.
4566 	 */
4567 	zone_count_t zone = 0;
4568 	struct slab_summary_entry *entries = depot->summary_entries;
4569 
4570 	if (depot->old_zone_count > 1) {
4571 		slab_count_t entry_number;
4572 
4573 		for (entry_number = 0; entry_number < MAX_VDO_SLABS; entry_number++) {
4574 			if (zone != 0) {
4575 				memcpy(entries + entry_number,
4576 				       entries + (zone * MAX_VDO_SLABS) + entry_number,
4577 				       sizeof(struct slab_summary_entry));
4578 			}
4579 
4580 			zone++;
4581 			if (zone == depot->old_zone_count)
4582 				zone = 0;
4583 		}
4584 	}
4585 
4586 	/* Copy the combined data to each zones's region of the buffer. */
4587 	for (zone = 1; zone < MAX_VDO_PHYSICAL_ZONES; zone++) {
4588 		memcpy(entries + (zone * MAX_VDO_SLABS), entries,
4589 		       MAX_VDO_SLABS * sizeof(struct slab_summary_entry));
4590 	}
4591 }
4592 
4593 /**
4594  * finish_loading_summary() - Finish loading slab summary data.
4595  * @completion: The vio which was used to read the summary data.
4596  *
4597  * Combines the slab summary data from all the previously written zones and copies the combined
4598  * summary to each partition's data region. Then writes the combined summary back out to disk. This
4599  * callback is registered in load_summary_endio().
4600  */
finish_loading_summary(struct vdo_completion * completion)4601 static void finish_loading_summary(struct vdo_completion *completion)
4602 {
4603 	struct slab_depot *depot = completion->vdo->depot;
4604 
4605 	/* Combine the summary from each zone so each zone is correct for all slabs. */
4606 	combine_summaries(depot);
4607 
4608 	/* Write the combined summary back out. */
4609 	vdo_submit_metadata_vio(as_vio(completion), depot->summary_origin,
4610 				write_summary_endio, handle_combining_error,
4611 				REQ_OP_WRITE);
4612 }
4613 
load_summary_endio(struct bio * bio)4614 static void load_summary_endio(struct bio *bio)
4615 {
4616 	struct vio *vio = bio->bi_private;
4617 	struct vdo *vdo = vio->completion.vdo;
4618 
4619 	continue_vio_after_io(vio, finish_loading_summary,
4620 			      vdo->thread_config.admin_thread);
4621 }
4622 
4623 /**
4624  * load_slab_summary() - Load the slab summary before the slab data.
4625  * @context: The slab depot.
4626  * @parent: The load operation.
4627  *
4628  * Implements vdo_action_preamble_fn.
4629  */
load_slab_summary(void * context,struct vdo_completion * parent)4630 static void load_slab_summary(void *context, struct vdo_completion *parent)
4631 {
4632 	int result;
4633 	struct vio *vio;
4634 	struct slab_depot *depot = context;
4635 	const struct admin_state_code *operation =
4636 		vdo_get_current_manager_operation(depot->action_manager);
4637 
4638 	result = create_multi_block_metadata_vio(depot->vdo, VIO_TYPE_SLAB_SUMMARY,
4639 						 VIO_PRIORITY_METADATA, parent,
4640 						 VDO_SLAB_SUMMARY_BLOCKS,
4641 						 (char *) depot->summary_entries, &vio);
4642 	if (result != VDO_SUCCESS) {
4643 		vdo_fail_completion(parent, result);
4644 		return;
4645 	}
4646 
4647 	if ((operation == VDO_ADMIN_STATE_FORMATTING) ||
4648 	    (operation == VDO_ADMIN_STATE_LOADING_FOR_REBUILD)) {
4649 		finish_loading_summary(&vio->completion);
4650 		return;
4651 	}
4652 
4653 	vdo_submit_metadata_vio(vio, depot->summary_origin, load_summary_endio,
4654 				handle_combining_error, REQ_OP_READ);
4655 }
4656 
4657 /* Implements vdo_zone_action_fn. */
load_allocator(void * context,zone_count_t zone_number,struct vdo_completion * parent)4658 static void load_allocator(void *context, zone_count_t zone_number,
4659 			   struct vdo_completion *parent)
4660 {
4661 	struct slab_depot *depot = context;
4662 
4663 	vdo_start_loading(&depot->allocators[zone_number].state,
4664 			  vdo_get_current_manager_operation(depot->action_manager),
4665 			  parent, initiate_load);
4666 }
4667 
4668 /**
4669  * vdo_load_slab_depot() - Asynchronously load any slab depot state that isn't included in the
4670  *                         super_block component.
4671  * @depot: The depot to load.
4672  * @operation: The type of load to perform.
4673  * @parent: The completion to notify when the load is complete.
4674  * @context: Additional context for the load operation; may be NULL.
4675  *
4676  * This method may be called only before entering normal operation from the load thread.
4677  */
vdo_load_slab_depot(struct slab_depot * depot,const struct admin_state_code * operation,struct vdo_completion * parent,void * context)4678 void vdo_load_slab_depot(struct slab_depot *depot,
4679 			 const struct admin_state_code *operation,
4680 			 struct vdo_completion *parent, void *context)
4681 {
4682 	if (!vdo_assert_load_operation(operation, parent))
4683 		return;
4684 
4685 	vdo_schedule_operation_with_context(depot->action_manager, operation,
4686 					    load_slab_summary, load_allocator,
4687 					    NULL, context, parent);
4688 }
4689 
4690 /* Implements vdo_zone_action_fn. */
prepare_to_allocate(void * context,zone_count_t zone_number,struct vdo_completion * parent)4691 static void prepare_to_allocate(void *context, zone_count_t zone_number,
4692 				struct vdo_completion *parent)
4693 {
4694 	struct slab_depot *depot = context;
4695 	struct block_allocator *allocator = &depot->allocators[zone_number];
4696 	int result;
4697 
4698 	result = vdo_prepare_slabs_for_allocation(allocator);
4699 	if (result != VDO_SUCCESS) {
4700 		vdo_fail_completion(parent, result);
4701 		return;
4702 	}
4703 
4704 	scrub_slabs(allocator, parent);
4705 }
4706 
4707 /**
4708  * vdo_prepare_slab_depot_to_allocate() - Prepare the slab depot to come online and start
4709  *                                        allocating blocks.
4710  * @depot: The depot to prepare.
4711  * @load_type: The load type.
4712  * @parent: The completion to notify when the operation is complete.
4713  *
4714  * This method may be called only before entering normal operation from the load thread. It must be
4715  * called before allocation may proceed.
4716  */
vdo_prepare_slab_depot_to_allocate(struct slab_depot * depot,enum slab_depot_load_type load_type,struct vdo_completion * parent)4717 void vdo_prepare_slab_depot_to_allocate(struct slab_depot *depot,
4718 					enum slab_depot_load_type load_type,
4719 					struct vdo_completion *parent)
4720 {
4721 	depot->load_type = load_type;
4722 	atomic_set(&depot->zones_to_scrub, depot->zone_count);
4723 	vdo_schedule_action(depot->action_manager, NULL,
4724 			    prepare_to_allocate, NULL, parent);
4725 }
4726 
4727 /**
4728  * vdo_update_slab_depot_size() - Update the slab depot to reflect its new size in memory.
4729  * @depot: The depot to update.
4730  *
4731  * This size is saved to disk as part of the super block.
4732  */
vdo_update_slab_depot_size(struct slab_depot * depot)4733 void vdo_update_slab_depot_size(struct slab_depot *depot)
4734 {
4735 	depot->last_block = depot->new_last_block;
4736 }
4737 
4738 /**
4739  * vdo_prepare_to_grow_slab_depot() - Allocate new memory needed for a resize of a slab depot to
4740  *                                    the given size.
4741  * @depot: The depot to prepare to resize.
4742  * @partition: The new depot partition.
4743  *
4744  * Return: VDO_SUCCESS or an error.
4745  */
vdo_prepare_to_grow_slab_depot(struct slab_depot * depot,const struct partition * partition)4746 int vdo_prepare_to_grow_slab_depot(struct slab_depot *depot,
4747 				   const struct partition *partition)
4748 {
4749 	struct slab_depot_state_2_0 new_state;
4750 	int result;
4751 	slab_count_t new_slab_count;
4752 
4753 	if ((partition->count >> depot->slab_size_shift) <= depot->slab_count)
4754 		return VDO_INCREMENT_TOO_SMALL;
4755 
4756 	/* Generate the depot configuration for the new block count. */
4757 	VDO_ASSERT_LOG_ONLY(depot->first_block == partition->offset,
4758 			    "New slab depot partition doesn't change origin");
4759 	result = vdo_configure_slab_depot(partition, depot->slab_config,
4760 					  depot->zone_count, &new_state);
4761 	if (result != VDO_SUCCESS)
4762 		return result;
4763 
4764 	new_slab_count = vdo_compute_slab_count(depot->first_block,
4765 						new_state.last_block,
4766 						depot->slab_size_shift);
4767 	if (new_slab_count <= depot->slab_count)
4768 		return vdo_log_error_strerror(VDO_INCREMENT_TOO_SMALL,
4769 					      "Depot can only grow");
4770 	if (new_slab_count == depot->new_slab_count) {
4771 		/* Check it out, we've already got all the new slabs allocated! */
4772 		return VDO_SUCCESS;
4773 	}
4774 
4775 	vdo_abandon_new_slabs(depot);
4776 	result = allocate_slabs(depot, new_slab_count);
4777 	if (result != VDO_SUCCESS) {
4778 		vdo_abandon_new_slabs(depot);
4779 		return result;
4780 	}
4781 
4782 	depot->new_size = partition->count;
4783 	depot->old_last_block = depot->last_block;
4784 	depot->new_last_block = new_state.last_block;
4785 
4786 	return VDO_SUCCESS;
4787 }
4788 
4789 /**
4790  * finish_registration() - Finish registering new slabs now that all of the allocators have
4791  *                         received their new slabs.
4792  * @context: The slab depot.
4793  *
4794  * Implements vdo_action_conclusion_fn.
4795  */
finish_registration(void * context)4796 static int finish_registration(void *context)
4797 {
4798 	struct slab_depot *depot = context;
4799 
4800 	WRITE_ONCE(depot->slab_count, depot->new_slab_count);
4801 	vdo_free(depot->slabs);
4802 	depot->slabs = depot->new_slabs;
4803 	depot->new_slabs = NULL;
4804 	depot->new_slab_count = 0;
4805 	return VDO_SUCCESS;
4806 }
4807 
4808 /* Implements vdo_zone_action_fn. */
register_new_slabs(void * context,zone_count_t zone_number,struct vdo_completion * parent)4809 static void register_new_slabs(void *context, zone_count_t zone_number,
4810 			       struct vdo_completion *parent)
4811 {
4812 	struct slab_depot *depot = context;
4813 	struct block_allocator *allocator = &depot->allocators[zone_number];
4814 	slab_count_t i;
4815 
4816 	for (i = depot->slab_count; i < depot->new_slab_count; i++) {
4817 		struct vdo_slab *slab = depot->new_slabs[i];
4818 
4819 		if (slab->allocator == allocator)
4820 			register_slab_with_allocator(allocator, slab);
4821 	}
4822 
4823 	vdo_finish_completion(parent);
4824 }
4825 
4826 /**
4827  * vdo_use_new_slabs() - Use the new slabs allocated for resize.
4828  * @depot: The depot.
4829  * @parent: The object to notify when complete.
4830  */
vdo_use_new_slabs(struct slab_depot * depot,struct vdo_completion * parent)4831 void vdo_use_new_slabs(struct slab_depot *depot, struct vdo_completion *parent)
4832 {
4833 	VDO_ASSERT_LOG_ONLY(depot->new_slabs != NULL, "Must have new slabs to use");
4834 	vdo_schedule_operation(depot->action_manager,
4835 			       VDO_ADMIN_STATE_SUSPENDED_OPERATION,
4836 			       NULL, register_new_slabs,
4837 			       finish_registration, parent);
4838 }
4839 
4840 /**
4841  * stop_scrubbing() - Tell the scrubber to stop scrubbing after it finishes the slab it is
4842  *                    currently working on.
4843  * @allocator: The block allocator owning the scrubber to stop.
4844  */
stop_scrubbing(struct block_allocator * allocator)4845 static void stop_scrubbing(struct block_allocator *allocator)
4846 {
4847 	struct slab_scrubber *scrubber = &allocator->scrubber;
4848 
4849 	if (vdo_is_state_quiescent(&scrubber->admin_state)) {
4850 		vdo_finish_completion(&allocator->completion);
4851 	} else {
4852 		vdo_start_draining(&scrubber->admin_state,
4853 				   VDO_ADMIN_STATE_SUSPENDING,
4854 				   &allocator->completion, NULL);
4855 	}
4856 }
4857 
4858 /* Implements vdo_admin_initiator_fn. */
initiate_summary_drain(struct admin_state * state)4859 static void initiate_summary_drain(struct admin_state *state)
4860 {
4861 	check_summary_drain_complete(container_of(state, struct block_allocator,
4862 						  summary_state));
4863 }
4864 
do_drain_step(struct vdo_completion * completion)4865 static void do_drain_step(struct vdo_completion *completion)
4866 {
4867 	struct block_allocator *allocator = vdo_as_block_allocator(completion);
4868 
4869 	vdo_prepare_completion_for_requeue(&allocator->completion, do_drain_step,
4870 					   handle_operation_error, allocator->thread_id,
4871 					   NULL);
4872 	switch (++allocator->drain_step) {
4873 	case VDO_DRAIN_ALLOCATOR_STEP_SCRUBBER:
4874 		stop_scrubbing(allocator);
4875 		return;
4876 
4877 	case VDO_DRAIN_ALLOCATOR_STEP_SLABS:
4878 		apply_to_slabs(allocator, do_drain_step);
4879 		return;
4880 
4881 	case VDO_DRAIN_ALLOCATOR_STEP_SUMMARY:
4882 		vdo_start_draining(&allocator->summary_state,
4883 				   vdo_get_admin_state_code(&allocator->state),
4884 				   completion, initiate_summary_drain);
4885 		return;
4886 
4887 	case VDO_DRAIN_ALLOCATOR_STEP_FINISHED:
4888 		VDO_ASSERT_LOG_ONLY(!is_vio_pool_busy(allocator->vio_pool),
4889 				    "vio pool not busy");
4890 		vdo_finish_draining_with_result(&allocator->state, completion->result);
4891 		return;
4892 
4893 	default:
4894 		vdo_finish_draining_with_result(&allocator->state, UDS_BAD_STATE);
4895 	}
4896 }
4897 
4898 /* Implements vdo_admin_initiator_fn. */
initiate_drain(struct admin_state * state)4899 static void initiate_drain(struct admin_state *state)
4900 {
4901 	struct block_allocator *allocator =
4902 		container_of(state, struct block_allocator, state);
4903 
4904 	allocator->drain_step = VDO_DRAIN_ALLOCATOR_START;
4905 	do_drain_step(&allocator->completion);
4906 }
4907 
4908 /*
4909  * Drain all allocator I/O. Depending upon the type of drain, some or all dirty metadata may be
4910  * written to disk. The type of drain will be determined from the state of the allocator's depot.
4911  *
4912  * Implements vdo_zone_action_fn.
4913  */
drain_allocator(void * context,zone_count_t zone_number,struct vdo_completion * parent)4914 static void drain_allocator(void *context, zone_count_t zone_number,
4915 			    struct vdo_completion *parent)
4916 {
4917 	struct slab_depot *depot = context;
4918 
4919 	vdo_start_draining(&depot->allocators[zone_number].state,
4920 			   vdo_get_current_manager_operation(depot->action_manager),
4921 			   parent, initiate_drain);
4922 }
4923 
4924 /**
4925  * vdo_drain_slab_depot() - Drain all slab depot I/O.
4926  * @depot: The depot to drain.
4927  * @operation: The drain operation (flush, rebuild, suspend, or save).
4928  * @parent: The completion to finish when the drain is complete.
4929  *
4930  * If saving, or flushing, all dirty depot metadata will be written out. If saving or suspending,
4931  * the depot will be left in a suspended state.
4932  */
vdo_drain_slab_depot(struct slab_depot * depot,const struct admin_state_code * operation,struct vdo_completion * parent)4933 void vdo_drain_slab_depot(struct slab_depot *depot,
4934 			  const struct admin_state_code *operation,
4935 			  struct vdo_completion *parent)
4936 {
4937 	vdo_schedule_operation(depot->action_manager, operation,
4938 			       NULL, drain_allocator, NULL, parent);
4939 }
4940 
4941 /**
4942  * resume_scrubbing() - Tell the scrubber to resume scrubbing if it has been stopped.
4943  * @allocator: The allocator being resumed.
4944  */
resume_scrubbing(struct block_allocator * allocator)4945 static void resume_scrubbing(struct block_allocator *allocator)
4946 {
4947 	int result;
4948 	struct slab_scrubber *scrubber = &allocator->scrubber;
4949 
4950 	if (!has_slabs_to_scrub(scrubber)) {
4951 		vdo_finish_completion(&allocator->completion);
4952 		return;
4953 	}
4954 
4955 	result = vdo_resume_if_quiescent(&scrubber->admin_state);
4956 	if (result != VDO_SUCCESS) {
4957 		vdo_fail_completion(&allocator->completion, result);
4958 		return;
4959 	}
4960 
4961 	scrub_next_slab(scrubber);
4962 	vdo_finish_completion(&allocator->completion);
4963 }
4964 
do_resume_step(struct vdo_completion * completion)4965 static void do_resume_step(struct vdo_completion *completion)
4966 {
4967 	struct block_allocator *allocator = vdo_as_block_allocator(completion);
4968 
4969 	vdo_prepare_completion_for_requeue(&allocator->completion, do_resume_step,
4970 					   handle_operation_error,
4971 					   allocator->thread_id, NULL);
4972 	switch (--allocator->drain_step) {
4973 	case VDO_DRAIN_ALLOCATOR_STEP_SUMMARY:
4974 		vdo_fail_completion(completion,
4975 				    vdo_resume_if_quiescent(&allocator->summary_state));
4976 		return;
4977 
4978 	case VDO_DRAIN_ALLOCATOR_STEP_SLABS:
4979 		apply_to_slabs(allocator, do_resume_step);
4980 		return;
4981 
4982 	case VDO_DRAIN_ALLOCATOR_STEP_SCRUBBER:
4983 		resume_scrubbing(allocator);
4984 		return;
4985 
4986 	case VDO_DRAIN_ALLOCATOR_START:
4987 		vdo_finish_resuming_with_result(&allocator->state, completion->result);
4988 		return;
4989 
4990 	default:
4991 		vdo_finish_resuming_with_result(&allocator->state, UDS_BAD_STATE);
4992 	}
4993 }
4994 
4995 /* Implements vdo_admin_initiator_fn. */
initiate_resume(struct admin_state * state)4996 static void initiate_resume(struct admin_state *state)
4997 {
4998 	struct block_allocator *allocator =
4999 		container_of(state, struct block_allocator, state);
5000 
5001 	allocator->drain_step = VDO_DRAIN_ALLOCATOR_STEP_FINISHED;
5002 	do_resume_step(&allocator->completion);
5003 }
5004 
5005 /* Implements vdo_zone_action_fn. */
resume_allocator(void * context,zone_count_t zone_number,struct vdo_completion * parent)5006 static void resume_allocator(void *context, zone_count_t zone_number,
5007 			     struct vdo_completion *parent)
5008 {
5009 	struct slab_depot *depot = context;
5010 
5011 	vdo_start_resuming(&depot->allocators[zone_number].state,
5012 			   vdo_get_current_manager_operation(depot->action_manager),
5013 			   parent, initiate_resume);
5014 }
5015 
5016 /**
5017  * vdo_resume_slab_depot() - Resume a suspended slab depot.
5018  * @depot: The depot to resume.
5019  * @parent: The completion to finish when the depot has resumed.
5020  */
vdo_resume_slab_depot(struct slab_depot * depot,struct vdo_completion * parent)5021 void vdo_resume_slab_depot(struct slab_depot *depot, struct vdo_completion *parent)
5022 {
5023 	if (vdo_is_read_only(depot->vdo)) {
5024 		vdo_continue_completion(parent, VDO_READ_ONLY);
5025 		return;
5026 	}
5027 
5028 	vdo_schedule_operation(depot->action_manager, VDO_ADMIN_STATE_RESUMING,
5029 			       NULL, resume_allocator, NULL, parent);
5030 }
5031 
5032 /**
5033  * vdo_commit_oldest_slab_journal_tail_blocks() - Commit all dirty tail blocks which are locking a
5034  *                                                given recovery journal block.
5035  * @depot: The depot.
5036  * @recovery_block_number: The sequence number of the recovery journal block whose locks should be
5037  *                         released.
5038  *
5039  * Context: This method must be called from the journal zone thread.
5040  */
vdo_commit_oldest_slab_journal_tail_blocks(struct slab_depot * depot,sequence_number_t recovery_block_number)5041 void vdo_commit_oldest_slab_journal_tail_blocks(struct slab_depot *depot,
5042 						sequence_number_t recovery_block_number)
5043 {
5044 	if (depot == NULL)
5045 		return;
5046 
5047 	depot->new_release_request = recovery_block_number;
5048 	vdo_schedule_default_action(depot->action_manager);
5049 }
5050 
5051 /* Implements vdo_zone_action_fn. */
scrub_all_unrecovered_slabs(void * context,zone_count_t zone_number,struct vdo_completion * parent)5052 static void scrub_all_unrecovered_slabs(void *context, zone_count_t zone_number,
5053 					struct vdo_completion *parent)
5054 {
5055 	struct slab_depot *depot = context;
5056 
5057 	scrub_slabs(&depot->allocators[zone_number], NULL);
5058 	vdo_launch_completion(parent);
5059 }
5060 
5061 /**
5062  * vdo_scrub_all_unrecovered_slabs() - Scrub all unrecovered slabs.
5063  * @depot: The depot to scrub.
5064  * @parent: The object to notify when scrubbing has been launched for all zones.
5065  */
vdo_scrub_all_unrecovered_slabs(struct slab_depot * depot,struct vdo_completion * parent)5066 void vdo_scrub_all_unrecovered_slabs(struct slab_depot *depot,
5067 				     struct vdo_completion *parent)
5068 {
5069 	vdo_schedule_action(depot->action_manager, NULL,
5070 			    scrub_all_unrecovered_slabs,
5071 			    NULL, parent);
5072 }
5073 
5074 /**
5075  * get_block_allocator_statistics() - Get the total of the statistics from all the block allocators
5076  *                                    in the depot.
5077  * @depot: The slab depot.
5078  *
5079  * Return: The statistics from all block allocators in the depot.
5080  */
5081 static struct block_allocator_statistics __must_check
get_block_allocator_statistics(const struct slab_depot * depot)5082 get_block_allocator_statistics(const struct slab_depot *depot)
5083 {
5084 	struct block_allocator_statistics totals;
5085 	zone_count_t zone;
5086 
5087 	memset(&totals, 0, sizeof(totals));
5088 
5089 	for (zone = 0; zone < depot->zone_count; zone++) {
5090 		const struct block_allocator *allocator = &depot->allocators[zone];
5091 		const struct block_allocator_statistics *stats = &allocator->statistics;
5092 
5093 		totals.slab_count += allocator->slab_count;
5094 		totals.slabs_opened += READ_ONCE(stats->slabs_opened);
5095 		totals.slabs_reopened += READ_ONCE(stats->slabs_reopened);
5096 	}
5097 
5098 	return totals;
5099 }
5100 
5101 /**
5102  * get_ref_counts_statistics() - Get the cumulative ref_counts statistics for the depot.
5103  * @depot: The slab depot.
5104  *
5105  * Return: The cumulative statistics for all ref_counts in the depot.
5106  */
5107 static struct ref_counts_statistics __must_check
get_ref_counts_statistics(const struct slab_depot * depot)5108 get_ref_counts_statistics(const struct slab_depot *depot)
5109 {
5110 	struct ref_counts_statistics totals;
5111 	zone_count_t zone;
5112 
5113 	memset(&totals, 0, sizeof(totals));
5114 
5115 	for (zone = 0; zone < depot->zone_count; zone++) {
5116 		totals.blocks_written +=
5117 			READ_ONCE(depot->allocators[zone].ref_counts_statistics.blocks_written);
5118 	}
5119 
5120 	return totals;
5121 }
5122 
5123 /**
5124  * get_slab_journal_statistics() - Get the aggregated slab journal statistics for the depot.
5125  * @depot: The slab depot.
5126  *
5127  * Return: The aggregated statistics for all slab journals in the depot.
5128  */
5129 static struct slab_journal_statistics __must_check
get_slab_journal_statistics(const struct slab_depot * depot)5130 get_slab_journal_statistics(const struct slab_depot *depot)
5131 {
5132 	struct slab_journal_statistics totals;
5133 	zone_count_t zone;
5134 
5135 	memset(&totals, 0, sizeof(totals));
5136 
5137 	for (zone = 0; zone < depot->zone_count; zone++) {
5138 		const struct slab_journal_statistics *stats =
5139 			&depot->allocators[zone].slab_journal_statistics;
5140 
5141 		totals.disk_full_count += READ_ONCE(stats->disk_full_count);
5142 		totals.flush_count += READ_ONCE(stats->flush_count);
5143 		totals.blocked_count += READ_ONCE(stats->blocked_count);
5144 		totals.blocks_written += READ_ONCE(stats->blocks_written);
5145 		totals.tail_busy_count += READ_ONCE(stats->tail_busy_count);
5146 	}
5147 
5148 	return totals;
5149 }
5150 
5151 /**
5152  * vdo_get_slab_depot_statistics() - Get all the vdo_statistics fields that are properties of the
5153  *                                   slab depot.
5154  * @depot: The slab depot.
5155  * @stats: The vdo statistics structure to partially fill.
5156  */
vdo_get_slab_depot_statistics(const struct slab_depot * depot,struct vdo_statistics * stats)5157 void vdo_get_slab_depot_statistics(const struct slab_depot *depot,
5158 				   struct vdo_statistics *stats)
5159 {
5160 	slab_count_t slab_count = READ_ONCE(depot->slab_count);
5161 	slab_count_t unrecovered = 0;
5162 	zone_count_t zone;
5163 
5164 	for (zone = 0; zone < depot->zone_count; zone++) {
5165 		/* The allocators are responsible for thread safety. */
5166 		unrecovered += READ_ONCE(depot->allocators[zone].scrubber.slab_count);
5167 	}
5168 
5169 	stats->recovery_percentage = (slab_count - unrecovered) * 100 / slab_count;
5170 	stats->allocator = get_block_allocator_statistics(depot);
5171 	stats->ref_counts = get_ref_counts_statistics(depot);
5172 	stats->slab_journal = get_slab_journal_statistics(depot);
5173 	stats->slab_summary = (struct slab_summary_statistics) {
5174 		.blocks_written = atomic64_read(&depot->summary_statistics.blocks_written),
5175 	};
5176 }
5177 
5178 /**
5179  * vdo_dump_slab_depot() - Dump the slab depot, in a thread-unsafe fashion.
5180  * @depot: The slab depot.
5181  */
vdo_dump_slab_depot(const struct slab_depot * depot)5182 void vdo_dump_slab_depot(const struct slab_depot *depot)
5183 {
5184 	vdo_log_info("vdo slab depot");
5185 	vdo_log_info("  zone_count=%u old_zone_count=%u slabCount=%u active_release_request=%llu new_release_request=%llu",
5186 		     (unsigned int) depot->zone_count,
5187 		     (unsigned int) depot->old_zone_count, READ_ONCE(depot->slab_count),
5188 		     (unsigned long long) depot->active_release_request,
5189 		     (unsigned long long) depot->new_release_request);
5190 }
5191