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