• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2020 The Pigweed Authors
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License"); you may not
4 // use this file except in compliance with the License. You may obtain a copy of
5 // the License at
6 //
7 //     https://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
11 // WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
12 // License for the specific language governing permissions and limitations under
13 // the License.
14 
15 #define PW_LOG_MODULE_NAME "KVS"
16 #define PW_LOG_LEVEL PW_KVS_LOG_LEVEL
17 
18 #include "pw_kvs/key_value_store.h"
19 
20 #include <algorithm>
21 #include <cinttypes>
22 #include <cstring>
23 #include <type_traits>
24 
25 #include "pw_assert/check.h"
26 #include "pw_kvs_private/config.h"
27 #include "pw_log/log.h"
28 #include "pw_status/try.h"
29 
30 namespace pw::kvs {
31 namespace {
32 
33 using std::byte;
34 
InvalidKey(Key key)35 constexpr bool InvalidKey(Key key) {
36   return key.empty() || (key.size() > internal::Entry::kMaxKeyLength);
37 }
38 
39 }  // namespace
40 
KeyValueStore(FlashPartition * partition,span<const EntryFormat> formats,const Options & options,size_t redundancy,Vector<SectorDescriptor> & sector_descriptor_list,const SectorDescriptor ** temp_sectors_to_skip,Vector<KeyDescriptor> & key_descriptor_list,Address * addresses)41 KeyValueStore::KeyValueStore(FlashPartition* partition,
42                              span<const EntryFormat> formats,
43                              const Options& options,
44                              size_t redundancy,
45                              Vector<SectorDescriptor>& sector_descriptor_list,
46                              const SectorDescriptor** temp_sectors_to_skip,
47                              Vector<KeyDescriptor>& key_descriptor_list,
48                              Address* addresses)
49     : partition_(*partition),
50       formats_(formats),
51       sectors_(sector_descriptor_list, *partition, temp_sectors_to_skip),
52       entry_cache_(key_descriptor_list, addresses, redundancy),
53       options_(options),
54       initialized_(InitializationState::kNotInitialized),
55       error_detected_(false),
56       internal_stats_({}),
57       last_transaction_id_(0) {}
58 
Init()59 Status KeyValueStore::Init() {
60   initialized_ = InitializationState::kNotInitialized;
61   error_detected_ = false;
62   last_transaction_id_ = 0;
63 
64   PW_LOG_INFO("Initializing key value store");
65   if (partition_.sector_count() > sectors_.max_size()) {
66     PW_LOG_ERROR(
67         "KVS init failed: kMaxUsableSectors (=%u) must be at least as "
68         "large as the number of sectors in the flash partition (=%u)",
69         unsigned(sectors_.max_size()),
70         unsigned(partition_.sector_count()));
71     return Status::FailedPrecondition();
72   }
73 
74   if (partition_.sector_count() < 2) {
75     PW_LOG_ERROR(
76         "KVS init failed: FlashParition sector count (=%u) must be at 2. KVS "
77         "requires at least 1 working sector + 1 free/reserved sector",
78         unsigned(partition_.sector_count()));
79     return Status::FailedPrecondition();
80   }
81 
82   const size_t sector_size_bytes = partition_.sector_size_bytes();
83 
84   // TODO(davidrogers): investigate doing this as a static assert/compile-time
85   // check.
86   if (sector_size_bytes > SectorDescriptor::max_sector_size()) {
87     PW_LOG_ERROR(
88         "KVS init failed: sector_size_bytes (=%u) is greater than maximum "
89         "allowed sector size (=%u)",
90         unsigned(sector_size_bytes),
91         unsigned(SectorDescriptor::max_sector_size()));
92     return Status::FailedPrecondition();
93   }
94 
95   Status metadata_result = InitializeMetadata();
96 
97   if (!error_detected_) {
98     initialized_ = InitializationState::kReady;
99   } else {
100     initialized_ = InitializationState::kNeedsMaintenance;
101 
102     if (options_.recovery != ErrorRecovery::kManual) {
103       size_t pre_fix_redundancy_errors =
104           internal_stats_.missing_redundant_entries_recovered;
105       Status recovery_status = FixErrors();
106 
107       if (recovery_status.ok()) {
108         if (metadata_result.IsOutOfRange()) {
109           internal_stats_.missing_redundant_entries_recovered =
110               pre_fix_redundancy_errors;
111           PW_LOG_INFO("KVS init: Redundancy level successfully updated");
112         } else {
113           PW_LOG_WARN("KVS init: Corruption detected and fully repaired");
114         }
115         initialized_ = InitializationState::kReady;
116       } else if (recovery_status.IsResourceExhausted()) {
117         PW_LOG_WARN("KVS init: Unable to maintain required free sector");
118       } else {
119         PW_LOG_WARN("KVS init: Corruption detected and unable repair");
120       }
121     } else {
122       PW_LOG_WARN(
123           "KVS init: Corruption detected, no repair attempted due to options");
124     }
125   }
126 
127   PW_LOG_INFO(
128       "KeyValueStore init complete: active keys %u, deleted keys %u, sectors "
129       "%u, logical sector size %u bytes",
130       unsigned(size()),
131       unsigned(entry_cache_.total_entries() - size()),
132       unsigned(sectors_.size()),
133       unsigned(partition_.sector_size_bytes()));
134 
135   // Report any corruption was not repaired.
136   if (error_detected_) {
137     PW_LOG_WARN(
138         "KVS init: Corruption found but not repaired, KVS unavailable until "
139         "successful maintenance.");
140     return Status::DataLoss();
141   }
142 
143   return OkStatus();
144 }
145 
InitializeMetadata()146 Status KeyValueStore::InitializeMetadata() {
147   const size_t sector_size_bytes = partition_.sector_size_bytes();
148 
149   sectors_.Reset();
150   entry_cache_.Reset();
151 
152   PW_LOG_DEBUG("First pass: Read all entries from all sectors");
153   Address sector_address = 0;
154 
155   size_t total_corrupt_bytes = 0;
156   size_t corrupt_entries = 0;
157   bool empty_sector_found = false;
158   size_t entry_copies_missing = 0;
159 
160   for (SectorDescriptor& sector : sectors_) {
161     Address entry_address = sector_address;
162 
163     size_t sector_corrupt_bytes = 0;
164 
165     for (int num_entries_in_sector = 0; true; num_entries_in_sector++) {
166       PW_LOG_DEBUG("Load entry: sector=%u, entry#=%d, address=%u",
167                    unsigned(sector_address),
168                    num_entries_in_sector,
169                    unsigned(entry_address));
170 
171       if (!sectors_.AddressInSector(sector, entry_address)) {
172         PW_LOG_DEBUG("Fell off end of sector; moving to the next sector");
173         break;
174       }
175 
176       Address next_entry_address;
177       Status status = LoadEntry(entry_address, &next_entry_address);
178       if (status.IsNotFound()) {
179         PW_LOG_DEBUG(
180             "Hit un-written data in sector; moving to the next sector");
181         break;
182       } else if (!status.ok()) {
183         // The entry could not be read, indicating likely data corruption within
184         // the sector. Try to scan the remainder of the sector for other
185         // entries.
186 
187         error_detected_ = true;
188         corrupt_entries++;
189 
190         status = ScanForEntry(sector,
191                               entry_address + Entry::kMinAlignmentBytes,
192                               &next_entry_address);
193         if (!status.ok()) {
194           // No further entries in this sector. Mark the remaining bytes in the
195           // sector as corrupt (since we can't reliably know the size of the
196           // corrupt entry).
197           sector_corrupt_bytes +=
198               sector_size_bytes - (entry_address - sector_address);
199           break;
200         }
201 
202         sector_corrupt_bytes += next_entry_address - entry_address;
203       }
204 
205       // Entry loaded successfully; so get ready to load the next one.
206       entry_address = next_entry_address;
207 
208       // Update of the number of writable bytes in this sector.
209       sector.set_writable_bytes(sector_size_bytes -
210                                 (entry_address - sector_address));
211     }
212 
213     if (sector_corrupt_bytes > 0) {
214       // If the sector contains corrupt data, prevent any further entries from
215       // being written to it by indicating that it has no space. This should
216       // also make it a decent GC candidate. Valid keys in the sector are still
217       // readable as normal.
218       sector.mark_corrupt();
219       error_detected_ = true;
220 
221       PW_LOG_WARN("Sector %u contains %uB of corrupt data",
222                   sectors_.Index(sector),
223                   unsigned(sector_corrupt_bytes));
224     }
225 
226     if (sector.Empty(sector_size_bytes)) {
227       empty_sector_found = true;
228     }
229     sector_address += sector_size_bytes;
230     total_corrupt_bytes += sector_corrupt_bytes;
231   }
232 
233   PW_LOG_DEBUG("Second pass: Count valid bytes in each sector");
234   Address newest_key = 0;
235 
236   // For every valid entry, for each address, count the valid bytes in that
237   // sector. If the address fails to read, remove the address and mark the
238   // sector as corrupt. Track which entry has the newest transaction ID for
239   // initializing last_new_sector_.
240   for (EntryMetadata& metadata : entry_cache_) {
241     if (metadata.addresses().size() < redundancy()) {
242       PW_LOG_DEBUG("Key 0x%08x missing copies, has %u, needs %u",
243                    unsigned(metadata.hash()),
244                    unsigned(metadata.addresses().size()),
245                    unsigned(redundancy()));
246       entry_copies_missing++;
247     }
248     size_t index = 0;
249     while (index < metadata.addresses().size()) {
250       Address address = metadata.addresses()[index];
251       Entry entry;
252 
253       Status read_result = Entry::Read(partition_, address, formats_, &entry);
254 
255       SectorDescriptor& sector = sectors_.FromAddress(address);
256 
257       if (read_result.ok()) {
258         sector.AddValidBytes(entry.size());
259         index++;
260       } else {
261         corrupt_entries++;
262         total_corrupt_bytes += sector.writable_bytes();
263         error_detected_ = true;
264         sector.mark_corrupt();
265 
266         // Remove the bad address and stay at this index. The removal
267         // replaces out the removed address with the back address so
268         // this index needs to be rechecked with the new address.
269         metadata.RemoveAddress(address);
270       }
271     }
272 
273     if (metadata.IsNewerThan(last_transaction_id_)) {
274       last_transaction_id_ = metadata.transaction_id();
275       newest_key = metadata.addresses().back();
276     }
277   }
278 
279   sectors_.set_last_new_sector(newest_key);
280 
281   if (!empty_sector_found) {
282     PW_LOG_DEBUG("No empty sector found");
283     error_detected_ = true;
284   }
285 
286   if (entry_copies_missing > 0) {
287     bool other_errors = error_detected_;
288     error_detected_ = true;
289 
290     if (!other_errors && entry_copies_missing == entry_cache_.total_entries()) {
291       PW_LOG_INFO(
292           "KVS configuration changed to redundancy of %u total copies per key",
293           unsigned(redundancy()));
294       return Status::OutOfRange();
295     }
296   }
297 
298   if (error_detected_) {
299     PW_LOG_WARN(
300         "Corruption detected. Found %u corrupt bytes, %u corrupt entries, "
301         "and %u keys missing redundant copies.",
302         unsigned(total_corrupt_bytes),
303         unsigned(corrupt_entries),
304         unsigned(entry_copies_missing));
305     return Status::FailedPrecondition();
306   }
307   return OkStatus();
308 }
309 
GetStorageStats() const310 KeyValueStore::StorageStats KeyValueStore::GetStorageStats() const {
311   StorageStats stats{};
312   const size_t sector_size = partition_.sector_size_bytes();
313   bool found_empty_sector = false;
314   stats.sector_erase_count = internal_stats_.sector_erase_count;
315   stats.corrupt_sectors_recovered = internal_stats_.corrupt_sectors_recovered;
316   stats.missing_redundant_entries_recovered =
317       internal_stats_.missing_redundant_entries_recovered;
318 
319   for (const SectorDescriptor& sector : sectors_) {
320     stats.in_use_bytes += sector.valid_bytes();
321     stats.reclaimable_bytes += sector.RecoverableBytes(sector_size);
322 
323     if (!found_empty_sector && sector.Empty(sector_size)) {
324       // The KVS tries to always keep an empty sector for GC, so don't count
325       // the first empty sector seen as writable space. However, a free sector
326       // cannot always be assumed to exist; if a GC operation fails, all sectors
327       // may be partially written, in which case the space reported might be
328       // inaccurate.
329       found_empty_sector = true;
330       continue;
331     }
332 
333     stats.writable_bytes += sector.writable_bytes();
334   }
335 
336   return stats;
337 }
338 
339 // Check KVS for any error conditions. Primarily intended for test and
340 // internal use.
CheckForErrors()341 bool KeyValueStore::CheckForErrors() {
342   // Check for corrupted sectors
343   for (SectorDescriptor& sector : sectors_) {
344     if (sector.corrupt()) {
345       error_detected_ = true;
346       return error_detected();
347     }
348   }
349 
350   // Check for missing redundancy.
351   if (redundancy() > 1) {
352     for (const EntryMetadata& metadata : entry_cache_) {
353       if (metadata.addresses().size() < redundancy()) {
354         error_detected_ = true;
355         return error_detected();
356       }
357     }
358   }
359 
360   return error_detected();
361 }
362 
LoadEntry(Address entry_address,Address * next_entry_address)363 Status KeyValueStore::LoadEntry(Address entry_address,
364                                 Address* next_entry_address) {
365   Entry entry;
366   PW_TRY(Entry::Read(partition_, entry_address, formats_, &entry));
367 
368   // Read the key from flash & validate the entry (which reads the value).
369   Entry::KeyBuffer key_buffer;
370   PW_TRY_ASSIGN(size_t key_length, entry.ReadKey(key_buffer));
371   const Key key(key_buffer.data(), key_length);
372 
373   PW_TRY(entry.VerifyChecksumInFlash());
374 
375   // A valid entry was found, so update the next entry address before doing any
376   // of the checks that happen in AddNewOrUpdateExisting.
377   *next_entry_address = entry.next_address();
378   return entry_cache_.AddNewOrUpdateExisting(
379       entry.descriptor(key), entry.address(), partition_.sector_size_bytes());
380 }
381 
382 // Scans flash memory within a sector to find a KVS entry magic.
ScanForEntry(const SectorDescriptor & sector,Address start_address,Address * next_entry_address)383 Status KeyValueStore::ScanForEntry(const SectorDescriptor& sector,
384                                    Address start_address,
385                                    Address* next_entry_address) {
386   PW_LOG_DEBUG("Scanning sector %u for entries starting from address %u",
387                sectors_.Index(sector),
388                unsigned(start_address));
389 
390   // Entries must start at addresses which are aligned on a multiple of
391   // Entry::kMinAlignmentBytes. However, that multiple can vary between entries.
392   // When scanning, we don't have an entry to tell us what the current alignment
393   // is, so the minimum alignment is used to be exhaustive.
394   for (Address address = AlignUp(start_address, Entry::kMinAlignmentBytes);
395        sectors_.AddressInSector(sector, address);
396        address += Entry::kMinAlignmentBytes) {
397     uint32_t magic;
398     StatusWithSize read_result =
399         partition_.Read(address, as_writable_bytes(span(&magic, 1)));
400     if (!read_result.ok()) {
401       continue;
402     }
403     if (formats_.KnownMagic(magic)) {
404       PW_LOG_DEBUG("Found entry magic at address %u", unsigned(address));
405       *next_entry_address = address;
406       return OkStatus();
407     }
408   }
409 
410   return Status::NotFound();
411 }
412 
413 #if PW_KVS_REMOVE_DELETED_KEYS_IN_HEAVY_MAINTENANCE
414 
RemoveDeletedKeyEntries()415 Status KeyValueStore::RemoveDeletedKeyEntries() {
416   for (internal::EntryCache::iterator it = entry_cache_.begin();
417        it != entry_cache_.end();
418        ++it) {
419     EntryMetadata& entry_metadata = *it;
420 
421     // The iterator we are given back from RemoveEntry could also be deleted,
422     // so loop until we find one that isn't deleted.
423     while (entry_metadata.state() == EntryState::kDeleted) {
424       // Read the original entry to get the size for sector accounting purposes.
425       Entry entry;
426       PW_TRY(ReadEntry(entry_metadata, entry));
427 
428       for (Address address : entry_metadata.addresses()) {
429         sectors_.FromAddress(address).RemoveValidBytes(entry.size());
430       }
431 
432       it = entry_cache_.RemoveEntry(it);
433 
434       if (it == entry_cache_.end()) {
435         return OkStatus();  // new iterator is the end, bail
436       }
437 
438       entry_metadata = *it;  // set entry_metadata to check for deletion again
439     }
440   }
441 
442   return OkStatus();
443 }
444 
445 #endif  // PW_KVS_REMOVE_DELETED_KEYS_IN_HEAVY_MAINTENANCE
446 
Get(Key key,span<byte> value_buffer,size_t offset_bytes) const447 StatusWithSize KeyValueStore::Get(Key key,
448                                   span<byte> value_buffer,
449                                   size_t offset_bytes) const {
450   PW_TRY_WITH_SIZE(CheckReadOperation(key));
451 
452   EntryMetadata metadata;
453   PW_TRY_WITH_SIZE(FindExisting(key, &metadata));
454 
455   return Get(key, metadata, value_buffer, offset_bytes);
456 }
457 
PutBytes(Key key,span<const byte> value)458 Status KeyValueStore::PutBytes(Key key, span<const byte> value) {
459   PW_TRY(CheckWriteOperation(key));
460   PW_LOG_DEBUG("Writing key/value; key length=%u, value length=%u",
461                unsigned(key.size()),
462                unsigned(value.size()));
463 
464   if (Entry::size(partition_, key, value) > partition_.sector_size_bytes()) {
465     PW_LOG_DEBUG("%u B value with %u B key cannot fit in one sector",
466                  unsigned(value.size()),
467                  unsigned(key.size()));
468     return Status::InvalidArgument();
469   }
470 
471   EntryMetadata metadata;
472   Status status = FindEntry(key, &metadata);
473 
474   if (status.ok()) {
475     // TODO(davidrogers): figure out logging how to support multiple addresses.
476     PW_LOG_DEBUG("Overwriting entry for key 0x%08x in %u sectors including %u",
477                  unsigned(metadata.hash()),
478                  unsigned(metadata.addresses().size()),
479                  sectors_.Index(metadata.first_address()));
480     return WriteEntryForExistingKey(metadata, EntryState::kValid, key, value);
481   }
482 
483   if (status.IsNotFound()) {
484     return WriteEntryForNewKey(key, value);
485   }
486 
487   return status;
488 }
489 
Delete(Key key)490 Status KeyValueStore::Delete(Key key) {
491   PW_TRY(CheckWriteOperation(key));
492 
493   EntryMetadata metadata;
494   PW_TRY(FindExisting(key, &metadata));
495 
496   // TODO(davidrogers): figure out logging how to support multiple addresses.
497   PW_LOG_DEBUG("Writing tombstone for key 0x%08x in %u sectors including %u",
498                unsigned(metadata.hash()),
499                unsigned(metadata.addresses().size()),
500                sectors_.Index(metadata.first_address()));
501   return WriteEntryForExistingKey(metadata, EntryState::kDeleted, key, {});
502 }
503 
ReadKey()504 void KeyValueStore::Item::ReadKey() {
505   key_buffer_.fill('\0');
506 
507   Entry entry;
508   if (kvs_.ReadEntry(*iterator_, entry).ok()) {
509     entry.ReadKey(key_buffer_)
510         .IgnoreError();  // TODO: b/242598609 - Handle Status properly
511   }
512 }
513 
operator ++()514 KeyValueStore::iterator& KeyValueStore::iterator::operator++() {
515   // Skip to the next entry that is valid (not deleted).
516   while (++item_.iterator_ != item_.kvs_.entry_cache_.end() &&
517          item_.iterator_->state() != EntryState::kValid) {
518   }
519   return *this;
520 }
521 
begin() const522 KeyValueStore::iterator KeyValueStore::begin() const {
523   internal::EntryCache::const_iterator cache_iterator = entry_cache_.begin();
524   // Skip over any deleted entries at the start of the descriptor list.
525   while (cache_iterator != entry_cache_.end() &&
526          cache_iterator->state() != EntryState::kValid) {
527     ++cache_iterator;
528   }
529   return iterator(*this, cache_iterator);
530 }
531 
ValueSize(Key key) const532 StatusWithSize KeyValueStore::ValueSize(Key key) const {
533   PW_TRY_WITH_SIZE(CheckReadOperation(key));
534 
535   EntryMetadata metadata;
536   PW_TRY_WITH_SIZE(FindExisting(key, &metadata));
537 
538   return ValueSize(metadata);
539 }
540 
ReadEntry(const EntryMetadata & metadata,Entry & entry) const541 Status KeyValueStore::ReadEntry(const EntryMetadata& metadata,
542                                 Entry& entry) const {
543   // Try to read an entry
544   Status read_result = Status::DataLoss();
545   for (Address address : metadata.addresses()) {
546     read_result = Entry::Read(partition_, address, formats_, &entry);
547     if (read_result.ok()) {
548       return read_result;
549     }
550 
551     // Found a bad address. Set the sector as corrupt.
552     error_detected_ = true;
553     sectors_.FromAddress(address).mark_corrupt();
554   }
555 
556   PW_LOG_ERROR("No valid entries for key. Data has been lost!");
557   return read_result;
558 }
559 
FindEntry(Key key,EntryMetadata * metadata_out) const560 Status KeyValueStore::FindEntry(Key key, EntryMetadata* metadata_out) const {
561   StatusWithSize find_result =
562       entry_cache_.Find(partition_, sectors_, formats_, key, metadata_out);
563 
564   if (find_result.size() > 0u) {
565     error_detected_ = true;
566   }
567   return find_result.status();
568 }
569 
FindExisting(Key key,EntryMetadata * metadata_out) const570 Status KeyValueStore::FindExisting(Key key, EntryMetadata* metadata_out) const {
571   Status status = FindEntry(key, metadata_out);
572 
573   // If the key's hash collides with an existing key or if the key is deleted,
574   // treat it as if it is not in the KVS.
575   if (status.IsAlreadyExists() ||
576       (status.ok() && metadata_out->state() == EntryState::kDeleted)) {
577     return Status::NotFound();
578   }
579   return status;
580 }
581 
Get(Key key,const EntryMetadata & metadata,span<std::byte> value_buffer,size_t offset_bytes) const582 StatusWithSize KeyValueStore::Get(Key key,
583                                   const EntryMetadata& metadata,
584                                   span<std::byte> value_buffer,
585                                   size_t offset_bytes) const {
586   Entry entry;
587 
588   PW_TRY_WITH_SIZE(ReadEntry(metadata, entry));
589 
590   StatusWithSize result = entry.ReadValue(value_buffer, offset_bytes);
591   if (result.ok() && options_.verify_on_read && offset_bytes == 0u) {
592     Status verify_result =
593         entry.VerifyChecksum(key, value_buffer.first(result.size()));
594     if (!verify_result.ok()) {
595       std::memset(value_buffer.data(), 0, result.size());
596       return StatusWithSize(verify_result, 0);
597     }
598 
599     return StatusWithSize(verify_result, result.size());
600   }
601   return result;
602 }
603 
FixedSizeGet(Key key,void * value,size_t size_bytes) const604 Status KeyValueStore::FixedSizeGet(Key key,
605                                    void* value,
606                                    size_t size_bytes) const {
607   PW_TRY(CheckWriteOperation(key));
608 
609   EntryMetadata metadata;
610   PW_TRY(FindExisting(key, &metadata));
611 
612   return FixedSizeGet(key, metadata, value, size_bytes);
613 }
614 
FixedSizeGet(Key key,const EntryMetadata & metadata,void * value,size_t size_bytes) const615 Status KeyValueStore::FixedSizeGet(Key key,
616                                    const EntryMetadata& metadata,
617                                    void* value,
618                                    size_t size_bytes) const {
619   // Ensure that the size of the stored value matches the size of the type.
620   // Otherwise, report error. This check avoids potential memory corruption.
621   PW_TRY_ASSIGN(const size_t actual_size, ValueSize(metadata));
622 
623   if (actual_size != size_bytes) {
624     PW_LOG_DEBUG("Requested %u B read, but value is %u B",
625                  unsigned(size_bytes),
626                  unsigned(actual_size));
627     return Status::InvalidArgument();
628   }
629 
630   StatusWithSize result =
631       Get(key, metadata, span(static_cast<byte*>(value), size_bytes), 0);
632 
633   return result.status();
634 }
635 
ValueSize(const EntryMetadata & metadata) const636 StatusWithSize KeyValueStore::ValueSize(const EntryMetadata& metadata) const {
637   Entry entry;
638   PW_TRY_WITH_SIZE(ReadEntry(metadata, entry));
639 
640   return StatusWithSize(entry.value_size());
641 }
642 
CheckWriteOperation(Key key) const643 Status KeyValueStore::CheckWriteOperation(Key key) const {
644   if (InvalidKey(key)) {
645     return Status::InvalidArgument();
646   }
647 
648   // For normal write operation the KVS must be fully ready.
649   if (!initialized()) {
650     return Status::FailedPrecondition();
651   }
652   return OkStatus();
653 }
654 
CheckReadOperation(Key key) const655 Status KeyValueStore::CheckReadOperation(Key key) const {
656   if (InvalidKey(key)) {
657     return Status::InvalidArgument();
658   }
659 
660   // Operations that are explicitly read-only can be done after init() has been
661   // called but not fully ready (when needing maintenance).
662   if (initialized_ == InitializationState::kNotInitialized) {
663     return Status::FailedPrecondition();
664   }
665   return OkStatus();
666 }
667 
WriteEntryForExistingKey(EntryMetadata & metadata,EntryState new_state,Key key,span<const byte> value)668 Status KeyValueStore::WriteEntryForExistingKey(EntryMetadata& metadata,
669                                                EntryState new_state,
670                                                Key key,
671                                                span<const byte> value) {
672   // Read the original entry to get the size for sector accounting purposes.
673   Entry entry;
674   PW_TRY(ReadEntry(metadata, entry));
675 
676   return WriteEntry(key, value, new_state, &metadata, &entry);
677 }
678 
WriteEntryForNewKey(Key key,span<const byte> value)679 Status KeyValueStore::WriteEntryForNewKey(Key key, span<const byte> value) {
680   // If there is no room in the cache for a new entry, it is possible some cache
681   // entries could be freed by removing deleted keys. If deleted key removal is
682   // enabled and the KVS is configured to make all possible writes succeed,
683   // attempt heavy maintenance now.
684 #if PW_KVS_REMOVE_DELETED_KEYS_IN_HEAVY_MAINTENANCE
685   if (options_.gc_on_write == GargbageCollectOnWrite::kAsManySectorsNeeded &&
686       entry_cache_.full()) {
687     Status maintenance_status = HeavyMaintenance();
688     if (!maintenance_status.ok()) {
689       PW_LOG_WARN("KVS Maintenance failed for write: %s",
690                   maintenance_status.str());
691       return maintenance_status;
692     }
693   }
694 #endif  // PW_KVS_REMOVE_DELETED_KEYS_IN_HEAVY_MAINTENANCE
695 
696   if (entry_cache_.full()) {
697     PW_LOG_WARN(
698         "KVS full: trying to store a new entry, but can't. Have %u entries",
699         unsigned(entry_cache_.total_entries()));
700     return Status::ResourceExhausted();
701   }
702 
703   return WriteEntry(key, value, EntryState::kValid);
704 }
705 
WriteEntry(Key key,span<const byte> value,EntryState new_state,EntryMetadata * prior_metadata,const Entry * prior_entry)706 Status KeyValueStore::WriteEntry(Key key,
707                                  span<const byte> value,
708                                  EntryState new_state,
709                                  EntryMetadata* prior_metadata,
710                                  const Entry* prior_entry) {
711   // If new entry and prior entry have matching value size, state, and checksum,
712   // check if the values match. Directly compare the prior and new values
713   // because the checksum can not be depended on to establish equality, it can
714   // only be depended on to establish inequality.
715   if (prior_entry != nullptr && prior_entry->value_size() == value.size() &&
716       prior_metadata->state() == new_state &&
717       prior_entry->ValueMatches(value).ok()) {
718     // The new value matches the prior value, don't need to write anything. Just
719     // keep the existing entry.
720     PW_LOG_DEBUG("Write for key 0x%08x with matching value skipped",
721                  unsigned(prior_metadata->hash()));
722     return OkStatus();
723   }
724 
725   // List of addresses for sectors with space for this entry.
726   Address* reserved_addresses = entry_cache_.TempReservedAddressesForWrite();
727 
728   // Find addresses to write the entry to. This may involve garbage collecting
729   // one or more sectors.
730   const size_t entry_size = Entry::size(partition_, key, value);
731   PW_TRY(GetAddressesForWrite(reserved_addresses, entry_size));
732 
733   // Write the entry at the first address that was found.
734   Entry entry = CreateEntry(reserved_addresses[0], key, value, new_state);
735   PW_TRY(AppendEntry(entry, key, value));
736 
737   // After writing the first entry successfully, update the key descriptors.
738   // Once a single new the entry is written, the old entries are invalidated.
739   size_t prior_size = prior_entry != nullptr ? prior_entry->size() : 0;
740   EntryMetadata new_metadata =
741       CreateOrUpdateKeyDescriptor(entry, key, prior_metadata, prior_size);
742 
743   // Write the additional copies of the entry, if redundancy is greater than 1.
744   for (size_t i = 1; i < redundancy(); ++i) {
745     entry.set_address(reserved_addresses[i]);
746     PW_TRY(AppendEntry(entry, key, value));
747     new_metadata.AddNewAddress(reserved_addresses[i]);
748   }
749   return OkStatus();
750 }
751 
CreateOrUpdateKeyDescriptor(const Entry & entry,Key key,EntryMetadata * prior_metadata,size_t prior_size)752 KeyValueStore::EntryMetadata KeyValueStore::CreateOrUpdateKeyDescriptor(
753     const Entry& entry,
754     Key key,
755     EntryMetadata* prior_metadata,
756     size_t prior_size) {
757   // If there is no prior descriptor, create a new one.
758   if (prior_metadata == nullptr) {
759     return entry_cache_.AddNew(entry.descriptor(key), entry.address());
760   }
761 
762   return UpdateKeyDescriptor(
763       entry, entry.address(), prior_metadata, prior_size);
764 }
765 
UpdateKeyDescriptor(const Entry & entry,Address new_address,EntryMetadata * prior_metadata,size_t prior_size)766 KeyValueStore::EntryMetadata KeyValueStore::UpdateKeyDescriptor(
767     const Entry& entry,
768     Address new_address,
769     EntryMetadata* prior_metadata,
770     size_t prior_size) {
771   // Remove valid bytes for the old entry and its copies, which are now stale.
772   for (Address address : prior_metadata->addresses()) {
773     sectors_.FromAddress(address).RemoveValidBytes(prior_size);
774   }
775 
776   prior_metadata->Reset(entry.descriptor(prior_metadata->hash()), new_address);
777   return *prior_metadata;
778 }
779 
GetAddressesForWrite(Address * write_addresses,size_t write_size)780 Status KeyValueStore::GetAddressesForWrite(Address* write_addresses,
781                                            size_t write_size) {
782   for (size_t i = 0; i < redundancy(); i++) {
783     SectorDescriptor* sector;
784     PW_TRY(GetSectorForWrite(&sector, write_size, span(write_addresses, i)));
785     write_addresses[i] = sectors_.NextWritableAddress(*sector);
786 
787     PW_LOG_DEBUG("Found space for entry in sector %u at address %u",
788                  sectors_.Index(sector),
789                  unsigned(write_addresses[i]));
790   }
791 
792   return OkStatus();
793 }
794 
795 // Finds a sector to use for writing a new entry to. Does automatic garbage
796 // collection if needed and allowed.
797 //
798 //                 OK: Sector found with needed space.
799 // RESOURCE_EXHAUSTED: No sector available with the needed space.
GetSectorForWrite(SectorDescriptor ** sector,size_t entry_size,span<const Address> reserved_addresses)800 Status KeyValueStore::GetSectorForWrite(
801     SectorDescriptor** sector,
802     size_t entry_size,
803     span<const Address> reserved_addresses) {
804   Status result = sectors_.FindSpace(sector, entry_size, reserved_addresses);
805 
806   size_t gc_sector_count = 0;
807   bool do_auto_gc = options_.gc_on_write != GargbageCollectOnWrite::kDisabled;
808 
809   // Do garbage collection as needed, so long as policy allows.
810   while (result.IsResourceExhausted() && do_auto_gc) {
811     if (options_.gc_on_write == GargbageCollectOnWrite::kOneSector) {
812       // If GC config option is kOneSector clear the flag to not do any more
813       // GC after this try.
814       do_auto_gc = false;
815     }
816     // Garbage collect and then try again to find the best sector.
817     Status gc_status = GarbageCollect(reserved_addresses);
818     if (!gc_status.ok()) {
819       if (gc_status.IsNotFound()) {
820         // Not enough space, and no reclaimable bytes, this KVS is full!
821         return Status::ResourceExhausted();
822       }
823       return gc_status;
824     }
825 
826     result = sectors_.FindSpace(sector, entry_size, reserved_addresses);
827 
828     gc_sector_count++;
829     // Allow total sectors + 2 number of GC cycles so that once reclaimable
830     // bytes in all the sectors have been reclaimed can try and free up space by
831     // moving entries for keys other than the one being worked on in to sectors
832     // that have copies of the key trying to be written.
833     if (gc_sector_count > (partition_.sector_count() + 2)) {
834       PW_LOG_ERROR("Did more GC sectors than total sectors!!!!");
835       return Status::ResourceExhausted();
836     }
837   }
838 
839   if (!result.ok()) {
840     PW_LOG_WARN("Unable to find sector to write %u B", unsigned(entry_size));
841   }
842   return result;
843 }
844 
MarkSectorCorruptIfNotOk(Status status,SectorDescriptor * sector)845 Status KeyValueStore::MarkSectorCorruptIfNotOk(Status status,
846                                                SectorDescriptor* sector) {
847   if (!status.ok()) {
848     PW_LOG_DEBUG("  Sector %u corrupt", sectors_.Index(sector));
849     sector->mark_corrupt();
850     error_detected_ = true;
851   }
852   return status;
853 }
854 
AppendEntry(const Entry & entry,Key key,span<const byte> value)855 Status KeyValueStore::AppendEntry(const Entry& entry,
856                                   Key key,
857                                   span<const byte> value) {
858   const StatusWithSize result = entry.Write(key, value);
859 
860   SectorDescriptor& sector = sectors_.FromAddress(entry.address());
861 
862   if (!result.ok()) {
863     PW_LOG_ERROR("Failed to write %u bytes at %#x. %u actually written",
864                  unsigned(entry.size()),
865                  unsigned(entry.address()),
866                  unsigned(result.size()));
867     PW_TRY(MarkSectorCorruptIfNotOk(result.status(), &sector));
868   }
869 
870   if (options_.verify_on_write) {
871     PW_TRY(MarkSectorCorruptIfNotOk(entry.VerifyChecksumInFlash(), &sector));
872   }
873 
874   sector.RemoveWritableBytes(result.size());
875   sector.AddValidBytes(result.size());
876   return OkStatus();
877 }
878 
CopyEntryToSector(Entry & entry,SectorDescriptor * new_sector,Address new_address)879 StatusWithSize KeyValueStore::CopyEntryToSector(Entry& entry,
880                                                 SectorDescriptor* new_sector,
881                                                 Address new_address) {
882   const StatusWithSize result = entry.Copy(new_address);
883 
884   PW_TRY_WITH_SIZE(MarkSectorCorruptIfNotOk(result.status(), new_sector));
885 
886   if (options_.verify_on_write) {
887     Entry new_entry;
888     PW_TRY_WITH_SIZE(MarkSectorCorruptIfNotOk(
889         Entry::Read(partition_, new_address, formats_, &new_entry),
890         new_sector));
891     // TODO(davidrogers): add test that catches doing the verify on the old
892     // entry.
893     PW_TRY_WITH_SIZE(MarkSectorCorruptIfNotOk(new_entry.VerifyChecksumInFlash(),
894                                               new_sector));
895   }
896   // Entry was written successfully; update descriptor's address and the sector
897   // descriptors to reflect the new entry.
898   new_sector->RemoveWritableBytes(result.size());
899   new_sector->AddValidBytes(result.size());
900 
901   return result;
902 }
903 
RelocateEntry(const EntryMetadata & metadata,KeyValueStore::Address & address,span<const Address> reserved_addresses)904 Status KeyValueStore::RelocateEntry(const EntryMetadata& metadata,
905                                     KeyValueStore::Address& address,
906                                     span<const Address> reserved_addresses) {
907   Entry entry;
908   PW_TRY(ReadEntry(metadata, entry));
909 
910   // Find a new sector for the entry and write it to the new location. For
911   // relocation the find should not not be a sector already containing the key
912   // but can be the always empty sector, since this is part of the GC process
913   // that will result in a new empty sector. Also find a sector that does not
914   // have reclaimable space (mostly for the full GC, where that would result in
915   // an immediate extra relocation).
916   SectorDescriptor* new_sector;
917 
918   PW_TRY(sectors_.FindSpaceDuringGarbageCollection(
919       &new_sector, entry.size(), metadata.addresses(), reserved_addresses));
920 
921   Address new_address = sectors_.NextWritableAddress(*new_sector);
922   PW_TRY_ASSIGN(const size_t result_size,
923                 CopyEntryToSector(entry, new_sector, new_address));
924   sectors_.FromAddress(address).RemoveValidBytes(result_size);
925   address = new_address;
926 
927   return OkStatus();
928 }
929 
FullMaintenanceHelper(MaintenanceType maintenance_type)930 Status KeyValueStore::FullMaintenanceHelper(MaintenanceType maintenance_type) {
931   if (initialized_ == InitializationState::kNotInitialized) {
932     return Status::FailedPrecondition();
933   }
934 
935   // Full maintenance can be a potentially heavy operation, and should be
936   // relatively infrequent, so log start/end at INFO level.
937   PW_LOG_INFO("Beginning full maintenance");
938   CheckForErrors();
939 
940   // Step 1: Repair errors
941   if (error_detected_) {
942     PW_TRY(Repair());
943   }
944 
945   // Step 2: Make sure all the entries are on the primary format.
946   StatusWithSize update_status = UpdateEntriesToPrimaryFormat();
947   Status overall_status = update_status.status();
948 
949   if (!overall_status.ok()) {
950     PW_LOG_ERROR("Failed to update all entries to the primary format");
951   }
952 
953   SectorDescriptor* sector = sectors_.last_new();
954 
955   // Calculate number of bytes for the threshold.
956   size_t threshold_bytes =
957       (partition_.size_bytes() * kGcUsageThresholdPercentage) / 100;
958 
959   // Is bytes in use over the threshold.
960   StorageStats stats = GetStorageStats();
961   bool over_usage_threshold = stats.in_use_bytes > threshold_bytes;
962   bool heavy = (maintenance_type == MaintenanceType::kHeavy);
963   bool force_gc = heavy || over_usage_threshold || (update_status.size() > 0);
964 
965   auto do_garbage_collect_pass = [&]() {
966     // TODO(drempel): look in to making an iterator method for cycling through
967     // sectors starting from last_new_sector_.
968     Status gc_status;
969     for (size_t j = 0; j < sectors_.size(); j++) {
970       sector += 1;
971       if (sector == sectors_.end()) {
972         sector = sectors_.begin();
973       }
974 
975       if (sector->RecoverableBytes(partition_.sector_size_bytes()) > 0 &&
976           (force_gc || sector->valid_bytes() == 0)) {
977         gc_status = GarbageCollectSector(*sector, {});
978         if (!gc_status.ok()) {
979           PW_LOG_ERROR("Failed to garbage collect all sectors");
980           break;
981         }
982       }
983     }
984     if (overall_status.ok()) {
985       overall_status = gc_status;
986     }
987   };
988 
989   // Step 3: Do full garbage collect pass for all sectors. This will erase all
990   // old/state entries from flash and leave only current/valid entries.
991   do_garbage_collect_pass();
992 
993 #if PW_KVS_REMOVE_DELETED_KEYS_IN_HEAVY_MAINTENANCE
994   // Step 4: (if heavy maintenance) garbage collect all the deleted keys.
995   if (heavy) {
996     // If enabled, remove deleted keys from the entry cache, including freeing
997     // sector bytes used by those keys. This must only be done directly after a
998     // full garbage collection, otherwise the current deleted entry could be
999     // garbage collected before the older stale entry producing a window for an
1000     // invalid/corrupted KVS state if there was a power-fault, crash or other
1001     // interruption.
1002     overall_status.Update(RemoveDeletedKeyEntries());
1003 
1004     // Do another garbage collect pass that will fully remove the deleted keys
1005     // from flash. Garbage collect will only touch sectors that have something
1006     // to garbage collect, which in this case is only sectors containing deleted
1007     // keys.
1008     do_garbage_collect_pass();
1009   }
1010 #endif  // PW_KVS_REMOVE_DELETED_KEYS_IN_HEAVY_MAINTENANCE
1011 
1012   if (overall_status.ok()) {
1013     PW_LOG_INFO("Full maintenance complete");
1014   } else {
1015     PW_LOG_ERROR("Full maintenance finished with some errors");
1016   }
1017   return overall_status;
1018 }
1019 
PartialMaintenance()1020 Status KeyValueStore::PartialMaintenance() {
1021   if (initialized_ == InitializationState::kNotInitialized) {
1022     return Status::FailedPrecondition();
1023   }
1024 
1025   CheckForErrors();
1026   // Do automatic repair, if KVS options allow for it.
1027   if (error_detected_ && options_.recovery != ErrorRecovery::kManual) {
1028     PW_TRY(Repair());
1029   }
1030   return GarbageCollect(span<const Address>());
1031 }
1032 
GarbageCollect(span<const Address> reserved_addresses)1033 Status KeyValueStore::GarbageCollect(span<const Address> reserved_addresses) {
1034   PW_LOG_DEBUG("Garbage Collect a single sector");
1035   for ([[maybe_unused]] Address address : reserved_addresses) {
1036     PW_LOG_DEBUG("   Avoid address %u", unsigned(address));
1037   }
1038 
1039   // Step 1: Find the sector to garbage collect
1040   SectorDescriptor* sector_to_gc =
1041       sectors_.FindSectorToGarbageCollect(reserved_addresses);
1042 
1043   if (sector_to_gc == nullptr) {
1044     // Nothing to GC.
1045     return Status::NotFound();
1046   }
1047 
1048   // Step 2: Garbage collect the selected sector.
1049   return GarbageCollectSector(*sector_to_gc, reserved_addresses);
1050 }
1051 
RelocateKeyAddressesInSector(SectorDescriptor & sector_to_gc,const EntryMetadata & metadata,span<const Address> reserved_addresses)1052 Status KeyValueStore::RelocateKeyAddressesInSector(
1053     SectorDescriptor& sector_to_gc,
1054     const EntryMetadata& metadata,
1055     span<const Address> reserved_addresses) {
1056   for (FlashPartition::Address& address : metadata.addresses()) {
1057     if (sectors_.AddressInSector(sector_to_gc, address)) {
1058       PW_LOG_DEBUG("  Relocate entry for Key 0x%08" PRIx32 ", sector %u",
1059                    metadata.hash(),
1060                    sectors_.Index(sectors_.FromAddress(address)));
1061       PW_TRY(RelocateEntry(metadata, address, reserved_addresses));
1062     }
1063   }
1064 
1065   return OkStatus();
1066 }
1067 
GarbageCollectSector(SectorDescriptor & sector_to_gc,span<const Address> reserved_addresses)1068 Status KeyValueStore::GarbageCollectSector(
1069     SectorDescriptor& sector_to_gc, span<const Address> reserved_addresses) {
1070   PW_LOG_DEBUG("  Garbage Collect sector %u", sectors_.Index(sector_to_gc));
1071 
1072   // Step 1: Move any valid entries in the GC sector to other sectors
1073   if (sector_to_gc.valid_bytes() != 0) {
1074     for (EntryMetadata& metadata : entry_cache_) {
1075       PW_TRY(RelocateKeyAddressesInSector(
1076           sector_to_gc, metadata, reserved_addresses));
1077     }
1078   }
1079 
1080   if (sector_to_gc.valid_bytes() != 0) {
1081     PW_LOG_ERROR(
1082         "  Failed to relocate valid entries from sector being garbage "
1083         "collected, %u valid bytes remain",
1084         unsigned(sector_to_gc.valid_bytes()));
1085     return Status::Internal();
1086   }
1087 
1088   // Step 2: Reinitialize the sector
1089   if (!sector_to_gc.Empty(partition_.sector_size_bytes())) {
1090     sector_to_gc.mark_corrupt();
1091     internal_stats_.sector_erase_count++;
1092     PW_TRY(partition_.Erase(sectors_.BaseAddress(sector_to_gc), 1));
1093     sector_to_gc.set_writable_bytes(partition_.sector_size_bytes());
1094   }
1095 
1096   PW_LOG_DEBUG("  Garbage Collect sector %u complete",
1097                sectors_.Index(sector_to_gc));
1098   return OkStatus();
1099 }
1100 
UpdateEntriesToPrimaryFormat()1101 StatusWithSize KeyValueStore::UpdateEntriesToPrimaryFormat() {
1102   size_t entries_updated = 0;
1103   for (EntryMetadata& prior_metadata : entry_cache_) {
1104     Entry entry;
1105     PW_TRY_WITH_SIZE(ReadEntry(prior_metadata, entry));
1106     if (formats_.primary().magic == entry.magic()) {
1107       // Ignore entries that are already on the primary format.
1108       continue;
1109     }
1110 
1111     PW_LOG_DEBUG(
1112         "Updating entry 0x%08x from old format [0x%08x] to new format "
1113         "[0x%08x]",
1114         unsigned(prior_metadata.hash()),
1115         unsigned(entry.magic()),
1116         unsigned(formats_.primary().magic));
1117 
1118     entries_updated++;
1119 
1120     last_transaction_id_ += 1;
1121     PW_TRY_WITH_SIZE(entry.Update(formats_.primary(), last_transaction_id_));
1122 
1123     // List of addresses for sectors with space for this entry.
1124     Address* reserved_addresses = entry_cache_.TempReservedAddressesForWrite();
1125 
1126     // Find addresses to write the entry to. This may involve garbage collecting
1127     // one or more sectors.
1128     PW_TRY_WITH_SIZE(GetAddressesForWrite(reserved_addresses, entry.size()));
1129 
1130     PW_TRY_WITH_SIZE(
1131         CopyEntryToSector(entry,
1132                           &sectors_.FromAddress(reserved_addresses[0]),
1133                           reserved_addresses[0]));
1134 
1135     // After writing the first entry successfully, update the key descriptors.
1136     // Once a single new the entry is written, the old entries are invalidated.
1137     EntryMetadata new_metadata = UpdateKeyDescriptor(
1138         entry, reserved_addresses[0], &prior_metadata, entry.size());
1139 
1140     // Write the additional copies of the entry, if redundancy is greater
1141     // than 1.
1142     for (size_t i = 1; i < redundancy(); ++i) {
1143       PW_TRY_WITH_SIZE(
1144           CopyEntryToSector(entry,
1145                             &sectors_.FromAddress(reserved_addresses[i]),
1146                             reserved_addresses[i]));
1147       new_metadata.AddNewAddress(reserved_addresses[i]);
1148     }
1149   }
1150 
1151   return StatusWithSize(entries_updated);
1152 }
1153 
1154 // Add any missing redundant entries/copies for a key.
AddRedundantEntries(EntryMetadata & metadata)1155 Status KeyValueStore::AddRedundantEntries(EntryMetadata& metadata) {
1156   Entry entry;
1157   PW_TRY(ReadEntry(metadata, entry));
1158   PW_TRY(entry.VerifyChecksumInFlash());
1159 
1160   while (metadata.addresses().size() < redundancy()) {
1161     SectorDescriptor* new_sector;
1162     PW_TRY(GetSectorForWrite(&new_sector, entry.size(), metadata.addresses()));
1163 
1164     Address new_address = sectors_.NextWritableAddress(*new_sector);
1165     PW_TRY(CopyEntryToSector(entry, new_sector, new_address));
1166 
1167     metadata.AddNewAddress(new_address);
1168   }
1169   return OkStatus();
1170 }
1171 
RepairCorruptSectors()1172 Status KeyValueStore::RepairCorruptSectors() {
1173   // Try to GC each corrupt sector, even if previous sectors fail. If GC of a
1174   // sector failed on the first pass, then do a second pass, since a later
1175   // sector might have cleared up space or otherwise unblocked the earlier
1176   // failed sector.
1177   Status repair_status = OkStatus();
1178 
1179   size_t loop_count = 0;
1180   do {
1181     loop_count++;
1182     // Error of RESOURCE_EXHAUSTED indicates no space found for relocation.
1183     // Reset back to OK for the next pass.
1184     if (repair_status.IsResourceExhausted()) {
1185       repair_status = OkStatus();
1186     }
1187 
1188     PW_LOG_DEBUG("   Pass %u", unsigned(loop_count));
1189     for (SectorDescriptor& sector : sectors_) {
1190       if (sector.corrupt()) {
1191         PW_LOG_DEBUG("   Found sector %u with corruption",
1192                      sectors_.Index(sector));
1193         Status sector_status = GarbageCollectSector(sector, {});
1194         if (sector_status.ok()) {
1195           internal_stats_.corrupt_sectors_recovered += 1;
1196         } else if (repair_status.ok() || repair_status.IsResourceExhausted()) {
1197           repair_status = sector_status;
1198         }
1199       }
1200     }
1201     PW_LOG_DEBUG("   Pass %u complete", unsigned(loop_count));
1202   } while (!repair_status.ok() && loop_count < 2);
1203 
1204   return repair_status;
1205 }
1206 
EnsureFreeSectorExists()1207 Status KeyValueStore::EnsureFreeSectorExists() {
1208   Status repair_status = OkStatus();
1209   bool empty_sector_found = false;
1210 
1211   PW_LOG_DEBUG("   Find empty sector");
1212   for (SectorDescriptor& sector : sectors_) {
1213     if (sector.Empty(partition_.sector_size_bytes())) {
1214       empty_sector_found = true;
1215       PW_LOG_DEBUG("   Empty sector found");
1216       break;
1217     }
1218   }
1219   if (empty_sector_found == false) {
1220     PW_LOG_DEBUG("   No empty sector found, attempting to GC a free sector");
1221     Status sector_status = GarbageCollect(span<const Address, 0>());
1222     if (repair_status.ok() && !sector_status.ok()) {
1223       PW_LOG_DEBUG("   Unable to free an empty sector");
1224       repair_status = sector_status;
1225     }
1226   }
1227 
1228   return repair_status;
1229 }
1230 
EnsureEntryRedundancy()1231 Status KeyValueStore::EnsureEntryRedundancy() {
1232   Status repair_status = OkStatus();
1233 
1234   if (redundancy() == 1) {
1235     PW_LOG_DEBUG("   Redundancy not in use, nothting to check");
1236     return OkStatus();
1237   }
1238 
1239   PW_LOG_DEBUG(
1240       "   Write any needed additional duplicate copies of keys to fulfill %u"
1241       " redundancy",
1242       unsigned(redundancy()));
1243   for (EntryMetadata& metadata : entry_cache_) {
1244     if (metadata.addresses().size() >= redundancy()) {
1245       continue;
1246     }
1247 
1248     PW_LOG_DEBUG("   Key with %u of %u copies found, adding missing copies",
1249                  unsigned(metadata.addresses().size()),
1250                  unsigned(redundancy()));
1251     Status fill_status = AddRedundantEntries(metadata);
1252     if (fill_status.ok()) {
1253       internal_stats_.missing_redundant_entries_recovered += 1;
1254       PW_LOG_DEBUG("   Key missing copies added");
1255     } else {
1256       PW_LOG_DEBUG("   Failed to add key missing copies");
1257       if (repair_status.ok()) {
1258         repair_status = fill_status;
1259       }
1260     }
1261   }
1262 
1263   return repair_status;
1264 }
1265 
FixErrors()1266 Status KeyValueStore::FixErrors() {
1267   PW_LOG_DEBUG("Fixing KVS errors");
1268 
1269   // Step 1: Garbage collect any sectors marked as corrupt.
1270   Status overall_status = RepairCorruptSectors();
1271 
1272   // Step 2: Make sure there is at least 1 empty sector. This needs to be a
1273   // seperate check of sectors from step 1, because a found empty sector might
1274   // get written to by a later GC that fails and does not result in a free
1275   // sector.
1276   Status repair_status = EnsureFreeSectorExists();
1277   if (overall_status.ok()) {
1278     overall_status = repair_status;
1279   }
1280 
1281   // Step 3: Make sure each stored key has the full number of redundant
1282   // entries.
1283   repair_status = EnsureEntryRedundancy();
1284   if (overall_status.ok()) {
1285     overall_status = repair_status;
1286   }
1287 
1288   if (overall_status.ok()) {
1289     error_detected_ = false;
1290     initialized_ = InitializationState::kReady;
1291   }
1292   return overall_status;
1293 }
1294 
Repair()1295 Status KeyValueStore::Repair() {
1296   // If errors have been detected, just reinit the KVS metadata. This does a
1297   // full deep error check and any needed repairs. Then repair any errors.
1298   PW_LOG_INFO("Starting KVS repair");
1299 
1300   PW_LOG_DEBUG("Reinitialize KVS metadata");
1301   InitializeMetadata()
1302       .IgnoreError();  // TODO: b/242598609 - Handle Status properly
1303 
1304   return FixErrors();
1305 }
1306 
CreateEntry(Address address,Key key,span<const byte> value,EntryState state)1307 KeyValueStore::Entry KeyValueStore::CreateEntry(Address address,
1308                                                 Key key,
1309                                                 span<const byte> value,
1310                                                 EntryState state) {
1311   // Always bump the transaction ID when creating a new entry.
1312   //
1313   // Burning transaction IDs prevents inconsistencies between flash and memory
1314   // that which could happen if a write succeeds, but for some reason the read
1315   // and verify step fails. Here's how this would happen:
1316   //
1317   //   1. The entry is written but for some reason the flash reports failure OR
1318   //      The write succeeds, but the read / verify operation fails.
1319   //   2. The transaction ID is NOT incremented, because of the failure
1320   //   3. (later) A new entry is written, re-using the transaction ID (oops)
1321   //
1322   // By always burning transaction IDs, the above problem can't happen.
1323   last_transaction_id_ += 1;
1324 
1325   if (state == EntryState::kDeleted) {
1326     return Entry::Tombstone(
1327         partition_, address, formats_.primary(), key, last_transaction_id_);
1328   }
1329   return Entry::Valid(partition_,
1330                       address,
1331                       formats_.primary(),
1332                       key,
1333                       value,
1334                       last_transaction_id_);
1335 }
1336 
LogDebugInfo() const1337 void KeyValueStore::LogDebugInfo() const {
1338   const size_t sector_size_bytes = partition_.sector_size_bytes();
1339   PW_LOG_DEBUG(
1340       "====================== KEY VALUE STORE DUMP =========================");
1341   PW_LOG_DEBUG(" ");
1342   PW_LOG_DEBUG("Flash partition:");
1343   PW_LOG_DEBUG("  Sector count     = %u", unsigned(partition_.sector_count()));
1344   PW_LOG_DEBUG("  Sector max count = %u", unsigned(sectors_.max_size()));
1345   PW_LOG_DEBUG("  Sectors in use   = %u", unsigned(sectors_.size()));
1346   PW_LOG_DEBUG("  Sector size      = %u", unsigned(sector_size_bytes));
1347   PW_LOG_DEBUG("  Total size       = %u", unsigned(partition_.size_bytes()));
1348   PW_LOG_DEBUG("  Alignment        = %u",
1349                unsigned(partition_.alignment_bytes()));
1350   PW_LOG_DEBUG(" ");
1351   PW_LOG_DEBUG("Key descriptors:");
1352   PW_LOG_DEBUG("  Entry count     = %u",
1353                unsigned(entry_cache_.total_entries()));
1354   PW_LOG_DEBUG("  Max entry count = %u", unsigned(entry_cache_.max_entries()));
1355   PW_LOG_DEBUG(" ");
1356   PW_LOG_DEBUG("      #     hash        version    address   address (hex)");
1357   size_t count = 0;
1358   for (const EntryMetadata& metadata : entry_cache_) {
1359     PW_LOG_DEBUG("   |%3zu: | %8zx  |%8zu  | %8zu | %8zx",
1360                  count++,
1361                  size_t(metadata.hash()),
1362                  size_t(metadata.transaction_id()),
1363                  size_t(metadata.first_address()),
1364                  size_t(metadata.first_address()));
1365   }
1366   PW_LOG_DEBUG(" ");
1367 
1368   PW_LOG_DEBUG("Sector descriptors:");
1369   PW_LOG_DEBUG("      #     tail free  valid    has_space");
1370   for (const SectorDescriptor& sd : sectors_) {
1371     PW_LOG_DEBUG("   |%3u: | %8zu  |%8zu  | %s",
1372                  sectors_.Index(sd),
1373                  size_t(sd.writable_bytes()),
1374                  sd.valid_bytes(),
1375                  sd.writable_bytes() ? "YES" : "");
1376   }
1377   PW_LOG_DEBUG(" ");
1378 
1379   // TODO(keir): This should stop logging after some threshold.
1380   // size_t dumped_bytes = 0;
1381   PW_LOG_DEBUG("Sector raw data:");
1382   for (size_t sector_id = 0; sector_id < sectors_.size(); ++sector_id) {
1383     // Read sector data. Yes, this will blow the stack on embedded.
1384     std::array<byte, 500> raw_sector_data;  // TODO!!!
1385     [[maybe_unused]] StatusWithSize sws =
1386         partition_.Read(sector_id * sector_size_bytes, raw_sector_data);
1387     PW_LOG_DEBUG("Read: %u bytes", unsigned(sws.size()));
1388 
1389     PW_LOG_DEBUG("  base    addr  offs   0  1  2  3  4  5  6  7");
1390     for (size_t i = 0; i < sector_size_bytes; i += 8) {
1391       PW_LOG_DEBUG("  %3zu %8zx %5zu | %02x %02x %02x %02x %02x %02x %02x %02x",
1392                    sector_id,
1393                    (sector_id * sector_size_bytes) + i,
1394                    i,
1395                    static_cast<unsigned int>(raw_sector_data[i + 0]),
1396                    static_cast<unsigned int>(raw_sector_data[i + 1]),
1397                    static_cast<unsigned int>(raw_sector_data[i + 2]),
1398                    static_cast<unsigned int>(raw_sector_data[i + 3]),
1399                    static_cast<unsigned int>(raw_sector_data[i + 4]),
1400                    static_cast<unsigned int>(raw_sector_data[i + 5]),
1401                    static_cast<unsigned int>(raw_sector_data[i + 6]),
1402                    static_cast<unsigned int>(raw_sector_data[i + 7]));
1403 
1404       // TODO(keir): Fix exit condition.
1405       if (i > 128) {
1406         break;
1407       }
1408     }
1409     PW_LOG_DEBUG(" ");
1410   }
1411 
1412   PW_LOG_DEBUG(
1413       "////////////////////// KEY VALUE STORE DUMP END /////////////////////");
1414 }
1415 
LogSectors() const1416 void KeyValueStore::LogSectors() const {
1417   PW_LOG_DEBUG("Sector descriptors: count %u", unsigned(sectors_.size()));
1418   for (auto& sector : sectors_) {
1419     PW_LOG_DEBUG(
1420         "  - Sector %u: valid %u, recoverable %u, free %u",
1421         sectors_.Index(sector),
1422         unsigned(sector.valid_bytes()),
1423         unsigned(sector.RecoverableBytes(partition_.sector_size_bytes())),
1424         unsigned(sector.writable_bytes()));
1425   }
1426 }
1427 
LogKeyDescriptor() const1428 void KeyValueStore::LogKeyDescriptor() const {
1429   PW_LOG_DEBUG("Key descriptors: count %u",
1430                unsigned(entry_cache_.total_entries()));
1431   for (const EntryMetadata& metadata : entry_cache_) {
1432     PW_LOG_DEBUG("  - Key: %s, hash %#x, transaction ID %u, first address %#x",
1433                  metadata.state() == EntryState::kDeleted ? "Deleted" : "Valid",
1434                  unsigned(metadata.hash()),
1435                  unsigned(metadata.transaction_id()),
1436                  unsigned(metadata.first_address()));
1437   }
1438 }
1439 
1440 }  // namespace pw::kvs
1441