Journal bucket is a circular buffer, the bucket can be like YYYNNNYY, which means the first valid journal in the 7th bucket, and the latest valid journal in third bucket, in this case, if we do not try we the zero index first, We may get a valid journal in the 7th bucket, then we call find_next_bit(bitmap,ca->sb.njournal_buckets, l + 1) to get the first invalid bucket after the 7th bucket, because all these buckets is valid, so no bit 1 in bitmap, thus find_next_bit() function would return with ca->sb.njournal_buckets (8). So, after that, bcache only read journal in 7th and 8the bucket, the first to the third buckets are lost. So, it is important to let developer know that, we need to try the zero index at first in the hash-search, and avoid any breaks in future's code modification. [ML: Fixed whitespace & formatting & file permissions]
Unified · Split
| @@ -170,6 +170,11 @@ int bch_journal_read(struct cache_set *c, struct list_head *list) · 169 unmodified lines | ||
| 170 | 170 | * find a sequence of buckets with valid journal entries |
| 171 | 171 | */ |
| 172 | 172 | for (i = 0; i < ca->sb.njournal_buckets; i++) { |
| 173 | + /* | |
| 174 | + * We must try the index l with ZERO first for | |
| 175 | + * correctness due to the scenario that the journal | |
| 176 | + * bucket is circular buffer which might have wrapped | |
| 177 | + */ | |
| 173 | 178 | l = (i * 2654435769U) % ca->sb.njournal_buckets; |
| 174 | 179 | |
| 175 | 180 | if (test_bit(l, bitmap)) |