[Mono-dev] do_rehash race

[Rodrigo Kumpera]

Good catch!

That's indeed a bug.

On Thu, May 28, 2015 at 2:35 PM, Neale Ferguson <neale at sinenomine.net>
wrote:

> Hi,
>  When a hash table exceeds a threshold a rehash operation is triggered. At
> the moment the new table is allocated and its address placed in the table
> field of the structure. The do_rehash also then copies the entries from
> the old table to the new. However, if there is another thread active that
> is doing lookups then there is a window where the new table is still being
> filled such that a lookup can fail. This is because the new table is made
> active before it has been copied. This proposed patch will fill the new
> table before swapping the old for the new table in the hash structure.
>
> Neale
>
> @@ -194,24 +196,24 @@ do_rehash (void *_data)
>         Slot **table;
>
>         /* printf ("Resizing diff=%d slots=%d\n", hash->in_use -
> hash->last_rehash, hash->table_size); */
> -       hash->last_rehash = hash->table_size;
>         current_size = hash->table_size;
> -       hash->table_size = data->new_size;
>         /* printf ("New size: %d\n", hash->table_size); */
>         table = hash->table;
> -       hash->table = data->table;
>
>         for (i = 0; i < current_size; i++){
>                 Slot *s, *next;
>
>                 for (s = table [i]; s != NULL; s = next){
> -                       guint hashcode = ((*hash->hash_func) (s->key)) %
> hash->table_size;
> +                       guint hashcode = ((*hash->hash_func) (s->key)) %
> data->new_size;
>                         next = s->next;
>
> -                       s->next = hash->table [hashcode];
> -                       hash->table [hashcode] = s;
> +                       s->next = data->table [hashcode];
> +                       data->table [hashcode] = s;
>                 }
>         }
> +       hash->table_size = data->new_size;
> +       hash->last_rehash = hash->table_size;
> +       hash->table = data->table;
>         return table;
>  }
>
>
> Neale
>
> _______________________________________________
> Mono-devel-list mailing list
> Mono-devel-list at lists.ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-devel-list
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ximian.com/pipermail/mono-devel-list/attachments/20150528/f9edaf91/attachment-0001.html>