On Tuesday 30 November 2004 5:05 pm, Alasdair G Kergon wrote: > On Mon, Nov 29, 2004 at 03:29:40AM +0100, Adrian Bunk wrote: > > Please apply or comment on it. > > Please check *why* the functions aren't used first. > > e.g. An alloc function with a corresponding free that > never gets called suggests a leak to me... That one isn't a leak (referring to "kmem_cache_t *exception_cache" in dm-snap.c). Items are allocated from this cache using alloc_exception(). This can happen either when an existing snapshot is activated and it's exception table is read from disk into an in-memory hash-table, or when a copy-on-write completes and a new exception is added to this hash-table. As long as the snapshot is active, this hash-table remains in memory and items cannot be removed from it. When the snapshot is deactivated, we call exit_exception_table() and pass it a pointer to this hash-table and exception_cache. This routine calls kmem_cache_free() directly instead of using the free_exception() routine. The reason it doesn't use free_exception() is that exit_exception_table() is used to tear down two different but somewhat similar hash-tables, each of which uses a different kmem_cache_t. So, it may be nice to keep the symmetric routines defined (alloc_exception() and free_exception()), but Adrian is correct in that the later is not being used, and it really can't be used without some more significant code changes. As for bs_bio_init(), it can be safely removed. It was just a duplicate of bio_init() from fs/bio.c. The use of that call in dm-io.c was changed to use bio_init(), but apparently the routine itself was never removed. -- Kevin Corry kevcorry@xxxxxxxxxx http://evms.sourceforge.net/