From: Petr Rockai Date: Mon, 06 Apr 2009 17:04:08 +0200 Subject: [PATCH 1/6] vg mempool: print pool leaks (debug helper) In-Reply-To: <49D9BDF5.6010205@redhat.com> (Milan Broz's message of "Mon, 06 Apr 2009 10:31:49 +0200") References: <49D9BDF5.6010205@redhat.com> Message-ID: <87r605bz87.fsf@eriador.mornfall.net> List-Id: To: lvm-devel@redhat.com MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Milan Broz writes: > Signed-off-by: Milan Broz One points needs to be kept in mind wrt. this patch -- it introduces a global list, making pool creation/destruction thread-unsafe (this *right now* does not hurt anyone, I believe, but it might so in future). There are some minor comments inline in the patch, see below. Other than that, Acked-By: Petr Ro?kai > diff --git a/libdm/ioctl/libdm-iface.c b/libdm/ioctl/libdm-iface.c > index fe86494..18c2d14 100644 > --- a/libdm/ioctl/libdm-iface.c > +++ b/libdm/ioctl/libdm-iface.c > @@ -1817,12 +1817,15 @@ void dm_lib_release(void) > update_devs(); > } > > +void dm_pools_dump(void); A better name might be in place. (dm_pools_print_leaks? dm_pools_announce_leaks? dm_pools_check_leaks? pick your favourite...) > + > void dm_lib_exit(void) > { > dm_lib_release(); > if (_dm_bitset) > dm_bitset_destroy(_dm_bitset); > _dm_bitset = NULL; > + dm_pools_dump(); > dm_dump_memory(); > _version_ok = 1; > _version_checked = 0; > diff --git a/libdm/mm/pool-debug.c b/libdm/mm/pool-debug.c > index 1c3d999..f775dea 100644 > --- a/libdm/mm/pool-debug.c > +++ b/libdm/mm/pool-debug.c > @@ -30,7 +30,9 @@ typedef struct { > } pool_stats; > > struct dm_pool { > + struct dm_list list; > const char *name; > + void *orig_pool; /* to pair it with first allocation call */ > > int begun; > struct block *object; > @@ -65,10 +67,13 @@ struct dm_pool *dm_pool_create(const char *name, size_t chunk_hint) > mem->stats.bytes = 0; > mem->stats.maxbytes = 0; > > + mem->orig_pool = mem; > + > #ifdef DEBUG_POOL > log_debug("Created mempool %s", name); > #endif > > + dm_list_add(&_dm_pools, &mem->list); > return mem; > } > > @@ -103,6 +108,7 @@ void dm_pool_destroy(struct dm_pool *p) > { > _pool_stats(p, "Destroying"); > _free_blocks(p, p->blocks); > + dm_list_del(&p->list); > dm_free(p); > } > > diff --git a/libdm/mm/pool-fast.c b/libdm/mm/pool-fast.c > index 9d9f808..d34cd9a 100644 > --- a/libdm/mm/pool-fast.c > +++ b/libdm/mm/pool-fast.c > @@ -21,6 +21,7 @@ struct chunk { > }; > > struct dm_pool { > + struct dm_list list; > struct chunk *chunk, *spare_chunk; /* spare_chunk is a one entry free > list to stop 'bobbling' */ > size_t chunk_size; > @@ -51,6 +52,7 @@ struct dm_pool *dm_pool_create(const char *name, size_t chunk_hint) > while (new_size < p->chunk_size) > new_size <<= 1; > p->chunk_size = new_size; > + dm_list_add(&_dm_pools, &p->list); > return p; > } > > @@ -65,6 +67,7 @@ void dm_pool_destroy(struct dm_pool *p) > c = pr; > } > > + dm_list_del(&p->list); > dm_free(p); > } > > diff --git a/libdm/mm/pool.c b/libdm/mm/pool.c > index 3528a91..13c1c06 100644 > --- a/libdm/mm/pool.c > +++ b/libdm/mm/pool.c > @@ -15,6 +15,9 @@ > > #include "dmlib.h" > > +static DM_LIST_INIT(_dm_pools); > +void dm_pools_dump(void); > + > #ifdef DEBUG_POOL > #include "pool-debug.c" > #else > @@ -52,3 +55,22 @@ void *dm_pool_zalloc(struct dm_pool *p, size_t s) > > return ptr; > } > + > +void dm_pools_dump(void) > +{ > + struct dm_pool *p; > + > + if (dm_list_empty(&_dm_pools)) > + return; > + > + log_error("You have a pool leak:"); > + dm_list_iterate_items(p, &_dm_pools) { > +#ifdef DEBUG_POOL > + log_error(" [%lx] %s (%u bytes)", > + (unsigned long)p->orig_pool, > + p->name, p->stats.bytes); > +#else > + log_error(" [%lx]", (unsigned long)p); > +#endif Would "%p" format make more sense instead of %lx and the cast here? I believe unsigned long is not guaranteed to be of the same size as a pointer. -- Peter Rockai | me()mornfall!net | prockai()redhat!com http://blog.mornfall.net | http://web.mornfall.net "In My Egotistical Opinion, most people's C programs should be indented six feet downward and covered with dirt." -- Blair P. Houghton on the subject of C program indentation