On Fri, Mar 04, 2011 at 01:45:06PM -0500, Neil Horman wrote: > On Fri, Mar 04, 2011 at 12:13:21PM -0500, J. Bruce Fields wrote: > > On Fri, Mar 04, 2011 at 11:44:13AM -0500, Neil Horman wrote: > > > We can't create a buffer with kmalloc and free it later in the tcp > > > ack path with put_page, so we need to either: > > > > But out of curiosity: why is there this rule? > > > Its an artifact the results from needing to free memory using a method > in keeping with the way in which it was allocated. To use this bug as > an example, the acl data was allocated by the VFS using kmalloc, which > gets data from the slab. Even though this data is a size that is a > multiple of a page, slab objects can be less than a page, and multiple > objects can be stored in a single page. As such, anything allocated > from the slab allocator needs to be freed by the slab allocator, so > that object reference counts internally maintained by the slab can be > kept accurate. OK. I guess my naive mental model was that the slab allocator was layered on top of the page allocator--so it got pages with alloc_pages() or equivalent, then handed out pieces of them to people using the slab allocator. So I assumed the slab allocator would hold a reference to the page like any other user would, in which case the tcp code could take a second reference of its own. --b. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
