On Sat, Sep 09, 2023 at 02:16:00AM +0100, Matthew Wilcox wrote: > On Fri, Sep 08, 2023 at 07:06:49PM -0400, Rafael Aquini wrote: > > This patch fixes this issue by properly checking shutdown_cache()'s > > return value before taking the kmem_cache_release() branch. > > Is this the right way to fix this problem? If the module destroys the > slab cache, it's not going to be possible to free any of the objects > still allocated from the cache. I feel that we should treat this as > implicitly freeing all the objects that were allocated from the cache > rather than saying the cache is still busy. > Leaving the cache with the unfreeable slabs "alone" is how it was historically done, and we have to fix this corner case opened by 0495e337b703 this way to address the corruption on stable releases without changing their established and expected behavior. I think your proposal for a different behavior upon cache destruction is something we should discuss for future releases, but it is orthogonal to this required follow-up fix.
