FS-Cache has a few bugs generally only get tried in error mode / under VMA preassure. The problem is exacerbated when majority of the IO is cacheable netfs IO. At that point there are few pages that are evictable due to GPF_NOFS allocation strategies. The first bug is a recursive hang caused by fscache write out to ext4 which in turn needs to allocate pages and ends up calling releasepages() on the netfs. The second two bugs are cookie ref count bugs that happen when in the error cases (mostly from related memory alloc failures). This is the second try at this with Sign-Off lines, suggestions from David and debug log on time out. With the log line we can further investigate how often this happens and be smarter with the timeout (if need be). Milosz Tanski (3): FS-Cache: Timeout for releasepage() FS-Cache: Reduce cookie ref count if submit fails. FS-Cache: refcount becomes corrupt under vma pressure. fs/fscache/object.c | 2 ++ fs/fscache/page.c | 26 ++++++++++++++++++++++---- 2 files changed, 24 insertions(+), 4 deletions(-) -- 1.7.9.5 -- Linux-cachefs mailing list Linux-cachefs@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cachefs
