On 1/22/25 5:16 PM, NeilBrown wrote:
On Thu, 23 Jan 2025, Chuck Lever wrote:On 1/21/25 10:54 PM, NeilBrown wrote:The nfsd filecache currently uses list_lru for tracking files recently used in NFSv3 requests which need to be "garbage collected" when they have becoming idle - unused for 2-4 seconds. I do not believe list_lru is a good tool for this. It does no allow the timeout which filecache requires so we have to add a timeout mechanism which holds the list_lru for while the whole list is scanned looking for entries that haven't been recently accessed. When the list is largish (even a few hundred) this can block new requests which need the lock to remove a file to access it. This patch removes the list_lru and instead uses 2 simple linked lists. When a file is accessed it is removed from whichever list it is one, then added to the tail of the first list. Every 2 seconds the second list is moved to the "freeme" list and the first list is moved to the second list. This avoids any need to walk a list to find old entries. These lists are per-netns rather than global as the freeme list is per-netns as the actual freeing is done in nfsd threads which are per-netns. This should not be applied until we resolve how to handle the race-detection code in nfsd_file_put(). However I'm posting it now to get any feedback so that a final version can be posted as soon as that issue is resolved. Thanks, NeilBrown [PATCH 1/4] nfsd: filecache: use nfsd_file_dispose_list() in [PATCH 2/4] nfsd: filecache: move globals nfsd_file_lru and [PATCH 3/4] nfsd: filecache: change garbage collection list [PATCH 4/4] nfsd: filecache: change garbage collection to a timer.Hi Neil - I would like Dave Chinner to chime in on this approach. When you resend, please Cc: him. Thanks!Sure, I can do that. But why Dave in particular? I would like to add a comment to the cover letter explaining to Dave what he was added to see and I don't know what to say.
Dave helped me with edead3a55804 ("NFSD: Fix the filecache LRU
shrinker") and a few other related fixes, and he is the original
author of the list_lru facility (see a38e40824844 ("list: add a new LRU
list type"). He might have additional insight on whether the filecache's
use of list_lru is salvageable or how a replacement of that mechanism
should work.
--
Chuck Lever
