Commit d6fc8821 authored by Kinglong Mee's avatar Kinglong Mee Committed by J. Bruce Fields

SUNRPC/Cache: Always treat the invalid cache as unexpired

When the first time pynfs runs after rpc/nfsd startup, always get the warning,

"Got error: Connection closed"

I found the problem is caused by,
1. A new startup of nfsd, rpc.mountd, etc,
2. A rpc request from client (pynfs test, or normal mounting),
3. An ip_map cache is created but invalid, so upcall to rpc.mountd,
4. rpc.mountd process the ip_map upcall, before write the valid data to nfsd,
   do auth_reload(), and check_useipaddr(),
5. For the first time, old_use_ipaddr = -1, it causes rpc.mountd do write_flush    that doing cache_clean,
6. The ip_map cache will be treat as expired and clean,
7. When rpc.mountd write the valid data to nfsd, a new ip_map is created
   and updated, the cache_check of old ip_map(doing the upcall) will
   return -ETIMEDOUT.
8. RPC layer return SVC_CLOSE and close the xprt after commit 4d712ef1
   "svcauth_gss: Close connection when dropping an incoming message"

NeilBrown suggest in another email,

"If CACHE_VALID is not set, then there is no data in the cache item,
 so there is nothing to expire. So it would be nice if cache items that
 don't have CACHE_VALID are never treated as expired."

v3, change the order of the two patches
v2, change the checking of CACHE_PENDING to CACHE_VALID
Reviewed-by: default avatarNeilBrown <neilb@suse.com>
Signed-off-by: default avatarKinglong Mee <kinglongmee@gmail.com>
Signed-off-by: default avatarJ. Bruce Fields <bfields@redhat.com>
parent 471a930a
......@@ -204,8 +204,11 @@ static inline void cache_put(struct cache_head *h, struct cache_detail *cd)
kref_put(&h->ref, cd->cache_put);
}
static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
static inline bool cache_is_expired(struct cache_detail *detail, struct cache_head *h)
{
if (!test_bit(CACHE_VALID, &h->flags))
return false;
return (h->expiry_time < seconds_since_boot()) ||
(detail->flush_time >= h->last_refresh);
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment