Skip to contentSkip to conversation

[3.6.11] NFS: Cache request denied due to non-unique superblock keys.

DISCUSSION Paweł Sikora 4 participants 6 messages Dec 19, 2012 – Dec 20, 2012

Subsystems (2)
NFS client NFS server
lore
Share & tools
lore

Paweł Sikora

author
#0 lore raw

Hi,

i'm testing the 3.6.11 kernel on some small internal nfs network (1 server <-> 2 clients) and observing such info in dmesg on the client side. could you put some light on this dmesg entry? is it a bug or misconfiguration? i can provide more information about nfs configuration if necessary but denying caching doesn't look nicely.

i'm using following nfs related packages:

autofs-5.0.5-1.x86_64
libnfsidmap-0.25-4.x86_64
nfs-utils-1.2.7-1.x86_64
nfs-utils-clients-1.2.7-1.x86_64
nfs-utils-common-1.2.7-1.x86_64

BR,
Paweł.

please CC me on reply.

#1 lore raw

Paweł Sikora wrote:

Hi,

i'm testing the 3.6.11 kernel on some small internal nfs network (1 server <-> 2 clients) and observing such info in dmesg on the client side. could you put some light on this dmesg entry?

Are you mounting the same file system more than once? Why? I thought I remembered a mount option to uniquify the mounts but I can't find it now.

#2 lore raw
In reply to Paweł Sikora
Quoted 9 lines · Paweł Sikora

Hi,

i'm testing the 3.6.11 kernel on some small internal nfs network (1 server <-> 2 clients) and observing such info in dmesg on the client side. could you put some light on this dmesg entry? is it a bug or misconfiguration? i can provide more information about nfs configuration if necessary but denying caching doesn't look nicely.

i'm using following nfs related packages:

Quoted 8 lines

autofs-5.0.5-1.x86_64
libnfsidmap-0.25-4.x86_64
nfs-utils-1.2.7-1.x86_64
nfs-utils-clients-1.2.7-1.x86_64
nfs-utils-common-1.2.7-1.x86_64

BR,
Paweł.

It's hard to guess when you don't tell us anything about what you are doing to trigger this message, but it sounds as if you are trying to use the cachefs mount options incorrectly.

Signature · 6 lines
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com
author
#3 lore raw
Quoted 19 lines · Trond

On Wed, 2012-12-19 at 19:58 +-0100, Pawe+AUI- Sikora wrote:

(quoted diff) +17 -0
+AD4- Hi,
+AD4- 
+AD4- i'm testing the 3.6.11 kernel on some small internal nfs network (1 server +ADw--+AD4- 2 clients)
+AD4- and observing such info in dmesg on the client side. could you put some light on this dmesg entry?
+AD4- is it a bug or misconfiguration? i can provide more information about nfs configuration
+AD4- if necessary but denying caching doesn't look nicely.
+AD4- 
+AD4- i'm using following nfs related packages:
+AD4- 
+AD4- autofs-5.0.5-1.x86+AF8-64
+AD4- libnfsidmap-0.25-4.x86+AF8-64
+AD4- nfs-utils-1.2.7-1.x86+AF8-64
+AD4- nfs-utils-clients-1.2.7-1.x86+AF8-64
+AD4- nfs-utils-common-1.2.7-1.x86+AF8-64
+AD4- 
+AD4- BR,
+AD4- Pawe+AUI-.
Quoted 3 lines

It's hard to guess when you don't tell us anything about what you are doing to trigger this message, but it sounds as if you are trying to use the cachefs mount options incorrectly.

on the server side there're following exports:

# exportfs -v
/home/atest/farm-local
                <world>(rw,async,wdelay,nohide,crossmnt,no_root_squash,no_subtree_check,insecure_locks)
/home/atest/farm
                <world>(rw,async,wdelay,nohide,crossmnt,no_root_squash,no_subtree_check,insecure_locks)
/home/users     <world>(rw,no_root_squash,no_subtree_check,insecure_locks)

in the exported /home/atest/* there're crossmounted/binded directories:

# fstab
/home/atest/farm-local/results  /home/atest/farm/results        none bind
/home/services/ftp              /var/lib/ftp                    none bind
/home/services/ftp              /home/atest/farm-local/ftp      none bind
# df -h
Filesystem                   Size  Used Avail Use% Mounted on
rootfs                       7.6G  4.5G  2.7G  63% /
/dev/md0                     7.6G  4.5G  2.7G  63% /
run                           32G  264K   32G   1% /run
devtmpfs                      32G     0   32G   0% /dev
/dev/md1                     7.2T  794G  6.4T  11% /home
none                          32G     0   32G   0% /dev/shm
/dev/md1                     7.2T  794G  6.4T  11% /home/atest/farm/results
/dev/md1                     7.2T  794G  6.4T  11% /var/lib/ftp
/dev/md1                     7.2T  794G  6.4T  11% /home/atest/farm-local/ftp

on the client side there's only a simple autofs configuration:

# cat /etc/autofs/auto.server
farm           -rw,soft,rsize=32768,wsize=32768,intr,vers=4,async,bg,noatime,retry=30,retrans=10        192.168.2.100:/home/atest/farm-local
home           -rw,soft,vers=4        192.168.2.100:/home/users
#4 lore raw
Quoted 53 lines · Paweł Sikora
Quoted 19 lines · Trond

On Wed, 2012-12-19 at 19:58 +--0100, Pawe+-AUI- Sikora wrote:

+-AD4- Hi,
+-AD4- 
+-AD4- i'm testing the 3.6.11 kernel on some small internal nfs network (1 server +-ADw--+-AD4- 2 clients)
+-AD4- and observing such info in dmesg on the client side. could you put some light on this dmesg entry?
+-AD4- is it a bug or misconfiguration? i can provide more information about nfs configuration
+-AD4- if necessary but denying caching doesn't look nicely.
+-AD4- 
+-AD4- i'm using following nfs related packages:
+-AD4- 
+-AD4- autofs-5.0.5-1.x86+-AF8-64
+-AD4- libnfsidmap-0.25-4.x86+-AF8-64
+-AD4- nfs-utils-1.2.7-1.x86+-AF8-64
+-AD4- nfs-utils-clients-1.2.7-1.x86+-AF8-64
+-AD4- nfs-utils-common-1.2.7-1.x86+-AF8-64
+-AD4- 
+-AD4- BR,
+-AD4- Pawe+-AUI-.
Quoted 3 lines

It's hard to guess when you don't tell us anything about what you are doing to trigger this message, but it sounds as if you are trying to use the cachefs mount options incorrectly.

on the server side there're following exports:

# exportfs -v
/home/atest/farm-local
                <world>(rw,async,wdelay,nohide,crossmnt,no_root_squash,no_subtree_check,insecure_locks)
/home/atest/farm
                <world>(rw,async,wdelay,nohide,crossmnt,no_root_squash,no_subtree_check,insecure_locks)
/home/users     <world>(rw,no_root_squash,no_subtree_check,insecure_locks)

in the exported /home/atest/* there're crossmounted/binded directories:

# fstab
/home/atest/farm-local/results  /home/atest/farm/results        none bind
/home/services/ftp              /var/lib/ftp                    none bind
/home/services/ftp              /home/atest/farm-local/ftp      none bind
# df -h
Filesystem                   Size  Used Avail Use% Mounted on
rootfs                       7.6G  4.5G  2.7G  63% /
/dev/md0                     7.6G  4.5G  2.7G  63% /
run                           32G  264K   32G   1% /run
devtmpfs                      32G     0   32G   0% /dev
/dev/md1                     7.2T  794G  6.4T  11% /home
none                          32G     0   32G   0% /dev/shm
/dev/md1                     7.2T  794G  6.4T  11% /home/atest/farm/results
/dev/md1                     7.2T  794G  6.4T  11% /var/lib/ftp
/dev/md1                     7.2T  794G  6.4T  11% /home/atest/farm-local/ftp
Quoted 6 lines

on the client side there's only a simple autofs configuration:

# cat /etc/autofs/auto.server
farm           -rw,soft,rsize=32768,wsize=32768,intr,vers=4,async,bg,noatime,retry=30,retrans=10        192.168.2.100:/home/atest/farm-local
home           -rw,soft,vers=4        192.168.2.100:/home/users

OK. If you are not specifying '-ofsc', then that warning is 100% confusing. How about the following patch?

8<---------------------------------------------------------------

Quoted 1 line

From e8ed657ba5edc8c9339e00ab2fd235ab1dd94cd5 Mon Sep 17 00:00:00 2001

From: Trond Myklebust <[email protected]>
Date: Wed, 19 Dec 2012 16:56:28 -0500
Subject: [PATCH] NFS: Kill fscache warnings when mounting without -ofsc
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The fscache code will currently bleat a "non-unique superblock keys" warning even if the user is mounting without the 'fsc' option. There should be no reason to even initialise the superblock cache cookie unless we're planning on using fscache for something, so ensure that we check for the NFS_OPTION_FSCACHE flag before calling into the fscache code.

Reported-by Paweł Sikora <[email protected]>
Signed-off-by Trond Myklebust <[email protected]>
Cc David Howells <[email protected]>
---
 fs/nfs/super.c | 19 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-----
 1 file changed, 15 insertions(+-), 4 deletions(-)
fs/nfs/super.c +0 -0
index aa5315b..c25cadf8 100644

--- a/fs/nfs/super.c
+-+-+- b/fs/nfs/super.c
@@ -2375,19 +-2375,30 @@ static void nfs_get_cache_cookie(struct super_block *sb,
struct nfs_parsed_mount_data *parsed, struct nfs_clone_mount *cloned)
{
+- struct nfs_server *nfss = NFS_SB(sb);
char *uniq = NULL;
int ulen = 0;

- if (parsed && parsed->fscache_uniq) {
- uniq = parsed->fscache_uniq;
- ulen = strlen(parsed->fscache_uniq);

+-	nfss->fscache_key = NULL;
+-	nfss->fscache = NULL;
+-
+-	if (parsed) {
+-		if (!(parsed->options & NFS_OPTION_FSCACHE))
+-			return;
+-		if (parsed->fscache_uniq) {
+-			uniq = parsed->fscache_uniq;
+-			ulen = strlen(parsed->fscache_uniq);
+-		}
 	} else if (cloned) {
 		struct nfs_server *mnt_s = NFS_SB(cloned->sb);
+-		if (!(mnt_s->options & NFS_OPTION_FSCACHE))
+-			return;

if (mnt_s->fscache_key) {
uniq = mnt_s->fscache_key->key.uniquifier; ulen = mnt_s->fscache_key->key.uniq_len;
};
- }

+-	} else
+-		return;

nfs_fscache_get_super_cookie(sb, uniq, ulen);
}

Signature · 10 lines
1.7.11.7


-- 
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com
Acked-by
#5 lore raw
Quoted 15 lines · Trond

From e8ed657ba5edc8c9339e00ab2fd235ab1dd94cd5 Mon Sep 17 00:00:00 2001
From: Trond Myklebust <[email protected]>
Date: Wed, 19 Dec 2012 16:56:28 -0500
Subject: [PATCH] NFS: Kill fscache warnings when mounting without -ofsc
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The fscache code will currently bleat a "non-unique superblock keys" warning even if the user is mounting without the 'fsc' option. There should be no reason to even initialise the superblock cache cookie unless we're planning on using fscache for something, so ensure that we check for the NFS_OPTION_FSCACHE flag before calling into the fscache code.

Quoted 3 lines
Reported-by Paweł Sikora <[email protected]>
Signed-off-by Trond Myklebust <[email protected]>
Cc David Howells <[email protected]>
Acked-by David Howells <[email protected]>