Return-Path: linux-nfs-owner@vger.kernel.org Received: from mail-ie0-f180.google.com ([209.85.223.180]:57925 "EHLO mail-ie0-f180.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755450AbaBFTxl (ORCPT ); Thu, 6 Feb 2014 14:53:41 -0500 Received: by mail-ie0-f180.google.com with SMTP id at1so943035iec.25 for ; Thu, 06 Feb 2014 11:53:40 -0800 (PST) From: Trond Myklebust To: Neil Brown Cc: linux-nfs@vger.kernel.org Subject: [PATCH] NFS: Do not set NFS_INO_INVALID_LABEL unless server supports labeled NFS Date: Thu, 6 Feb 2014 14:53:38 -0500 Message-Id: <1391716418-7353-1-git-send-email-trond.myklebust@primarydata.com> In-Reply-To: <20140206135101.1cc83442@notabene.brown> References: <20140206135101.1cc83442@notabene.brown> Sender: linux-nfs-owner@vger.kernel.org List-ID: Commit aa9c2669626c (NFS: Client implementation of Labeled-NFS) introduces a performance regression. When nfs_zap_caches_locked is called, it sets the NFS_INO_INVALID_LABEL flag irrespectively of whether or not the NFS server supports security labels. Since that flag is never cleared, it means that all calls to nfs_revalidate_inode() will now trigger an on-the-wire GETATTR call. This patch ensures that we never set the NFS_INO_INVALID_LABEL unless the server advertises support for labeled NFS. It also causes nfs_setsecurity() to clear NFS_INO_INVALID_LABEL when it has successfully set the security label for the inode. Finally it gets rid of the NFS_INO_INVALID_LABEL cruft from nfs_update_inode, which has nothing to do with labeled NFS. Reported-by: Neil Brown Cc: stable@vger.kernel.org # 3.11+ Signed-off-by: Trond Myklebust --- Hi Neil, Does this fix the GETATTR regression that you reported? Cheers Trond fs/nfs/inode.c | 14 ++++++++++---- fs/nfs/internal.h | 9 +++++++++ 2 files changed, 19 insertions(+), 4 deletions(-) diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c index 29cb93653b3c..4636b828e957 100644 --- a/fs/nfs/inode.c +++ b/fs/nfs/inode.c @@ -169,7 +169,6 @@ static void nfs_zap_caches_locked(struct inode *inode) if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) { nfs_fscache_invalidate(inode); nfsi->cache_validity |= NFS_INO_INVALID_ATTR - | NFS_INO_INVALID_LABEL | NFS_INO_INVALID_DATA | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL @@ -177,10 +176,10 @@ static void nfs_zap_caches_locked(struct inode *inode) nfs_zap_readdir_cookie(nfsi); } else nfsi->cache_validity |= NFS_INO_INVALID_ATTR - | NFS_INO_INVALID_LABEL | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL | NFS_INO_REVAL_PAGECACHE; + nfs_zap_label_cache_locked(nfsi); } void nfs_zap_caches(struct inode *inode) @@ -272,6 +271,13 @@ nfs_init_locked(struct inode *inode, void *opaque) } #ifdef CONFIG_NFS_V4_SECURITY_LABEL +static void nfs_clear_label_invalid(struct inode *inode) +{ + spin_lock(&inode->i_lock); + NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL; + spin_unlock(&inode->i_lock); +} + void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, struct nfs4_label *label) { @@ -289,6 +295,7 @@ void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr, __func__, (char *)label->label, label->len, error); + nfs_clear_label_invalid(inode); } } @@ -1654,7 +1661,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) inode->i_blocks = fattr->du.nfs2.blocks; /* Update attrtimeo value if we're out of the unstable period */ - if (invalid & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) { + if (invalid & NFS_INO_INVALID_ATTR) { nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); nfsi->attrtimeo_timestamp = now; @@ -1667,7 +1674,6 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) } } invalid &= ~NFS_INO_INVALID_ATTR; - invalid &= ~NFS_INO_INVALID_LABEL; /* Don't invalidate the data if we were to blame */ if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h index 8b5cc04a8611..fafdddac8271 100644 --- a/fs/nfs/internal.h +++ b/fs/nfs/internal.h @@ -279,9 +279,18 @@ static inline void nfs4_label_free(struct nfs4_label *label) } return; } + +static inline void nfs_zap_label_cache_locked(struct nfs_inode *nfsi) +{ + if (nfs_server_capable(&nfsi->vfs_inode, NFS_CAP_SECURITY_LABEL)) + nfsi->cache_validity |= NFS_INO_INVALID_LABEL; +} #else static inline struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags) { return NULL; } static inline void nfs4_label_free(void *label) {} +static inline void nfs_zap_label_cache_locked(struct nfs_inode *nfsi) +{ +} #endif /* CONFIG_NFS_V4_SECURITY_LABEL */ /* proc.c */ -- 1.8.5.3