Return-Path: linux-nfs-owner@vger.kernel.org Received: from e28smtp08.in.ibm.com ([122.248.162.8]:42855 "EHLO e28smtp08.in.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751016AbaEAPpU (ORCPT ); Thu, 1 May 2014 11:45:20 -0400 Received: from /spool/local by e28smtp08.in.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Thu, 1 May 2014 21:15:18 +0530 From: "Aneesh Kumar K.V" To: Dave Chinner Cc: agruen@kernel.org, bfields@fieldses.org, akpm@linux-foundation.org, viro@zeniv.linux.org.uk, dhowells@redhat.com, linux-fsdevel@vger.kernel.org, linux-nfs@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH -V1 19/22] vfs: Cache richacl in struct inode In-Reply-To: <20140429005203.GT15995@dastard> References: <1398615293-22931-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com> <1398615293-22931-20-git-send-email-aneesh.kumar@linux.vnet.ibm.com> <20140429005203.GT15995@dastard> Date: Thu, 01 May 2014 21:15:11 +0530 Message-ID: <87a9b1bj08.fsf@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: text/plain Sender: linux-nfs-owner@vger.kernel.org List-ID: Dave Chinner writes: > On Sun, Apr 27, 2014 at 09:44:50PM +0530, Aneesh Kumar K.V wrote: >> From: Andreas Gruenbacher >> >> Cache richacls in struct inode so that this doesn't have to be done >> individually in each filesystem. >> >> Signed-off-by: Andreas Gruenbacher >> Signed-off-by: Aneesh Kumar K.V >> --- >> fs/inode.c | 25 ++++++++++++++++++----- >> include/linux/fs.h | 12 +++++++++-- >> include/linux/richacl.h | 53 +++++++++++++++++++++++++++++++++++++++++++++++++ >> 3 files changed, 83 insertions(+), 7 deletions(-) >> >> diff --git a/fs/inode.c b/fs/inode.c >> index f96d2a6f88cc..b96a16d5c653 100644 >> --- a/fs/inode.c >> +++ b/fs/inode.c >> @@ -18,6 +18,7 @@ >> #include /* for inode_has_buffers */ >> #include >> #include >> +#include >> #include "internal.h" >> >> /* >> @@ -185,7 +186,12 @@ int inode_init_always(struct super_block *sb, struct inode *inode) >> inode->i_mapping = mapping; >> INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ >> #ifdef CONFIG_FS_POSIX_ACL >> - inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; >> + if (IS_POSIXACL(inode)) >> + inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; >> +#endif >> +#ifdef CONFIG_FS_RICHACL >> + if (IS_RICHACL(inode)) >> + inode->i_richacl = ACL_NOT_CACHED; >> #endif > > That's just plain wrong. i think this is what Christoph didn't like. > An inode can have either a POSIX ACL or a RICH ACL, so there is no > need for multiple pointers in the inode for them. We don't really have multiple pointers union { #ifdef CONFIG_FS_POSIX_ACL struct { struct posix_acl *i_acl; struct posix_acl *i_default_acl; }; #endif #ifdef CONFIG_FS_RICHACL struct richacl *i_richacl; #endif }; But I understand what you are suggesting here. Will try to rework and see how much easy/simpler it makes the code. > >> >> #ifdef CONFIG_FSNOTIFY >> @@ -240,10 +246,19 @@ void __destroy_inode(struct inode *inode) >> } >> >> #ifdef CONFIG_FS_POSIX_ACL >> - if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) >> - posix_acl_release(inode->i_acl); >> - if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) >> - posix_acl_release(inode->i_default_acl); >> + if (IS_POSIXACL(inode)) { >> + if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) >> + posix_acl_release(inode->i_acl); >> + if (inode->i_default_acl && >> + inode->i_default_acl != ACL_NOT_CACHED) >> + posix_acl_release(inode->i_default_acl); >> + } >> +#endif >> +#ifdef CONFIG_FS_RICHACL >> + if (IS_RICHACL(inode)) { >> + if (inode->i_richacl && inode->i_richacl != ACL_NOT_CACHED) >> + richacl_put(inode->i_richacl); >> + } >> #endif > > if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) > acl_release(inode->i_acl); > > And all the mess of working out what acl needs releasing get taken > out of of this code. Understood. Will try to see how much I can simplify. > >> index 22d85798b520..95df64d21e55 100644 >> --- a/include/linux/fs.h >> +++ b/include/linux/fs.h >> @@ -492,6 +492,7 @@ static inline int mapping_writably_mapped(struct address_space *mapping) >> #endif >> >> struct posix_acl; >> +struct richacl; >> #define ACL_NOT_CACHED ((void *)(-1)) >> >> #define IOP_FASTPERM 0x0001 >> @@ -510,10 +511,17 @@ struct inode { >> kgid_t i_gid; >> unsigned int i_flags; >> >> + union { >> #ifdef CONFIG_FS_POSIX_ACL >> - struct posix_acl *i_acl; >> - struct posix_acl *i_default_acl; >> + struct { >> + struct posix_acl *i_acl; >> + struct posix_acl *i_default_acl; >> + }; >> #endif >> +#ifdef CONFIG_FS_RICHACL >> + struct richacl *i_richacl; >> +#endif >> + }; > >> >> const struct inode_operations *i_op; >> struct super_block *i_sb; >> diff --git a/include/linux/richacl.h b/include/linux/richacl.h >> index 88f95d78b897..a7db341e4ee9 100644 >> --- a/include/linux/richacl.h >> +++ b/include/linux/richacl.h >> @@ -191,6 +191,59 @@ richacl_put(struct richacl *acl) >> kfree(acl); >> } >> >> +#ifdef CONFIG_FS_RICHACL >> +static inline struct richacl *get_cached_richacl(struct inode *inode) >> +{ >> + struct richacl **p, *acl; >> + >> + p = &inode->i_richacl; >> + acl = ACCESS_ONCE(*p); >> + if (acl) { >> + spin_lock(&inode->i_lock); >> + acl = *p; >> + if (acl != ACL_NOT_CACHED) >> + acl = richacl_get(acl); >> + spin_unlock(&inode->i_lock); >> + } >> + return acl; >> +} >> + >> +static inline void set_cached_richacl(struct inode *inode, >> + struct richacl *acl) >> +{ >> + struct richacl *old = NULL; >> + spin_lock(&inode->i_lock); >> + old = inode->i_richacl; >> + inode->i_richacl = richacl_get(acl); >> + spin_unlock(&inode->i_lock); >> + if (old != ACL_NOT_CACHED) >> + richacl_put(old); >> +} >> + >> +static inline void forget_cached_richacl(struct inode *inode) >> +{ >> + struct richacl *old = NULL; >> + spin_lock(&inode->i_lock); >> + old = inode->i_richacl; >> + inode->i_richacl = ACL_NOT_CACHED; >> + spin_unlock(&inode->i_lock); >> + if (old != ACL_NOT_CACHED) >> + richacl_put(old); >> +} >> + >> +static inline int negative_cached_richacl(struct inode *inode) >> +{ >> + struct richacl **p, *acl; >> + >> + p = &inode->i_richacl; >> + acl = ACCESS_ONCE(*p); >> + if (acl) >> + return 0; >> + return 1; >> +} >> + >> +#endif > > This is all just copy-and-paste from the posix_acl code with the RCU > path-walk awareness removed from it. There should only be a single > RCU aware version of these functions that does not care what type of > ACL is in use. > Ok. -aneesh