Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Rasmus Villemoes <linux@rasmusvillemoes.dk>
To:     Alexander Viro <viro@zeniv.linux.org.uk>
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        Rasmus Villemoes <linux@rasmusvillemoes.dk>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH resend] fs/namei.c: micro-optimize acl_permission_check
Date:   Fri,  5 Jun 2020 16:23:00 +0200
Message-Id: <20200605142300.14591-1-linux@rasmusvillemoes.dk>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Just something like open(/usr/include/sys/stat.h) causes five calls of
generic_permission -> acl_permission_check -> in_group_p; if the
compiler first tried /usr/local/include/..., that would be a few
more.

Altogether, on a bog-standard Ubuntu 20.04 install, a workload
consisting of compiling lots of userspace programs (i.e., calling lots
of short-lived programs that all need to get their shared libs mapped
in, and the compilers poking around looking for system headers - lots
of /usr/lib, /usr/bin, /usr/include/ accesses) puts in_group_p around
0.1% according to perf top. With an artificial load such as

  while true ; do find /usr/ -print0 | xargs -0 stat > /dev/null ; done

that jumps to over 0.4%.

System-installed files are almost always 0755 (directories and
binaries) or 0644, so in most cases, we can avoid the binary search
and the cost of pulling the cred->groups array and in_group_p() .text
into the cpu cache.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
---
 fs/namei.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/fs/namei.c b/fs/namei.c
index d81f73ff1a8b..c6f0c6643db5 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -303,7 +303,12 @@ static int acl_permission_check(struct inode *inode, int mask)
 				return error;
 		}
 
-		if (in_group_p(inode->i_gid))
+		/*
+		 * If the "group" and "other" permissions are the same,
+		 * there's no point calling in_group_p() to decide which
+		 * set to use.
+		 */
+		if ((((mode >> 3) ^ mode) & 7) && in_group_p(inode->i_gid))
 			mode >>= 3;
 	}
 
-- 
2.23.0