From patchwork Fri Jul 25 06:31:59 2014 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Junxiao Bi X-Patchwork-Id: 4621501 Return-Path: X-Original-To: patchwork-ocfs2-devel@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork1.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.19.201]) by patchwork1.web.kernel.org (Postfix) with ESMTP id A63D29F23C for ; Fri, 25 Jul 2014 06:33:11 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id C6E58201E4 for ; Fri, 25 Jul 2014 06:33:10 +0000 (UTC) Received: from userp1040.oracle.com (userp1040.oracle.com [156.151.31.81]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id AF2DF201DD for ; Fri, 25 Jul 2014 06:33:09 +0000 (UTC) Received: from acsinet22.oracle.com (acsinet22.oracle.com [141.146.126.238]) by userp1040.oracle.com (Sentrion-MTA-4.3.2/Sentrion-MTA-4.3.2) with ESMTP id s6P6WJrI031360 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK); Fri, 25 Jul 2014 06:32:21 GMT Received: from oss.oracle.com (oss-external.oracle.com [137.254.96.51]) by acsinet22.oracle.com (8.14.4+Sun/8.14.4) with ESMTP id s6P6WGxK009584 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Fri, 25 Jul 2014 06:32:17 GMT Received: from localhost ([127.0.0.1] helo=oss.oracle.com) by oss.oracle.com with esmtp (Exim 4.63) (envelope-from ) id 1XAZ3A-0000yD-Fv; Thu, 24 Jul 2014 23:32:16 -0700 Received: from ucsinet21.oracle.com ([156.151.31.93]) by oss.oracle.com with esmtp (Exim 4.63) (envelope-from ) id 1XAZ2o-0000wx-G4 for ocfs2-devel@oss.oracle.com; Thu, 24 Jul 2014 23:31:54 -0700 Received: from userz7022.oracle.com (userz7022.oracle.com [156.151.31.86]) by ucsinet21.oracle.com (8.14.4+Sun/8.14.4) with ESMTP id s6P6Vrrq003531 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Fri, 25 Jul 2014 06:31:53 GMT Received: from abhmp0017.oracle.com (abhmp0017.oracle.com [141.146.116.23]) by userz7022.oracle.com (8.14.5+Sun/8.14.4) with ESMTP id s6P6Vq2B001567; Fri, 25 Jul 2014 06:31:52 GMT Received: from bijx-OptiPlex-780.cn.oracle.com (/10.182.39.153) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Thu, 24 Jul 2014 23:31:51 -0700 From: Junxiao Bi To: ocfs2-devel@oss.oracle.com Date: Fri, 25 Jul 2014 14:31:59 +0800 Message-Id: <1406269919-13234-1-git-send-email-junxiao.bi@oracle.com> X-Mailer: git-send-email 1.7.9.5 Cc: mfasheh@suse.com, joe.jin@oracle.com Subject: [Ocfs2-devel] [PATCH v2] ocfs2: reflink: fix slow unlink for refcounted file X-BeenThere: ocfs2-devel@oss.oracle.com X-Mailman-Version: 2.1.9 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , MIME-Version: 1.0 Sender: ocfs2-devel-bounces@oss.oracle.com Errors-To: ocfs2-devel-bounces@oss.oracle.com X-Source-IP: acsinet22.oracle.com [141.146.126.238] X-Spam-Status: No, score=-4.2 required=5.0 tests=BAYES_00, RCVD_IN_DNSWL_MED, RP_MATCHES_RCVD, UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP When running ocfs2 test suite multiple nodes reflink stress test, for a 4 nodes cluster, every unlink() for refcounted file needs about 700s. The slow unlink is caused by the contention of refcount tree lock since all nodes are unlink files using the same refcount tree. When the unlinking file have many extents(over 1600 in our test), most of the extents has refcounted flag set. In ocfs2_commit_truncate(), it will execute the following call trace for every extents. This means it needs get and released refcount tree lock about 1600 times. And when several nodes are do this at the same time, the performance will be very low. . ocfs2_remove_btree_range() ----ocfs2_lock_refcount_tree() ------ocfs2_refcount_lock() --------__ocfs2_cluster_lock() ocfs2_refcount_lock() is costly, move it to ocfs2_commit_truncate() to do lock/unlock once can improve a lot performance. Signed-off-by: Junxiao Bi --- fs/ocfs2/alloc.c | 28 +++++++++++++++++++++------- fs/ocfs2/alloc.h | 2 +- fs/ocfs2/dir.c | 2 +- fs/ocfs2/file.c | 2 +- 4 files changed, 24 insertions(+), 10 deletions(-) diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c index 9d8fcf2..b8afbd7 100644 --- a/fs/ocfs2/alloc.c +++ b/fs/ocfs2/alloc.c @@ -5653,7 +5653,7 @@ int ocfs2_remove_btree_range(struct inode *inode, struct ocfs2_extent_tree *et, u32 cpos, u32 phys_cpos, u32 len, int flags, struct ocfs2_cached_dealloc_ctxt *dealloc, - u64 refcount_loc) + u64 refcount_loc, bool refcount_tree_locked) { int ret, credits = 0, extra_blocks = 0; u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); @@ -5667,11 +5667,13 @@ int ocfs2_remove_btree_range(struct inode *inode, BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)); - ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, - &ref_tree, NULL); - if (ret) { - mlog_errno(ret); - goto bail; + if (!refcount_tree_locked) { + ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, + &ref_tree, NULL); + if (ret) { + mlog_errno(ret); + goto bail; + } } ret = ocfs2_prepare_refcount_change_for_del(inode, @@ -7012,6 +7014,7 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb, u64 refcount_loc = le64_to_cpu(di->i_refcount_loc); struct ocfs2_extent_tree et; struct ocfs2_cached_dealloc_ctxt dealloc; + struct ocfs2_refcount_tree *ref_tree = NULL; ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); ocfs2_init_dealloc_ctxt(&dealloc); @@ -7121,9 +7124,18 @@ start: phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno); + if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) { + status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, + &ref_tree, NULL); + if (status) { + mlog_errno(status); + goto bail; + } + } + status = ocfs2_remove_btree_range(inode, &et, trunc_cpos, phys_cpos, trunc_len, flags, &dealloc, - refcount_loc); + refcount_loc, true); if (status < 0) { mlog_errno(status); goto bail; @@ -7138,6 +7150,8 @@ start: goto start; bail: + if (ref_tree) + ocfs2_unlock_refcount_tree(osb, ref_tree, 1); ocfs2_schedule_truncate_log_flush(osb, 1); diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h index ca381c5..fb09b97 100644 --- a/fs/ocfs2/alloc.h +++ b/fs/ocfs2/alloc.h @@ -142,7 +142,7 @@ int ocfs2_remove_btree_range(struct inode *inode, struct ocfs2_extent_tree *et, u32 cpos, u32 phys_cpos, u32 len, int flags, struct ocfs2_cached_dealloc_ctxt *dealloc, - u64 refcount_loc); + u64 refcount_loc, bool refcount_tree_locked); int ocfs2_num_free_extents(struct ocfs2_super *osb, struct ocfs2_extent_tree *et); diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c index 0717662..e737a4c 100644 --- a/fs/ocfs2/dir.c +++ b/fs/ocfs2/dir.c @@ -4477,7 +4477,7 @@ int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh) p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno); ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0, - &dealloc, 0); + &dealloc, 0, false); if (ret) { mlog_errno(ret); goto out; diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c index 2930e23..319c858 100644 --- a/fs/ocfs2/file.c +++ b/fs/ocfs2/file.c @@ -1804,7 +1804,7 @@ static int ocfs2_remove_inode_range(struct inode *inode, ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos, phys_cpos, trunc_len, flags, - &dealloc, refcount_loc); + &dealloc, refcount_loc, false); if (ret < 0) { mlog_errno(ret); goto out;