Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751624AbdLKTu7 (ORCPT ); Mon, 11 Dec 2017 14:50:59 -0500 Received: from mail-qt0-f195.google.com ([209.85.216.195]:40088 "EHLO mail-qt0-f195.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750886AbdLKTu4 (ORCPT ); Mon, 11 Dec 2017 14:50:56 -0500 X-Google-Smtp-Source: ACJfBotBCHHA4LDoQ54aANOETh5jdpo+RD/QcSnqlksxjGYMIpHR2bfcJx+fhecg+V2TJSajZguTOA== Date: Mon, 11 Dec 2017 11:50:52 -0800 From: Tejun Heo To: Jens Axboe , Jan Kara Cc: cgroups@vger.kernel.org, xuejiufei , kernel-team@fb.com, linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org Subject: [PATCH] writeback: synchronize sync(2) against cgroup writeback membership switches Message-ID: <20171211195052.GN2421075@devbig577.frc2.facebook.com> References: <20171205182007.GV2421075@devbig577.frc2.facebook.com> <8844b550-d91c-38d5-997a-a899d1e4aa42@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <8844b550-d91c-38d5-997a-a899d1e4aa42@gmail.com> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5159 Lines: 158 sync_inodes_sb() can race against cgwb (cgroup writeback) membership switches and fail to writeback some inodes. For example, if an inode switches to another wb while sync_inodes_sb() is in progress, the new wb might not be visible to bdi_split_work_to_wbs() at all or the inode might jump from a wb which hasn't issued writebacks yet to one which already has. This patch adds backing_dev_info->wb_switch_rwsem to synchronize cgwb switch path against sync_inodes_sb() so that sync_inodes_sb() is guaranteed to see all the target wbs and inodes can't jump wbs to escape syncing. Signed-off-by: Tejun Heo Reported-by: Jiufei Xue Link: http://lkml.kernel.org/r/dc694ae2-f07f-61e1-7097-7c8411cee12d@gmail.com --- fs/fs-writeback.c | 50 +++++++++++++++++++++++++++++++++++++-- include/linux/backing-dev-defs.h | 1 mm/backing-dev.c | 3 ++ 3 files changed, 52 insertions(+), 2 deletions(-) --- a/fs/fs-writeback.c +++ b/fs/fs-writeback.c @@ -331,11 +331,22 @@ struct inode_switch_wbs_context { struct work_struct work; }; +static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) +{ + down_write(&bdi->wb_switch_rwsem); +} + +static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) +{ + up_write(&bdi->wb_switch_rwsem); +} + static void inode_switch_wbs_work_fn(struct work_struct *work) { struct inode_switch_wbs_context *isw = container_of(work, struct inode_switch_wbs_context, work); struct inode *inode = isw->inode; + struct backing_dev_info *bdi = inode_to_bdi(inode); struct address_space *mapping = inode->i_mapping; struct bdi_writeback *old_wb = inode->i_wb; struct bdi_writeback *new_wb = isw->new_wb; @@ -344,6 +355,12 @@ static void inode_switch_wbs_work_fn(str void **slot; /* + * If @inode switches cgwb membership while sync_inodes_sb() is + * being issued, sync_inodes_sb() might miss it. Synchronize. + */ + down_read(&bdi->wb_switch_rwsem); + + /* * By the time control reaches here, RCU grace period has passed * since I_WB_SWITCH assertion and all wb stat update transactions * between unlocked_inode_to_wb_begin/end() are guaranteed to be @@ -435,6 +452,8 @@ skip_switch: spin_unlock(&new_wb->list_lock); spin_unlock(&old_wb->list_lock); + up_read(&bdi->wb_switch_rwsem); + if (switched) { wb_wakeup(new_wb); wb_put(old_wb); @@ -475,9 +494,18 @@ static void inode_switch_wbs(struct inod if (inode->i_state & I_WB_SWITCH) return; + /* + * Avoid starting new switches while sync_inodes_sb() is in + * progress. Otherwise, if the down_write protected issue path + * blocks heavily, we might end up starting a large number of + * switches which will block on the rwsem. + */ + if (!down_read_trylock(&bdi->wb_switch_rwsem)) + return; + isw = kzalloc(sizeof(*isw), GFP_ATOMIC); if (!isw) - return; + goto out_unlock; /* find and pin the new wb */ rcu_read_lock(); @@ -511,12 +539,14 @@ static void inode_switch_wbs(struct inod * Let's continue after I_WB_SWITCH is guaranteed to be visible. */ call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn); - return; + goto out_unlock; out_free: if (isw->new_wb) wb_put(isw->new_wb); kfree(isw); +out_unlock: + up_read(&bdi->wb_switch_rwsem); } /** @@ -893,6 +923,9 @@ fs_initcall(cgroup_writeback_init); #else /* CONFIG_CGROUP_WRITEBACK */ +static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) { } +static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) { } + static struct bdi_writeback * locked_inode_to_wb_and_lock_list(struct inode *inode) __releases(&inode->i_lock) @@ -2422,8 +2455,21 @@ void sync_inodes_sb(struct super_block * return; WARN_ON(!rwsem_is_locked(&sb->s_umount)); + /* + * Protect against inode wb switch in inode_switch_wbs_work_fn(). + * We want to synchronize syncs against switches. Synchronizing + * among syncs isn't necessary but it shouldn't matter especially + * as we're only protecting the issuing. + * + * Once we grab the rwsem, bdi_split_work_to_wbs() is guaranteed to + * see all the target wbs. If the wb membership hasn't changed, + * through the ordering visible to the caller; otherwise, through + * the transitive ordering via the rwsem. + */ + bdi_down_write_wb_switch_rwsem(bdi); bdi_split_work_to_wbs(bdi, &work, false); wb_wait_for_completion(bdi, &done); + bdi_up_write_wb_switch_rwsem(bdi); wait_sb_inodes(sb); } --- a/include/linux/backing-dev-defs.h +++ b/include/linux/backing-dev-defs.h @@ -189,6 +189,7 @@ struct backing_dev_info { #ifdef CONFIG_CGROUP_WRITEBACK struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */ struct rb_root cgwb_congested_tree; /* their congested states */ + struct rw_semaphore wb_switch_rwsem; /* no cgwb switch while syncing */ #else struct bdi_writeback_congested *wb_congested; #endif --- a/mm/backing-dev.c +++ b/mm/backing-dev.c @@ -813,6 +813,9 @@ static int cgwb_bdi_init(struct backing_ wb_congested_put(bdi->wb_congested); return err; } + + init_rwsem(&bdi->wb_switch_rwsem); + return 0; }