From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail.kernel.org ([198.145.29.99]:39458 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750977AbeERMeS (ORCPT ); Fri, 18 May 2018 08:34:18 -0400 From: Jeff Layton To: linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org Cc: viro@ZenIV.linux.org.uk, willy@infradead.org, andres@anarazel.de Subject: [RFC PATCH 00/11] vfs: have syncfs return an error when inode writeback fails Date: Fri, 18 May 2018 08:34:04 -0400 Message-Id: <20180518123415.28181-1-jlayton@kernel.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: From: Jeff Layton Currently, syncfs does not return errors when one of the inodes fails to be written back. It will return errors based on the legacy AS_EIO and AS_ENOSPC flags when syncing out the block device fails, but that's not particularly helpful for filesystems that aren't backed by a blockdev and more granular errors are often not caught. At LSF/MM this year, we had a discussion about tracking writeback errors on a per-sb basis to help remedy this. This patchset is an inital foray into doing that: The basic idea is to add a new errseq_t to struct super_block, and use that to track writeback errors for the filesystem as a whole. Because syncfs is not a common operation, it's not desirable to grow struct file in order to house an errseq_t cursor for it. Willy's suggestion was to reuse file->f_wb_err, but only when the file is opened with O_PATH. fsync always returns an error on an O_PATH fd, so this should be safe, and seems semantically sane, given that syncfs operates on the whole filesystem. The patchset starts by pushing the __sync_blockdev calls down into the various sync_fs operations. It then plumbs in an extra errseq_t pointer into the sync_fs ops, and the relevant callers. It then adds the new errseq_t field to struct super_block, and adds the machinery to handle syncfs on an O_PATH open correctly. Lastly, it has some filesystems return stored writeback errors in sync_fs ops when they are passed a non-NULL errseq_t pointer. This is just a start (more filesystems will need similar patches to the later ones in the pile). This all works vs. the fstests testcase I have. I haven't submitted that testcase yet, as I want to get consensus about the interface. Does it make sense to treat O_PATH opens specially like this? Another thing on the wishlist for this is to expose the per-sb errseq_t via another interface so that applications could periodically check to see whether there has been a writeback error without syncing out the whole fs. This patch doesn't add that yet, but once sb->s_wb_err is in place and tracking errors it should be simple to do (possibly via a new generic filesystem ioctl). Jeff Layton (11): vfs: push __sync_blockdev calls down into sync_fs routines vfs: add a new errseq_t pointer to sync_fs prototype vfs: add an errseq_t pointer to sync_filesystem vfs: add errseq_t pointer to __sync_filesystem fs: track per-sb writeback errors and report them to syncfs buffer: record blockdev write errors in super_block that backs them ext4: have sync_fs op report writeback errors when passed a since pointer xfs: have sync_fs op report writeback errors when passed a since pointer btrfs: have sync_fs op report writeback errors when passed a since pointer ext2: have sync_fs op report writeback errors when passed a since pointer vfs: have call_sync_fs op report writeback errors when passed a since pointer drivers/staging/ncpfs/inode.c | 2 +- fs/adfs/super.c | 2 +- fs/affs/super.c | 6 +++--- fs/befs/linuxvfs.c | 2 +- fs/block_dev.c | 3 ++- fs/btrfs/ctree.h | 2 +- fs/btrfs/ioctl.c | 2 +- fs/btrfs/super.c | 28 +++++++++++++++++++--------- fs/buffer.c | 2 ++ fs/cachefiles/interface.c | 2 +- fs/ceph/super.c | 2 +- fs/cifs/cifsfs.c | 2 +- fs/coda/inode.c | 2 +- fs/cramfs/inode.c | 2 +- fs/debugfs/inode.c | 2 +- fs/efs/super.c | 2 +- fs/exofs/super.c | 2 +- fs/ext2/super.c | 19 +++++++++++++------ fs/ext4/super.c | 19 +++++++++++-------- fs/f2fs/f2fs.h | 2 +- fs/f2fs/file.c | 6 +++--- fs/f2fs/namei.c | 16 ++++++++-------- fs/f2fs/segment.c | 2 +- fs/f2fs/super.c | 19 +++++++++++-------- fs/fat/inode.c | 2 +- fs/freevxfs/vxfs_super.c | 2 +- fs/fuse/inode.c | 2 +- fs/gfs2/super.c | 8 +++++--- fs/hfs/super.c | 6 +++--- fs/hfsplus/super.c | 10 +++++----- fs/hpfs/super.c | 2 +- fs/internal.h | 7 ------- fs/isofs/inode.c | 2 +- fs/jffs2/super.c | 4 ++-- fs/jfs/super.c | 7 +++---- fs/minix/inode.c | 2 +- fs/nfs/super.c | 2 +- fs/nilfs2/super.c | 9 +++++---- fs/ntfs/super.c | 2 +- fs/ocfs2/super.c | 8 ++++---- fs/open.c | 6 +++--- fs/openpromfs/inode.c | 2 +- fs/overlayfs/super.c | 4 ++-- fs/proc/root.c | 2 +- fs/pstore/inode.c | 2 +- fs/qnx4/inode.c | 2 +- fs/qnx6/inode.c | 2 +- fs/quota/dquot.c | 11 ++++------- fs/reiserfs/super.c | 8 ++++---- fs/romfs/super.c | 2 +- fs/squashfs/super.c | 2 +- fs/super.c | 4 ++-- fs/sync.c | 28 +++++++++++++++++----------- fs/sysv/inode.c | 7 +++---- fs/tracefs/inode.c | 2 +- fs/ubifs/super.c | 4 ++-- fs/udf/super.c | 6 +++--- fs/ufs/super.c | 6 +++--- fs/xfs/xfs_super.c | 18 +++++++++++++----- include/linux/fs.h | 35 +++++++++++++++++++++++++++++++++-- include/linux/pagemap.h | 5 ++++- 61 files changed, 223 insertions(+), 158 deletions(-) -- 2.17.0