From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-io0-f174.google.com ([209.85.223.174]:33066 "EHLO mail-io0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755466AbcGFWae (ORCPT ); Wed, 6 Jul 2016 18:30:34 -0400 Received: by mail-io0-f174.google.com with SMTP id t74so8026281ioi.0 for ; Wed, 06 Jul 2016 15:30:34 -0700 (PDT) From: Trond Myklebust To: linux-nfs@vger.kernel.org Subject: [PATCH v4 00/28] NFS writeback performance patches for v4.8 Date: Wed, 6 Jul 2016 18:29:37 -0400 Message-Id: <1467844205-76852-1-git-send-email-trond.myklebust@primarydata.com> Sender: linux-nfs-owner@vger.kernel.org List-ID: These patches are about improving the NFS I/O. 1) a new locking scheme between O_DIRECT and buffered I/O in order to allow parallelisation of O_DIRECT. 2) Remove instances of inode_lock that are preventing parallelism of writes and attribute revalidation. 3) Remove a redundant write throttling scheme that was conspiring with the FLUSH_COND_STABLE mode to drive writeback performance down. 4) Cache attributes more aggressively when we can assume close to open cache consistency. 5) Don't force data sync to disk where it is not needed. If attributes need to be up to date, we can usually get by with unstable writes to the server, particularly when we're using NFSv4 and can rely on stateids to cause operations to fail on a server reboot. 6) Fix a number of bugs around pNFS and attributes. Some (but not all) pNFS dialects may need data sync + layoutcommit in order to ensure that data and attribute updates are visible. In particular fix one bug was causing file size changes to be clobbered when layoutcommit was pending. Trond Myklebust (28): NFS: Don't flush caches for a getattr that races with writeback NFS: Cache access checks more aggressively NFS: Cache aggressively when file is open for writing NFS: Kill NFS_INO_NFS_INO_FLUSHING: it is a performance killer NFS: writepage of a single page should not be synchronous NFS: Don't hold the inode lock across fsync() NFS: Don't call COMMIT in ->releasepage() pNFS/files: Fix layoutcommit after a commit to DS pNFS/flexfiles: Fix layoutcommit after a commit to DS pNFS/flexfiles: Clean up calls to pnfs_set_layoutcommit() pNFS: Files and flexfiles always need to commit before layoutcommit pNFS: Ensure we layoutcommit before revalidating attributes pNFS: pnfs_layoutcommit_outstanding() is no longer used when !CONFIG_NFS_V4_1 NFS: Fix O_DIRECT verifier problems NFS: Ensure we reset the write verifier 'committed' value on resend. NFS: Remove racy size manipulations in O_DIRECT NFS Cleanup: move call to generic_write_checks() into fs/nfs/direct.c NFS: Move buffered I/O locking into nfs_file_write() NFS: Do not serialise O_DIRECT reads and writes NFS: Cleanup nfs_direct_complete() NFS: Remove redundant waits for O_DIRECT in fsync() and write_begin() NFS: Remove unused function nfs_revalidate_mapping_protected() NFS: Do not aggressively cache file attributes in the case of O_DIRECT NFS: Getattr doesn't require data sync semantics NFSv4.2: Fix a race in nfs42_proc_deallocate() NFSv4.2: Fix writeback races in nfs4_copy_file_range NFSv4.2: llseek(SEEK_HOLE) and llseek(SEEK_DATA) don't require data sync NFS nfs_vm_page_mkwrite: Don't freeze me, Bro... fs/nfs/Makefile | 2 +- fs/nfs/dir.c | 52 +++++++----- fs/nfs/direct.c | 93 +++++++-------------- fs/nfs/file.c | 96 ++++++--------------- fs/nfs/filelayout/filelayout.c | 12 +-- fs/nfs/flexfilelayout/flexfilelayout.c | 23 +++--- fs/nfs/inode.c | 133 ++++++++++++++--------------- fs/nfs/internal.h | 40 +++++++++ fs/nfs/io.c | 147 +++++++++++++++++++++++++++++++++ fs/nfs/nfs42proc.c | 21 ++++- fs/nfs/nfs4file.c | 14 +--- fs/nfs/nfs4xdr.c | 11 ++- fs/nfs/nfstrace.h | 1 - fs/nfs/pnfs.c | 5 +- fs/nfs/pnfs.h | 7 -- fs/nfs/pnfs_nfs.c | 7 ++ fs/nfs/write.c | 33 +++++--- include/linux/nfs_fs.h | 3 +- 18 files changed, 420 insertions(+), 280 deletions(-) create mode 100644 fs/nfs/io.c -- 2.7.4