From patchwork Wed Dec 13 15:23:10 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: David Howells X-Patchwork-Id: 13491412 Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.129.124]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 14F3D31596 for ; Wed, 13 Dec 2023 15:24:03 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=redhat.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=redhat.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (1024-bit key) header.d=redhat.com header.i=@redhat.com header.b="VQrkEmEu" DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1702481042; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding; bh=nElvBfJsqrTBHn7ca9kS/IDQA0m2TnErz3scXuPoO1A=; b=VQrkEmEuUNBt6qOcrfQ/lKPkk1t85/rd7DVJNTnjVeF1UDdBbAut5GK+PvzV6bBdGE3dYF cKC0Yt+3suasA8T4+xUq92KqGgJeSkRDe2ll3JW3BAqC+2rSHVdGgmIJG0dNJYId1cmbEJ ri/4dptnfyAHPN69MehSQWoG9LKpRuc= Received: from mimecast-mx02.redhat.com (mimecast-mx02.redhat.com [66.187.233.88]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.3, cipher=TLS_AES_256_GCM_SHA384) id us-mta-83-bNT4a_5wMzSOH84dTC7R4A-1; Wed, 13 Dec 2023 10:23:57 -0500 X-MC-Unique: bNT4a_5wMzSOH84dTC7R4A-1 Received: from smtp.corp.redhat.com (int-mx08.intmail.prod.int.rdu2.redhat.com [10.11.54.8]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by mimecast-mx02.redhat.com (Postfix) with ESMTPS id D3D8C86301A; Wed, 13 Dec 2023 15:23:55 +0000 (UTC) Received: from warthog.procyon.org.com (unknown [10.42.28.2]) by smtp.corp.redhat.com (Postfix) with ESMTP id B84EDC1596E; Wed, 13 Dec 2023 15:23:52 +0000 (UTC) From: David Howells To: Jeff Layton , Steve French Cc: David Howells , Matthew Wilcox , Marc Dionne , Paulo Alcantara , Shyam Prasad N , Tom Talpey , Dominique Martinet , Eric Van Hensbergen , Ilya Dryomov , Christian Brauner , linux-cachefs@redhat.com, linux-afs@lists.infradead.org, linux-cifs@vger.kernel.org, linux-nfs@vger.kernel.org, ceph-devel@vger.kernel.org, v9fs@lists.linux.dev, linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, netdev@vger.kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH v4 00/39] netfs, afs, 9p: Delegate high-level I/O to netfslib Date: Wed, 13 Dec 2023 15:23:10 +0000 Message-ID: <20231213152350.431591-1-dhowells@redhat.com> Precedence: bulk X-Mailing-List: v9fs@lists.linux.dev List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-Scanned-By: MIMEDefang 3.4.1 on 10.11.54.8 Hi Jeff, Steve, Dominique, I have been working on my netfslib helpers to the point that I can run xfstests on AFS to completion (both with write-back buffering and, with a small patch, write-through buffering in the pagecache). I have a patch for 9P, but am currently unable to test it. The patches remove a little over 800 lines from AFS, 300 from 9P, albeit with around 3000 lines added to netfs. Hopefully, I will be able to remove a bunch of lines from Ceph too. I've split the CIFS patches out to a separate branch, cifs-netfs, where a further 2000+ lines are removed. I can run a certain amount of xfstests on CIFS, though I'm running into ksmbd issues and not all the tests work correctly because of issues between fallocate and what the SMB protocol actually supports. I've also dropped the content-crypto patches out for the moment as they're only usable by the ceph changes which I'm still working on. The patch to use PG_writeback instead of PG_fscache for writing to the cache has also been deferred, pending 9p, afs, ceph and cifs all being converted. The main aims of these patches are to get high-level I/O and knowledge of the pagecache out of the filesystem drivers as much as possible and to get rid, as much of possible, of the knowledge that pages/folios exist. Further, I would like to see ->write_begin, ->write_end and ->launder_folio go away. Features that are added by these patches to that which is already there in netfslib: (1) NFS-style (and Ceph-style) locking around DIO vs buffered I/O calls to prevent these from happening at the same time. mmap'd I/O can, of necessity, happen at any time ignoring these locks. (2) Support for unbuffered I/O. The data is kept in the bounce buffer and the pagecache is not used. This can be turned on with an inode flag. (3) Support for direct I/O. This is basically unbuffered I/O with some extra restrictions and no RMW. (4) Support for using a bounce buffer in an operation. The bounce buffer may be bigger than the target data/buffer, allowing for crypto rounding. (5) ->write_begin() and ->write_end() are ignored in favour of merging all of that into one function, netfs_perform_write(), thereby avoiding the function pointer traversals. (6) Support for write-through caching in the pagecache. netfs_perform_write() adds the pages is modifies to an I/O operation as it goes and directly marks them writeback rather than dirty. When writing back from write-through, it limits the range written back. This should allow CIFS to deal with byte-range mandatory locks correctly. (7) O_*SYNC and RWF_*SYNC writes use write-through rather than writing to the pagecache and then flushing afterwards. An AIO O_*SYNC write will notify of completion when the sub-writes all complete. (8) Support for write-streaming where modifed data is held in !uptodate folios, with a private struct attached indicating the range that is valid. (9) Support for write grouping, multiplexing a pointer to a group in the folio private data with the write-streaming data. The writepages algorithm only writes stuff back that's in the nominated group. This is intended for use by Ceph to write is snaps in order. (10) Skipping reads for which we know the server could only supply zeros or EOF (for instance if we've done a local write that leaves a hole in the file and extends the local inode size). General notes: (1) The fscache module is merged into the netfslib module to avoid cyclic exported symbol usage that prevents either module from being loaded. (2) Some helpers from fscache are reassigned to netfslib by name. (3) netfslib now makes use of folio->private, which means the filesystem can't use it. (4) The filesystem provides wrappers to call the write helpers, allowing it to do pre-validation, oplock/capability fetching and the passing in of write group info. (5) I want to try flushing the data when tearing down an inode before invalidating it to try and render launder_folio unnecessary. (6) Write-through caching will generate and dispatch write subrequests as it gathers enough data to hit wsize and has whole pages that at least span that size. This needs to be a bit more flexible, allowing for a filesystem such as CIFS to have a variable wsize. (7) The filesystem driver is just given read and write calls with an iov_iter describing the data/buffer to use. Ideally, they don't see pages or folios at all. A function, extract_iter_to_sg(), is already available to decant part of an iterator into a scatterlist for crypto purposes. 9P notes: (1) I haven't managed to test this as I haven't been able to get Ganesha to work correctly with 9P. (2) Writes should now occur in larger-than-page-sized chunks. (3) It should be possible to turn on multipage folio support in 9P now. Changes ======= ver #4) - Slimmed down the branch: - Split the cifs-related patches off to a separate branch (cifs-netfs) - Deferred the content-encryption to the in-progress ceph changes. - Deferred the use-PG_writeback rather than PG_fscache patch - Rebased on a later linux-next with afs-rotation patches. ver #3) - Moved the fscache module into netfslib to avoid export cycles. - Fixed a bunch of bugs. - Got CIFS to pass as much of xfstests as possible. - Added a patch to make 9P use all the helpers. - Added a patch to stop using PG_fscache, but rather dirty pages on reading and have writepages write to the cache. ver #2) - Folded the addition of NETFS_RREQ_NONBLOCK/BLOCKED into first patch that uses them. - Folded addition of rsize member into first user. - Don't set rsize in ceph (yet) and set it in kafs to 256KiB. cifs sets it dynamically. - Moved direct_bv next to direct_bv_count in struct netfs_io_request and labelled it with a __counted_by(). - Passed flags into netfs_xa_store_and_mark() rather than two bools. - Removed netfs_set_up_buffer() as it wasn't used. David Link: https://lore.kernel.org/r/20231013160423.2218093-1-dhowells@redhat.com/ # v1 Link: https://lore.kernel.org/r/20231117211544.1740466-1-dhowells@redhat.com/ # v2 David Howells (39): netfs, fscache: Move fs/fscache/* into fs/netfs/ netfs, fscache: Combine fscache with netfs netfs, fscache: Remove ->begin_cache_operation netfs, fscache: Move /proc/fs/fscache to /proc/fs/netfs and put in a symlink netfs: Move pinning-for-writeback from fscache to netfs netfs: Add a procfile to list in-progress requests netfs: Allow the netfs to make the io (sub)request alloc larger netfs: Add a ->free_subrequest() op afs: Don't use folio->private to record partial modification netfs: Provide invalidate_folio and release_folio calls netfs: Implement unbuffered/DIO vs buffered I/O locking netfs: Add iov_iters to (sub)requests to describe various buffers netfs: Add support for DIO buffering netfs: Provide tools to create a buffer in an xarray netfs: Add bounce buffering support netfs: Add func to calculate pagecount/size-limited span of an iterator netfs: Limit subrequest by size or number of segments netfs: Export netfs_put_subrequest() and some tracepoints netfs: Extend the netfs_io_*request structs to handle writes netfs: Add a hook to allow tell the netfs to update its i_size netfs: Make netfs_put_request() handle a NULL pointer netfs: Make the refcounting of netfs_begin_read() easier to use netfs: Prep to use folio->private for write grouping and streaming write netfs: Dispatch write requests to process a writeback slice netfs: Provide func to copy data to pagecache for buffered write netfs: Make netfs_read_folio() handle streaming-write pages netfs: Allocate multipage folios in the writepath netfs: Implement support for unbuffered/DIO read netfs: Implement unbuffered/DIO write support netfs: Implement buffered write API netfs: Allow buffered shared-writeable mmap through netfs_page_mkwrite() netfs: Provide netfs_file_read_iter() netfs, cachefiles: Pass upper bound length to allow expansion netfs: Provide a writepages implementation netfs: Provide a launder_folio implementation netfs: Implement a write-through caching option netfs: Optimise away reads above the point at which there can be no data afs: Use the netfs write helpers 9p: Use netfslib read/write_iter Documentation/filesystems/netfs_library.rst | 23 +- MAINTAINERS | 2 +- fs/9p/vfs_addr.c | 352 +---- fs/9p/vfs_file.c | 89 +- fs/9p/vfs_inode.c | 5 +- fs/9p/vfs_super.c | 14 +- fs/Kconfig | 1 - fs/Makefile | 1 - fs/afs/file.c | 213 +-- fs/afs/inode.c | 26 +- fs/afs/internal.h | 72 +- fs/afs/super.c | 2 +- fs/afs/write.c | 826 +---------- fs/cachefiles/internal.h | 2 +- fs/cachefiles/io.c | 10 +- fs/cachefiles/ondemand.c | 2 +- fs/ceph/addr.c | 25 +- fs/ceph/cache.h | 35 +- fs/ceph/inode.c | 2 +- fs/fs-writeback.c | 10 +- fs/fscache/Kconfig | 40 - fs/fscache/Makefile | 16 - fs/fscache/internal.h | 277 ---- fs/netfs/Kconfig | 39 + fs/netfs/Makefile | 22 +- fs/netfs/buffered_read.c | 229 ++- fs/netfs/buffered_write.c | 1247 +++++++++++++++++ fs/netfs/direct_read.c | 252 ++++ fs/netfs/direct_write.c | 170 +++ fs/{fscache/cache.c => netfs/fscache_cache.c} | 0 .../cookie.c => netfs/fscache_cookie.c} | 0 fs/netfs/fscache_internal.h | 14 + fs/{fscache/io.c => netfs/fscache_io.c} | 42 +- fs/{fscache/main.c => netfs/fscache_main.c} | 25 +- fs/{fscache/proc.c => netfs/fscache_proc.c} | 23 +- fs/{fscache/stats.c => netfs/fscache_stats.c} | 4 +- .../volume.c => netfs/fscache_volume.c} | 0 fs/netfs/internal.h | 288 ++++ fs/netfs/io.c | 214 ++- fs/netfs/iterator.c | 97 ++ fs/netfs/locking.c | 215 +++ fs/netfs/main.c | 110 ++ fs/netfs/misc.c | 260 ++++ fs/netfs/objects.c | 63 +- fs/netfs/output.c | 478 +++++++ fs/netfs/stats.c | 31 +- fs/nfs/Kconfig | 4 +- fs/nfs/fscache.c | 7 - fs/smb/client/cifsfs.c | 9 +- fs/smb/client/file.c | 18 +- fs/smb/client/fscache.c | 2 +- include/linux/fs.h | 2 +- include/linux/fscache.h | 45 - include/linux/netfs.h | 176 ++- include/linux/writeback.h | 2 +- include/trace/events/afs.h | 31 - include/trace/events/netfs.h | 155 +- mm/filemap.c | 1 + 58 files changed, 4197 insertions(+), 2123 deletions(-) delete mode 100644 fs/fscache/Kconfig delete mode 100644 fs/fscache/Makefile delete mode 100644 fs/fscache/internal.h create mode 100644 fs/netfs/buffered_write.c create mode 100644 fs/netfs/direct_read.c create mode 100644 fs/netfs/direct_write.c rename fs/{fscache/cache.c => netfs/fscache_cache.c} (100%) rename fs/{fscache/cookie.c => netfs/fscache_cookie.c} (100%) create mode 100644 fs/netfs/fscache_internal.h rename fs/{fscache/io.c => netfs/fscache_io.c} (86%) rename fs/{fscache/main.c => netfs/fscache_main.c} (84%) rename fs/{fscache/proc.c => netfs/fscache_proc.c} (58%) rename fs/{fscache/stats.c => netfs/fscache_stats.c} (97%) rename fs/{fscache/volume.c => netfs/fscache_volume.c} (100%) create mode 100644 fs/netfs/locking.c create mode 100644 fs/netfs/misc.c create mode 100644 fs/netfs/output.c Reviewed-by: Jeff Layton