KVM: Don't pre-reserve xarray entries when storing empty/NULL attributesgrafted

Skip the xarray reservation loop when clearing all memory attributes, as
storing NULL only erases the entry and never needs to allocate, so no
reservation (and no cleanup of a failed one) is required in that case.

Suggested-by: Sean Christopherson <seanjc@google.com>
Cc: David Ballesteros <davimaba.v@proton.me>
Signed-off-by: Zeng Chi <zengchi@kylinos.cn>
Link: https://patch.msgid.link/20260921102442.1232375-1-zeng_chi911@163.com
[sean: split to separate patch]
Signed-off-by: Sean Christopherson <seanjc@google.com>

29 files changed, 10413 insertions, 0 deletions

diff --git a/fs/netfs/Kconfig b/fs/netfs/Kconfig
new file mode 100644
index 000000000..7701c037c
--- /dev/null
+++ b/fs/netfs/Kconfig
@@ -0,0 +1,60 @@
+# SPDX-License-Identifier: GPL-2.0-only
+
+config NETFS_SUPPORT
+	tristate
+	help
+	  This option enables support for network filesystems, including
+	  helpers for high-level buffered I/O, abstracting out read
+	  segmentation, local caching and transparent huge page support.
+
+config NETFS_STATS
+	bool "Gather statistical information on local caching"
+	depends on NETFS_SUPPORT && PROC_FS
+	help
+	  This option causes statistical information to be gathered on local
+	  caching and exported through file:
+
+		/proc/fs/fscache/stats
+
+	  The gathering of statistics adds a certain amount of overhead to
+	  execution as there are a quite a few stats gathered, and on a
+	  multi-CPU system these may be on cachelines that keep bouncing
+	  between CPUs.  On the other hand, the stats are very useful for
+	  debugging purposes.  Saying 'Y' here is recommended.
+
+config NETFS_DEBUG
+	bool "Enable dynamic debugging netfslib and FS-Cache"
+	depends on NETFS_SUPPORT
+	help
+	  This permits debugging to be dynamically enabled in the local caching
+	  management module.  If this is set, the debugging output may be
+	  enabled by setting bits in /sys/module/netfs/parameters/debug.
+
+config FSCACHE
+	bool "General filesystem local caching manager"
+	depends on NETFS_SUPPORT
+	help
+	  This option enables a generic filesystem caching manager that can be
+	  used by various network and other filesystems to cache data locally.
+	  Different sorts of caches can be plugged in, depending on the
+	  resources available.
+
+	  See Documentation/filesystems/caching/fscache.rst for more information.
+
+config FSCACHE_STATS
+	bool "Gather statistical information on local caching"
+	depends on FSCACHE && PROC_FS
+	select NETFS_STATS
+	help
+	  This option causes statistical information to be gathered on local
+	  caching and exported through file:
+
+		/proc/fs/fscache/stats
+
+	  The gathering of statistics adds a certain amount of overhead to
+	  execution as there are a quite a few stats gathered, and on a
+	  multi-CPU system these may be on cachelines that keep bouncing
+	  between CPUs.  On the other hand, the stats are very useful for
+	  debugging purposes.  Saying 'Y' here is recommended.
+
+	  See Documentation/filesystems/caching/fscache.rst for more information.
diff --git a/fs/netfs/Makefile b/fs/netfs/Makefile
new file mode 100644
index 000000000..b43188d64
--- /dev/null
+++ b/fs/netfs/Makefile
@@ -0,0 +1,36 @@
+# SPDX-License-Identifier: GPL-2.0
+
+netfs-y := \
+	buffered_read.o \
+	buffered_write.o \
+	direct_read.o \
+	direct_write.o \
+	iterator.o \
+	locking.o \
+	main.o \
+	misc.o \
+	objects.o \
+	read_collect.o \
+	read_pgpriv2.o \
+	read_retry.o \
+	read_single.o \
+	rolling_buffer.o \
+	write_collect.o \
+	write_issue.o \
+	write_retry.o
+
+netfs-$(CONFIG_NETFS_STATS) += stats.o
+
+netfs-$(CONFIG_FSCACHE) += \
+	fscache_cache.o \
+	fscache_cookie.o \
+	fscache_io.o \
+	fscache_main.o \
+	fscache_volume.o
+
+ifeq ($(CONFIG_PROC_FS),y)
+netfs-$(CONFIG_FSCACHE) += fscache_proc.o
+endif
+netfs-$(CONFIG_FSCACHE_STATS) += fscache_stats.o
+
+obj-$(CONFIG_NETFS_SUPPORT) += netfs.o
diff --git a/fs/netfs/buffered_read.c b/fs/netfs/buffered_read.c
new file mode 100644
index 000000000..7fdfa4f27
--- /dev/null
+++ b/fs/netfs/buffered_read.c
@@ -0,0 +1,840 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Network filesystem high-level buffered read support.
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/task_io_accounting_ops.h>
+#include "internal.h"
+
+static void netfs_cache_expand_readahead(struct netfs_io_request *rreq,
+					 unsigned long long *_start,
+					 unsigned long long *_len,
+					 unsigned long long i_size)
+{
+	struct netfs_cache_resources *cres = &rreq->cache_resources;
+
+	if (cres->ops && cres->ops->expand_readahead)
+		cres->ops->expand_readahead(cres, _start, _len, i_size);
+}
+
+static void netfs_rreq_expand(struct netfs_io_request *rreq,
+			      struct readahead_control *ractl)
+{
+	/* Give the cache a chance to change the request parameters.  The
+	 * resultant request must contain the original region.
+	 */
+	netfs_cache_expand_readahead(rreq, &rreq->start, &rreq->len, rreq->i_size);
+
+	/* Give the netfs a chance to change the request parameters.  The
+	 * resultant request must contain the original region.
+	 */
+	if (rreq->netfs_ops->expand_readahead)
+		rreq->netfs_ops->expand_readahead(rreq);
+
+	/* Expand the request if the cache wants it to start earlier.  Note
+	 * that the expansion may get further extended if the VM wishes to
+	 * insert THPs and the preferred start and/or end wind up in the middle
+	 * of THPs.
+	 *
+	 * If this is the case, however, the THP size should be an integer
+	 * multiple of the cache granule size, so we get a whole number of
+	 * granules to deal with.
+	 */
+	if (rreq->start  != readahead_pos(ractl) ||
+	    rreq->len != readahead_length(ractl)) {
+		readahead_expand(ractl, rreq->start, rreq->len);
+		rreq->start  = readahead_pos(ractl);
+		rreq->len = readahead_length(ractl);
+
+		trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
+				 netfs_read_trace_expanded);
+	}
+}
+
+/*
+ * Begin an operation, and fetch the stored zero point value from the cookie if
+ * available.
+ */
+static int netfs_begin_cache_read(struct netfs_io_request *rreq, struct netfs_inode *ctx)
+{
+	return fscache_begin_read_operation(&rreq->cache_resources, netfs_i_cookie(ctx));
+}
+
+/*
+ * netfs_prepare_read_iterator - Prepare the subreq iterator for I/O
+ * @subreq: The subrequest to be set up
+ *
+ * Prepare the I/O iterator representing the read buffer on a subrequest for
+ * the filesystem to use for I/O (it can be passed directly to a socket).  This
+ * is intended to be called from the ->issue_read() method once the filesystem
+ * has trimmed the request to the size it wants.
+ *
+ * Returns the limited size if successful and -ENOMEM if insufficient memory
+ * available.
+ *
+ * [!] NOTE: This must be run in the same thread as ->issue_read() was called
+ * in as we access the readahead_control struct.
+ */
+static ssize_t netfs_prepare_read_iterator(struct netfs_io_subrequest *subreq,
+					   struct readahead_control *ractl)
+{
+	struct netfs_io_request *rreq = subreq->rreq;
+	size_t rsize = subreq->len;
+
+	if (subreq->source == NETFS_DOWNLOAD_FROM_SERVER)
+		rsize = umin(rsize, rreq->io_streams[0].sreq_max_len);
+
+	if (ractl) {
+		/* If we don't have sufficient folios in the rolling buffer,
+		 * extract a folioq's worth from the readahead region at a time
+		 * into the buffer.  Note that this acquires a ref on each page
+		 * that we will need to release later - but we don't want to do
+		 * that until after we've started the I/O.
+		 */
+		struct folio_batch put_batch;
+
+		folio_batch_init(&put_batch);
+		while (rreq->submitted < subreq->start + rsize) {
+			ssize_t added;
+
+			added = rolling_buffer_load_from_ra(&rreq->buffer, ractl,
+							    &put_batch);
+			if (added < 0) {
+				folio_batch_release(&put_batch);
+				return added;
+			}
+			rreq->submitted += added;
+		}
+		folio_batch_release(&put_batch);
+	}
+
+	subreq->len = rsize;
+	if (unlikely(rreq->io_streams[0].sreq_max_segs)) {
+		size_t limit = netfs_limit_iter(&rreq->buffer.iter, 0, rsize,
+						rreq->io_streams[0].sreq_max_segs);
+
+		if (limit < rsize) {
+			subreq->len = limit;
+			trace_netfs_sreq(subreq, netfs_sreq_trace_limited);
+		}
+	}
+
+	subreq->io_iter	= rreq->buffer.iter;
+
+	iov_iter_truncate(&subreq->io_iter, subreq->len);
+	rolling_buffer_advance(&rreq->buffer, subreq->len);
+	return subreq->len;
+}
+
+static enum netfs_io_source netfs_cache_prepare_read(struct netfs_io_request *rreq,
+						     struct netfs_io_subrequest *subreq,
+						     loff_t i_size)
+{
+	struct netfs_cache_resources *cres = &rreq->cache_resources;
+	enum netfs_io_source source;
+
+	if (!cres->ops)
+		return NETFS_DOWNLOAD_FROM_SERVER;
+	source = cres->ops->prepare_read(subreq, i_size);
+	trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
+	return source;
+
+}
+
+/*
+ * Issue a read against the cache.
+ * - Eats the caller's ref on subreq.
+ */
+static void netfs_read_cache_to_pagecache(struct netfs_io_request *rreq,
+					  struct netfs_io_subrequest *subreq)
+{
+	struct netfs_cache_resources *cres = &rreq->cache_resources;
+
+	netfs_stat(&netfs_n_rh_read);
+	cres->ops->read(cres, subreq->start, &subreq->io_iter, NETFS_READ_HOLE_IGNORE,
+			netfs_cache_read_terminated, subreq);
+}
+
+void netfs_queue_read(struct netfs_io_request *rreq,
+		      struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+
+	__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+
+	/* We add to the end of the list whilst the collector may be walking
+	 * the list.  The collector only goes nextwards and uses the lock to
+	 * remove entries off of the front.
+	 */
+	spin_lock(&rreq->lock);
+	/* Write IN_PROGRESS before pointer to new subreq */
+	list_add_tail_release(&subreq->rreq_link, &stream->subrequests);
+	if (list_is_first(&subreq->rreq_link, &stream->subrequests)) {
+		if (!stream->active) {
+			stream->collected_to = subreq->start;
+			/* Store list pointers before active flag */
+			smp_store_release(&stream->active, true);
+		}
+	}
+
+	spin_unlock(&rreq->lock);
+}
+
+static void netfs_issue_read(struct netfs_io_request *rreq,
+			     struct netfs_io_subrequest *subreq)
+{
+	switch (subreq->source) {
+	case NETFS_DOWNLOAD_FROM_SERVER:
+		rreq->netfs_ops->issue_read(subreq);
+		break;
+	case NETFS_READ_FROM_CACHE:
+		netfs_read_cache_to_pagecache(rreq, subreq);
+		break;
+	default:
+		__set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
+		subreq->error = 0;
+		iov_iter_zero(subreq->len, &subreq->io_iter);
+		subreq->transferred = subreq->len;
+		netfs_read_subreq_terminated(subreq);
+		break;
+	}
+}
+
+/*
+ * Perform a read to the pagecache from a series of sources of different types,
+ * slicing up the region to be read according to available cache blocks and
+ * network rsize.
+ */
+static void netfs_read_to_pagecache(struct netfs_io_request *rreq,
+				    struct readahead_control *ractl)
+{
+	unsigned long long start = rreq->start;
+	ssize_t size = rreq->len;
+	int ret = 0;
+
+	do {
+		struct netfs_io_subrequest *subreq;
+		enum netfs_io_source source = NETFS_SOURCE_UNKNOWN;
+		ssize_t slice;
+
+		subreq = netfs_alloc_subrequest(rreq);
+		if (!subreq) {
+			ret = -ENOMEM;
+			break;
+		}
+
+		subreq->start	= start;
+		subreq->len	= size;
+
+		netfs_queue_read(rreq, subreq);
+
+		source = netfs_cache_prepare_read(rreq, subreq, rreq->i_size);
+		subreq->source = source;
+		if (source == NETFS_DOWNLOAD_FROM_SERVER) {
+			unsigned long long zero_point = netfs_read_zero_point(rreq->inode);
+			unsigned long long zp = umin(zero_point, rreq->i_size);
+			size_t len = subreq->len;
+
+			if (unlikely(rreq->origin == NETFS_READ_SINGLE))
+				zp = rreq->i_size;
+			if (subreq->start >= zp) {
+				subreq->source = source = NETFS_FILL_WITH_ZEROES;
+				goto fill_with_zeroes;
+			}
+
+			if (len > zp - subreq->start)
+				len = zp - subreq->start;
+			if (len == 0) {
+				pr_err("ZERO-LEN READ: R=%08x[%x] l=%zx/%zx s=%llx z=%llx i=%llx",
+				       rreq->debug_id, subreq->debug_index,
+				       subreq->len, size,
+				       subreq->start, zero_point, rreq->i_size);
+				netfs_cancel_read(subreq, ret);
+				break;
+			}
+			subreq->len = len;
+
+			netfs_stat(&netfs_n_rh_download);
+			if (rreq->netfs_ops->prepare_read) {
+				ret = rreq->netfs_ops->prepare_read(subreq);
+				if (ret < 0) {
+					netfs_cancel_read(subreq, ret);
+					break;
+				}
+				trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
+			}
+			goto issue;
+		}
+
+	fill_with_zeroes:
+		if (source == NETFS_FILL_WITH_ZEROES) {
+			subreq->source = NETFS_FILL_WITH_ZEROES;
+			trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
+			netfs_stat(&netfs_n_rh_zero);
+			goto issue;
+		}
+
+		if (source == NETFS_READ_FROM_CACHE) {
+			trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
+			goto issue;
+		}
+
+		pr_err("Unexpected read source %u\n", source);
+		WARN_ON_ONCE(1);
+		netfs_cancel_read(subreq, ret);
+		break;
+
+	issue:
+		slice = netfs_prepare_read_iterator(subreq, ractl);
+		if (slice < 0) {
+			ret = slice;
+			netfs_cancel_read(subreq, ret);
+			break;
+		}
+		start += slice;
+		size -= slice;
+		if (size <= 0) {
+			smp_wmb(); /* Write lists before ALL_QUEUED. */
+			set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+		}
+
+		netfs_issue_read(rreq, subreq);
+
+		if (test_bit(NETFS_RREQ_PAUSE, &rreq->flags))
+			netfs_wait_for_paused_read(rreq);
+		if (test_bit(NETFS_RREQ_FAILED, &rreq->flags))
+			break;
+		cond_resched();
+	} while (size > 0);
+
+	if (unlikely(size > 0)) {
+		smp_wmb(); /* Write lists before ALL_QUEUED. */
+		set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+		netfs_wake_collector(rreq);
+	}
+
+	/* Defer error return as we may need to wait for outstanding I/O. */
+	cmpxchg(&rreq->error, 0, ret);
+}
+
+/**
+ * netfs_readahead - Helper to manage a read request
+ * @ractl: The description of the readahead request
+ *
+ * Fulfil a readahead request by drawing data from the cache if possible, or
+ * the netfs if not.  Space beyond the EOF is zero-filled.  Multiple I/O
+ * requests from different sources will get munged together.  If necessary, the
+ * readahead window can be expanded in either direction to a more convenient
+ * alighment for RPC efficiency or to make storage in the cache feasible.
+ *
+ * The calling netfs must initialise a netfs context contiguous to the vfs
+ * inode before calling this.
+ *
+ * This is usable whether or not caching is enabled.
+ */
+void netfs_readahead(struct readahead_control *ractl)
+{
+	struct netfs_io_request *rreq;
+	struct netfs_inode *ictx = netfs_inode(ractl->mapping->host);
+	unsigned long long start = readahead_pos(ractl);
+	size_t size = readahead_length(ractl);
+	int ret;
+
+	rreq = netfs_alloc_request(ractl->mapping, ractl->file, start, size,
+				   NETFS_READAHEAD);
+	if (IS_ERR(rreq))
+		return;
+
+	__set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags);
+
+	ret = netfs_begin_cache_read(rreq, ictx);
+	if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
+		goto cleanup_free;
+
+	netfs_stat(&netfs_n_rh_readahead);
+	trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
+			 netfs_read_trace_readahead);
+
+	netfs_rreq_expand(rreq, ractl);
+
+	rreq->submitted = rreq->start;
+	if (rolling_buffer_init(&rreq->buffer, rreq->debug_id, ITER_DEST, rreq->gfp) < 0)
+		goto cleanup_free;
+	netfs_read_to_pagecache(rreq, ractl);
+
+	return netfs_put_request(rreq, netfs_rreq_trace_put_return);
+
+cleanup_free:
+	return netfs_put_failed_request(rreq);
+}
+EXPORT_SYMBOL(netfs_readahead);
+
+/*
+ * Create a rolling buffer with a single occupying folio.
+ */
+static int netfs_create_singular_buffer(struct netfs_io_request *rreq, struct folio *folio,
+					unsigned int rollbuf_flags)
+{
+	ssize_t added;
+
+	if (rolling_buffer_init(&rreq->buffer, rreq->debug_id, ITER_DEST, rreq->gfp) < 0)
+		return -ENOMEM;
+
+	added = rolling_buffer_append(&rreq->buffer, folio, rollbuf_flags, rreq->gfp);
+	if (added < 0)
+		return added;
+	rreq->submitted = rreq->start + added;
+	return 0;
+}
+
+/*
+ * Read into gaps in a folio partially filled by a streaming write.
+ */
+static int netfs_read_gaps(struct file *file, struct folio *folio)
+{
+	struct netfs_io_request *rreq;
+	struct address_space *mapping = folio->mapping;
+	struct netfs_group *group = netfs_folio_group(folio);
+	struct netfs_folio *finfo = netfs_folio_info(folio);
+	struct netfs_inode *ctx = netfs_inode(mapping->host);
+	struct folio *sink = NULL;
+	struct bio_vec *bvec;
+	unsigned int from = finfo->dirty_offset;
+	unsigned int to = from + finfo->dirty_len;
+	unsigned int off = 0, i = 0;
+	size_t flen = folio_size(folio);
+	size_t nr_bvec = flen / PAGE_SIZE + 2;
+	size_t part;
+	int ret;
+
+	_enter("%lx", folio->index);
+
+	rreq = netfs_alloc_request(mapping, file, folio_pos(folio), flen, NETFS_READ_GAPS);
+	if (IS_ERR(rreq)) {
+		ret = PTR_ERR(rreq);
+		goto alloc_error;
+	}
+
+	ret = netfs_begin_cache_read(rreq, ctx);
+	if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
+		goto discard;
+
+	netfs_stat(&netfs_n_rh_read_folio);
+	trace_netfs_read(rreq, rreq->start, rreq->len, netfs_read_trace_read_gaps);
+
+	/* Fiddle the buffer so that a gap at the beginning and/or a gap at the
+	 * end get copied to, but the middle is discarded.
+	 */
+	ret = -ENOMEM;
+	bvec = kmalloc_objs(*bvec, nr_bvec);
+	if (!bvec)
+		goto discard;
+
+	sink = folio_alloc(GFP_KERNEL, 0);
+	if (!sink) {
+		kfree(bvec);
+		goto discard;
+	}
+
+	trace_netfs_folio(folio, netfs_folio_trace_read_gaps);
+
+	rreq->direct_bv = bvec;
+	rreq->direct_bv_count = nr_bvec;
+	if (from > 0) {
+		bvec_set_folio(&bvec[i++], folio, from, 0);
+		off = from;
+	}
+	while (off < to) {
+		part = min_t(size_t, to - off, PAGE_SIZE);
+		bvec_set_folio(&bvec[i++], sink, part, 0);
+		off += part;
+	}
+	if (to < flen)
+		bvec_set_folio(&bvec[i++], folio, flen - to, to);
+	iov_iter_bvec(&rreq->buffer.iter, ITER_DEST, bvec, i, rreq->len);
+	rreq->submitted = rreq->start + flen;
+
+	netfs_read_to_pagecache(rreq, NULL);
+
+	ret = netfs_wait_for_read(rreq);
+	if (ret >= 0) {
+		if (group)
+			folio_change_private(folio, group);
+		else
+			folio_detach_private(folio);
+		kfree(finfo);
+		trace_netfs_folio(folio, netfs_folio_trace_filled_gaps);
+		flush_dcache_folio(folio);
+		folio_mark_uptodate(folio);
+	}
+
+	if (sink)
+		folio_put(sink);
+	folio_unlock(folio);
+	netfs_put_request(rreq, netfs_rreq_trace_put_return);
+	return ret < 0 ? ret : 0;
+
+discard:
+	netfs_put_failed_request(rreq);
+alloc_error:
+	folio_unlock(folio);
+	return ret;
+}
+
+/**
+ * netfs_read_folio - Helper to manage a read_folio request
+ * @file: The file to read from
+ * @folio: The folio to read
+ *
+ * Fulfil a read_folio request by drawing data from the cache if
+ * possible, or the netfs if not.  Space beyond the EOF is zero-filled.
+ * Multiple I/O requests from different sources will get munged together.
+ *
+ * The calling netfs must initialise a netfs context contiguous to the vfs
+ * inode before calling this.
+ *
+ * This is usable whether or not caching is enabled.
+ */
+int netfs_read_folio(struct file *file, struct folio *folio)
+{
+	struct address_space *mapping = folio->mapping;
+	struct netfs_io_request *rreq;
+	struct netfs_inode *ctx = netfs_inode(mapping->host);
+	int ret;
+
+	folio_wait_writeback(folio);
+
+	if (folio_test_dirty(folio))
+		return netfs_read_gaps(file, folio);
+
+	_enter("%lx", folio->index);
+
+	rreq = netfs_alloc_request(mapping, file,
+				   folio_pos(folio), folio_size(folio),
+				   NETFS_READPAGE);
+	if (IS_ERR(rreq)) {
+		ret = PTR_ERR(rreq);
+		goto alloc_error;
+	}
+
+	ret = netfs_begin_cache_read(rreq, ctx);
+	if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
+		goto discard;
+
+	netfs_stat(&netfs_n_rh_read_folio);
+	trace_netfs_read(rreq, rreq->start, rreq->len, netfs_read_trace_readpage);
+
+	/* Set up the output buffer */
+	ret = netfs_create_singular_buffer(rreq, folio, 0);
+	if (ret < 0)
+		goto discard;
+
+	netfs_read_to_pagecache(rreq, NULL);
+	ret = netfs_wait_for_read(rreq);
+	netfs_put_request(rreq, netfs_rreq_trace_put_return);
+	return ret < 0 ? ret : 0;
+
+discard:
+	netfs_put_failed_request(rreq);
+alloc_error:
+	folio_unlock(folio);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_read_folio);
+
+/*
+ * Prepare a folio for writing without reading first
+ * @folio: The folio being prepared
+ * @pos: starting position for the write
+ * @len: length of write
+ * @always_fill: T if the folio should always be completely filled/cleared
+ *
+ * In some cases, write_begin doesn't need to read at all:
+ * - full folio write
+ * - write that lies in a folio that is completely beyond EOF
+ * - write that covers the folio from start to EOF or beyond it
+ *
+ * If any of these criteria are met, then zero out the unwritten parts
+ * of the folio and return true. Otherwise, return false.
+ */
+static bool netfs_skip_folio_read(struct folio *folio, loff_t pos, size_t len,
+				 bool always_fill)
+{
+	struct inode *inode = folio_inode(folio);
+	loff_t i_size = i_size_read(inode);
+	size_t offset = offset_in_folio(folio, pos);
+	size_t plen = folio_size(folio);
+
+	if (unlikely(always_fill)) {
+		if (pos - offset + len <= i_size)
+			return false; /* Page entirely before EOF */
+		folio_zero_segment(folio, 0, plen);
+		folio_mark_uptodate(folio);
+		return true;
+	}
+
+	/* Full folio write */
+	if (offset == 0 && len >= plen)
+		return true;
+
+	/* Page entirely beyond the end of the file */
+	if (pos - offset >= i_size)
+		goto zero_out;
+
+	/* Write that covers from the start of the folio to EOF or beyond */
+	if (offset == 0 && (pos + len) >= i_size)
+		goto zero_out;
+
+	return false;
+zero_out:
+	folio_zero_segments(folio, 0, offset, offset + len, plen);
+	return true;
+}
+
+/**
+ * netfs_write_begin - Helper to prepare for writing [DEPRECATED]
+ * @ctx: The netfs context
+ * @file: The file to read from
+ * @mapping: The mapping to read from
+ * @pos: File position at which the write will begin
+ * @len: The length of the write (may extend beyond the end of the folio chosen)
+ * @_folio: Where to put the resultant folio
+ * @_fsdata: Place for the netfs to store a cookie
+ *
+ * Pre-read data for a write-begin request by drawing data from the cache if
+ * possible, or the netfs if not.  Space beyond the EOF is zero-filled.
+ * Multiple I/O requests from different sources will get munged together.
+ *
+ * The calling netfs must provide a table of operations, only one of which,
+ * issue_read, is mandatory.
+ *
+ * The check_write_begin() operation can be provided to check for and flush
+ * conflicting writes once the folio is grabbed and locked.  It is passed a
+ * pointer to the fsdata cookie that gets returned to the VM to be passed to
+ * write_end.  It is permitted to sleep.  It should return 0 if the request
+ * should go ahead or it may return an error.  It may also unlock and put the
+ * folio, provided it sets ``*foliop`` to NULL, in which case a return of 0
+ * will cause the folio to be re-got and the process to be retried.
+ *
+ * The calling netfs must initialise a netfs context contiguous to the vfs
+ * inode before calling this.
+ *
+ * This is usable whether or not caching is enabled.
+ *
+ * Note that this should be considered deprecated and netfs_perform_write()
+ * used instead.
+ */
+int netfs_write_begin(struct netfs_inode *ctx,
+		      struct file *file, struct address_space *mapping,
+		      loff_t pos, unsigned int len, struct folio **_folio,
+		      void **_fsdata)
+{
+	struct netfs_io_request *rreq;
+	struct folio *folio;
+	pgoff_t index = pos >> PAGE_SHIFT;
+	int ret;
+
+retry:
+	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
+				    mapping_gfp_mask(mapping));
+	if (IS_ERR(folio))
+		return PTR_ERR(folio);
+
+	if (ctx->ops->check_write_begin) {
+		/* Allow the netfs (eg. ceph) to flush conflicts. */
+		ret = ctx->ops->check_write_begin(file, pos, len, &folio, _fsdata);
+		if (ret < 0) {
+			trace_netfs_failure(NULL, NULL, ret, netfs_fail_check_write_begin);
+			goto error;
+		}
+		if (!folio)
+			goto retry;
+	}
+
+	if (folio_test_uptodate(folio))
+		goto have_folio;
+
+	/* If the folio is beyond the EOF, we want to clear it - unless it's
+	 * within the cache granule containing the EOF, in which case we need
+	 * to preload the granule.
+	 */
+	if (!netfs_is_cache_maybe_enabled(ctx) &&
+	    netfs_skip_folio_read(folio, pos, len, false)) {
+		netfs_stat(&netfs_n_rh_write_zskip);
+		goto have_folio_no_wait;
+	}
+
+	rreq = netfs_alloc_request(mapping, file,
+				   folio_pos(folio), folio_size(folio),
+				   NETFS_READ_FOR_WRITE);
+	if (IS_ERR(rreq)) {
+		ret = PTR_ERR(rreq);
+		goto error;
+	}
+	rreq->no_unlock_folio	= folio;
+	__set_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags);
+
+	ret = netfs_begin_cache_read(rreq, ctx);
+	if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
+		goto error_put;
+
+	netfs_stat(&netfs_n_rh_write_begin);
+	trace_netfs_read(rreq, pos, len, netfs_read_trace_write_begin);
+
+	/* Set up the output buffer */
+	ret = netfs_create_singular_buffer(rreq, folio, 0);
+	if (ret < 0)
+		goto error_put;
+
+	netfs_read_to_pagecache(rreq, NULL);
+	ret = netfs_wait_for_read(rreq);
+	netfs_put_request(rreq, netfs_rreq_trace_put_return);
+	if (ret < 0)
+		goto error;
+
+have_folio:
+	ret = folio_wait_private_2_killable(folio);
+	if (ret < 0)
+		goto error;
+have_folio_no_wait:
+	*_folio = folio;
+	_leave(" = 0");
+	return 0;
+
+error_put:
+	netfs_put_failed_request(rreq);
+error:
+	if (folio) {
+		folio_unlock(folio);
+		folio_put(folio);
+	}
+	_leave(" = %d", ret);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_write_begin);
+
+/*
+ * Preload the data into a folio we're proposing to write into.
+ */
+int netfs_prefetch_for_write(struct file *file, struct folio *folio,
+			     size_t offset, size_t len)
+{
+	struct netfs_io_request *rreq;
+	struct address_space *mapping = folio->mapping;
+	struct netfs_inode *ctx = netfs_inode(mapping->host);
+	unsigned long long start = folio_pos(folio);
+	size_t flen = folio_size(folio);
+	int ret;
+
+	_enter("%zx @%llx", flen, start);
+
+	ret = -ENOMEM;
+
+	rreq = netfs_alloc_request(mapping, file, start, flen,
+				   NETFS_READ_FOR_WRITE);
+	if (IS_ERR(rreq)) {
+		ret = PTR_ERR(rreq);
+		goto error;
+	}
+
+	rreq->no_unlock_folio = folio;
+	__set_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags);
+	ret = netfs_begin_cache_read(rreq, ctx);
+	if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
+		goto error_put;
+
+	netfs_stat(&netfs_n_rh_write_begin);
+	trace_netfs_read(rreq, start, flen, netfs_read_trace_prefetch_for_write);
+
+	/* Set up the output buffer */
+	ret = netfs_create_singular_buffer(rreq, folio, NETFS_ROLLBUF_PAGECACHE_MARK);
+	if (ret < 0)
+		goto error_put;
+
+	netfs_read_to_pagecache(rreq, NULL);
+	ret = netfs_wait_for_read(rreq);
+	netfs_put_request(rreq, netfs_rreq_trace_put_return);
+	return ret < 0 ? ret : 0;
+
+error_put:
+	netfs_put_failed_request(rreq);
+error:
+	_leave(" = %d", ret);
+	return ret;
+}
+
+/**
+ * netfs_buffered_read_iter - Filesystem buffered I/O read routine
+ * @iocb: kernel I/O control block
+ * @iter: destination for the data read
+ *
+ * This is the ->read_iter() routine for all filesystems that can use the page
+ * cache directly.
+ *
+ * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall be
+ * returned when no data can be read without waiting for I/O requests to
+ * complete; it doesn't prevent readahead.
+ *
+ * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O requests
+ * shall be made for the read or for readahead.  When no data can be read,
+ * -EAGAIN shall be returned.  When readahead would be triggered, a partial,
+ * possibly empty read shall be returned.
+ *
+ * Return:
+ * * number of bytes copied, even for partial reads
+ * * negative error code (or 0 if IOCB_NOIO) if nothing was read
+ */
+ssize_t netfs_buffered_read_iter(struct kiocb *iocb, struct iov_iter *iter)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct netfs_inode *ictx = netfs_inode(inode);
+	ssize_t ret;
+
+	if (WARN_ON_ONCE((iocb->ki_flags & IOCB_DIRECT) ||
+			 test_bit(NETFS_ICTX_UNBUFFERED, &ictx->flags)))
+		return -EINVAL;
+
+	ret = netfs_start_io_read(inode);
+	if (ret == 0) {
+		ret = filemap_read(iocb, iter, 0);
+		netfs_end_io_read(inode);
+	}
+	return ret;
+}
+EXPORT_SYMBOL(netfs_buffered_read_iter);
+
+/**
+ * netfs_file_read_iter - Generic filesystem read routine
+ * @iocb: kernel I/O control block
+ * @iter: destination for the data read
+ *
+ * This is the ->read_iter() routine for all filesystems that can use the page
+ * cache directly.
+ *
+ * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall be
+ * returned when no data can be read without waiting for I/O requests to
+ * complete; it doesn't prevent readahead.
+ *
+ * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O requests
+ * shall be made for the read or for readahead.  When no data can be read,
+ * -EAGAIN shall be returned.  When readahead would be triggered, a partial,
+ * possibly empty read shall be returned.
+ *
+ * Return:
+ * * number of bytes copied, even for partial reads
+ * * negative error code (or 0 if IOCB_NOIO) if nothing was read
+ */
+ssize_t netfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
+{
+	struct netfs_inode *ictx = netfs_inode(iocb->ki_filp->f_mapping->host);
+
+	if ((iocb->ki_flags & IOCB_DIRECT) ||
+	    test_bit(NETFS_ICTX_UNBUFFERED, &ictx->flags))
+		return netfs_unbuffered_read_iter(iocb, iter);
+
+	return netfs_buffered_read_iter(iocb, iter);
+}
+EXPORT_SYMBOL(netfs_file_read_iter);
diff --git a/fs/netfs/buffered_write.c b/fs/netfs/buffered_write.c
new file mode 100644
index 000000000..2cdb68e6b
--- /dev/null
+++ b/fs/netfs/buffered_write.c
@@ -0,0 +1,616 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Network filesystem high-level buffered write support.
+ *
+ * Copyright (C) 2023 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+/*
+ * Grab a folio for writing and lock it.  Attempt to allocate as large a folio
+ * as possible to hold as much of the remaining length as possible in one go.
+ */
+static struct folio *netfs_grab_folio_for_write(struct address_space *mapping,
+						loff_t pos, size_t part)
+{
+	pgoff_t index = pos / PAGE_SIZE;
+	fgf_t fgp_flags = FGP_WRITEBEGIN;
+
+	if (mapping_large_folio_support(mapping))
+		fgp_flags |= fgf_set_order(pos % PAGE_SIZE + part);
+
+	return __filemap_get_folio(mapping, index, fgp_flags,
+				   mapping_gfp_mask(mapping));
+}
+
+/*
+ * Update i_size and estimate the update to i_blocks to reflect the additional
+ * data written into the pagecache until we can find out from the server what
+ * the values actually are.
+ */
+void netfs_update_i_size(struct netfs_inode *ctx, struct inode *inode,
+			 loff_t pos, size_t copied)
+{
+	loff_t i_size, end = pos + copied;
+	blkcnt_t add;
+	size_t gap;
+
+	if (end <= i_size_read(inode))
+		return;
+
+	if (ctx->ops->update_i_size) {
+		ctx->ops->update_i_size(inode, end);
+		return;
+	}
+
+	spin_lock(&inode->i_lock);
+
+	i_size = i_size_read(inode);
+	if (end > i_size) {
+		i_size_write(inode, end);
+#if IS_ENABLED(CONFIG_FSCACHE)
+		fscache_update_cookie(ctx->cache, NULL, &end);
+#endif
+
+		gap = SECTOR_SIZE - (i_size & (SECTOR_SIZE - 1));
+		if (copied > gap) {
+			add = DIV_ROUND_UP(copied - gap, SECTOR_SIZE);
+
+			inode->i_blocks = min_t(blkcnt_t,
+						DIV_ROUND_UP(end, SECTOR_SIZE),
+						inode->i_blocks + add);
+		}
+	}
+	spin_unlock(&inode->i_lock);
+}
+
+/**
+ * netfs_perform_write - Copy data into the pagecache.
+ * @iocb: The operation parameters
+ * @iter: The source buffer
+ * @netfs_group: Grouping for dirty folios (eg. ceph snaps).
+ *
+ * Copy data into pagecache folios attached to the inode specified by @iocb.
+ * The caller must hold appropriate inode locks.
+ *
+ * Dirty folios are tagged with a netfs_folio struct if they're not up to date
+ * to indicate the range modified.  Dirty folios may also be tagged with a
+ * netfs-specific grouping such that data from an old group gets flushed before
+ * a new one is started.
+ */
+ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter,
+			    struct netfs_group *netfs_group)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file_inode(file);
+	struct address_space *mapping = inode->i_mapping;
+	struct netfs_inode *ctx = netfs_inode(inode);
+	struct writeback_control wbc = {
+		.sync_mode	= WB_SYNC_NONE,
+		.for_sync	= true,
+		.nr_to_write	= LONG_MAX,
+		.range_start	= iocb->ki_pos,
+		.range_end	= iocb->ki_pos + iter->count,
+	};
+	struct netfs_io_request *wreq = NULL;
+	struct folio *folio = NULL, *writethrough = NULL;
+	unsigned int bdp_flags = (iocb->ki_flags & IOCB_NOWAIT) ? BDP_ASYNC : 0;
+	ssize_t written = 0, ret, ret2;
+	loff_t pos = iocb->ki_pos;
+	size_t max_chunk = mapping_max_folio_size(mapping);
+	bool maybe_trouble = false;
+
+	if (unlikely(iocb->ki_flags & (IOCB_DSYNC | IOCB_SYNC))
+	    ) {
+		wbc_attach_fdatawrite_inode(&wbc, mapping->host);
+
+		ret = filemap_write_and_wait_range(mapping, pos, pos + iter->count);
+		if (ret < 0) {
+			wbc_detach_inode(&wbc);
+			goto out;
+		}
+
+		wreq = netfs_begin_writethrough(iocb, iter->count);
+		if (IS_ERR(wreq)) {
+			wbc_detach_inode(&wbc);
+			ret = PTR_ERR(wreq);
+			wreq = NULL;
+			goto out;
+		}
+		if (!is_sync_kiocb(iocb))
+			wreq->iocb = iocb;
+		netfs_stat(&netfs_n_wh_writethrough);
+	} else {
+		netfs_stat(&netfs_n_wh_buffered_write);
+	}
+
+	do {
+		enum netfs_folio_trace trace;
+		struct netfs_folio *finfo;
+		struct netfs_group *group;
+		unsigned long long fpos;
+		size_t flen;
+		size_t offset;	/* Offset into pagecache folio */
+		size_t part;	/* Bytes to write to folio */
+		size_t copied;	/* Bytes copied from user */
+		void *priv;
+
+		offset = pos & (max_chunk - 1);
+		part = min(max_chunk - offset, iov_iter_count(iter));
+
+		/* Bring in the user pages that we will copy from _first_ lest
+		 * we hit a nasty deadlock on copying from the same page as
+		 * we're writing to, without it being marked uptodate.
+		 *
+		 * Not only is this an optimisation, but it is also required to
+		 * check that the address is actually valid, when atomic
+		 * usercopies are used below.
+		 *
+		 * We rely on the page being held onto long enough by the LRU
+		 * that we can grab it below if this causes it to be read.
+		 */
+		ret = -EFAULT;
+		if (unlikely(fault_in_iov_iter_readable(iter, part) == part))
+			break;
+
+		folio = netfs_grab_folio_for_write(mapping, pos, part);
+		if (IS_ERR(folio)) {
+			ret = PTR_ERR(folio);
+			break;
+		}
+
+		flen = folio_size(folio);
+		fpos = folio_pos(folio);
+		offset = pos - fpos;
+		part = min_t(size_t, flen - offset, part);
+
+		/* Wait for writeback to complete.  The writeback engine owns
+		 * the info in folio->private and may change it until it
+		 * removes the WB mark.
+		 */
+		if (folio_get_private(folio) &&
+		    folio_wait_writeback_killable(folio)) {
+			ret = written ? -EINTR : -ERESTARTSYS;
+			goto error_folio_unlock;
+		}
+
+		if (signal_pending(current)) {
+			ret = written ? -EINTR : -ERESTARTSYS;
+			goto error_folio_unlock;
+		}
+
+		finfo = netfs_folio_info(folio);
+		group = netfs_folio_group(folio);
+
+		/* If the requested group differs from the group set on the
+		 * page, then we need to flush out the folio if it has a group
+		 * set (ie. is non-NULL).  Note that COPY_TO_CACHE is a special
+		 * case, being a netfs annotation rather than an actual group.
+		 *
+		 * The filesystem isn't permitted to mix writes with groups and
+		 * writes without groups as the NULL group is used to indicate
+		 * that no group is set.
+		 */
+		if (unlikely(group != netfs_group) &&
+		    group != NETFS_FOLIO_COPY_TO_CACHE &&
+		    group) {
+			WARN_ON_ONCE(!netfs_group);
+			goto flush_content;
+		}
+
+		/* Decide how we should modify a folio.  We might be attempting
+		 * to do write-streaming, as we don't want to a local RMW cycle
+		 * if we can avoid it.  If we're doing local caching or content
+		 * crypto, we award that priority over avoiding RMW.  If the
+		 * file is open readably, then we let ->read_folio() fill in
+		 * the gaps.
+		 */
+		if (folio_test_uptodate(folio)) {
+			if (mapping_writably_mapped(mapping))
+				flush_dcache_folio(folio);
+			copied = copy_folio_from_iter_atomic(folio, offset, part, iter);
+			if (unlikely(copied == 0))
+				goto copy_failed;
+			trace = netfs_folio_is_uptodate;
+			goto copied_uptodate;
+		}
+
+		/* If the page is above the zero-point then we assume that the
+		 * server would just return a block of zeros or a short read if
+		 * we try to read it.
+		 */
+		if (fpos >= netfs_read_zero_point(inode)) {
+			folio_zero_segment(folio, 0, offset);
+			copied = copy_folio_from_iter_atomic(folio, offset, part, iter);
+			if (unlikely(copied == 0))
+				goto copy_failed;
+			folio_zero_segment(folio, offset + copied, flen);
+			if (finfo)
+				trace = netfs_modify_and_clear_rm_finfo;
+			else
+				trace = netfs_modify_and_clear;
+			goto mark_uptodate;
+		}
+
+		/* See if we can write a whole folio in one go. */
+		if (!maybe_trouble && offset == 0 && part >= flen) {
+			copied = copy_folio_from_iter_atomic(folio, offset, part, iter);
+			if (likely(copied == part)) {
+				if (finfo)
+					trace = netfs_whole_folio_modify_filled;
+				else
+					trace = netfs_whole_folio_modify;
+				goto mark_uptodate;
+			}
+			if (copied == 0)
+				goto copy_failed;
+			if (!finfo || copied <= finfo->dirty_offset) {
+				maybe_trouble = true;
+				iov_iter_revert(iter, copied);
+				copied = 0;
+				folio_unlock(folio);
+				goto retry;
+			}
+
+			/* We overwrote some existing dirty data, so we have to
+			 * accept the partial write.
+			 */
+			finfo->dirty_len += finfo->dirty_offset;
+			if (finfo->dirty_len == flen) {
+				trace = netfs_whole_folio_modify_filled_efault;
+				goto mark_uptodate;
+			}
+			if (copied > finfo->dirty_len)
+				finfo->dirty_len = copied;
+			finfo->dirty_offset = 0;
+			trace = netfs_whole_folio_modify_efault;
+			goto copied;
+		}
+
+		/* We don't want to do a streaming write on a file that loses
+		 * caching service temporarily because the backing store got
+		 * culled.
+		 */
+		if (netfs_is_cache_maybe_enabled(ctx)) {
+			if (finfo) {
+				netfs_stat(&netfs_n_wh_wstream_conflict);
+				goto flush_content;
+			}
+			ret = netfs_prefetch_for_write(file, folio, offset, part);
+			if (ret < 0) {
+				_debug("prefetch = %zd", ret);
+				goto error_folio_unlock;
+			}
+			/* Note that copy-to-cache may have been set. */
+
+			copied = copy_folio_from_iter_atomic(folio, offset, part, iter);
+			if (unlikely(copied == 0))
+				goto copy_failed;
+			trace = netfs_just_prefetch;
+			goto copied_uptodate;
+		}
+
+		/* Do a streaming write on a folio that has nothing in it yet. */
+		if (!finfo) {
+			ret = -EIO;
+			if (WARN_ON(folio_get_private(folio)))
+				goto error_folio_unlock;
+			copied = copy_folio_from_iter_atomic(folio, offset, part, iter);
+			if (unlikely(copied == 0))
+				goto copy_failed;
+			if (offset == 0 && copied == flen) {
+				trace = netfs_streaming_filled_page;
+				goto mark_uptodate;
+			}
+
+			finfo = kzalloc_obj(*finfo);
+			if (!finfo) {
+				iov_iter_revert(iter, copied);
+				ret = -ENOMEM;
+				goto error_folio_unlock;
+			}
+			finfo->netfs_group = netfs_get_group(netfs_group);
+			finfo->dirty_offset = offset;
+			finfo->dirty_len = copied;
+			folio_attach_private(folio, (void *)((unsigned long)finfo |
+							     NETFS_FOLIO_INFO));
+			trace = netfs_streaming_write;
+			goto copied;
+		}
+
+		/* We can continue a streaming write only if it continues on
+		 * from the previous.  If it overlaps, we must flush lest we
+		 * suffer a partial copy and disjoint dirty regions.
+		 */
+		if (offset == finfo->dirty_offset + finfo->dirty_len) {
+			copied = copy_folio_from_iter_atomic(folio, offset, part, iter);
+			if (unlikely(copied == 0))
+				goto copy_failed;
+			finfo->dirty_len += copied;
+			if (finfo->dirty_offset == 0 && finfo->dirty_len == flen) {
+				trace = netfs_streaming_cont_filled_page;
+				goto mark_uptodate;
+			}
+			trace = netfs_streaming_write_cont;
+			goto copied;
+		}
+
+		/* Incompatible write; flush the folio and try again. */
+	flush_content:
+		trace_netfs_folio(folio, netfs_flush_content);
+		folio_unlock(folio);
+		folio_put(folio);
+		ret = filemap_write_and_wait_range(mapping, fpos, fpos + flen - 1);
+		if (ret < 0)
+			goto out;
+		continue;
+
+		/* Mark a folio as being up to data when we've filled it
+		 * completely.  If the folio has a group attached, then it must
+		 * be the same group, otherwise we should have flushed it out
+		 * above.  We have to get rid of the netfs_folio struct if
+		 * there was one.
+		 */
+	mark_uptodate:
+		folio_mark_uptodate(folio);
+
+	copied_uptodate:
+		priv = folio_get_private(folio);
+		if (likely(priv == netfs_group)) {
+			/* Already set correctly; no change required. */
+		} else if (priv == NETFS_FOLIO_COPY_TO_CACHE) {
+			if (!netfs_group)
+				folio_detach_private(folio);
+			else
+				folio_change_private(folio, netfs_get_group(netfs_group));
+		} else if (!priv) {
+			folio_attach_private(folio, netfs_get_group(netfs_group));
+		} else {
+			WARN_ON_ONCE(!finfo);
+			if (netfs_group)
+				/* finfo->netfs_group has a ref */
+				folio_change_private(folio, netfs_group);
+			else
+				folio_detach_private(folio);
+			kfree(finfo);
+		}
+
+	copied:
+		trace_netfs_folio(folio, trace);
+		flush_dcache_folio(folio);
+
+		/* Update the inode size if we moved the EOF marker */
+		netfs_update_i_size(ctx, inode, pos, copied);
+		pos += copied;
+		written += copied;
+
+		if (likely(!wreq)) {
+			folio_mark_dirty(folio);
+			folio_unlock(folio);
+		} else {
+			netfs_advance_writethrough(wreq, &wbc, folio, copied,
+						   offset + copied == flen,
+						   &writethrough);
+			/* Folio unlocked */
+		}
+	retry:
+		folio_put(folio);
+		folio = NULL;
+
+		ret = balance_dirty_pages_ratelimited_flags(mapping, bdp_flags);
+		if (unlikely(ret < 0))
+			break;
+
+		cond_resched();
+	} while (iov_iter_count(iter));
+
+out:
+	if (likely(written)) {
+		/* Set indication that ctime and mtime got updated in case
+		 * close is deferred.
+		 */
+		set_bit(NETFS_ICTX_MODIFIED_ATTR, &ctx->flags);
+		if (unlikely(ctx->ops->post_modify))
+			ctx->ops->post_modify(inode);
+	}
+
+	if (unlikely(wreq)) {
+		ret2 = netfs_end_writethrough(wreq, &wbc, writethrough);
+		wbc_detach_inode(&wbc);
+		if (ret2 == -EIOCBQUEUED)
+			return ret2;
+		if (ret == 0 && ret2 < 0)
+			ret = ret2;
+	}
+
+	iocb->ki_pos += written;
+	_leave(" = %zd [%zd]", written, ret);
+	return written ? written : ret;
+
+copy_failed:
+	ret = -EFAULT;
+error_folio_unlock:
+	folio_unlock(folio);
+	folio_put(folio);
+	goto out;
+}
+EXPORT_SYMBOL(netfs_perform_write);
+
+/**
+ * netfs_buffered_write_iter_locked - write data to a file
+ * @iocb:	IO state structure (file, offset, etc.)
+ * @from:	iov_iter with data to write
+ * @netfs_group: Grouping for dirty folios (eg. ceph snaps).
+ *
+ * This function does all the work needed for actually writing data to a
+ * file. It does all basic checks, removes SUID from the file, updates
+ * modification times and calls proper subroutines depending on whether we
+ * do direct IO or a standard buffered write.
+ *
+ * The caller must hold appropriate locks around this function and have called
+ * generic_write_checks() already.  The caller is also responsible for doing
+ * any necessary syncing afterwards.
+ *
+ * This function does *not* take care of syncing data in case of O_SYNC write.
+ * A caller has to handle it. This is mainly due to the fact that we want to
+ * avoid syncing under i_rwsem.
+ *
+ * Return:
+ * * number of bytes written, even for truncated writes
+ * * negative error code if no data has been written at all
+ */
+ssize_t netfs_buffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *from,
+					 struct netfs_group *netfs_group)
+{
+	struct file *file = iocb->ki_filp;
+	ssize_t ret;
+
+	trace_netfs_write_iter(iocb, from);
+
+	ret = file_remove_privs(file);
+	if (ret)
+		return ret;
+
+	ret = file_update_time(file);
+	if (ret)
+		return ret;
+
+	return netfs_perform_write(iocb, from, netfs_group);
+}
+EXPORT_SYMBOL(netfs_buffered_write_iter_locked);
+
+/**
+ * netfs_file_write_iter - write data to a file
+ * @iocb: IO state structure
+ * @from: iov_iter with data to write
+ *
+ * Perform a write to a file, writing into the pagecache if possible and doing
+ * an unbuffered write instead if not.
+ *
+ * Return:
+ * * Negative error code if no data has been written at all of
+ *   vfs_fsync_range() failed for a synchronous write
+ * * Number of bytes written, even for truncated writes
+ */
+ssize_t netfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	struct netfs_inode *ictx = netfs_inode(inode);
+	ssize_t ret;
+
+	_enter("%llx,%zx,%llx", iocb->ki_pos, iov_iter_count(from), i_size_read(inode));
+
+	if (!iov_iter_count(from))
+		return 0;
+
+	if ((iocb->ki_flags & IOCB_DIRECT) ||
+	    test_bit(NETFS_ICTX_UNBUFFERED, &ictx->flags))
+		return netfs_unbuffered_write_iter(iocb, from);
+
+	ret = netfs_start_io_write(inode);
+	if (ret < 0)
+		return ret;
+
+	ret = generic_write_checks(iocb, from);
+	if (ret > 0)
+		ret = netfs_buffered_write_iter_locked(iocb, from, NULL);
+	netfs_end_io_write(inode);
+	if (ret > 0)
+		ret = generic_write_sync(iocb, ret);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_file_write_iter);
+
+/*
+ * Notification that a previously read-only page is about to become writable.
+ * The caller indicates the precise page that needs to be written to, but
+ * we only track group on a per-folio basis, so we block more often than
+ * we might otherwise.
+ */
+vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group)
+{
+	struct netfs_group *group;
+	struct folio *folio = page_folio(vmf->page);
+	struct file *file = vmf->vma->vm_file;
+	struct address_space *mapping = file->f_mapping;
+	struct inode *inode = file_inode(file);
+	struct netfs_inode *ictx = netfs_inode(inode);
+	vm_fault_t ret = VM_FAULT_NOPAGE;
+	void *priv;
+	int err;
+
+	_enter("%lx", folio->index);
+
+	sb_start_pagefault(inode->i_sb);
+
+	if (folio_lock_killable(folio) < 0)
+		goto out;
+	if (folio->mapping != mapping)
+		goto unlock;
+	if (folio_wait_writeback_killable(folio) < 0)
+		goto unlock;
+
+	/* Can we see a streaming write here? */
+	if (WARN_ON(!folio_test_uptodate(folio))) {
+		ret = VM_FAULT_SIGBUS;
+		goto unlock;
+	}
+
+	group = netfs_folio_group(folio);
+	if (group &&
+	    group != netfs_group &&
+	    group != NETFS_FOLIO_COPY_TO_CACHE) {
+		folio_unlock(folio);
+		err = filemap_fdatawrite_range(mapping,
+					       folio_pos(folio),
+					       folio_next_pos(folio));
+		switch (err) {
+		case 0:
+			ret = VM_FAULT_RETRY;
+			goto out;
+		case -ENOMEM:
+			ret = VM_FAULT_OOM;
+			goto out;
+		default:
+			ret = VM_FAULT_SIGBUS;
+			goto out;
+		}
+	}
+
+	if (folio_test_dirty(folio))
+		trace_netfs_folio(folio, netfs_folio_trace_mkwrite_plus);
+	else
+		trace_netfs_folio(folio, netfs_folio_trace_mkwrite);
+
+	priv = folio_get_private(folio);
+	if (priv != netfs_group) {
+		if (!netfs_group && priv == NETFS_FOLIO_COPY_TO_CACHE)
+			folio_detach_private(folio);
+		else if (netfs_group && priv == NETFS_FOLIO_COPY_TO_CACHE)
+			folio_change_private(folio, netfs_get_group(netfs_group));
+		else if (netfs_group && !priv)
+			folio_attach_private(folio, netfs_get_group(netfs_group));
+		else
+			WARN_ON_ONCE(1);
+	}
+
+	file_update_time(file);
+	set_bit(NETFS_ICTX_MODIFIED_ATTR, &ictx->flags);
+	if (ictx->ops->post_modify)
+		ictx->ops->post_modify(inode);
+	ret = VM_FAULT_LOCKED;
+out:
+	sb_end_pagefault(inode->i_sb);
+	return ret;
+unlock:
+	folio_unlock(folio);
+	goto out;
+}
+EXPORT_SYMBOL(netfs_page_mkwrite);
diff --git a/fs/netfs/direct_read.c b/fs/netfs/direct_read.c
new file mode 100644
index 000000000..6a8fb0d55
--- /dev/null
+++ b/fs/netfs/direct_read.c
@@ -0,0 +1,255 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Direct I/O support.
+ *
+ * Copyright (C) 2023 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include <linux/uio.h>
+#include <linux/sched/mm.h>
+#include <linux/task_io_accounting_ops.h>
+#include <linux/netfs.h>
+#include "internal.h"
+
+static void netfs_prepare_dio_read_iterator(struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_request *rreq = subreq->rreq;
+	size_t rsize;
+
+	rsize = umin(subreq->len, rreq->io_streams[0].sreq_max_len);
+	subreq->len = rsize;
+
+	if (unlikely(rreq->io_streams[0].sreq_max_segs)) {
+		size_t limit = netfs_limit_iter(&rreq->buffer.iter, 0, rsize,
+						rreq->io_streams[0].sreq_max_segs);
+
+		if (limit < rsize) {
+			subreq->len = limit;
+			trace_netfs_sreq(subreq, netfs_sreq_trace_limited);
+		}
+	}
+
+	trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
+
+	subreq->io_iter	= rreq->buffer.iter;
+	iov_iter_truncate(&subreq->io_iter, subreq->len);
+	iov_iter_advance(&rreq->buffer.iter, subreq->len);
+}
+
+/*
+ * Perform a read to a buffer from the server, slicing up the region to be read
+ * according to the network rsize.
+ */
+static void netfs_dispatch_unbuffered_reads(struct netfs_io_request *rreq)
+{
+	unsigned long long start = rreq->start;
+	ssize_t size = rreq->len;
+	int ret;
+
+	do {
+		struct netfs_io_subrequest *subreq;
+		ssize_t slice;
+
+		subreq = netfs_alloc_subrequest(rreq);
+		if (!subreq) {
+			/* Stash the error in the request if there's not
+			 * already an error set.
+			 */
+			cmpxchg(&rreq->error, 0, -ENOMEM);
+			break;
+		}
+
+		subreq->source	= NETFS_DOWNLOAD_FROM_SERVER;
+		subreq->start	= start;
+		subreq->len	= size;
+
+		netfs_queue_read(rreq, subreq);
+
+		netfs_stat(&netfs_n_rh_download);
+		if (rreq->netfs_ops->prepare_read) {
+			ret = rreq->netfs_ops->prepare_read(subreq);
+			if (ret < 0) {
+				netfs_cancel_read(subreq, ret);
+				break;
+			}
+		}
+
+		netfs_prepare_dio_read_iterator(subreq);
+		slice = subreq->len;
+		size -= slice;
+		start += slice;
+		rreq->submitted += slice;
+		if (size <= 0) {
+			smp_wmb(); /* Write lists before ALL_QUEUED. */
+			set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+		}
+
+		rreq->netfs_ops->issue_read(subreq);
+
+		if (test_bit(NETFS_RREQ_PAUSE, &rreq->flags))
+			netfs_wait_for_paused_read(rreq);
+		if (test_bit(NETFS_RREQ_FAILED, &rreq->flags))
+			break;
+		cond_resched();
+	} while (size > 0);
+
+	if (unlikely(size > 0)) {
+		smp_wmb(); /* Write lists before ALL_QUEUED. */
+		set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+		netfs_wake_collector(rreq);
+	}
+}
+
+/*
+ * Perform a read to an application buffer, bypassing the pagecache and the
+ * local disk cache.
+ */
+static ssize_t netfs_unbuffered_read(struct netfs_io_request *rreq, bool sync)
+{
+	ssize_t ret;
+
+	_enter("R=%x %llx-%llx",
+	       rreq->debug_id, rreq->start, rreq->start + rreq->len - 1);
+
+	if (rreq->len == 0) {
+		pr_err("Zero-sized read [R=%x]\n", rreq->debug_id);
+		netfs_put_request(rreq, netfs_rreq_trace_put_discard);
+		return -EIO;
+	}
+
+	// TODO: Use bounce buffer if requested
+
+	inode_dio_begin(rreq->inode);
+	netfs_dispatch_unbuffered_reads(rreq);
+
+	/* The collector will get run, even if we don't manage to submit any
+	 * subreqs, so we shouldn't call inode_dio_end() here.
+	 */
+
+	if (sync)
+		ret = netfs_wait_for_read(rreq);
+	else
+		ret = -EIOCBQUEUED;
+
+	_leave(" = %zd", ret);
+	return ret;
+}
+
+/**
+ * netfs_unbuffered_read_iter_locked - Perform an unbuffered or direct I/O read
+ * @iocb: The I/O control descriptor describing the read
+ * @iter: The output buffer (also specifies read length)
+ *
+ * Perform an unbuffered I/O or direct I/O from the file in @iocb to the
+ * output buffer.  No use is made of the pagecache.
+ *
+ * The caller must hold any appropriate locks.
+ */
+ssize_t netfs_unbuffered_read_iter_locked(struct kiocb *iocb, struct iov_iter *iter)
+{
+	struct netfs_io_request *rreq;
+	ssize_t ret;
+	size_t orig_count = iov_iter_count(iter);
+	bool sync = is_sync_kiocb(iocb);
+
+	_enter("");
+
+	if (!orig_count)
+		return 0; /* Don't update atime */
+
+	ret = kiocb_write_and_wait(iocb, orig_count);
+	if (ret < 0)
+		return ret;
+	file_accessed(iocb->ki_filp);
+
+	rreq = netfs_alloc_request(iocb->ki_filp->f_mapping, iocb->ki_filp,
+				   iocb->ki_pos, orig_count,
+				   iocb->ki_flags & IOCB_DIRECT ?
+				   NETFS_DIO_READ : NETFS_UNBUFFERED_READ);
+	if (IS_ERR(rreq))
+		return PTR_ERR(rreq);
+
+	netfs_stat(&netfs_n_rh_dio_read);
+	trace_netfs_read(rreq, rreq->start, rreq->len, netfs_read_trace_dio_read);
+
+	/* If this is an async op, we have to keep track of the destination
+	 * buffer for ourselves as the caller's iterator will be trashed when
+	 * we return.
+	 *
+	 * In such a case, extract an iterator to represent as much of the the
+	 * output buffer as we can manage.  Note that the extraction might not
+	 * be able to allocate a sufficiently large bvec array and may shorten
+	 * the request.
+	 */
+	if (user_backed_iter(iter)) {
+		ret = netfs_extract_user_iter(iter, rreq->len, &rreq->buffer.iter, 0);
+		if (ret < 0)
+			goto error_put;
+		rreq->direct_bv = (struct bio_vec *)rreq->buffer.iter.bvec;
+		rreq->direct_bv_count = ret;
+		rreq->direct_bv_unpin = iov_iter_extract_will_pin(iter);
+		rreq->len = iov_iter_count(&rreq->buffer.iter);
+	} else {
+		rreq->buffer.iter = *iter;
+		rreq->len = orig_count;
+		rreq->direct_bv_unpin = false;
+		iov_iter_advance(iter, orig_count);
+	}
+
+	// TODO: Set up bounce buffer if needed
+
+	if (!sync) {
+		rreq->iocb = iocb;
+		__set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags);
+	}
+
+	ret = netfs_unbuffered_read(rreq, sync);
+	if (ret < 0)
+		goto out; /* May be -EIOCBQUEUED */
+	if (sync) {
+		// TODO: Copy from bounce buffer
+		iocb->ki_pos += rreq->transferred;
+		ret = rreq->transferred;
+	}
+
+out:
+	netfs_put_request(rreq, netfs_rreq_trace_put_return);
+	if (ret > 0)
+		orig_count -= ret;
+	return ret;
+
+error_put:
+	netfs_put_failed_request(rreq);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_unbuffered_read_iter_locked);
+
+/**
+ * netfs_unbuffered_read_iter - Perform an unbuffered or direct I/O read
+ * @iocb: The I/O control descriptor describing the read
+ * @iter: The output buffer (also specifies read length)
+ *
+ * Perform an unbuffered I/O or direct I/O from the file in @iocb to the
+ * output buffer.  No use is made of the pagecache.
+ */
+ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	ssize_t ret;
+
+	if (!iter->count)
+		return 0; /* Don't update atime */
+
+	ret = netfs_start_io_direct(inode);
+	if (ret == 0) {
+		ret = netfs_unbuffered_read_iter_locked(iocb, iter);
+		netfs_end_io_direct(inode);
+	}
+	return ret;
+}
+EXPORT_SYMBOL(netfs_unbuffered_read_iter);
diff --git a/fs/netfs/direct_write.c b/fs/netfs/direct_write.c
new file mode 100644
index 000000000..c16fbad28
--- /dev/null
+++ b/fs/netfs/direct_write.c
@@ -0,0 +1,387 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Unbuffered and direct write support.
+ *
+ * Copyright (C) 2023 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/uio.h>
+#include "internal.h"
+
+/*
+ * Perform the cleanup rituals after an unbuffered write is complete.
+ */
+static void netfs_unbuffered_write_done(struct netfs_io_request *wreq)
+{
+	struct netfs_inode *ictx = netfs_inode(wreq->inode);
+
+	_enter("R=%x", wreq->debug_id);
+
+	/* Okay, declare that all I/O is complete. */
+	trace_netfs_rreq(wreq, netfs_rreq_trace_write_done);
+
+	if (!wreq->error)
+		netfs_update_i_size(ictx, &ictx->inode, wreq->start, wreq->transferred);
+
+	if (wreq->origin == NETFS_DIO_WRITE &&
+	    wreq->mapping->nrpages) {
+		/* mmap may have got underfoot and we may now have folios
+		 * locally covering the region we just wrote.  Attempt to
+		 * discard the folios, but leave in place any modified locally.
+		 * ->write_iter() is prevented from interfering by the DIO
+		 * counter.
+		 */
+		pgoff_t first = wreq->start >> PAGE_SHIFT;
+		pgoff_t last = (wreq->start + wreq->transferred - 1) >> PAGE_SHIFT;
+
+		invalidate_inode_pages2_range(wreq->mapping, first, last);
+	}
+
+	if (wreq->origin == NETFS_DIO_WRITE)
+		inode_dio_end(wreq->inode);
+
+	_debug("finished");
+	netfs_wake_rreq_flag(wreq, NETFS_RREQ_IN_PROGRESS, netfs_rreq_trace_wake_ip);
+	/* As we cleared NETFS_RREQ_IN_PROGRESS, we acquired its ref. */
+
+	if (wreq->iocb) {
+		size_t written = umin(wreq->transferred, wreq->len);
+
+		wreq->iocb->ki_pos += written;
+		if (wreq->iocb->ki_complete) {
+			trace_netfs_rreq(wreq, netfs_rreq_trace_ki_complete);
+			wreq->iocb->ki_complete(wreq->iocb, wreq->error ?: written);
+		}
+		wreq->iocb = VFS_PTR_POISON;
+	}
+
+	netfs_clear_subrequests(wreq);
+}
+
+/*
+ * Collect the subrequest results of unbuffered write subrequests.
+ */
+static void netfs_unbuffered_write_collect(struct netfs_io_request *wreq,
+					   struct netfs_io_stream *stream,
+					   struct netfs_io_subrequest *subreq)
+{
+	trace_netfs_collect_sreq(wreq, subreq);
+
+	spin_lock(&wreq->lock);
+	list_del_init(&subreq->rreq_link);
+	spin_unlock(&wreq->lock);
+
+	wreq->transferred += subreq->transferred;
+	iov_iter_advance(&wreq->buffer.iter, subreq->transferred);
+
+	stream->collected_to = subreq->start + subreq->transferred;
+	wreq->collected_to = stream->collected_to;
+	netfs_put_subrequest(subreq, netfs_sreq_trace_put_done);
+
+	trace_netfs_collect_stream(wreq, stream);
+	trace_netfs_collect_state(wreq, wreq->collected_to, 0);
+}
+
+/*
+ * Write data to the server without going through the pagecache and without
+ * writing it to the local cache.  We dispatch the subrequests serially and
+ * wait for each to complete before dispatching the next, lest we leave a gap
+ * in the data written due to a failure such as ENOSPC.  We could, however
+ * attempt to do preparation such as content encryption for the next subreq
+ * whilst the current is in progress.
+ */
+static int netfs_unbuffered_write(struct netfs_io_request *wreq)
+{
+	struct netfs_io_subrequest *subreq = NULL;
+	struct netfs_io_stream *stream = &wreq->io_streams[0];
+	int ret;
+
+	_enter("%llx", wreq->len);
+
+	if (wreq->origin == NETFS_DIO_WRITE)
+		inode_dio_begin(wreq->inode);
+
+	stream->collected_to = wreq->start;
+
+	for (;;) {
+		bool retry = false;
+
+		if (!subreq) {
+			netfs_prepare_write(wreq, stream, wreq->start + wreq->transferred);
+			subreq = stream->construct;
+			stream->construct = NULL;
+		}
+
+		/* Check if (re-)preparation failed. */
+		if (unlikely(test_bit(NETFS_SREQ_FAILED, &subreq->flags))) {
+			netfs_write_subrequest_terminated(subreq, subreq->error);
+			wreq->error = subreq->error;
+			break;
+		}
+
+		iov_iter_truncate(&subreq->io_iter, wreq->len - wreq->transferred);
+		if (!iov_iter_count(&subreq->io_iter))
+			break;
+
+		subreq->len = netfs_limit_iter(&subreq->io_iter, 0,
+					       stream->sreq_max_len,
+					       stream->sreq_max_segs);
+		iov_iter_truncate(&subreq->io_iter, subreq->len);
+		stream->submit_extendable_to = subreq->len;
+
+		trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
+		stream->issue_write(subreq);
+
+		/* Async, need to wait. */
+		netfs_wait_for_in_progress_stream(wreq, stream);
+
+		if (test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
+			retry = true;
+		} else if (test_bit(NETFS_SREQ_FAILED, &subreq->flags)) {
+			ret = subreq->error;
+			wreq->error = ret;
+			netfs_see_subrequest(subreq, netfs_sreq_trace_see_failed);
+			subreq = NULL;
+			break;
+		}
+		ret = 0;
+
+		if (!retry) {
+			netfs_unbuffered_write_collect(wreq, stream, subreq);
+			subreq = NULL;
+			if (wreq->transferred >= wreq->len)
+				break;
+			if (!wreq->iocb && signal_pending(current)) {
+				ret = wreq->transferred ? -EINTR : -ERESTARTSYS;
+				trace_netfs_rreq(wreq, netfs_rreq_trace_intr);
+				break;
+			}
+			continue;
+		}
+
+		/* We need to retry the last subrequest, so first reset the
+		 * iterator, taking into account what, if anything, we managed
+		 * to transfer.
+		 */
+		subreq->error = -EAGAIN;
+		trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
+		if (subreq->transferred > 0) {
+			iov_iter_advance(&wreq->buffer.iter, subreq->transferred);
+			wreq->transferred += subreq->transferred;
+		}
+
+		if (stream->source == NETFS_UPLOAD_TO_SERVER &&
+		    wreq->netfs_ops->retry_request)
+			wreq->netfs_ops->retry_request(wreq, stream);
+
+		__clear_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+		__clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+		__clear_bit(NETFS_SREQ_BOUNDARY, &subreq->flags);
+		__clear_bit(NETFS_SREQ_FAILED, &subreq->flags);
+		subreq->io_iter		= wreq->buffer.iter;
+		subreq->start		= wreq->start + wreq->transferred;
+		subreq->len		= wreq->len   - wreq->transferred;
+		subreq->transferred	= 0;
+		subreq->retry_count	+= 1;
+		stream->sreq_max_len	= UINT_MAX;
+		stream->sreq_max_segs	= INT_MAX;
+
+		netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
+
+		if (stream->prepare_write)
+			stream->prepare_write(subreq);
+		__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+		netfs_stat(&netfs_n_wh_retry_write_subreq);
+	}
+
+	netfs_unbuffered_write_done(wreq);
+	_leave(" = %d", ret);
+	return ret;
+}
+
+static void netfs_unbuffered_write_async(struct work_struct *work)
+{
+	struct netfs_io_request *wreq = container_of(work, struct netfs_io_request, work);
+
+	netfs_unbuffered_write(wreq);
+	netfs_put_request(wreq, netfs_rreq_trace_put_complete);
+}
+
+/*
+ * Perform an unbuffered write where we may have to do an RMW operation on an
+ * encrypted file.  This can also be used for direct I/O writes.
+ */
+ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *iter,
+						  struct netfs_group *netfs_group)
+{
+	struct netfs_io_request *wreq;
+	unsigned long long start = iocb->ki_pos;
+	unsigned long long end = start + iov_iter_count(iter);
+	ssize_t ret, n;
+	size_t len = iov_iter_count(iter);
+	bool async = !is_sync_kiocb(iocb);
+
+	_enter("");
+
+	/* We're going to need a bounce buffer if what we transmit is going to
+	 * be different in some way to the source buffer, e.g. because it gets
+	 * encrypted/compressed or because it needs expanding to a block size.
+	 */
+	// TODO
+
+	_debug("uw %llx-%llx", start, end);
+
+	wreq = netfs_create_write_req(iocb->ki_filp->f_mapping, iocb->ki_filp, start,
+				      iocb->ki_flags & IOCB_DIRECT ?
+				      NETFS_DIO_WRITE : NETFS_UNBUFFERED_WRITE);
+	if (IS_ERR(wreq))
+		return PTR_ERR(wreq);
+
+	wreq->io_streams[0].avail = true;
+	trace_netfs_write(wreq, (iocb->ki_flags & IOCB_DIRECT ?
+				 netfs_write_trace_dio_write :
+				 netfs_write_trace_unbuffered_write));
+
+	{
+		/* If this is an async op and we're not using a bounce buffer,
+		 * we have to save the source buffer as the iterator is only
+		 * good until we return.  In such a case, extract an iterator
+		 * to represent as much of the the output buffer as we can
+		 * manage.  Note that the extraction might not be able to
+		 * allocate a sufficiently large bvec array and may shorten the
+		 * request.
+		 */
+		if (user_backed_iter(iter)) {
+			n = netfs_extract_user_iter(iter, len, &wreq->buffer.iter, 0);
+			if (n < 0) {
+				ret = n;
+				goto error_put;
+			}
+			wreq->direct_bv = (struct bio_vec *)wreq->buffer.iter.bvec;
+			wreq->direct_bv_count = n;
+			wreq->direct_bv_unpin = iov_iter_extract_will_pin(iter);
+		} else {
+			/* If this is a kernel-generated async DIO request,
+			 * assume that any resources the iterator points to
+			 * (eg. a bio_vec array) will persist till the end of
+			 * the op.
+			 */
+			wreq->buffer.iter = *iter;
+		}
+
+		wreq->len = iov_iter_count(&wreq->buffer.iter);
+	}
+
+	__set_bit(NETFS_RREQ_USE_IO_ITER, &wreq->flags);
+
+	/* Copy the data into the bounce buffer and encrypt it. */
+	// TODO
+
+	/* Dispatch the write. */
+	__set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags);
+
+	if (async) {
+		INIT_WORK(&wreq->work, netfs_unbuffered_write_async);
+		wreq->iocb = iocb;
+		queue_work(system_dfl_wq, &wreq->work);
+		ret = -EIOCBQUEUED;
+	} else {
+		ret = netfs_unbuffered_write(wreq);
+		if (ret < 0) {
+			_debug("begin = %zd", ret);
+		} else {
+			iocb->ki_pos += wreq->transferred;
+			ret = wreq->transferred ?: wreq->error;
+		}
+
+		netfs_put_request(wreq, netfs_rreq_trace_put_complete);
+	}
+
+	netfs_put_request(wreq, netfs_rreq_trace_put_return);
+	return ret;
+
+error_put:
+	netfs_put_failed_request(wreq);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_unbuffered_write_iter_locked);
+
+/**
+ * netfs_unbuffered_write_iter - Unbuffered write to a file
+ * @iocb: IO state structure
+ * @from: iov_iter with data to write
+ *
+ * Do an unbuffered write to a file, writing the data directly to the server
+ * and not lodging the data in the pagecache.
+ *
+ * Return:
+ * * Negative error code if no data has been written at all of
+ *   vfs_fsync_range() failed for a synchronous write
+ * * Number of bytes written, even for truncated writes
+ */
+ssize_t netfs_unbuffered_write_iter(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct address_space *mapping = file->f_mapping;
+	struct inode *inode = mapping->host;
+	struct netfs_inode *ictx = netfs_inode(inode);
+	ssize_t ret;
+	loff_t pos = iocb->ki_pos;
+	unsigned long long end = pos + iov_iter_count(from) - 1;
+
+	_enter("%llx,%zx,%llx", pos, iov_iter_count(from), i_size_read(inode));
+
+	if (!iov_iter_count(from))
+		return 0;
+
+	trace_netfs_write_iter(iocb, from);
+	netfs_stat(&netfs_n_wh_dio_write);
+
+	ret = netfs_start_io_direct(inode);
+	if (ret < 0)
+		return ret;
+	ret = generic_write_checks(iocb, from);
+	if (ret <= 0)
+		goto out;
+	ret = file_remove_privs(file);
+	if (ret < 0)
+		goto out;
+	ret = file_update_time(file);
+	if (ret < 0)
+		goto out;
+	if (iocb->ki_flags & IOCB_NOWAIT) {
+		/* We could block if there are any pages in the range. */
+		ret = -EAGAIN;
+		if (filemap_range_has_page(mapping, pos, end))
+			if (filemap_invalidate_inode(inode, true, pos, end))
+				goto out;
+	} else {
+		ret = filemap_write_and_wait_range(mapping, pos, end);
+		if (ret < 0)
+			goto out;
+	}
+
+	/*
+	 * After a write we want buffered reads to be sure to go to disk to get
+	 * the new data.  We invalidate clean cached page from the region we're
+	 * about to write.  We do this *before* the write so that we can return
+	 * without clobbering -EIOCBQUEUED from ->direct_IO().
+	 */
+	ret = filemap_invalidate_inode(inode, true, pos, end);
+	if (ret < 0)
+		goto out;
+	end = iocb->ki_pos + iov_iter_count(from);
+	spin_lock(&inode->i_lock);
+	if (end > ictx->_zero_point)
+		netfs_write_zero_point(inode, end);
+	spin_unlock(&inode->i_lock);
+
+	fscache_invalidate(netfs_i_cookie(ictx), NULL, i_size_read(inode),
+			   FSCACHE_INVAL_DIO_WRITE);
+	ret = netfs_unbuffered_write_iter_locked(iocb, from, NULL);
+out:
+	netfs_end_io_direct(inode);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_unbuffered_write_iter);
diff --git a/fs/netfs/fscache_cache.c b/fs/netfs/fscache_cache.c
new file mode 100644
index 000000000..15b54d133
--- /dev/null
+++ b/fs/netfs/fscache_cache.c
@@ -0,0 +1,429 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* FS-Cache cache handling
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define FSCACHE_DEBUG_LEVEL CACHE
+#include <linux/export.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+static LIST_HEAD(fscache_caches);
+DECLARE_RWSEM(fscache_addremove_sem);
+EXPORT_SYMBOL(fscache_addremove_sem);
+DECLARE_WAIT_QUEUE_HEAD(fscache_clearance_waiters);
+EXPORT_SYMBOL(fscache_clearance_waiters);
+
+static atomic_t fscache_cache_debug_id;
+
+/*
+ * Allocate a cache cookie.
+ */
+static struct fscache_cache *fscache_alloc_cache(const char *name)
+{
+	struct fscache_cache *cache;
+
+	cache = kzalloc_obj(*cache);
+	if (cache) {
+		if (name) {
+			cache->name = kstrdup(name, GFP_KERNEL);
+			if (!cache->name) {
+				kfree(cache);
+				return NULL;
+			}
+		}
+		refcount_set(&cache->ref, 1);
+		INIT_LIST_HEAD(&cache->cache_link);
+		cache->debug_id = atomic_inc_return(&fscache_cache_debug_id);
+	}
+	return cache;
+}
+
+static bool fscache_get_cache_maybe(struct fscache_cache *cache,
+				    enum fscache_cache_trace where)
+{
+	bool success;
+	int ref;
+
+	success = __refcount_inc_not_zero(&cache->ref, &ref);
+	if (success)
+		trace_fscache_cache(cache->debug_id, ref + 1, where);
+	return success;
+}
+
+/*
+ * Look up a cache cookie.
+ */
+struct fscache_cache *fscache_lookup_cache(const char *name, bool is_cache)
+{
+	struct fscache_cache *candidate, *cache, *unnamed = NULL;
+
+	/* firstly check for the existence of the cache under read lock */
+	down_read(&fscache_addremove_sem);
+
+	list_for_each_entry(cache, &fscache_caches, cache_link) {
+		if (cache->name && name && strcmp(cache->name, name) == 0 &&
+		    fscache_get_cache_maybe(cache, fscache_cache_get_acquire))
+			goto got_cache_r;
+		if (!cache->name && !name &&
+		    fscache_get_cache_maybe(cache, fscache_cache_get_acquire))
+			goto got_cache_r;
+	}
+
+	if (!name) {
+		list_for_each_entry(cache, &fscache_caches, cache_link) {
+			if (cache->name &&
+			    fscache_get_cache_maybe(cache, fscache_cache_get_acquire))
+				goto got_cache_r;
+		}
+	}
+
+	up_read(&fscache_addremove_sem);
+
+	/* the cache does not exist - create a candidate */
+	candidate = fscache_alloc_cache(name);
+	if (!candidate)
+		return ERR_PTR(-ENOMEM);
+
+	/* write lock, search again and add if still not present */
+	down_write(&fscache_addremove_sem);
+
+	list_for_each_entry(cache, &fscache_caches, cache_link) {
+		if (cache->name && name && strcmp(cache->name, name) == 0 &&
+		    fscache_get_cache_maybe(cache, fscache_cache_get_acquire))
+			goto got_cache_w;
+		if (!cache->name) {
+			unnamed = cache;
+			if (!name &&
+			    fscache_get_cache_maybe(cache, fscache_cache_get_acquire))
+				goto got_cache_w;
+		}
+	}
+
+	if (unnamed && is_cache &&
+	    fscache_get_cache_maybe(unnamed, fscache_cache_get_acquire))
+		goto use_unnamed_cache;
+
+	if (!name) {
+		list_for_each_entry(cache, &fscache_caches, cache_link) {
+			if (cache->name &&
+			    fscache_get_cache_maybe(cache, fscache_cache_get_acquire))
+				goto got_cache_w;
+		}
+	}
+
+	list_add_tail(&candidate->cache_link, &fscache_caches);
+	trace_fscache_cache(candidate->debug_id,
+			    refcount_read(&candidate->ref),
+			    fscache_cache_new_acquire);
+	up_write(&fscache_addremove_sem);
+	return candidate;
+
+got_cache_r:
+	up_read(&fscache_addremove_sem);
+	return cache;
+use_unnamed_cache:
+	cache = unnamed;
+	cache->name = candidate->name;
+	candidate->name = NULL;
+got_cache_w:
+	up_write(&fscache_addremove_sem);
+	kfree(candidate->name);
+	kfree(candidate);
+	return cache;
+}
+
+/**
+ * fscache_acquire_cache - Acquire a cache-level cookie.
+ * @name: The name of the cache.
+ *
+ * Get a cookie to represent an actual cache.  If a name is given and there is
+ * a nameless cache record available, this will acquire that and set its name,
+ * directing all the volumes using it to this cache.
+ *
+ * The cache will be switched over to the preparing state if not currently in
+ * use, otherwise -EBUSY will be returned.
+ */
+struct fscache_cache *fscache_acquire_cache(const char *name)
+{
+	struct fscache_cache *cache;
+
+	ASSERT(name);
+	cache = fscache_lookup_cache(name, true);
+	if (IS_ERR(cache))
+		return cache;
+
+	if (!fscache_set_cache_state_maybe(cache,
+					   FSCACHE_CACHE_IS_NOT_PRESENT,
+					   FSCACHE_CACHE_IS_PREPARING)) {
+		pr_warn("Cache tag %s in use\n", name);
+		fscache_put_cache(cache, fscache_cache_put_cache);
+		return ERR_PTR(-EBUSY);
+	}
+
+	return cache;
+}
+EXPORT_SYMBOL(fscache_acquire_cache);
+
+/**
+ * fscache_put_cache - Release a cache-level cookie.
+ * @cache: The cache cookie to be released
+ * @where: An indication of where the release happened
+ *
+ * Release the caller's reference on a cache-level cookie.  The @where
+ * indication should give information about the circumstances in which the call
+ * occurs and will be logged through a tracepoint.
+ */
+void fscache_put_cache(struct fscache_cache *cache,
+		       enum fscache_cache_trace where)
+{
+	unsigned int debug_id;
+	bool zero;
+	int ref;
+
+	if (IS_ERR_OR_NULL(cache))
+		return;
+
+	debug_id = cache->debug_id;
+	zero = __refcount_dec_and_test(&cache->ref, &ref);
+	trace_fscache_cache(debug_id, ref - 1, where);
+
+	if (zero) {
+		down_write(&fscache_addremove_sem);
+		list_del_init(&cache->cache_link);
+		up_write(&fscache_addremove_sem);
+		kfree(cache->name);
+		kfree(cache);
+	}
+}
+
+/**
+ * fscache_relinquish_cache - Reset cache state and release cookie
+ * @cache: The cache cookie to be released
+ *
+ * Reset the state of a cache and release the caller's reference on a cache
+ * cookie.
+ */
+void fscache_relinquish_cache(struct fscache_cache *cache)
+{
+	enum fscache_cache_trace where =
+		(cache->state == FSCACHE_CACHE_IS_PREPARING) ?
+		fscache_cache_put_prep_failed :
+		fscache_cache_put_relinquish;
+
+	cache->ops = NULL;
+	cache->cache_priv = NULL;
+	fscache_set_cache_state(cache, FSCACHE_CACHE_IS_NOT_PRESENT);
+	fscache_put_cache(cache, where);
+}
+EXPORT_SYMBOL(fscache_relinquish_cache);
+
+/**
+ * fscache_add_cache - Declare a cache as being open for business
+ * @cache: The cache-level cookie representing the cache
+ * @ops: Table of cache operations to use
+ * @cache_priv: Private data for the cache record
+ *
+ * Add a cache to the system, making it available for netfs's to use.
+ *
+ * See Documentation/filesystems/caching/backend-api.rst for a complete
+ * description.
+ */
+int fscache_add_cache(struct fscache_cache *cache,
+		      const struct fscache_cache_ops *ops,
+		      void *cache_priv)
+{
+	int n_accesses;
+
+	_enter("{%s,%s}", ops->name, cache->name);
+
+	BUG_ON(fscache_cache_state(cache) != FSCACHE_CACHE_IS_PREPARING);
+
+	/* Get a ref on the cache cookie and keep its n_accesses counter raised
+	 * by 1 to prevent wakeups from transitioning it to 0 until we're
+	 * withdrawing caching services from it.
+	 */
+	n_accesses = atomic_inc_return(&cache->n_accesses);
+	trace_fscache_access_cache(cache->debug_id, refcount_read(&cache->ref),
+				   n_accesses, fscache_access_cache_pin);
+
+	down_write(&fscache_addremove_sem);
+
+	cache->ops = ops;
+	cache->cache_priv = cache_priv;
+	fscache_set_cache_state(cache, FSCACHE_CACHE_IS_ACTIVE);
+
+	up_write(&fscache_addremove_sem);
+	pr_notice("Cache \"%s\" added (type %s)\n", cache->name, ops->name);
+	_leave(" = 0 [%s]", cache->name);
+	return 0;
+}
+EXPORT_SYMBOL(fscache_add_cache);
+
+/**
+ * fscache_begin_cache_access - Pin a cache so it can be accessed
+ * @cache: The cache-level cookie
+ * @why: An indication of the circumstances of the access for tracing
+ *
+ * Attempt to pin the cache to prevent it from going away whilst we're
+ * accessing it and returns true if successful.  This works as follows:
+ *
+ *  (1) If the cache tests as not live (state is not FSCACHE_CACHE_IS_ACTIVE),
+ *      then we return false to indicate access was not permitted.
+ *
+ *  (2) If the cache tests as live, then we increment the n_accesses count and
+ *      then recheck the liveness, ending the access if it ceased to be live.
+ *
+ *  (3) When we end the access, we decrement n_accesses and wake up the any
+ *      waiters if it reaches 0.
+ *
+ *  (4) Whilst the cache is caching, n_accesses is kept artificially
+ *      incremented to prevent wakeups from happening.
+ *
+ *  (5) When the cache is taken offline, the state is changed to prevent new
+ *      accesses, n_accesses is decremented and we wait for n_accesses to
+ *      become 0.
+ */
+bool fscache_begin_cache_access(struct fscache_cache *cache, enum fscache_access_trace why)
+{
+	int n_accesses;
+
+	if (!fscache_cache_is_live(cache))
+		return false;
+
+	n_accesses = atomic_inc_return(&cache->n_accesses);
+	smp_mb__after_atomic(); /* Reread live flag after n_accesses */
+	trace_fscache_access_cache(cache->debug_id, refcount_read(&cache->ref),
+				   n_accesses, why);
+	if (!fscache_cache_is_live(cache)) {
+		fscache_end_cache_access(cache, fscache_access_unlive);
+		return false;
+	}
+	return true;
+}
+
+/**
+ * fscache_end_cache_access - Unpin a cache at the end of an access.
+ * @cache: The cache-level cookie
+ * @why: An indication of the circumstances of the access for tracing
+ *
+ * Unpin a cache after we've accessed it.  The @why indicator is merely
+ * provided for tracing purposes.
+ */
+void fscache_end_cache_access(struct fscache_cache *cache, enum fscache_access_trace why)
+{
+	int n_accesses;
+
+	smp_mb__before_atomic();
+	n_accesses = atomic_dec_return(&cache->n_accesses);
+	trace_fscache_access_cache(cache->debug_id, refcount_read(&cache->ref),
+				   n_accesses, why);
+	if (n_accesses == 0)
+		wake_up_var(&cache->n_accesses);
+}
+
+/**
+ * fscache_io_error - Note a cache I/O error
+ * @cache: The record describing the cache
+ *
+ * Note that an I/O error occurred in a cache and that it should no longer be
+ * used for anything.  This also reports the error into the kernel log.
+ *
+ * See Documentation/filesystems/caching/backend-api.rst for a complete
+ * description.
+ */
+void fscache_io_error(struct fscache_cache *cache)
+{
+	if (fscache_set_cache_state_maybe(cache,
+					  FSCACHE_CACHE_IS_ACTIVE,
+					  FSCACHE_CACHE_GOT_IOERROR))
+		pr_err("Cache '%s' stopped due to I/O error\n",
+		       cache->name);
+}
+EXPORT_SYMBOL(fscache_io_error);
+
+/**
+ * fscache_withdraw_cache - Withdraw a cache from the active service
+ * @cache: The cache cookie
+ *
+ * Begin the process of withdrawing a cache from service.  This stops new
+ * cache-level and volume-level accesses from taking place and waits for
+ * currently ongoing cache-level accesses to end.
+ */
+void fscache_withdraw_cache(struct fscache_cache *cache)
+{
+	int n_accesses;
+
+	pr_notice("Withdrawing cache \"%s\" (%u objs)\n",
+		  cache->name, atomic_read(&cache->object_count));
+
+	fscache_set_cache_state(cache, FSCACHE_CACHE_IS_WITHDRAWN);
+
+	/* Allow wakeups on dec-to-0 */
+	n_accesses = atomic_dec_return(&cache->n_accesses);
+	trace_fscache_access_cache(cache->debug_id, refcount_read(&cache->ref),
+				   n_accesses, fscache_access_cache_unpin);
+
+	wait_var_event(&cache->n_accesses,
+		       atomic_read(&cache->n_accesses) == 0);
+}
+EXPORT_SYMBOL(fscache_withdraw_cache);
+
+#ifdef CONFIG_PROC_FS
+static const char fscache_cache_states[NR__FSCACHE_CACHE_STATE] __nonstring = "-PAEW";
+
+/*
+ * Generate a list of caches in /proc/fs/fscache/caches
+ */
+static int fscache_caches_seq_show(struct seq_file *m, void *v)
+{
+	struct fscache_cache *cache;
+
+	if (v == &fscache_caches) {
+		seq_puts(m,
+			 "CACHE    REF   VOLS  OBJS  ACCES S NAME\n"
+			 "======== ===== ===== ===== ===== = ===============\n"
+			 );
+		return 0;
+	}
+
+	cache = list_entry(v, struct fscache_cache, cache_link);
+	seq_printf(m,
+		   "%08x %5d %5d %5d %5d %c %s\n",
+		   cache->debug_id,
+		   refcount_read(&cache->ref),
+		   atomic_read(&cache->n_volumes),
+		   atomic_read(&cache->object_count),
+		   atomic_read(&cache->n_accesses),
+		   fscache_cache_states[cache->state],
+		   cache->name ?: "-");
+	return 0;
+}
+
+static void *fscache_caches_seq_start(struct seq_file *m, loff_t *_pos)
+	__acquires(fscache_addremove_sem)
+{
+	down_read(&fscache_addremove_sem);
+	return seq_list_start_head(&fscache_caches, *_pos);
+}
+
+static void *fscache_caches_seq_next(struct seq_file *m, void *v, loff_t *_pos)
+{
+	return seq_list_next(v, &fscache_caches, _pos);
+}
+
+static void fscache_caches_seq_stop(struct seq_file *m, void *v)
+	__releases(fscache_addremove_sem)
+{
+	up_read(&fscache_addremove_sem);
+}
+
+const struct seq_operations fscache_caches_seq_ops = {
+	.start  = fscache_caches_seq_start,
+	.next   = fscache_caches_seq_next,
+	.stop   = fscache_caches_seq_stop,
+	.show   = fscache_caches_seq_show,
+};
+#endif /* CONFIG_PROC_FS */
diff --git a/fs/netfs/fscache_cookie.c b/fs/netfs/fscache_cookie.c
new file mode 100644
index 000000000..3d56fc734
--- /dev/null
+++ b/fs/netfs/fscache_cookie.c
@@ -0,0 +1,1184 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* netfs cookie management
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ *
+ * See Documentation/filesystems/caching/netfs-api.rst for more information on
+ * the netfs API.
+ */
+
+#define FSCACHE_DEBUG_LEVEL COOKIE
+#include <linux/module.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+struct kmem_cache *fscache_cookie_jar;
+
+static void fscache_cookie_lru_timed_out(struct timer_list *timer);
+static void fscache_cookie_lru_worker(struct work_struct *work);
+static void fscache_cookie_worker(struct work_struct *work);
+static void fscache_unhash_cookie(struct fscache_cookie *cookie);
+static void fscache_perform_invalidation(struct fscache_cookie *cookie);
+
+#define fscache_cookie_hash_shift 15
+static struct hlist_bl_head fscache_cookie_hash[1 << fscache_cookie_hash_shift];
+static LIST_HEAD(fscache_cookies);
+static DEFINE_RWLOCK(fscache_cookies_lock);
+static LIST_HEAD(fscache_cookie_lru);
+static DEFINE_SPINLOCK(fscache_cookie_lru_lock);
+DEFINE_TIMER(fscache_cookie_lru_timer, fscache_cookie_lru_timed_out);
+static DECLARE_WORK(fscache_cookie_lru_work, fscache_cookie_lru_worker);
+static const char fscache_cookie_states[FSCACHE_COOKIE_STATE__NR] __nonstring = "-LCAIFUWRD";
+static unsigned int fscache_lru_cookie_timeout = 10 * HZ;
+
+void fscache_print_cookie(struct fscache_cookie *cookie, char prefix)
+{
+	const u8 *k;
+
+	pr_err("%c-cookie c=%08x [fl=%lx na=%u nA=%u s=%c]\n",
+	       prefix,
+	       cookie->debug_id,
+	       cookie->flags,
+	       atomic_read(&cookie->n_active),
+	       atomic_read(&cookie->n_accesses),
+	       fscache_cookie_states[cookie->state]);
+	pr_err("%c-cookie V=%08x [%s]\n",
+	       prefix,
+	       cookie->volume->debug_id,
+	       cookie->volume->key);
+
+	k = (cookie->key_len <= sizeof(cookie->inline_key)) ?
+		cookie->inline_key : cookie->key;
+	pr_err("%c-key=[%u] '%*phN'\n", prefix, cookie->key_len, cookie->key_len, k);
+}
+
+static void fscache_free_cookie(struct fscache_cookie *cookie)
+{
+	if (WARN_ON_ONCE(!list_empty(&cookie->commit_link))) {
+		spin_lock(&fscache_cookie_lru_lock);
+		list_del_init(&cookie->commit_link);
+		spin_unlock(&fscache_cookie_lru_lock);
+		fscache_stat_d(&fscache_n_cookies_lru);
+		fscache_stat(&fscache_n_cookies_lru_removed);
+	}
+
+	if (WARN_ON_ONCE(test_bit(FSCACHE_COOKIE_IS_HASHED, &cookie->flags))) {
+		fscache_print_cookie(cookie, 'F');
+		return;
+	}
+
+	write_lock(&fscache_cookies_lock);
+	list_del(&cookie->proc_link);
+	write_unlock(&fscache_cookies_lock);
+	if (cookie->aux_len > sizeof(cookie->inline_aux))
+		kfree(cookie->aux);
+	if (cookie->key_len > sizeof(cookie->inline_key))
+		kfree(cookie->key);
+	fscache_stat_d(&fscache_n_cookies);
+	kmem_cache_free(fscache_cookie_jar, cookie);
+}
+
+static void __fscache_queue_cookie(struct fscache_cookie *cookie)
+{
+	if (!queue_work(fscache_wq, &cookie->work))
+		fscache_put_cookie(cookie, fscache_cookie_put_over_queued);
+}
+
+static void fscache_queue_cookie(struct fscache_cookie *cookie,
+				 enum fscache_cookie_trace where)
+{
+	fscache_get_cookie(cookie, where);
+	__fscache_queue_cookie(cookie);
+}
+
+/*
+ * Initialise the access gate on a cookie by setting a flag to prevent the
+ * state machine from being queued when the access counter transitions to 0.
+ * We're only interested in this when we withdraw caching services from the
+ * cookie.
+ */
+static void fscache_init_access_gate(struct fscache_cookie *cookie)
+{
+	int n_accesses;
+
+	n_accesses = atomic_read(&cookie->n_accesses);
+	trace_fscache_access(cookie->debug_id, refcount_read(&cookie->ref),
+			     n_accesses, fscache_access_cache_pin);
+	set_bit(FSCACHE_COOKIE_NO_ACCESS_WAKE, &cookie->flags);
+}
+
+/**
+ * fscache_end_cookie_access - Unpin a cache at the end of an access.
+ * @cookie: A data file cookie
+ * @why: An indication of the circumstances of the access for tracing
+ *
+ * Unpin a cache cookie after we've accessed it and bring a deferred
+ * relinquishment or withdrawal state into effect.
+ *
+ * The @why indicator is provided for tracing purposes.
+ */
+void fscache_end_cookie_access(struct fscache_cookie *cookie,
+			       enum fscache_access_trace why)
+{
+	int n_accesses;
+
+	smp_mb__before_atomic();
+	n_accesses = atomic_dec_return(&cookie->n_accesses);
+	trace_fscache_access(cookie->debug_id, refcount_read(&cookie->ref),
+			     n_accesses, why);
+	if (n_accesses == 0 &&
+	    !test_bit(FSCACHE_COOKIE_NO_ACCESS_WAKE, &cookie->flags))
+		fscache_queue_cookie(cookie, fscache_cookie_get_end_access);
+}
+EXPORT_SYMBOL(fscache_end_cookie_access);
+
+/*
+ * Pin the cache behind a cookie so that we can access it.
+ */
+static void __fscache_begin_cookie_access(struct fscache_cookie *cookie,
+					  enum fscache_access_trace why)
+{
+	int n_accesses;
+
+	n_accesses = atomic_inc_return(&cookie->n_accesses);
+	smp_mb__after_atomic(); /* (Future) read state after is-caching.
+				 * Reread n_accesses after is-caching
+				 */
+	trace_fscache_access(cookie->debug_id, refcount_read(&cookie->ref),
+			     n_accesses, why);
+}
+
+/**
+ * fscache_begin_cookie_access - Pin a cache so data can be accessed
+ * @cookie: A data file cookie
+ * @why: An indication of the circumstances of the access for tracing
+ *
+ * Attempt to pin the cache to prevent it from going away whilst we're
+ * accessing data and returns true if successful.  This works as follows:
+ *
+ *  (1) If the cookie is not being cached (ie. FSCACHE_COOKIE_IS_CACHING is not
+ *      set), we return false to indicate access was not permitted.
+ *
+ *  (2) If the cookie is being cached, we increment its n_accesses count and
+ *      then recheck the IS_CACHING flag, ending the access if it got cleared.
+ *
+ *  (3) When we end the access, we decrement the cookie's n_accesses and wake
+ *      up the any waiters if it reaches 0.
+ *
+ *  (4) Whilst the cookie is actively being cached, its n_accesses is kept
+ *      artificially incremented to prevent wakeups from happening.
+ *
+ *  (5) When the cache is taken offline or if the cookie is culled, the flag is
+ *      cleared to prevent new accesses, the cookie's n_accesses is decremented
+ *      and we wait for it to become 0.
+ *
+ * The @why indicator are merely provided for tracing purposes.
+ */
+bool fscache_begin_cookie_access(struct fscache_cookie *cookie,
+				 enum fscache_access_trace why)
+{
+	if (!test_bit(FSCACHE_COOKIE_IS_CACHING, &cookie->flags))
+		return false;
+	__fscache_begin_cookie_access(cookie, why);
+	if (!test_bit(FSCACHE_COOKIE_IS_CACHING, &cookie->flags) ||
+	    !fscache_cache_is_live(cookie->volume->cache)) {
+		fscache_end_cookie_access(cookie, fscache_access_unlive);
+		return false;
+	}
+	return true;
+}
+
+static inline void wake_up_cookie_state(struct fscache_cookie *cookie)
+{
+	/* Use a barrier to ensure that waiters see the state variable
+	 * change, as spin_unlock doesn't guarantee a barrier.
+	 *
+	 * See comments over wake_up_bit() and waitqueue_active().
+	 */
+	smp_mb();
+	wake_up_var(&cookie->state);
+}
+
+/*
+ * Change the state a cookie is at and wake up anyone waiting for that.  Impose
+ * an ordering between the stuff stored in the cookie and the state member.
+ * Paired with fscache_cookie_state().
+ */
+static void __fscache_set_cookie_state(struct fscache_cookie *cookie,
+				       enum fscache_cookie_state state)
+{
+	smp_store_release(&cookie->state, state);
+}
+
+static void fscache_set_cookie_state(struct fscache_cookie *cookie,
+				     enum fscache_cookie_state state)
+{
+	spin_lock(&cookie->lock);
+	__fscache_set_cookie_state(cookie, state);
+	spin_unlock(&cookie->lock);
+	wake_up_cookie_state(cookie);
+}
+
+/**
+ * fscache_cookie_lookup_negative - Note negative lookup
+ * @cookie: The cookie that was being looked up
+ *
+ * Note that some part of the metadata path in the cache doesn't exist and so
+ * we can release any waiting readers in the certain knowledge that there's
+ * nothing for them to actually read.
+ *
+ * This function uses no locking and must only be called from the state machine.
+ */
+void fscache_cookie_lookup_negative(struct fscache_cookie *cookie)
+{
+	set_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
+	fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_CREATING);
+}
+EXPORT_SYMBOL(fscache_cookie_lookup_negative);
+
+/**
+ * fscache_resume_after_invalidation - Allow I/O to resume after invalidation
+ * @cookie: The cookie that was invalidated
+ *
+ * Tell fscache that invalidation is sufficiently complete that I/O can be
+ * allowed again.
+ */
+void fscache_resume_after_invalidation(struct fscache_cookie *cookie)
+{
+	fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_ACTIVE);
+}
+EXPORT_SYMBOL(fscache_resume_after_invalidation);
+
+/**
+ * fscache_caching_failed - Report that a failure stopped caching on a cookie
+ * @cookie: The cookie that was affected
+ *
+ * Tell fscache that caching on a cookie needs to be stopped due to some sort
+ * of failure.
+ *
+ * This function uses no locking and must only be called from the state machine.
+ */
+void fscache_caching_failed(struct fscache_cookie *cookie)
+{
+	clear_bit(FSCACHE_COOKIE_IS_CACHING, &cookie->flags);
+	fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_FAILED);
+	trace_fscache_cookie(cookie->debug_id, refcount_read(&cookie->ref),
+				fscache_cookie_failed);
+}
+EXPORT_SYMBOL(fscache_caching_failed);
+
+/*
+ * Set the index key in a cookie.  The cookie struct has space for a 16-byte
+ * key plus length and hash, but if that's not big enough, it's instead a
+ * pointer to a buffer containing 3 bytes of hash, 1 byte of length and then
+ * the key data.
+ */
+static int fscache_set_key(struct fscache_cookie *cookie,
+			   const void *index_key, size_t index_key_len)
+{
+	void *buf;
+	size_t buf_size;
+
+	buf_size = round_up(index_key_len, sizeof(__le32));
+
+	if (index_key_len > sizeof(cookie->inline_key)) {
+		buf = kzalloc(buf_size, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+		cookie->key = buf;
+	} else {
+		buf = cookie->inline_key;
+	}
+
+	memcpy(buf, index_key, index_key_len);
+	cookie->key_hash = fscache_hash(cookie->volume->key_hash,
+					buf, buf_size);
+	return 0;
+}
+
+static bool fscache_cookie_same(const struct fscache_cookie *a,
+				const struct fscache_cookie *b)
+{
+	const void *ka, *kb;
+
+	if (a->key_hash	!= b->key_hash ||
+	    a->volume	!= b->volume ||
+	    a->key_len	!= b->key_len)
+		return false;
+
+	if (a->key_len <= sizeof(a->inline_key)) {
+		ka = &a->inline_key;
+		kb = &b->inline_key;
+	} else {
+		ka = a->key;
+		kb = b->key;
+	}
+	return memcmp(ka, kb, a->key_len) == 0;
+}
+
+static atomic_t fscache_cookie_debug_id = ATOMIC_INIT(1);
+
+/*
+ * Allocate a cookie.
+ */
+static struct fscache_cookie *fscache_alloc_cookie(
+	struct fscache_volume *volume,
+	u8 advice,
+	const void *index_key, size_t index_key_len,
+	const void *aux_data, size_t aux_data_len,
+	loff_t object_size)
+{
+	struct fscache_cookie *cookie;
+
+	/* allocate and initialise a cookie */
+	cookie = kmem_cache_zalloc(fscache_cookie_jar, GFP_KERNEL);
+	if (!cookie)
+		return NULL;
+	fscache_stat(&fscache_n_cookies);
+
+	cookie->volume		= volume;
+	cookie->advice		= advice;
+	cookie->key_len		= index_key_len;
+	cookie->aux_len		= aux_data_len;
+	cookie->object_size	= object_size;
+	if (object_size == 0)
+		__set_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
+
+	if (fscache_set_key(cookie, index_key, index_key_len) < 0)
+		goto nomem;
+
+	if (cookie->aux_len <= sizeof(cookie->inline_aux)) {
+		memcpy(cookie->inline_aux, aux_data, cookie->aux_len);
+	} else {
+		cookie->aux = kmemdup(aux_data, cookie->aux_len, GFP_KERNEL);
+		if (!cookie->aux)
+			goto nomem;
+	}
+
+	refcount_set(&cookie->ref, 1);
+	cookie->debug_id = atomic_inc_return(&fscache_cookie_debug_id);
+	spin_lock_init(&cookie->lock);
+	INIT_LIST_HEAD(&cookie->commit_link);
+	INIT_WORK(&cookie->work, fscache_cookie_worker);
+	__fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_QUIESCENT);
+
+	write_lock(&fscache_cookies_lock);
+	list_add_tail(&cookie->proc_link, &fscache_cookies);
+	write_unlock(&fscache_cookies_lock);
+	fscache_see_cookie(cookie, fscache_cookie_new_acquire);
+	return cookie;
+
+nomem:
+	fscache_free_cookie(cookie);
+	return NULL;
+}
+
+static inline bool fscache_cookie_is_dropped(struct fscache_cookie *cookie)
+{
+	return READ_ONCE(cookie->state) == FSCACHE_COOKIE_STATE_DROPPED;
+}
+
+static void fscache_wait_on_collision(struct fscache_cookie *candidate,
+				      struct fscache_cookie *wait_for)
+{
+	enum fscache_cookie_state *statep = &wait_for->state;
+
+	wait_var_event_timeout(statep, fscache_cookie_is_dropped(wait_for),
+			       20 * HZ);
+	if (!fscache_cookie_is_dropped(wait_for)) {
+		pr_notice("Potential collision c=%08x old: c=%08x",
+			  candidate->debug_id, wait_for->debug_id);
+		wait_var_event(statep, fscache_cookie_is_dropped(wait_for));
+	}
+}
+
+/*
+ * Attempt to insert the new cookie into the hash.  If there's a collision, we
+ * wait for the old cookie to complete if it's being relinquished and an error
+ * otherwise.
+ */
+static bool fscache_hash_cookie(struct fscache_cookie *candidate)
+{
+	struct fscache_cookie *cursor, *wait_for = NULL;
+	struct hlist_bl_head *h;
+	struct hlist_bl_node *p;
+	unsigned int bucket;
+
+	bucket = candidate->key_hash & (ARRAY_SIZE(fscache_cookie_hash) - 1);
+	h = &fscache_cookie_hash[bucket];
+
+	hlist_bl_lock(h);
+	hlist_bl_for_each_entry(cursor, p, h, hash_link) {
+		if (fscache_cookie_same(candidate, cursor)) {
+			if (!test_bit(FSCACHE_COOKIE_RELINQUISHED, &cursor->flags))
+				goto collision;
+			wait_for = fscache_get_cookie(cursor,
+						      fscache_cookie_get_hash_collision);
+			break;
+		}
+	}
+
+	fscache_get_volume(candidate->volume, fscache_volume_get_cookie);
+	atomic_inc(&candidate->volume->n_cookies);
+	hlist_bl_add_head(&candidate->hash_link, h);
+	set_bit(FSCACHE_COOKIE_IS_HASHED, &candidate->flags);
+	hlist_bl_unlock(h);
+
+	if (wait_for) {
+		fscache_wait_on_collision(candidate, wait_for);
+		fscache_put_cookie(wait_for, fscache_cookie_put_hash_collision);
+	}
+	return true;
+
+collision:
+	trace_fscache_cookie(cursor->debug_id, refcount_read(&cursor->ref),
+			     fscache_cookie_collision);
+	pr_err("Duplicate cookie detected\n");
+	fscache_print_cookie(cursor, 'O');
+	fscache_print_cookie(candidate, 'N');
+	hlist_bl_unlock(h);
+	return false;
+}
+
+/*
+ * Request a cookie to represent a data storage object within a volume.
+ *
+ * We never let on to the netfs about errors.  We may set a negative cookie
+ * pointer, but that's okay
+ */
+struct fscache_cookie *__fscache_acquire_cookie(
+	struct fscache_volume *volume,
+	u8 advice,
+	const void *index_key, size_t index_key_len,
+	const void *aux_data, size_t aux_data_len,
+	loff_t object_size)
+{
+	struct fscache_cookie *cookie;
+
+	_enter("V=%x", volume->debug_id);
+
+	if (!index_key || !index_key_len || index_key_len > 255 || aux_data_len > 255)
+		return NULL;
+	if (!aux_data || !aux_data_len) {
+		aux_data = NULL;
+		aux_data_len = 0;
+	}
+
+	fscache_stat(&fscache_n_acquires);
+
+	cookie = fscache_alloc_cookie(volume, advice,
+				      index_key, index_key_len,
+				      aux_data, aux_data_len,
+				      object_size);
+	if (!cookie) {
+		fscache_stat(&fscache_n_acquires_oom);
+		return NULL;
+	}
+
+	if (!fscache_hash_cookie(cookie)) {
+		fscache_see_cookie(cookie, fscache_cookie_discard);
+		fscache_free_cookie(cookie);
+		return NULL;
+	}
+
+	trace_fscache_acquire(cookie);
+	fscache_stat(&fscache_n_acquires_ok);
+	_leave(" = c=%08x", cookie->debug_id);
+	return cookie;
+}
+EXPORT_SYMBOL(__fscache_acquire_cookie);
+
+/*
+ * Prepare a cache object to be written to.
+ */
+static void fscache_prepare_to_write(struct fscache_cookie *cookie)
+{
+	cookie->volume->cache->ops->prepare_to_write(cookie);
+}
+
+/*
+ * Look up a cookie in the cache.
+ */
+static void fscache_perform_lookup(struct fscache_cookie *cookie)
+{
+	enum fscache_access_trace trace = fscache_access_lookup_cookie_end_failed;
+	bool need_withdraw = false;
+
+	_enter("");
+
+	if (!cookie->volume->cache_priv) {
+		fscache_create_volume(cookie->volume, true);
+		if (!cookie->volume->cache_priv) {
+			fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_QUIESCENT);
+			goto out;
+		}
+	}
+
+	if (!cookie->volume->cache->ops->lookup_cookie(cookie)) {
+		if (cookie->state != FSCACHE_COOKIE_STATE_FAILED)
+			fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_QUIESCENT);
+		need_withdraw = true;
+		_leave(" [fail]");
+		goto out;
+	}
+
+	fscache_see_cookie(cookie, fscache_cookie_see_active);
+	spin_lock(&cookie->lock);
+	if (test_and_clear_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags))
+		__fscache_set_cookie_state(cookie,
+					   FSCACHE_COOKIE_STATE_INVALIDATING);
+	else
+		__fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_ACTIVE);
+	spin_unlock(&cookie->lock);
+	wake_up_cookie_state(cookie);
+	trace = fscache_access_lookup_cookie_end;
+
+out:
+	fscache_end_cookie_access(cookie, trace);
+	if (need_withdraw)
+		fscache_withdraw_cookie(cookie);
+	fscache_end_volume_access(cookie->volume, cookie, trace);
+}
+
+/*
+ * Begin the process of looking up a cookie.  We offload the actual process to
+ * a worker thread.
+ */
+static bool fscache_begin_lookup(struct fscache_cookie *cookie, bool will_modify)
+{
+	if (will_modify) {
+		set_bit(FSCACHE_COOKIE_LOCAL_WRITE, &cookie->flags);
+		set_bit(FSCACHE_COOKIE_DO_PREP_TO_WRITE, &cookie->flags);
+	}
+	if (!fscache_begin_volume_access(cookie->volume, cookie,
+					 fscache_access_lookup_cookie))
+		return false;
+
+	__fscache_begin_cookie_access(cookie, fscache_access_lookup_cookie);
+	__fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_LOOKING_UP);
+	set_bit(FSCACHE_COOKIE_IS_CACHING, &cookie->flags);
+	set_bit(FSCACHE_COOKIE_HAS_BEEN_CACHED, &cookie->flags);
+	return true;
+}
+
+/*
+ * Start using the cookie for I/O.  This prevents the backing object from being
+ * reaped by VM pressure.
+ */
+void __fscache_use_cookie(struct fscache_cookie *cookie, bool will_modify)
+{
+	enum fscache_cookie_state state;
+	bool queue = false;
+	int n_active;
+
+	_enter("c=%08x", cookie->debug_id);
+
+	if (WARN(test_bit(FSCACHE_COOKIE_RELINQUISHED, &cookie->flags),
+		 "Trying to use relinquished cookie\n"))
+		return;
+
+	spin_lock(&cookie->lock);
+
+	n_active = atomic_inc_return(&cookie->n_active);
+	trace_fscache_active(cookie->debug_id, refcount_read(&cookie->ref),
+			     n_active, atomic_read(&cookie->n_accesses),
+			     will_modify ?
+			     fscache_active_use_modify : fscache_active_use);
+
+again:
+	state = fscache_cookie_state(cookie);
+	switch (state) {
+	case FSCACHE_COOKIE_STATE_QUIESCENT:
+		queue = fscache_begin_lookup(cookie, will_modify);
+		break;
+
+	case FSCACHE_COOKIE_STATE_LOOKING_UP:
+	case FSCACHE_COOKIE_STATE_CREATING:
+		if (will_modify)
+			set_bit(FSCACHE_COOKIE_LOCAL_WRITE, &cookie->flags);
+		break;
+	case FSCACHE_COOKIE_STATE_ACTIVE:
+	case FSCACHE_COOKIE_STATE_INVALIDATING:
+		if (will_modify &&
+		    !test_and_set_bit(FSCACHE_COOKIE_LOCAL_WRITE, &cookie->flags)) {
+			set_bit(FSCACHE_COOKIE_DO_PREP_TO_WRITE, &cookie->flags);
+			queue = true;
+		}
+		/*
+		 * We could race with cookie_lru which may set LRU_DISCARD bit
+		 * but has yet to run the cookie state machine.  If this happens
+		 * and another thread tries to use the cookie, clear LRU_DISCARD
+		 * so we don't end up withdrawing the cookie while in use.
+		 */
+		if (test_and_clear_bit(FSCACHE_COOKIE_DO_LRU_DISCARD, &cookie->flags))
+			fscache_see_cookie(cookie, fscache_cookie_see_lru_discard_clear);
+		break;
+
+	case FSCACHE_COOKIE_STATE_FAILED:
+	case FSCACHE_COOKIE_STATE_WITHDRAWING:
+		break;
+
+	case FSCACHE_COOKIE_STATE_LRU_DISCARDING:
+		spin_unlock(&cookie->lock);
+		wait_var_event(&cookie->state,
+			       fscache_cookie_state(cookie) !=
+			       FSCACHE_COOKIE_STATE_LRU_DISCARDING);
+		spin_lock(&cookie->lock);
+		goto again;
+
+	case FSCACHE_COOKIE_STATE_DROPPED:
+	case FSCACHE_COOKIE_STATE_RELINQUISHING:
+		WARN(1, "Can't use cookie in state %u\n", state);
+		break;
+	}
+
+	spin_unlock(&cookie->lock);
+	if (queue)
+		fscache_queue_cookie(cookie, fscache_cookie_get_use_work);
+	_leave("");
+}
+EXPORT_SYMBOL(__fscache_use_cookie);
+
+static void fscache_unuse_cookie_locked(struct fscache_cookie *cookie)
+{
+	clear_bit(FSCACHE_COOKIE_DISABLED, &cookie->flags);
+	if (!test_bit(FSCACHE_COOKIE_IS_CACHING, &cookie->flags))
+		return;
+
+	cookie->unused_at = jiffies;
+	spin_lock(&fscache_cookie_lru_lock);
+	if (list_empty(&cookie->commit_link)) {
+		fscache_get_cookie(cookie, fscache_cookie_get_lru);
+		fscache_stat(&fscache_n_cookies_lru);
+	}
+	list_move_tail(&cookie->commit_link, &fscache_cookie_lru);
+
+	spin_unlock(&fscache_cookie_lru_lock);
+	timer_reduce(&fscache_cookie_lru_timer,
+		     jiffies + fscache_lru_cookie_timeout);
+}
+
+/*
+ * Stop using the cookie for I/O.
+ */
+void __fscache_unuse_cookie(struct fscache_cookie *cookie,
+			    const void *aux_data, const loff_t *object_size)
+{
+	unsigned int debug_id = cookie->debug_id;
+	unsigned int r = refcount_read(&cookie->ref);
+	unsigned int a = atomic_read(&cookie->n_accesses);
+	unsigned int c;
+
+	if (aux_data || object_size)
+		__fscache_update_cookie(cookie, aux_data, object_size);
+
+	/* Subtract 1 from counter unless that drops it to 0 (ie. it was 1) */
+	c = atomic_fetch_add_unless(&cookie->n_active, -1, 1);
+	if (c != 1) {
+		trace_fscache_active(debug_id, r, c - 1, a, fscache_active_unuse);
+		return;
+	}
+
+	spin_lock(&cookie->lock);
+	r = refcount_read(&cookie->ref);
+	a = atomic_read(&cookie->n_accesses);
+	c = atomic_dec_return(&cookie->n_active);
+	trace_fscache_active(debug_id, r, c, a, fscache_active_unuse);
+	if (c == 0)
+		fscache_unuse_cookie_locked(cookie);
+	spin_unlock(&cookie->lock);
+}
+EXPORT_SYMBOL(__fscache_unuse_cookie);
+
+/*
+ * Perform work upon the cookie, such as committing its cache state,
+ * relinquishing it or withdrawing the backing cache.  We're protected from the
+ * cache going away under us as object withdrawal must come through this
+ * non-reentrant work item.
+ */
+static void fscache_cookie_state_machine(struct fscache_cookie *cookie)
+{
+	enum fscache_cookie_state state;
+	bool wake = false;
+
+	_enter("c=%x", cookie->debug_id);
+
+again:
+	spin_lock(&cookie->lock);
+again_locked:
+	state = cookie->state;
+	switch (state) {
+	case FSCACHE_COOKIE_STATE_QUIESCENT:
+		/* The QUIESCENT state is jumped to the LOOKING_UP state by
+		 * fscache_use_cookie().
+		 */
+
+		if (atomic_read(&cookie->n_accesses) == 0 &&
+		    test_bit(FSCACHE_COOKIE_DO_RELINQUISH, &cookie->flags)) {
+			__fscache_set_cookie_state(cookie,
+						   FSCACHE_COOKIE_STATE_RELINQUISHING);
+			wake = true;
+			goto again_locked;
+		}
+		break;
+
+	case FSCACHE_COOKIE_STATE_LOOKING_UP:
+		spin_unlock(&cookie->lock);
+		fscache_init_access_gate(cookie);
+		fscache_perform_lookup(cookie);
+		goto again;
+
+	case FSCACHE_COOKIE_STATE_INVALIDATING:
+		spin_unlock(&cookie->lock);
+		fscache_perform_invalidation(cookie);
+		goto again;
+
+	case FSCACHE_COOKIE_STATE_ACTIVE:
+		if (test_and_clear_bit(FSCACHE_COOKIE_DO_PREP_TO_WRITE, &cookie->flags)) {
+			spin_unlock(&cookie->lock);
+			fscache_prepare_to_write(cookie);
+			spin_lock(&cookie->lock);
+		}
+		if (test_bit(FSCACHE_COOKIE_DO_LRU_DISCARD, &cookie->flags)) {
+			if (atomic_read(&cookie->n_accesses) != 0)
+				/* still being accessed: postpone it */
+				break;
+
+			__fscache_set_cookie_state(cookie,
+						   FSCACHE_COOKIE_STATE_LRU_DISCARDING);
+			wake = true;
+			goto again_locked;
+		}
+		fallthrough;
+
+	case FSCACHE_COOKIE_STATE_FAILED:
+		if (test_and_clear_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags))
+			fscache_end_cookie_access(cookie, fscache_access_invalidate_cookie_end);
+
+		if (atomic_read(&cookie->n_accesses) != 0)
+			break;
+		if (test_bit(FSCACHE_COOKIE_DO_RELINQUISH, &cookie->flags)) {
+			__fscache_set_cookie_state(cookie,
+						   FSCACHE_COOKIE_STATE_RELINQUISHING);
+			wake = true;
+			goto again_locked;
+		}
+		if (test_bit(FSCACHE_COOKIE_DO_WITHDRAW, &cookie->flags)) {
+			__fscache_set_cookie_state(cookie,
+						   FSCACHE_COOKIE_STATE_WITHDRAWING);
+			wake = true;
+			goto again_locked;
+		}
+		break;
+
+	case FSCACHE_COOKIE_STATE_LRU_DISCARDING:
+	case FSCACHE_COOKIE_STATE_RELINQUISHING:
+	case FSCACHE_COOKIE_STATE_WITHDRAWING:
+		if (cookie->cache_priv) {
+			spin_unlock(&cookie->lock);
+			cookie->volume->cache->ops->withdraw_cookie(cookie);
+			spin_lock(&cookie->lock);
+		}
+
+		if (test_and_clear_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags))
+			fscache_end_cookie_access(cookie, fscache_access_invalidate_cookie_end);
+
+		switch (state) {
+		case FSCACHE_COOKIE_STATE_RELINQUISHING:
+			fscache_see_cookie(cookie, fscache_cookie_see_relinquish);
+			fscache_unhash_cookie(cookie);
+			__fscache_set_cookie_state(cookie,
+						   FSCACHE_COOKIE_STATE_DROPPED);
+			wake = true;
+			goto out;
+		case FSCACHE_COOKIE_STATE_LRU_DISCARDING:
+			fscache_see_cookie(cookie, fscache_cookie_see_lru_discard);
+			break;
+		case FSCACHE_COOKIE_STATE_WITHDRAWING:
+			fscache_see_cookie(cookie, fscache_cookie_see_withdraw);
+			break;
+		default:
+			BUG();
+		}
+
+		clear_bit(FSCACHE_COOKIE_NEEDS_UPDATE, &cookie->flags);
+		clear_bit(FSCACHE_COOKIE_DO_WITHDRAW, &cookie->flags);
+		clear_bit(FSCACHE_COOKIE_DO_LRU_DISCARD, &cookie->flags);
+		clear_bit(FSCACHE_COOKIE_DO_PREP_TO_WRITE, &cookie->flags);
+		set_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
+		__fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_QUIESCENT);
+		wake = true;
+		goto again_locked;
+
+	case FSCACHE_COOKIE_STATE_DROPPED:
+		break;
+
+	default:
+		WARN_ONCE(1, "Cookie %x in unexpected state %u\n",
+			  cookie->debug_id, state);
+		break;
+	}
+
+out:
+	spin_unlock(&cookie->lock);
+	if (wake)
+		wake_up_cookie_state(cookie);
+	_leave("");
+}
+
+static void fscache_cookie_worker(struct work_struct *work)
+{
+	struct fscache_cookie *cookie = container_of(work, struct fscache_cookie, work);
+
+	fscache_see_cookie(cookie, fscache_cookie_see_work);
+	fscache_cookie_state_machine(cookie);
+	fscache_put_cookie(cookie, fscache_cookie_put_work);
+}
+
+/*
+ * Wait for the object to become inactive.  The cookie's work item will be
+ * scheduled when someone transitions n_accesses to 0 - but if someone's
+ * already done that, schedule it anyway.
+ */
+static void __fscache_withdraw_cookie(struct fscache_cookie *cookie)
+{
+	int n_accesses;
+	bool unpinned;
+
+	unpinned = test_and_clear_bit(FSCACHE_COOKIE_NO_ACCESS_WAKE, &cookie->flags);
+
+	/* Need to read the access count after unpinning */
+	n_accesses = atomic_read(&cookie->n_accesses);
+	if (unpinned)
+		trace_fscache_access(cookie->debug_id, refcount_read(&cookie->ref),
+				     n_accesses, fscache_access_cache_unpin);
+	if (n_accesses == 0)
+		fscache_queue_cookie(cookie, fscache_cookie_get_end_access);
+}
+
+static void fscache_cookie_lru_do_one(struct fscache_cookie *cookie)
+{
+	fscache_see_cookie(cookie, fscache_cookie_see_lru_do_one);
+
+	spin_lock(&cookie->lock);
+	if (cookie->state != FSCACHE_COOKIE_STATE_ACTIVE ||
+	    time_before(jiffies, cookie->unused_at + fscache_lru_cookie_timeout) ||
+	    atomic_read(&cookie->n_active) > 0) {
+		spin_unlock(&cookie->lock);
+		fscache_stat(&fscache_n_cookies_lru_removed);
+	} else {
+		set_bit(FSCACHE_COOKIE_DO_LRU_DISCARD, &cookie->flags);
+		spin_unlock(&cookie->lock);
+		fscache_stat(&fscache_n_cookies_lru_expired);
+		_debug("lru c=%x", cookie->debug_id);
+		__fscache_withdraw_cookie(cookie);
+	}
+
+	fscache_put_cookie(cookie, fscache_cookie_put_lru);
+}
+
+static void fscache_cookie_lru_worker(struct work_struct *work)
+{
+	struct fscache_cookie *cookie;
+	unsigned long unused_at;
+
+	spin_lock(&fscache_cookie_lru_lock);
+
+	while (!list_empty(&fscache_cookie_lru)) {
+		cookie = list_first_entry(&fscache_cookie_lru,
+					  struct fscache_cookie, commit_link);
+		unused_at = cookie->unused_at + fscache_lru_cookie_timeout;
+		if (time_before(jiffies, unused_at)) {
+			timer_reduce(&fscache_cookie_lru_timer, unused_at);
+			break;
+		}
+
+		list_del_init(&cookie->commit_link);
+		fscache_stat_d(&fscache_n_cookies_lru);
+		spin_unlock(&fscache_cookie_lru_lock);
+		fscache_cookie_lru_do_one(cookie);
+		spin_lock(&fscache_cookie_lru_lock);
+	}
+
+	spin_unlock(&fscache_cookie_lru_lock);
+}
+
+static void fscache_cookie_lru_timed_out(struct timer_list *timer)
+{
+	queue_work(fscache_wq, &fscache_cookie_lru_work);
+}
+
+static void fscache_cookie_drop_from_lru(struct fscache_cookie *cookie)
+{
+	bool need_put = false;
+
+	if (!list_empty(&cookie->commit_link)) {
+		spin_lock(&fscache_cookie_lru_lock);
+		if (!list_empty(&cookie->commit_link)) {
+			list_del_init(&cookie->commit_link);
+			fscache_stat_d(&fscache_n_cookies_lru);
+			fscache_stat(&fscache_n_cookies_lru_dropped);
+			need_put = true;
+		}
+		spin_unlock(&fscache_cookie_lru_lock);
+		if (need_put)
+			fscache_put_cookie(cookie, fscache_cookie_put_lru);
+	}
+}
+
+/*
+ * Remove a cookie from the hash table.
+ */
+static void fscache_unhash_cookie(struct fscache_cookie *cookie)
+{
+	struct hlist_bl_head *h;
+	unsigned int bucket;
+
+	bucket = cookie->key_hash & (ARRAY_SIZE(fscache_cookie_hash) - 1);
+	h = &fscache_cookie_hash[bucket];
+
+	hlist_bl_lock(h);
+	hlist_bl_del(&cookie->hash_link);
+	clear_bit(FSCACHE_COOKIE_IS_HASHED, &cookie->flags);
+	hlist_bl_unlock(h);
+	fscache_stat(&fscache_n_relinquishes_dropped);
+}
+
+static void fscache_drop_withdraw_cookie(struct fscache_cookie *cookie)
+{
+	fscache_cookie_drop_from_lru(cookie);
+	__fscache_withdraw_cookie(cookie);
+}
+
+/**
+ * fscache_withdraw_cookie - Mark a cookie for withdrawal
+ * @cookie: The cookie to be withdrawn.
+ *
+ * Allow the cache backend to withdraw the backing for a cookie for its own
+ * reasons, even if that cookie is in active use.
+ */
+void fscache_withdraw_cookie(struct fscache_cookie *cookie)
+{
+	set_bit(FSCACHE_COOKIE_DO_WITHDRAW, &cookie->flags);
+	fscache_drop_withdraw_cookie(cookie);
+}
+EXPORT_SYMBOL(fscache_withdraw_cookie);
+
+/*
+ * Allow the netfs to release a cookie back to the cache.
+ * - the object will be marked as recyclable on disk if retire is true
+ */
+void __fscache_relinquish_cookie(struct fscache_cookie *cookie, bool retire)
+{
+	fscache_stat(&fscache_n_relinquishes);
+	if (retire)
+		fscache_stat(&fscache_n_relinquishes_retire);
+
+	_enter("c=%08x{%d},%d",
+	       cookie->debug_id, atomic_read(&cookie->n_active), retire);
+
+	if (WARN(test_and_set_bit(FSCACHE_COOKIE_RELINQUISHED, &cookie->flags),
+		 "Cookie c=%x already relinquished\n", cookie->debug_id))
+		return;
+
+	if (retire)
+		set_bit(FSCACHE_COOKIE_RETIRED, &cookie->flags);
+	trace_fscache_relinquish(cookie, retire);
+
+	ASSERTCMP(atomic_read(&cookie->n_active), ==, 0);
+	ASSERTCMP(atomic_read(&cookie->volume->n_cookies), >, 0);
+	atomic_dec(&cookie->volume->n_cookies);
+
+	if (test_bit(FSCACHE_COOKIE_HAS_BEEN_CACHED, &cookie->flags)) {
+		set_bit(FSCACHE_COOKIE_DO_RELINQUISH, &cookie->flags);
+		fscache_drop_withdraw_cookie(cookie);
+	} else {
+		fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_DROPPED);
+		fscache_unhash_cookie(cookie);
+	}
+	fscache_put_cookie(cookie, fscache_cookie_put_relinquish);
+}
+EXPORT_SYMBOL(__fscache_relinquish_cookie);
+
+/*
+ * Drop a reference to a cookie.
+ */
+void fscache_put_cookie(struct fscache_cookie *cookie,
+			enum fscache_cookie_trace where)
+{
+	struct fscache_volume *volume = cookie->volume;
+	unsigned int cookie_debug_id = cookie->debug_id;
+	bool zero;
+	int ref;
+
+	zero = __refcount_dec_and_test(&cookie->ref, &ref);
+	trace_fscache_cookie(cookie_debug_id, ref - 1, where);
+	if (zero) {
+		fscache_free_cookie(cookie);
+		fscache_put_volume(volume, fscache_volume_put_cookie);
+	}
+}
+EXPORT_SYMBOL(fscache_put_cookie);
+
+/*
+ * Get a reference to a cookie.
+ */
+struct fscache_cookie *fscache_get_cookie(struct fscache_cookie *cookie,
+					  enum fscache_cookie_trace where)
+{
+	int ref;
+
+	__refcount_inc(&cookie->ref, &ref);
+	trace_fscache_cookie(cookie->debug_id, ref + 1, where);
+	return cookie;
+}
+EXPORT_SYMBOL(fscache_get_cookie);
+
+/*
+ * Ask the cache to effect invalidation of a cookie.
+ */
+static void fscache_perform_invalidation(struct fscache_cookie *cookie)
+{
+	if (!cookie->volume->cache->ops->invalidate_cookie(cookie))
+		fscache_caching_failed(cookie);
+	fscache_end_cookie_access(cookie, fscache_access_invalidate_cookie_end);
+}
+
+/*
+ * Invalidate an object.
+ */
+void __fscache_invalidate(struct fscache_cookie *cookie,
+			  const void *aux_data, loff_t new_size,
+			  unsigned int flags)
+{
+	bool is_caching;
+
+	_enter("c=%x", cookie->debug_id);
+
+	fscache_stat(&fscache_n_invalidates);
+
+	if (WARN(test_bit(FSCACHE_COOKIE_RELINQUISHED, &cookie->flags),
+		 "Trying to invalidate relinquished cookie\n"))
+		return;
+
+	if ((flags & FSCACHE_INVAL_DIO_WRITE) &&
+	    test_and_set_bit(FSCACHE_COOKIE_DISABLED, &cookie->flags))
+		return;
+
+	spin_lock(&cookie->lock);
+	set_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
+	fscache_update_aux(cookie, aux_data, &new_size);
+	cookie->inval_counter++;
+	trace_fscache_invalidate(cookie, new_size);
+
+	switch (cookie->state) {
+	case FSCACHE_COOKIE_STATE_INVALIDATING: /* is_still_valid will catch it */
+	default:
+		spin_unlock(&cookie->lock);
+		_leave(" [no %u]", cookie->state);
+		return;
+
+	case FSCACHE_COOKIE_STATE_LOOKING_UP:
+		if (!test_and_set_bit(FSCACHE_COOKIE_DO_INVALIDATE, &cookie->flags))
+			__fscache_begin_cookie_access(cookie, fscache_access_invalidate_cookie);
+		fallthrough;
+	case FSCACHE_COOKIE_STATE_CREATING:
+		spin_unlock(&cookie->lock);
+		_leave(" [look %x]", cookie->inval_counter);
+		return;
+
+	case FSCACHE_COOKIE_STATE_ACTIVE:
+		is_caching = fscache_begin_cookie_access(
+			cookie, fscache_access_invalidate_cookie);
+		if (is_caching)
+			__fscache_set_cookie_state(cookie, FSCACHE_COOKIE_STATE_INVALIDATING);
+		spin_unlock(&cookie->lock);
+		wake_up_cookie_state(cookie);
+
+		if (is_caching)
+			fscache_queue_cookie(cookie, fscache_cookie_get_inval_work);
+		_leave(" [inv]");
+		return;
+	}
+}
+EXPORT_SYMBOL(__fscache_invalidate);
+
+#ifdef CONFIG_PROC_FS
+/*
+ * Generate a list of extant cookies in /proc/fs/fscache/cookies
+ */
+static int fscache_cookies_seq_show(struct seq_file *m, void *v)
+{
+	struct fscache_cookie *cookie;
+	unsigned int keylen = 0, auxlen = 0;
+	u8 *p;
+
+	if (v == &fscache_cookies) {
+		seq_puts(m,
+			 "COOKIE   VOLUME   REF ACT ACC S FL DEF             \n"
+			 "======== ======== === === === = == ================\n"
+			 );
+		return 0;
+	}
+
+	cookie = list_entry(v, struct fscache_cookie, proc_link);
+
+	seq_printf(m,
+		   "%08x %08x %3d %3d %3d %c %02lx",
+		   cookie->debug_id,
+		   cookie->volume->debug_id,
+		   refcount_read(&cookie->ref),
+		   atomic_read(&cookie->n_active),
+		   atomic_read(&cookie->n_accesses),
+		   fscache_cookie_states[cookie->state],
+		   cookie->flags);
+
+	keylen = cookie->key_len;
+	auxlen = cookie->aux_len;
+
+	if (keylen > 0 || auxlen > 0) {
+		seq_puts(m, " ");
+		p = keylen <= sizeof(cookie->inline_key) ?
+			cookie->inline_key : cookie->key;
+		for (; keylen > 0; keylen--)
+			seq_printf(m, "%02x", *p++);
+		if (auxlen > 0) {
+			seq_puts(m, ", ");
+			p = auxlen <= sizeof(cookie->inline_aux) ?
+				cookie->inline_aux : cookie->aux;
+			for (; auxlen > 0; auxlen--)
+				seq_printf(m, "%02x", *p++);
+		}
+	}
+
+	seq_puts(m, "\n");
+	return 0;
+}
+
+static void *fscache_cookies_seq_start(struct seq_file *m, loff_t *_pos)
+	__acquires(fscache_cookies_lock)
+{
+	read_lock(&fscache_cookies_lock);
+	return seq_list_start_head(&fscache_cookies, *_pos);
+}
+
+static void *fscache_cookies_seq_next(struct seq_file *m, void *v, loff_t *_pos)
+{
+	return seq_list_next(v, &fscache_cookies, _pos);
+}
+
+static void fscache_cookies_seq_stop(struct seq_file *m, void *v)
+	__releases(rcu)
+{
+	read_unlock(&fscache_cookies_lock);
+}
+
+
+const struct seq_operations fscache_cookies_seq_ops = {
+	.start  = fscache_cookies_seq_start,
+	.next   = fscache_cookies_seq_next,
+	.stop   = fscache_cookies_seq_stop,
+	.show   = fscache_cookies_seq_show,
+};
+#endif
diff --git a/fs/netfs/fscache_internal.h b/fs/netfs/fscache_internal.h
new file mode 100644
index 000000000..a09b948fc
--- /dev/null
+++ b/fs/netfs/fscache_internal.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* Internal definitions for FS-Cache
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include "internal.h"
+
+#ifdef pr_fmt
+#undef pr_fmt
+#endif
+
+#define pr_fmt(fmt) "FS-Cache: " fmt
diff --git a/fs/netfs/fscache_io.c b/fs/netfs/fscache_io.c
new file mode 100644
index 000000000..37f05b4d3
--- /dev/null
+++ b/fs/netfs/fscache_io.c
@@ -0,0 +1,290 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Cache data I/O routines
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+#define FSCACHE_DEBUG_LEVEL OPERATION
+#include <linux/fscache-cache.h>
+#include <linux/uio.h>
+#include <linux/bvec.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+/**
+ * fscache_wait_for_operation - Wait for an object become accessible
+ * @cres: The cache resources for the operation being performed
+ * @want_state: The minimum state the object must be at
+ *
+ * See if the target cache object is at the specified minimum state of
+ * accessibility yet, and if not, wait for it.
+ */
+bool fscache_wait_for_operation(struct netfs_cache_resources *cres,
+				enum fscache_want_state want_state)
+{
+	struct fscache_cookie *cookie = fscache_cres_cookie(cres);
+	enum fscache_cookie_state state;
+
+again:
+	if (!fscache_cache_is_live(cookie->volume->cache)) {
+		_leave(" [broken]");
+		return false;
+	}
+
+	state = fscache_cookie_state(cookie);
+	_enter("c=%08x{%u},%x", cookie->debug_id, state, want_state);
+
+	switch (state) {
+	case FSCACHE_COOKIE_STATE_CREATING:
+	case FSCACHE_COOKIE_STATE_INVALIDATING:
+		if (want_state == FSCACHE_WANT_PARAMS)
+			goto ready; /* There can be no content */
+		fallthrough;
+	case FSCACHE_COOKIE_STATE_LOOKING_UP:
+	case FSCACHE_COOKIE_STATE_LRU_DISCARDING:
+		wait_var_event(&cookie->state,
+			       fscache_cookie_state(cookie) != state);
+		goto again;
+
+	case FSCACHE_COOKIE_STATE_ACTIVE:
+		goto ready;
+	case FSCACHE_COOKIE_STATE_DROPPED:
+	case FSCACHE_COOKIE_STATE_RELINQUISHING:
+	default:
+		_leave(" [not live]");
+		return false;
+	}
+
+ready:
+	if (!cres->cache_priv2)
+		return cookie->volume->cache->ops->begin_operation(cres, want_state);
+	return true;
+}
+EXPORT_SYMBOL(fscache_wait_for_operation);
+
+/*
+ * Begin an I/O operation on the cache, waiting till we reach the right state.
+ *
+ * Attaches the resources required to the operation resources record.
+ */
+static int fscache_begin_operation(struct netfs_cache_resources *cres,
+				   struct fscache_cookie *cookie,
+				   enum fscache_want_state want_state,
+				   enum fscache_access_trace why)
+{
+	enum fscache_cookie_state state;
+	long timeo;
+	bool once_only = false;
+
+	cres->ops		= NULL;
+	cres->cache_priv	= cookie;
+	cres->cache_priv2	= NULL;
+	cres->debug_id		= cookie->debug_id;
+	cres->inval_counter	= cookie->inval_counter;
+
+	if (!fscache_begin_cookie_access(cookie, why)) {
+		cres->cache_priv = NULL;
+		return -ENOBUFS;
+	}
+
+again:
+	spin_lock(&cookie->lock);
+
+	state = fscache_cookie_state(cookie);
+	_enter("c=%08x{%u},%x", cookie->debug_id, state, want_state);
+
+	switch (state) {
+	case FSCACHE_COOKIE_STATE_LOOKING_UP:
+	case FSCACHE_COOKIE_STATE_LRU_DISCARDING:
+	case FSCACHE_COOKIE_STATE_INVALIDATING:
+		goto wait_for_file_wrangling;
+	case FSCACHE_COOKIE_STATE_CREATING:
+		if (want_state == FSCACHE_WANT_PARAMS)
+			goto ready; /* There can be no content */
+		goto wait_for_file_wrangling;
+	case FSCACHE_COOKIE_STATE_ACTIVE:
+		goto ready;
+	case FSCACHE_COOKIE_STATE_DROPPED:
+	case FSCACHE_COOKIE_STATE_RELINQUISHING:
+		WARN(1, "Can't use cookie in state %u\n", cookie->state);
+		goto not_live;
+	default:
+		goto not_live;
+	}
+
+ready:
+	spin_unlock(&cookie->lock);
+	if (!cookie->volume->cache->ops->begin_operation(cres, want_state))
+		goto failed;
+	return 0;
+
+wait_for_file_wrangling:
+	spin_unlock(&cookie->lock);
+	trace_fscache_access(cookie->debug_id, refcount_read(&cookie->ref),
+			     atomic_read(&cookie->n_accesses),
+			     fscache_access_io_wait);
+	timeo = wait_var_event_timeout(&cookie->state,
+				       fscache_cookie_state(cookie) != state, 20 * HZ);
+	if (timeo <= 1 && !once_only) {
+		pr_warn("%s: cookie state change wait timed out: cookie->state=%u state=%u",
+			__func__, fscache_cookie_state(cookie), state);
+		fscache_print_cookie(cookie, 'O');
+		once_only = true;
+	}
+	goto again;
+
+not_live:
+	spin_unlock(&cookie->lock);
+failed:
+	cres->cache_priv = NULL;
+	cres->ops = NULL;
+	fscache_end_cookie_access(cookie, fscache_access_io_not_live);
+	_leave(" = -ENOBUFS");
+	return -ENOBUFS;
+}
+
+int __fscache_begin_read_operation(struct netfs_cache_resources *cres,
+				   struct fscache_cookie *cookie)
+{
+	return fscache_begin_operation(cres, cookie, FSCACHE_WANT_PARAMS,
+				       fscache_access_io_read);
+}
+EXPORT_SYMBOL(__fscache_begin_read_operation);
+
+int __fscache_begin_write_operation(struct netfs_cache_resources *cres,
+				    struct fscache_cookie *cookie)
+{
+	return fscache_begin_operation(cres, cookie, FSCACHE_WANT_PARAMS,
+				       fscache_access_io_write);
+}
+EXPORT_SYMBOL(__fscache_begin_write_operation);
+
+struct fscache_write_request {
+	struct netfs_cache_resources cache_resources;
+	struct address_space	*mapping;
+	loff_t			start;
+	size_t			len;
+	bool			set_bits;
+	bool			using_pgpriv2;
+	netfs_io_terminated_t	term_func;
+	void			*term_func_priv;
+};
+
+void __fscache_clear_page_bits(struct address_space *mapping,
+			       loff_t start, size_t len)
+{
+	pgoff_t first = start / PAGE_SIZE;
+	pgoff_t last = (start + len - 1) / PAGE_SIZE;
+	struct page *page;
+
+	if (len) {
+		XA_STATE(xas, &mapping->i_pages, first);
+
+		rcu_read_lock();
+		xas_for_each(&xas, page, last) {
+			folio_end_private_2(page_folio(page));
+		}
+		rcu_read_unlock();
+	}
+}
+EXPORT_SYMBOL(__fscache_clear_page_bits);
+
+/*
+ * Deal with the completion of writing the data to the cache.
+ */
+static void fscache_wreq_done(void *priv, ssize_t transferred_or_error)
+{
+	struct fscache_write_request *wreq = priv;
+
+	if (wreq->using_pgpriv2)
+		fscache_clear_page_bits(wreq->mapping, wreq->start, wreq->len,
+					wreq->set_bits);
+
+	if (wreq->term_func)
+		wreq->term_func(wreq->term_func_priv, transferred_or_error);
+	fscache_end_operation(&wreq->cache_resources);
+	kfree(wreq);
+}
+
+void __fscache_write_to_cache(struct fscache_cookie *cookie,
+			      struct address_space *mapping,
+			      loff_t start, size_t len, loff_t i_size,
+			      netfs_io_terminated_t term_func,
+			      void *term_func_priv,
+			      bool using_pgpriv2, bool cond)
+{
+	struct fscache_write_request *wreq;
+	struct netfs_cache_resources *cres;
+	struct iov_iter iter;
+	int ret = -ENOBUFS;
+
+	if (len == 0)
+		goto abandon;
+
+	_enter("%llx,%zx", start, len);
+
+	wreq = kzalloc_obj(struct fscache_write_request, GFP_NOFS);
+	if (!wreq)
+		goto abandon;
+	wreq->mapping		= mapping;
+	wreq->start		= start;
+	wreq->len		= len;
+	wreq->using_pgpriv2	= using_pgpriv2;
+	wreq->set_bits		= cond;
+	wreq->term_func		= term_func;
+	wreq->term_func_priv	= term_func_priv;
+
+	cres = &wreq->cache_resources;
+	if (fscache_begin_operation(cres, cookie, FSCACHE_WANT_WRITE,
+				    fscache_access_io_write) < 0)
+		goto abandon_free;
+
+	ret = cres->ops->prepare_write(cres, &start, &len, len, i_size, false);
+	if (ret < 0)
+		goto abandon_end;
+
+	/* TODO: Consider clearing page bits now for space the write isn't
+	 * covering.  This is more complicated than it appears when THPs are
+	 * taken into account.
+	 */
+
+	iov_iter_xarray(&iter, ITER_SOURCE, &mapping->i_pages, start, len);
+	fscache_write(cres, start, &iter, fscache_wreq_done, wreq);
+	return;
+
+abandon_end:
+	return fscache_wreq_done(wreq, ret);
+abandon_free:
+	kfree(wreq);
+abandon:
+	if (using_pgpriv2)
+		fscache_clear_page_bits(mapping, start, len, cond);
+	if (term_func)
+		term_func(term_func_priv, ret);
+}
+EXPORT_SYMBOL(__fscache_write_to_cache);
+
+/*
+ * Change the size of a backing object.
+ */
+void __fscache_resize_cookie(struct fscache_cookie *cookie, loff_t new_size)
+{
+	struct netfs_cache_resources cres;
+
+	trace_fscache_resize(cookie, new_size);
+	if (fscache_begin_operation(&cres, cookie, FSCACHE_WANT_WRITE,
+				    fscache_access_io_resize) == 0) {
+		fscache_stat(&fscache_n_resizes);
+		set_bit(FSCACHE_COOKIE_NEEDS_UPDATE, &cookie->flags);
+
+		/* We cannot defer a resize as we need to do it inside the
+		 * netfs's inode lock so that we're serialised with respect to
+		 * writes.
+		 */
+		cookie->volume->cache->ops->resize_cookie(&cres, new_size);
+		fscache_end_operation(&cres);
+	} else {
+		fscache_stat(&fscache_n_resizes_null);
+	}
+}
+EXPORT_SYMBOL(__fscache_resize_cookie);
diff --git a/fs/netfs/fscache_main.c b/fs/netfs/fscache_main.c
new file mode 100644
index 000000000..49849005e
--- /dev/null
+++ b/fs/netfs/fscache_main.c
@@ -0,0 +1,109 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* General filesystem local caching manager
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define FSCACHE_DEBUG_LEVEL CACHE
+#include <linux/module.h>
+#include <linux/init.h>
+#include "internal.h"
+#define CREATE_TRACE_POINTS
+#include <trace/events/fscache.h>
+
+EXPORT_TRACEPOINT_SYMBOL(fscache_access_cache);
+EXPORT_TRACEPOINT_SYMBOL(fscache_access_volume);
+EXPORT_TRACEPOINT_SYMBOL(fscache_access);
+
+struct workqueue_struct *fscache_wq;
+EXPORT_SYMBOL(fscache_wq);
+
+/*
+ * Mixing scores (in bits) for (7,20):
+ * Input delta: 1-bit      2-bit
+ * 1 round:     330.3     9201.6
+ * 2 rounds:   1246.4    25475.4
+ * 3 rounds:   1907.1    31295.1
+ * 4 rounds:   2042.3    31718.6
+ * Perfect:    2048      31744
+ *            (32*64)   (32*31/2 * 64)
+ */
+#define HASH_MIX(x, y, a)	\
+	(	x ^= (a),	\
+	y ^= x,	x = rol32(x, 7),\
+	x += y,	y = rol32(y,20),\
+	y *= 9			)
+
+static inline unsigned int fold_hash(unsigned long x, unsigned long y)
+{
+	/* Use arch-optimized multiply if one exists */
+	return __hash_32(y ^ __hash_32(x));
+}
+
+/*
+ * Generate a hash.  This is derived from full_name_hash(), but we want to be
+ * sure it is arch independent and that it doesn't change as bits of the
+ * computed hash value might appear on disk.  The caller must guarantee that
+ * the source data is a multiple of four bytes in size.
+ */
+unsigned int fscache_hash(unsigned int salt, const void *data, size_t len)
+{
+	const __le32 *p = data;
+	unsigned int a, x = 0, y = salt, n = len / sizeof(__le32);
+
+	for (; n; n--) {
+		a = le32_to_cpu(*p++);
+		HASH_MIX(x, y, a);
+	}
+	return fold_hash(x, y);
+}
+
+/*
+ * initialise the fs caching module
+ */
+int __init fscache_init(void)
+{
+	int ret = -ENOMEM;
+
+	fscache_wq = alloc_workqueue("fscache", WQ_UNBOUND | WQ_FREEZABLE, 0);
+	if (!fscache_wq)
+		goto error_wq;
+
+	ret = fscache_proc_init();
+	if (ret < 0)
+		goto error_proc;
+
+	fscache_cookie_jar = kmem_cache_create("fscache_cookie_jar",
+					       sizeof(struct fscache_cookie),
+					       0, 0, NULL);
+	if (!fscache_cookie_jar) {
+		pr_notice("Failed to allocate a cookie jar\n");
+		ret = -ENOMEM;
+		goto error_cookie_jar;
+	}
+
+	pr_notice("FS-Cache loaded\n");
+	return 0;
+
+error_cookie_jar:
+	fscache_proc_cleanup();
+error_proc:
+	destroy_workqueue(fscache_wq);
+error_wq:
+	return ret;
+}
+
+/*
+ * clean up on module removal
+ */
+void __exit fscache_exit(void)
+{
+	_enter("");
+
+	kmem_cache_destroy(fscache_cookie_jar);
+	fscache_proc_cleanup();
+	timer_shutdown_sync(&fscache_cookie_lru_timer);
+	destroy_workqueue(fscache_wq);
+	pr_notice("FS-Cache unloaded\n");
+}
diff --git a/fs/netfs/fscache_proc.c b/fs/netfs/fscache_proc.c
new file mode 100644
index 000000000..874d951bc
--- /dev/null
+++ b/fs/netfs/fscache_proc.c
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* FS-Cache statistics viewing interface
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define FSCACHE_DEBUG_LEVEL CACHE
+#include <linux/module.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include "internal.h"
+
+/*
+ * Add files to /proc/fs/netfs/.
+ */
+int __init fscache_proc_init(void)
+{
+	if (!proc_symlink("fs/fscache", NULL, "netfs"))
+		goto error_sym;
+
+	if (!proc_create_seq("fs/netfs/caches", S_IFREG | 0444, NULL,
+			     &fscache_caches_seq_ops))
+		goto error;
+
+	if (!proc_create_seq("fs/netfs/volumes", S_IFREG | 0444, NULL,
+			     &fscache_volumes_seq_ops))
+		goto error;
+
+	if (!proc_create_seq("fs/netfs/cookies", S_IFREG | 0444, NULL,
+			     &fscache_cookies_seq_ops))
+		goto error;
+	return 0;
+
+error:
+	remove_proc_entry("fs/fscache", NULL);
+error_sym:
+	return -ENOMEM;
+}
+
+/*
+ * Clean up the /proc/fs/fscache symlink.
+ */
+void fscache_proc_cleanup(void)
+{
+	remove_proc_subtree("fs/fscache", NULL);
+}
diff --git a/fs/netfs/fscache_stats.c b/fs/netfs/fscache_stats.c
new file mode 100644
index 000000000..add21abdf
--- /dev/null
+++ b/fs/netfs/fscache_stats.c
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* FS-Cache statistics
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define FSCACHE_DEBUG_LEVEL CACHE
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include "internal.h"
+
+/*
+ * operation counters
+ */
+atomic_t fscache_n_volumes;
+atomic_t fscache_n_volumes_collision;
+atomic_t fscache_n_volumes_nomem;
+atomic_t fscache_n_cookies;
+atomic_t fscache_n_cookies_lru;
+atomic_t fscache_n_cookies_lru_expired;
+atomic_t fscache_n_cookies_lru_removed;
+atomic_t fscache_n_cookies_lru_dropped;
+
+atomic_t fscache_n_acquires;
+atomic_t fscache_n_acquires_ok;
+atomic_t fscache_n_acquires_oom;
+
+atomic_t fscache_n_invalidates;
+
+atomic_t fscache_n_updates;
+EXPORT_SYMBOL(fscache_n_updates);
+
+atomic_t fscache_n_relinquishes;
+atomic_t fscache_n_relinquishes_retire;
+atomic_t fscache_n_relinquishes_dropped;
+
+atomic_t fscache_n_resizes;
+atomic_t fscache_n_resizes_null;
+
+atomic_t fscache_n_read;
+EXPORT_SYMBOL(fscache_n_read);
+atomic_t fscache_n_write;
+EXPORT_SYMBOL(fscache_n_write);
+atomic_t fscache_n_no_write_space;
+EXPORT_SYMBOL(fscache_n_no_write_space);
+atomic_t fscache_n_no_create_space;
+EXPORT_SYMBOL(fscache_n_no_create_space);
+atomic_t fscache_n_culled;
+EXPORT_SYMBOL(fscache_n_culled);
+atomic_t fscache_n_dio_misfit;
+EXPORT_SYMBOL(fscache_n_dio_misfit);
+
+/*
+ * display the general statistics
+ */
+int fscache_stats_show(struct seq_file *m)
+{
+	seq_puts(m, "-- FS-Cache statistics --\n");
+	seq_printf(m, "Cookies: n=%d v=%d vcol=%u voom=%u\n",
+		   atomic_read(&fscache_n_cookies),
+		   atomic_read(&fscache_n_volumes),
+		   atomic_read(&fscache_n_volumes_collision),
+		   atomic_read(&fscache_n_volumes_nomem)
+		   );
+
+	seq_printf(m, "Acquire: n=%u ok=%u oom=%u\n",
+		   atomic_read(&fscache_n_acquires),
+		   atomic_read(&fscache_n_acquires_ok),
+		   atomic_read(&fscache_n_acquires_oom));
+
+	seq_printf(m, "LRU    : n=%u exp=%u rmv=%u drp=%u at=%ld\n",
+		   atomic_read(&fscache_n_cookies_lru),
+		   atomic_read(&fscache_n_cookies_lru_expired),
+		   atomic_read(&fscache_n_cookies_lru_removed),
+		   atomic_read(&fscache_n_cookies_lru_dropped),
+		   timer_pending(&fscache_cookie_lru_timer) ?
+		   fscache_cookie_lru_timer.expires - jiffies : 0);
+
+	seq_printf(m, "Invals : n=%u\n",
+		   atomic_read(&fscache_n_invalidates));
+
+	seq_printf(m, "Updates: n=%u rsz=%u rsn=%u\n",
+		   atomic_read(&fscache_n_updates),
+		   atomic_read(&fscache_n_resizes),
+		   atomic_read(&fscache_n_resizes_null));
+
+	seq_printf(m, "Relinqs: n=%u rtr=%u drop=%u\n",
+		   atomic_read(&fscache_n_relinquishes),
+		   atomic_read(&fscache_n_relinquishes_retire),
+		   atomic_read(&fscache_n_relinquishes_dropped));
+
+	seq_printf(m, "NoSpace: nwr=%u ncr=%u cull=%u\n",
+		   atomic_read(&fscache_n_no_write_space),
+		   atomic_read(&fscache_n_no_create_space),
+		   atomic_read(&fscache_n_culled));
+
+	seq_printf(m, "IO     : rd=%u wr=%u mis=%u\n",
+		   atomic_read(&fscache_n_read),
+		   atomic_read(&fscache_n_write),
+		   atomic_read(&fscache_n_dio_misfit));
+	return 0;
+}
diff --git a/fs/netfs/fscache_volume.c b/fs/netfs/fscache_volume.c
new file mode 100644
index 000000000..78412fb85
--- /dev/null
+++ b/fs/netfs/fscache_volume.c
@@ -0,0 +1,532 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Volume-level cache cookie handling.
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#define FSCACHE_DEBUG_LEVEL COOKIE
+#include <linux/export.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+#define fscache_volume_hash_shift 10
+static struct hlist_bl_head fscache_volume_hash[1 << fscache_volume_hash_shift];
+static atomic_t fscache_volume_debug_id;
+static LIST_HEAD(fscache_volumes);
+
+static void fscache_create_volume_work(struct work_struct *work);
+
+struct fscache_volume *fscache_get_volume(struct fscache_volume *volume,
+					  enum fscache_volume_trace where)
+{
+	int ref;
+
+	__refcount_inc(&volume->ref, &ref);
+	trace_fscache_volume(volume->debug_id, ref + 1, where);
+	return volume;
+}
+
+struct fscache_volume *fscache_try_get_volume(struct fscache_volume *volume,
+					      enum fscache_volume_trace where)
+{
+	int ref;
+
+	if (!__refcount_inc_not_zero(&volume->ref, &ref))
+		return NULL;
+
+	trace_fscache_volume(volume->debug_id, ref + 1, where);
+	return volume;
+}
+EXPORT_SYMBOL(fscache_try_get_volume);
+
+static void fscache_see_volume(struct fscache_volume *volume,
+			       enum fscache_volume_trace where)
+{
+	int ref = refcount_read(&volume->ref);
+
+	trace_fscache_volume(volume->debug_id, ref, where);
+}
+
+/*
+ * Pin the cache behind a volume so that we can access it.
+ */
+static void __fscache_begin_volume_access(struct fscache_volume *volume,
+					  struct fscache_cookie *cookie,
+					  enum fscache_access_trace why)
+{
+	int n_accesses;
+
+	n_accesses = atomic_inc_return(&volume->n_accesses);
+	smp_mb__after_atomic();
+	trace_fscache_access_volume(volume->debug_id, cookie ? cookie->debug_id : 0,
+				    refcount_read(&volume->ref),
+				    n_accesses, why);
+}
+
+/**
+ * fscache_begin_volume_access - Pin a cache so a volume can be accessed
+ * @volume: The volume cookie
+ * @cookie: A datafile cookie for a tracing reference (or NULL)
+ * @why: An indication of the circumstances of the access for tracing
+ *
+ * Attempt to pin the cache to prevent it from going away whilst we're
+ * accessing a volume and returns true if successful.  This works as follows:
+ *
+ *  (1) If the cache tests as not live (state is not FSCACHE_CACHE_IS_ACTIVE),
+ *      then we return false to indicate access was not permitted.
+ *
+ *  (2) If the cache tests as live, then we increment the volume's n_accesses
+ *      count and then recheck the cache liveness, ending the access if it
+ *      ceased to be live.
+ *
+ *  (3) When we end the access, we decrement the volume's n_accesses and wake
+ *      up the any waiters if it reaches 0.
+ *
+ *  (4) Whilst the cache is caching, the volume's n_accesses is kept
+ *      artificially incremented to prevent wakeups from happening.
+ *
+ *  (5) When the cache is taken offline, the state is changed to prevent new
+ *      accesses, the volume's n_accesses is decremented and we wait for it to
+ *      become 0.
+ *
+ * The datafile @cookie and the @why indicator are merely provided for tracing
+ * purposes.
+ */
+bool fscache_begin_volume_access(struct fscache_volume *volume,
+				 struct fscache_cookie *cookie,
+				 enum fscache_access_trace why)
+{
+	if (!fscache_cache_is_live(volume->cache))
+		return false;
+	__fscache_begin_volume_access(volume, cookie, why);
+	if (!fscache_cache_is_live(volume->cache)) {
+		fscache_end_volume_access(volume, cookie, fscache_access_unlive);
+		return false;
+	}
+	return true;
+}
+
+/**
+ * fscache_end_volume_access - Unpin a cache at the end of an access.
+ * @volume: The volume cookie
+ * @cookie: A datafile cookie for a tracing reference (or NULL)
+ * @why: An indication of the circumstances of the access for tracing
+ *
+ * Unpin a cache volume after we've accessed it.  The datafile @cookie and the
+ * @why indicator are merely provided for tracing purposes.
+ */
+void fscache_end_volume_access(struct fscache_volume *volume,
+			       struct fscache_cookie *cookie,
+			       enum fscache_access_trace why)
+{
+	int n_accesses;
+
+	smp_mb__before_atomic();
+	n_accesses = atomic_dec_return(&volume->n_accesses);
+	trace_fscache_access_volume(volume->debug_id, cookie ? cookie->debug_id : 0,
+				    refcount_read(&volume->ref),
+				    n_accesses, why);
+	if (n_accesses == 0)
+		wake_up_var(&volume->n_accesses);
+}
+EXPORT_SYMBOL(fscache_end_volume_access);
+
+static bool fscache_volume_same(const struct fscache_volume *a,
+				const struct fscache_volume *b)
+{
+	size_t klen;
+
+	if (a->key_hash	!= b->key_hash ||
+	    a->cache	!= b->cache ||
+	    a->key[0]	!= b->key[0])
+		return false;
+
+	klen = round_up(a->key[0] + 1, sizeof(__le32));
+	return memcmp(a->key, b->key, klen) == 0;
+}
+
+static bool fscache_is_acquire_pending(struct fscache_volume *volume)
+{
+	return test_bit(FSCACHE_VOLUME_ACQUIRE_PENDING, &volume->flags);
+}
+
+static void fscache_wait_on_volume_collision(struct fscache_volume *candidate,
+					     unsigned int collidee_debug_id)
+{
+	wait_on_bit_timeout(&candidate->flags, FSCACHE_VOLUME_ACQUIRE_PENDING,
+			    TASK_UNINTERRUPTIBLE, 20 * HZ);
+	if (fscache_is_acquire_pending(candidate)) {
+		pr_notice("Potential volume collision new=%08x old=%08x",
+			  candidate->debug_id, collidee_debug_id);
+		fscache_stat(&fscache_n_volumes_collision);
+		wait_on_bit(&candidate->flags, FSCACHE_VOLUME_ACQUIRE_PENDING,
+			    TASK_UNINTERRUPTIBLE);
+	}
+}
+
+/*
+ * Attempt to insert the new volume into the hash.  If there's a collision, we
+ * wait for the old volume to complete if it's being relinquished and an error
+ * otherwise.
+ */
+static bool fscache_hash_volume(struct fscache_volume *candidate)
+{
+	struct fscache_volume *cursor;
+	struct hlist_bl_head *h;
+	struct hlist_bl_node *p;
+	unsigned int bucket, collidee_debug_id = 0;
+
+	bucket = candidate->key_hash & (ARRAY_SIZE(fscache_volume_hash) - 1);
+	h = &fscache_volume_hash[bucket];
+
+	hlist_bl_lock(h);
+	hlist_bl_for_each_entry(cursor, p, h, hash_link) {
+		if (fscache_volume_same(candidate, cursor)) {
+			if (!test_bit(FSCACHE_VOLUME_RELINQUISHED, &cursor->flags))
+				goto collision;
+			fscache_see_volume(cursor, fscache_volume_get_hash_collision);
+			set_bit(FSCACHE_VOLUME_COLLIDED_WITH, &cursor->flags);
+			set_bit(FSCACHE_VOLUME_ACQUIRE_PENDING, &candidate->flags);
+			collidee_debug_id = cursor->debug_id;
+			break;
+		}
+	}
+
+	hlist_bl_add_head(&candidate->hash_link, h);
+	hlist_bl_unlock(h);
+
+	if (fscache_is_acquire_pending(candidate))
+		fscache_wait_on_volume_collision(candidate, collidee_debug_id);
+	return true;
+
+collision:
+	fscache_see_volume(cursor, fscache_volume_collision);
+	hlist_bl_unlock(h);
+	return false;
+}
+
+/*
+ * Allocate and initialise a volume representation cookie.
+ */
+static struct fscache_volume *fscache_alloc_volume(const char *volume_key,
+						   const char *cache_name,
+						   const void *coherency_data,
+						   size_t coherency_len)
+{
+	struct fscache_volume *volume;
+	struct fscache_cache *cache;
+	size_t klen, hlen;
+	u8 *key;
+
+	klen = strlen(volume_key);
+	if (klen > NAME_MAX)
+		return NULL;
+
+	if (!coherency_data)
+		coherency_len = 0;
+
+	cache = fscache_lookup_cache(cache_name, false);
+	if (IS_ERR(cache))
+		return NULL;
+
+	volume = kzalloc_flex(*volume, coherency, coherency_len);
+	if (!volume)
+		goto err_cache;
+
+	volume->cache = cache;
+	volume->coherency_len = coherency_len;
+	if (coherency_data)
+		memcpy(volume->coherency, coherency_data, coherency_len);
+	INIT_LIST_HEAD(&volume->proc_link);
+	INIT_WORK(&volume->work, fscache_create_volume_work);
+	refcount_set(&volume->ref, 1);
+	spin_lock_init(&volume->lock);
+
+	/* Stick the length on the front of the key and pad it out to make
+	 * hashing easier.
+	 */
+	hlen = round_up(1 + klen + 1, sizeof(__le32));
+	key = kzalloc(hlen, GFP_KERNEL);
+	if (!key)
+		goto err_vol;
+	key[0] = klen;
+	memcpy(key + 1, volume_key, klen);
+
+	volume->key = key;
+	volume->key_hash = fscache_hash(0, key, hlen);
+
+	volume->debug_id = atomic_inc_return(&fscache_volume_debug_id);
+	down_write(&fscache_addremove_sem);
+	atomic_inc(&cache->n_volumes);
+	list_add_tail(&volume->proc_link, &fscache_volumes);
+	fscache_see_volume(volume, fscache_volume_new_acquire);
+	fscache_stat(&fscache_n_volumes);
+	up_write(&fscache_addremove_sem);
+	_leave(" = v=%x", volume->debug_id);
+	return volume;
+
+err_vol:
+	kfree(volume);
+err_cache:
+	fscache_put_cache(cache, fscache_cache_put_alloc_volume);
+	fscache_stat(&fscache_n_volumes_nomem);
+	return NULL;
+}
+
+/*
+ * Create a volume's representation on disk.  Have a volume ref and a cache
+ * access we have to release.
+ */
+static void fscache_create_volume_work(struct work_struct *work)
+{
+	const struct fscache_cache_ops *ops;
+	struct fscache_volume *volume =
+		container_of(work, struct fscache_volume, work);
+
+	fscache_see_volume(volume, fscache_volume_see_create_work);
+
+	ops = volume->cache->ops;
+	if (ops->acquire_volume)
+		ops->acquire_volume(volume);
+	fscache_end_cache_access(volume->cache,
+				 fscache_access_acquire_volume_end);
+
+	clear_and_wake_up_bit(FSCACHE_VOLUME_CREATING, &volume->flags);
+	fscache_put_volume(volume, fscache_volume_put_create_work);
+}
+
+/*
+ * Dispatch a worker thread to create a volume's representation on disk.
+ */
+void fscache_create_volume(struct fscache_volume *volume, bool wait)
+{
+	if (test_and_set_bit(FSCACHE_VOLUME_CREATING, &volume->flags))
+		goto maybe_wait;
+	if (volume->cache_priv)
+		goto no_wait; /* We raced */
+	if (!fscache_begin_cache_access(volume->cache,
+					fscache_access_acquire_volume))
+		goto no_wait;
+
+	fscache_get_volume(volume, fscache_volume_get_create_work);
+	if (!schedule_work(&volume->work))
+		fscache_put_volume(volume, fscache_volume_put_create_work);
+
+maybe_wait:
+	if (wait) {
+		fscache_see_volume(volume, fscache_volume_wait_create_work);
+		wait_on_bit(&volume->flags, FSCACHE_VOLUME_CREATING,
+			    TASK_UNINTERRUPTIBLE);
+	}
+	return;
+no_wait:
+	clear_and_wake_up_bit(FSCACHE_VOLUME_CREATING, &volume->flags);
+}
+
+/*
+ * Acquire a volume representation cookie and link it to a (proposed) cache.
+ */
+struct fscache_volume *__fscache_acquire_volume(const char *volume_key,
+						const char *cache_name,
+						const void *coherency_data,
+						size_t coherency_len)
+{
+	struct fscache_volume *volume;
+
+	volume = fscache_alloc_volume(volume_key, cache_name,
+				      coherency_data, coherency_len);
+	if (!volume)
+		return ERR_PTR(-ENOMEM);
+
+	if (!fscache_hash_volume(volume)) {
+		fscache_put_volume(volume, fscache_volume_put_hash_collision);
+		return ERR_PTR(-EBUSY);
+	}
+
+	fscache_create_volume(volume, false);
+	return volume;
+}
+EXPORT_SYMBOL(__fscache_acquire_volume);
+
+static void fscache_wake_pending_volume(struct fscache_volume *volume,
+					struct hlist_bl_head *h)
+{
+	struct fscache_volume *cursor;
+	struct hlist_bl_node *p;
+
+	hlist_bl_for_each_entry(cursor, p, h, hash_link) {
+		if (fscache_volume_same(cursor, volume)) {
+			fscache_see_volume(cursor, fscache_volume_see_hash_wake);
+			clear_and_wake_up_bit(FSCACHE_VOLUME_ACQUIRE_PENDING,
+					      &cursor->flags);
+			return;
+		}
+	}
+}
+
+/*
+ * Remove a volume cookie from the hash table.
+ */
+static void fscache_unhash_volume(struct fscache_volume *volume)
+{
+	struct hlist_bl_head *h;
+	unsigned int bucket;
+
+	bucket = volume->key_hash & (ARRAY_SIZE(fscache_volume_hash) - 1);
+	h = &fscache_volume_hash[bucket];
+
+	hlist_bl_lock(h);
+	hlist_bl_del(&volume->hash_link);
+	if (test_bit(FSCACHE_VOLUME_COLLIDED_WITH, &volume->flags))
+		fscache_wake_pending_volume(volume, h);
+	hlist_bl_unlock(h);
+}
+
+/*
+ * Drop a cache's volume attachments.
+ */
+static void fscache_free_volume(struct fscache_volume *volume)
+{
+	struct fscache_cache *cache = volume->cache;
+
+	if (volume->cache_priv) {
+		__fscache_begin_volume_access(volume, NULL,
+					      fscache_access_relinquish_volume);
+		if (volume->cache_priv)
+			cache->ops->free_volume(volume);
+		fscache_end_volume_access(volume, NULL,
+					  fscache_access_relinquish_volume_end);
+	}
+
+	down_write(&fscache_addremove_sem);
+	list_del_init(&volume->proc_link);
+	atomic_dec(&volume->cache->n_volumes);
+	up_write(&fscache_addremove_sem);
+
+	if (!hlist_bl_unhashed(&volume->hash_link))
+		fscache_unhash_volume(volume);
+
+	trace_fscache_volume(volume->debug_id, 0, fscache_volume_free);
+	kfree(volume->key);
+	kfree(volume);
+	fscache_stat_d(&fscache_n_volumes);
+	fscache_put_cache(cache, fscache_cache_put_volume);
+}
+
+/*
+ * Drop a reference to a volume cookie.
+ */
+void fscache_put_volume(struct fscache_volume *volume,
+			enum fscache_volume_trace where)
+{
+	if (volume) {
+		unsigned int debug_id = volume->debug_id;
+		bool zero;
+		int ref;
+
+		zero = __refcount_dec_and_test(&volume->ref, &ref);
+		trace_fscache_volume(debug_id, ref - 1, where);
+		if (zero)
+			fscache_free_volume(volume);
+	}
+}
+EXPORT_SYMBOL(fscache_put_volume);
+
+/*
+ * Relinquish a volume representation cookie.
+ */
+void __fscache_relinquish_volume(struct fscache_volume *volume,
+				 const void *coherency_data,
+				 bool invalidate)
+{
+	if (WARN_ON(test_and_set_bit(FSCACHE_VOLUME_RELINQUISHED, &volume->flags)))
+		return;
+
+	if (invalidate) {
+		set_bit(FSCACHE_VOLUME_INVALIDATE, &volume->flags);
+	} else if (coherency_data) {
+		memcpy(volume->coherency, coherency_data, volume->coherency_len);
+	}
+
+	fscache_put_volume(volume, fscache_volume_put_relinquish);
+}
+EXPORT_SYMBOL(__fscache_relinquish_volume);
+
+/**
+ * fscache_withdraw_volume - Withdraw a volume from being cached
+ * @volume: Volume cookie
+ *
+ * Withdraw a cache volume from service, waiting for all accesses to complete
+ * before returning.
+ */
+void fscache_withdraw_volume(struct fscache_volume *volume)
+{
+	int n_accesses;
+
+	_debug("withdraw V=%x", volume->debug_id);
+
+	/* Allow wakeups on dec-to-0 */
+	n_accesses = atomic_dec_return(&volume->n_accesses);
+	trace_fscache_access_volume(volume->debug_id, 0,
+				    refcount_read(&volume->ref),
+				    n_accesses, fscache_access_cache_unpin);
+
+	wait_var_event(&volume->n_accesses,
+		       atomic_read(&volume->n_accesses) == 0);
+}
+EXPORT_SYMBOL(fscache_withdraw_volume);
+
+#ifdef CONFIG_PROC_FS
+/*
+ * Generate a list of volumes in /proc/fs/fscache/volumes
+ */
+static int fscache_volumes_seq_show(struct seq_file *m, void *v)
+{
+	struct fscache_volume *volume;
+
+	if (v == &fscache_volumes) {
+		seq_puts(m,
+			 "VOLUME   REF   nCOOK ACC FL CACHE           KEY\n"
+			 "======== ===== ===== === == =============== ================\n");
+		return 0;
+	}
+
+	volume = list_entry(v, struct fscache_volume, proc_link);
+	seq_printf(m,
+		   "%08x %5d %5d %3d %02lx %-15.15s %s\n",
+		   volume->debug_id,
+		   refcount_read(&volume->ref),
+		   atomic_read(&volume->n_cookies),
+		   atomic_read(&volume->n_accesses),
+		   volume->flags,
+		   volume->cache->name ?: "-",
+		   volume->key + 1);
+	return 0;
+}
+
+static void *fscache_volumes_seq_start(struct seq_file *m, loff_t *_pos)
+	__acquires(&fscache_addremove_sem)
+{
+	down_read(&fscache_addremove_sem);
+	return seq_list_start_head(&fscache_volumes, *_pos);
+}
+
+static void *fscache_volumes_seq_next(struct seq_file *m, void *v, loff_t *_pos)
+{
+	return seq_list_next(v, &fscache_volumes, _pos);
+}
+
+static void fscache_volumes_seq_stop(struct seq_file *m, void *v)
+	__releases(&fscache_addremove_sem)
+{
+	up_read(&fscache_addremove_sem);
+}
+
+const struct seq_operations fscache_volumes_seq_ops = {
+	.start  = fscache_volumes_seq_start,
+	.next   = fscache_volumes_seq_next,
+	.stop   = fscache_volumes_seq_stop,
+	.show   = fscache_volumes_seq_show,
+};
+#endif /* CONFIG_PROC_FS */
diff --git a/fs/netfs/internal.h b/fs/netfs/internal.h
new file mode 100644
index 000000000..420ee7b26
--- /dev/null
+++ b/fs/netfs/internal.h
@@ -0,0 +1,553 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* Internal definitions for network filesystem support
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/slab.h>
+#include <linux/seq_file.h>
+#include <linux/folio_queue.h>
+#include <linux/netfs.h>
+#include <linux/fscache.h>
+#include <linux/fscache-cache.h>
+#include <trace/events/netfs.h>
+#include <trace/events/fscache.h>
+
+#ifdef pr_fmt
+#undef pr_fmt
+#endif
+
+#define pr_fmt(fmt) "netfs: " fmt
+
+/*
+ * buffered_read.c
+ */
+void netfs_queue_read(struct netfs_io_request *rreq,
+		      struct netfs_io_subrequest *subreq);
+void netfs_cache_read_terminated(void *priv, ssize_t transferred_or_error);
+int netfs_prefetch_for_write(struct file *file, struct folio *folio,
+			     size_t offset, size_t len);
+
+/*
+ * buffered_write.c
+ */
+void netfs_update_i_size(struct netfs_inode *ctx, struct inode *inode,
+			 loff_t pos, size_t copied);
+
+/*
+ * main.c
+ */
+extern unsigned int netfs_debug;
+extern struct list_head netfs_io_requests;
+extern spinlock_t netfs_proc_lock;
+extern mempool_t netfs_request_pool;
+extern mempool_t netfs_subrequest_pool;
+extern mempool_t netfs_folioq_pool;
+
+#ifdef CONFIG_PROC_FS
+static inline void netfs_proc_add_rreq(struct netfs_io_request *rreq)
+{
+	spin_lock(&netfs_proc_lock);
+	list_add_tail_rcu(&rreq->proc_link, &netfs_io_requests);
+	spin_unlock(&netfs_proc_lock);
+}
+static inline void netfs_proc_del_rreq(struct netfs_io_request *rreq)
+{
+	if (!list_empty(&rreq->proc_link)) {
+		spin_lock(&netfs_proc_lock);
+		list_del_rcu(&rreq->proc_link);
+		spin_unlock(&netfs_proc_lock);
+	}
+}
+#else
+static inline void netfs_proc_add_rreq(struct netfs_io_request *rreq) {}
+static inline void netfs_proc_del_rreq(struct netfs_io_request *rreq) {}
+#endif
+
+/*
+ * misc.c
+ */
+struct folio_queue *netfs_buffer_make_space(struct netfs_io_request *rreq,
+					    enum netfs_folioq_trace trace);
+void netfs_reset_iter(struct netfs_io_subrequest *subreq);
+void netfs_wake_collector(struct netfs_io_request *rreq);
+void netfs_subreq_clear_in_progress(struct netfs_io_subrequest *subreq);
+void netfs_wait_for_in_progress_stream(struct netfs_io_request *rreq,
+				       struct netfs_io_stream *stream);
+ssize_t netfs_wait_for_read(struct netfs_io_request *rreq);
+ssize_t netfs_wait_for_write(struct netfs_io_request *rreq);
+void netfs_wait_for_paused_read(struct netfs_io_request *rreq);
+void netfs_wait_for_paused_write(struct netfs_io_request *rreq);
+
+/*
+ * objects.c
+ */
+struct netfs_io_request *netfs_alloc_request(struct address_space *mapping,
+					     struct file *file,
+					     loff_t start, size_t len,
+					     enum netfs_io_origin origin);
+void netfs_get_request(struct netfs_io_request *rreq, enum netfs_rreq_ref_trace what);
+void netfs_clear_subrequests(struct netfs_io_request *rreq);
+void netfs_put_request(struct netfs_io_request *rreq, enum netfs_rreq_ref_trace what);
+void netfs_put_failed_request(struct netfs_io_request *rreq);
+struct netfs_io_subrequest *netfs_alloc_subrequest(struct netfs_io_request *rreq);
+
+static inline void netfs_see_request(struct netfs_io_request *rreq,
+				     enum netfs_rreq_ref_trace what)
+{
+	trace_netfs_rreq_ref(rreq->debug_id, refcount_read(&rreq->ref), what);
+}
+
+static inline void netfs_see_subrequest(struct netfs_io_subrequest *subreq,
+					enum netfs_sreq_ref_trace what)
+{
+	trace_netfs_sreq_ref(subreq->rreq->debug_id, subreq->debug_index,
+			     refcount_read(&subreq->ref), what);
+}
+
+/*
+ * read_collect.c
+ */
+bool netfs_read_collection(struct netfs_io_request *rreq);
+void netfs_read_collection_worker(struct work_struct *work);
+void netfs_cancel_read(struct netfs_io_subrequest *subreq, int error);
+void netfs_cache_read_terminated(void *priv, ssize_t transferred_or_error);
+
+/*
+ * read_pgpriv2.c
+ */
+void netfs_pgpriv2_copy_to_cache(struct netfs_io_request *rreq, struct folio *folio);
+void netfs_pgpriv2_end_copy_to_cache(struct netfs_io_request *rreq);
+bool netfs_pgpriv2_unlock_copied_folios(struct netfs_io_request *wreq);
+
+/*
+ * read_retry.c
+ */
+void netfs_retry_reads(struct netfs_io_request *rreq);
+void netfs_unlock_abandoned_read_pages(struct netfs_io_request *rreq);
+
+/*
+ * stats.c
+ */
+#ifdef CONFIG_NETFS_STATS
+extern atomic_t netfs_n_rh_dio_read;
+extern atomic_t netfs_n_rh_readahead;
+extern atomic_t netfs_n_rh_read_folio;
+extern atomic_t netfs_n_rh_read_single;
+extern atomic_t netfs_n_rh_rreq;
+extern atomic_t netfs_n_rh_sreq;
+extern atomic_t netfs_n_rh_download;
+extern atomic_t netfs_n_rh_download_done;
+extern atomic_t netfs_n_rh_download_failed;
+extern atomic_t netfs_n_rh_download_instead;
+extern atomic_t netfs_n_rh_read;
+extern atomic_t netfs_n_rh_read_done;
+extern atomic_t netfs_n_rh_read_failed;
+extern atomic_t netfs_n_rh_zero;
+extern atomic_t netfs_n_rh_short_read;
+extern atomic_t netfs_n_rh_write;
+extern atomic_t netfs_n_rh_write_begin;
+extern atomic_t netfs_n_rh_write_done;
+extern atomic_t netfs_n_rh_write_failed;
+extern atomic_t netfs_n_rh_write_zskip;
+extern atomic_t netfs_n_rh_retry_read_req;
+extern atomic_t netfs_n_rh_retry_read_subreq;
+extern atomic_t netfs_n_wh_buffered_write;
+extern atomic_t netfs_n_wh_writethrough;
+extern atomic_t netfs_n_wh_dio_write;
+extern atomic_t netfs_n_wh_writepages;
+extern atomic_t netfs_n_wh_copy_to_cache;
+extern atomic_t netfs_n_wh_wstream_conflict;
+extern atomic_t netfs_n_wh_upload;
+extern atomic_t netfs_n_wh_upload_done;
+extern atomic_t netfs_n_wh_upload_failed;
+extern atomic_t netfs_n_wh_write;
+extern atomic_t netfs_n_wh_write_done;
+extern atomic_t netfs_n_wh_write_failed;
+extern atomic_t netfs_n_wh_retry_write_req;
+extern atomic_t netfs_n_wh_retry_write_subreq;
+extern atomic_t netfs_n_wb_lock_skip;
+extern atomic_t netfs_n_wb_lock_wait;
+extern atomic_t netfs_n_folioq;
+
+int netfs_stats_show(struct seq_file *m, void *v);
+
+static inline void netfs_stat(atomic_t *stat)
+{
+	atomic_inc(stat);
+}
+
+static inline void netfs_stat_d(atomic_t *stat)
+{
+	atomic_dec(stat);
+}
+
+#else
+#define netfs_stat(x) do {} while(0)
+#define netfs_stat_d(x) do {} while(0)
+#endif
+
+/*
+ * write_collect.c
+ */
+int netfs_folio_written_back(struct folio *folio);
+bool netfs_write_collection(struct netfs_io_request *wreq);
+void netfs_write_collection_worker(struct work_struct *work);
+
+/*
+ * write_issue.c
+ */
+struct netfs_io_request *netfs_create_write_req(struct address_space *mapping,
+						struct file *file,
+						loff_t start,
+						enum netfs_io_origin origin);
+void netfs_prepare_write(struct netfs_io_request *wreq,
+			 struct netfs_io_stream *stream,
+			 loff_t start);
+void netfs_reissue_write(struct netfs_io_stream *stream,
+			 struct netfs_io_subrequest *subreq,
+			 struct iov_iter *source);
+void netfs_issue_write(struct netfs_io_request *wreq,
+		       struct netfs_io_stream *stream);
+size_t netfs_advance_write(struct netfs_io_request *wreq,
+			   struct netfs_io_stream *stream,
+			   loff_t start, size_t len, bool to_eof);
+struct netfs_io_request *netfs_begin_writethrough(struct kiocb *iocb, size_t len);
+int netfs_advance_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
+			       struct folio *folio, size_t copied, bool to_page_end,
+			       struct folio **writethrough_cache);
+ssize_t netfs_end_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
+			       struct folio *writethrough_cache);
+
+/*
+ * write_retry.c
+ */
+void netfs_retry_writes(struct netfs_io_request *wreq);
+
+/*
+ * Miscellaneous functions.
+ */
+static inline bool netfs_is_cache_enabled(struct netfs_inode *ctx)
+{
+#if IS_ENABLED(CONFIG_FSCACHE)
+	struct fscache_cookie *cookie = ctx->cache;
+
+	return fscache_cookie_valid(cookie) && cookie->cache_priv &&
+		fscache_cookie_enabled(cookie);
+#else
+	return false;
+#endif
+}
+
+static inline bool netfs_is_cache_maybe_enabled(struct netfs_inode *ctx)
+{
+#if IS_ENABLED(CONFIG_FSCACHE)
+	struct fscache_cookie *cookie = ctx->cache;
+
+	return fscache_cookie_valid(cookie) &&
+		test_bit(FSCACHE_COOKIE_IS_CACHING, &cookie->flags);
+#else
+	return false;
+#endif
+}
+
+/*
+ * Get a ref on a netfs group attached to a dirty page (e.g. a ceph snap).
+ */
+static inline struct netfs_group *netfs_get_group(struct netfs_group *netfs_group)
+{
+	if (netfs_group && netfs_group != NETFS_FOLIO_COPY_TO_CACHE)
+		refcount_inc(&netfs_group->ref);
+	return netfs_group;
+}
+
+/*
+ * Dispose of a netfs group attached to a dirty page (e.g. a ceph snap).
+ */
+static inline void netfs_put_group(struct netfs_group *netfs_group)
+{
+	if (netfs_group &&
+	    netfs_group != NETFS_FOLIO_COPY_TO_CACHE &&
+	    refcount_dec_and_test(&netfs_group->ref))
+		netfs_group->free(netfs_group);
+}
+
+/*
+ * Dispose of a netfs group attached to a dirty page (e.g. a ceph snap).
+ */
+static inline void netfs_put_group_many(struct netfs_group *netfs_group, int nr)
+{
+	if (netfs_group &&
+	    netfs_group != NETFS_FOLIO_COPY_TO_CACHE &&
+	    refcount_sub_and_test(nr, &netfs_group->ref))
+		netfs_group->free(netfs_group);
+}
+
+/*
+ * Clear and wake up a NETFS_RREQ_* flag bit on a request.
+ */
+static inline void netfs_wake_rreq_flag(struct netfs_io_request *rreq,
+					unsigned int rreq_flag,
+					enum netfs_rreq_trace trace)
+{
+	if (test_bit(rreq_flag, &rreq->flags)) {
+		clear_bit_unlock(rreq_flag, &rreq->flags);
+		smp_mb__after_atomic(); /* Set flag before task state */
+		trace_netfs_rreq(rreq, trace);
+		wake_up(&rreq->waitq);
+	}
+}
+
+/*
+ * Test the NETFS_RREQ_IN_PROGRESS flag, inserting an appropriate barrier.
+ */
+static inline bool netfs_check_rreq_in_progress(const struct netfs_io_request *rreq)
+{
+	/* Order read of flags before read of anything else, such as error. */
+	return test_bit_acquire(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
+}
+
+/*
+ * Test the NETFS_SREQ_IN_PROGRESS flag, inserting an appropriate barrier.
+ */
+static inline bool netfs_check_subreq_in_progress(const struct netfs_io_subrequest *subreq)
+{
+	/* Order read of flags before read of anything else, such as error. */
+	return test_bit_acquire(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+}
+
+/*
+ * fscache-cache.c
+ */
+#ifdef CONFIG_PROC_FS
+extern const struct seq_operations fscache_caches_seq_ops;
+#endif
+bool fscache_begin_cache_access(struct fscache_cache *cache, enum fscache_access_trace why);
+void fscache_end_cache_access(struct fscache_cache *cache, enum fscache_access_trace why);
+struct fscache_cache *fscache_lookup_cache(const char *name, bool is_cache);
+void fscache_put_cache(struct fscache_cache *cache, enum fscache_cache_trace where);
+
+static inline enum fscache_cache_state fscache_cache_state(const struct fscache_cache *cache)
+{
+	return smp_load_acquire(&cache->state);
+}
+
+static inline bool fscache_cache_is_live(const struct fscache_cache *cache)
+{
+	return fscache_cache_state(cache) == FSCACHE_CACHE_IS_ACTIVE;
+}
+
+static inline void fscache_set_cache_state(struct fscache_cache *cache,
+					   enum fscache_cache_state new_state)
+{
+	smp_store_release(&cache->state, new_state);
+
+}
+
+static inline bool fscache_set_cache_state_maybe(struct fscache_cache *cache,
+						 enum fscache_cache_state old_state,
+						 enum fscache_cache_state new_state)
+{
+	return try_cmpxchg_release(&cache->state, &old_state, new_state);
+}
+
+/*
+ * fscache-cookie.c
+ */
+extern struct kmem_cache *fscache_cookie_jar;
+#ifdef CONFIG_PROC_FS
+extern const struct seq_operations fscache_cookies_seq_ops;
+#endif
+extern struct timer_list fscache_cookie_lru_timer;
+
+extern void fscache_print_cookie(struct fscache_cookie *cookie, char prefix);
+extern bool fscache_begin_cookie_access(struct fscache_cookie *cookie,
+					enum fscache_access_trace why);
+
+static inline void fscache_see_cookie(struct fscache_cookie *cookie,
+				      enum fscache_cookie_trace where)
+{
+	trace_fscache_cookie(cookie->debug_id, refcount_read(&cookie->ref),
+			     where);
+}
+
+/*
+ * fscache-main.c
+ */
+extern unsigned int fscache_hash(unsigned int salt, const void *data, size_t len);
+#ifdef CONFIG_FSCACHE
+int __init fscache_init(void);
+void __exit fscache_exit(void);
+#else
+static inline int fscache_init(void) { return 0; }
+static inline void fscache_exit(void) {}
+#endif
+
+/*
+ * fscache-proc.c
+ */
+#ifdef CONFIG_PROC_FS
+extern int __init fscache_proc_init(void);
+extern void fscache_proc_cleanup(void);
+#else
+#define fscache_proc_init()	(0)
+#define fscache_proc_cleanup()	do {} while (0)
+#endif
+
+/*
+ * fscache-stats.c
+ */
+#ifdef CONFIG_FSCACHE_STATS
+extern atomic_t fscache_n_volumes;
+extern atomic_t fscache_n_volumes_collision;
+extern atomic_t fscache_n_volumes_nomem;
+extern atomic_t fscache_n_cookies;
+extern atomic_t fscache_n_cookies_lru;
+extern atomic_t fscache_n_cookies_lru_expired;
+extern atomic_t fscache_n_cookies_lru_removed;
+extern atomic_t fscache_n_cookies_lru_dropped;
+
+extern atomic_t fscache_n_acquires;
+extern atomic_t fscache_n_acquires_ok;
+extern atomic_t fscache_n_acquires_oom;
+
+extern atomic_t fscache_n_invalidates;
+
+extern atomic_t fscache_n_relinquishes;
+extern atomic_t fscache_n_relinquishes_retire;
+extern atomic_t fscache_n_relinquishes_dropped;
+
+extern atomic_t fscache_n_resizes;
+extern atomic_t fscache_n_resizes_null;
+
+static inline void fscache_stat(atomic_t *stat)
+{
+	atomic_inc(stat);
+}
+
+static inline void fscache_stat_d(atomic_t *stat)
+{
+	atomic_dec(stat);
+}
+
+#define __fscache_stat(stat) (stat)
+
+int fscache_stats_show(struct seq_file *m);
+#else
+
+#define __fscache_stat(stat) (NULL)
+#define fscache_stat(stat) do {} while (0)
+#define fscache_stat_d(stat) do {} while (0)
+
+static inline int fscache_stats_show(struct seq_file *m) { return 0; }
+#endif
+
+/*
+ * fscache-volume.c
+ */
+#ifdef CONFIG_PROC_FS
+extern const struct seq_operations fscache_volumes_seq_ops;
+#endif
+
+struct fscache_volume *fscache_get_volume(struct fscache_volume *volume,
+					  enum fscache_volume_trace where);
+bool fscache_begin_volume_access(struct fscache_volume *volume,
+				 struct fscache_cookie *cookie,
+				 enum fscache_access_trace why);
+void fscache_create_volume(struct fscache_volume *volume, bool wait);
+
+/*****************************************************************************/
+/*
+ * debug tracing
+ */
+#define dbgprintk(FMT, ...) \
+	printk("[%-6.6s] "FMT"\n", current->comm, ##__VA_ARGS__)
+
+#define kenter(FMT, ...) dbgprintk("==> %s("FMT")", __func__, ##__VA_ARGS__)
+#define kleave(FMT, ...) dbgprintk("<== %s()"FMT"", __func__, ##__VA_ARGS__)
+#define kdebug(FMT, ...) dbgprintk(FMT, ##__VA_ARGS__)
+
+#ifdef __KDEBUG
+#define _enter(FMT, ...) kenter(FMT, ##__VA_ARGS__)
+#define _leave(FMT, ...) kleave(FMT, ##__VA_ARGS__)
+#define _debug(FMT, ...) kdebug(FMT, ##__VA_ARGS__)
+
+#elif defined(CONFIG_NETFS_DEBUG)
+#define _enter(FMT, ...)			\
+do {						\
+	if (netfs_debug)			\
+		kenter(FMT, ##__VA_ARGS__);	\
+} while (0)
+
+#define _leave(FMT, ...)			\
+do {						\
+	if (netfs_debug)			\
+		kleave(FMT, ##__VA_ARGS__);	\
+} while (0)
+
+#define _debug(FMT, ...)			\
+do {						\
+	if (netfs_debug)			\
+		kdebug(FMT, ##__VA_ARGS__);	\
+} while (0)
+
+#else
+#define _enter(FMT, ...) no_printk("==> %s("FMT")", __func__, ##__VA_ARGS__)
+#define _leave(FMT, ...) no_printk("<== %s()"FMT"", __func__, ##__VA_ARGS__)
+#define _debug(FMT, ...) no_printk(FMT, ##__VA_ARGS__)
+#endif
+
+/*
+ * assertions
+ */
+#if 1 /* defined(__KDEBUGALL) */
+
+#define ASSERT(X)							\
+do {									\
+	if (unlikely(!(X))) {						\
+		pr_err("\n");					\
+		pr_err("Assertion failed\n");	\
+		BUG();							\
+	}								\
+} while (0)
+
+#define ASSERTCMP(X, OP, Y)						\
+do {									\
+	if (unlikely(!((X) OP (Y)))) {					\
+		pr_err("\n");					\
+		pr_err("Assertion failed\n");	\
+		pr_err("%lx " #OP " %lx is false\n",		\
+		       (unsigned long)(X), (unsigned long)(Y));		\
+		BUG();							\
+	}								\
+} while (0)
+
+#define ASSERTIF(C, X)							\
+do {									\
+	if (unlikely((C) && !(X))) {					\
+		pr_err("\n");					\
+		pr_err("Assertion failed\n");	\
+		BUG();							\
+	}								\
+} while (0)
+
+#define ASSERTIFCMP(C, X, OP, Y)					\
+do {									\
+	if (unlikely((C) && !((X) OP (Y)))) {				\
+		pr_err("\n");					\
+		pr_err("Assertion failed\n");	\
+		pr_err("%lx " #OP " %lx is false\n",		\
+		       (unsigned long)(X), (unsigned long)(Y));		\
+		BUG();							\
+	}								\
+} while (0)
+
+#else
+
+#define ASSERT(X)			do {} while (0)
+#define ASSERTCMP(X, OP, Y)		do {} while (0)
+#define ASSERTIF(C, X)			do {} while (0)
+#define ASSERTIFCMP(C, X, OP, Y)	do {} while (0)
+
+#endif /* assert or not */
diff --git a/fs/netfs/iterator.c b/fs/netfs/iterator.c
new file mode 100644
index 000000000..b375567e0
--- /dev/null
+++ b/fs/netfs/iterator.c
@@ -0,0 +1,308 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Iterator helpers.
+ *
+ * Copyright (C) 2022 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/uio.h>
+#include <linux/scatterlist.h>
+#include <linux/netfs.h>
+#include "internal.h"
+
+/**
+ * netfs_extract_user_iter - Extract the pages from a user iterator into a bvec
+ * @orig: The original iterator
+ * @orig_len: The amount of iterator to copy
+ * @new: The iterator to be set up
+ * @extraction_flags: Flags to qualify the request
+ *
+ * Extract the page fragments from the given amount of the source iterator and
+ * build up a second iterator that refers to all of those bits.  This allows
+ * the original iterator to be disposed of.
+ *
+ * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA be
+ * allowed on the pages extracted.
+ *
+ * On success, the number of elements in the bvec is returned, the original
+ * iterator will have been advanced by the amount extracted.
+ *
+ * The iov_iter_extract_mode() function should be used to query how cleanup
+ * should be performed.
+ */
+ssize_t netfs_extract_user_iter(struct iov_iter *orig, size_t orig_len,
+				struct iov_iter *new,
+				iov_iter_extraction_t extraction_flags)
+{
+	struct bio_vec *bv = NULL;
+	struct page **pages;
+	unsigned int cur_npages;
+	unsigned int max_pages;
+	unsigned int npages = 0;
+	unsigned int i;
+	ssize_t ret = 0;
+	size_t count = orig_len, offset, len;
+	size_t bv_size, pg_size;
+
+	if (WARN_ON_ONCE(!iter_is_ubuf(orig) && !iter_is_iovec(orig)))
+		return -EIO;
+
+	max_pages = iov_iter_npages(orig, INT_MAX);
+	bv_size = array_size(max_pages, sizeof(*bv));
+	bv = kvmalloc(bv_size, GFP_KERNEL);
+	if (!bv)
+		return -ENOMEM;
+
+	/* Put the page list at the end of the bvec list storage.  bvec
+	 * elements are larger than page pointers, so as long as we work
+	 * 0->last, we should be fine.
+	 */
+	pg_size = array_size(max_pages, sizeof(*pages));
+	pages = (void *)bv + bv_size - pg_size;
+
+	while (count && npages < max_pages) {
+		ret = iov_iter_extract_pages(orig, &pages, count,
+					     max_pages - npages, extraction_flags,
+					     &offset);
+		if (unlikely(ret <= 0)) {
+			ret = ret ?: -EIO;
+			break;
+		}
+
+		if (WARN(ret > count,
+			 "%s: extract_pages overrun %zd > %zu bytes\n",
+			 __func__, ret, count)) {
+			ret = -EIO;
+			break;
+		}
+
+		cur_npages = DIV_ROUND_UP(offset + ret, PAGE_SIZE);
+		if (WARN(cur_npages > max_pages - npages,
+			 "%s: extract_pages overrun %u > %u pages\n",
+			 __func__, npages + cur_npages, max_pages)) {
+			ret = -EIO;
+			break;
+		}
+
+		count -= ret;
+		ret += offset;
+
+		for (i = 0; i < cur_npages; i++) {
+			len = ret > PAGE_SIZE ? PAGE_SIZE : ret;
+			bvec_set_page(bv + npages + i, *pages++, len - offset, offset);
+			ret -= len;
+			offset = 0;
+		}
+
+		npages += cur_npages;
+	}
+
+	/* Note: Don't try to clean up after EIO.  Either we got no pages, so
+	 * nothing to clean up, or we got a buffer overrun, memory corruption
+	 * and can't trust the stuff in the buffer (a WARN was emitted).
+	 */
+
+	if (ret < 0 && (ret == -ENOMEM || npages == 0)) {
+		for (i = 0; i < npages; i++)
+			unpin_user_page(bv[i].bv_page);
+		kvfree(bv);
+		return ret;
+	}
+
+	iov_iter_bvec(new, orig->data_source, bv, npages, orig_len - count);
+	return npages;
+}
+EXPORT_SYMBOL_GPL(netfs_extract_user_iter);
+
+/*
+ * Select the span of a bvec iterator we're going to use.  Limit it by both maximum
+ * size and maximum number of segments.  Returns the size of the span in bytes.
+ */
+static size_t netfs_limit_bvec(const struct iov_iter *iter, size_t start_offset,
+			       size_t max_size, size_t max_segs)
+{
+	const struct bio_vec *bvecs = iter->bvec;
+	unsigned int nbv = iter->nr_segs, ix = 0, nsegs = 0;
+	size_t len, span = 0, n = iter->count;
+	size_t skip = iter->iov_offset + start_offset;
+
+	if (WARN_ON(!iov_iter_is_bvec(iter)) ||
+	    WARN_ON(start_offset > n) ||
+	    n == 0)
+		return 0;
+
+	while (n && ix < nbv && skip) {
+		len = bvecs[ix].bv_len;
+		if (skip < len)
+			break;
+		skip -= len;
+		n -= len;
+		ix++;
+	}
+
+	while (n && ix < nbv) {
+		len = min3(n, bvecs[ix].bv_len - skip, max_size);
+		span += len;
+		nsegs++;
+		ix++;
+		if (span >= max_size || nsegs >= max_segs)
+			break;
+		skip = 0;
+		n -= len;
+	}
+
+	return min(span, max_size);
+}
+
+/*
+ * Select the span of a kvec iterator we're going to use.  Limit it by both
+ * maximum size and maximum number of segments.  Returns the size of the span
+ * in bytes.
+ */
+static size_t netfs_limit_kvec(const struct iov_iter *iter, size_t start_offset,
+			       size_t max_size, size_t max_segs)
+{
+	const struct kvec *kvecs = iter->kvec;
+	unsigned int nkv = iter->nr_segs, ix = 0, nsegs = 0;
+	size_t len, span = 0, n = iter->count;
+	size_t skip = iter->iov_offset + start_offset;
+
+	if (WARN_ON(!iov_iter_is_kvec(iter)) ||
+	    WARN_ON(start_offset > n) ||
+	    n == 0)
+		return 0;
+
+	while (n && ix < nkv && skip) {
+		len = kvecs[ix].iov_len;
+		if (skip < len)
+			break;
+		skip -= len;
+		n -= len;
+		ix++;
+	}
+
+	while (n && ix < nkv) {
+		len = min3(n, kvecs[ix].iov_len - skip, max_size);
+		span += len;
+		nsegs++;
+		ix++;
+		if (span >= max_size || nsegs >= max_segs)
+			break;
+		skip = 0;
+		n -= len;
+	}
+
+	return min(span, max_size);
+}
+
+/*
+ * Select the span of an xarray iterator we're going to use.  Limit it by both
+ * maximum size and maximum number of segments.  It is assumed that segments
+ * can be larger than a page in size, provided they're physically contiguous.
+ * Returns the size of the span in bytes.
+ */
+static size_t netfs_limit_xarray(const struct iov_iter *iter, size_t start_offset,
+				 size_t max_size, size_t max_segs)
+{
+	struct folio *folio;
+	unsigned int nsegs = 0;
+	loff_t pos = iter->xarray_start + iter->iov_offset;
+	pgoff_t index = pos / PAGE_SIZE;
+	size_t span = 0, n = iter->count;
+
+	XA_STATE(xas, iter->xarray, index);
+
+	if (WARN_ON(!iov_iter_is_xarray(iter)) ||
+	    WARN_ON(start_offset > n) ||
+	    n == 0)
+		return 0;
+	max_size = min(max_size, n - start_offset);
+
+	rcu_read_lock();
+	xas_for_each(&xas, folio, ULONG_MAX) {
+		size_t offset, flen, len;
+		if (xas_retry(&xas, folio))
+			continue;
+		if (WARN_ON(xa_is_value(folio)))
+			break;
+		if (WARN_ON(folio_test_hugetlb(folio)))
+			break;
+
+		flen = folio_size(folio);
+		offset = offset_in_folio(folio, pos);
+		len = min(max_size, flen - offset);
+		span += len;
+		nsegs++;
+		if (span >= max_size || nsegs >= max_segs)
+			break;
+	}
+
+	rcu_read_unlock();
+	return min(span, max_size);
+}
+
+/*
+ * Select the span of a folio queue iterator we're going to use.  Limit it by
+ * both maximum size and maximum number of segments.  Returns the size of the
+ * span in bytes.
+ */
+static size_t netfs_limit_folioq(const struct iov_iter *iter, size_t start_offset,
+				 size_t max_size, size_t max_segs)
+{
+	const struct folio_queue *folioq = iter->folioq;
+	unsigned int nsegs = 0;
+	unsigned int slot = iter->folioq_slot;
+	size_t span = 0, n = iter->count;
+
+	if (WARN_ON(!iov_iter_is_folioq(iter)) ||
+	    WARN_ON(start_offset > n) ||
+	    n == 0)
+		return 0;
+	max_size = umin(max_size, n - start_offset);
+
+	if (slot >= folioq_nr_slots(folioq)) {
+		folioq = folioq->next;
+		slot = 0;
+	}
+
+	start_offset += iter->iov_offset;
+	do {
+		size_t flen = folioq_folio_size(folioq, slot);
+
+		if (start_offset < flen) {
+			span += flen - start_offset;
+			nsegs++;
+			start_offset = 0;
+		} else {
+			start_offset -= flen;
+		}
+		if (span >= max_size || nsegs >= max_segs)
+			break;
+
+		slot++;
+		if (slot >= folioq_nr_slots(folioq)) {
+			folioq = folioq->next;
+			slot = 0;
+		}
+	} while (folioq);
+
+	return umin(span, max_size);
+}
+
+size_t netfs_limit_iter(const struct iov_iter *iter, size_t start_offset,
+			size_t max_size, size_t max_segs)
+{
+	if (iov_iter_is_folioq(iter))
+		return netfs_limit_folioq(iter, start_offset, max_size, max_segs);
+	if (iov_iter_is_bvec(iter))
+		return netfs_limit_bvec(iter, start_offset, max_size, max_segs);
+	if (iov_iter_is_xarray(iter))
+		return netfs_limit_xarray(iter, start_offset, max_size, max_segs);
+	if (iov_iter_is_kvec(iter))
+		return netfs_limit_kvec(iter, start_offset, max_size, max_segs);
+	BUG();
+}
+EXPORT_SYMBOL(netfs_limit_iter);
diff --git a/fs/netfs/locking.c b/fs/netfs/locking.c
new file mode 100644
index 000000000..4e3be2b81
--- /dev/null
+++ b/fs/netfs/locking.c
@@ -0,0 +1,300 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * I/O and data path helper functionality.
+ *
+ * Borrowed from NFS Copyright (c) 2016 Trond Myklebust
+ */
+
+#include <linux/kernel.h>
+#include <linux/netfs.h>
+#include "internal.h"
+
+struct netfs_wb_waiter {
+	struct list_head	link;		/* Link in ictx->wb_queue */
+	struct task_struct	*waiter;	/* Waiter task; cleared when lock granted */
+};
+
+/*
+ * inode_dio_wait_interruptible - wait for outstanding DIO requests to finish
+ * @inode: inode to wait for
+ *
+ * Waits for all pending direct I/O requests to finish so that we can
+ * proceed with a truncate or equivalent operation.
+ *
+ * Must be called under a lock that serializes taking new references
+ * to i_dio_count, usually by inode->i_mutex.
+ */
+static int netfs_inode_dio_wait_interruptible(struct inode *inode)
+{
+	if (inode_dio_finished(inode))
+		return 0;
+
+	inode_dio_wait_interruptible(inode);
+	return !inode_dio_finished(inode) ? -ERESTARTSYS : 0;
+}
+
+/* Call with exclusively locked inode->i_rwsem */
+static int netfs_block_o_direct(struct netfs_inode *ictx)
+{
+	if (!test_bit(NETFS_ICTX_ODIRECT, &ictx->flags))
+		return 0;
+	clear_bit(NETFS_ICTX_ODIRECT, &ictx->flags);
+	return netfs_inode_dio_wait_interruptible(&ictx->inode);
+}
+
+/**
+ * netfs_start_io_read - declare the file is being used for buffered reads
+ * @inode: file inode
+ *
+ * Declare that a buffered read operation is about to start, and ensure
+ * that we block all direct I/O.
+ * On exit, the function ensures that the NETFS_ICTX_ODIRECT flag is unset,
+ * and holds a shared lock on inode->i_rwsem to ensure that the flag
+ * cannot be changed.
+ * In practice, this means that buffered read operations are allowed to
+ * execute in parallel, thanks to the shared lock, whereas direct I/O
+ * operations need to wait to grab an exclusive lock in order to set
+ * NETFS_ICTX_ODIRECT.
+ * Note that buffered writes and truncates both take a write lock on
+ * inode->i_rwsem, meaning that those are serialised w.r.t. the reads.
+ */
+int netfs_start_io_read(struct inode *inode)
+	__acquires(inode->i_rwsem)
+{
+	struct netfs_inode *ictx = netfs_inode(inode);
+
+	/* Be an optimist! */
+	if (down_read_interruptible(&inode->i_rwsem) < 0)
+		return -ERESTARTSYS;
+	if (test_bit(NETFS_ICTX_ODIRECT, &ictx->flags) == 0)
+		return 0;
+	up_read(&inode->i_rwsem);
+
+	/* Slow path.... */
+	if (down_write_killable(&inode->i_rwsem) < 0)
+		return -ERESTARTSYS;
+	if (netfs_block_o_direct(ictx) < 0) {
+		up_write(&inode->i_rwsem);
+		return -ERESTARTSYS;
+	}
+	downgrade_write(&inode->i_rwsem);
+	return 0;
+}
+EXPORT_SYMBOL(netfs_start_io_read);
+
+/**
+ * netfs_end_io_read - declare that the buffered read operation is done
+ * @inode: file inode
+ *
+ * Declare that a buffered read operation is done, and release the shared
+ * lock on inode->i_rwsem.
+ */
+void netfs_end_io_read(struct inode *inode)
+	__releases(inode->i_rwsem)
+{
+	up_read(&inode->i_rwsem);
+}
+EXPORT_SYMBOL(netfs_end_io_read);
+
+/**
+ * netfs_start_io_write - declare the file is being used for buffered writes
+ * @inode: file inode
+ *
+ * Declare that a buffered read operation is about to start, and ensure
+ * that we block all direct I/O.
+ */
+int netfs_start_io_write(struct inode *inode)
+	__acquires(inode->i_rwsem)
+{
+	struct netfs_inode *ictx = netfs_inode(inode);
+
+	if (down_write_killable(&inode->i_rwsem) < 0)
+		return -ERESTARTSYS;
+	if (netfs_block_o_direct(ictx) < 0) {
+		up_write(&inode->i_rwsem);
+		return -ERESTARTSYS;
+	}
+	downgrade_write(&inode->i_rwsem);
+	return 0;
+}
+EXPORT_SYMBOL(netfs_start_io_write);
+
+/**
+ * netfs_end_io_write - declare that the buffered write operation is done
+ * @inode: file inode
+ *
+ * Declare that a buffered write operation is done, and release the
+ * lock on inode->i_rwsem.
+ */
+void netfs_end_io_write(struct inode *inode)
+	__releases(inode->i_rwsem)
+{
+	up_read(&inode->i_rwsem);
+}
+EXPORT_SYMBOL(netfs_end_io_write);
+
+/* Call with exclusively locked inode->i_rwsem */
+static int netfs_block_buffered(struct inode *inode)
+{
+	struct netfs_inode *ictx = netfs_inode(inode);
+	int ret;
+
+	if (!test_bit(NETFS_ICTX_ODIRECT, &ictx->flags)) {
+		set_bit(NETFS_ICTX_ODIRECT, &ictx->flags);
+		if (inode->i_mapping->nrpages != 0) {
+			unmap_mapping_range(inode->i_mapping, 0, 0, 0);
+			ret = filemap_fdatawait(inode->i_mapping);
+			if (ret < 0) {
+				clear_bit(NETFS_ICTX_ODIRECT, &ictx->flags);
+				return ret;
+			}
+		}
+	}
+	return 0;
+}
+
+/**
+ * netfs_start_io_direct - declare the file is being used for direct i/o
+ * @inode: file inode
+ *
+ * Declare that a direct I/O operation is about to start, and ensure
+ * that we block all buffered I/O.
+ * On exit, the function ensures that the NETFS_ICTX_ODIRECT flag is set,
+ * and holds a shared lock on inode->i_rwsem to ensure that the flag
+ * cannot be changed.
+ * In practice, this means that direct I/O operations are allowed to
+ * execute in parallel, thanks to the shared lock, whereas buffered I/O
+ * operations need to wait to grab an exclusive lock in order to clear
+ * NETFS_ICTX_ODIRECT.
+ * Note that buffered writes and truncates both take a write lock on
+ * inode->i_rwsem, meaning that those are serialised w.r.t. O_DIRECT.
+ */
+int netfs_start_io_direct(struct inode *inode)
+	__acquires(inode->i_rwsem)
+{
+	struct netfs_inode *ictx = netfs_inode(inode);
+	int ret;
+
+	/* Be an optimist! */
+	if (down_read_interruptible(&inode->i_rwsem) < 0)
+		return -ERESTARTSYS;
+	if (test_bit(NETFS_ICTX_ODIRECT, &ictx->flags) != 0)
+		return 0;
+	up_read(&inode->i_rwsem);
+
+	/* Slow path.... */
+	if (down_write_killable(&inode->i_rwsem) < 0)
+		return -ERESTARTSYS;
+	ret = netfs_block_buffered(inode);
+	if (ret < 0) {
+		up_write(&inode->i_rwsem);
+		return ret;
+	}
+	downgrade_write(&inode->i_rwsem);
+	return 0;
+}
+EXPORT_SYMBOL(netfs_start_io_direct);
+
+/**
+ * netfs_end_io_direct - declare that the direct i/o operation is done
+ * @inode: file inode
+ *
+ * Declare that a direct I/O operation is done, and release the shared
+ * lock on inode->i_rwsem.
+ */
+void netfs_end_io_direct(struct inode *inode)
+	__releases(inode->i_rwsem)
+{
+	up_read(&inode->i_rwsem);
+}
+EXPORT_SYMBOL(netfs_end_io_direct);
+
+/*
+ * Wait to have exclusive access to writeback.
+ */
+static bool netfs_wb_begin_wait(struct netfs_inode *ictx)
+{
+	struct netfs_wb_waiter waiter = {};
+	struct task_struct *tsk = current;
+	bool got = false;
+
+	spin_lock(&ictx->lock);
+
+	if (test_and_set_bit_lock(NETFS_ICTX_WB_LOCK, &ictx->flags)) {
+		get_task_struct(tsk);
+		waiter.waiter = tsk;
+		list_add_tail(&waiter.link, &ictx->wb_queue);
+	} else {
+		got = true;
+	}
+	spin_unlock(&ictx->lock);
+
+	if (!got) {
+		for (;;) {
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			/* Read waiter before accessing inode state. */
+			if (smp_load_acquire(&waiter.waiter) == NULL)
+				break;
+			schedule();
+		}
+	}
+	__set_current_state(TASK_RUNNING);
+	return true;
+}
+
+/**
+ * netfs_wb_begin - Begin writeback, waiting if need be
+ * @ictx: The inode to get writeback access on
+ * @nowait: Return failure immediately rather than waiting if true
+ *
+ * Begin writeback to an inode, waiting for exclusive access if @nowait is
+ * false.  This prevents collection from being done out of order with respect
+ * to the issuance of write subrequests.
+ *
+ * Note that writeback may be ended in a different process (e.g. the collection
+ * function on a workqueue) than started it.
+ *
+ * Return: True if can proceed, false if denied.
+ */
+bool netfs_wb_begin(struct netfs_inode *ictx, bool nowait)
+{
+	if (!test_and_set_bit_lock(NETFS_ICTX_WB_LOCK, &ictx->flags))
+		return true;
+	if (nowait) {
+		netfs_stat(&netfs_n_wb_lock_skip);
+		return false;
+	}
+	netfs_stat(&netfs_n_wb_lock_wait);
+	return netfs_wb_begin_wait(ictx);
+}
+EXPORT_SYMBOL(netfs_wb_begin);
+
+/* netfs_wb_end - End writeback
+ * @ictx: The inode we have writeback access to
+ *
+ * End writeback access on an inode, waking up the next writeback request.
+ */
+void netfs_wb_end(struct netfs_inode *ictx)
+{
+	struct netfs_wb_waiter *waiter;
+	struct task_struct *tsk;
+
+	WARN_ON_ONCE(!test_bit(NETFS_ICTX_WB_LOCK, &ictx->flags));
+
+	spin_lock(&ictx->lock);
+
+	waiter = list_first_entry_or_null(&ictx->wb_queue, struct netfs_wb_waiter, link);
+	if (waiter) {
+		list_del(&waiter->link);
+		tsk = waiter->waiter;
+		/* Write inode state before clearing waiter. */
+		smp_store_release(&waiter->waiter, NULL);
+		wake_up_process(tsk);
+		put_task_struct(tsk);
+	} else {
+		clear_bit_unlock(NETFS_ICTX_WB_LOCK, &ictx->flags);
+	}
+
+	spin_unlock(&ictx->lock);
+}
+EXPORT_SYMBOL(netfs_wb_end);
diff --git a/fs/netfs/main.c b/fs/netfs/main.c
new file mode 100644
index 000000000..927badf39
--- /dev/null
+++ b/fs/netfs/main.c
@@ -0,0 +1,183 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Miscellaneous bits for the netfs support library.
+ *
+ * Copyright (C) 2022 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/module.h>
+#include <linux/export.h>
+#include <linux/mempool.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include "internal.h"
+#define CREATE_TRACE_POINTS
+#include <trace/events/netfs.h>
+
+MODULE_DESCRIPTION("Network fs support");
+MODULE_AUTHOR("Red Hat, Inc.");
+MODULE_LICENSE("GPL");
+
+EXPORT_TRACEPOINT_SYMBOL(netfs_sreq);
+
+unsigned netfs_debug;
+module_param_named(debug, netfs_debug, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(netfs_debug, "Netfs support debugging mask");
+
+static struct kmem_cache *netfs_request_slab;
+static struct kmem_cache *netfs_subrequest_slab;
+mempool_t netfs_request_pool;
+mempool_t netfs_subrequest_pool;
+mempool_t netfs_folioq_pool;
+
+#ifdef CONFIG_PROC_FS
+LIST_HEAD(netfs_io_requests);
+DEFINE_SPINLOCK(netfs_proc_lock);
+
+static const char *netfs_origins[nr__netfs_io_origin] = {
+	[NETFS_READAHEAD]		= "RA",
+	[NETFS_READPAGE]		= "RP",
+	[NETFS_READ_GAPS]		= "RG",
+	[NETFS_READ_SINGLE]		= "R1",
+	[NETFS_READ_FOR_WRITE]		= "RW",
+	[NETFS_UNBUFFERED_READ]		= "UR",
+	[NETFS_DIO_READ]		= "DR",
+	[NETFS_WRITEBACK]		= "WB",
+	[NETFS_WRITEBACK_SINGLE]	= "W1",
+	[NETFS_WRITETHROUGH]		= "WT",
+	[NETFS_UNBUFFERED_WRITE]	= "UW",
+	[NETFS_DIO_WRITE]		= "DW",
+	[NETFS_PGPRIV2_COPY_TO_CACHE]	= "2C",
+};
+
+/*
+ * Generate a list of I/O requests in /proc/fs/netfs/requests
+ */
+static int netfs_requests_seq_show(struct seq_file *m, void *v)
+{
+	struct netfs_io_request *rreq;
+
+	if (v == &netfs_io_requests) {
+		seq_puts(m,
+			 "REQUEST  OR REF FLAG ERR  OPS COVERAGE\n"
+			 "======== == === ==== ==== === =========\n"
+			 );
+		return 0;
+	}
+
+	rreq = list_entry(v, struct netfs_io_request, proc_link);
+	seq_printf(m,
+		   "%08x %s %3d %4lx %4ld %3d @%04llx %llx/%llx",
+		   rreq->debug_id,
+		   netfs_origins[rreq->origin],
+		   refcount_read(&rreq->ref),
+		   rreq->flags,
+		   rreq->error,
+		   0,
+		   rreq->start, rreq->submitted, rreq->len);
+	seq_putc(m, '\n');
+	return 0;
+}
+
+static void *netfs_requests_seq_start(struct seq_file *m, loff_t *_pos)
+	__acquires(rcu)
+{
+	rcu_read_lock();
+	return seq_list_start_head(&netfs_io_requests, *_pos);
+}
+
+static void *netfs_requests_seq_next(struct seq_file *m, void *v, loff_t *_pos)
+{
+	return seq_list_next(v, &netfs_io_requests, _pos);
+}
+
+static void netfs_requests_seq_stop(struct seq_file *m, void *v)
+	__releases(rcu)
+{
+	rcu_read_unlock();
+}
+
+static const struct seq_operations netfs_requests_seq_ops = {
+	.start  = netfs_requests_seq_start,
+	.next   = netfs_requests_seq_next,
+	.stop   = netfs_requests_seq_stop,
+	.show   = netfs_requests_seq_show,
+};
+#endif /* CONFIG_PROC_FS */
+
+static int __init netfs_init(void)
+{
+	int ret = -ENOMEM;
+
+	if (mempool_init_kmalloc_pool(&netfs_folioq_pool, 100, sizeof(struct folio_queue)) < 0)
+		goto error_folioq_pool;
+
+	netfs_request_slab = kmem_cache_create("netfs_request",
+					       sizeof(struct netfs_io_request), 0,
+					       SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT,
+					       NULL);
+	if (!netfs_request_slab)
+		goto error_req;
+
+	if (mempool_init_slab_pool(&netfs_request_pool, 100, netfs_request_slab) < 0)
+		goto error_reqpool;
+
+	netfs_subrequest_slab = kmem_cache_create("netfs_subrequest",
+						  sizeof(struct netfs_io_subrequest) + 16, 0,
+						  SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT,
+						  NULL);
+	if (!netfs_subrequest_slab)
+		goto error_subreq;
+
+	if (mempool_init_slab_pool(&netfs_subrequest_pool, 100, netfs_subrequest_slab) < 0)
+		goto error_subreqpool;
+
+#ifdef CONFIG_PROC_FS
+	if (!proc_mkdir("fs/netfs", NULL))
+		goto error_proc;
+	if (!proc_create_seq("fs/netfs/requests", S_IFREG | 0444, NULL,
+			     &netfs_requests_seq_ops))
+		goto error_procfile;
+#endif
+#ifdef CONFIG_FSCACHE_STATS
+	if (!proc_create_single("fs/netfs/stats", S_IFREG | 0444, NULL,
+				netfs_stats_show))
+		goto error_procfile;
+#endif
+
+	ret = fscache_init();
+	if (ret < 0)
+		goto error_fscache;
+	return 0;
+
+error_fscache:
+#ifdef CONFIG_PROC_FS
+error_procfile:
+	remove_proc_subtree("fs/netfs", NULL);
+error_proc:
+#endif
+	mempool_exit(&netfs_subrequest_pool);
+error_subreqpool:
+	kmem_cache_destroy(netfs_subrequest_slab);
+error_subreq:
+	mempool_exit(&netfs_request_pool);
+error_reqpool:
+	kmem_cache_destroy(netfs_request_slab);
+error_req:
+	mempool_exit(&netfs_folioq_pool);
+error_folioq_pool:
+	return ret;
+}
+fs_initcall(netfs_init);
+
+static void __exit netfs_exit(void)
+{
+	fscache_exit();
+	remove_proc_subtree("fs/netfs", NULL);
+	mempool_exit(&netfs_subrequest_pool);
+	kmem_cache_destroy(netfs_subrequest_slab);
+	mempool_exit(&netfs_request_pool);
+	kmem_cache_destroy(netfs_request_slab);
+	mempool_exit(&netfs_folioq_pool);
+}
+module_exit(netfs_exit);
diff --git a/fs/netfs/misc.c b/fs/netfs/misc.c
new file mode 100644
index 000000000..5d554512e
--- /dev/null
+++ b/fs/netfs/misc.c
@@ -0,0 +1,565 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Miscellaneous routines.
+ *
+ * Copyright (C) 2023 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/swap.h>
+#include "internal.h"
+
+/**
+ * netfs_alloc_folioq_buffer - Allocate buffer space into a folio queue
+ * @mapping: Address space to set on the folio (or NULL).
+ * @_buffer: Pointer to the folio queue to add to (may point to a NULL; updated).
+ * @_cur_size: Current size of the buffer (updated).
+ * @size: Target size of the buffer.
+ * @gfp: The allocation constraints.
+ */
+int netfs_alloc_folioq_buffer(struct address_space *mapping,
+			      struct folio_queue **_buffer,
+			      size_t *_cur_size, ssize_t size, gfp_t gfp)
+{
+	struct folio_queue *tail = *_buffer, *p;
+
+	size = round_up(size, PAGE_SIZE);
+	if (*_cur_size >= size)
+		return 0;
+
+	if (tail)
+		while (tail->next)
+			tail = tail->next;
+
+	do {
+		struct folio *folio;
+		int order = 0, slot;
+
+		if (!tail || folioq_full(tail)) {
+			p = netfs_folioq_alloc(0, GFP_NOFS, netfs_trace_folioq_alloc_buffer);
+			if (!p)
+				return -ENOMEM;
+			if (tail) {
+				tail->next = p;
+				p->prev = tail;
+			} else {
+				*_buffer = p;
+			}
+			tail = p;
+		}
+
+		if (size - *_cur_size > PAGE_SIZE)
+			order = umin(ilog2(size - *_cur_size) - PAGE_SHIFT,
+				     MAX_PAGECACHE_ORDER);
+
+		folio = folio_alloc(gfp, order);
+		if (!folio && order > 0)
+			folio = folio_alloc(gfp, 0);
+		if (!folio)
+			return -ENOMEM;
+
+		folio->mapping = mapping;
+		folio->index = *_cur_size / PAGE_SIZE;
+		trace_netfs_folio(folio, netfs_folio_trace_alloc_buffer);
+		slot = folioq_append_mark(tail, folio);
+		*_cur_size += folioq_folio_size(tail, slot);
+	} while (*_cur_size < size);
+
+	return 0;
+}
+EXPORT_SYMBOL(netfs_alloc_folioq_buffer);
+
+/**
+ * netfs_free_folioq_buffer - Free a folio queue.
+ * @fq: The start of the folio queue to free
+ *
+ * Free up a chain of folio_queues and, if marked, the marked folios they point
+ * to.
+ */
+void netfs_free_folioq_buffer(struct folio_queue *fq)
+{
+	struct folio_queue *next;
+	struct folio_batch fbatch;
+
+	folio_batch_init(&fbatch);
+
+	for (; fq; fq = next) {
+		for (int slot = 0; slot < folioq_count(fq); slot++) {
+			struct folio *folio = folioq_folio(fq, slot);
+
+			if (!folio ||
+			    !folioq_is_marked(fq, slot))
+				continue;
+
+			trace_netfs_folio(folio, netfs_folio_trace_put);
+			if (folio_batch_add(&fbatch, folio))
+				folio_batch_release(&fbatch);
+		}
+
+		netfs_stat_d(&netfs_n_folioq);
+		next = fq->next;
+		kfree(fq);
+	}
+
+	folio_batch_release(&fbatch);
+}
+EXPORT_SYMBOL(netfs_free_folioq_buffer);
+
+/*
+ * Reset the subrequest iterator to refer just to the region remaining to be
+ * read.  The iterator may or may not have been advanced by socket ops or
+ * extraction ops to an extent that may or may not match the amount actually
+ * read.
+ */
+void netfs_reset_iter(struct netfs_io_subrequest *subreq)
+{
+	struct iov_iter *io_iter = &subreq->io_iter;
+	size_t remain = subreq->len - subreq->transferred;
+
+	if (io_iter->count > remain)
+		iov_iter_advance(io_iter, io_iter->count - remain);
+	else if (io_iter->count < remain)
+		iov_iter_revert(io_iter, remain - io_iter->count);
+	iov_iter_truncate(&subreq->io_iter, remain);
+}
+
+/**
+ * netfs_dirty_folio - Mark folio dirty and pin a cache object for writeback
+ * @mapping: The mapping the folio belongs to.
+ * @folio: The folio being dirtied.
+ *
+ * Set the dirty flag on a folio and pin an in-use cache object in memory so
+ * that writeback can later write to it.  This is intended to be called from
+ * the filesystem's ->dirty_folio() method.
+ *
+ * Return: true if the dirty flag was set on the folio, false otherwise.
+ */
+bool netfs_dirty_folio(struct address_space *mapping, struct folio *folio)
+{
+	struct inode *inode = mapping->host;
+	struct netfs_inode *ictx = netfs_inode(inode);
+	struct fscache_cookie *cookie = netfs_i_cookie(ictx);
+	bool need_use = false;
+
+	_enter("");
+
+	if (!filemap_dirty_folio(mapping, folio))
+		return false;
+	if (!fscache_cookie_valid(cookie))
+		return true;
+
+	if (!(inode_state_read_once(inode) & I_PINNING_NETFS_WB)) {
+		spin_lock(&inode->i_lock);
+		if (!(inode_state_read(inode) & I_PINNING_NETFS_WB)) {
+			inode_state_set(inode, I_PINNING_NETFS_WB);
+			need_use = true;
+		}
+		spin_unlock(&inode->i_lock);
+
+		if (need_use)
+			fscache_use_cookie(cookie, true);
+	}
+	return true;
+}
+EXPORT_SYMBOL(netfs_dirty_folio);
+
+/**
+ * netfs_unpin_writeback - Unpin writeback resources
+ * @inode: The inode on which the cookie resides
+ * @wbc: The writeback control
+ *
+ * Unpin the writeback resources pinned by netfs_dirty_folio().  This is
+ * intended to be called as/by the netfs's ->write_inode() method.
+ */
+int netfs_unpin_writeback(struct inode *inode, struct writeback_control *wbc)
+{
+	struct fscache_cookie *cookie = netfs_i_cookie(netfs_inode(inode));
+
+	if (wbc->unpinned_netfs_wb)
+		fscache_unuse_cookie(cookie, NULL, NULL);
+	return 0;
+}
+EXPORT_SYMBOL(netfs_unpin_writeback);
+
+/**
+ * netfs_clear_inode_writeback - Clear writeback resources pinned by an inode
+ * @inode: The inode to clean up
+ * @aux: Auxiliary data to apply to the inode
+ *
+ * Clear any writeback resources held by an inode when the inode is evicted.
+ * This must be called before clear_inode() is called.
+ */
+void netfs_clear_inode_writeback(struct inode *inode, const void *aux)
+{
+	struct fscache_cookie *cookie = netfs_i_cookie(netfs_inode(inode));
+
+	if (inode_state_read_once(inode) & I_PINNING_NETFS_WB) {
+		loff_t i_size = i_size_read(inode);
+		fscache_unuse_cookie(cookie, aux, &i_size);
+	}
+}
+EXPORT_SYMBOL(netfs_clear_inode_writeback);
+
+/**
+ * netfs_invalidate_folio - Invalidate or partially invalidate a folio
+ * @folio: Folio proposed for release
+ * @offset: Offset of the invalidated region
+ * @length: Length of the invalidated region
+ *
+ * Invalidate part or all of a folio for a network filesystem.  The folio will
+ * be removed afterwards if the invalidated region covers the entire folio.
+ */
+void netfs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
+{
+	struct netfs_folio *finfo;
+	struct inode *inode = folio_inode(folio);
+	struct netfs_inode *ctx = netfs_inode(inode);
+	size_t flen = folio_size(folio);
+
+	_enter("{%lx},%zx,%zx", folio->index, offset, length);
+
+	if (offset == 0 && length == flen) {
+		unsigned long long i_size, remote_i_size, zero_point;
+		unsigned long long fpos = folio_pos(folio), end;
+
+		netfs_read_sizes(inode, &i_size, &remote_i_size, &zero_point);
+		end = umin(fpos + flen, i_size);
+		if (fpos < i_size && end > zero_point) {
+			spin_lock(&inode->i_lock);
+			end = umin(fpos + flen, inode->i_size);
+			if (fpos < i_size && end > ctx->_zero_point)
+				netfs_write_zero_point(inode, end);
+			spin_unlock(&inode->i_lock);
+		}
+	}
+
+	folio_wait_private_2(folio); /* [DEPRECATED] */
+
+	if (!folio_test_private(folio))
+		return;
+
+	finfo = netfs_folio_info(folio);
+
+	if (offset == 0 && length >= flen)
+		goto erase_completely;
+
+	if (finfo) {
+		/* We have a partially uptodate page from a streaming write. */
+		unsigned int fstart = finfo->dirty_offset;
+		unsigned int fend = fstart + finfo->dirty_len;
+		unsigned int iend = offset + length;
+
+		if (offset >= fend)
+			return;
+		if (iend <= fstart)
+			return;
+
+		/* The invalidation region overlaps the data.  If the region
+		 * covers the start of the data, we either move along the start
+		 * or just erase the data entirely.
+		 */
+		if (offset <= fstart) {
+			if (iend >= fend)
+				goto erase_completely;
+			/* Move the start of the data. */
+			finfo->dirty_len = fend - iend;
+			finfo->dirty_offset = iend;
+			trace_netfs_folio(folio, netfs_folio_trace_invalidate_front);
+			return;
+		}
+
+		/* Reduce the length of the data if the invalidation region
+		 * covers the tail part.
+		 */
+		if (iend >= fend) {
+			finfo->dirty_len = offset - fstart;
+			trace_netfs_folio(folio, netfs_folio_trace_invalidate_tail);
+			return;
+		}
+
+		/* A partial write was split.  The caller has already zeroed
+		 * it, so just absorb the hole.
+		 */
+		trace_netfs_folio(folio, netfs_folio_trace_invalidate_middle);
+	}
+	return;
+
+erase_completely:
+	netfs_put_group(netfs_folio_group(folio));
+	folio_detach_private(folio);
+	folio_clear_uptodate(folio);
+	folio_cancel_dirty(folio);
+	kfree(finfo);
+	trace_netfs_folio(folio, netfs_folio_trace_invalidate_all);
+}
+EXPORT_SYMBOL(netfs_invalidate_folio);
+
+/**
+ * netfs_release_folio - Try to release a folio
+ * @folio: Folio proposed for release
+ * @gfp: Flags qualifying the release
+ *
+ * Request release of a folio and clean up its private state if it's not busy.
+ * Returns true if the folio can now be released, false if not
+ */
+bool netfs_release_folio(struct folio *folio, gfp_t gfp)
+{
+	struct inode *inode = folio_inode(folio);
+	struct netfs_inode *ctx = netfs_inode(inode);
+	unsigned long long i_size, remote_i_size, zero_point, end;
+
+	if (folio_test_dirty(folio))
+		return false;
+
+	netfs_read_sizes(inode, &i_size, &remote_i_size, &zero_point);
+	end = folio_next_pos(folio);
+	if (end > zero_point) {
+		spin_lock(&inode->i_lock);
+		end = umin(end, ctx->_remote_i_size);
+		if (end > ctx->_zero_point)
+			netfs_write_zero_point(inode, end);
+		spin_unlock(&inode->i_lock);
+	}
+
+	if (folio_test_private(folio))
+		return false;
+	if (unlikely(folio_test_private_2(folio))) { /* [DEPRECATED] */
+		if (current_is_kswapd() || !(gfp & __GFP_FS))
+			return false;
+		folio_wait_private_2(folio);
+	}
+	fscache_note_page_release(netfs_i_cookie(ctx));
+	return true;
+}
+EXPORT_SYMBOL(netfs_release_folio);
+
+/*
+ * Wake the collection work item.
+ */
+void netfs_wake_collector(struct netfs_io_request *rreq)
+{
+	if (test_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags) &&
+	    !test_bit(NETFS_RREQ_RETRYING, &rreq->flags)) {
+		queue_work(system_dfl_wq, &rreq->work);
+	} else {
+		trace_netfs_rreq(rreq, netfs_rreq_trace_wake_queue);
+		wake_up(&rreq->waitq);
+	}
+}
+
+/*
+ * Mark a subrequest as no longer being in progress and, if need be, wake the
+ * collector.
+ */
+void netfs_subreq_clear_in_progress(struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_request *rreq = subreq->rreq;
+	struct netfs_io_stream *stream = &rreq->io_streams[subreq->stream_nr];
+
+	clear_bit_unlock(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+	smp_mb__after_atomic(); /* Clear IN_PROGRESS before task state */
+
+	/* If we are at the head of the queue, wake up the collector. */
+	if (list_is_first(&subreq->rreq_link, &stream->subrequests) ||
+	    test_bit(NETFS_RREQ_RETRYING, &rreq->flags))
+		netfs_wake_collector(rreq);
+}
+
+/*
+ * Wait for all outstanding I/O in a stream to quiesce.
+ */
+void netfs_wait_for_in_progress_stream(struct netfs_io_request *rreq,
+				       struct netfs_io_stream *stream)
+{
+	struct netfs_io_subrequest *subreq;
+	DEFINE_WAIT(myself);
+
+	list_for_each_entry(subreq, &stream->subrequests, rreq_link) {
+		smp_rmb(); /* Read ->next before IN_PROGRESS. */
+		if (!netfs_check_subreq_in_progress(subreq))
+			continue;
+
+		trace_netfs_rreq(rreq, netfs_rreq_trace_wait_quiesce);
+		for (;;) {
+			prepare_to_wait(&rreq->waitq, &myself, TASK_UNINTERRUPTIBLE);
+
+			if (!netfs_check_subreq_in_progress(subreq))
+				break;
+
+			trace_netfs_sreq(subreq, netfs_sreq_trace_wait_for);
+			schedule();
+		}
+	}
+
+	trace_netfs_rreq(rreq, netfs_rreq_trace_waited_quiesce);
+	finish_wait(&rreq->waitq, &myself);
+}
+
+/*
+ * Perform collection in app thread if not offloaded to workqueue.
+ */
+static int netfs_collect_in_app(struct netfs_io_request *rreq,
+				bool (*collector)(struct netfs_io_request *rreq))
+{
+	bool need_collect = false, inactive = true, done = true;
+
+	if (!netfs_check_rreq_in_progress(rreq)) {
+		trace_netfs_rreq(rreq, netfs_rreq_trace_recollect);
+		return 1; /* Done */
+	}
+
+	for (int i = 0; i < NR_IO_STREAMS; i++) {
+		struct netfs_io_subrequest *subreq;
+		struct netfs_io_stream *stream = &rreq->io_streams[i];
+
+		if (!stream->active)
+			continue;
+		inactive = false;
+		trace_netfs_collect_stream(rreq, stream);
+		subreq = list_first_entry_or_null(&stream->subrequests,
+						  struct netfs_io_subrequest,
+						  rreq_link);
+		if (subreq &&
+		    (!netfs_check_subreq_in_progress(subreq) ||
+		     test_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags))) {
+			need_collect = true;
+			break;
+		}
+		if (subreq || !test_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags))
+			done = false;
+	}
+
+	if (!need_collect && !inactive && !done)
+		return 0; /* Sleep */
+
+	__set_current_state(TASK_RUNNING);
+	if (collector(rreq)) {
+		/* Drop the ref from the NETFS_RREQ_IN_PROGRESS flag. */
+		netfs_put_request(rreq, netfs_rreq_trace_put_work_ip);
+		return 1; /* Done */
+	}
+
+	if (inactive) {
+		WARN(true, "Failed to collect inactive req R=%08x\n",
+		     rreq->debug_id);
+		cond_resched();
+	}
+	return 2; /* Again */
+}
+
+/*
+ * Wait for a request to complete, successfully or otherwise.
+ */
+static ssize_t netfs_wait_for_in_progress(struct netfs_io_request *rreq,
+					  bool (*collector)(struct netfs_io_request *rreq))
+{
+	DEFINE_WAIT(myself);
+	ssize_t ret;
+
+	for (;;) {
+		prepare_to_wait(&rreq->waitq, &myself, TASK_UNINTERRUPTIBLE);
+
+		if (!test_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags)) {
+			switch (netfs_collect_in_app(rreq, collector)) {
+			case 0:
+				break;
+			case 1:
+				goto all_collected;
+			case 2:
+				if (!netfs_check_rreq_in_progress(rreq))
+					break;
+				cond_resched();
+				continue;
+			}
+		}
+
+		if (!netfs_check_rreq_in_progress(rreq))
+			break;
+
+		trace_netfs_rreq(rreq, netfs_rreq_trace_wait_ip);
+		schedule();
+	}
+
+all_collected:
+	trace_netfs_rreq(rreq, netfs_rreq_trace_waited_ip);
+	finish_wait(&rreq->waitq, &myself);
+
+	ret = rreq->error;
+	if (ret == 0) {
+		ret = rreq->transferred;
+		switch (rreq->origin) {
+		case NETFS_DIO_READ:
+		case NETFS_DIO_WRITE:
+		case NETFS_READ_SINGLE:
+		case NETFS_UNBUFFERED_READ:
+		case NETFS_UNBUFFERED_WRITE:
+			break;
+		default:
+			if (rreq->submitted < rreq->len) {
+				trace_netfs_failure(rreq, NULL, ret, netfs_fail_short_read);
+				ret = -EIO;
+			}
+			break;
+		}
+	}
+
+	return ret;
+}
+
+ssize_t netfs_wait_for_read(struct netfs_io_request *rreq)
+{
+	return netfs_wait_for_in_progress(rreq, netfs_read_collection);
+}
+
+ssize_t netfs_wait_for_write(struct netfs_io_request *rreq)
+{
+	return netfs_wait_for_in_progress(rreq, netfs_write_collection);
+}
+
+/*
+ * Wait for a paused operation to unpause or complete in some manner.
+ */
+static void netfs_wait_for_pause(struct netfs_io_request *rreq,
+				 bool (*collector)(struct netfs_io_request *rreq))
+{
+	DEFINE_WAIT(myself);
+
+	for (;;) {
+		trace_netfs_rreq(rreq, netfs_rreq_trace_wait_pause);
+		prepare_to_wait(&rreq->waitq, &myself, TASK_UNINTERRUPTIBLE);
+
+		if (!test_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags)) {
+			switch (netfs_collect_in_app(rreq, collector)) {
+			case 0:
+				break;
+			case 1:
+				goto all_collected;
+			case 2:
+				if (!netfs_check_rreq_in_progress(rreq) ||
+				    !test_bit(NETFS_RREQ_PAUSE, &rreq->flags))
+					break;
+				cond_resched();
+				continue;
+			}
+		}
+
+		if (!netfs_check_rreq_in_progress(rreq) ||
+		    !test_bit(NETFS_RREQ_PAUSE, &rreq->flags))
+			break;
+
+		schedule();
+	}
+
+all_collected:
+	trace_netfs_rreq(rreq, netfs_rreq_trace_waited_pause);
+	finish_wait(&rreq->waitq, &myself);
+}
+
+void netfs_wait_for_paused_read(struct netfs_io_request *rreq)
+{
+	return netfs_wait_for_pause(rreq, netfs_read_collection);
+}
+
+void netfs_wait_for_paused_write(struct netfs_io_request *rreq)
+{
+	return netfs_wait_for_pause(rreq, netfs_write_collection);
+}
diff --git a/fs/netfs/objects.c b/fs/netfs/objects.c
new file mode 100644
index 000000000..01461a746
--- /dev/null
+++ b/fs/netfs/objects.c
@@ -0,0 +1,260 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Object lifetime handling and tracing.
+ *
+ * Copyright (C) 2022 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/slab.h>
+#include <linux/mempool.h>
+#include "internal.h"
+
+static void netfs_free_request(struct work_struct *work);
+
+/*
+ * Allocate an I/O request and initialise it.
+ */
+struct netfs_io_request *netfs_alloc_request(struct address_space *mapping,
+					     struct file *file,
+					     loff_t start, size_t len,
+					     enum netfs_io_origin origin)
+{
+	static atomic_t debug_ids;
+	struct inode *inode = file ? file_inode(file) : mapping->host;
+	struct netfs_inode *ctx = netfs_inode(inode);
+	struct netfs_io_request *rreq;
+	mempool_t *mempool = ctx->ops->request_pool ?: &netfs_request_pool;
+	struct kmem_cache *cache = mempool->pool_data;
+	gfp_t gfp = GFP_KERNEL;
+	int ret;
+
+	/* Writeback is part of memory reclaim and must not fail due to ENOMEM. */
+	if (origin == NETFS_WRITEBACK || origin == NETFS_WRITEBACK_SINGLE) {
+		gfp = GFP_NOFS; /* Allows use of mempools. */
+
+		rreq = mempool_alloc(mempool, gfp);
+	} else {
+		rreq = mempool->alloc(gfp, mempool->pool_data);
+		if (!rreq)
+			return ERR_PTR(-ENOMEM);
+	}
+
+	memset(rreq, 0, kmem_cache_size(cache));
+	INIT_WORK(&rreq->cleanup_work, netfs_free_request);
+	rreq->gfp	= gfp;
+	rreq->start	= start;
+	rreq->len	= len;
+	rreq->origin	= origin;
+	rreq->netfs_ops	= ctx->ops;
+	rreq->mapping	= mapping;
+	rreq->inode	= inode;
+	rreq->i_size	= i_size_read(inode);
+	rreq->debug_id	= atomic_inc_return(&debug_ids);
+	rreq->wsize	= INT_MAX;
+	rreq->io_streams[0].sreq_max_len = ULONG_MAX;
+	rreq->io_streams[0].sreq_max_segs = 0;
+	spin_lock_init(&rreq->lock);
+	INIT_LIST_HEAD(&rreq->io_streams[0].subrequests);
+	INIT_LIST_HEAD(&rreq->io_streams[1].subrequests);
+	init_waitqueue_head(&rreq->waitq);
+	refcount_set(&rreq->ref, 2);
+
+	if (origin == NETFS_READAHEAD ||
+	    origin == NETFS_READPAGE ||
+	    origin == NETFS_READ_GAPS ||
+	    origin == NETFS_READ_SINGLE ||
+	    origin == NETFS_READ_FOR_WRITE ||
+	    origin == NETFS_UNBUFFERED_READ ||
+	    origin == NETFS_DIO_READ) {
+		INIT_WORK(&rreq->work, netfs_read_collection_worker);
+		rreq->io_streams[0].avail = true;
+	} else {
+		INIT_WORK(&rreq->work, netfs_write_collection_worker);
+	}
+
+	__set_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
+	if (rreq->netfs_ops->init_request) {
+		ret = rreq->netfs_ops->init_request(rreq, file);
+		if (ret < 0) {
+			mempool_free(rreq, rreq->netfs_ops->request_pool ?: &netfs_request_pool);
+			return ERR_PTR(ret);
+		}
+	}
+
+	atomic_inc(&ctx->io_count);
+	trace_netfs_rreq_ref(rreq->debug_id, refcount_read(&rreq->ref), netfs_rreq_trace_new);
+	netfs_proc_add_rreq(rreq);
+	netfs_stat(&netfs_n_rh_rreq);
+	return rreq;
+}
+
+void netfs_get_request(struct netfs_io_request *rreq, enum netfs_rreq_ref_trace what)
+{
+	int r;
+
+	__refcount_inc(&rreq->ref, &r);
+	trace_netfs_rreq_ref(rreq->debug_id, r + 1, what);
+}
+
+void netfs_clear_subrequests(struct netfs_io_request *rreq)
+{
+	struct netfs_io_subrequest *subreq;
+	struct netfs_io_stream *stream;
+	int s;
+
+	for (s = 0; s < ARRAY_SIZE(rreq->io_streams); s++) {
+		stream = &rreq->io_streams[s];
+		while (!list_empty(&stream->subrequests)) {
+			subreq = list_first_entry(&stream->subrequests,
+						  struct netfs_io_subrequest, rreq_link);
+			list_del(&subreq->rreq_link);
+			netfs_put_subrequest(subreq, netfs_sreq_trace_put_clear);
+		}
+	}
+}
+
+static void netfs_free_request_rcu(struct rcu_head *rcu)
+{
+	struct netfs_io_request *rreq = container_of(rcu, struct netfs_io_request, rcu);
+
+	mempool_free(rreq, rreq->netfs_ops->request_pool ?: &netfs_request_pool);
+	netfs_stat_d(&netfs_n_rh_rreq);
+}
+
+static void netfs_deinit_request(struct netfs_io_request *rreq)
+{
+	struct netfs_inode *ictx = netfs_inode(rreq->inode);
+	unsigned int i;
+
+	trace_netfs_rreq(rreq, netfs_rreq_trace_free);
+
+	/* Cancel/flush the result collection worker.  That does not carry a
+	 * ref of its own, so we must wait for it somewhere.
+	 */
+	cancel_work_sync(&rreq->work);
+
+	netfs_proc_del_rreq(rreq);
+	netfs_clear_subrequests(rreq);
+	if (rreq->netfs_ops->free_request)
+		rreq->netfs_ops->free_request(rreq);
+	if (rreq->cache_resources.ops)
+		rreq->cache_resources.ops->end_operation(&rreq->cache_resources);
+	if (rreq->direct_bv) {
+		for (i = 0; i < rreq->direct_bv_count; i++) {
+			if (rreq->direct_bv[i].bv_page) {
+				if (rreq->direct_bv_unpin)
+					unpin_user_page(rreq->direct_bv[i].bv_page);
+			}
+		}
+		kvfree(rreq->direct_bv);
+	}
+	rolling_buffer_clear(&rreq->buffer);
+
+	if (atomic_dec_and_test(&ictx->io_count))
+		wake_up_var(&ictx->io_count);
+}
+
+static void netfs_free_request(struct work_struct *work)
+{
+	struct netfs_io_request *rreq =
+		container_of(work, struct netfs_io_request, cleanup_work);
+
+	netfs_deinit_request(rreq);
+	call_rcu(&rreq->rcu, netfs_free_request_rcu);
+}
+
+void netfs_put_request(struct netfs_io_request *rreq, enum netfs_rreq_ref_trace what)
+{
+	unsigned int debug_id;
+	bool dead;
+	int r;
+
+	if (rreq) {
+		debug_id = rreq->debug_id;
+		dead = __refcount_dec_and_test(&rreq->ref, &r);
+		trace_netfs_rreq_ref(debug_id, r - 1, what);
+		if (dead)
+			WARN_ON(!queue_work(system_dfl_wq, &rreq->cleanup_work));
+	}
+}
+
+/*
+ * Free a request (synchronously) that was just allocated but has
+ * failed before it could be submitted.
+ */
+void netfs_put_failed_request(struct netfs_io_request *rreq)
+{
+	int r = refcount_read(&rreq->ref);
+
+	/* new requests have two references (see
+	 * netfs_alloc_request(), and this function is only allowed on
+	 * new request objects
+	 */
+	WARN_ON_ONCE(r != 2);
+
+	trace_netfs_rreq_ref(rreq->debug_id, r, netfs_rreq_trace_put_failed);
+	netfs_free_request(&rreq->cleanup_work);
+}
+
+/*
+ * Allocate and partially initialise an I/O request structure.
+ */
+struct netfs_io_subrequest *netfs_alloc_subrequest(struct netfs_io_request *rreq)
+{
+	struct netfs_io_subrequest *subreq;
+	mempool_t *mempool = rreq->netfs_ops->subrequest_pool ?: &netfs_subrequest_pool;
+	struct kmem_cache *cache = mempool->pool_data;
+
+	if (rreq->gfp == GFP_KERNEL)
+		subreq = mempool->alloc(rreq->gfp, mempool->pool_data);
+	else
+		subreq = mempool_alloc(mempool, rreq->gfp);
+	if (!subreq)
+		return NULL;
+
+	memset(subreq, 0, kmem_cache_size(cache));
+	INIT_WORK(&subreq->work, NULL);
+	INIT_LIST_HEAD(&subreq->rreq_link);
+	refcount_set(&subreq->ref, 2);
+	subreq->rreq = rreq;
+	subreq->debug_index = atomic_inc_return(&rreq->subreq_counter);
+	netfs_get_request(rreq, netfs_rreq_trace_get_subreq);
+	netfs_stat(&netfs_n_rh_sreq);
+	return subreq;
+}
+
+void netfs_get_subrequest(struct netfs_io_subrequest *subreq,
+			  enum netfs_sreq_ref_trace what)
+{
+	int r;
+
+	__refcount_inc(&subreq->ref, &r);
+	trace_netfs_sreq_ref(subreq->rreq->debug_id, subreq->debug_index, r + 1,
+			     what);
+}
+
+static void netfs_free_subrequest(struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_request *rreq = subreq->rreq;
+
+	trace_netfs_sreq(subreq, netfs_sreq_trace_free);
+	if (rreq->netfs_ops->free_subrequest)
+		rreq->netfs_ops->free_subrequest(subreq);
+	mempool_free(subreq, rreq->netfs_ops->subrequest_pool ?: &netfs_subrequest_pool);
+	netfs_stat_d(&netfs_n_rh_sreq);
+	netfs_put_request(rreq, netfs_rreq_trace_put_subreq);
+}
+
+void netfs_put_subrequest(struct netfs_io_subrequest *subreq,
+			  enum netfs_sreq_ref_trace what)
+{
+	unsigned int debug_index = subreq->debug_index;
+	unsigned int debug_id = subreq->rreq->debug_id;
+	bool dead;
+	int r;
+
+	dead = __refcount_dec_and_test(&subreq->ref, &r);
+	trace_netfs_sreq_ref(debug_id, debug_index, r - 1, what);
+	if (dead)
+		netfs_free_subrequest(subreq);
+}
diff --git a/fs/netfs/read_collect.c b/fs/netfs/read_collect.c
new file mode 100644
index 000000000..23660a590
--- /dev/null
+++ b/fs/netfs/read_collect.c
@@ -0,0 +1,608 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Network filesystem read subrequest result collection, assessment and
+ * retrying.
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include <linux/task_io_accounting_ops.h>
+#include "internal.h"
+
+/* Notes made in the collector */
+#define HIT_PENDING	0x01	/* A front op was still pending */
+#define MADE_PROGRESS	0x04	/* Made progress cleaning up a stream or the folio set */
+#define BUFFERED	0x08	/* The pagecache needs cleaning up */
+#define NEED_RETRY	0x10	/* A front op requests retrying */
+#define COPY_TO_CACHE	0x40	/* Need to copy subrequest to cache */
+#define ABANDON_SREQ	0x80	/* Need to abandon untransferred part of subrequest */
+
+/*
+ * Clear the unread part of an I/O request.
+ */
+static void netfs_clear_unread(struct netfs_io_subrequest *subreq)
+{
+	netfs_reset_iter(subreq);
+	WARN_ON_ONCE(subreq->len - subreq->transferred != iov_iter_count(&subreq->io_iter));
+	iov_iter_zero(iov_iter_count(&subreq->io_iter), &subreq->io_iter);
+	if (subreq->start + subreq->transferred >= subreq->rreq->i_size)
+		__set_bit(NETFS_SREQ_HIT_EOF, &subreq->flags);
+}
+
+/*
+ * Flush, mark and unlock a folio that's now completely read.  If we want to
+ * cache the folio, we set the group to NETFS_FOLIO_COPY_TO_CACHE, mark it
+ * dirty and let writeback handle it.
+ */
+static void netfs_unlock_read_folio(struct netfs_io_request *rreq,
+				    struct folio_queue *folioq,
+				    int slot)
+{
+	struct netfs_folio *finfo;
+	struct folio *folio = folioq_folio(folioq, slot);
+
+	if (unlikely(folio_pos(folio) < rreq->abandon_to)) {
+		trace_netfs_folio(folio, netfs_folio_trace_abandon);
+		goto just_unlock;
+	}
+
+	flush_dcache_folio(folio);
+	folio_mark_uptodate(folio);
+
+	if (!test_bit(NETFS_RREQ_USE_PGPRIV2, &rreq->flags)) {
+		finfo = netfs_folio_info(folio);
+		if (finfo) {
+			trace_netfs_folio(folio, netfs_folio_trace_filled_gaps);
+			if (finfo->netfs_group)
+				folio_change_private(folio, finfo->netfs_group);
+			else
+				folio_detach_private(folio);
+			kfree(finfo);
+		}
+
+		if (test_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags)) {
+			if (!WARN_ON_ONCE(folio_get_private(folio) != NULL)) {
+				trace_netfs_folio(folio, netfs_folio_trace_copy_to_cache);
+				folio_attach_private(folio, NETFS_FOLIO_COPY_TO_CACHE);
+				folio_mark_dirty(folio);
+			}
+		} else {
+			trace_netfs_folio(folio, netfs_folio_trace_read_done);
+		}
+
+		folioq_clear(folioq, slot);
+	} else {
+		// TODO: Use of PG_private_2 is deprecated.
+		if (test_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags))
+			netfs_pgpriv2_copy_to_cache(rreq, folio);
+	}
+
+just_unlock:
+	if (folio == rreq->no_unlock_folio &&
+	    test_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags)) {
+		_debug("no unlock");
+	} else {
+		trace_netfs_folio(folio, netfs_folio_trace_read_unlock);
+		folio_unlock(folio);
+	}
+
+	folioq_clear(folioq, slot);
+}
+
+/*
+ * Unlock any folios we've finished with.
+ */
+static void netfs_read_unlock_folios(struct netfs_io_request *rreq,
+				     unsigned int *notes)
+{
+	struct folio_queue *folioq = rreq->buffer.tail;
+	unsigned long long collected_to = rreq->collected_to;
+	unsigned int slot = rreq->buffer.first_tail_slot;
+
+	if (rreq->cleaned_to >= rreq->collected_to)
+		return;
+
+	// TODO: Begin decryption
+
+	if (slot >= folioq_nr_slots(folioq)) {
+		folioq = rolling_buffer_delete_spent(&rreq->buffer);
+		if (!folioq) {
+			rreq->front_folio_order = 0;
+			return;
+		}
+		slot = 0;
+	}
+
+	for (;;) {
+		struct folio *folio;
+		unsigned long long fpos, fend;
+		unsigned int order;
+		size_t fsize;
+
+		if (*notes & COPY_TO_CACHE)
+			set_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags);
+
+		folio = folioq_folio(folioq, slot);
+		if (WARN_ONCE(!folio_test_locked(folio),
+			      "R=%08x: folio %lx is not locked\n",
+			      rreq->debug_id, folio->index))
+			trace_netfs_folio(folio, netfs_folio_trace_not_locked);
+
+		order = folioq_folio_order(folioq, slot);
+		rreq->front_folio_order = order;
+		fsize = PAGE_SIZE << order;
+		fpos = folio_pos(folio);
+		fend = fpos + fsize;
+
+		trace_netfs_collect_folio(rreq, folio, fend, collected_to);
+
+		/* Unlock any folio we've transferred all of. */
+		if (collected_to < fend)
+			break;
+
+		netfs_unlock_read_folio(rreq, folioq, slot);
+		WRITE_ONCE(rreq->cleaned_to, fpos + fsize);
+		*notes |= MADE_PROGRESS;
+
+		clear_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags);
+
+		/* Clean up the head folioq.  If we clear an entire folioq, then
+		 * we can get rid of it provided it's not also the tail folioq
+		 * being filled by the issuer.
+		 */
+		folioq_clear(folioq, slot);
+		slot++;
+		if (slot >= folioq_nr_slots(folioq)) {
+			folioq = rolling_buffer_delete_spent(&rreq->buffer);
+			if (!folioq)
+				goto done;
+			slot = 0;
+			trace_netfs_folioq(folioq, netfs_trace_folioq_read_progress);
+		}
+
+		if (fpos + fsize >= collected_to)
+			break;
+	}
+
+	rreq->buffer.tail = folioq;
+done:
+	rreq->buffer.first_tail_slot = slot;
+}
+
+/*
+ * Collect and assess the results of various read subrequests.  We may need to
+ * retry some of the results.
+ *
+ * Note that we have a sequence of subrequests, which may be drawing on
+ * different sources and may or may not be the same size or starting position
+ * and may not even correspond in boundary alignment.
+ */
+static void netfs_collect_read_results(struct netfs_io_request *rreq)
+{
+	struct netfs_io_subrequest *front, *remove;
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+	unsigned int notes;
+
+	_enter("%llx-%llx", rreq->start, rreq->start + rreq->len);
+	trace_netfs_rreq(rreq, netfs_rreq_trace_collect);
+	trace_netfs_collect(rreq);
+
+reassess:
+	if (rreq->origin == NETFS_READAHEAD ||
+	    rreq->origin == NETFS_READPAGE ||
+	    rreq->origin == NETFS_READ_FOR_WRITE)
+		notes = BUFFERED;
+	else
+		notes = 0;
+
+	/* Remove completed subrequests from the front of the stream and
+	 * advance the completion point.  We stop when we hit something that's
+	 * in progress.  The issuer thread may be adding stuff to the tail
+	 * whilst we're doing this.
+	 */
+	front = list_first_entry_or_null_acquire(&stream->subrequests,
+						 struct netfs_io_subrequest, rreq_link);
+	/* Read first subreq pointer before IN_PROGRESS flag. */
+
+	while (front) {
+		size_t transferred;
+
+		trace_netfs_collect_sreq(rreq, front);
+		_debug("sreq [%x] %llx %zx/%zx",
+		       front->debug_index, front->start, front->transferred, front->len);
+
+		if (stream->collected_to < front->start) {
+			trace_netfs_collect_gap(rreq, stream, front->start, 'F');
+			stream->collected_to = front->start;
+		}
+
+		if (netfs_check_subreq_in_progress(front))
+			notes |= HIT_PENDING;
+		smp_rmb(); /* Read counters after IN_PROGRESS flag. */
+		transferred = READ_ONCE(front->transferred);
+
+		/* If we can now collect the next folio, do so.  We don't want
+		 * to defer this as we have to decide whether we need to copy
+		 * to the cache or not, and that may differ between adjacent
+		 * subreqs.
+		 */
+		if (notes & BUFFERED) {
+			size_t fsize = PAGE_SIZE << rreq->front_folio_order;
+
+			/* Clear the tail of a short read. */
+			if (!(notes & HIT_PENDING) &&
+			    front->error == 0 &&
+			    transferred < front->len &&
+			    (test_bit(NETFS_SREQ_HIT_EOF, &front->flags) ||
+			     test_bit(NETFS_SREQ_CLEAR_TAIL, &front->flags))) {
+				netfs_clear_unread(front);
+				transferred = front->transferred = front->len;
+				trace_netfs_sreq(front, netfs_sreq_trace_clear);
+			}
+
+			stream->collected_to = front->start + transferred;
+			rreq->collected_to = stream->collected_to;
+
+			if (test_bit(NETFS_SREQ_COPY_TO_CACHE, &front->flags))
+				notes |= COPY_TO_CACHE;
+
+			if (test_bit(NETFS_SREQ_FAILED, &front->flags)) {
+				rreq->abandon_to = front->start + front->len;
+				front->transferred = front->len;
+				transferred = front->len;
+				trace_netfs_rreq(rreq, netfs_rreq_trace_set_abandon);
+			}
+			if (front->start + transferred >= rreq->cleaned_to + fsize ||
+			    test_bit(NETFS_SREQ_HIT_EOF, &front->flags))
+				netfs_read_unlock_folios(rreq, &notes);
+		} else {
+			stream->collected_to = front->start + transferred;
+			rreq->collected_to = stream->collected_to;
+		}
+
+		/* Stall if the front is still undergoing I/O. */
+		if (notes & HIT_PENDING)
+			break;
+
+		if (test_bit(NETFS_SREQ_FAILED, &front->flags)) {
+			if (!stream->failed) {
+				stream->error = front->error;
+				rreq->error = front->error;
+				set_bit(NETFS_RREQ_FAILED, &rreq->flags);
+				stream->failed = true;
+			}
+			notes |= MADE_PROGRESS | ABANDON_SREQ;
+		} else if (test_bit(NETFS_SREQ_NEED_RETRY, &front->flags)) {
+			stream->need_retry = true;
+			notes |= NEED_RETRY | MADE_PROGRESS;
+			break;
+		} else if (test_bit(NETFS_RREQ_SHORT_TRANSFER, &rreq->flags)) {
+			notes |= MADE_PROGRESS;
+		} else {
+			if (!stream->failed) {
+				stream->transferred += transferred;
+				stream->transferred_valid = true;
+			}
+			if (front->transferred < front->len)
+				set_bit(NETFS_RREQ_SHORT_TRANSFER, &rreq->flags);
+			notes |= MADE_PROGRESS;
+		}
+
+		/* Remove if completely consumed. */
+		stream->source = front->source;
+		spin_lock(&rreq->lock);
+
+		remove = front;
+		trace_netfs_sreq(front,
+				 notes & ABANDON_SREQ ?
+				 netfs_sreq_trace_abandoned : netfs_sreq_trace_consumed);
+		list_del_init(&front->rreq_link);
+		front = list_first_entry_or_null(&stream->subrequests,
+						 struct netfs_io_subrequest, rreq_link);
+		spin_unlock(&rreq->lock);
+		netfs_put_subrequest(remove,
+				     notes & ABANDON_SREQ ?
+				     netfs_sreq_trace_put_abandon :
+				     netfs_sreq_trace_put_done);
+	}
+
+	trace_netfs_collect_stream(rreq, stream);
+	trace_netfs_collect_state(rreq, rreq->collected_to, notes);
+
+	if (!(notes & BUFFERED))
+		rreq->cleaned_to = rreq->collected_to;
+
+	if (notes & NEED_RETRY)
+		goto need_retry;
+	if (notes & MADE_PROGRESS) {
+		netfs_wake_rreq_flag(rreq, NETFS_RREQ_PAUSE, netfs_rreq_trace_unpause);
+		//cond_resched();
+		goto reassess;
+	}
+
+out:
+	_leave(" = %x", notes);
+	return;
+
+need_retry:
+	/* Okay...  We're going to have to retry parts of the stream.  Note
+	 * that any partially completed op will have had any wholly transferred
+	 * folios removed from it.
+	 */
+	_debug("retry");
+	netfs_retry_reads(rreq);
+	goto out;
+}
+
+/*
+ * Do page flushing and suchlike after DIO.
+ */
+static void netfs_rreq_assess_dio(struct netfs_io_request *rreq)
+{
+	unsigned int i;
+
+	if (rreq->origin == NETFS_UNBUFFERED_READ ||
+	    rreq->origin == NETFS_DIO_READ) {
+		for (i = 0; i < rreq->direct_bv_count; i++) {
+			flush_dcache_page(rreq->direct_bv[i].bv_page);
+			// TODO: cifs marks pages in the destination buffer
+			// dirty under some circumstances after a read.  Do we
+			// need to do that too?
+			set_page_dirty(rreq->direct_bv[i].bv_page);
+		}
+	}
+
+	if (rreq->iocb) {
+		rreq->iocb->ki_pos += rreq->transferred;
+		if (rreq->iocb->ki_complete) {
+			trace_netfs_rreq(rreq, netfs_rreq_trace_ki_complete);
+			rreq->iocb->ki_complete(
+				rreq->iocb, rreq->error ? rreq->error : rreq->transferred);
+		}
+	}
+	if (rreq->netfs_ops->done)
+		rreq->netfs_ops->done(rreq);
+	if (rreq->origin == NETFS_UNBUFFERED_READ ||
+	    rreq->origin == NETFS_DIO_READ)
+		inode_dio_end(rreq->inode);
+}
+
+/*
+ * Do processing after reading a monolithic single object.
+ */
+static void netfs_rreq_assess_single(struct netfs_io_request *rreq)
+{
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+
+	if (!rreq->error && stream->source == NETFS_DOWNLOAD_FROM_SERVER &&
+	    fscache_resources_valid(&rreq->cache_resources)) {
+		trace_netfs_rreq(rreq, netfs_rreq_trace_dirty);
+		netfs_single_mark_inode_dirty(rreq->inode);
+	}
+
+	if (rreq->iocb) {
+		rreq->iocb->ki_pos += rreq->transferred;
+		if (rreq->iocb->ki_complete) {
+			trace_netfs_rreq(rreq, netfs_rreq_trace_ki_complete);
+			rreq->iocb->ki_complete(
+				rreq->iocb, rreq->error ? rreq->error : rreq->transferred);
+		}
+	}
+	if (rreq->netfs_ops->done)
+		rreq->netfs_ops->done(rreq);
+}
+
+/*
+ * Perform the collection of subrequests and folios.
+ *
+ * Note that we're in normal kernel thread context at this point, possibly
+ * running on a workqueue.
+ */
+bool netfs_read_collection(struct netfs_io_request *rreq)
+{
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+
+	netfs_collect_read_results(rreq);
+
+	/* We're done when the app thread has finished posting subreqs and the
+	 * queue is empty.
+	 */
+	if (!test_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags))
+		return false;
+	smp_rmb(); /* Read ALL_QUEUED before subreq lists. */
+
+	if (!list_empty(&stream->subrequests))
+		return false;
+
+	/* Okay, declare that all I/O is complete. */
+	rreq->transferred = stream->transferred;
+	trace_netfs_rreq(rreq, netfs_rreq_trace_complete);
+
+	//netfs_rreq_is_still_valid(rreq);
+
+	switch (rreq->origin) {
+	case NETFS_UNBUFFERED_READ:
+	case NETFS_DIO_READ:
+	case NETFS_READ_GAPS:
+		netfs_rreq_assess_dio(rreq);
+		break;
+	case NETFS_READ_SINGLE:
+		netfs_rreq_assess_single(rreq);
+		break;
+	default:
+		break;
+	}
+	task_io_account_read(rreq->transferred);
+
+	netfs_wake_rreq_flag(rreq, NETFS_RREQ_IN_PROGRESS, netfs_rreq_trace_wake_ip);
+	/* As we cleared NETFS_RREQ_IN_PROGRESS, we acquired its ref. */
+
+	trace_netfs_rreq(rreq, netfs_rreq_trace_done);
+	netfs_clear_subrequests(rreq);
+	netfs_unlock_abandoned_read_pages(rreq);
+	if (unlikely(rreq->copy_to_cache))
+		netfs_pgpriv2_end_copy_to_cache(rreq);
+	return true;
+}
+
+void netfs_read_collection_worker(struct work_struct *work)
+{
+	struct netfs_io_request *rreq = container_of(work, struct netfs_io_request, work);
+
+	netfs_see_request(rreq, netfs_rreq_trace_see_work);
+	if (netfs_check_rreq_in_progress(rreq)) {
+		if (netfs_read_collection(rreq))
+			/* Drop the ref from the IN_PROGRESS flag. */
+			netfs_put_request(rreq, netfs_rreq_trace_put_work_ip);
+		else
+			netfs_see_request(rreq, netfs_rreq_trace_see_work_complete);
+	}
+}
+
+/**
+ * netfs_read_subreq_progress - Note progress of a read operation.
+ * @subreq: The read request that has terminated.
+ *
+ * This tells the read side of netfs lib that a contributory I/O operation has
+ * made some progress and that it may be possible to unlock some folios.
+ *
+ * Before calling, the filesystem should update subreq->transferred to track
+ * the amount of data copied into the output buffer.
+ */
+void netfs_read_subreq_progress(struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_request *rreq = subreq->rreq;
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+	size_t fsize = PAGE_SIZE << rreq->front_folio_order;
+
+	trace_netfs_sreq(subreq, netfs_sreq_trace_progress);
+
+	/* If we are at the head of the queue, wake up the collector,
+	 * getting a ref to it if we were the ones to do so.
+	 */
+	if (subreq->start + subreq->transferred > rreq->cleaned_to + fsize &&
+	    (rreq->origin == NETFS_READAHEAD ||
+	     rreq->origin == NETFS_READPAGE ||
+	     rreq->origin == NETFS_READ_FOR_WRITE) &&
+	    list_is_first(&subreq->rreq_link, &stream->subrequests)
+	    ) {
+		__set_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+		netfs_wake_collector(rreq);
+	}
+}
+EXPORT_SYMBOL(netfs_read_subreq_progress);
+
+/**
+ * netfs_read_subreq_terminated - Note the termination of an I/O operation.
+ * @subreq: The I/O request that has terminated.
+ *
+ * This tells the read helper that a contributory I/O operation has terminated,
+ * one way or another, and that it should integrate the results.
+ *
+ * The caller indicates the outcome of the operation through @subreq->error,
+ * supplying 0 to indicate a successful or retryable transfer (if
+ * NETFS_SREQ_NEED_RETRY is set) or a negative error code.  The helper will
+ * look after reissuing I/O operations as appropriate and writing downloaded
+ * data to the cache.
+ *
+ * Before calling, the filesystem should update subreq->transferred to track
+ * the amount of data copied into the output buffer.
+ */
+void netfs_read_subreq_terminated(struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_request *rreq = subreq->rreq;
+
+	switch (subreq->source) {
+	case NETFS_READ_FROM_CACHE:
+		netfs_stat(&netfs_n_rh_read_done);
+		break;
+	case NETFS_DOWNLOAD_FROM_SERVER:
+		netfs_stat(&netfs_n_rh_download_done);
+		break;
+	default:
+		break;
+	}
+
+	/* Deal with retry requests, short reads and errors.  If we retry
+	 * but don't make progress, we abandon the attempt.
+	 */
+	if (!subreq->error && subreq->transferred < subreq->len) {
+		if (test_bit(NETFS_SREQ_HIT_EOF, &subreq->flags)) {
+			trace_netfs_sreq(subreq, netfs_sreq_trace_hit_eof);
+		} else if (test_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags)) {
+			trace_netfs_sreq(subreq, netfs_sreq_trace_need_clear);
+		} else if (test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
+			trace_netfs_sreq(subreq, netfs_sreq_trace_need_retry);
+		} else if (test_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags)) {
+			__set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+			trace_netfs_sreq(subreq, netfs_sreq_trace_partial_read);
+		} else {
+			__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+			subreq->error = -ENODATA;
+			trace_netfs_sreq(subreq, netfs_sreq_trace_short);
+		}
+	}
+
+	/* If need retry is set, error should not matter unless we hit too many
+	 * retries. Pause the generation of new subreqs
+	 */
+	if (test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
+		trace_netfs_rreq(rreq, netfs_rreq_trace_set_pause);
+		set_bit(NETFS_RREQ_PAUSE, &rreq->flags);
+		goto skip_error_checks;
+	}
+
+	if (unlikely(subreq->error < 0)) {
+		trace_netfs_failure(rreq, subreq, subreq->error, netfs_fail_read);
+		if (subreq->source == NETFS_READ_FROM_CACHE) {
+			netfs_stat(&netfs_n_rh_read_failed);
+			__set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+		} else {
+			netfs_stat(&netfs_n_rh_download_failed);
+			__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+		}
+		trace_netfs_rreq(rreq, netfs_rreq_trace_set_pause);
+		set_bit(NETFS_RREQ_PAUSE, &rreq->flags);
+	}
+
+skip_error_checks:
+	trace_netfs_sreq(subreq, netfs_sreq_trace_terminated);
+	netfs_subreq_clear_in_progress(subreq);
+	netfs_put_subrequest(subreq, netfs_sreq_trace_put_terminated);
+}
+EXPORT_SYMBOL(netfs_read_subreq_terminated);
+
+/*
+ * Cancel a read subrequest due to preparation failure.
+ */
+void netfs_cancel_read(struct netfs_io_subrequest *subreq, int error)
+{
+	trace_netfs_sreq(subreq, netfs_sreq_trace_cancel);
+	subreq->error = error;
+	__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+	netfs_read_subreq_terminated(subreq);
+}
+
+/*
+ * Handle termination of a read from the cache.
+ */
+void netfs_cache_read_terminated(void *priv, ssize_t transferred_or_error)
+{
+	struct netfs_io_subrequest *subreq = priv;
+
+	if (transferred_or_error > 0) {
+		subreq->error = 0;
+		if (transferred_or_error > 0) {
+			subreq->transferred += transferred_or_error;
+			__set_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+		}
+	} else {
+		subreq->error = transferred_or_error;
+	}
+	netfs_read_subreq_terminated(subreq);
+}
diff --git a/fs/netfs/read_pgpriv2.c b/fs/netfs/read_pgpriv2.c
new file mode 100644
index 000000000..c31190993
--- /dev/null
+++ b/fs/netfs/read_pgpriv2.c
@@ -0,0 +1,233 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Read with PG_private_2 [DEPRECATED].
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include <linux/task_io_accounting_ops.h>
+#include "internal.h"
+
+/*
+ * [DEPRECATED] Copy a folio to the cache with PG_private_2 set.
+ */
+static void netfs_pgpriv2_copy_folio(struct netfs_io_request *creq, struct folio *folio)
+{
+	struct netfs_io_stream *cache = &creq->io_streams[1];
+	size_t fsize = folio_size(folio), flen = fsize;
+	loff_t fpos = folio_pos(folio), i_size;
+	bool to_eof = false;
+
+	_enter("");
+
+	/* netfs_perform_write() may shift i_size around the page or from out
+	 * of the page to beyond it, but cannot move i_size into or through the
+	 * page since we have it locked.
+	 */
+	i_size = i_size_read(creq->inode);
+
+	if (fpos >= i_size) {
+		/* mmap beyond eof. */
+		_debug("beyond eof");
+		folio_end_private_2(folio);
+		return;
+	}
+
+	if (fpos + fsize > creq->i_size)
+		creq->i_size = i_size;
+
+	if (flen > i_size - fpos) {
+		flen = i_size - fpos;
+		to_eof = true;
+	} else if (flen == i_size - fpos) {
+		to_eof = true;
+	}
+
+	_debug("folio %zx %zx", flen, fsize);
+
+	trace_netfs_folio(folio, netfs_folio_trace_store_copy);
+
+	/* Attach the folio to the rolling buffer. */
+	if (rolling_buffer_append(&creq->buffer, folio, 0, creq->gfp) < 0) {
+		folio_end_private_2(folio);
+		clear_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &creq->flags);
+		return;
+	}
+
+	cache->submit_extendable_to = fsize;
+	cache->submit_off = 0;
+	cache->submit_len = flen;
+
+	/* Attach the folio to one or more subrequests.  For a big folio, we
+	 * could end up with thousands of subrequests if the wsize is small -
+	 * but we might need to wait during the creation of subrequests for
+	 * network resources (eg. SMB credits).
+	 */
+	do {
+		ssize_t part;
+
+		creq->buffer.iter.iov_offset = cache->submit_off;
+
+		atomic64_set(&creq->issued_to, fpos + cache->submit_off);
+		cache->submit_extendable_to = fsize - cache->submit_off;
+		part = netfs_advance_write(creq, cache, fpos + cache->submit_off,
+					   cache->submit_len, to_eof);
+		cache->submit_off += part;
+		if (part > cache->submit_len)
+			cache->submit_len = 0;
+		else
+			cache->submit_len -= part;
+	} while (cache->submit_len > 0);
+
+	creq->buffer.iter.iov_offset = 0;
+	rolling_buffer_advance(&creq->buffer, fsize);
+	atomic64_set(&creq->issued_to, fpos + fsize);
+
+	if (flen < fsize)
+		netfs_issue_write(creq, cache);
+}
+
+/*
+ * [DEPRECATED] Set up copying to the cache.
+ */
+static struct netfs_io_request *netfs_pgpriv2_begin_copy_to_cache(
+	struct netfs_io_request *rreq, struct folio *folio)
+{
+	struct netfs_io_request *creq;
+
+	if (!fscache_resources_valid(&rreq->cache_resources))
+		goto cancel;
+
+	creq = netfs_create_write_req(rreq->mapping, NULL, folio_pos(folio),
+				      NETFS_PGPRIV2_COPY_TO_CACHE);
+	if (IS_ERR(creq))
+		goto cancel;
+
+	if (!creq->io_streams[1].avail)
+		goto cancel_put;
+
+	__set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &creq->flags);
+	trace_netfs_copy2cache(rreq, creq);
+	trace_netfs_write(creq, netfs_write_trace_copy_to_cache);
+	netfs_stat(&netfs_n_wh_copy_to_cache);
+	rreq->copy_to_cache = creq;
+	return creq;
+
+cancel_put:
+	netfs_put_failed_request(creq);
+cancel:
+	rreq->copy_to_cache = ERR_PTR(-ENOBUFS);
+	clear_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags);
+	return ERR_PTR(-ENOBUFS);
+}
+
+/*
+ * [DEPRECATED] Mark page as requiring copy-to-cache using PG_private_2 and add
+ * it to the copy write request.
+ */
+void netfs_pgpriv2_copy_to_cache(struct netfs_io_request *rreq, struct folio *folio)
+{
+	struct netfs_io_request *creq = rreq->copy_to_cache;
+
+	if (!creq)
+		creq = netfs_pgpriv2_begin_copy_to_cache(rreq, folio);
+	if (IS_ERR(creq))
+		return;
+
+	trace_netfs_folio(folio, netfs_folio_trace_copy_to_cache);
+	folio_start_private_2(folio);
+	netfs_pgpriv2_copy_folio(creq, folio);
+}
+
+/*
+ * [DEPRECATED] End writing to the cache, flushing out any outstanding writes.
+ */
+void netfs_pgpriv2_end_copy_to_cache(struct netfs_io_request *rreq)
+{
+	struct netfs_io_request *creq = rreq->copy_to_cache;
+
+	if (IS_ERR_OR_NULL(creq))
+		return;
+
+	netfs_issue_write(creq, &creq->io_streams[1]);
+	smp_wmb(); /* Write lists before ALL_QUEUED. */
+	set_bit(NETFS_RREQ_ALL_QUEUED, &creq->flags);
+	trace_netfs_rreq(rreq, netfs_rreq_trace_end_copy_to_cache);
+	if (list_empty_careful(&creq->io_streams[1].subrequests))
+		netfs_wake_collector(creq);
+
+	netfs_put_request(creq, netfs_rreq_trace_put_return);
+	creq->copy_to_cache = NULL;
+}
+
+/*
+ * [DEPRECATED] Remove the PG_private_2 mark from any folios we've finished
+ * copying.
+ */
+bool netfs_pgpriv2_unlock_copied_folios(struct netfs_io_request *creq)
+{
+	struct folio_queue *folioq = creq->buffer.tail;
+	unsigned long long collected_to = creq->collected_to;
+	unsigned int slot = creq->buffer.first_tail_slot;
+	bool made_progress = false;
+
+	if (slot >= folioq_nr_slots(folioq)) {
+		folioq = rolling_buffer_delete_spent(&creq->buffer);
+		slot = 0;
+	}
+
+	for (;;) {
+		struct folio *folio;
+		unsigned long long fpos, fend;
+		size_t fsize, flen;
+
+		folio = folioq_folio(folioq, slot);
+		if (WARN_ONCE(!folio_test_private_2(folio),
+			      "R=%08x: folio %lx is not marked private_2\n",
+			      creq->debug_id, folio->index))
+			trace_netfs_folio(folio, netfs_folio_trace_not_under_wback);
+
+		fpos = folio_pos(folio);
+		fsize = folio_size(folio);
+		flen = fsize;
+
+		fend = min_t(unsigned long long, fpos + flen, creq->i_size);
+
+		trace_netfs_collect_folio(creq, folio, fend, collected_to);
+
+		/* Unlock any folio we've transferred all of. */
+		if (collected_to < fend)
+			break;
+
+		trace_netfs_folio(folio, netfs_folio_trace_end_copy);
+		folio_end_private_2(folio);
+		creq->cleaned_to = fpos + fsize;
+		made_progress = true;
+
+		/* Clean up the head folioq.  If we clear an entire folioq, then
+		 * we can get rid of it provided it's not also the tail folioq
+		 * being filled by the issuer.
+		 */
+		folioq_clear(folioq, slot);
+		slot++;
+		if (slot >= folioq_nr_slots(folioq)) {
+			folioq = rolling_buffer_delete_spent(&creq->buffer);
+			if (!folioq)
+				goto done;
+			slot = 0;
+		}
+
+		if (fpos + fsize >= collected_to)
+			break;
+	}
+
+	creq->buffer.tail = folioq;
+done:
+	creq->buffer.first_tail_slot = slot;
+	return made_progress;
+}
diff --git a/fs/netfs/read_retry.c b/fs/netfs/read_retry.c
new file mode 100644
index 000000000..2b42758e0
--- /dev/null
+++ b/fs/netfs/read_retry.c
@@ -0,0 +1,312 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Network filesystem read subrequest retrying.
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+static void netfs_reissue_read(struct netfs_io_request *rreq,
+			       struct netfs_io_subrequest *subreq)
+{
+	subreq->error = 0;
+	__clear_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+	__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+	netfs_stat(&netfs_n_rh_retry_read_subreq);
+	subreq->rreq->netfs_ops->issue_read(subreq);
+}
+
+/*
+ * Go through the list of failed/short reads, retrying all retryable ones.  We
+ * need to switch failed cache reads to network downloads.
+ */
+static void netfs_retry_read_subrequests(struct netfs_io_request *rreq)
+{
+	struct netfs_io_subrequest *subreq;
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+	struct list_head *next;
+
+	_enter("R=%x", rreq->debug_id);
+
+	if (list_empty(&stream->subrequests))
+		return;
+
+	if (rreq->netfs_ops->retry_request)
+		rreq->netfs_ops->retry_request(rreq, NULL);
+
+	/* If there's no renegotiation to do, just resend each retryable subreq
+	 * up to the first permanently failed one.
+	 */
+	if (!rreq->netfs_ops->prepare_read &&
+	    !rreq->cache_resources.ops) {
+		list_for_each_entry(subreq, &stream->subrequests, rreq_link) {
+			if (test_bit(NETFS_SREQ_FAILED, &subreq->flags))
+				break;
+			if (__test_and_clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
+				__clear_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+				subreq->retry_count++;
+				netfs_reset_iter(subreq);
+				netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
+				netfs_reissue_read(rreq, subreq);
+			}
+		}
+		return;
+	}
+
+	/* Okay, we need to renegotiate all the download requests and flip any
+	 * failed cache reads over to being download requests and negotiate
+	 * those also.  All fully successful subreqs have been removed from the
+	 * list and any spare data from those has been donated.
+	 *
+	 * What we do is decant the list and rebuild it one subreq at a time so
+	 * that we don't end up with donations jumping over a gap we're busy
+	 * populating with smaller subrequests.  In the event that the subreq
+	 * we just launched finishes before we insert the next subreq, it'll
+	 * fill in rreq->prev_donated instead.
+	 *
+	 * Note: Alternatively, we could split the tail subrequest right before
+	 * we reissue it and fix up the donations under lock.
+	 */
+	next = stream->subrequests.next;
+
+	do {
+		struct netfs_io_subrequest *from, *to, *tmp;
+		struct iov_iter source;
+		unsigned long long start, len;
+		size_t part;
+		bool boundary = false, subreq_superfluous = false;
+
+		/* Go through the subreqs and find the next span of contiguous
+		 * buffer that we then rejig (cifs, for example, needs the
+		 * rsize renegotiating) and reissue.
+		 */
+		from = list_entry(next, struct netfs_io_subrequest, rreq_link);
+		to = from;
+		start = from->start + from->transferred;
+		len   = from->len   - from->transferred;
+
+		_debug("from R=%08x[%x] s=%llx ctl=%zx/%zx",
+		       rreq->debug_id, from->debug_index,
+		       from->start, from->transferred, from->len);
+
+		if (test_bit(NETFS_SREQ_FAILED, &from->flags) ||
+		    !test_bit(NETFS_SREQ_NEED_RETRY, &from->flags)) {
+			subreq = from;
+			goto abandon;
+		}
+
+		for (;;) {
+			/* Read pointer to subreq before reading subreq state. */
+			next = smp_load_acquire(&next->next);
+			if (next == &stream->subrequests)
+				break;
+
+			subreq = list_entry(next, struct netfs_io_subrequest, rreq_link);
+			if (subreq->start + subreq->transferred != start + len ||
+			    test_bit(NETFS_SREQ_BOUNDARY, &subreq->flags) ||
+			    !test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags))
+				break;
+			to = subreq;
+			len += to->len;
+		}
+
+		_debug(" - range: %llx-%llx %llx", start, start + len - 1, len);
+
+		/* Determine the set of buffers we're going to use.  Each
+		 * subreq gets a subset of a single overall contiguous buffer.
+		 */
+		netfs_reset_iter(from);
+		source = from->io_iter;
+		source.count = len;
+
+		/* Work through the sublist. */
+		subreq = from;
+		list_for_each_entry_from(subreq, &stream->subrequests, rreq_link) {
+			if (!len) {
+				subreq_superfluous = true;
+				break;
+			}
+			subreq->source	= NETFS_DOWNLOAD_FROM_SERVER;
+			subreq->start	= start - subreq->transferred;
+			subreq->len	= len   + subreq->transferred;
+			__clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+			__clear_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+			subreq->retry_count++;
+
+			trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
+
+			/* Renegotiate max_len (rsize) */
+			stream->sreq_max_len = subreq->len;
+			if (rreq->netfs_ops->prepare_read &&
+			    rreq->netfs_ops->prepare_read(subreq) < 0) {
+				trace_netfs_sreq(subreq, netfs_sreq_trace_reprep_failed);
+				__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+				goto abandon;
+			}
+
+			part = umin(len, stream->sreq_max_len);
+			if (unlikely(stream->sreq_max_segs))
+				part = netfs_limit_iter(&source, 0, part, stream->sreq_max_segs);
+			subreq->len = subreq->transferred + part;
+			subreq->io_iter = source;
+			iov_iter_truncate(&subreq->io_iter, part);
+			iov_iter_advance(&source, part);
+			len -= part;
+			start += part;
+			if (!len) {
+				if (boundary)
+					__set_bit(NETFS_SREQ_BOUNDARY, &subreq->flags);
+			} else {
+				__clear_bit(NETFS_SREQ_BOUNDARY, &subreq->flags);
+			}
+
+			netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
+			netfs_reissue_read(rreq, subreq);
+			if (subreq == to) {
+				subreq_superfluous = false;
+				break;
+			}
+		}
+
+		/* If we managed to use fewer subreqs, we can discard the
+		 * excess; if we used the same number, then we're done.
+		 */
+		if (!len) {
+			if (!subreq_superfluous)
+				continue;
+			list_for_each_entry_safe_from(subreq, tmp,
+						      &stream->subrequests, rreq_link) {
+				trace_netfs_sreq(subreq, netfs_sreq_trace_superfluous);
+				spin_lock(&rreq->lock);
+				list_del(&subreq->rreq_link);
+				spin_unlock(&rreq->lock);
+				netfs_put_subrequest(subreq, netfs_sreq_trace_put_done);
+				if (subreq == to)
+					break;
+			}
+			subreq = NULL;
+			continue;
+		}
+
+		/* We ran out of subrequests, so we need to allocate some more
+		 * and insert them after.
+		 */
+		do {
+			subreq = netfs_alloc_subrequest(rreq);
+			if (!subreq) {
+				subreq = to;
+				goto abandon_after;
+			}
+			subreq->source		= NETFS_DOWNLOAD_FROM_SERVER;
+			subreq->start		= start;
+			subreq->len		= len;
+			subreq->stream_nr	= stream->stream_nr;
+			subreq->retry_count	= 1;
+
+			trace_netfs_sreq_ref(rreq->debug_id, subreq->debug_index,
+					     refcount_read(&subreq->ref),
+					     netfs_sreq_trace_new);
+
+			spin_lock(&rreq->lock);
+			list_add(&subreq->rreq_link, &to->rreq_link);
+			spin_unlock(&rreq->lock);
+			to = subreq;
+			trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
+
+			stream->sreq_max_len	= umin(len, rreq->rsize);
+			stream->sreq_max_segs	= 0;
+			if (unlikely(stream->sreq_max_segs))
+				part = netfs_limit_iter(&source, 0, part, stream->sreq_max_segs);
+
+			netfs_stat(&netfs_n_rh_download);
+			if (rreq->netfs_ops->prepare_read(subreq) < 0) {
+				trace_netfs_sreq(subreq, netfs_sreq_trace_reprep_failed);
+				__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+				goto abandon;
+			}
+
+			part = umin(len, stream->sreq_max_len);
+			subreq->len = subreq->transferred + part;
+			subreq->io_iter = source;
+			iov_iter_truncate(&subreq->io_iter, part);
+			iov_iter_advance(&source, part);
+
+			len -= part;
+			start += part;
+			if (!len && boundary) {
+				__set_bit(NETFS_SREQ_BOUNDARY, &to->flags);
+				boundary = false;
+			}
+
+			netfs_reissue_read(rreq, subreq);
+		} while (len);
+
+	} while (!list_is_head(next, &stream->subrequests));
+
+	return;
+
+	/* If we hit an error, fail all remaining incomplete subrequests */
+abandon_after:
+	if (list_is_last(&subreq->rreq_link, &stream->subrequests))
+		return;
+	subreq = list_next_entry(subreq, rreq_link);
+abandon:
+	list_for_each_entry_from(subreq, &stream->subrequests, rreq_link) {
+		if (!test_bit(NETFS_SREQ_FAILED, &subreq->flags) &&
+		    !test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags))
+			continue;
+		subreq->error = -ENOMEM;
+		__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+		__clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+	}
+}
+
+/*
+ * Retry reads.
+ */
+void netfs_retry_reads(struct netfs_io_request *rreq)
+{
+	struct netfs_io_stream *stream = &rreq->io_streams[0];
+
+	netfs_stat(&netfs_n_rh_retry_read_req);
+
+	/* Wait for all outstanding I/O to quiesce before performing retries as
+	 * we may need to renegotiate the I/O sizes.
+	 */
+	set_bit(NETFS_RREQ_RETRYING, &rreq->flags);
+	netfs_wait_for_in_progress_stream(rreq, stream);
+	clear_bit(NETFS_RREQ_RETRYING, &rreq->flags);
+
+	trace_netfs_rreq(rreq, netfs_rreq_trace_resubmit);
+	netfs_retry_read_subrequests(rreq);
+}
+
+/*
+ * Unlock any the pages that haven't been unlocked yet due to abandoned
+ * subrequests.
+ */
+void netfs_unlock_abandoned_read_pages(struct netfs_io_request *rreq)
+{
+	struct folio_queue *p;
+
+	for (p = rreq->buffer.tail; p; p = p->next) {
+		for (int slot = 0; slot < folioq_count(p); slot++) {
+			struct folio *folio = folioq_folio(p, slot);
+
+			if (folio && !folioq_is_marked2(p, slot)) {
+				if (folio == rreq->no_unlock_folio &&
+				    test_bit(NETFS_RREQ_NO_UNLOCK_FOLIO,
+					     &rreq->flags)) {
+					_debug("no unlock");
+				} else {
+					trace_netfs_folio(folio,
+						netfs_folio_trace_abandon);
+					folio_unlock(folio);
+				}
+			}
+		}
+	}
+}
diff --git a/fs/netfs/read_single.c b/fs/netfs/read_single.c
new file mode 100644
index 000000000..8833550d2
--- /dev/null
+++ b/fs/netfs/read_single.c
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Single, monolithic object support (e.g. AFS directory).
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include <linux/uio.h>
+#include <linux/sched/mm.h>
+#include <linux/task_io_accounting_ops.h>
+#include <linux/netfs.h>
+#include "internal.h"
+
+/**
+ * netfs_single_mark_inode_dirty - Mark a single, monolithic object inode dirty
+ * @inode: The inode to mark
+ *
+ * Mark an inode that contains a single, monolithic object as dirty so that its
+ * writepages op will get called.  If set, the SINGLE_NO_UPLOAD flag indicates
+ * that the object will only be written to the cache and not uploaded (e.g. AFS
+ * directory contents).
+ */
+void netfs_single_mark_inode_dirty(struct inode *inode)
+{
+	struct netfs_inode *ictx = netfs_inode(inode);
+	bool cache_only = test_bit(NETFS_ICTX_SINGLE_NO_UPLOAD, &ictx->flags);
+	bool caching = fscache_cookie_enabled(netfs_i_cookie(netfs_inode(inode)));
+
+	if (cache_only && !caching)
+		return;
+
+	mark_inode_dirty(inode);
+
+	if (caching && !(inode_state_read_once(inode) & I_PINNING_NETFS_WB)) {
+		bool need_use = false;
+
+		spin_lock(&inode->i_lock);
+		if (!(inode_state_read(inode) & I_PINNING_NETFS_WB)) {
+			inode_state_set(inode, I_PINNING_NETFS_WB);
+			need_use = true;
+		}
+		spin_unlock(&inode->i_lock);
+
+		if (need_use)
+			fscache_use_cookie(netfs_i_cookie(ictx), true);
+	}
+
+}
+EXPORT_SYMBOL(netfs_single_mark_inode_dirty);
+
+static int netfs_single_begin_cache_read(struct netfs_io_request *rreq, struct netfs_inode *ctx)
+{
+	return fscache_begin_read_operation(&rreq->cache_resources, netfs_i_cookie(ctx));
+}
+
+static void netfs_single_cache_prepare_read(struct netfs_io_request *rreq,
+					    struct netfs_io_subrequest *subreq)
+{
+	struct netfs_cache_resources *cres = &rreq->cache_resources;
+
+	if (!cres->ops) {
+		subreq->source = NETFS_DOWNLOAD_FROM_SERVER;
+		return;
+	}
+	subreq->source = cres->ops->prepare_read(subreq, rreq->i_size);
+	trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
+
+}
+
+static void netfs_single_read_cache(struct netfs_io_request *rreq,
+				    struct netfs_io_subrequest *subreq)
+{
+	struct netfs_cache_resources *cres = &rreq->cache_resources;
+
+	_enter("R=%08x[%x]", rreq->debug_id, subreq->debug_index);
+	netfs_stat(&netfs_n_rh_read);
+	cres->ops->read(cres, subreq->start, &subreq->io_iter, NETFS_READ_HOLE_FAIL,
+			netfs_cache_read_terminated, subreq);
+}
+
+/*
+ * Perform a read to a buffer from the cache or the server.  Only a single
+ * subreq is permitted as the object must be fetched in a single transaction.
+ */
+static int netfs_single_dispatch_read(struct netfs_io_request *rreq)
+{
+	struct netfs_io_subrequest *subreq;
+	int ret = 0;
+
+	subreq = netfs_alloc_subrequest(rreq);
+	if (!subreq)
+		return -ENOMEM;
+
+	subreq->source	= NETFS_SOURCE_UNKNOWN;
+	subreq->start	= 0;
+	subreq->len	= rreq->len;
+	subreq->io_iter	= rreq->buffer.iter;
+
+	netfs_queue_read(rreq, subreq);
+
+	netfs_single_cache_prepare_read(rreq, subreq);
+	switch (subreq->source) {
+	case NETFS_DOWNLOAD_FROM_SERVER:
+		netfs_stat(&netfs_n_rh_download);
+		if (rreq->netfs_ops->prepare_read) {
+			ret = rreq->netfs_ops->prepare_read(subreq);
+			if (ret < 0)
+				goto cancel;
+		}
+
+		smp_wmb(); /* Write lists before ALL_QUEUED. */
+		set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+		rreq->netfs_ops->issue_read(subreq);
+		rreq->submitted += subreq->len;
+		break;
+	case NETFS_READ_FROM_CACHE:
+		smp_wmb(); /* Write lists before ALL_QUEUED. */
+		set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+		trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
+		netfs_single_read_cache(rreq, subreq);
+		rreq->submitted += subreq->len;
+		ret = 0;
+		break;
+	default:
+		pr_warn("Unexpected single-read source %u\n", subreq->source);
+		WARN_ON_ONCE(true);
+		ret = -EIO;
+		goto cancel;
+	}
+
+	return ret;
+cancel:
+	netfs_cancel_read(subreq, ret);
+	smp_wmb(); /* Write lists before ALL_QUEUED. */
+	set_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags);
+	netfs_wake_collector(rreq);
+	return ret;
+}
+
+/**
+ * netfs_read_single - Synchronously read a single blob of pages.
+ * @inode: The inode to read from.
+ * @file: The file we're using to read or NULL.
+ * @iter: The buffer we're reading into.
+ *
+ * Fulfil a read request for a single monolithic object by drawing data from
+ * the cache if possible, or the netfs if not.  The buffer may be larger than
+ * the file content; unused beyond the EOF will be zero-filled.  The content
+ * will be read with a single I/O request (though this may be retried).
+ *
+ * The calling netfs must initialise a netfs context contiguous to the vfs
+ * inode before calling this.
+ *
+ * This is usable whether or not caching is enabled.  If caching is enabled,
+ * the data will be stored as a single object into the cache.
+ */
+ssize_t netfs_read_single(struct inode *inode, struct file *file, struct iov_iter *iter)
+{
+	struct netfs_io_request *rreq;
+	struct netfs_inode *ictx = netfs_inode(inode);
+	ssize_t ret;
+
+	rreq = netfs_alloc_request(inode->i_mapping, file, 0, iov_iter_count(iter),
+				   NETFS_READ_SINGLE);
+	if (IS_ERR(rreq))
+		return PTR_ERR(rreq);
+
+	ret = netfs_single_begin_cache_read(rreq, ictx);
+	if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
+		goto cleanup_free;
+
+	netfs_stat(&netfs_n_rh_read_single);
+	trace_netfs_read(rreq, 0, rreq->len, netfs_read_trace_read_single);
+
+	rreq->buffer.iter = *iter;
+	netfs_single_dispatch_read(rreq);
+
+	ret = netfs_wait_for_read(rreq);
+	netfs_put_request(rreq, netfs_rreq_trace_put_return);
+	return ret;
+
+cleanup_free:
+	netfs_put_failed_request(rreq);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_read_single);
diff --git a/fs/netfs/rolling_buffer.c b/fs/netfs/rolling_buffer.c
new file mode 100644
index 000000000..8c0026836
--- /dev/null
+++ b/fs/netfs/rolling_buffer.c
@@ -0,0 +1,226 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Rolling buffer helpers
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/bitops.h>
+#include <linux/mempool.h>
+#include <linux/pagemap.h>
+#include <linux/rolling_buffer.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+static atomic_t debug_ids;
+
+/**
+ * netfs_folioq_alloc - Allocate a folio_queue struct
+ * @rreq_id: Associated debugging ID for tracing purposes
+ * @gfp: Allocation constraints
+ * @trace: Trace tag to indicate the purpose of the allocation
+ *
+ * Allocate, initialise and account the folio_queue struct and log a trace line
+ * to mark the allocation.
+ */
+struct folio_queue *netfs_folioq_alloc(unsigned int rreq_id, gfp_t gfp,
+				       unsigned int /*enum netfs_folioq_trace*/ trace)
+{
+	struct folio_queue *fq;
+
+	if (gfp == GFP_KERNEL)
+		fq = netfs_folioq_pool.alloc(gfp, netfs_folioq_pool.pool_data);
+	else
+		fq = mempool_alloc(&netfs_folioq_pool, gfp);
+	if (fq) {
+		netfs_stat(&netfs_n_folioq);
+		folioq_init(fq, rreq_id);
+		fq->debug_id = atomic_inc_return(&debug_ids);
+		trace_netfs_folioq(fq, trace);
+	}
+	return fq;
+}
+EXPORT_SYMBOL(netfs_folioq_alloc);
+
+/**
+ * netfs_folioq_free - Free a folio_queue struct
+ * @folioq: The object to free
+ * @trace: Trace tag to indicate which free
+ *
+ * Free and unaccount the folio_queue struct.
+ */
+void netfs_folioq_free(struct folio_queue *folioq,
+		       unsigned int /*enum netfs_trace_folioq*/ trace)
+{
+	trace_netfs_folioq(folioq, trace);
+	netfs_stat_d(&netfs_n_folioq);
+	mempool_free(folioq, &netfs_folioq_pool);
+}
+EXPORT_SYMBOL(netfs_folioq_free);
+
+/*
+ * Initialise a rolling buffer.  We allocate an empty folio queue struct to so
+ * that the pointers can be independently driven by the producer and the
+ * consumer.
+ */
+int rolling_buffer_init(struct rolling_buffer *roll, unsigned int rreq_id,
+			unsigned int direction, gfp_t gfp)
+{
+	struct folio_queue *fq;
+
+	fq = netfs_folioq_alloc(rreq_id, gfp, netfs_trace_folioq_rollbuf_init);
+	if (!fq)
+		return -ENOMEM;
+
+	roll->head = fq;
+	roll->tail = fq;
+	iov_iter_folio_queue(&roll->iter, direction, fq, 0, 0, 0);
+	return 0;
+}
+
+/*
+ * Add another folio_queue to a rolling buffer if there's no space left.
+ */
+int rolling_buffer_make_space(struct rolling_buffer *roll, gfp_t gfp)
+{
+	struct folio_queue *fq, *head = roll->head;
+
+	if (!folioq_full(head))
+		return 0;
+
+	fq = netfs_folioq_alloc(head->rreq_id, gfp, netfs_trace_folioq_make_space);
+	if (!fq)
+		return -ENOMEM;
+	fq->prev = head;
+
+	roll->head = fq;
+	if (folioq_full(head)) {
+		/* Make sure we don't leave the master iterator pointing to a
+		 * block that might get immediately consumed.
+		 */
+		if (roll->iter.folioq == head &&
+		    roll->iter.folioq_slot == folioq_nr_slots(head)) {
+			roll->iter.folioq = fq;
+			roll->iter.folioq_slot = 0;
+		}
+	}
+
+	/* Make sure the initialisation is stored before the next pointer.
+	 *
+	 * [!] NOTE: After we set head->next, the consumer is at liberty to
+	 * immediately delete the old head.
+	 */
+	smp_store_release(&head->next, fq);
+	return 0;
+}
+
+/*
+ * Decant the list of folios to read into a rolling buffer.
+ */
+ssize_t rolling_buffer_load_from_ra(struct rolling_buffer *roll,
+				    struct readahead_control *ractl,
+				    struct folio_batch *put_batch)
+{
+	struct folio_queue *fq;
+	struct page **vec;
+	int nr, ix, to;
+	ssize_t size = 0;
+
+	if (rolling_buffer_make_space(roll, GFP_KERNEL) < 0)
+		return -ENOMEM;
+
+	fq = roll->head;
+	vec = (struct page **)fq->vec.folios;
+	nr = __readahead_batch(ractl, vec + folio_batch_count(&fq->vec),
+			       folio_batch_space(&fq->vec));
+	ix = fq->vec.nr;
+	to = ix + nr;
+	fq->vec.nr = to;
+	for (; ix < to; ix++) {
+		struct folio *folio = folioq_folio(fq, ix);
+		unsigned int order = folio_order(folio);
+
+		fq->orders[ix] = order;
+		size += PAGE_SIZE << order;
+		trace_netfs_folio(folio, netfs_folio_trace_read);
+		if (!folio_batch_add(put_batch, folio))
+			folio_batch_release(put_batch);
+	}
+	WRITE_ONCE(roll->iter.count, roll->iter.count + size);
+
+	/* Store the counter after setting the slot. */
+	smp_store_release(&roll->next_head_slot, to);
+	return size;
+}
+
+/*
+ * Append a folio to the rolling buffer.
+ */
+ssize_t rolling_buffer_append(struct rolling_buffer *roll, struct folio *folio,
+			      unsigned int flags, gfp_t gfp)
+{
+	ssize_t size = folio_size(folio);
+	int slot;
+
+	if (rolling_buffer_make_space(roll, gfp) < 0)
+		return -ENOMEM;
+
+	slot = folioq_append(roll->head, folio);
+	if (flags & ROLLBUF_MARK_1)
+		folioq_mark(roll->head, slot);
+	if (flags & ROLLBUF_MARK_2)
+		folioq_mark2(roll->head, slot);
+
+	WRITE_ONCE(roll->iter.count, roll->iter.count + size);
+
+	/* Store the counter after setting the slot. */
+	smp_store_release(&roll->next_head_slot, slot);
+	return size;
+}
+
+/*
+ * Delete a spent buffer from a rolling queue and return the next in line.  We
+ * don't return the last buffer to keep the pointers independent, but return
+ * NULL instead.
+ */
+struct folio_queue *rolling_buffer_delete_spent(struct rolling_buffer *roll)
+{
+	struct folio_queue *spent = roll->tail, *next = READ_ONCE(spent->next);
+
+	if (!next)
+		return NULL;
+	next->prev = NULL;
+	netfs_folioq_free(spent, netfs_trace_folioq_delete);
+	roll->tail = next;
+	return next;
+}
+
+/*
+ * Clear out a rolling queue.  Folios that have mark 1 set are put.
+ */
+void rolling_buffer_clear(struct rolling_buffer *roll)
+{
+	struct folio_batch fbatch;
+	struct folio_queue *p;
+
+	folio_batch_init(&fbatch);
+
+	while ((p = roll->tail)) {
+		roll->tail = p->next;
+		for (int slot = 0; slot < folioq_count(p); slot++) {
+			struct folio *folio = folioq_folio(p, slot);
+
+			if (!folio)
+				continue;
+			if (folioq_is_marked(p, slot)) {
+				trace_netfs_folio(folio, netfs_folio_trace_put);
+				if (!folio_batch_add(&fbatch, folio))
+					folio_batch_release(&fbatch);
+			}
+		}
+
+		netfs_folioq_free(p, netfs_trace_folioq_clear);
+	}
+
+	folio_batch_release(&fbatch);
+}
diff --git a/fs/netfs/stats.c b/fs/netfs/stats.c
new file mode 100644
index 000000000..ab6b916ad
--- /dev/null
+++ b/fs/netfs/stats.c
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/* Netfs support statistics
+ *
+ * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/seq_file.h>
+#include "internal.h"
+
+atomic_t netfs_n_rh_dio_read;
+atomic_t netfs_n_rh_readahead;
+atomic_t netfs_n_rh_read_folio;
+atomic_t netfs_n_rh_read_single;
+atomic_t netfs_n_rh_rreq;
+atomic_t netfs_n_rh_sreq;
+atomic_t netfs_n_rh_download;
+atomic_t netfs_n_rh_download_done;
+atomic_t netfs_n_rh_download_failed;
+atomic_t netfs_n_rh_download_instead;
+atomic_t netfs_n_rh_read;
+atomic_t netfs_n_rh_read_done;
+atomic_t netfs_n_rh_read_failed;
+atomic_t netfs_n_rh_zero;
+atomic_t netfs_n_rh_short_read;
+atomic_t netfs_n_rh_write;
+atomic_t netfs_n_rh_write_begin;
+atomic_t netfs_n_rh_write_done;
+atomic_t netfs_n_rh_write_failed;
+atomic_t netfs_n_rh_write_zskip;
+atomic_t netfs_n_rh_retry_read_req;
+atomic_t netfs_n_rh_retry_read_subreq;
+atomic_t netfs_n_wh_buffered_write;
+atomic_t netfs_n_wh_writethrough;
+atomic_t netfs_n_wh_dio_write;
+atomic_t netfs_n_wh_writepages;
+atomic_t netfs_n_wh_copy_to_cache;
+atomic_t netfs_n_wh_wstream_conflict;
+atomic_t netfs_n_wh_upload;
+atomic_t netfs_n_wh_upload_done;
+atomic_t netfs_n_wh_upload_failed;
+atomic_t netfs_n_wh_write;
+atomic_t netfs_n_wh_write_done;
+atomic_t netfs_n_wh_write_failed;
+atomic_t netfs_n_wh_retry_write_req;
+atomic_t netfs_n_wh_retry_write_subreq;
+atomic_t netfs_n_wb_lock_skip;
+atomic_t netfs_n_wb_lock_wait;
+atomic_t netfs_n_folioq;
+
+int netfs_stats_show(struct seq_file *m, void *v)
+{
+	seq_printf(m, "Reads  : DR=%u RA=%u RF=%u RS=%u WB=%u WBZ=%u\n",
+		   atomic_read(&netfs_n_rh_dio_read),
+		   atomic_read(&netfs_n_rh_readahead),
+		   atomic_read(&netfs_n_rh_read_folio),
+		   atomic_read(&netfs_n_rh_read_single),
+		   atomic_read(&netfs_n_rh_write_begin),
+		   atomic_read(&netfs_n_rh_write_zskip));
+	seq_printf(m, "Writes : BW=%u WT=%u DW=%u WP=%u 2C=%u\n",
+		   atomic_read(&netfs_n_wh_buffered_write),
+		   atomic_read(&netfs_n_wh_writethrough),
+		   atomic_read(&netfs_n_wh_dio_write),
+		   atomic_read(&netfs_n_wh_writepages),
+		   atomic_read(&netfs_n_wh_copy_to_cache));
+	seq_printf(m, "ZeroOps: ZR=%u sh=%u sk=%u\n",
+		   atomic_read(&netfs_n_rh_zero),
+		   atomic_read(&netfs_n_rh_short_read),
+		   atomic_read(&netfs_n_rh_write_zskip));
+	seq_printf(m, "DownOps: DL=%u ds=%u df=%u di=%u\n",
+		   atomic_read(&netfs_n_rh_download),
+		   atomic_read(&netfs_n_rh_download_done),
+		   atomic_read(&netfs_n_rh_download_failed),
+		   atomic_read(&netfs_n_rh_download_instead));
+	seq_printf(m, "CaRdOps: RD=%u rs=%u rf=%u\n",
+		   atomic_read(&netfs_n_rh_read),
+		   atomic_read(&netfs_n_rh_read_done),
+		   atomic_read(&netfs_n_rh_read_failed));
+	seq_printf(m, "UpldOps: UL=%u us=%u uf=%u\n",
+		   atomic_read(&netfs_n_wh_upload),
+		   atomic_read(&netfs_n_wh_upload_done),
+		   atomic_read(&netfs_n_wh_upload_failed));
+	seq_printf(m, "CaWrOps: WR=%u ws=%u wf=%u\n",
+		   atomic_read(&netfs_n_wh_write),
+		   atomic_read(&netfs_n_wh_write_done),
+		   atomic_read(&netfs_n_wh_write_failed));
+	seq_printf(m, "Retries: rq=%u rs=%u wq=%u ws=%u\n",
+		   atomic_read(&netfs_n_rh_retry_read_req),
+		   atomic_read(&netfs_n_rh_retry_read_subreq),
+		   atomic_read(&netfs_n_wh_retry_write_req),
+		   atomic_read(&netfs_n_wh_retry_write_subreq));
+	seq_printf(m, "Objs   : rr=%u sr=%u foq=%u wsc=%u\n",
+		   atomic_read(&netfs_n_rh_rreq),
+		   atomic_read(&netfs_n_rh_sreq),
+		   atomic_read(&netfs_n_folioq),
+		   atomic_read(&netfs_n_wh_wstream_conflict));
+	seq_printf(m, "WbLock : skip=%u wait=%u\n",
+		   atomic_read(&netfs_n_wb_lock_skip),
+		   atomic_read(&netfs_n_wb_lock_wait));
+	return fscache_stats_show(m);
+}
+EXPORT_SYMBOL(netfs_stats_show);
diff --git a/fs/netfs/write_collect.c b/fs/netfs/write_collect.c
new file mode 100644
index 000000000..210eb8f39
--- /dev/null
+++ b/fs/netfs/write_collect.c
@@ -0,0 +1,525 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Network filesystem write subrequest result collection, assessment
+ * and retrying.
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+/* Notes made in the collector */
+#define HIT_PENDING		0x01	/* A front op was still pending */
+#define NEED_REASSESS		0x02	/* Need to loop round and reassess */
+#define MADE_PROGRESS		0x04	/* Made progress cleaning up a stream or the folio set */
+#define NEED_UNLOCK		0x08	/* The pagecache needs unlocking */
+#define NEED_RETRY		0x10	/* A front op requests retrying */
+#define SAW_FAILURE		0x20	/* One stream or hit a permanent failure */
+
+static void netfs_dump_request(const struct netfs_io_request *rreq)
+{
+	pr_err("Request R=%08x r=%d fl=%lx or=%x e=%ld\n",
+	       rreq->debug_id, refcount_read(&rreq->ref), rreq->flags,
+	       rreq->origin, rreq->error);
+	pr_err("  st=%llx tsl=%zx/%llx/%llx\n",
+	       rreq->start, rreq->transferred, rreq->submitted, rreq->len);
+	pr_err("  cci=%llx/%llx/%llx\n",
+	       rreq->cleaned_to, rreq->collected_to, atomic64_read(&rreq->issued_to));
+	pr_err("  iw=%pSR\n", rreq->netfs_ops->issue_write);
+	for (int i = 0; i < NR_IO_STREAMS; i++) {
+		const struct netfs_io_subrequest *sreq;
+		const struct netfs_io_stream *s = &rreq->io_streams[i];
+
+		pr_err("  str[%x] s=%x e=%d acnf=%u,%u,%u,%u\n",
+		       s->stream_nr, s->source, s->error,
+		       s->avail, s->active, s->need_retry, s->failed);
+		pr_err("  str[%x] ct=%llx t=%zx\n",
+		       s->stream_nr, s->collected_to, s->transferred);
+		list_for_each_entry(sreq, &s->subrequests, rreq_link) {
+			pr_err("  sreq[%x:%x] sc=%u s=%llx t=%zx/%zx r=%d f=%lx\n",
+			       sreq->stream_nr, sreq->debug_index, sreq->source,
+			       sreq->start, sreq->transferred, sreq->len,
+			       refcount_read(&sreq->ref), sreq->flags);
+		}
+	}
+}
+
+/*
+ * Successful completion of write of a folio to the server and/or cache.  Note
+ * that we are not allowed to lock the folio here on pain of deadlocking with
+ * truncate.
+ */
+int netfs_folio_written_back(struct folio *folio)
+{
+	enum netfs_folio_trace why = netfs_folio_trace_clear;
+	struct inode *inode = folio_inode(folio);
+	struct netfs_inode *ictx = netfs_inode(inode);
+	struct netfs_folio *finfo;
+	struct netfs_group *group = NULL;
+	int gcount = 0;
+
+	if ((finfo = netfs_folio_info(folio))) {
+		/* Streaming writes cannot be redirtied whilst under writeback,
+		 * so discard the streaming record.
+		 */
+		unsigned long long fend;
+
+		fend = folio_pos(folio) + finfo->dirty_offset + finfo->dirty_len;
+		spin_lock(&ictx->inode.i_lock);
+		if (fend > ictx->_zero_point)
+			netfs_write_zero_point(inode, fend);
+		spin_unlock(&ictx->inode.i_lock);
+
+		folio_detach_private(folio);
+		group = finfo->netfs_group;
+		gcount++;
+		kfree(finfo);
+		why = netfs_folio_trace_clear_s;
+		goto end_wb;
+	}
+
+	if ((group = netfs_folio_group(folio))) {
+		if (group == NETFS_FOLIO_COPY_TO_CACHE) {
+			why = netfs_folio_trace_clear_cc;
+			folio_detach_private(folio);
+			goto end_wb;
+		}
+
+		/* Need to detach the group pointer if the page didn't get
+		 * redirtied.  If it has been redirtied, then it must be within
+		 * the same group.
+		 */
+		why = netfs_folio_trace_redirtied;
+		if (!folio_test_dirty(folio)) {
+			folio_detach_private(folio);
+			gcount++;
+			why = netfs_folio_trace_clear_g;
+		}
+	}
+
+end_wb:
+	trace_netfs_folio(folio, why);
+	folio_end_writeback(folio);
+	return gcount;
+}
+
+/*
+ * Unlock any folios we've finished with.
+ */
+static void netfs_writeback_unlock_folios(struct netfs_io_request *wreq,
+					  unsigned int *notes)
+{
+	struct folio_queue *folioq = wreq->buffer.tail;
+	unsigned long long collected_to = wreq->collected_to;
+	unsigned int slot = wreq->buffer.first_tail_slot;
+
+	if (WARN_ON_ONCE(!folioq)) {
+		pr_err("[!] Writeback unlock found empty rolling buffer!\n");
+		netfs_dump_request(wreq);
+		return;
+	}
+
+	if (wreq->origin == NETFS_PGPRIV2_COPY_TO_CACHE) {
+		if (netfs_pgpriv2_unlock_copied_folios(wreq))
+			*notes |= MADE_PROGRESS;
+		return;
+	}
+
+	if (slot >= folioq_nr_slots(folioq)) {
+		folioq = rolling_buffer_delete_spent(&wreq->buffer);
+		if (!folioq)
+			return;
+		slot = 0;
+	}
+
+	for (;;) {
+		struct folio *folio;
+		struct netfs_folio *finfo;
+		unsigned long long fpos, fend;
+		size_t fsize, flen;
+
+		folio = folioq_folio(folioq, slot);
+		if (WARN_ONCE(!folio_test_writeback(folio),
+			      "R=%08x: folio %lx is not under writeback\n",
+			      wreq->debug_id, folio->index))
+			trace_netfs_folio(folio, netfs_folio_trace_not_under_wback);
+
+		fpos = folio_pos(folio);
+		fsize = folio_size(folio);
+		finfo = netfs_folio_info(folio);
+		flen = finfo ? finfo->dirty_offset + finfo->dirty_len : fsize;
+
+		fend = min_t(unsigned long long, fpos + flen, wreq->i_size);
+
+		trace_netfs_collect_folio(wreq, folio, fend, collected_to);
+
+		/* Unlock any folio we've transferred all of. */
+		if (collected_to < fend)
+			break;
+
+		wreq->nr_group_rel += netfs_folio_written_back(folio);
+		wreq->cleaned_to = fpos + fsize;
+		*notes |= MADE_PROGRESS;
+
+		/* Clean up the head folioq.  If we clear an entire folioq, then
+		 * we can get rid of it provided it's not also the tail folioq
+		 * being filled by the issuer.
+		 */
+		folioq_clear(folioq, slot);
+		slot++;
+		if (slot >= folioq_nr_slots(folioq)) {
+			folioq = rolling_buffer_delete_spent(&wreq->buffer);
+			if (!folioq)
+				goto done;
+			slot = 0;
+		}
+
+		if (fpos + fsize >= collected_to)
+			break;
+	}
+
+	wreq->buffer.tail = folioq;
+done:
+	wreq->buffer.first_tail_slot = slot;
+}
+
+/*
+ * Collect and assess the results of various write subrequests.  We may need to
+ * retry some of the results - or even do an RMW cycle for content crypto.
+ *
+ * Note that we have a number of parallel, overlapping lists of subrequests,
+ * one to the server and one to the local cache for example, which may not be
+ * the same size or starting position and may not even correspond in boundary
+ * alignment.
+ */
+static void netfs_collect_write_results(struct netfs_io_request *wreq)
+{
+	struct netfs_io_subrequest *front, *remove;
+	struct netfs_io_stream *stream;
+	unsigned long long collected_to, issued_to;
+	unsigned int notes;
+	int s;
+
+	_enter("%llx-%llx", wreq->start, wreq->start + wreq->len);
+	trace_netfs_collect(wreq);
+	trace_netfs_rreq(wreq, netfs_rreq_trace_collect);
+
+reassess_streams:
+	issued_to = atomic64_read(&wreq->issued_to);
+	smp_rmb();
+	collected_to = ULLONG_MAX;
+	if (wreq->origin == NETFS_WRITEBACK ||
+	    wreq->origin == NETFS_WRITETHROUGH ||
+	    wreq->origin == NETFS_PGPRIV2_COPY_TO_CACHE)
+		notes = NEED_UNLOCK;
+	else
+		notes = 0;
+
+	/* Remove completed subrequests from the front of the streams and
+	 * advance the completion point on each stream.  We stop when we hit
+	 * something that's in progress.  The issuer thread may be adding stuff
+	 * to the tail whilst we're doing this.
+	 */
+	for (s = 0; s < NR_IO_STREAMS; s++) {
+		stream = &wreq->io_streams[s];
+		/* Read active flag before list pointers */
+		if (!smp_load_acquire(&stream->active))
+			continue;
+
+		front = list_first_entry_or_null_acquire(&stream->subrequests,
+							 struct netfs_io_subrequest, rreq_link);
+		/* Read first subreq pointer before IN_PROGRESS flag. */
+
+		while (front) {
+			trace_netfs_collect_sreq(wreq, front);
+			//_debug("sreq [%x] %llx %zx/%zx",
+			//       front->debug_index, front->start, front->transferred, front->len);
+
+			if (stream->collected_to < front->start) {
+				trace_netfs_collect_gap(wreq, stream, issued_to, 'F');
+				stream->collected_to = front->start;
+			}
+
+			/* Stall if the front is still undergoing I/O. */
+			if (netfs_check_subreq_in_progress(front)) {
+				notes |= HIT_PENDING;
+				break;
+			}
+			smp_rmb(); /* Read counters after I-P flag. */
+
+			if (stream->failed) {
+				stream->collected_to = front->start + front->len;
+				notes |= MADE_PROGRESS | SAW_FAILURE;
+				goto cancel;
+			}
+			if (front->start + front->transferred > stream->collected_to) {
+				stream->collected_to = front->start + front->transferred;
+				stream->transferred = stream->collected_to - wreq->start;
+				stream->transferred_valid = true;
+				notes |= MADE_PROGRESS;
+			}
+			if (test_bit(NETFS_SREQ_FAILED, &front->flags)) {
+				stream->failed = true;
+				stream->error = front->error;
+				if (stream->source == NETFS_UPLOAD_TO_SERVER)
+					mapping_set_error(wreq->mapping, front->error);
+				notes |= NEED_REASSESS | SAW_FAILURE;
+				break;
+			}
+			if (front->transferred < front->len) {
+				stream->need_retry = true;
+				notes |= NEED_RETRY | MADE_PROGRESS;
+				break;
+			}
+
+		cancel:
+			/* Remove if completely consumed. */
+			spin_lock(&wreq->lock);
+
+			remove = front;
+			list_del_init(&front->rreq_link);
+			front = list_first_entry_or_null(&stream->subrequests,
+							 struct netfs_io_subrequest, rreq_link);
+			spin_unlock(&wreq->lock);
+			netfs_put_subrequest(remove,
+					     notes & SAW_FAILURE ?
+					     netfs_sreq_trace_put_cancel :
+					     netfs_sreq_trace_put_done);
+		}
+
+		/* If we have an empty stream, we need to jump it forward
+		 * otherwise the collection point will never advance.
+		 */
+		if (!front && issued_to > stream->collected_to) {
+			trace_netfs_collect_gap(wreq, stream, issued_to, 'E');
+			stream->collected_to = issued_to;
+		}
+
+		if (stream->collected_to < collected_to)
+			collected_to = stream->collected_to;
+	}
+
+	if (collected_to != ULLONG_MAX && collected_to > wreq->collected_to)
+		wreq->collected_to = collected_to;
+
+	for (s = 0; s < NR_IO_STREAMS; s++) {
+		stream = &wreq->io_streams[s];
+		if (stream->active)
+			trace_netfs_collect_stream(wreq, stream);
+	}
+
+	trace_netfs_collect_state(wreq, wreq->collected_to, notes);
+
+	/* Unlock any folios that we have now finished with. */
+	if (notes & NEED_UNLOCK) {
+		if (wreq->cleaned_to < wreq->collected_to)
+			netfs_writeback_unlock_folios(wreq, &notes);
+	} else {
+		wreq->cleaned_to = wreq->collected_to;
+	}
+
+	// TODO: Discard encryption buffers
+
+	if (notes & NEED_RETRY)
+		goto need_retry;
+
+	if (notes & MADE_PROGRESS) {
+		netfs_wake_rreq_flag(wreq, NETFS_RREQ_PAUSE, netfs_rreq_trace_unpause);
+		//cond_resched();
+		goto reassess_streams;
+	}
+
+	if (notes & NEED_REASSESS) {
+		//cond_resched();
+		goto reassess_streams;
+	}
+
+out:
+	netfs_put_group_many(wreq->group, wreq->nr_group_rel);
+	wreq->nr_group_rel = 0;
+	_leave(" = %x", notes);
+	return;
+
+need_retry:
+	/* Okay...  We're going to have to retry one or both streams.  Note
+	 * that any partially completed op will have had any wholly transferred
+	 * folios removed from it.
+	 */
+	_debug("retry");
+	netfs_retry_writes(wreq);
+	goto out;
+}
+
+/*
+ * Perform the collection of subrequests, folios and encryption buffers.
+ */
+bool netfs_write_collection(struct netfs_io_request *wreq)
+{
+	struct netfs_inode *ictx = netfs_inode(wreq->inode);
+	size_t transferred;
+	bool transferred_valid = false;
+	int s;
+
+	_enter("R=%x", wreq->debug_id);
+
+	netfs_collect_write_results(wreq);
+
+	/* We're done when the app thread has finished posting subreqs and all
+	 * the queues in all the streams are empty.
+	 */
+	if (!test_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags))
+		return false;
+	smp_rmb(); /* Read ALL_QUEUED before lists. */
+
+	transferred = LONG_MAX;
+	for (s = 0; s < NR_IO_STREAMS; s++) {
+		struct netfs_io_stream *stream = &wreq->io_streams[s];
+		if (!stream->active)
+			continue;
+		if (!list_empty(&stream->subrequests))
+			return false;
+		if (stream->transferred_valid &&
+		    stream->transferred < transferred) {
+			transferred = stream->transferred;
+			transferred_valid = true;
+		}
+	}
+
+	/* Okay, declare that all I/O is complete. */
+	if (transferred_valid)
+		wreq->transferred = transferred;
+	trace_netfs_rreq(wreq, netfs_rreq_trace_write_done);
+
+	if (wreq->io_streams[1].active &&
+	    wreq->io_streams[1].failed &&
+	    ictx->ops->invalidate_cache) {
+		/* Cache write failure doesn't prevent writeback completion
+		 * unless we're in disconnected mode.
+		 */
+		ictx->ops->invalidate_cache(wreq);
+	}
+
+	_debug("finished");
+	netfs_wake_rreq_flag(wreq, NETFS_RREQ_IN_PROGRESS, netfs_rreq_trace_wake_ip);
+	/* As we cleared NETFS_RREQ_IN_PROGRESS, we acquired its ref. */
+
+	switch (wreq->origin) {
+	case NETFS_WRITEBACK:
+	case NETFS_WRITEBACK_SINGLE:
+	case NETFS_WRITETHROUGH:
+		netfs_wb_end(ictx);
+		break;
+	default:
+		break;
+	}
+
+	if (wreq->iocb) {
+		size_t written = min(wreq->transferred, wreq->len);
+		wreq->iocb->ki_pos += written;
+		if (wreq->iocb->ki_complete) {
+			trace_netfs_rreq(wreq, netfs_rreq_trace_ki_complete);
+			wreq->iocb->ki_complete(
+				wreq->iocb, wreq->error ? wreq->error : written);
+		}
+		wreq->iocb = VFS_PTR_POISON;
+	}
+
+	netfs_clear_subrequests(wreq);
+	return true;
+}
+
+void netfs_write_collection_worker(struct work_struct *work)
+{
+	struct netfs_io_request *rreq = container_of(work, struct netfs_io_request, work);
+
+	netfs_see_request(rreq, netfs_rreq_trace_see_work);
+	if (netfs_check_rreq_in_progress(rreq)) {
+		if (netfs_write_collection(rreq))
+			/* Drop the ref from the IN_PROGRESS flag. */
+			netfs_put_request(rreq, netfs_rreq_trace_put_work_ip);
+		else
+			netfs_see_request(rreq, netfs_rreq_trace_see_work_complete);
+	}
+}
+
+/**
+ * netfs_write_subrequest_terminated - Note the termination of a write operation.
+ * @_op: The I/O request that has terminated.
+ * @transferred_or_error: The amount of data transferred or an error code.
+ *
+ * This tells the library that a contributory write I/O operation has
+ * terminated, one way or another, and that it should collect the results.
+ *
+ * The caller indicates in @transferred_or_error the outcome of the operation,
+ * supplying a positive value to indicate the number of bytes transferred or a
+ * negative error code.  The library will look after reissuing I/O operations
+ * as appropriate and writing downloaded data to the cache.
+ *
+ * When this is called, ownership of the subrequest is transferred back to the
+ * library, along with a ref.
+ *
+ * Note that %_op is a void* so that the function can be passed to
+ * kiocb::term_func without the need for a casting wrapper.
+ */
+void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error)
+{
+	struct netfs_io_subrequest *subreq = _op;
+	struct netfs_io_request *wreq = subreq->rreq;
+
+	_enter("%x[%x] %zd", wreq->debug_id, subreq->debug_index, transferred_or_error);
+
+	switch (subreq->source) {
+	case NETFS_UPLOAD_TO_SERVER:
+		netfs_stat(&netfs_n_wh_upload_done);
+		break;
+	case NETFS_WRITE_TO_CACHE:
+		netfs_stat(&netfs_n_wh_write_done);
+		break;
+	default:
+		BUG();
+	}
+
+	if (IS_ERR_VALUE(transferred_or_error)) {
+		subreq->error = transferred_or_error;
+		/* if need retry is set, error should not matter */
+		if (!test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
+			set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+			trace_netfs_failure(wreq, subreq, transferred_or_error, netfs_fail_write);
+		}
+
+		switch (subreq->source) {
+		case NETFS_WRITE_TO_CACHE:
+			netfs_stat(&netfs_n_wh_write_failed);
+			break;
+		case NETFS_UPLOAD_TO_SERVER:
+			netfs_stat(&netfs_n_wh_upload_failed);
+			break;
+		default:
+			break;
+		}
+		trace_netfs_rreq(wreq, netfs_rreq_trace_set_pause);
+		set_bit(NETFS_RREQ_PAUSE, &wreq->flags);
+	} else {
+		if (WARN(transferred_or_error > subreq->len - subreq->transferred,
+			 "Subreq excess write: R=%x[%x] %zd > %zu - %zu",
+			 wreq->debug_id, subreq->debug_index,
+			 transferred_or_error, subreq->len, subreq->transferred))
+			transferred_or_error = subreq->len - subreq->transferred;
+
+		subreq->error = 0;
+		subreq->transferred += transferred_or_error;
+
+		if (subreq->transferred < subreq->len)
+			set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+	}
+
+	trace_netfs_sreq(subreq, netfs_sreq_trace_terminated);
+	netfs_subreq_clear_in_progress(subreq);
+	netfs_put_subrequest(subreq, netfs_sreq_trace_put_terminated);
+}
+EXPORT_SYMBOL(netfs_write_subrequest_terminated);
diff --git a/fs/netfs/write_issue.c b/fs/netfs/write_issue.c
new file mode 100644
index 000000000..2d9cfcd43
--- /dev/null
+++ b/fs/netfs/write_issue.c
@@ -0,0 +1,906 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Network filesystem high-level (buffered) writeback.
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ *
+ *
+ * To support network filesystems with local caching, we manage a situation
+ * that can be envisioned like the following:
+ *
+ *               +---+---+-----+-----+---+----------+
+ *    Folios:    |   |   |     |     |   |          |
+ *               +---+---+-----+-----+---+----------+
+ *
+ *                 +------+------+     +----+----+
+ *    Upload:      |      |      |.....|    |    |
+ *  (Stream 0)     +------+------+     +----+----+
+ *
+ *               +------+------+------+------+------+
+ *    Cache:     |      |      |      |      |      |
+ *  (Stream 1)   +------+------+------+------+------+
+ *
+ * Where we have a sequence of folios of varying sizes that we need to overlay
+ * with multiple parallel streams of I/O requests, where the I/O requests in a
+ * stream may also be of various sizes (in cifs, for example, the sizes are
+ * negotiated with the server; in something like ceph, they may represent the
+ * sizes of storage objects).
+ *
+ * The sequence in each stream may contain gaps and noncontiguous subrequests
+ * may be glued together into single vectored write RPCs.
+ */
+
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include "internal.h"
+
+/*
+ * Kill all dirty folios in the event of an unrecoverable error, starting with
+ * a locked folio we've already obtained from writeback_iter().
+ */
+static void netfs_kill_dirty_pages(struct address_space *mapping,
+				   struct writeback_control *wbc,
+				   struct folio *folio)
+{
+	int error = 0;
+
+	do {
+		enum netfs_folio_trace why = netfs_folio_trace_kill;
+		struct netfs_group *group = NULL;
+		struct netfs_folio *finfo = NULL;
+		void *priv;
+
+		priv = folio_detach_private(folio);
+		if (priv) {
+			finfo = __netfs_folio_info(priv);
+			if (finfo) {
+				/* Kill folio from streaming write. */
+				group = finfo->netfs_group;
+				why = netfs_folio_trace_kill_s;
+			} else {
+				group = priv;
+				if (group == NETFS_FOLIO_COPY_TO_CACHE) {
+					/* Kill copy-to-cache folio */
+					why = netfs_folio_trace_kill_cc;
+					group = NULL;
+				} else {
+					/* Kill folio with group */
+					why = netfs_folio_trace_kill_g;
+				}
+			}
+		}
+
+		trace_netfs_folio(folio, why);
+
+		folio_start_writeback(folio);
+		folio_unlock(folio);
+		folio_end_writeback(folio);
+
+		netfs_put_group(group);
+		kfree(finfo);
+
+	} while ((folio = writeback_iter(mapping, wbc, folio, &error)));
+}
+
+/*
+ * Create a write request and set it up appropriately for the origin type.
+ */
+struct netfs_io_request *netfs_create_write_req(struct address_space *mapping,
+						struct file *file,
+						loff_t start,
+						enum netfs_io_origin origin)
+{
+	struct netfs_io_request *wreq;
+	struct netfs_inode *ictx;
+	bool is_cacheable = (origin == NETFS_WRITEBACK ||
+			     origin == NETFS_WRITEBACK_SINGLE ||
+			     origin == NETFS_WRITETHROUGH ||
+			     origin == NETFS_PGPRIV2_COPY_TO_CACHE);
+
+	wreq = netfs_alloc_request(mapping, file, start, 0, origin);
+	if (IS_ERR(wreq))
+		return wreq;
+
+	_enter("R=%x", wreq->debug_id);
+
+	ictx = netfs_inode(wreq->inode);
+	if (is_cacheable)
+		fscache_begin_write_operation(&wreq->cache_resources, netfs_i_cookie(ictx));
+	if (rolling_buffer_init(&wreq->buffer, wreq->debug_id, ITER_SOURCE, wreq->gfp) < 0)
+		goto nomem;
+
+	wreq->cleaned_to = wreq->start;
+
+	wreq->io_streams[0].stream_nr		= 0;
+	wreq->io_streams[0].source		= NETFS_UPLOAD_TO_SERVER;
+	wreq->io_streams[0].prepare_write	= ictx->ops->prepare_write;
+	wreq->io_streams[0].issue_write		= ictx->ops->issue_write;
+	wreq->io_streams[0].collected_to	= start;
+	wreq->io_streams[0].transferred		= 0;
+
+	wreq->io_streams[1].stream_nr		= 1;
+	wreq->io_streams[1].source		= NETFS_WRITE_TO_CACHE;
+	wreq->io_streams[1].collected_to	= start;
+	wreq->io_streams[1].transferred		= 0;
+	if (fscache_resources_valid(&wreq->cache_resources)) {
+		wreq->io_streams[1].avail	= true;
+		wreq->io_streams[1].active	= true;
+		wreq->io_streams[1].prepare_write = wreq->cache_resources.ops->prepare_write_subreq;
+		wreq->io_streams[1].issue_write = wreq->cache_resources.ops->issue_write;
+	}
+
+	return wreq;
+nomem:
+	netfs_put_failed_request(wreq);
+	return ERR_PTR(-ENOMEM);
+}
+
+/**
+ * netfs_prepare_write_failed - Note write preparation failed
+ * @subreq: The subrequest to mark
+ *
+ * Mark a subrequest to note that preparation for write failed.
+ */
+void netfs_prepare_write_failed(struct netfs_io_subrequest *subreq)
+{
+	__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
+	trace_netfs_sreq(subreq, netfs_sreq_trace_prep_failed);
+}
+EXPORT_SYMBOL(netfs_prepare_write_failed);
+
+/*
+ * Prepare a write subrequest.  We need to allocate a new subrequest
+ * if we don't have one.
+ */
+void netfs_prepare_write(struct netfs_io_request *wreq,
+			 struct netfs_io_stream *stream,
+			 loff_t start)
+{
+	struct netfs_io_subrequest *subreq;
+	struct iov_iter *wreq_iter = &wreq->buffer.iter;
+
+	/* Make sure we don't point the iterator at a used-up folio_queue
+	 * struct being used as a placeholder to prevent the queue from
+	 * collapsing.  In such a case, extend the queue.
+	 */
+	if (iov_iter_is_folioq(wreq_iter) &&
+	    wreq_iter->folioq_slot >= folioq_nr_slots(wreq_iter->folioq))
+		rolling_buffer_make_space(&wreq->buffer, wreq->gfp);
+
+	subreq = netfs_alloc_subrequest(wreq);
+	subreq->source		= stream->source;
+	subreq->start		= start;
+	subreq->stream_nr	= stream->stream_nr;
+	subreq->io_iter		= *wreq_iter;
+
+	_enter("R=%x[%x]", wreq->debug_id, subreq->debug_index);
+
+	trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
+
+	stream->sreq_max_len	= UINT_MAX;
+	stream->sreq_max_segs	= INT_MAX;
+	switch (stream->source) {
+	case NETFS_UPLOAD_TO_SERVER:
+		netfs_stat(&netfs_n_wh_upload);
+		stream->sreq_max_len = wreq->wsize;
+		break;
+	case NETFS_WRITE_TO_CACHE:
+		netfs_stat(&netfs_n_wh_write);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		break;
+	}
+
+	if (stream->prepare_write)
+		stream->prepare_write(subreq);
+
+	__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+
+	/* We add to the end of the list whilst the collector may be walking
+	 * the list.  The collector only goes nextwards and uses the lock to
+	 * remove entries off of the front.
+	 */
+	spin_lock(&wreq->lock);
+	/* Write IN_PROGRESS before pointer to new subreq */
+	list_add_tail_release(&subreq->rreq_link, &stream->subrequests);
+	if (list_is_first(&subreq->rreq_link, &stream->subrequests)) {
+		if (!stream->active) {
+			stream->collected_to = subreq->start;
+			/* Write list pointers before active flag */
+			smp_store_release(&stream->active, true);
+		}
+	}
+
+	spin_unlock(&wreq->lock);
+
+	stream->construct = subreq;
+}
+
+/*
+ * Set the I/O iterator for the filesystem/cache to use and dispatch the I/O
+ * operation.  The operation may be asynchronous and should call
+ * netfs_write_subrequest_terminated() when complete.
+ */
+static void netfs_do_issue_write(struct netfs_io_stream *stream,
+				 struct netfs_io_subrequest *subreq)
+{
+	struct netfs_io_request *wreq = subreq->rreq;
+
+	_enter("R=%x[%x],%zx", wreq->debug_id, subreq->debug_index, subreq->len);
+
+	if (test_bit(NETFS_SREQ_FAILED, &subreq->flags))
+		return netfs_write_subrequest_terminated(subreq, subreq->error);
+
+	trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
+	stream->issue_write(subreq);
+}
+
+void netfs_reissue_write(struct netfs_io_stream *stream,
+			 struct netfs_io_subrequest *subreq,
+			 struct iov_iter *source)
+{
+	size_t size = subreq->len - subreq->transferred;
+
+	// TODO: Use encrypted buffer
+	subreq->io_iter = *source;
+	iov_iter_advance(source, size);
+	iov_iter_truncate(&subreq->io_iter, size);
+
+	subreq->retry_count++;
+	subreq->error = 0;
+	__clear_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
+	__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
+	netfs_stat(&netfs_n_wh_retry_write_subreq);
+	netfs_do_issue_write(stream, subreq);
+}
+
+void netfs_issue_write(struct netfs_io_request *wreq,
+		       struct netfs_io_stream *stream)
+{
+	struct netfs_io_subrequest *subreq = stream->construct;
+
+	if (!subreq)
+		return;
+	stream->construct = NULL;
+	subreq->io_iter.count = subreq->len;
+	netfs_do_issue_write(stream, subreq);
+}
+
+/*
+ * Add data to the write subrequest, dispatching each as we fill it up or if it
+ * is discontiguous with the previous.  We only fill one part at a time so that
+ * we can avoid overrunning the credits obtained (cifs) and try to parallelise
+ * content-crypto preparation with network writes.
+ */
+size_t netfs_advance_write(struct netfs_io_request *wreq,
+			   struct netfs_io_stream *stream,
+			   loff_t start, size_t len, bool to_eof)
+{
+	struct netfs_io_subrequest *subreq = stream->construct;
+	size_t part;
+
+	if (!stream->avail) {
+		_leave("no write");
+		return len;
+	}
+
+	_enter("R=%x[%x]", wreq->debug_id, subreq ? subreq->debug_index : 0);
+
+	if (subreq && start != subreq->start + subreq->len) {
+		netfs_issue_write(wreq, stream);
+		subreq = NULL;
+	}
+
+	if (!stream->construct)
+		netfs_prepare_write(wreq, stream, start);
+	subreq = stream->construct;
+
+	part = umin(stream->sreq_max_len - subreq->len, len);
+	_debug("part %zx/%zx %zx/%zx", subreq->len, stream->sreq_max_len, part, len);
+	subreq->len += part;
+	subreq->nr_segs++;
+	stream->submit_extendable_to -= part;
+
+	if (subreq->len >= stream->sreq_max_len ||
+	    subreq->nr_segs >= stream->sreq_max_segs ||
+	    to_eof) {
+		netfs_issue_write(wreq, stream);
+		subreq = NULL;
+	}
+
+	return part;
+}
+
+/*
+ * Write some of a pending folio data back to the server.
+ */
+static int netfs_write_folio(struct netfs_io_request *wreq,
+			     struct writeback_control *wbc,
+			     struct folio *folio)
+{
+	struct netfs_io_stream *upload = &wreq->io_streams[0];
+	struct netfs_io_stream *cache  = &wreq->io_streams[1];
+	struct netfs_io_stream *stream;
+	struct netfs_group *fgroup; /* TODO: Use this with ceph */
+	struct netfs_folio *finfo;
+	size_t iter_off = 0;
+	size_t fsize = folio_size(folio), flen = fsize, foff = 0;
+	loff_t fpos = folio_pos(folio), i_size;
+	bool to_eof = false, streamw = false;
+	bool debug = false;
+
+	_enter("");
+
+	if (rolling_buffer_make_space(&wreq->buffer, wreq->gfp) < 0)
+		return -ENOMEM;
+
+	/* netfs_perform_write() may shift i_size around the page or from out
+	 * of the page to beyond it, but cannot move i_size into or through the
+	 * page since we have it locked.
+	 */
+	i_size = i_size_read(wreq->inode);
+
+	if (fpos >= i_size) {
+		/* mmap beyond eof. */
+		_debug("beyond eof");
+		folio_start_writeback(folio);
+		folio_unlock(folio);
+		wreq->nr_group_rel += netfs_folio_written_back(folio);
+		netfs_put_group_many(wreq->group, wreq->nr_group_rel);
+		wreq->nr_group_rel = 0;
+		return 0;
+	}
+
+	if (fpos + fsize > wreq->i_size)
+		wreq->i_size = i_size;
+
+	fgroup = netfs_folio_group(folio);
+	finfo = netfs_folio_info(folio);
+	if (finfo) {
+		foff = finfo->dirty_offset;
+		flen = foff + finfo->dirty_len;
+		streamw = true;
+	}
+
+	if (wreq->origin == NETFS_WRITETHROUGH) {
+		to_eof = false;
+		if (flen > i_size - fpos)
+			flen = i_size - fpos;
+	} else if (flen > i_size - fpos) {
+		flen = i_size - fpos;
+		if (!streamw)
+			folio_zero_segment(folio, flen, fsize);
+		to_eof = true;
+	} else if (flen == i_size - fpos) {
+		to_eof = true;
+	}
+	flen -= foff;
+
+	_debug("folio %zx %zx %zx", foff, flen, fsize);
+
+	/* Deal with discontinuities in the stream of dirty pages.  These can
+	 * arise from a number of sources:
+	 *
+	 * (1) Intervening non-dirty pages from random-access writes, multiple
+	 *     flushers writing back different parts simultaneously and manual
+	 *     syncing.
+	 *
+	 * (2) Partially-written pages from write-streaming.
+	 *
+	 * (3) Pages that belong to a different write-back group (eg.  Ceph
+	 *     snapshots).
+	 *
+	 * (4) Actually-clean pages that were marked for write to the cache
+	 *     when they were read.  Note that these appear as a special
+	 *     write-back group.
+	 */
+	if (fgroup == NETFS_FOLIO_COPY_TO_CACHE) {
+		netfs_issue_write(wreq, upload);
+	} else if (fgroup != wreq->group) {
+		/* We can't write this page to the server yet. */
+		kdebug("wrong group");
+		folio_redirty_for_writepage(wbc, folio);
+		folio_unlock(folio);
+		netfs_issue_write(wreq, upload);
+		netfs_issue_write(wreq, cache);
+		return 0;
+	}
+
+	if (foff > 0)
+		netfs_issue_write(wreq, upload);
+	if (streamw)
+		netfs_issue_write(wreq, cache);
+
+	folio_start_writeback(folio);
+	folio_unlock(folio);
+
+	if (fgroup == NETFS_FOLIO_COPY_TO_CACHE) {
+		if (!cache->avail) {
+			trace_netfs_folio(folio, netfs_folio_trace_cancel_copy);
+			netfs_issue_write(wreq, upload);
+			netfs_folio_written_back(folio);
+			return 0;
+		}
+		trace_netfs_folio(folio, netfs_folio_trace_store_copy);
+	} else if (!upload->avail && !cache->avail) {
+		trace_netfs_folio(folio, netfs_folio_trace_cancel_store);
+		netfs_folio_written_back(folio);
+		return 0;
+	} else if (!upload->construct) {
+		trace_netfs_folio(folio, netfs_folio_trace_store);
+	} else {
+		trace_netfs_folio(folio, netfs_folio_trace_store_plus);
+	}
+
+	/* Attach the folio to the rolling buffer. */
+	rolling_buffer_append(&wreq->buffer, folio, 0, wreq->gfp);
+
+	/* Move the submission point forward to allow for write-streaming data
+	 * not starting at the front of the page.  We don't do write-streaming
+	 * with the cache as the cache requires DIO alignment.
+	 *
+	 * Also skip uploading for data that's been read and just needs copying
+	 * to the cache.
+	 */
+	for (int s = 0; s < NR_IO_STREAMS; s++) {
+		stream = &wreq->io_streams[s];
+		stream->submit_off = foff;
+		stream->submit_len = flen;
+		if (!stream->avail ||
+		    (stream->source == NETFS_WRITE_TO_CACHE && streamw) ||
+		    (stream->source == NETFS_UPLOAD_TO_SERVER &&
+		     fgroup == NETFS_FOLIO_COPY_TO_CACHE)) {
+			stream->submit_off = UINT_MAX;
+			stream->submit_len = 0;
+		}
+	}
+
+	/* Attach the folio to one or more subrequests.  For a big folio, we
+	 * could end up with thousands of subrequests if the wsize is small -
+	 * but we might need to wait during the creation of subrequests for
+	 * network resources (eg. SMB credits).
+	 */
+	for (;;) {
+		ssize_t part;
+		size_t lowest_off = ULONG_MAX;
+		int choose_s = -1;
+
+		/* Always add to the lowest-submitted stream first. */
+		for (int s = 0; s < NR_IO_STREAMS; s++) {
+			stream = &wreq->io_streams[s];
+			if (stream->submit_len > 0 &&
+			    stream->submit_off < lowest_off) {
+				lowest_off = stream->submit_off;
+				choose_s = s;
+			}
+		}
+
+		if (choose_s < 0)
+			break;
+		stream = &wreq->io_streams[choose_s];
+
+		/* Advance the iterator(s). */
+		if (stream->submit_off > iter_off) {
+			rolling_buffer_advance(&wreq->buffer, stream->submit_off - iter_off);
+			iter_off = stream->submit_off;
+		}
+
+		atomic64_set(&wreq->issued_to, fpos + stream->submit_off);
+		stream->submit_extendable_to = fsize - stream->submit_off;
+		part = netfs_advance_write(wreq, stream, fpos + stream->submit_off,
+					   stream->submit_len, to_eof);
+		stream->submit_off += part;
+		if (part > stream->submit_len)
+			stream->submit_len = 0;
+		else
+			stream->submit_len -= part;
+		if (part > 0)
+			debug = true;
+	}
+
+	if (fsize > iter_off)
+		rolling_buffer_advance(&wreq->buffer, fsize - iter_off);
+	atomic64_set(&wreq->issued_to, fpos + fsize);
+
+	if (!debug)
+		kdebug("R=%x: No submit", wreq->debug_id);
+
+	if (foff + flen < fsize)
+		for (int s = 0; s < NR_IO_STREAMS; s++)
+			netfs_issue_write(wreq, &wreq->io_streams[s]);
+
+	_leave(" = 0");
+	return 0;
+}
+
+/*
+ * End the issuing of writes, letting the collector know we're done.
+ */
+static void netfs_end_issue_write(struct netfs_io_request *wreq)
+{
+	bool needs_poke = true;
+
+	smp_wmb(); /* Write subreq lists before ALL_QUEUED. */
+	set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
+
+	for (int s = 0; s < NR_IO_STREAMS; s++) {
+		struct netfs_io_stream *stream = &wreq->io_streams[s];
+
+		if (!stream->active)
+			continue;
+		if (!list_empty(&stream->subrequests))
+			needs_poke = false;
+		netfs_issue_write(wreq, stream);
+	}
+
+	if (needs_poke)
+		netfs_wake_collector(wreq);
+}
+
+/*
+ * Write some of the pending data back to the server
+ */
+int netfs_writepages(struct address_space *mapping,
+		     struct writeback_control *wbc)
+{
+	struct netfs_inode *ictx = netfs_inode(mapping->host);
+	struct netfs_io_request *wreq = NULL;
+	struct folio *folio;
+	int error = 0;
+
+	if (!netfs_wb_begin(ictx, wbc->sync_mode == WB_SYNC_NONE))
+		return 0;
+
+	/* Need the first folio to be able to set up the op. */
+	folio = writeback_iter(mapping, wbc, NULL, &error);
+	if (!folio)
+		goto out;
+
+	wreq = netfs_create_write_req(mapping, NULL, folio_pos(folio), NETFS_WRITEBACK);
+	if (IS_ERR(wreq)) {
+		error = PTR_ERR(wreq);
+		goto couldnt_start;
+	}
+
+	__set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &wreq->flags);
+	trace_netfs_write(wreq, netfs_write_trace_writeback);
+	netfs_stat(&netfs_n_wh_writepages);
+
+	do {
+		_debug("wbiter %lx %llx", folio->index, atomic64_read(&wreq->issued_to));
+
+		/* It appears we don't have to handle cyclic writeback wrapping. */
+		WARN_ON_ONCE(wreq && folio_pos(folio) < atomic64_read(&wreq->issued_to));
+
+		if (netfs_folio_group(folio) != NETFS_FOLIO_COPY_TO_CACHE &&
+		    unlikely(!test_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags))) {
+			set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags);
+			wreq->netfs_ops->begin_writeback(wreq);
+		}
+
+		error = netfs_write_folio(wreq, wbc, folio);
+		if (error == -ENOMEM) {
+			folio_redirty_for_writepage(wbc, folio);
+			folio_unlock(folio);
+		}
+	} while ((folio = writeback_iter(mapping, wbc, folio, &error)));
+
+	netfs_end_issue_write(wreq);
+	netfs_wake_collector(wreq);
+
+	netfs_put_request(wreq, netfs_rreq_trace_put_return);
+	_leave(" = %d", error);
+	return error;
+
+couldnt_start:
+	if (error == -ENOMEM) {
+		folio_redirty_for_writepage(wbc, folio);
+		folio_unlock(folio);
+		folio = writeback_iter(mapping, wbc, folio, &error);
+		WARN_ON_ONCE(folio != NULL);
+	} else {
+		netfs_kill_dirty_pages(mapping, wbc, folio);
+	}
+out:
+	netfs_wb_end(ictx);
+	_leave(" = %d", error);
+	return error;
+}
+EXPORT_SYMBOL(netfs_writepages);
+
+/*
+ * Begin a write operation for writing through the pagecache.
+ */
+struct netfs_io_request *netfs_begin_writethrough(struct kiocb *iocb, size_t len)
+{
+	struct netfs_io_request *wreq = NULL;
+	struct netfs_inode *ictx = netfs_inode(file_inode(iocb->ki_filp));
+
+	netfs_wb_begin(ictx, false);
+
+	wreq = netfs_create_write_req(iocb->ki_filp->f_mapping, iocb->ki_filp,
+				      iocb->ki_pos, NETFS_WRITETHROUGH);
+	if (IS_ERR(wreq)) {
+		netfs_wb_end(ictx);
+		return wreq;
+	}
+
+	wreq->io_streams[0].avail = true;
+	__set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &wreq->flags);
+	trace_netfs_write(wreq, netfs_write_trace_writethrough);
+	return wreq;
+}
+
+/*
+ * Advance the state of the write operation used when writing through the
+ * pagecache.  Data has been copied into the pagecache that we need to append
+ * to the request.  If we've added more than wsize then we need to create a new
+ * subrequest.
+ */
+int netfs_advance_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
+			       struct folio *folio, size_t copied, bool to_page_end,
+			       struct folio **writethrough_cache)
+{
+	int ret;
+
+	_enter("R=%x ic=%zu ws=%u cp=%zu tp=%u",
+	       wreq->debug_id, wreq->buffer.iter.count, wreq->wsize, copied, to_page_end);
+
+	/* The folio is locked. */
+
+	if (*writethrough_cache != folio) {
+		if (*writethrough_cache) {
+			/* Did the folio get moved? */
+			folio_put(*writethrough_cache);
+			*writethrough_cache = NULL;
+		}
+		/* We can make multiple writes to the folio... */
+		if (wreq->len == 0)
+			trace_netfs_folio(folio, netfs_folio_trace_wthru);
+		else
+			trace_netfs_folio(folio, netfs_folio_trace_wthru_plus);
+		*writethrough_cache = folio;
+		folio_get(folio);
+	}
+
+	wreq->len += copied;
+
+	if (!to_page_end) {
+		folio_mark_dirty(folio);
+		folio_unlock(folio);
+		return 0;
+	}
+
+	ret = netfs_write_folio(wreq, wbc, folio);
+	folio_put(*writethrough_cache);
+	*writethrough_cache = NULL;
+	wreq->submitted = wreq->len;
+	return ret;
+}
+
+/*
+ * End a write operation used when writing through the pagecache.
+ */
+ssize_t netfs_end_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
+			       struct folio *writethrough_cache)
+{
+	ssize_t ret;
+
+	_enter("R=%x", wreq->debug_id);
+
+	if (writethrough_cache) {
+		folio_lock(writethrough_cache);
+		netfs_write_folio(wreq, wbc, writethrough_cache);
+		folio_put(writethrough_cache);
+		wreq->submitted = wreq->len;
+	}
+
+	netfs_end_issue_write(wreq);
+
+	if (wreq->iocb)
+		ret = -EIOCBQUEUED;
+	else
+		ret = netfs_wait_for_write(wreq);
+	netfs_put_request(wreq, netfs_rreq_trace_put_return);
+	return ret;
+}
+
+/*
+ * Write some of a pending folio data back to the server and/or the cache.
+ */
+static int netfs_write_folio_single(struct netfs_io_request *wreq,
+				    struct folio *folio)
+{
+	struct netfs_io_stream *upload = &wreq->io_streams[0];
+	struct netfs_io_stream *cache  = &wreq->io_streams[1];
+	struct netfs_io_stream *stream;
+	size_t iter_off = 0;
+	size_t fsize = folio_size(folio), flen;
+	loff_t fpos = folio_pos(folio);
+	ssize_t ret;
+	bool to_eof = false;
+	bool no_debug = false;
+
+	_enter("");
+
+	flen = folio_size(folio);
+	if (flen > wreq->i_size - fpos) {
+		flen = wreq->i_size - fpos;
+		folio_zero_segment(folio, flen, fsize);
+		to_eof = true;
+	} else if (flen == wreq->i_size - fpos) {
+		to_eof = true;
+	}
+
+	_debug("folio %zx/%zx", flen, fsize);
+
+	if (!upload->avail && !cache->avail) {
+		trace_netfs_folio(folio, netfs_folio_trace_cancel_store);
+		return 0;
+	}
+
+	if (!upload->construct)
+		trace_netfs_folio(folio, netfs_folio_trace_store);
+	else
+		trace_netfs_folio(folio, netfs_folio_trace_store_plus);
+
+	/* Attach the folio to the rolling buffer. */
+	folio_get(folio);
+	ret = rolling_buffer_append(&wreq->buffer, folio, NETFS_ROLLBUF_PUT_MARK, wreq->gfp);
+	if (ret < 0) {
+		folio_put(folio);
+		return ret;
+	}
+
+	/* Move the submission point forward to allow for write-streaming data
+	 * not starting at the front of the page.  We don't do write-streaming
+	 * with the cache as the cache requires DIO alignment.
+	 *
+	 * Also skip uploading for data that's been read and just needs copying
+	 * to the cache.
+	 */
+	for (int s = 0; s < NR_IO_STREAMS; s++) {
+		stream = &wreq->io_streams[s];
+		stream->submit_off = 0;
+		stream->submit_len = flen;
+		if (!stream->avail) {
+			stream->submit_off = UINT_MAX;
+			stream->submit_len = 0;
+		}
+	}
+
+	/* Attach the folio to one or more subrequests.  For a big folio, we
+	 * could end up with thousands of subrequests if the wsize is small -
+	 * but we might need to wait during the creation of subrequests for
+	 * network resources (eg. SMB credits).
+	 */
+	for (;;) {
+		ssize_t part;
+		size_t lowest_off = ULONG_MAX;
+		int choose_s = -1;
+
+		/* Always add to the lowest-submitted stream first. */
+		for (int s = 0; s < NR_IO_STREAMS; s++) {
+			stream = &wreq->io_streams[s];
+			if (stream->submit_len > 0 &&
+			    stream->submit_off < lowest_off) {
+				lowest_off = stream->submit_off;
+				choose_s = s;
+			}
+		}
+
+		if (choose_s < 0)
+			break;
+		stream = &wreq->io_streams[choose_s];
+
+		/* Advance the iterator(s). */
+		if (stream->submit_off > iter_off) {
+			rolling_buffer_advance(&wreq->buffer, stream->submit_off - iter_off);
+			iter_off = stream->submit_off;
+		}
+
+		atomic64_set(&wreq->issued_to, fpos + stream->submit_off);
+		stream->submit_extendable_to = fsize - stream->submit_off;
+		part = netfs_advance_write(wreq, stream, fpos + stream->submit_off,
+					   stream->submit_len, to_eof);
+		stream->submit_off += part;
+		if (part > stream->submit_len)
+			stream->submit_len = 0;
+		else
+			stream->submit_len -= part;
+		if (part > 0)
+			no_debug = true;
+	}
+
+	wreq->buffer.iter.iov_offset = 0;
+	if (fsize > iter_off)
+		rolling_buffer_advance(&wreq->buffer, fsize - iter_off);
+	atomic64_set(&wreq->issued_to, fpos + fsize);
+
+	if (!no_debug)
+		kdebug("R=%x: No submit", wreq->debug_id);
+	_leave(" = 0");
+	return 0;
+}
+
+/**
+ * netfs_writeback_single - Write back a monolithic payload
+ * @mapping: The mapping to write from
+ * @wbc: Hints from the VM
+ * @iter: Data to write, must be ITER_FOLIOQ.
+ *
+ * Write a monolithic, non-pagecache object back to the server and/or
+ * the cache.
+ *
+ * Return: 0 if successful; 1 if skipped due to lock conflict and WB_SYNC_NONE;
+ * or a negative error code.
+ */
+int netfs_writeback_single(struct address_space *mapping,
+			   struct writeback_control *wbc,
+			   struct iov_iter *iter)
+{
+	struct netfs_io_request *wreq;
+	struct netfs_inode *ictx = netfs_inode(mapping->host);
+	struct folio_queue *fq;
+	size_t size = iov_iter_count(iter);
+	int ret;
+
+	if (WARN_ON_ONCE(!iov_iter_is_folioq(iter)))
+		return -EIO;
+
+	if (!netfs_wb_begin(ictx, wbc->sync_mode == WB_SYNC_NONE)) {
+		/* The VFS will have undirtied the inode. */
+		netfs_single_mark_inode_dirty(&ictx->inode);
+		return 1;
+	}
+
+	wreq = netfs_create_write_req(mapping, NULL, 0, NETFS_WRITEBACK_SINGLE);
+	if (IS_ERR(wreq)) {
+		ret = PTR_ERR(wreq);
+		goto couldnt_start;
+	}
+
+	__set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &wreq->flags);
+	trace_netfs_write(wreq, netfs_write_trace_writeback_single);
+	netfs_stat(&netfs_n_wh_writepages);
+
+	if (__test_and_set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags))
+		wreq->netfs_ops->begin_writeback(wreq);
+
+	for (fq = (struct folio_queue *)iter->folioq; fq; fq = fq->next) {
+		for (int slot = 0; slot < folioq_count(fq); slot++) {
+			struct folio *folio = folioq_folio(fq, slot);
+			size_t part = umin(folioq_folio_size(fq, slot), size);
+
+			_debug("wbiter %lx %llx", folio->index, atomic64_read(&wreq->issued_to));
+
+			ret = netfs_write_folio_single(wreq, folio);
+			if (ret < 0)
+				goto stop;
+			size -= part;
+			if (size <= 0)
+				goto stop;
+		}
+	}
+
+stop:
+	for (int s = 0; s < NR_IO_STREAMS; s++)
+		netfs_issue_write(wreq, &wreq->io_streams[s]);
+	smp_wmb(); /* Write lists before ALL_QUEUED. */
+	set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
+
+	netfs_wake_collector(wreq);
+
+	netfs_put_request(wreq, netfs_rreq_trace_put_return);
+	_leave(" = %d", ret);
+	return ret;
+
+couldnt_start:
+	netfs_wb_end(ictx);
+	_leave(" = %d", ret);
+	return ret;
+}
+EXPORT_SYMBOL(netfs_writeback_single);
diff --git a/fs/netfs/write_retry.c b/fs/netfs/write_retry.c
new file mode 100644
index 000000000..058bc7a16
--- /dev/null
+++ b/fs/netfs/write_retry.c
@@ -0,0 +1,238 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Network filesystem write retrying.
+ *
+ * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include "internal.h"
+
+/*
+ * Perform retries on the streams that need it.
+ */
+static void netfs_retry_write_stream(struct netfs_io_request *wreq,
+				     struct netfs_io_stream *stream)
+{
+	struct list_head *next;
+
+	_enter("R=%x[%x:]", wreq->debug_id, stream->stream_nr);
+
+	if (list_empty(&stream->subrequests))
+		return;
+
+	if (stream->source == NETFS_UPLOAD_TO_SERVER &&
+	    wreq->netfs_ops->retry_request)
+		wreq->netfs_ops->retry_request(wreq, stream);
+
+	if (unlikely(stream->failed))
+		return;
+
+	/* If there's no renegotiation to do, just resend each failed subreq. */
+	if (!stream->prepare_write) {
+		struct netfs_io_subrequest *subreq;
+
+		list_for_each_entry(subreq, &stream->subrequests, rreq_link) {
+			if (test_bit(NETFS_SREQ_FAILED, &subreq->flags))
+				break;
+			if (__test_and_clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
+				struct iov_iter source;
+
+				netfs_reset_iter(subreq);
+				source = subreq->io_iter;
+				netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
+				netfs_reissue_write(stream, subreq, &source);
+			}
+		}
+		return;
+	}
+
+	next = stream->subrequests.next;
+
+	do {
+		struct netfs_io_subrequest *subreq = NULL, *from, *to, *tmp;
+		struct iov_iter source;
+		unsigned long long start, len;
+		size_t part;
+		bool boundary = false;
+
+		/* Go through the stream and find the next span of contiguous
+		 * data that we then rejig (cifs, for example, needs the wsize
+		 * renegotiating) and reissue.
+		 */
+		from = list_entry(next, struct netfs_io_subrequest, rreq_link);
+		to = from;
+		start = from->start + from->transferred;
+		len   = from->len   - from->transferred;
+
+		if (test_bit(NETFS_SREQ_FAILED, &from->flags) ||
+		    !test_bit(NETFS_SREQ_NEED_RETRY, &from->flags))
+			return;
+
+		for (;;) {
+			/* Read pointer to subreq before reading subreq state. */
+			next = smp_load_acquire(&next->next);
+			if (next == &stream->subrequests)
+				break;
+
+			subreq = list_entry(next, struct netfs_io_subrequest, rreq_link);
+			if (subreq->start + subreq->transferred != start + len ||
+			    test_bit(NETFS_SREQ_BOUNDARY, &subreq->flags) ||
+			    !test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags))
+				break;
+			to = subreq;
+			len += to->len;
+		}
+
+		/* Determine the set of buffers we're going to use.  Each
+		 * subreq gets a subset of a single overall contiguous buffer.
+		 */
+		netfs_reset_iter(from);
+		source = from->io_iter;
+		source.count = len;
+
+		/* Work through the sublist. */
+		subreq = from;
+		list_for_each_entry_from(subreq, &stream->subrequests, rreq_link) {
+			if (!len)
+				break;
+
+			subreq->start	= start;
+			subreq->len	= len;
+			__clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
+			trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
+
+			/* Renegotiate max_len (wsize) */
+			stream->sreq_max_len = len;
+			stream->prepare_write(subreq);
+
+			part = umin(len, stream->sreq_max_len);
+			if (unlikely(stream->sreq_max_segs))
+				part = netfs_limit_iter(&source, 0, part, stream->sreq_max_segs);
+			subreq->len = part;
+			subreq->transferred = 0;
+			len -= part;
+			start += part;
+			if (len && subreq == to &&
+			    __test_and_clear_bit(NETFS_SREQ_BOUNDARY, &to->flags))
+				boundary = true;
+
+			netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
+			netfs_reissue_write(stream, subreq, &source);
+			if (subreq == to)
+				break;
+		}
+
+		/* If we managed to use fewer subreqs, we can discard the
+		 * excess; if we used the same number, then we're done.
+		 */
+		if (!len) {
+			if (subreq == to)
+				continue;
+			list_for_each_entry_safe_from(subreq, tmp,
+						      &stream->subrequests, rreq_link) {
+				trace_netfs_sreq(subreq, netfs_sreq_trace_discard);
+				spin_lock(&wreq->lock);
+				list_del(&subreq->rreq_link);
+				spin_unlock(&wreq->lock);
+				netfs_put_subrequest(subreq, netfs_sreq_trace_put_done);
+				if (subreq == to)
+					break;
+			}
+			continue;
+		}
+
+		/* We ran out of subrequests, so we need to allocate some more
+		 * and insert them after.
+		 */
+		do {
+			subreq = netfs_alloc_subrequest(wreq);
+			subreq->source		= to->source;
+			subreq->start		= start;
+			subreq->stream_nr	= to->stream_nr;
+			subreq->retry_count	= 1;
+
+			trace_netfs_sreq_ref(wreq->debug_id, subreq->debug_index,
+					     refcount_read(&subreq->ref),
+					     netfs_sreq_trace_new);
+			trace_netfs_sreq(subreq, netfs_sreq_trace_split);
+
+			spin_lock(&wreq->lock);
+			list_add(&subreq->rreq_link, &to->rreq_link);
+			spin_unlock(&wreq->lock);
+			to = subreq;
+			trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
+
+			stream->sreq_max_len	= len;
+			stream->sreq_max_segs	= INT_MAX;
+			switch (stream->source) {
+			case NETFS_UPLOAD_TO_SERVER:
+				netfs_stat(&netfs_n_wh_upload);
+				stream->sreq_max_len = umin(len, wreq->wsize);
+				break;
+			case NETFS_WRITE_TO_CACHE:
+				netfs_stat(&netfs_n_wh_write);
+				break;
+			default:
+				WARN_ON_ONCE(1);
+			}
+
+			stream->prepare_write(subreq);
+
+			part = umin(len, stream->sreq_max_len);
+			subreq->len = subreq->transferred + part;
+			len -= part;
+			start += part;
+			if (!len && boundary) {
+				__set_bit(NETFS_SREQ_BOUNDARY, &to->flags);
+				boundary = false;
+			}
+
+			netfs_reissue_write(stream, subreq, &source);
+			if (!len)
+				break;
+
+		} while (len);
+
+	} while (!list_is_head(next, &stream->subrequests));
+}
+
+/*
+ * Perform retries on the streams that need it.  If we're doing content
+ * encryption and the server copy changed due to a third-party write, we may
+ * need to do an RMW cycle and also rewrite the data to the cache.
+ */
+void netfs_retry_writes(struct netfs_io_request *wreq)
+{
+	struct netfs_io_stream *stream;
+	int s;
+
+	netfs_stat(&netfs_n_wh_retry_write_req);
+
+	/* Wait for all outstanding I/O to quiesce before performing retries as
+	 * we may need to renegotiate the I/O sizes.
+	 */
+	set_bit(NETFS_RREQ_RETRYING, &wreq->flags);
+	for (s = 0; s < NR_IO_STREAMS; s++) {
+		stream = &wreq->io_streams[s];
+		if (stream->active)
+			netfs_wait_for_in_progress_stream(wreq, stream);
+	}
+	clear_bit(NETFS_RREQ_RETRYING, &wreq->flags);
+
+	// TODO: Enc: Fetch changed partial pages
+	// TODO: Enc: Reencrypt content if needed.
+	// TODO: Enc: Wind back transferred point.
+	// TODO: Enc: Mark cache pages for retry.
+
+	for (s = 0; s < NR_IO_STREAMS; s++) {
+		stream = &wreq->io_streams[s];
+		if (stream->need_retry) {
+			stream->need_retry = false;
+			netfs_retry_write_stream(wreq, stream);
+		}
+	}
+}