fs/netfs/direct_read.c

// 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"

/*
 * If we did a direct read to a bounce buffer (say we needed to decrypt it),
 * copy the data obtained to the destination iterator.
 */
int netfs_dio_copy_bounce_to_dest(struct netfs_io_request *rreq, struct kiocb *iocb)
{
	struct iov_iter *dest_iter = &rreq->buffer.iter;
	unsigned long long start = rreq->start;

	_enter("%zx/%llx @%llx", rreq->transferred, rreq->len, start);

	if (!iocb)
		iocb = rreq->iocb;

	if (!test_bit(NETFS_RREQ_USE_BOUNCE_BUFFER, &rreq->flags))
		return 0;
	if (!test_bit(NETFS_RREQ_CRYPT_IN_PLACE, &rreq->flags)) {
		if (!rreq->transferred) {
			trace_netfs_failure(rreq, NULL, -EIO, netfs_fail_dio_read_zero);
			return -EIO;
		}
		return 0;
	}

	if (start < iocb->ki_pos) {
		if (rreq->transferred <= iocb->ki_pos - start) {
			trace_netfs_failure(rreq, NULL, -EIO, netfs_fail_dio_read_short);
			return -EIO;
		}
		rreq->len = rreq->transferred;
		rreq->transferred -= iocb->ki_pos - start;
	}

	if (rreq->transferred > iov_iter_count(dest_iter))
		rreq->transferred = iov_iter_count(dest_iter);

	_debug("xfer %zx/%llx @%llx", rreq->transferred, rreq->len, iocb->ki_pos);

	if (!rreq->transferred || !iov_iter_count(dest_iter)) {
		trace_netfs_failure(rreq, NULL, -EIO, netfs_fail_dio_read_zero);
		return -EIO;
	}

	return rolling_buffer_copy_to_iter(&rreq->bounce, dest_iter,
					   iocb->ki_pos, rreq->transferred);
}

/*
 * Prepare the buffer iterator for the network filesystem.  We may limit the
 * size of the request to the maximum read size, the maximum number of segments
 * or to trim it to a block size.
 */
void netfs_prepare_dio_read_iterator(struct netfs_io_subrequest *subreq, size_t bsize)
{
	struct netfs_io_request *rreq = subreq->rreq;
	struct iov_iter *iter = &rreq->buffer.iter;
	size_t rsize;

	if (test_bit(NETFS_RREQ_USE_BOUNCE_BUFFER, &rreq->flags))
		iter = &rreq->bounce.iter;

	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(iter, 0, rsize,
						rreq->io_streams[0].sreq_max_segs);

		if (limit < rsize) {
			subreq->len = limit;
			trace_netfs_sreq(subreq, netfs_sreq_trace_limited);
		}
	}

	/* Don't split a crypto block across multiple transfers unless the
	 * transport won't support a transfer that big.
	 */
	if (subreq->len > bsize)
		subreq->len = round_down(subreq->len, bsize);

	trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);

	subreq->io_iter	= *iter;
	iov_iter_truncate(&subreq->io_iter, subreq->len);
	iov_iter_advance(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 int netfs_dispatch_unbuffered_reads(struct netfs_io_request *rreq)
{
	struct netfs_io_stream *stream = &rreq->io_streams[0];
	unsigned long long start = rreq->start;
	ssize_t size = rreq->len;
	size_t bsize = rreq->crypto_bsize;
	int ret = 0;

	/* Pad out an encrypted transfer to the block size. */
	start = round_down(start, bsize);
	size += rreq->start - start;
	size = round_up(size, bsize);

	do {
		struct netfs_io_subrequest *subreq;
		ssize_t slice;

		subreq = netfs_alloc_subrequest(rreq, NETFS_DOWNLOAD_FROM_SERVER);
		if (!subreq) {
			ret = -ENOMEM;
			break;
		}

		subreq->start	= start;
		subreq->len	= size;

		__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);

		spin_lock(&rreq->lock);
		list_add_tail(&subreq->rreq_link, &stream->subrequests);
		if (list_is_first(&subreq->rreq_link, &stream->subrequests)) {
			stream->front = subreq;
			if (!stream->active) {
				stream->collected_to = stream->front->start;
				/* Store list pointers before active flag */
				smp_store_release(&stream->active, true);
			}
		}
		trace_netfs_sreq(subreq, netfs_sreq_trace_added);
		spin_unlock(&rreq->lock);

		netfs_stat(&netfs_n_rh_download);
		if (rreq->netfs_ops->prepare_read) {
			ret = rreq->netfs_ops->prepare_read(subreq);
			if (ret < 0) {
				netfs_put_subrequest(subreq, netfs_sreq_trace_put_cancel);
				break;
			}
		}

		/* Make sure we have sufficient bounce bufferage. */
		if (test_bit(NETFS_RREQ_USE_BOUNCE_BUFFER, &rreq->flags)) {
			ret = netfs_alloc_bounce(rreq, subreq->start + subreq->len, GFP_KERNEL);
			if (ret < 0) {
				netfs_put_subrequest(subreq, netfs_sreq_trace_put_cancel);
				break;
			}
		}

		netfs_prepare_dio_read_iterator(subreq, bsize);
		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;
		if (test_bit(NETFS_RREQ_BLOCKED, &rreq->flags) &&
		    test_bit(NETFS_RREQ_NONBLOCK, &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);
	}

	return ret;
}

/*
 * 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);
		return -EIO;
	}

	// TODO: Use bounce buffer if requested

	inode_dio_begin(rreq->inode);

	ret = netfs_dispatch_unbuffered_reads(rreq);

	if (!rreq->submitted) {
		netfs_put_request(rreq, netfs_rreq_trace_put_no_submit);
		inode_dio_end(rreq->inode);
		ret = 0;
		goto out;
	}

	if (sync)
		ret = netfs_wait_for_read(rreq);
	else
		ret = -EIOCBQUEUED;
out:
	_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, NULL,
				   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 out;
		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);
	}

	/* If we're going to do decryption or decompression, we're going to
	 * need a bounce buffer.  If the output buffer is correctly aligned and
	 * correctly sized for the crypto algorithm, we get a free copy between
	 * buffers from the crypto; if misaligned, we decrypt in place in the
	 * bounce buffer and then copy.
	 */
	if (test_bit(NETFS_RREQ_CONTENT_ENCRYPTION, &rreq->flags)) {
		if (!netfs_is_crypto_aligned(rreq, iter))
			__set_bit(NETFS_RREQ_CRYPT_IN_PLACE, &rreq->flags);
		__set_bit(NETFS_RREQ_USE_BOUNCE_BUFFER, &rreq->flags);
	}

	/* Set up the bounce buffer if we need it.  Allow for padding the
	 * request out to the crypo block size and allocate at least one folio
	 * into it.
	 */
	if (test_bit(NETFS_RREQ_USE_BOUNCE_BUFFER, &rreq->flags)) {
		size_t bsize = rreq->crypto_bsize;

		rreq->bounce_alloc_to = round_down(rreq->start, bsize);

		ret = rolling_buffer_init(&rreq->bounce, rreq->debug_id, ITER_DEST);
		if (ret < 0)
			goto out;

		ret = netfs_alloc_bounce(rreq, rreq->bounce_alloc_to + bsize, GFP_KERNEL);
		if (ret < 0)
			goto out;
	}

	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) {
		ret = netfs_dio_copy_bounce_to_dest(rreq, iocb);
		if (ret == 0) {
			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;
}
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);

/**
 * netfs_unbuffered_read_from_inode - Perform an unbuffered sync I/O read
 * @inode: The inode being accessed
 * @pos: The file position to read from
 * @iter: The output buffer (also specifies read length)
 * @nohole: True to return short/ENODATA if hole encountered
 *
 * Perform a synchronous unbuffered I/O from the inode to the output buffer.
 * No use is made of the pagecache.  The output buffer must be suitably aligned
 * if content encryption is to be used.  If @nohole is true then the read will
 * stop short if a hole is encountered and return -ENODATA if the read begins
 * with a hole.
 *
 * The caller must hold any appropriate locks.
 */
ssize_t netfs_unbuffered_read_from_inode(struct inode *inode, loff_t pos,
					 struct iov_iter *iter, bool nohole)
{
	struct netfs_io_request *rreq;
	ssize_t ret;
	size_t orig_count = iov_iter_count(iter);

	_enter("");

	if (WARN_ON(user_backed_iter(iter)))
		return -EIO;

	if (!orig_count)
		return 0; /* Don't update atime */

	ret = filemap_write_and_wait_range(inode->i_mapping, pos, orig_count);
	if (ret < 0)
		return ret;
	inode_update_time(inode, S_ATIME);

	rreq = netfs_alloc_request(inode->i_mapping, NULL, pos, orig_count,
				   NULL, NETFS_UNBUFFERED_READ);
	if (IS_ERR(rreq))
		return PTR_ERR(rreq);

	ret = -EIO;
	if (test_bit(NETFS_RREQ_CONTENT_ENCRYPTION, &rreq->flags) &&
	    WARN_ON(!netfs_is_crypto_aligned(rreq, iter)))
		goto out;

	netfs_stat(&netfs_n_rh_dio_read);
	trace_netfs_read(rreq, rreq->start, rreq->len,
			 netfs_read_trace_unbuffered_read_from_inode);

	rreq->buffer.iter	= *iter;
	rreq->len		= orig_count;
	rreq->direct_bv_unpin	= false;
	iov_iter_advance(iter, orig_count);

	if (nohole)
		__set_bit(NETFS_RREQ_NO_READ_HOLE, &rreq->flags);

	/* We're going to do the crypto in place in the destination buffer. */
	if (test_bit(NETFS_RREQ_CONTENT_ENCRYPTION, &rreq->flags))
		__set_bit(NETFS_RREQ_CRYPT_IN_PLACE, &rreq->flags);

	ret = netfs_dispatch_unbuffered_reads(rreq);

	if (!rreq->submitted) {
		netfs_put_request(rreq, netfs_rreq_trace_put_no_submit);
		goto out;
	}

	ret = netfs_wait_for_read(rreq);
out:
	netfs_put_request(rreq, netfs_rreq_trace_put_return);
	return ret;
}
EXPORT_SYMBOL(netfs_unbuffered_read_from_inode);