xfs: implement buffered writes to zoned RT devices

Implement buffered writes including page faults and block zeroing for
zoned RT devices.  Buffered writes to zoned RT devices are split into
three phases:

 1) a reservation for the worst case data block usage is taken before
    acquiring the iolock.  When there are enough free blocks but not
    enough available one, garbage collection is kicked off to free the
    space before continuing with the write.  If there isn't enough
    freeable space, the block reservation is reduced and a short write
    will happen as expected by normal Linux write semantics.
 2) with the iolock held, the generic iomap buffered write code is
    called, which through the iomap_begin operation usually just inserts
    delalloc extents for the range in a single iteration.  Only for
    overwrites of existing data that are not block aligned, or zeroing
    operations the existing extent mapping is read to fill out the srcmap
    and to figure out if zeroing is required.
 3) the ->map_blocks callback to the generic iomap writeback code
    calls into the zoned space allocator to actually allocate on-disk
    space for the range before kicking of the writeback.

Note that because all writes are out of place, truncate or hole punches
that are not aligned to block size boundaries need to allocate space.
For block zeroing from truncate, ->setattr is called with the iolock
(aka i_rwsem) already held, so a hacky deviation from the above
scheme is needed.  In this case the space reservations is called with
the iolock held, but is required not to block and can dip into the
reserved block pool.  This can lead to -ENOSPC when truncating a
file, which is unfortunate.  But fixing the calling conventions in
the VFS is probably much easier with code requiring it already in
mainline.

Similarly because all writes are out place, the zoned allocator can't
support unwritten extents and thus the FALLOC_FL_ALLOCATE_RANGE range
mode of fallocate.  Other fallocate modes that would reserved space
but don't need to to provide proper semantics do work but do not
reserve space.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

1 files changed, 216 insertions, 25 deletions

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index 07ee980cd8d3..512d2f56b206 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -25,6 +25,7 @@
 #include "xfs_iomap.h"
 #include "xfs_reflink.h"
 #include "xfs_file.h"
+#include "xfs_zone_alloc.h"
 
 #include <linux/dax.h>
 #include <linux/falloc.h>
@@ -360,7 +361,8 @@ xfs_file_write_zero_eof(
 	struct iov_iter		*from,
 	unsigned int		*iolock,
 	size_t			count,
-	bool			*drained_dio)
+	bool			*drained_dio,
+	struct xfs_zone_alloc_ctx *ac)
 {
 	struct xfs_inode	*ip = XFS_I(iocb->ki_filp->f_mapping->host);
 	loff_t			isize;
@@ -414,7 +416,7 @@ xfs_file_write_zero_eof(
 	trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
 
 	xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
-	error = xfs_zero_range(ip, isize, iocb->ki_pos - isize, NULL);
+	error = xfs_zero_range(ip, isize, iocb->ki_pos - isize, ac, NULL);
 	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
 
 	return error;
@@ -431,7 +433,8 @@ STATIC ssize_t
 xfs_file_write_checks(
 	struct kiocb		*iocb,
 	struct iov_iter		*from,
-	unsigned int		*iolock)
+	unsigned int		*iolock,
+	struct xfs_zone_alloc_ctx *ac)
 {
 	struct inode		*inode = iocb->ki_filp->f_mapping->host;
 	size_t			count = iov_iter_count(from);
@@ -481,7 +484,7 @@ restart:
 	 */
 	if (iocb->ki_pos > i_size_read(inode)) {
 		error = xfs_file_write_zero_eof(iocb, from, iolock, count,
-				&drained_dio);
+				&drained_dio, ac);
 		if (error == 1)
 			goto restart;
 		if (error)
@@ -491,6 +494,48 @@ restart:
 	return kiocb_modified(iocb);
 }
 
+static ssize_t
+xfs_zoned_write_space_reserve(
+	struct xfs_inode		*ip,
+	struct kiocb			*iocb,
+	struct iov_iter			*from,
+	unsigned int			flags,
+	struct xfs_zone_alloc_ctx	*ac)
+{
+	loff_t				count = iov_iter_count(from);
+	int				error;
+
+	if (iocb->ki_flags & IOCB_NOWAIT)
+		flags |= XFS_ZR_NOWAIT;
+
+	/*
+	 * Check the rlimit and LFS boundary first so that we don't over-reserve
+	 * by possibly a lot.
+	 *
+	 * The generic write path will redo this check later, and it might have
+	 * changed by then.  If it got expanded we'll stick to our earlier
+	 * smaller limit, and if it is decreased the new smaller limit will be
+	 * used and our extra space reservation will be returned after finishing
+	 * the write.
+	 */
+	error = generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, &count);
+	if (error)
+		return error;
+
+	/*
+	 * Sloppily round up count to file system blocks.
+	 *
+	 * This will often reserve an extra block, but that avoids having to look
+	 * at the start offset, which isn't stable for O_APPEND until taking the
+	 * iolock.  Also we need to reserve a block each for zeroing the old
+	 * EOF block and the new start block if they are unaligned.
+	 *
+	 * Any remaining block will be returned after the write.
+	 */
+	return xfs_zoned_space_reserve(ip,
+			XFS_B_TO_FSB(ip->i_mount, count) + 1 + 2, flags, ac);
+}
+
 static int
 xfs_dio_write_end_io(
 	struct kiocb		*iocb,
@@ -597,7 +642,7 @@ xfs_file_dio_write_aligned(
 	ret = xfs_ilock_iocb_for_write(iocb, &iolock);
 	if (ret)
 		return ret;
-	ret = xfs_file_write_checks(iocb, from, &iolock);
+	ret = xfs_file_write_checks(iocb, from, &iolock, NULL);
 	if (ret)
 		goto out_unlock;
 
@@ -675,7 +720,7 @@ retry_exclusive:
 		goto out_unlock;
 	}
 
-	ret = xfs_file_write_checks(iocb, from, &iolock);
+	ret = xfs_file_write_checks(iocb, from, &iolock, NULL);
 	if (ret)
 		goto out_unlock;
 
@@ -749,7 +794,7 @@ xfs_file_dax_write(
 	ret = xfs_ilock_iocb(iocb, iolock);
 	if (ret)
 		return ret;
-	ret = xfs_file_write_checks(iocb, from, &iolock);
+	ret = xfs_file_write_checks(iocb, from, &iolock, NULL);
 	if (ret)
 		goto out;
 
@@ -793,7 +838,7 @@ write_retry:
 	if (ret)
 		return ret;
 
-	ret = xfs_file_write_checks(iocb, from, &iolock);
+	ret = xfs_file_write_checks(iocb, from, &iolock, NULL);
 	if (ret)
 		goto out;
 
@@ -841,6 +886,67 @@ out:
 }
 
 STATIC ssize_t
+xfs_file_buffered_write_zoned(
+	struct kiocb		*iocb,
+	struct iov_iter		*from)
+{
+	struct xfs_inode	*ip = XFS_I(iocb->ki_filp->f_mapping->host);
+	struct xfs_mount	*mp = ip->i_mount;
+	unsigned int		iolock = XFS_IOLOCK_EXCL;
+	bool			cleared_space = false;
+	struct xfs_zone_alloc_ctx ac = { };
+	ssize_t			ret;
+
+	ret = xfs_zoned_write_space_reserve(ip, iocb, from, XFS_ZR_GREEDY, &ac);
+	if (ret < 0)
+		return ret;
+
+	ret = xfs_ilock_iocb(iocb, iolock);
+	if (ret)
+		goto out_unreserve;
+
+	ret = xfs_file_write_checks(iocb, from, &iolock, &ac);
+	if (ret)
+		goto out_unlock;
+
+	/*
+	 * Truncate the iter to the length that we were actually able to
+	 * allocate blocks for.  This needs to happen after
+	 * xfs_file_write_checks, because that assigns ki_pos for O_APPEND
+	 * writes.
+	 */
+	iov_iter_truncate(from,
+			XFS_FSB_TO_B(mp, ac.reserved_blocks) -
+			(iocb->ki_pos & mp->m_blockmask));
+	if (!iov_iter_count(from))
+		goto out_unlock;
+
+retry:
+	trace_xfs_file_buffered_write(iocb, from);
+	ret = iomap_file_buffered_write(iocb, from,
+			&xfs_buffered_write_iomap_ops, &ac);
+	if (ret == -ENOSPC && !cleared_space) {
+		/*
+		 * Kick off writeback to convert delalloc space and release the
+		 * usually too pessimistic indirect block reservations.
+		 */
+		xfs_flush_inodes(mp);
+		cleared_space = true;
+		goto retry;
+	}
+
+out_unlock:
+	xfs_iunlock(ip, iolock);
+out_unreserve:
+	xfs_zoned_space_unreserve(ip, &ac);
+	if (ret > 0) {
+		XFS_STATS_ADD(mp, xs_write_bytes, ret);
+		ret = generic_write_sync(iocb, ret);
+	}
+	return ret;
+}
+
+STATIC ssize_t
 xfs_file_write_iter(
 	struct kiocb		*iocb,
 	struct iov_iter		*from)
@@ -887,6 +993,8 @@ xfs_file_write_iter(
 			return ret;
 	}
 
+	if (xfs_is_zoned_inode(ip))
+		return xfs_file_buffered_write_zoned(iocb, from);
 	return xfs_file_buffered_write(iocb, from);
 }
 
@@ -941,7 +1049,8 @@ static int
 xfs_falloc_collapse_range(
 	struct file		*file,
 	loff_t			offset,
-	loff_t			len)
+	loff_t			len,
+	struct xfs_zone_alloc_ctx *ac)
 {
 	struct inode		*inode = file_inode(file);
 	loff_t			new_size = i_size_read(inode) - len;
@@ -957,7 +1066,7 @@ xfs_falloc_collapse_range(
 	if (offset + len >= i_size_read(inode))
 		return -EINVAL;
 
-	error = xfs_collapse_file_space(XFS_I(inode), offset, len);
+	error = xfs_collapse_file_space(XFS_I(inode), offset, len, ac);
 	if (error)
 		return error;
 	return xfs_falloc_setsize(file, new_size);
@@ -1013,7 +1122,8 @@ xfs_falloc_zero_range(
 	struct file		*file,
 	int			mode,
 	loff_t			offset,
-	loff_t			len)
+	loff_t			len,
+	struct xfs_zone_alloc_ctx *ac)
 {
 	struct inode		*inode = file_inode(file);
 	unsigned int		blksize = i_blocksize(inode);
@@ -1026,7 +1136,7 @@ xfs_falloc_zero_range(
 	if (error)
 		return error;
 
-	error = xfs_free_file_space(XFS_I(inode), offset, len);
+	error = xfs_free_file_space(XFS_I(inode), offset, len, ac);
 	if (error)
 		return error;
 
@@ -1097,22 +1207,18 @@ xfs_falloc_allocate_range(
 		 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
 
 STATIC long
-xfs_file_fallocate(
+__xfs_file_fallocate(
 	struct file		*file,
 	int			mode,
 	loff_t			offset,
-	loff_t			len)
+	loff_t			len,
+	struct xfs_zone_alloc_ctx *ac)
 {
 	struct inode		*inode = file_inode(file);
 	struct xfs_inode	*ip = XFS_I(inode);
 	long			error;
 	uint			iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
 
-	if (!S_ISREG(inode->i_mode))
-		return -EINVAL;
-	if (mode & ~XFS_FALLOC_FL_SUPPORTED)
-		return -EOPNOTSUPP;
-
 	xfs_ilock(ip, iolock);
 	error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
 	if (error)
@@ -1133,16 +1239,16 @@ xfs_file_fallocate(
 
 	switch (mode & FALLOC_FL_MODE_MASK) {
 	case FALLOC_FL_PUNCH_HOLE:
-		error = xfs_free_file_space(ip, offset, len);
+		error = xfs_free_file_space(ip, offset, len, ac);
 		break;
 	case FALLOC_FL_COLLAPSE_RANGE:
-		error = xfs_falloc_collapse_range(file, offset, len);
+		error = xfs_falloc_collapse_range(file, offset, len, ac);
 		break;
 	case FALLOC_FL_INSERT_RANGE:
 		error = xfs_falloc_insert_range(file, offset, len);
 		break;
 	case FALLOC_FL_ZERO_RANGE:
-		error = xfs_falloc_zero_range(file, mode, offset, len);
+		error = xfs_falloc_zero_range(file, mode, offset, len, ac);
 		break;
 	case FALLOC_FL_UNSHARE_RANGE:
 		error = xfs_falloc_unshare_range(file, mode, offset, len);
@@ -1163,6 +1269,54 @@ out_unlock:
 	return error;
 }
 
+static long
+xfs_file_zoned_fallocate(
+	struct file		*file,
+	int			mode,
+	loff_t			offset,
+	loff_t			len)
+{
+	struct xfs_zone_alloc_ctx ac = { };
+	struct xfs_inode	*ip = XFS_I(file_inode(file));
+	int			error;
+
+	error = xfs_zoned_space_reserve(ip, 2, XFS_ZR_RESERVED, &ac);
+	if (error)
+		return error;
+	error = __xfs_file_fallocate(file, mode, offset, len, &ac);
+	xfs_zoned_space_unreserve(ip, &ac);
+	return error;
+}
+
+static long
+xfs_file_fallocate(
+	struct file		*file,
+	int			mode,
+	loff_t			offset,
+	loff_t			len)
+{
+	struct inode		*inode = file_inode(file);
+
+	if (!S_ISREG(inode->i_mode))
+		return -EINVAL;
+	if (mode & ~XFS_FALLOC_FL_SUPPORTED)
+		return -EOPNOTSUPP;
+
+	/*
+	 * For zoned file systems, zeroing the first and last block of a hole
+	 * punch requires allocating a new block to rewrite the remaining data
+	 * and new zeroes out of place.  Get a reservations for those before
+	 * taking the iolock.  Dip into the reserved pool because we are
+	 * expected to be able to punch a hole even on a completely full
+	 * file system.
+	 */
+	if (xfs_is_zoned_inode(XFS_I(inode)) &&
+	    (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
+		     FALLOC_FL_COLLAPSE_RANGE)))
+		return xfs_file_zoned_fallocate(file, mode, offset, len);
+	return __xfs_file_fallocate(file, mode, offset, len, NULL);
+}
+
 STATIC int
 xfs_file_fadvise(
 	struct file	*file,
@@ -1488,9 +1642,10 @@ xfs_dax_read_fault(
  *         i_lock (XFS - extent map serialisation)
  */
 static vm_fault_t
-xfs_write_fault(
+__xfs_write_fault(
 	struct vm_fault		*vmf,
-	unsigned int		order)
+	unsigned int		order,
+	struct xfs_zone_alloc_ctx *ac)
 {
 	struct inode		*inode = file_inode(vmf->vma->vm_file);
 	struct xfs_inode	*ip = XFS_I(inode);
@@ -1528,13 +1683,49 @@ xfs_write_fault(
 		ret = xfs_dax_fault_locked(vmf, order, true);
 	else
 		ret = iomap_page_mkwrite(vmf, &xfs_buffered_write_iomap_ops,
-				NULL);
+				ac);
 	xfs_iunlock(ip, lock_mode);
 
 	sb_end_pagefault(inode->i_sb);
 	return ret;
 }
 
+static vm_fault_t
+xfs_write_fault_zoned(
+	struct vm_fault		*vmf,
+	unsigned int		order)
+{
+	struct xfs_inode	*ip = XFS_I(file_inode(vmf->vma->vm_file));
+	unsigned int		len = folio_size(page_folio(vmf->page));
+	struct xfs_zone_alloc_ctx ac = { };
+	int			error;
+	vm_fault_t		ret;
+
+	/*
+	 * This could over-allocate as it doesn't check for truncation.
+	 *
+	 * But as the overallocation is limited to less than a folio and will be
+	 * release instantly that's just fine.
+	 */
+	error = xfs_zoned_space_reserve(ip, XFS_B_TO_FSB(ip->i_mount, len), 0,
+			&ac);
+	if (error < 0)
+		return vmf_fs_error(error);
+	ret = __xfs_write_fault(vmf, order, &ac);
+	xfs_zoned_space_unreserve(ip, &ac);
+	return ret;
+}
+
+static vm_fault_t
+xfs_write_fault(
+	struct vm_fault		*vmf,
+	unsigned int		order)
+{
+	if (xfs_is_zoned_inode(XFS_I(file_inode(vmf->vma->vm_file))))
+		return xfs_write_fault_zoned(vmf, order);
+	return __xfs_write_fault(vmf, order, NULL);
+}
+
 static inline bool
 xfs_is_write_fault(
 	struct vm_fault		*vmf)