Skip to contentSkip to conversation

iomap: replace folio_batch allocation with stack allocation

PATCH Superseded Brian Foster 2 participants 4 messages Nov 11, 2025 – Nov 12, 2025

Subsystems (3)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iomap: replace folio_batch allocation with stack allocation

Zhang Yi points out that the dynamic folio_batch allocation in iomap_fill_dirty_folios() is problematic for the ext4 on iomap work that is under development because it doesn't sufficiently handle the allocation failure case (by allowing a retry, for example).

The dynamic allocation was initially added for simplicity and to help indicate whether the batch was used or not by the calling fs. To address this issue, put the batch on the stack of iomap_zero_range() and use a flag to control whether the batch should be used in the iomap folio lookup path. This keeps things simple and eliminates the concern for ext4 on iomap.

---

Hi all,

This is patch 1 from the previously posted RFC series here [1]. This was initially motivated by the ext4-on-iomap WIP, but the folio batch allocation has also since been called out separately here [2]. The broader series at [1] still needs work, but since this is a straight cleanup I'm posting it as such for standalone consideration.

Thoughts, reviews, flames appreciated.

Brian

[1] https://lore.kernel.org/linux-fsdevel/[email protected]/
[2] https://lore.kernel.org/linux-xfs/[email protected]/

 fs/iomap/buffered-io.c | 45 ++++++++++++++++++++++++++++--------------
 fs/iomap/iter.c        |  6 +++---
 fs/xfs/xfs_iomap.c     | 11 ++++++-----
 include/linux/iomap.h  |  8 ++++++--
 4 files changed, 45 insertions(+), 25 deletions(-)
fs/iomap/buffered-io.c +30 -15
index 9b0b9cf7caa7..618f7f02f1bd 100644
@@ -772,7 +772,7 @@ static struct folio *__iomap_get_folio(struct iomap_iter *iter,
 	if (!mapping_large_folio_support(iter->inode->i_mapping))
 		len = min_t(size_t, len, PAGE_SIZE - offset_in_page(pos));
 
-	if (iter->fbatch) {
+	if (iter->iomap.flags & IOMAP_F_FOLIO_BATCH) {
 		struct folio *folio = folio_batch_next(iter->fbatch);
 
 		if (!folio)
@@ -869,7 +869,7 @@ static int iomap_write_begin(struct iomap_iter *iter,
 	 * process so return and let the caller iterate and refill the batch.
 	 */
 	if (!folio) {
-		WARN_ON_ONCE(!iter->fbatch);
+		WARN_ON_ONCE(!(iter->iomap.flags & IOMAP_F_FOLIO_BATCH));
 		return 0;
 	}
 
@@ -1483,23 +1483,34 @@ static int iomap_zero_iter(struct iomap_iter *iter, bool *did_zero,
 	return status;
 }
 
-loff_t
+/**
+ * iomap_fill_dirty_folios - fill a folio batch with dirty folios
+ * @iter: Iteration structure
+ * @start: Start offset of range. Updated based on lookup progress.
+ * @end: End offset of range
+ *
+ * Returns the associated control flag if the folio batch is available and the
+ * lookup performed. The caller is responsible to set the flag on the associated
+ * iomap.
+ */
+unsigned int
 iomap_fill_dirty_folios(
 	struct iomap_iter	*iter,
-	loff_t			offset,
-	loff_t			length)
+	loff_t			*start,
+	loff_t			end)
 {
 	struct address_space	*mapping = iter->inode->i_mapping;
-	pgoff_t			start = offset >> PAGE_SHIFT;
-	pgoff_t			end = (offset + length - 1) >> PAGE_SHIFT;
+	pgoff_t			pstart = *start >> PAGE_SHIFT;
+	pgoff_t			pend = (end - 1) >> PAGE_SHIFT;
 
-	iter->fbatch = kmalloc(sizeof(struct folio_batch), GFP_KERNEL);
-	if (!iter->fbatch)
-		return offset + length;
-	folio_batch_init(iter->fbatch);
+	if (!iter->fbatch) {
+		*start = end;
+		return 0;
+	}
 
-	filemap_get_folios_dirty(mapping, &start, end, iter->fbatch);
-	return (start << PAGE_SHIFT);
+	filemap_get_folios_dirty(mapping, &pstart, pend, iter->fbatch);
+	*start = (pstart << PAGE_SHIFT);
+	return IOMAP_F_FOLIO_BATCH;
 }
 EXPORT_SYMBOL_GPL(iomap_fill_dirty_folios);
 
@@ -1508,17 +1519,21 @@ iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
 		const struct iomap_ops *ops,
 		const struct iomap_write_ops *write_ops, void *private)
 {
+	struct folio_batch fbatch;
 	struct iomap_iter iter = {
 		.inode		= inode,
 		.pos		= pos,
 		.len		= len,
 		.flags		= IOMAP_ZERO,
 		.private	= private,
+		.fbatch		= &fbatch,
 	};
 	struct address_space *mapping = inode->i_mapping;
 	int ret;
 	bool range_dirty;
 
+	folio_batch_init(&fbatch);
+
 	/*
 	 * To avoid an unconditional flush, check pagecache state and only flush
 	 * if dirty and the fs returns a mapping that might convert on
@@ -1529,11 +1544,11 @@ iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
 	while ((ret = iomap_iter(&iter, ops)) > 0) {
 		const struct iomap *srcmap = iomap_iter_srcmap(&iter);
 
-		if (WARN_ON_ONCE(iter.fbatch &&
+		if (WARN_ON_ONCE((iter.iomap.flags & IOMAP_F_FOLIO_BATCH) &&
 				 srcmap->type != IOMAP_UNWRITTEN))
 			return -EIO;
 
-		if (!iter.fbatch &&
+		if (!(iter.iomap.flags & IOMAP_F_FOLIO_BATCH) &&
 		    (srcmap->type == IOMAP_HOLE ||
 		     srcmap->type == IOMAP_UNWRITTEN)) {
 			s64 status;
fs/iomap/iter.c +3 -3
index 8692e5e41c6d..c04796f6e57f 100644
@@ -8,10 +8,10 @@
 
 static inline void iomap_iter_reset_iomap(struct iomap_iter *iter)
 {
-	if (iter->fbatch) {
+	if (iter->iomap.flags & IOMAP_F_FOLIO_BATCH) {
 		folio_batch_release(iter->fbatch);
-		kfree(iter->fbatch);
-		iter->fbatch = NULL;
+		folio_batch_reinit(iter->fbatch);
+		iter->iomap.flags &= ~IOMAP_F_FOLIO_BATCH;
 	}
 
 	iter->status = 0;
fs/xfs/xfs_iomap.c +6 -5
index 04f39ea15898..639a85bd260a 100644
@@ -1831,7 +1831,6 @@ xfs_buffered_write_iomap_begin(
 	 */
 	if (flags & IOMAP_ZERO) {
 		xfs_fileoff_t eof_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
-		u64 end;
 
 		if (isnullstartblock(imap.br_startblock) &&
 		    offset_fsb >= eof_fsb)
@@ -1851,12 +1850,14 @@ xfs_buffered_write_iomap_begin(
 		 */
 		if (imap.br_state == XFS_EXT_UNWRITTEN &&
 		    offset_fsb < eof_fsb) {
-			loff_t len = min(count,
-					 XFS_FSB_TO_B(mp, imap.br_blockcount));
+			loff_t foffset = offset, fend;
 
-			end = iomap_fill_dirty_folios(iter, offset, len);
+			fend = offset +
+			       min(count, XFS_FSB_TO_B(mp, imap.br_blockcount));
+			iomap_flags |= iomap_fill_dirty_folios(iter, &foffset,
+							       fend);
 			end_fsb = min_t(xfs_fileoff_t, end_fsb,
-					XFS_B_TO_FSB(mp, end));
+					XFS_B_TO_FSB(mp, foffset));
 		}
 
 		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
include/linux/iomap.h +6 -2
index 8b1ac08c7474..fc3aa944d22f 100644
@@ -88,6 +88,9 @@ struct vm_fault;
 /*
  * Flags set by the core iomap code during operations:
  *
+ * IOMAP_F_FOLIO_BATCH indicates that the folio batch mechanism is active
+ * for this operation, set by iomap_fill_dirty_folios().
+ *
  * IOMAP_F_SIZE_CHANGED indicates to the iomap_end method that the file size
  * has changed as the result of this write operation.
  *
@@ -95,6 +98,7 @@ struct vm_fault;
  * range it covers needs to be remapped by the high level before the operation
  * can proceed.
  */
+#define IOMAP_F_FOLIO_BATCH	(1U << 13)
 #define IOMAP_F_SIZE_CHANGED	(1U << 14)
 #define IOMAP_F_STALE		(1U << 15)
 
@@ -352,8 +356,8 @@ bool iomap_dirty_folio(struct address_space *mapping, struct folio *folio);
 int iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len,
 		const struct iomap_ops *ops,
 		const struct iomap_write_ops *write_ops);
-loff_t iomap_fill_dirty_folios(struct iomap_iter *iter, loff_t offset,
-		loff_t length);
+unsigned int iomap_fill_dirty_folios(struct iomap_iter *iter, loff_t *start,
+		loff_t end);
 int iomap_zero_range(struct inode *inode, loff_t pos, loff_t len,
 		bool *did_zero, const struct iomap_ops *ops,
 		const struct iomap_write_ops *write_ops, void *private);
Signature · 1 line
2.51.1
#1 lore raw
Quoted 4 lines · Brian Foster
fs/xfs/xfs_iomap.c +0 -2
 		if (imap.br_state == XFS_EXT_UNWRITTEN &&
 		    offset_fsb < eof_fsb) {
-			loff_t len = min(count,
-					 XFS_FSB_TO_B(mp, imap.br_blockcount));

fs/xfs/xfs_iomap.c xfs_buffered_write_iomap_begin( · line 1860

Quoted 10 lines
fs/xfs/xfs_iomap.c +6 -2
+			loff_t foffset = offset, fend;
 
-			end = iomap_fill_dirty_folios(iter, offset, len);
+			fend = offset +
+			       min(count, XFS_FSB_TO_B(mp, imap.br_blockcount));
+			iomap_flags |= iomap_fill_dirty_folios(iter, &foffset,
+							       fend);
 			end_fsb = min_t(xfs_fileoff_t, end_fsb,
-					XFS_B_TO_FSB(mp, end));
+					XFS_B_TO_FSB(mp, foffset));

Maybe it's just me, but I found the old calling convention a lot more logic. Why not keep it and extend it for passing the flags as an in/out argument? That would also keep the churn down a bit.

Otherwise the change looks good.

#2 lore raw
Quoted 7 lines · Christoph Hellwig
Quoted 6 lines · Brian Foster
fs/xfs/xfs_iomap.c +1 -2
 		if (imap.br_state == XFS_EXT_UNWRITTEN &&
 		    offset_fsb < eof_fsb) {
-			loff_t len = min(count,
-					 XFS_FSB_TO_B(mp, imap.br_blockcount));
+			loff_t foffset = offset, fend;
Quoted 13 lines

fs/xfs/xfs_iomap.c xfs_buffered_write_iomap_begin( · line 1860

Quoted 8 lines
fs/xfs/xfs_iomap.c +5 -2
-			end = iomap_fill_dirty_folios(iter, offset, len);
+			fend = offset +
+			       min(count, XFS_FSB_TO_B(mp, imap.br_blockcount));
+			iomap_flags |= iomap_fill_dirty_folios(iter, &foffset,
+							       fend);
 			end_fsb = min_t(xfs_fileoff_t, end_fsb,
-					XFS_B_TO_FSB(mp, end));
+					XFS_B_TO_FSB(mp, foffset));

Maybe it's just me, but I found the old calling convention a lot more logic. Why not keep it and extend it for passing the flags as an in/out argument? That would also keep the churn down a bit.

Hmm.. well I never really loved the flag return (or the end return), but I wanted to make the iomap helper more consistent with the underlying filemap helper because I think that reduces unnecessary complexity. I suppose we could also make the flags an out param and either return void or just pass through the filemap helper return (i.e. folio count)...

Brian

Quoted 2 lines

Otherwise the change looks good.

#3 lore raw
Quoted 3 lines · Brian Foster

Hmm.. well I never really loved the flag return (or the end return), but I wanted to make the iomap helper more consistent with the underlying filemap helper because I think that reduces unnecessary complexity.

Probably. But the weird filemap (in general, not just the helper) calling convention of using page indices and non-inclusive ends doesn't really help with making it useful for higher level code.

Quoted 3 lines

I
suppose we could also make the flags an out param and either return void or just pass through the filemap helper return (i.e. folio count)...

That would seem more logical to me. Besides not really liking the new convention overly much (but that's a minor nitpick) my main issue was that this was hidden in a functional change without any explanation.