Merge patch series "further iomap large atomic writes changes"

John Garry <john.g.garry@oracle.com> says:

These iomap changes are spun-off the XFS large atomic writes series at
https://lore.kernel.org/linux-xfs/86a64256-497a-453b-bbba-a5ac6b4cb056@oracle.com/T/#ma99c763221de9d49ea2ccfca9ff9b8d71c8b2677

The XFS parts there are not ready yet, but it is worth having the iomap
changes queued in advance.

Some much earlier changes from that same series were already queued in the
vfs tree, and these patches rework those changes - specifically the
first patch in this series does.

The most other significant change is the patch to rework how the bio flags
are set in the DIO patch.

* patches from https://lore.kernel.org/r/20250320120250.4087011-1-john.g.garry@oracle.com:
  iomap: rework IOMAP atomic flags
  iomap: comment on atomic write checks in iomap_dio_bio_iter()
  iomap: inline iomap_dio_bio_opflags()

Link: https://lore.kernel.org/r/20250320120250.4087011-1-john.g.garry@oracle.com
Signed-off-by: Christian Brauner <brauner@kernel.org>

6 files changed, 91 insertions, 93 deletions

diff --git a/Documentation/filesystems/iomap/operations.rst b/Documentation/filesystems/iomap/operations.rst
index b08a79d11d9f..3b628e370d88 100644
--- a/Documentation/filesystems/iomap/operations.rst
+++ b/Documentation/filesystems/iomap/operations.rst
@@ -514,29 +514,32 @@ IOMAP_WRITE`` with any combination of the following enhancements:
    if the mapping is unwritten and the filesystem cannot handle zeroing
    the unaligned regions without exposing stale contents.
 
- * ``IOMAP_ATOMIC_HW``: This write is being issued with torn-write
-   protection based on HW-offload support.
-   Only a single bio can be created for the write, and the write must
-   not be split into multiple I/O requests, i.e. flag REQ_ATOMIC must be
-   set.
+ * ``IOMAP_ATOMIC``: This write is being issued with torn-write
+   protection.
+   Torn-write protection may be provided based on HW-offload or by a
+   software mechanism provided by the filesystem.
+
+   For HW-offload based support, only a single bio can be created for the
+   write, and the write must not be split into multiple I/O requests, i.e.
+   flag REQ_ATOMIC must be set.
    The file range to write must be aligned to satisfy the requirements
    of both the filesystem and the underlying block device's atomic
    commit capabilities.
    If filesystem metadata updates are required (e.g. unwritten extent
-   conversion or copy on write), all updates for the entire file range
+   conversion or copy-on-write), all updates for the entire file range
    must be committed atomically as well.
-   Only one space mapping is allowed per untorn write.
-   Untorn writes may be longer than a single file block. In all cases,
+   Untorn-writes may be longer than a single file block. In all cases,
    the mapping start disk block must have at least the same alignment as
    the write offset.
-
- * ``IOMAP_ATOMIC_SW``: This write is being issued with torn-write
-   protection via a software mechanism provided by the filesystem.
-   All the disk block alignment and single bio restrictions which apply
-   to IOMAP_ATOMIC_HW do not apply here.
-   SW-based untorn writes would typically be used as a fallback when
-   HW-based untorn writes may not be issued, e.g. the range of the write
-   covers multiple extents, meaning that it is not possible to issue
+   The filesystems must set IOMAP_F_ATOMIC_BIO to inform iomap core of an
+   untorn-write based on HW-offload.
+
+   For untorn-writes based on a software mechanism provided by the
+   filesystem, all the disk block alignment and single bio restrictions
+   which apply for HW-offload based untorn-writes do not apply.
+   The mechanism would typically be used as a fallback for when
+   HW-offload based untorn-writes may not be issued, e.g. the range of the
+   write covers multiple extents, meaning that it is not possible to issue
    a single bio.
    All filesystem metadata updates for the entire file range must be
    committed atomically as well.
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index ba2f1e3db7c7..d04d8a7f12e7 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3290,6 +3290,10 @@ static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
 	if (map->m_flags & EXT4_MAP_NEW)
 		iomap->flags |= IOMAP_F_NEW;
 
+	/* HW-offload atomics are always used */
+	if (flags & IOMAP_ATOMIC)
+		iomap->flags |= IOMAP_F_ATOMIC_BIO;
+
 	if (flags & IOMAP_DAX)
 		iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
 	else
@@ -3467,7 +3471,7 @@ static inline bool ext4_want_directio_fallback(unsigned flags, ssize_t written)
 		return false;
 
 	/* atomic writes are all-or-nothing */
-	if (flags & IOMAP_ATOMIC_HW)
+	if (flags & IOMAP_ATOMIC)
 		return false;
 
 	/* can only try again if we wrote nothing */
diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index 5299f70428ef..6ac7a1534f7c 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -312,27 +312,20 @@ static int iomap_dio_zero(const struct iomap_iter *iter, struct iomap_dio *dio,
 }
 
 /*
- * Figure out the bio's operation flags from the dio request, the
- * mapping, and whether or not we want FUA.  Note that we can end up
- * clearing the WRITE_THROUGH flag in the dio request.
+ * Use a FUA write if we need datasync semantics and this is a pure data I/O
+ * that doesn't require any metadata updates (including after I/O completion
+ * such as unwritten extent conversion) and the underlying device either
+ * doesn't have a volatile write cache or supports FUA.
+ * This allows us to avoid cache flushes on I/O completion.
  */
-static inline blk_opf_t iomap_dio_bio_opflags(struct iomap_dio *dio,
-		const struct iomap *iomap, bool use_fua, bool atomic_hw)
+static inline bool iomap_dio_can_use_fua(const struct iomap *iomap,
+		struct iomap_dio *dio)
 {
-	blk_opf_t opflags = REQ_SYNC | REQ_IDLE;
-
-	if (!(dio->flags & IOMAP_DIO_WRITE))
-		return REQ_OP_READ;
-
-	opflags |= REQ_OP_WRITE;
-	if (use_fua)
-		opflags |= REQ_FUA;
-	else
-		dio->flags &= ~IOMAP_DIO_WRITE_THROUGH;
-	if (atomic_hw)
-		opflags |= REQ_ATOMIC;
-
-	return opflags;
+	if (iomap->flags & (IOMAP_F_SHARED | IOMAP_F_DIRTY))
+		return false;
+	if (!(dio->flags & IOMAP_DIO_WRITE_THROUGH))
+		return false;
+	return !bdev_write_cache(iomap->bdev) || bdev_fua(iomap->bdev);
 }
 
 static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
@@ -340,52 +333,64 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
 	const struct iomap *iomap = &iter->iomap;
 	struct inode *inode = iter->inode;
 	unsigned int fs_block_size = i_blocksize(inode), pad;
-	bool atomic_hw = iter->flags & IOMAP_ATOMIC_HW;
 	const loff_t length = iomap_length(iter);
 	loff_t pos = iter->pos;
-	blk_opf_t bio_opf;
+	blk_opf_t bio_opf = REQ_SYNC | REQ_IDLE;
 	struct bio *bio;
 	bool need_zeroout = false;
-	bool use_fua = false;
 	int nr_pages, ret = 0;
 	u64 copied = 0;
 	size_t orig_count;
 
-	if (atomic_hw && length != iter->len)
-		return -EINVAL;
-
 	if ((pos | length) & (bdev_logical_block_size(iomap->bdev) - 1) ||
 	    !bdev_iter_is_aligned(iomap->bdev, dio->submit.iter))
 		return -EINVAL;
 
-	if (iomap->type == IOMAP_UNWRITTEN) {
-		dio->flags |= IOMAP_DIO_UNWRITTEN;
-		need_zeroout = true;
-	}
+	if (dio->flags & IOMAP_DIO_WRITE) {
+		bio_opf |= REQ_OP_WRITE;
+
+		if (iomap->flags & IOMAP_F_ATOMIC_BIO) {
+			/*
+			 * Ensure that the mapping covers the full write
+			 * length, otherwise it won't be submitted as a single
+			 * bio, which is required to use hardware atomics.
+			 */
+			if (length != iter->len)
+				return -EINVAL;
+			bio_opf |= REQ_ATOMIC;
+		}
 
-	if (iomap->flags & IOMAP_F_SHARED)
-		dio->flags |= IOMAP_DIO_COW;
+		if (iomap->type == IOMAP_UNWRITTEN) {
+			dio->flags |= IOMAP_DIO_UNWRITTEN;
+			need_zeroout = true;
+		}
+
+		if (iomap->flags & IOMAP_F_SHARED)
+			dio->flags |= IOMAP_DIO_COW;
+
+		if (iomap->flags & IOMAP_F_NEW) {
+			need_zeroout = true;
+		} else if (iomap->type == IOMAP_MAPPED) {
+			if (iomap_dio_can_use_fua(iomap, dio))
+				bio_opf |= REQ_FUA;
+			else
+				dio->flags &= ~IOMAP_DIO_WRITE_THROUGH;
+		}
 
-	if (iomap->flags & IOMAP_F_NEW) {
-		need_zeroout = true;
-	} else if (iomap->type == IOMAP_MAPPED) {
 		/*
-		 * Use a FUA write if we need datasync semantics, this is a pure
-		 * data IO that doesn't require any metadata updates (including
-		 * after IO completion such as unwritten extent conversion) and
-		 * the underlying device either supports FUA or doesn't have
-		 * a volatile write cache. This allows us to avoid cache flushes
-		 * on IO completion. If we can't use writethrough and need to
-		 * sync, disable in-task completions as dio completion will
-		 * need to call generic_write_sync() which will do a blocking
-		 * fsync / cache flush call.
+		 * We can only do deferred completion for pure overwrites that
+		 * don't require additional I/O at completion time.
+		 *
+		 * This rules out writes that need zeroing or extent conversion,
+		 * extend the file size, or issue metadata I/O or cache flushes
+		 * during completion processing.
 		 */
-		if (!(iomap->flags & (IOMAP_F_SHARED|IOMAP_F_DIRTY)) &&
-		    (dio->flags & IOMAP_DIO_WRITE_THROUGH) &&
-		    (bdev_fua(iomap->bdev) || !bdev_write_cache(iomap->bdev)))
-			use_fua = true;
-		else if (dio->flags & IOMAP_DIO_NEED_SYNC)
+		if (need_zeroout || (pos >= i_size_read(inode)) ||
+		    ((dio->flags & IOMAP_DIO_NEED_SYNC) &&
+		     !(bio_opf & REQ_FUA)))
 			dio->flags &= ~IOMAP_DIO_CALLER_COMP;
+	} else {
+		bio_opf |= REQ_OP_READ;
 	}
 
 	/*
@@ -400,18 +405,6 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
 		goto out;
 
 	/*
-	 * We can only do deferred completion for pure overwrites that
-	 * don't require additional IO at completion. This rules out
-	 * writes that need zeroing or extent conversion, extend
-	 * the file size, or issue journal IO or cache flushes
-	 * during completion processing.
-	 */
-	if (need_zeroout ||
-	    ((dio->flags & IOMAP_DIO_NEED_SYNC) && !use_fua) ||
-	    ((dio->flags & IOMAP_DIO_WRITE) && pos >= i_size_read(inode)))
-		dio->flags &= ~IOMAP_DIO_CALLER_COMP;
-
-	/*
 	 * The rules for polled IO completions follow the guidelines as the
 	 * ones we set for inline and deferred completions. If none of those
 	 * are available for this IO, clear the polled flag.
@@ -428,8 +421,6 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
 			goto out;
 	}
 
-	bio_opf = iomap_dio_bio_opflags(dio, iomap, use_fua, atomic_hw);
-
 	nr_pages = bio_iov_vecs_to_alloc(dio->submit.iter, BIO_MAX_VECS);
 	do {
 		size_t n;
@@ -461,9 +452,9 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio)
 		}
 
 		n = bio->bi_iter.bi_size;
-		if (WARN_ON_ONCE(atomic_hw && n != length)) {
+		if (WARN_ON_ONCE((bio_opf & REQ_ATOMIC) && n != length)) {
 			/*
-			 * This bio should have covered the complete length,
+			 * An atomic write bio must cover the complete length,
 			 * which it doesn't, so error. We may need to zero out
 			 * the tail (complete FS block), similar to when
 			 * bio_iov_iter_get_pages() returns an error, above.
@@ -686,10 +677,8 @@ __iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 			iomi.flags |= IOMAP_OVERWRITE_ONLY;
 		}
 
-		if (dio_flags & IOMAP_DIO_ATOMIC_SW)
-			iomi.flags |= IOMAP_ATOMIC_SW;
-		else if (iocb->ki_flags & IOCB_ATOMIC)
-			iomi.flags |= IOMAP_ATOMIC_HW;
+		if (iocb->ki_flags & IOCB_ATOMIC)
+			iomi.flags |= IOMAP_ATOMIC;
 
 		/* for data sync or sync, we need sync completion processing */
 		if (iocb_is_dsync(iocb)) {
diff --git a/fs/iomap/trace.h b/fs/iomap/trace.h
index 69af89044ebd..9eab2c8ac3c5 100644
--- a/fs/iomap/trace.h
+++ b/fs/iomap/trace.h
@@ -99,7 +99,7 @@ DEFINE_RANGE_EVENT(iomap_dio_rw_queued);
 	{ IOMAP_FAULT,		"FAULT" }, \
 	{ IOMAP_DIRECT,		"DIRECT" }, \
 	{ IOMAP_NOWAIT,		"NOWAIT" }, \
-	{ IOMAP_ATOMIC_HW,	"ATOMIC_HW" }
+	{ IOMAP_ATOMIC,		"ATOMIC" }
 
 #define IOMAP_F_FLAGS_STRINGS \
 	{ IOMAP_F_NEW,		"NEW" }, \
diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index 5dd0922fe2d1..ee40dc509413 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -828,6 +828,10 @@ xfs_direct_write_iomap_begin(
 	if (offset + length > i_size_read(inode))
 		iomap_flags |= IOMAP_F_DIRTY;
 
+	/* HW-offload atomics are always used in this path */
+	if (flags & IOMAP_ATOMIC)
+		iomap_flags |= IOMAP_F_ATOMIC_BIO;
+
 	/*
 	 * COW writes may allocate delalloc space or convert unwritten COW
 	 * extents, so we need to make sure to take the lock exclusively here.
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 9cd93530013c..02fe001feebb 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -60,6 +60,9 @@ struct vm_fault;
  * IOMAP_F_ANON_WRITE indicates that (write) I/O does not have a target block
  * assigned to it yet and the file system will do that in the bio submission
  * handler, splitting the I/O as needed.
+ *
+ * IOMAP_F_ATOMIC_BIO indicates that (write) I/O will be issued as an atomic
+ * bio, i.e. set REQ_ATOMIC.
  */
 #define IOMAP_F_NEW		(1U << 0)
 #define IOMAP_F_DIRTY		(1U << 1)
@@ -73,6 +76,7 @@ struct vm_fault;
 #define IOMAP_F_XATTR		(1U << 5)
 #define IOMAP_F_BOUNDARY	(1U << 6)
 #define IOMAP_F_ANON_WRITE	(1U << 7)
+#define IOMAP_F_ATOMIC_BIO	(1U << 8)
 
 /*
  * Flags set by the core iomap code during operations:
@@ -189,9 +193,8 @@ struct iomap_folio_ops {
 #else
 #define IOMAP_DAX		0
 #endif /* CONFIG_FS_DAX */
-#define IOMAP_ATOMIC_HW		(1 << 9) /* HW-based torn-write protection */
+#define IOMAP_ATOMIC		(1 << 9) /* torn-write protection */
 #define IOMAP_DONTCACHE		(1 << 10)
-#define IOMAP_ATOMIC_SW		(1 << 11)/* SW-based torn-write protection */
 
 struct iomap_ops {
 	/*
@@ -503,11 +506,6 @@ struct iomap_dio_ops {
  */
 #define IOMAP_DIO_PARTIAL		(1 << 2)
 
-/*
- * Use software-based torn-write protection.
- */
-#define IOMAP_DIO_ATOMIC_SW		(1 << 3)
-
 ssize_t iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
 		const struct iomap_ops *ops, const struct iomap_dio_ops *dops,
 		unsigned int dio_flags, void *private, size_t done_before);