Merge branch 'iommufd/arm-smmuv3-nested' of iommu/linux into iommufd for-next

Common SMMUv3 patches for the following patches adding nesting, shared
branch with the iommu tree.

* 'iommufd/arm-smmuv3-nested' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/iommu/linux:
  iommu/arm-smmu-v3: Expose the arm_smmu_attach interface
  iommu/arm-smmu-v3: Implement IOMMU_HWPT_ALLOC_NEST_PARENT
  iommu/arm-smmu-v3: Support IOMMU_GET_HW_INFO via struct arm_smmu_hw_info
  iommu/arm-smmu-v3: Report IOMMU_CAP_ENFORCE_CACHE_COHERENCY for CANWBS
  ACPI/IORT: Support CANWBS memory access flag
  ACPICA: IORT: Update for revision E.f
  vfio: Remove VFIO_TYPE1_NESTING_IOMMU
  ...

Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

1 files changed, 27 insertions, 4 deletions

diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
index 7f48ba6c1c77..27306d98ec43 100644
--- a/fs/btrfs/send.c
+++ b/fs/btrfs/send.c
@@ -346,8 +346,10 @@ struct name_cache_entry {
 	u64 parent_gen;
 	int ret;
 	int need_later_update;
+	/* Name length without NUL terminator. */
 	int name_len;
-	char name[] __counted_by(name_len);
+	/* Not NUL terminated. */
+	char name[] __counted_by(name_len) __nonstring;
 };
 
 /* See the comment at lru_cache.h about struct btrfs_lru_cache_entry. */
@@ -2388,7 +2390,7 @@ out_cache:
 	/*
 	 * Store the result of the lookup in the name cache.
 	 */
-	nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
+	nce = kmalloc(sizeof(*nce) + fs_path_len(dest), GFP_KERNEL);
 	if (!nce) {
 		ret = -ENOMEM;
 		goto out;
@@ -2400,7 +2402,7 @@ out_cache:
 	nce->parent_gen = *parent_gen;
 	nce->name_len = fs_path_len(dest);
 	nce->ret = ret;
-	strcpy(nce->name, dest->start);
+	memcpy(nce->name, dest->start, nce->name_len);
 
 	if (ino < sctx->send_progress)
 		nce->need_later_update = 0;
@@ -6187,8 +6189,29 @@ static int send_write_or_clone(struct send_ctx *sctx,
 	if (ret < 0)
 		return ret;
 
-	if (clone_root->offset + num_bytes == info.size)
+	if (clone_root->offset + num_bytes == info.size) {
+		/*
+		 * The final size of our file matches the end offset, but it may
+		 * be that its current size is larger, so we have to truncate it
+		 * to any value between the start offset of the range and the
+		 * final i_size, otherwise the clone operation is invalid
+		 * because it's unaligned and it ends before the current EOF.
+		 * We do this truncate to the final i_size when we finish
+		 * processing the inode, but it's too late by then. And here we
+		 * truncate to the start offset of the range because it's always
+		 * sector size aligned while if it were the final i_size it
+		 * would result in dirtying part of a page, filling part of a
+		 * page with zeroes and then having the clone operation at the
+		 * receiver trigger IO and wait for it due to the dirty page.
+		 */
+		if (sctx->parent_root != NULL) {
+			ret = send_truncate(sctx, sctx->cur_ino,
+					    sctx->cur_inode_gen, offset);
+			if (ret < 0)
+				return ret;
+		}
 		goto clone_data;
+	}
 
 write_data:
 	ret = send_extent_data(sctx, path, offset, num_bytes);