[1/2] block/fuse: Let PUNCH_HOLE write zeroes

Commit Message

Hanna Czenczek Feb. 27, 2023, 10:47 a.m. UTC

fallocate(2) says about PUNCH_HOLE: "After a successful call, subsequent
reads from this range will return zeros."  As it is, PUNCH_HOLE is
implemented as a call to blk_pdiscard(), which does not guarantee this.

We must call blk_pwrite_zeroes() instead.  The difference to ZERO_RANGE
is that we pass the `BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK` flags to
the call -- the storage is supposed to be unmapped, and a slow fallback
by actually writing zeroes as data is not allowed.

Closes: https://gitlab.com/qemu-project/qemu/-/issues/1507
Signed-off-by: Hanna Czenczek <[e-mail]>
---
 block/export/fuse.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

Patch

diff --git a/block/export/fuse.c b/block/export/fuse.c
index e5fc4af165..06fa41079e 100644
--- a/block/export/fuse.c
+++ b/block/export/fuse.c
@@ -673,7 +673,16 @@  static void fuse_fallocate(fuse_req_t req, fuse_ino_t inode, int mode,
         do {
             int size = MIN(length, BDRV_REQUEST_MAX_BYTES);
 
-            ret = blk_pdiscard(exp->common.blk, offset, size);
+            ret = blk_pwrite_zeroes(exp->common.blk, offset, size,
+                                    BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK);
+            if (ret == -ENOTSUP) {
+                /*
+                 * fallocate() specifies to return EOPNOTSUPP for unsupported
+                 * operations
+                 */
+                ret = -EOPNOTSUPP;
+            }
+
             offset += size;
             length -= size;
         } while (ret == 0 && length > 0);