diff mbox series

[RFC,3/3] block/file-posix: Let post-EOF fallocate serialize

Message ID 20191025095849.25283-4-mreitz@redhat.com
State New
Headers show
Series block/file-posix: Work around XFS bug | expand

Commit Message

Max Reitz Oct. 25, 2019, 9:58 a.m. UTC 
The XFS kernel driver has a bug that may cause data corruption for qcow2
images as of qemu commit c8bb23cbdbe32f.  We can work around it by
treating post-EOF fallocates as serializing up until infinity (INT64_MAX
in practice).

Signed-off-by: Max Reitz <mreitz@redhat.com>
---
 block/file-posix.c | 42 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)

Comments

Nir Soffer Oct. 26, 2019, 5:28 p.m. UTC | #1 
On Fri, Oct 25, 2019 at 1:24 PM Max Reitz <mreitz@redhat.com> wrote:
>
> The XFS kernel driver has a bug that may cause data corruption for qcow2
> images as of qemu commit c8bb23cbdbe32f.  We can work around it by
> treating post-EOF fallocates as serializing up until infinity (INT64_MAX
> in practice).
>
> Signed-off-by: Max Reitz <mreitz@redhat.com>
> ---
>  block/file-posix.c | 42 ++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 42 insertions(+)
>
> diff --git a/block/file-posix.c b/block/file-posix.c
> index 5cd54c8bff..1f5a01df70 100644
> --- a/block/file-posix.c
> +++ b/block/file-posix.c
> @@ -2713,6 +2713,48 @@ raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
>      RawPosixAIOData acb;
>      ThreadPoolFunc *handler;
>
> +#ifdef CONFIG_FALLOCATE
> +    if (s->is_xfs && s->use_linux_aio &&

This limit the fix to local xfs filesystem, but the fix may be needed
to remote filesystem such
as gluster over xfs.

> +        offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE)
> +    {
> +        BdrvTrackedRequest *req;
> +        uint64_t end;
> +
> +        /*
> +         * The Linux XFS driver has a bug where it will discard writes
> +         * submitted through the AIO interface if they happen beyond a
> +         * concurrently running fallocate() that increases the file
> +         * length (i.e., both the write and the fallocate() happen
> +         * beyond the EOF).
> +         *
> +         * To work around it, we look for the tracked request for this
> +         * zero write, extend it until INT64_MAX (effectively
> +         * infinity), and mark it as serializing.
> +         *
> +         * TODO: Detect whether this has been fixed in the XFS driver.
> +         */
> +
> +        QLIST_FOREACH(req, &bs->tracked_requests, list) {
> +            if (req->co == qemu_coroutine_self() &&
> +                req->type == BDRV_TRACKED_WRITE)
> +            {
> +                break;
> +            }
> +        }
> +
> +        assert(req);
> +        assert(req->offset <= offset);
> +        assert(req->offset + req->bytes >= offset + bytes);
> +
> +        end = INT64_MAX & -(uint64_t)bs->bl.request_alignment;
> +        req->bytes = end - req->offset;
> +        req->overlap_bytes = req->bytes;
> +
> +        bdrv_mark_request_serialising(req, bs->bl.request_alignment);
> +        bdrv_wait_serialising_requests(req);
> +    }
> +#endif
> +
>      acb = (RawPosixAIOData) {
>          .bs             = bs,
>          .aio_fildes     = s->fd,
> --
> 2.21.0
>
>
diff mbox series

Patch

diff --git a/block/file-posix.c b/block/file-posix.c
index 5cd54c8bff..1f5a01df70 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -2713,6 +2713,48 @@  raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
     RawPosixAIOData acb;
     ThreadPoolFunc *handler;
 
+#ifdef CONFIG_FALLOCATE
+    if (s->is_xfs && s->use_linux_aio &&
+        offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE)
+    {
+        BdrvTrackedRequest *req;
+        uint64_t end;
+
+        /*
+         * The Linux XFS driver has a bug where it will discard writes
+         * submitted through the AIO interface if they happen beyond a
+         * concurrently running fallocate() that increases the file
+         * length (i.e., both the write and the fallocate() happen
+         * beyond the EOF).
+         *
+         * To work around it, we look for the tracked request for this
+         * zero write, extend it until INT64_MAX (effectively
+         * infinity), and mark it as serializing.
+         *
+         * TODO: Detect whether this has been fixed in the XFS driver.
+         */
+
+        QLIST_FOREACH(req, &bs->tracked_requests, list) {
+            if (req->co == qemu_coroutine_self() &&
+                req->type == BDRV_TRACKED_WRITE)
+            {
+                break;
+            }
+        }
+
+        assert(req);
+        assert(req->offset <= offset);
+        assert(req->offset + req->bytes >= offset + bytes);
+
+        end = INT64_MAX & -(uint64_t)bs->bl.request_alignment;
+        req->bytes = end - req->offset;
+        req->overlap_bytes = req->bytes;
+
+        bdrv_mark_request_serialising(req, bs->bl.request_alignment);
+        bdrv_wait_serialising_requests(req);
+    }
+#endif
+
     acb = (RawPosixAIOData) {
         .bs             = bs,
         .aio_fildes     = s->fd,