[PULL 0/1] Block layer patches

[PULL 0/1] Block layer patches

[Kevin Wolf]

The following changes since commit 66ec38b6fa593afcc26a2dfb5d1f9871dd15f527:

  Merge tag 'pull-target-arm-20251201' of https://gitlab.com/pm215/qemu into staging (2025-12-01 12:08:52 -0800)

are available in the Git repository at:

  https://repo.or.cz/qemu/kevin.git tags/for-upstream

for you to fetch changes up to d94ea1a10bdfb99c5822e64a5b6ee3e352719326:

  file-posix: Handle suspended dm-multipath better for SG_IO (2025-12-02 16:51:47 +0100)

----------------------------------------------------------------
Block layer patches

- SCSI passthrough: Fix errors on temporarily suspended dm-multipath

----------------------------------------------------------------
Kevin Wolf (1):
      file-posix: Handle suspended dm-multipath better for SG_IO

 block/file-posix.c | 56 +++++++++++++++++++++++++++++++++---------------------
 1 file changed, 34 insertions(+), 22 deletions(-)

[PULL 1/1] file-posix: Handle suspended dm-multipath better for SG_IO

[Kevin Wolf]

When introducing DM_MPATH_PROBE_PATHS, we already anticipated that
dm-multipath devices might be suspended for a short time when the DM
tables are reloaded and that they return -EAGAIN in this case. The
behaviour promised in the comment wasn't actually implemented, though:
We don't get SG_IO_MAX_RETRIES in practice, because after the first
1ms sleep, DM_MPATH_PROBE_PATHS is called and if that still fails with
-EAGAIN, we error out immediately without any retry.

However, meanwhile it has also turned out that libmpathpersist (which is
used by qemu-pr-helper) may need to perform more complex recovery
operations to get reservations back to expected state if a path failure
happened in the middle of a PR operation. In this case, the device is
suspended for a longer time compared to the case we originally expected.

This patch changes hdev_co_ioctl() to treat -EAGAIN separately so that
it doesn't result in an immediate failure if the device is suspended for
more than 1ms, and moves to incremental backoff to cover both quick and
slow cases without excessive delays.

Buglink: https://issues.redhat.com/browse/RHEL-121543
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20251128221440.89125-1-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 block/file-posix.c | 56 ++++++++++++++++++++++++++++------------------
 1 file changed, 34 insertions(+), 22 deletions(-)

diff --git a/block/file-posix.c b/block/file-posix.c
index c9e367a222..6265d2e248 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -4288,25 +4288,8 @@ hdev_open_Mac_error:
 static bool coroutine_fn sgio_path_error(int ret, sg_io_hdr_t *io_hdr)
 {
     if (ret < 0) {
-        switch (ret) {
-        case -ENODEV:
-            return true;
-        case -EAGAIN:
-            /*
-             * The device is probably suspended. This happens while the dm table
-             * is reloaded, e.g. because a path is added or removed. This is an
-             * operation that should complete within 1ms, so just wait a bit and
-             * retry.
-             *
-             * If the device was suspended for another reason, we'll wait and
-             * retry SG_IO_MAX_RETRIES times. This is a tolerable delay before
-             * we return an error and potentially stop the VM.
-             */
-            qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000);
-            return true;
-        default:
-            return false;
-        }
+        /* Path errors sometimes result in -ENODEV */
+        return ret == -ENODEV;
     }
 
     if (io_hdr->host_status != SCSI_HOST_OK) {
@@ -4375,6 +4358,7 @@ hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 {
     BDRVRawState *s = bs->opaque;
     RawPosixAIOData acb;
+    uint64_t eagain_sleep_ns = 1 * SCALE_MS;
     int retries = SG_IO_MAX_RETRIES;
     int ret;
 
@@ -4403,9 +4387,37 @@ hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
         },
     };
 
-    do {
-        ret = raw_thread_pool_submit(handle_aiocb_ioctl, &acb);
-    } while (req == SG_IO && retries-- && hdev_co_ioctl_sgio_retry(&acb, ret));
+retry:
+    ret = raw_thread_pool_submit(handle_aiocb_ioctl, &acb);
+    if (req == SG_IO && s->use_mpath) {
+        if (ret == -EAGAIN && eagain_sleep_ns < NANOSECONDS_PER_SECOND) {
+            /*
+             * If this is a multipath device, it is probably suspended.
+             *
+             * This can happen while the dm table is reloaded, e.g. because a
+             * path is added or removed. This is an operation that should
+             * complete within 1ms, so just wait a bit and retry.
+             *
+             * There are also some cases in which libmpathpersist must recover
+             * from path failure during its operation, which can leave the
+             * device suspended for a bit longer while the library brings back
+             * reservations into the expected state.
+             *
+             * Use increasing delays to cover both cases without waiting
+             * excessively, and stop after a bit more than a second (1023 ms).
+             * This is a tolerable delay before we return an error and
+             * potentially stop the VM.
+             */
+            qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, eagain_sleep_ns);
+            eagain_sleep_ns *= 2;
+            goto retry;
+        }
+
+        /* Even for ret == 0, the SG_IO header can contain an error */
+        if (retries-- && hdev_co_ioctl_sgio_retry(&acb, ret)) {
+            goto retry;
+        }
+    }
 
     return ret;
 }
-- 
2.52.0

Re: [PULL 0/1] Block layer patches

[Richard Henderson]

On 12/2/25 10:26, Kevin Wolf wrote:
> The following changes since commit 66ec38b6fa593afcc26a2dfb5d1f9871dd15f527:
> 
>    Merge tag 'pull-target-arm-20251201' ofhttps://gitlab.com/pm215/qemu into staging (2025-12-01 12:08:52 -0800)
> 
> are available in the Git repository at:
> 
>    https://repo.or.cz/qemu/kevin.git tags/for-upstream
> 
> for you to fetch changes up to d94ea1a10bdfb99c5822e64a5b6ee3e352719326:
> 
>    file-posix: Handle suspended dm-multipath better for SG_IO (2025-12-02 16:51:47 +0100)
> 
> ----------------------------------------------------------------
> Block layer patches
> 
> - SCSI passthrough: Fix errors on temporarily suspended dm-multipath


Applied, thanks.  Please update https://wiki.qemu.org/ChangeLog/10.2 as appropriate.

r~