Merge tag 'for-upstream' of https://repo.or.cz/qemu/kevin into staging

Block layer patches

- scsi-disk: Add native FUA write support, enable FUA by default
- qemu-img: fix offset calculation in bench
- file-posix: allow BLKZEROOUT with -t writeback
- file-posix: Probe paths and retry SG_IO on potential path errors

# -----BEGIN PGP SIGNATURE-----
#
# iQJFBAABCAAvFiEE3D3rFZqa+V09dFb+fwmycsiPL9YFAmgvbSARHGt3b2xmQHJl
# ZGhhdC5jb20ACgkQfwmycsiPL9ZDmhAAm34XjSe/aQJG76Nll18eO8lnoWidsnjX
# OoSI3//O77dE0AuMOxSVvfve+ZkV4ehZDeo+GeREBZh3TGBPHf+elVfo3XEu/L5u
# +jWTNgoNe7pdlkxB4pv9boHIy2C7+/odVNtmVggxuNy2kyQoKP0tXHEjqiHjNzB6
# jCHXegklC9psXt+wgQGJpFkeI8r2OtABkEP15PtI02gvQyk+spBg0sr6pB5FP2ZY
# y8YWgOXeXis6QHqARMpqoiRGxhCQrWuKuzwZfeyNmLvYLWCBqmt3Opk+3kMPU6NB
# KrXZHCJXtaIlLO7YeurPXcnFQsJ94IY7x1TvIza5mgY+ct7mal2uN4u29PnhnLLm
# eXSgSiXhb8h9PY8KfOQfU9brclijcbV8Rn7sSP6WPX00bHspib275gNG4RLPxnxi
# AezWfBg1IOYAvliwq99ZY+Ts+faezo7XiNQbfNpZ82pzxhO6IqemSH2IcDS6SpjO
# mNoGLNmCi3CpQw1bdlnqwiU1OxWsHK2627VGyLZXPnOrYPr+erN/A4Nucxr2bzJk
# 69dKg0/ekTucTtKiF1uRl/bkhTOHplOogyyuKala6ogsthvBV8jmGAEEqZnMTVtx
# opAl7MSpVa4FWQ7C2LHi5vYEVNGC8OLXYBy09N73m/Q+Fs+z6bViFy3BQdUO+Nh1
# OjmUGzSrqRo=
# =yjOc
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 22 May 2025 14:29:52 EDT
# gpg:                using RSA key DC3DEB159A9AF95D3D7456FE7F09B272C88F2FD6
# gpg:                issuer "kwolf@redhat.com"
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full]
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* tag 'for-upstream' of https://repo.or.cz/qemu/kevin:
  file-posix: Probe paths and retry SG_IO on potential path errors
  file-posix: allow BLKZEROOUT with -t writeback
  qemu-img: fix offset calculation in bench
  scsi-disk: Advertise FUA support by default
  scsi-disk: Add native FUA write support

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

4 files changed, 135 insertions, 54 deletions

diff --git a/block/file-posix.c b/block/file-posix.c
index ec95b74869..9b5f08ccb2 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -41,6 +41,7 @@
 
 #include "scsi/pr-manager.h"
 #include "scsi/constants.h"
+#include "scsi/utils.h"
 
 #if defined(__APPLE__) && (__MACH__)
 #include <sys/ioctl.h>
@@ -72,6 +73,7 @@
 #include <linux/blkzoned.h>
 #endif
 #include <linux/cdrom.h>
+#include <linux/dm-ioctl.h>
 #include <linux/fd.h>
 #include <linux/fs.h>
 #include <linux/hdreg.h>
@@ -138,6 +140,22 @@
 #define RAW_LOCK_PERM_BASE             100
 #define RAW_LOCK_SHARED_BASE           200
 
+/*
+ * Multiple retries are mostly meant for two separate scenarios:
+ *
+ * - DM_MPATH_PROBE_PATHS returns success, but before SG_IO completes, another
+ *   path goes down.
+ *
+ * - DM_MPATH_PROBE_PATHS failed all paths in the current path group, so we have
+ *   to send another SG_IO to switch to another path group to probe the paths in
+ *   it.
+ *
+ * Even if each path is in a separate path group (path_grouping_policy set to
+ * failover), it's rare to have more than eight path groups - and even then
+ * pretty unlikely that only bad path groups would be chosen in eight retries.
+ */
+#define SG_IO_MAX_RETRIES 8
+
 typedef struct BDRVRawState {
     int fd;
     bool use_lock;
@@ -165,6 +183,7 @@ typedef struct BDRVRawState {
     bool use_linux_aio:1;
     bool has_laio_fdsync:1;
     bool use_linux_io_uring:1;
+    bool use_mpath:1;
     int page_cache_inconsistent; /* errno from fdatasync failure */
     bool has_fallocate;
     bool needs_alignment;
@@ -785,17 +804,6 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
     }
 #endif
 
-    if (S_ISBLK(st.st_mode)) {
-#ifdef __linux__
-        /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache.  Do
-         * not rely on the contents of discarded blocks unless using O_DIRECT.
-         * Same for BLKZEROOUT.
-         */
-        if (!(bs->open_flags & BDRV_O_NOCACHE)) {
-            s->has_write_zeroes = false;
-        }
-#endif
-    }
 #ifdef __FreeBSD__
     if (S_ISCHR(st.st_mode)) {
         /*
@@ -4264,15 +4272,105 @@ hdev_open_Mac_error:
     /* Since this does ioctl the device must be already opened */
     bs->sg = hdev_is_sg(bs);
 
+    /* sg devices aren't even block devices and can't use dm-mpath */
+    s->use_mpath = !bs->sg;
+
     return ret;
 }
 
 #if defined(__linux__)
+#if defined(DM_MPATH_PROBE_PATHS)
+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;
+        }
+    }
+
+    if (io_hdr->host_status != SCSI_HOST_OK) {
+        return true;
+    }
+
+    switch (io_hdr->status) {
+    case GOOD:
+    case CONDITION_GOOD:
+    case INTERMEDIATE_GOOD:
+    case INTERMEDIATE_C_GOOD:
+    case RESERVATION_CONFLICT:
+    case COMMAND_TERMINATED:
+        return false;
+    case CHECK_CONDITION:
+        return !scsi_sense_buf_is_guest_recoverable(io_hdr->sbp,
+                                                    io_hdr->mx_sb_len);
+    default:
+        return true;
+    }
+}
+
+static bool coroutine_fn hdev_co_ioctl_sgio_retry(RawPosixAIOData *acb, int ret)
+{
+    BDRVRawState *s = acb->bs->opaque;
+    RawPosixAIOData probe_acb;
+
+    if (!s->use_mpath) {
+        return false;
+    }
+
+    if (!sgio_path_error(ret, acb->ioctl.buf)) {
+        return false;
+    }
+
+    probe_acb = (RawPosixAIOData) {
+        .bs         = acb->bs,
+        .aio_type   = QEMU_AIO_IOCTL,
+        .aio_fildes = s->fd,
+        .aio_offset = 0,
+        .ioctl      = {
+            .buf        = NULL,
+            .cmd        = DM_MPATH_PROBE_PATHS,
+        },
+    };
+
+    ret = raw_thread_pool_submit(handle_aiocb_ioctl, &probe_acb);
+    if (ret == -ENOTTY) {
+        s->use_mpath = false;
+    } else if (ret == -EAGAIN) {
+        /* The device might be suspended for a table reload, worth retrying */
+        return true;
+    }
+
+    return ret == 0;
+}
+#else
+static bool coroutine_fn hdev_co_ioctl_sgio_retry(RawPosixAIOData *acb, int ret)
+{
+    return false;
+}
+#endif /* DM_MPATH_PROBE_PATHS */
+
 static int coroutine_fn
 hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 {
     BDRVRawState *s = bs->opaque;
     RawPosixAIOData acb;
+    int retries = SG_IO_MAX_RETRIES;
     int ret;
 
     ret = fd_open(bs);
@@ -4300,7 +4398,11 @@ hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
         },
     };
 
-    return raw_thread_pool_submit(handle_aiocb_ioctl, &acb);
+    do {
+        ret = raw_thread_pool_submit(handle_aiocb_ioctl, &acb);
+    } while (req == SG_IO && retries-- && hdev_co_ioctl_sgio_retry(&acb, ret));
+
+    return ret;
 }
 #endif /* linux */
 
diff --git a/hw/core/machine.c b/hw/core/machine.c
index b8ae155dfa..c3f3a5020d 100644
--- a/hw/core/machine.c
+++ b/hw/core/machine.c
@@ -37,7 +37,9 @@
 #include "hw/virtio/virtio-iommu.h"
 #include "audio/audio.h"
 
-GlobalProperty hw_compat_10_0[] = {};
+GlobalProperty hw_compat_10_0[] = {
+    { "scsi-hd", "dpofua", "off" },
+};
 const size_t hw_compat_10_0_len = G_N_ELEMENTS(hw_compat_10_0);
 
 GlobalProperty hw_compat_9_2[] = {
diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index cb4af1b715..b4782c6248 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -74,7 +74,7 @@ struct SCSIDiskClass {
      */
     DMAIOFunc       *dma_readv;
     DMAIOFunc       *dma_writev;
-    bool            (*need_fua_emulation)(SCSICommand *cmd);
+    bool            (*need_fua)(SCSICommand *cmd);
     void            (*update_sense)(SCSIRequest *r);
 };
 
@@ -85,7 +85,7 @@ typedef struct SCSIDiskReq {
     uint32_t sector_count;
     uint32_t buflen;
     bool started;
-    bool need_fua_emulation;
+    bool need_fua;
     struct iovec iov;
     QEMUIOVector qiov;
     BlockAcctCookie acct;
@@ -389,24 +389,6 @@ static bool scsi_is_cmd_fua(SCSICommand *cmd)
     }
 }
 
-static void scsi_write_do_fua(SCSIDiskReq *r)
-{
-    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
-
-    assert(r->req.aiocb == NULL);
-    assert(!r->req.io_canceled);
-
-    if (r->need_fua_emulation) {
-        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
-                         BLOCK_ACCT_FLUSH);
-        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
-        return;
-    }
-
-    scsi_req_complete(&r->req, GOOD);
-    scsi_req_unref(&r->req);
-}
-
 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
 {
     assert(r->req.aiocb == NULL);
@@ -416,12 +398,7 @@ static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
 
     r->sector += r->sector_count;
     r->sector_count = 0;
-    if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
-        scsi_write_do_fua(r);
-        return;
-    } else {
-        scsi_req_complete(&r->req, GOOD);
-    }
+    scsi_req_complete(&r->req, GOOD);
 
 done:
     scsi_req_unref(&r->req);
@@ -564,7 +541,7 @@ static void scsi_read_data(SCSIRequest *req)
 
     first = !r->started;
     r->started = true;
-    if (first && r->need_fua_emulation) {
+    if (first && r->need_fua) {
         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
                          BLOCK_ACCT_FLUSH);
         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
@@ -589,8 +566,7 @@ static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
     r->sector += n;
     r->sector_count -= n;
     if (r->sector_count == 0) {
-        scsi_write_do_fua(r);
-        return;
+        scsi_req_complete(&r->req, GOOD);
     } else {
         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
@@ -623,6 +599,7 @@ static void scsi_write_data(SCSIRequest *req)
     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
+    BlockCompletionFunc *cb;
 
     /* No data transfer may already be in progress */
     assert(r->req.aiocb == NULL);
@@ -648,11 +625,10 @@ static void scsi_write_data(SCSIRequest *req)
 
     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
         r->req.cmd.buf[0] == VERIFY_16) {
-        if (r->req.sg) {
-            scsi_dma_complete_noio(r, 0);
-        } else {
-            scsi_write_complete_noio(r, 0);
-        }
+        block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
+                         BLOCK_ACCT_FLUSH);
+        cb = r->req.sg ? scsi_dma_complete : scsi_write_complete;
+        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, cb, r);
         return;
     }
 
@@ -2391,7 +2367,7 @@ static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
         return 0;
     }
-    r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
+    r->need_fua = sdc->need_fua(&r->req.cmd);
     if (r->sector_count == 0) {
         scsi_req_complete(&r->req, GOOD);
     }
@@ -3137,7 +3113,8 @@ BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
 {
     SCSIDiskReq *r = opaque;
     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
-    return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
+    int flags = r->need_fua ? BDRV_REQ_FUA : 0;
+    return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, flags, cb, cb_opaque);
 }
 
 static char *scsi_property_get_loadparm(Object *obj, Error **errp)
@@ -3186,7 +3163,7 @@ static void scsi_disk_base_class_initfn(ObjectClass *klass, const void *data)
     device_class_set_legacy_reset(dc, scsi_disk_reset);
     sdc->dma_readv = scsi_dma_readv;
     sdc->dma_writev = scsi_dma_writev;
-    sdc->need_fua_emulation = scsi_is_cmd_fua;
+    sdc->need_fua  = scsi_is_cmd_fua;
 }
 
 static const TypeInfo scsi_disk_base_info = {
@@ -3215,7 +3192,7 @@ static const Property scsi_hd_properties[] = {
     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
                     SCSI_DISK_F_REMOVABLE, false),
     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
-                    SCSI_DISK_F_DPOFUA, false),
+                    SCSI_DISK_F_DPOFUA, true),
     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
@@ -3338,7 +3315,7 @@ static void scsi_block_class_initfn(ObjectClass *klass, const void *data)
     sdc->dma_readv   = scsi_block_dma_readv;
     sdc->dma_writev  = scsi_block_dma_writev;
     sdc->update_sense = scsi_block_update_sense;
-    sdc->need_fua_emulation = scsi_block_no_fua;
+    sdc->need_fua    = scsi_block_no_fua;
     dc->desc = "SCSI block device passthrough";
     device_class_set_props(dc, scsi_block_properties);
     dc->vmsd  = &vmstate_scsi_disk_state;
diff --git a/qemu-img.c b/qemu-img.c
index 76ac5d3028..139eeb5039 100644
--- a/qemu-img.c
+++ b/qemu-img.c
@@ -4488,10 +4488,10 @@ static void bench_cb(void *opaque, int ret)
          */
         b->in_flight++;
         b->offset += b->step;
-        if (b->image_size == 0) {
+        if (b->image_size <= b->bufsize) {
             b->offset = 0;
         } else {
-            b->offset %= b->image_size;
+            b->offset %= b->image_size - b->bufsize;
         }
         if (b->write) {
             acb = blk_aio_pwritev(b->blk, offset, b->qiov, 0, bench_cb, b);