Libguestfs - May 2020 - Re: [PATCH nbdkit 4/4] fua: Add unsafe fuamode=discard.

Two hopefully useful additions to the fua filter.  The second one is
kind of like cache=unsafe in qemu, in that it exchanges correctness
for speed.  Useful for data which is easily recreated in the event of
a crash or for people who like living on the edge and have good
backups.

Rich.

Break up the large paragraph and split up some longer sentences to
make it easier to read.
---
 filters/fua/nbdkit-fua-filter.pod | 51 ++++++++++++++++++++-----------
 1 file changed, 34 insertions(+), 17 deletions(-)

diff --git a/filters/fua/nbdkit-fua-filter.pod
b/filters/fua/nbdkit-fua-filter.pod
index 0411f9f4..6e8f87e2 100644
--- a/filters/fua/nbdkit-fua-filter.pod
+++ b/filters/fua/nbdkit-fua-filter.pod
@@ -1,6 +1,6 @@
 =head1 NAME
 
-nbdkit-fua-filter - nbdkit FUA filter
+nbdkit-fua-filter - modify nbdkit flush and Forced Unit Access (FUA)
 
 =head1 SYNOPSIS
 
@@ -9,47 +9,60 @@ nbdkit-fua-filter - nbdkit FUA filter
 =head1 DESCRIPTION
 
 C<nbdkit-fua-filter> is a filter that intentionally modifies handling
-of the 'Forced Unit Access' (FUA) flag across the NBD protocol.  It is
-mainly useful for testing client or server fallbacks, and for
+of the S<“Forced Unit Access”> (FUA) flag across the NBD protocol.  It
+is mainly useful for testing client or server fallbacks, and for
 evaluating timing differences between proper use of FUA compared to a
 full flush.
 
 =head1 PARAMETERS
 
+The C<fuamode> parameter is optional and controls which mode the
+filter will use.
+
 =over 4
 
 =item B<fuamode=none>
 
+FUA support is not advertised to the client.  This is the default if
+the C<fuamode> parameter is not specified.
+
 =item B<fuamode=emulate>
 
+The filter will emulate FUA support using the plugin’s C<.flush>
+callback, regardless of whether the plugin itself supports more
+efficient FUA.  It refuses to load if the plugin does not support
+flush.
+
 =item B<fuamode=native>
 
+The filter will advertise native FUA support to the client and earlier
+filters in the chain.  This is useful for comparing optimizations of
+FUA handling when splitting large requests into sub-requests.  It
+refuses to load if the plugin’s C<.can_fua> callback returns
+C<NBDKIT_FUA_NONE>.
+
 =item B<fuamode=force>
 
-Optional, controls which mode the filter will use.  Mode B<none>
-(default) means that FUA support is not advertised to the client.
-Mode B<emulate> causes the filter to emulate FUA support using the
-plugin's C<flush> callback, regardless of whether the plugin itself
-supports more efficient FUA (or refuses to load if the plugin's
-C<can_flush> callback returns false).  Mode B<native> causes the
-filter to advertise native FUA support to earlier filters, useful for
-comparing optimizations of FUA handling when splitting large requests
-into sub-requests (or refuses to load if the plugin's C<can_fua>
-callback returns C<NBDKIT_FUA_NONE>).  Mode B<force> causes the
filter
-to request FUA on all write transactions, even when the client did not
-request it, and in turn treats client flush requests as a no-op (or
-refuses to load if the plugin's C<can_fua> callback returns
-C<NBDKIT_FUA_NONE>).
+The filter will request FUA on all write transactions, even when the
+client did not request it.  In turn client flush requests become
+no-ops.  It refuses to load if the plugin's C<.can_fua> callback
+returns C<NBDKIT_FUA_NONE>.
 
 =back
 
 =head1 EXAMPLES
 
+=over 4
+
+=item *
+
 Serve the file F<disk.img>, but force the client to submit explicit
 flush requests instead of using C<NBD_CMD_FLAG_FUA>:
 
  nbdkit --filter=fua file disk.img
 
+=item *
+
 Observe that the blocksize filter optimizes its handling of the FUA
 flag based on whether it knows nbdkit will be emulating FUA with a
 flush, by comparing the log filter output on top of different fua
@@ -60,12 +73,16 @@ filter modes:
  nbdkit --filter=blocksize --filter=log --filter=fua file disk.img \
    maxlen=4k logfile=fua_native fuamode=native
 
+=item *
+
 Serve the file F<disk.img> in write-through mode, where all writes
 from the client are immediately flushed to disk as if the client had
 always requested FUA:
 
  nbdkit --filter=fua file fuamode=force disk.img
 
+=back
+
 =head1 FILES
 
 =over 4
-- 
2.25.0

Passes FUA and flush requests unchanged.  A convenient way to turn the
filter into a no-op without rewriting the whole nbdkit command line.
---
 filters/fua/nbdkit-fua-filter.pod |  5 +++++
 filters/fua/fua.c                 | 12 +++++++++++-
 tests/test-fua.sh                 | 21 ++++++++++++++++++---
 3 files changed, 34 insertions(+), 4 deletions(-)

diff --git a/filters/fua/nbdkit-fua-filter.pod
b/filters/fua/nbdkit-fua-filter.pod
index 6e8f87e2..3d20b56a 100644
--- a/filters/fua/nbdkit-fua-filter.pod
+++ b/filters/fua/nbdkit-fua-filter.pod
@@ -48,6 +48,11 @@ client did not request it.  In turn client flush requests
become
 no-ops.  It refuses to load if the plugin's C<.can_fua> callback
 returns C<NBDKIT_FUA_NONE>.
 
+=item B<fuamode=pass>
+
+Pass through FUA and flush requests unchanged.  Turns the filter into
+a no-op.
+
 =back
 
 =head1 EXAMPLES
diff --git a/filters/fua/fua.c b/filters/fua/fua.c
index cee05eed..6bc62a02 100644
--- a/filters/fua/fua.c
+++ b/filters/fua/fua.c
@@ -46,6 +46,7 @@ static enum FuaMode {
   EMULATE,
   NATIVE,
   FORCE,
+  PASS,
 } fuamode;
 
 static int
@@ -61,6 +62,8 @@ fua_config (nbdkit_next_config *next, void *nxdata,
       fuamode = NATIVE;
     else if (strcmp (value, "force") == 0)
       fuamode = FORCE;
+    else if (strcmp (value, "pass") == 0)
+      fuamode = PASS;
     else {
       nbdkit_error ("unknown fuamode '%s'", value);
       return -1;
@@ -71,7 +74,8 @@ fua_config (nbdkit_next_config *next, void *nxdata,
 }
 
 #define fua_config_help \
-  "fuamode=<MODE>       One of 'none' (default),
'emulate', 'native', 'force'.\n" \
+  "fuamode=<MODE>       One of 'none' (default),
'emulate', 'native',\n" \
+  "                       'force', 'pass'."
 
 /* Check that desired mode is supported by plugin. */
 static int
@@ -86,6 +90,7 @@ fua_prepare (struct nbdkit_next_ops *next_ops, void *nxdata,
void *handle,
 
   switch (fuamode) {
   case NONE:
+  case PASS:
     break;
   case EMULATE:
     r = next_ops->can_flush (nxdata);
@@ -132,6 +137,8 @@ fua_can_fua (struct nbdkit_next_ops *next_ops, void *nxdata,
void *handle)
   case NATIVE:
   case FORCE:
     return NBDKIT_FUA_NATIVE;
+  case PASS:
+    return next_ops->can_fua (nxdata);
   }
   abort ();
 }
@@ -155,6 +162,7 @@ fua_pwrite (struct nbdkit_next_ops *next_ops, void *nxdata,
     }
     break;
   case NATIVE:
+  case PASS:
     break;
   case FORCE:
     flags |= NBDKIT_FLAG_FUA;
@@ -194,6 +202,7 @@ fua_trim (struct nbdkit_next_ops *next_ops, void *nxdata,
     }
     break;
   case NATIVE:
+  case PASS:
     break;
   case FORCE:
     flags |= NBDKIT_FLAG_FUA;
@@ -224,6 +233,7 @@ fua_zero (struct nbdkit_next_ops *next_ops, void *nxdata,
     }
     break;
   case NATIVE:
+  case PASS:
     break;
   case FORCE:
     flags |= NBDKIT_FLAG_FUA;
diff --git a/tests/test-fua.sh b/tests/test-fua.sh
index 1c869e96..5fb5e929 100755
--- a/tests/test-fua.sh
+++ b/tests/test-fua.sh
@@ -39,7 +39,8 @@ files="fua.img
        fua1.log fua1.pid
        fua2.log fua2.pid
        fua3.log fua3.pid
-       fua4.log fua4.pid"
+       fua4.log fua4.pid
+       fua5.log fua5.pid"
 rm -f $files
 
 # Prep images, and check that qemu-io understands the actions we plan on
@@ -63,16 +64,19 @@ cleanup ()
     cat fua3.log || :
     echo "Log 4 file contents:"
     cat fua4.log || :
+    echo "Log 5 file contents:"
+    cat fua5.log || :
     rm -f $files
     rm -rf $sockdir
 }
 cleanup_fn cleanup
 
-# Run four parallel nbdkit; to compare the logs and see what changes.
+# Run parallel nbdkit; to compare the logs and see what changes.
 # 1: fuamode=none (default): client should send flush instead
 # 2: fuamode=emulate: log shows that blocksize optimizes fua to flush
 # 3: fuamode=native: log shows that blocksize preserves fua
 # 4: fuamode=force: log shows that fua is always enabled
+# 5: fuamode=pass: fua flag and flush unchanged
 start_nbdkit -P fua1.pid -U $sockdir/fua1.sock \
              --filter=log --filter=fua \
              file logfile=fua1.log fua.img
@@ -85,10 +89,13 @@ start_nbdkit -P fua3.pid -U $sockdir/fua3.sock \
 start_nbdkit -P fua4.pid -U $sockdir/fua4.sock \
              --filter=fua --filter=log \
              file logfile=fua4.log fua.img fuamode=force
+start_nbdkit -P fua5.pid -U $sockdir/fua5.sock \
+             --filter=fua --filter=log \
+             file logfile=fua5.log fua.img fuamode=pass
 
 # Perform a flush, write, and zero, first without then with FUA
 for f in '' -f; do
-    for i in {1..4}; do
+    for i in {1..5}; do
 	qemu-io -f raw -t none -c flush -c "w $f 0 64k" -c "w -z $f 64k
64k" \
 		 "nbd+unix://?socket=$sockdir/fua$i.sock"
     done
@@ -124,3 +131,11 @@ if grep 'Flush' fua4.log; then
     echo "filter should have elided flush"
     exit 1
 fi
+
+# Test 5: Flush should be passed through.
+# There should also be one set of fua=0 and a second set of fua=1.
+grep 'Flush' fua5.log
+grep 'connection=1 Write.*fua=0' fua5.log
+grep 'connection=2 Write.*fua=1' fua5.log
+grep 'connection=1 Zero.*fua=0' fua5.log
+grep 'connection=2 Zero.*fua=1' fua5.log
-- 
2.25.0

This causes conflicts if a cleanup hook happens to use a variable with
the same name.
---
 tests/functions.sh.in | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/tests/functions.sh.in b/tests/functions.sh.in
index e2ef9701..b5c7d88b 100644
--- a/tests/functions.sh.in
+++ b/tests/functions.sh.in
@@ -43,17 +43,17 @@ cleanup_fn ()
 
 _run_cleanup_hooks ()
 {
-    local status=$? i
+    local _status=$? _i
 
     set +e
     trap '' INT QUIT TERM EXIT ERR
-    echo $0: run cleanup hooks: exit code $status
+    echo $0: run cleanup hooks: exit code $_status
 
-    for (( i = 0; i < ${#_cleanup_hook[@]}; ++i )); do
-        ${_cleanup_hook[i]}
+    for (( _i = 0; _i < ${#_cleanup_hook[@]}; ++_i )); do
+        ${_cleanup_hook[_i]}
     done
 
-    exit $status
+    exit $_status
 }
 trap _run_cleanup_hooks INT QUIT TERM EXIT ERR
 
-- 
2.25.0

This drops all FUA and flush requests.
---
 filters/fua/nbdkit-fua-filter.pod | 27 +++++++++++++++++----
 filters/fua/fua.c                 | 39 +++++++++++++++++++++++++++----
 tests/test-fua.sh                 | 33 ++++++++++++++++----------
 3 files changed, 79 insertions(+), 20 deletions(-)

diff --git a/filters/fua/nbdkit-fua-filter.pod
b/filters/fua/nbdkit-fua-filter.pod
index 3d20b56a..4f564fab 100644
--- a/filters/fua/nbdkit-fua-filter.pod
+++ b/filters/fua/nbdkit-fua-filter.pod
@@ -9,10 +9,12 @@ nbdkit-fua-filter - modify nbdkit flush and Forced Unit Access
(FUA)
 =head1 DESCRIPTION
 
 C<nbdkit-fua-filter> is a filter that intentionally modifies handling
-of the S<“Forced Unit Access”> (FUA) flag across the NBD protocol.  It
-is mainly useful for testing client or server fallbacks, and for
-evaluating timing differences between proper use of FUA compared to a
-full flush.
+of the S<“Forced Unit Access”> (FUA) flag across the NBD protocol.
+
+This filter can be used to disable FUA and flush requests for speed
+(although this is unsafe).  Also it can be used to test client or
+server fallbacks, and for evaluating timing differences between proper
+use of FUA compared to a full flush.
 
 =head1 PARAMETERS
 
@@ -53,6 +55,15 @@ returns C<NBDKIT_FUA_NONE>.
 Pass through FUA and flush requests unchanged.  Turns the filter into
 a no-op.
 
+=item B<fuamode=discard>
+
+The filter will discard FUA and flush requests.
+
+B<This mode is unsafe>: If the NBD disk contains a filesystem then you
+will likely lose data in the event of a crash.  It should only be used
+for ephemeral data which you can easily recreate, such as caches,
+builds, test data, etc.
+
 =back
 
 =head1 EXAMPLES
@@ -86,6 +97,14 @@ always requested FUA:
 
  nbdkit --filter=fua file fuamode=force disk.img
 
+=item *
+
+Serve the file F<disk.img> discarding all FUA and flush requests.
+This can greatly improve performance, but you will likely lose data if
+there is a crash, so it is not safe.
+
+ nbdkit --filter=discard file fuamode=force disk.img
+
 =back
 
 =head1 FILES
diff --git a/filters/fua/fua.c b/filters/fua/fua.c
index 6bc62a02..229d83db 100644
--- a/filters/fua/fua.c
+++ b/filters/fua/fua.c
@@ -47,6 +47,7 @@ static enum FuaMode {
   NATIVE,
   FORCE,
   PASS,
+  DISCARD,
 } fuamode;
 
 static int
@@ -64,6 +65,8 @@ fua_config (nbdkit_next_config *next, void *nxdata,
       fuamode = FORCE;
     else if (strcmp (value, "pass") == 0)
       fuamode = PASS;
+    else if (strcmp (value, "discard") == 0)
+      fuamode = DISCARD;
     else {
       nbdkit_error ("unknown fuamode '%s'", value);
       return -1;
@@ -91,6 +94,7 @@ fua_prepare (struct nbdkit_next_ops *next_ops, void *nxdata,
void *handle,
   switch (fuamode) {
   case NONE:
   case PASS:
+  case DISCARD:
     break;
   case EMULATE:
     r = next_ops->can_flush (nxdata);
@@ -120,9 +124,17 @@ fua_prepare (struct nbdkit_next_ops *next_ops, void
*nxdata, void *handle,
 static int
 fua_can_flush (struct nbdkit_next_ops *next_ops, void *nxdata, void *handle)
 {
-  if (fuamode == FORCE)
+  switch (fuamode) {
+  case FORCE:
+  case DISCARD:
     return 1; /* Advertise our no-op flush, even if plugin lacks it */
-  return next_ops->can_flush (nxdata);
+  case NONE:
+  case EMULATE:
+  case NATIVE:
+  case PASS:
+    return next_ops->can_flush (nxdata);
+  }
+  abort ();
 }
 
 /* Advertise desired fua mode. */
@@ -136,6 +148,7 @@ fua_can_fua (struct nbdkit_next_ops *next_ops, void *nxdata,
void *handle)
     return NBDKIT_FUA_EMULATE;
   case NATIVE:
   case FORCE:
+  case DISCARD:
     return NBDKIT_FUA_NATIVE;
   case PASS:
     return next_ops->can_fua (nxdata);
@@ -167,6 +180,9 @@ fua_pwrite (struct nbdkit_next_ops *next_ops, void *nxdata,
   case FORCE:
     flags |= NBDKIT_FLAG_FUA;
     break;
+  case DISCARD:
+    flags &= ~NBDKIT_FLAG_FUA;
+    break;
   }
   r = next_ops->pwrite (nxdata, buf, count, offs, flags, err);
   if (r != -1 && need_flush)
@@ -178,9 +194,18 @@ static int
 fua_flush (struct nbdkit_next_ops *next_ops, void *nxdata,
            void *handle, uint32_t flags, int *err)
 {
-  if (fuamode == FORCE)
+  switch (fuamode) {
+  case FORCE:
     return 0; /* Nothing to flush, since all writes already used FUA */
-  return next_ops->flush (nxdata, flags, err);
+  case DISCARD:
+    return 0; /* Drop flushes! */
+  case NONE:
+  case EMULATE:
+  case NATIVE:
+  case PASS:
+    return next_ops->flush (nxdata, flags, err);
+  }
+  abort ();
 }
 
 static int
@@ -207,6 +232,9 @@ fua_trim (struct nbdkit_next_ops *next_ops, void *nxdata,
   case FORCE:
     flags |= NBDKIT_FLAG_FUA;
     break;
+  case DISCARD:
+    flags &= ~NBDKIT_FLAG_FUA;
+    break;
   }
   r = next_ops->trim (nxdata, count, offs, flags, err);
   if (r != -1 && need_flush)
@@ -238,6 +266,9 @@ fua_zero (struct nbdkit_next_ops *next_ops, void *nxdata,
   case FORCE:
     flags |= NBDKIT_FLAG_FUA;
     break;
+  case DISCARD:
+    flags &= ~NBDKIT_FLAG_FUA;
+    break;
   }
   r = next_ops->zero (nxdata, count, offs, flags, err);
   if (r != -1 && need_flush)
diff --git a/tests/test-fua.sh b/tests/test-fua.sh
index 5fb5e929..bf64e8da 100755
--- a/tests/test-fua.sh
+++ b/tests/test-fua.sh
@@ -40,7 +40,8 @@ files="fua.img
        fua2.log fua2.pid
        fua3.log fua3.pid
        fua4.log fua4.pid
-       fua5.log fua5.pid"
+       fua5.log fua5.pid
+       fua6.log fua6.pid"
 rm -f $files
 
 # Prep images, and check that qemu-io understands the actions we plan on
@@ -56,16 +57,10 @@ fi
 # on exit.
 cleanup ()
 {
-    echo "Log 1 file contents:"
-    cat fua1.log || :
-    echo "Log 2 file contents:"
-    cat fua2.log || :
-    echo "Log 3 file contents:"
-    cat fua3.log || :
-    echo "Log 4 file contents:"
-    cat fua4.log || :
-    echo "Log 5 file contents:"
-    cat fua5.log || :
+    for i in {1..6}; do
+        echo "Log $i file contents:"
+        cat fua$i.log || :
+    done
     rm -f $files
     rm -rf $sockdir
 }
@@ -77,6 +72,7 @@ cleanup_fn cleanup
 # 3: fuamode=native: log shows that blocksize preserves fua
 # 4: fuamode=force: log shows that fua is always enabled
 # 5: fuamode=pass: fua flag and flush unchanged
+# 6: fuamode=discard: discard all fua and flush
 start_nbdkit -P fua1.pid -U $sockdir/fua1.sock \
              --filter=log --filter=fua \
              file logfile=fua1.log fua.img
@@ -92,10 +88,13 @@ start_nbdkit -P fua4.pid -U $sockdir/fua4.sock \
 start_nbdkit -P fua5.pid -U $sockdir/fua5.sock \
              --filter=fua --filter=log \
              file logfile=fua5.log fua.img fuamode=pass
+start_nbdkit -P fua6.pid -U $sockdir/fua6.sock \
+             --filter=fua --filter=log \
+             file logfile=fua6.log fua.img fuamode=discard
 
 # Perform a flush, write, and zero, first without then with FUA
 for f in '' -f; do
-    for i in {1..5}; do
+    for i in {1..6}; do
 	qemu-io -f raw -t none -c flush -c "w $f 0 64k" -c "w -z $f 64k
64k" \
 		 "nbd+unix://?socket=$sockdir/fua$i.sock"
     done
@@ -139,3 +138,13 @@ grep 'connection=1 Write.*fua=0' fua5.log
 grep 'connection=2 Write.*fua=1' fua5.log
 grep 'connection=1 Zero.*fua=0' fua5.log
 grep 'connection=2 Zero.*fua=1' fua5.log
+
+# Test 6: Flush and fua=1 must not appear.
+if grep 'Flush' fua6.log; then
+    echo "filter should have elided flush"
+    exit 1
+fi
+if grep -E '(Write|Zero).*fua=1' fua6.log; then
+    echo "filter should have elided fua"
+    exit 1
+fi
-- 
2.25.0

On 5/22/20 4:32 PM, Richard W.M. Jones wrote:
> This causes conflicts if a cleanup hook happens to use a variable with
> the same name.
I don't envy the debugging you used to finally spot this.  Using
'local'
in the cleanup hook helps, but so does this rename, so I like this approach.
> ---
>   tests/functions.sh.in | 10 +++++-----
>   1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/tests/functions.sh.in b/tests/functions.sh.in
> index e2ef9701..b5c7d88b 100644
> --- a/tests/functions.sh.in
> +++ b/tests/functions.sh.in
> @@ -43,17 +43,17 @@ cleanup_fn ()
>   
>   _run_cleanup_hooks ()
>   {
> -    local status=$? i
> +    local _status=$? _i
>   
>       set +e
>       trap '' INT QUIT TERM EXIT ERR
> -    echo $0: run cleanup hooks: exit code $status
> +    echo $0: run cleanup hooks: exit code $_status
>   
> -    for (( i = 0; i < ${#_cleanup_hook[@]}; ++i )); do
> -        ${_cleanup_hook[i]}
> +    for (( _i = 0; _i < ${#_cleanup_hook[@]}; ++_i )); do
> +        ${_cleanup_hook[_i]}
>       done
>   
> -    exit $status
> +    exit $_status
>   }
>   trap _run_cleanup_hooks INT QUIT TERM EXIT ERR
>   
> 
-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org

On 5/22/20 4:32 PM, Richard W.M. Jones wrote:
> This drops all FUA and flush requests.
> ---
>   filters/fua/nbdkit-fua-filter.pod | 27 +++++++++++++++++----
>   filters/fua/fua.c                 | 39 +++++++++++++++++++++++++++----
>   tests/test-fua.sh                 | 33 ++++++++++++++++----------
>   3 files changed, 79 insertions(+), 20 deletions(-)
Overall, the series looks good to me.  I don't know if there are any 
other man pages where we should call attention to the fact about the new 
mode.
> 
> diff --git a/filters/fua/nbdkit-fua-filter.pod
b/filters/fua/nbdkit-fua-filter.pod
> index 3d20b56a..4f564fab 100644
> --- a/filters/fua/nbdkit-fua-filter.pod
> +++ b/filters/fua/nbdkit-fua-filter.pod
> @@ -9,10 +9,12 @@ nbdkit-fua-filter - modify nbdkit flush and Forced Unit
Access (FUA)
>   =head1 DESCRIPTION
>   
>   C<nbdkit-fua-filter> is a filter that intentionally modifies
handling
> -of the S<“Forced Unit Access”> (FUA) flag across the NBD protocol. 
It
> -is mainly useful for testing client or server fallbacks, and for
> -evaluating timing differences between proper use of FUA compared to a
> -full flush.
> +of the S<“Forced Unit Access”> (FUA) flag across the NBD protocol.
> +
> +This filter can be used to disable FUA and flush requests for speed
> +(although this is unsafe).  Also it can be used to test client or
> +server fallbacks, and for evaluating timing differences between proper
> +use of FUA compared to a full flush.
>   
>   =head1 PARAMETERS
>   
> @@ -53,6 +55,15 @@ returns C<NBDKIT_FUA_NONE>.
>   Pass through FUA and flush requests unchanged.  Turns the filter into
>   a no-op.
>   
> +=item B<fuamode=discard>
> +
> +The filter will discard FUA and flush requests.
> +
> +B<This mode is unsafe>: If the NBD disk contains a filesystem then
you
> +will likely lose data in the event of a crash.  It should only be used
> +for ephemeral data which you can easily recreate, such as caches,
> +builds, test data, etc.
Also, we may want to mention when this mode was introduced (since if you 
don't have 1.22 installed, you can't use it yet).

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org