Skip to contentSkip to conversation

Remove backend xen store entry on domain destroy

PATCH In review Oleksandr Grytsov 5 participants 15 messages Nov 21, 2019 – Nov 25, 2019

lore
Share & tools
lore
#0 lore raw

v2 cover Remove backend xen store entry on domain destroy

From: Oleksandr Grytsov <[email protected]>

Changes since v1:

* add commit to rename VKB backend type "linux" to "pv";
* add default case to handle adding/removing PV devices in add_device, remove_device functions (libxl_device.c);
* add comment about removing num_vifs, num_vbds into commit message.

Oleksandr Grytsov (3):
libxl: introduce new backend type VINPUT
libxl: rename VKB backend type "linux" to "pv"
libxl: make default path to add/remove all PV devices

 docs/man/xl.cfg.5.pod.in             |  4 +-
 tools/libxl/libxl_device.c           | 63 +++++++++++-----------------
 tools/libxl/libxl_types.idl          |  2 +-
 tools/libxl/libxl_types_internal.idl |  1 +
 tools/libxl/libxl_vkb.c              | 29 ++++++++-----
 5 files changed, 47 insertions(+), 52 deletions(-)
Signature · 7 lines
2.17.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#1 lore raw

PATCH v2 1/3 libxl: introduce new backend type VINPUT

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKBD devices: with QEMU backend and user space PV backend. In current implementation they can't be distinguished as both use VKBD backend type. As result, user space PV KBD backend is started and stopped as QEMU backend. This commit adds new device kind VINPUT to be used as backend type for user space PV KBD backend.

---
 tools/libxl/libxl_types_internal.idl |  1 +
 tools/libxl/libxl_vkb.c              | 29 ++++++++++++++++++----------
 2 files changed, 20 insertions(+), 10 deletions(-)
tools/libxl/libxl_types_internal.idl +1 -0
index cb85c3b37f..3593e21dbb 100644
@@ -31,6 +31,7 @@ libxl__device_kind = Enumeration("device_kind", [
     (13, "VUART"),
     (14, "PVCALLS"),
     (15, "VSND"),
+    (16, "VINPUT"),
     ])
 
 libxl__console_backend = Enumeration("console_backend", [
tools/libxl/libxl_vkb.c +19 -10
index 26376a7eef..4c44a813c1 100644
@@ -38,9 +38,6 @@ static int libxl__set_xenstore_vkb(libxl__gc *gc, uint32_t domid,
                                    flexarray_t *back, flexarray_t *front,
                                    flexarray_t *ro_front)
 {
-    flexarray_append_pair(back, "backend-type",
-                          (char *)libxl_vkb_backend_to_string(vkb->backend_type));
-
     if (vkb->unique_id) {
         flexarray_append_pair(back, XENKBD_FIELD_UNIQUE_ID, vkb->unique_id);
     }
@@ -93,7 +90,8 @@ static int libxl__vkb_from_xenstore(libxl__gc *gc, const char *libxl_path,
                                     libxl_devid devid,
                                     libxl_device_vkb *vkb)
 {
-    const char *be_path, *be_type, *fe_path, *tmp;
+    const char *be_path, *fe_path, *tmp;
+    libxl__device dev;
     int rc;
 
     vkb->devid = devid;
@@ -111,13 +109,11 @@ static int libxl__vkb_from_xenstore(libxl__gc *gc, const char *libxl_path,
     rc = libxl__backendpath_parse_domid(gc, be_path, &vkb->backend_domid);
     if (rc) goto out;
 
-    rc = libxl__xs_read_mandatory(gc, XBT_NULL,
-                                  GCSPRINTF("%s/backend-type", be_path),
-                                  &be_type);
+    rc = libxl__parse_backend_path(gc, be_path, &dev);
     if (rc) goto out;
 
-    rc = libxl_vkb_backend_from_string(be_type, &vkb->backend_type);
-    if (rc) goto out;
+    vkb->backend_type = dev.backend_kind == LIBXL__DEVICE_KIND_VINPUT ?
+                                            LIBXL_VKB_BACKEND_LINUX : LIBXL_VKB_BACKEND_QEMU;
 
     vkb->unique_id = xs_read(CTX->xsh, XBT_NULL, GCSPRINTF("%s/"XENKBD_FIELD_UNIQUE_ID, be_path), NULL);
 
@@ -218,6 +214,20 @@ out:
     return rc;
 }
 
+static int libxl__device_from_vkb(libxl__gc *gc, uint32_t domid,
+                                  libxl_device_vkb *type, libxl__device *device)
+{
+    device->backend_devid   = type->devid;
+    device->backend_domid   = type->backend_domid;
+    device->backend_kind    = type->backend_type == LIBXL_VKB_BACKEND_LINUX ?
+                              LIBXL__DEVICE_KIND_VINPUT : LIBXL__DEVICE_KIND_VKBD;
+    device->devid           = type->devid;
+    device->domid           = domid;
+    device->kind            = LIBXL__DEVICE_KIND_VKBD;
+
+    return 0;
+}
+
 int libxl_device_vkb_add(libxl_ctx *ctx, uint32_t domid, libxl_device_vkb *vkb,
                          const libxl_asyncop_how *ao_how)
 {
@@ -318,7 +328,6 @@ out:
      return rc;
 }
 
-static LIBXL_DEFINE_DEVICE_FROM_TYPE(vkb)
 static LIBXL_DEFINE_UPDATE_DEVID(vkb)
 
 #define libxl__add_vkbs NULL
Signature · 7 lines
2.17.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 1/3
#2 lore raw

Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT

Quoted 10 lines · Oleksandr Grytsov

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKBD devices: with QEMU backend and user space PV backend. In current implementation they can't be distinguished as both use VKBD backend type. As result, user space PV KBD backend is started and stopped as QEMU backend. This commit adds new device kind VINPUT to be used as backend type for user space PV KBD backend.

That patch removes the "backend-type" node from xenstore but it wasn't part of the api (kbdif.h) and qemu doesn't read, it so I guess that's fine:

Thanks,

Signature · 6 lines
Anthony PERARD

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#3 lore raw

Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT

Quoted 9 lines · Anthony PERARD
Quoted 8 lines · Oleksandr Grytsov

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKBD devices: with QEMU backend and user space PV backend. In current implementation they can't be distinguished as both use VKBD backend type. As result, user space PV KBD backend is started and stopped as QEMU backend. This commit adds new device kind VINPUT to be used as backend type for user space PV KBD backend.

Quoted 8 lines
Quoted 2 lines

That patch removes the "backend-type" node from xenstore but it wasn't part of the api (kbdif.h) and qemu doesn't read, it so I guess that's fine:

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#4 lore raw

Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT

Quoted 13 lines · Jürgen Groß
Quoted 12 lines · Anthony PERARD
Quoted 10 lines · Oleksandr Grytsov

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKBD devices: with QEMU backend and user space PV backend. In current implementation they can't be distinguished as both use VKBD backend type. As result, user space PV KBD backend is started and stopped as QEMU backend. This commit adds new device kind VINPUT to be used as backend type for user space PV KBD backend.

Quoted 7 lines
Quoted 5 lines

That patch removes the "backend-type" node from xenstore but it wasn't part of the api (kbdif.h) and qemu doesn't read, it so I guess that's fine:

I take it this applies to both patch 1 and 3?

Wei.

Quoted 3 lines

Juergen

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#5 lore raw

Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT

Wei Liu writes ("Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT"):

Quoted 4 lines · Jürgen Groß
Quoted 1 line

I take it this applies to both patch 1 and 3?

In the absence of a reply to the contrary by 21:00 UTC today, I will assume this to be the case and push this to staging. I hope that meets with everyone's approval.

Ian.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#6 lore raw

Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT

Quoted 9 lines · Ian Jackson

Wei Liu writes ("Re: [Xen-devel] [PATCH v2 1/3] libxl: introduce new backend type VINPUT"):

Quoted 4 lines · Jürgen Groß
Quoted 1 line

I take it this applies to both patch 1 and 3?

In the absence of a reply to the contrary by 21:00 UTC today, I will assume this to be the case and push this to staging. I hope that meets with everyone's approval.

Got confirmation from Juergen on IRC, so I have just pushed patch 1 and 3.

Wei.

Quoted 2 lines

Ian.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#7 lore raw

PATCH v2 2/3 libxl: rename VKB backend type "linux" to "pv"

In reply to Oleksandr Grytsov

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKB backends: QEMU and user space PV backend. For PV backend "linux" enum is used but this name is confused. Rename "linux" enum to "pv" as it better matches user space PV backend.

---
 docs/man/xl.cfg.5.pod.in    | 4 ++--
 tools/libxl/libxl_types.idl | 2 +-
 tools/libxl/libxl_vkb.c     | 4 ++--
 3 files changed, 5 insertions(+), 5 deletions(-)
docs/man/xl.cfg.5.pod.in +2 -2
index 245d3f9472..37bdc595d3 100644
@@ -1562,8 +1562,8 @@ Specifies the backend domain name or id.
 
 =item B<backend-type=type>
 
-Specifies the backend type: qemu - for QEMU backend or linux - for Linux PV
-domain.
+Specifies the backend type: qemu - for QEMU backend or pv - for PV
+backend.
 
 =item B<feature-disable-keyboard=BOOLEAN>
 
tools/libxl/libxl_types.idl +1 -1
index 0546d7865a..751dad8870 100644
@@ -260,7 +260,7 @@ libxl_vuart_type = Enumeration("vuart_type", [
 libxl_vkb_backend = Enumeration("vkb_backend", [
     (0, "UNKNOWN"),
     (1, "QEMU"),
-    (2, "LINUX")
+    (2, "PV")
     ])
 
 libxl_passthrough = Enumeration("passthrough", [
tools/libxl/libxl_vkb.c +2 -2
index 4c44a813c1..5f018f55b7 100644
@@ -113,7 +113,7 @@ static int libxl__vkb_from_xenstore(libxl__gc *gc, const char *libxl_path,
     if (rc) goto out;
 
     vkb->backend_type = dev.backend_kind == LIBXL__DEVICE_KIND_VINPUT ?
-                                            LIBXL_VKB_BACKEND_LINUX : LIBXL_VKB_BACKEND_QEMU;
+                                            LIBXL_VKB_BACKEND_PV : LIBXL_VKB_BACKEND_QEMU;
 
     vkb->unique_id = xs_read(CTX->xsh, XBT_NULL, GCSPRINTF("%s/"XENKBD_FIELD_UNIQUE_ID, be_path), NULL);
 
@@ -219,7 +219,7 @@ static int libxl__device_from_vkb(libxl__gc *gc, uint32_t domid,
 {
     device->backend_devid   = type->devid;
     device->backend_domid   = type->backend_domid;
-    device->backend_kind    = type->backend_type == LIBXL_VKB_BACKEND_LINUX ?
+    device->backend_kind    = type->backend_type == LIBXL_VKB_BACKEND_PV ?
                               LIBXL__DEVICE_KIND_VINPUT : LIBXL__DEVICE_KIND_VKBD;
     device->devid           = type->devid;
     device->domid           = domid;
Signature · 7 lines
2.17.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 2/3
#8 lore raw

Re: [Xen-devel] [PATCH v2 2/3] libxl: rename VKB backend type "linux" to "pv"

Oleksandr Grytsov writes ("[PATCH v2 2/3] libxl: rename VKB backend type "linux" to "pv""):

Quoted 7 lines

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKB backends: QEMU and user space PV backend. For PV backend "linux" enum is used but this name is confused. Rename "linux" enum to "pv" as it better matches user space PV backend.

Thanks.

Ian.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#9 lore raw

Re: [Xen-devel] [PATCH v2 2/3] libxl: rename VKB backend type "linux" to "pv"

Quoted 11 lines · Ian Jackson

Oleksandr Grytsov writes ("[PATCH v2 2/3] libxl: rename VKB backend type "linux" to "pv""):

Quoted 7 lines

From: Oleksandr Grytsov <[email protected]>

There are two kind of VKB backends: QEMU and user space PV backend. For PV backend "linux" enum is used but this name is confused. Rename "linux" enum to "pv" as it better matches user space PV backend.

The LINUX type was introduced back in 2018.

This patch should have provided at least a

  #define ...LINUX ...PV

in libxl.h for backward compatibility purpose.

Wei.

Quoted 4 lines

Thanks.

Ian.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#10 lore raw

Re: [Xen-devel] [PATCH v2 2/3] libxl: rename VKB backend type "linux" to "pv"

Wei Liu writes ("Re: [PATCH v2 2/3] libxl: rename VKB backend type "linux" to "pv""):

Quoted 1 line

The LINUX type was introduced back in 2018.

Oh. Yes. I used a wrong git-describe rune to try to find out where it had been introduced. The answer in fact is that it was in 4.12.0-rc1. Thanks for pointing this out.

Thanks also for the 2nd pair of eyes on
[PATCH v2 3/3] libxl: make default path to add/remove all PV devices

Quoted 3 lines

This patch should have provided at least a
#define ...LINUX ...PV
in libxl.h for backward compatibility purpose.

Under the circumstances, given that I suggested the rename during the 4.13 freeze, and the (IMO wrong) name was shipped in 4.12, it seems that we should drop this patch for 4.13. This is something that perhaps we should think about for the future but if it's been released already the cost of messing about with the name is sharply larger, and it may not be worth it at all. Sorry for messing you about more, Oleksandr.

The other two patches 1/ and 3/ apply and work just fine without this middle one, so I don't think we need a resend.

I'm still hoping for a 2nd pair of eyes on
[PATCH v2 1/3] libxl: introduce new backend type VINPUT

and a release ack for both 1/3 and 3/3.

Ian.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#11 lore raw

PATCH v2 3/3 libxl: make default path to add/remove all PV devices

In reply to Oleksandr Grytsov

From: Oleksandr Grytsov <[email protected]>

Adding/removing device is handled for specific devices only: VBD, VIF, QDISK. This commit adds default case to handle adding/removing for all PV devices by default, except QDISK device, which requires special handling. If any other device is required a special handling it should be done by implementing separate case (similar to QDISK device). The default behaviour for adding device is to wait when the backend goes to XenbusStateInitWait and the default behaviour on removing device is to start generic device remove procedure.

Also this commit fixes removing guest function: before the guest was removed when all VIF and VBD devices are removed. The fix removes guest when all created devices are removed. This is done by checking the guest device list instead of checking num_vifs and num_vbds. num_vifs and num_vbds variables are removed as redundant in this case.

---
 tools/libxl/libxl_device.c | 63 +++++++++++++++-----------------------
 1 file changed, 24 insertions(+), 39 deletions(-)
tools/libxl/libxl_device.c +24 -39
index 1402b61a81..9d05d2fd13 100644
@@ -1477,7 +1477,7 @@ typedef struct libxl__ddomain_device {
  */
 typedef struct libxl__ddomain_guest {
     uint32_t domid;
-    int num_vifs, num_vbds, num_qdisks;
+    int num_qdisks;
     LIBXL_SLIST_HEAD(, struct libxl__ddomain_device) devices;
     LIBXL_SLIST_ENTRY(struct libxl__ddomain_guest) next;
 } libxl__ddomain_guest;
@@ -1530,8 +1530,7 @@ static void check_and_maybe_remove_guest(libxl__gc *gc,
 {
     assert(ddomain);
 
-    if (dguest != NULL &&
-        dguest->num_vifs + dguest->num_vbds + dguest->num_qdisks == 0) {
+    if (dguest != NULL && LIBXL_SLIST_FIRST(&dguest->devices) == NULL) {
         LIBXL_SLIST_REMOVE(&ddomain->guests, dguest, libxl__ddomain_guest,
                            next);
         LOGD(DEBUG, dguest->domid, "Removed domain from the list of active guests");
@@ -1571,24 +1570,6 @@ static int add_device(libxl__egc *egc, libxl__ao *ao,
          libxl__device_backend_path(gc, dev));
 
     switch(dev->backend_kind) {
-    case LIBXL__DEVICE_KIND_VBD:
-    case LIBXL__DEVICE_KIND_VIF:
-        if (dev->backend_kind == LIBXL__DEVICE_KIND_VBD) dguest->num_vbds++;
-        if (dev->backend_kind == LIBXL__DEVICE_KIND_VIF) dguest->num_vifs++;
-
-        GCNEW(aodev);
-        libxl__prepare_ao_device(ao, aodev);
-        /*
-         * Clone the libxl__device to avoid races if remove_device is called
-         * before the device addition has finished.
-         */
-        GCNEW(aodev->dev);
-        *aodev->dev = *dev;
-        aodev->action = LIBXL__DEVICE_ACTION_ADD;
-        aodev->callback = device_complete;
-        libxl__wait_device_connection(egc, aodev);
-
-        break;
     case LIBXL__DEVICE_KIND_QDISK:
         if (dguest->num_qdisks == 0) {
             GCNEW(dmss);
@@ -1599,10 +1580,19 @@ static int add_device(libxl__egc *egc, libxl__ao *ao,
             libxl__spawn_qdisk_backend(egc, dmss);
         }
         dguest->num_qdisks++;
-
         break;
     default:
-        rc = 1;
+        GCNEW(aodev);
+        libxl__prepare_ao_device(ao, aodev);
+        /*
+         * Clone the libxl__device to avoid races if remove_device is called
+         * before the device addition has finished.
+         */
+        GCNEW(aodev->dev);
+        *aodev->dev = *dev;
+        aodev->action = LIBXL__DEVICE_ACTION_ADD;
+        aodev->callback = device_complete;
+        libxl__wait_device_connection(egc, aodev);
         break;
     }
 
@@ -1619,11 +1609,17 @@ static int remove_device(libxl__egc *egc, libxl__ao *ao,
     int rc = 0;
 
     switch(ddev->dev->backend_kind) {
-    case LIBXL__DEVICE_KIND_VBD:
-    case LIBXL__DEVICE_KIND_VIF:
-        if (dev->backend_kind == LIBXL__DEVICE_KIND_VBD) dguest->num_vbds--;
-        if (dev->backend_kind == LIBXL__DEVICE_KIND_VIF) dguest->num_vifs--;
-
+    case LIBXL__DEVICE_KIND_QDISK:
+        if (--dguest->num_qdisks == 0) {
+            rc = libxl__destroy_qdisk_backend(gc, dev->domid);
+            if (rc)
+                goto out;
+        }
+        libxl__device_destroy(gc, dev);
+        /* Return > 0, no ao has been dispatched */
+        rc = 1;
+        break;
+    default:
         GCNEW(aodev);
         libxl__prepare_ao_device(ao, aodev);
         /*
@@ -1636,17 +1632,6 @@ static int remove_device(libxl__egc *egc, libxl__ao *ao,
         aodev->callback = device_complete;
         libxl__initiate_device_generic_remove(egc, aodev);
         break;
-    case LIBXL__DEVICE_KIND_QDISK:
-        if (--dguest->num_qdisks == 0) {
-            rc = libxl__destroy_qdisk_backend(gc, dev->domid);
-            if (rc)
-                goto out;
-        }
-        libxl__device_destroy(gc, dev);
-        /* Fall through to return > 0, no ao has been dispatched */
-    default:
-        rc = 1;
-        break;
     }
 
     /*
Signature · 7 lines
2.17.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v2 3/3
#12 lore raw

Re: [Xen-devel] [PATCH v2 3/3] libxl: make default path to add/remove all PV devices

Oleksandr Grytsov writes ("[PATCH v2 3/3] libxl: make default path to add/remove all PV devices"):

Quoted 11 lines

From: Oleksandr Grytsov <[email protected]>

Adding/removing device is handled for specific devices only: VBD, VIF, QDISK. This commit adds default case to handle adding/removing for all PV devices by default, except QDISK device, which requires special handling. If any other device is required a special handling it should be done by implementing separate case (similar to QDISK device). The default behaviour for adding device is to wait when the backend goes to XenbusStateInitWait and the default behaviour on removing device is to start generic device remove procedure.

Quoted 5 lines

Also this commit fixes removing guest function: before the guest was removed when all VIF and VBD devices are removed. The fix removes guest when all created devices are removed. This is done by checking the guest device list instead of checking num_vifs and num_vbds. num_vifs and num_vbds variables are removed as redundant in this case.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Acked-by on v2 3/3
#13 lore raw

Re: [Xen-devel] [PATCH v2 3/3] libxl: make default path to add/remove all PV devices

In reply to Oleksandr Grytsov
Quoted 11 lines · Oleksandr Grytsov

From: Oleksandr Grytsov <[email protected]>

Adding/removing device is handled for specific devices only: VBD, VIF, QDISK. This commit adds default case to handle adding/removing for all PV devices by default, except QDISK device, which requires special handling. If any other device is required a special handling it should be done by implementing separate case (similar to QDISK device). The default behaviour for adding device is to wait when the backend goes to XenbusStateInitWait and the default behaviour on removing device is to start generic device remove procedure.

Quoted 7 lines

Also this commit fixes removing guest function: before the guest was removed when all VIF and VBD devices are removed. The fix removes guest when all created devices are removed. This is done by checking the guest device list instead of checking num_vifs and num_vbds. num_vifs and num_vbds variables are removed as redundant in this case.

I agree with this approach in general, but I haven't looked closely into the implementation. FWIW:

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#14 lore raw
In reply to Oleksandr Grytsov

Oleksandr Grytsov writes ("[PATCH v2 0/3] Remove backend xen store entry on domain destroy"):

Quoted 8 lines

From: Oleksandr Grytsov <[email protected]>

Changes since v1:

* add commit to rename VKB backend type "linux" to "pv";
* add default case to handle adding/removing PV devices in add_device, remove_device functions (libxl_device.c);
* add comment about removing num_vifs, num_vbds into commit message.

Thank you for this. As you see, I approve. We need an ack from the Release Manager, so I have addresed Juergen (and also added the missing `for-4.13' tag to the subject).

Given where we are in the release cycle with this (which I stress is not really Oleksandr's fault) I think I would like to get a 2nd review from another libxl maintainer. Anthony, or Wei, would you like to take a look ?

Thanks,
Ian.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel