Skip to contentSkip to conversation

xen: use qdev_unplug() insteda of g_free() in xen_pv_find_xendev()

PATCH Superseded Juergen Gross 2 participants 4 messages Jan 30, 2017

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen: use qdev_unplug() insteda of g_free() in xen_pv_find_xendev()

The error exits of xen_pv_find_xendev() free the new xen-device via g_free() which is wrong.

As the xen-device has been initialized as qdev it must be removed via qdev_unplug().

This bug has been introduced with commit 3a6c9172ac5951e6dac2b3f6 ("xen: create qdev for each backend device").

---
 hw/xen/xen_backend.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
hw/xen/xen_backend.c +2 -2
index d119004..030772b 100644
@@ -145,7 +145,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
     xendev->evtchndev = xenevtchn_open(NULL, 0);
     if (xendev->evtchndev == NULL) {
         xen_pv_printf(NULL, 0, "can't open evtchn device\n");
-        g_free(xendev);
+        qdev_unplug(&xendev->qdev, NULL);
         return NULL;
     }
     fcntl(xenevtchn_fd(xendev->evtchndev), F_SETFD, FD_CLOEXEC);
@@ -155,7 +155,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
         if (xendev->gnttabdev == NULL) {
             xen_pv_printf(NULL, 0, "can't open gnttab device\n");
             xenevtchn_close(xendev->evtchndev);
-            g_free(xendev);
+            qdev_unplug(&xendev->qdev, NULL);
             return NULL;
         }
     } else {
Signature · 1 line
2.10.2
#1 lore raw
Quoted 30 lines · Juergen Gross

The error exits of xen_pv_find_xendev() free the new xen-device via g_free() which is wrong.

As the xen-device has been initialized as qdev it must be removed via qdev_unplug().

This bug has been introduced with commit 3a6c9172ac5951e6dac2b3f6 ("xen: create qdev for each backend device").

---
 hw/xen/xen_backend.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
hw/xen/xen_backend.c +1 -1
index d119004..030772b 100644
@@ -145,7 +145,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
     xendev->evtchndev = xenevtchn_open(NULL, 0);
     if (xendev->evtchndev == NULL) {
         xen_pv_printf(NULL, 0, "can't open evtchn device\n");
-        g_free(xendev);
+        qdev_unplug(&xendev->qdev, NULL);
         return NULL;
     }
     fcntl(xenevtchn_fd(xendev->evtchndev), F_SETFD, FD_CLOEXEC);

hw/xen/xen_backend.c static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev, · line 161

Quoted 8 lines
hw/xen/xen_backend.c +1 -1
@@ -155,7 +155,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
         if (xendev->gnttabdev == NULL) {
             xen_pv_printf(NULL, 0, "can't open gnttab device\n");
             xenevtchn_close(xendev->evtchndev);
-            g_free(xendev);
+            qdev_unplug(&xendev->qdev, NULL);
             return NULL;
         }
     } else {

I think this will leak memory (and that we're already leaking memory), because the code is creating the xendev with

    xendev = g_malloc0(ops->size);
    object_initialize(&xendev->qdev, ops->size, TYPE_XENBACKEND);

This is saying "I own and am responsible for freeing the memory that the device object lives in". If you want the object system to handle freeing the memory for you when the reference count goes to zero, then you need to create it with
xendev = object_new()
which we can't do here because we're allocating ops->size bytes rather than just the size of the object type.

Two options I think:
(1) have your code do

    OBJECT(xendev)->free = g_free;

after the object_initialize (to tell the object system how
to free this object)
(2) call both qdev_unplug and g_free

I think (1) is better because it will definitely work even if the qdev bus system is holding on to an object reference after it returns from qdev_unplug() for some reason, and it will also mean we free the memory when we do a qdev_unplug in xen_pv_del_xendev(), rather than leaking it.

Side note: Using DEVICE(xendev) is better QOM style than &xendev->qdev.

(The weirdness with doing a g_malloc0(ops->size) appears to be because Xen backends have their own setup for saying "allocate this many bytes for my data", rather than having XenBlkDev, XenConsole, etc be QOM subclasses of XenDevice. Presumably that code all predates QEMU's QOMification.)

We also do OBJECT(thing)->free = something in hw/core/sysbus.c; maybe we should have a function to abstract this away? Not sure.

thanks
-- PMM

#2 lore raw
Quoted 60 lines · Peter Maydell
Quoted 30 lines · Juergen Gross

The error exits of xen_pv_find_xendev() free the new xen-device via g_free() which is wrong.

As the xen-device has been initialized as qdev it must be removed via qdev_unplug().

This bug has been introduced with commit 3a6c9172ac5951e6dac2b3f6 ("xen: create qdev for each backend device").

---
 hw/xen/xen_backend.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
hw/xen/xen_backend.c +1 -1
index d119004..030772b 100644
@@ -145,7 +145,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
     xendev->evtchndev = xenevtchn_open(NULL, 0);
     if (xendev->evtchndev == NULL) {
         xen_pv_printf(NULL, 0, "can't open evtchn device\n");
-        g_free(xendev);
+        qdev_unplug(&xendev->qdev, NULL);
         return NULL;
     }
     fcntl(xenevtchn_fd(xendev->evtchndev), F_SETFD, FD_CLOEXEC);

hw/xen/xen_backend.c static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev, · line 161

Quoted 8 lines
hw/xen/xen_backend.c +1 -1
@@ -155,7 +155,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
         if (xendev->gnttabdev == NULL) {
             xen_pv_printf(NULL, 0, "can't open gnttab device\n");
             xenevtchn_close(xendev->evtchndev);
-            g_free(xendev);
+            qdev_unplug(&xendev->qdev, NULL);
             return NULL;
         }
     } else {

I think this will leak memory (and that we're already leaking memory), because the code is creating the xendev with

    xendev = g_malloc0(ops->size);
    object_initialize(&xendev->qdev, ops->size, TYPE_XENBACKEND);

This is saying "I own and am responsible for freeing the memory that the device object lives in". If you want the object system to handle freeing the memory for you when the reference count goes to zero, then you need to create it with
xendev = object_new()
which we can't do here because we're allocating ops->size bytes rather than just the size of the object type.

Two options I think:
(1) have your code do

    OBJECT(xendev)->free = g_free;

after the object_initialize (to tell the object system how
to free this object)
(2) call both qdev_unplug and g_free

Quoted 5 lines

I think (1) is better because it will definitely work even if the qdev bus system is holding on to an object reference after it returns from qdev_unplug() for some reason, and it will also mean we free the memory when we do a qdev_unplug in xen_pv_del_xendev(), rather than leaking it.

I agree. I'll send V2 with (1) included.

Quoted 2 lines

Side note: Using DEVICE(xendev) is better QOM style than &xendev->qdev.

Okay. Will change.

Thanks for the very precise description.

Juergen

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

#3 lore raw
In reply to Juergen Gross
Quoted 30 lines · Juergen Gross

The error exits of xen_pv_find_xendev() free the new xen-device via g_free() which is wrong.

As the xen-device has been initialized as qdev it must be removed via qdev_unplug().

This bug has been introduced with commit 3a6c9172ac5951e6dac2b3f6 ("xen: create qdev for each backend device").

---
 hw/xen/xen_backend.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
hw/xen/xen_backend.c +1 -1
index d119004..030772b 100644
@@ -145,7 +145,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
     xendev->evtchndev = xenevtchn_open(NULL, 0);
     if (xendev->evtchndev == NULL) {
         xen_pv_printf(NULL, 0, "can't open evtchn device\n");
-        g_free(xendev);
+        qdev_unplug(&xendev->qdev, NULL);
         return NULL;
     }
     fcntl(xenevtchn_fd(xendev->evtchndev), F_SETFD, FD_CLOEXEC);

hw/xen/xen_backend.c static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev, · line 161

Quoted 8 lines
hw/xen/xen_backend.c +1 -1
@@ -155,7 +155,7 @@ static struct XenDevice *xen_be_get_xendev(const char *type, int dom, int dev,
         if (xendev->gnttabdev == NULL) {
             xen_pv_printf(NULL, 0, "can't open gnttab device\n");
             xenevtchn_close(xendev->evtchndev);
-            g_free(xendev);
+            qdev_unplug(&xendev->qdev, NULL);
             return NULL;
         }
     } else {

I think this will leak memory (and that we're already leaking memory), because the code is creating the xendev with

    xendev = g_malloc0(ops->size);
    object_initialize(&xendev->qdev, ops->size, TYPE_XENBACKEND);

This is saying "I own and am responsible for freeing the memory that the device object lives in". If you want the object system to handle freeing the memory for you when the reference count goes to zero, then you need to create it with
xendev = object_new()
which we can't do here because we're allocating ops->size bytes rather than just the size of the object type.

Two options I think:
(1) have your code do

    OBJECT(xendev)->free = g_free;

after the object_initialize (to tell the object system how
to free this object)
(2) call both qdev_unplug and g_free

I think (1) is better because it will definitely work even if the qdev bus system is holding on to an object reference after it returns from qdev_unplug() for some reason, and it will also mean we free the memory when we do a qdev_unplug in xen_pv_del_xendev(), rather than leaking it.

Side note: Using DEVICE(xendev) is better QOM style than &xendev->qdev.

(The weirdness with doing a g_malloc0(ops->size) appears to be because Xen backends have their own setup for saying "allocate this many bytes for my data", rather than having XenBlkDev, XenConsole, etc be QOM subclasses of XenDevice. Presumably that code all predates QEMU's QOMification.)

We also do OBJECT(thing)->free = something in hw/core/sysbus.c; maybe we should have a function to abstract this away? Not sure.

thanks
-- PMM

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