Skip to contentSkip to conversation

numa: Clean up query-memdev error handling, plug leak

PATCH Superseded Markus Armbruster 2 participants 7 messages Nov 20, 2015

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 numa: Clean up query-memdev error handling, plug leak

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Screwed up in commit 76b5d85 "qmp: add query-memdev".

---
 numa.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
numa.c +4 -6
index fdfe294..a584e8e 100644
@@ -568,6 +568,7 @@ static int query_memdev(Object *obj, void *opaque)
 
     return 0;
 error:
+    error_free(err);
     g_free(m->value);
     g_free(m);
 
@@ -576,15 +577,12 @@ error:
 
 MemdevList *qmp_query_memdev(Error **errp)
 {
-    Object *obj;
+    Object *obj = object_get_objects_root();
     MemdevList *list = NULL;
 
-    obj = object_get_objects_root();
-    if (obj == NULL) {
-        return NULL;
-    }
-
     if (object_child_foreach(obj, query_memdev, &list) != 0) {
+        /* FIXME propagate the error query_memdev() throws away */
+        error_setg(errp, "Unknown error");
         goto error;
     }
 
Signature · 1 line
2.4.3
Reviewed-by on v1 1/1
#1 lore raw
Quoted 15 lines · Markus Armbruster

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Quoted 3 lines

Screwed up in commit 76b5d85 "qmp: add query-memdev".

Do you know how to trigger a query_memdev() error today, or is just theoretical?

Signature · 1 line
Eduardo
#2 lore raw
Quoted 16 lines · Eduardo Habkost
Quoted 15 lines · Markus Armbruster

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Quoted 8 lines
Quoted 3 lines

Screwed up in commit 76b5d85 "qmp: add query-memdev".

Do you know how to trigger a query_memdev() error today, or is just theoretical?

Theoretical; I tested by injecting an error with gdb.

query_memdev() fails exactly when some object_property_get_FOO() fails. If we decide such a failure would always be a programming error, we can pass &error_abort and simplify things. Opinions?

#3 lore raw
Quoted 24 lines · Markus Armbruster
Quoted 22 lines · Eduardo Habkost
Quoted 15 lines · Markus Armbruster

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Quoted 3 lines

Screwed up in commit 76b5d85 "qmp: add query-memdev".

Quoted 8 lines
Quoted 2 lines

Do you know how to trigger a query_memdev() error today, or is just theoretical?

Theoretical; I tested by injecting an error with gdb.

query_memdev() fails exactly when some object_property_get_FOO() fails. If we decide such a failure would always be a programming error, we can pass &error_abort and simplify things. Opinions?

The hostmem-backend property getters should never fail. Using &error_abort on query_memdev() would make everything simpler.

(I would even use the HostMemoryBackend struct fields directly, instead of QOM properties. Is there a good reason to use QOM to fetch the data that's readily available in the C struct?)

Signature · 1 line
Eduardo
#4 lore raw
Quoted 34 lines · Eduardo Habkost
Quoted 24 lines · Markus Armbruster
Quoted 22 lines · Eduardo Habkost
Quoted 15 lines · Markus Armbruster

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Quoted 3 lines

Screwed up in commit 76b5d85 "qmp: add query-memdev".

Quoted 8 lines
Quoted 2 lines

Do you know how to trigger a query_memdev() error today, or is just theoretical?

Theoretical; I tested by injecting an error with gdb.

query_memdev() fails exactly when some object_property_get_FOO() fails. If we decide such a failure would always be a programming error, we can pass &error_abort and simplify things. Opinions?

Quoted 6 lines

The hostmem-backend property getters should never fail. Using &error_abort on query_memdev() would make everything simpler.

(I would even use the HostMemoryBackend struct fields directly, instead of QOM properties. Is there a good reason to use QOM to fetch the data that's readily available in the C struct?)

I can't see why not, sketch appended. Note that I still go through the property for host_nodes, because I couldn't see how to do that simpler. If you like this patch, I'll post it as v2.

numa: Clean up query-memdev error handling

qmp_query_memdev() has two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. It looks like it could, but that's just because its code is pointlessly complicated. Simplify it, and drop the useless error handling.

Messed up in commit 76b5d85 "qmp: add query-memdev".

---
 numa.c | 69 ++++++++++--------------------------------------------------------
 1 file changed, 10 insertions(+), 59 deletions(-)
numa.c +10 -59
index fdfe294..4e9be9f 100644
@@ -517,80 +517,31 @@ static int query_memdev(Object *obj, void *opaque)
 {
     MemdevList **list = opaque;
     MemdevList *m = NULL;
-    Error *err = NULL;
 
     if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
+        HostMemoryBackend *backend = MEMORY_BACKEND(obj);
+
         m = g_malloc0(sizeof(*m));
-
         m->value = g_malloc0(sizeof(*m->value));
-
-        m->value->size = object_property_get_int(obj, "size",
-                                                 &err);
-        if (err) {
-            goto error;
-        }
-
-        m->value->merge = object_property_get_bool(obj, "merge",
-                                                   &err);
-        if (err) {
-            goto error;
-        }
-
-        m->value->dump = object_property_get_bool(obj, "dump",
-                                                  &err);
-        if (err) {
-            goto error;
-        }
-
-        m->value->prealloc = object_property_get_bool(obj,
-                                                      "prealloc", &err);
-        if (err) {
-            goto error;
-        }
-
-        m->value->policy = object_property_get_enum(obj,
-                                                    "policy",
-                                                    "HostMemPolicy",
-                                                    &err);
-        if (err) {
-            goto error;
-        }
-
+        m->value->size = backend->size;
+        m->value->merge = backend->merge;
+        m->value->dump = backend->dump;
+        m->value->prealloc = backend->prealloc;
+        m->value->policy = backend->policy;
         object_property_get_uint16List(obj, "host-nodes",
-                                       &m->value->host_nodes, &err);
-        if (err) {
-            goto error;
-        }
-
+                                       &m->value->host_nodes, &error_abort);
         m->next = *list;
         *list = m;
     }
 
     return 0;
-error:
-    g_free(m->value);
-    g_free(m);
-
-    return -1;
 }
 
 MemdevList *qmp_query_memdev(Error **errp)
 {
-    Object *obj;
+    Object *obj = object_get_objects_root();
     MemdevList *list = NULL;
 
-    obj = object_get_objects_root();
-    if (obj == NULL) {
-        return NULL;
-    }
-
-    if (object_child_foreach(obj, query_memdev, &list) != 0) {
-        goto error;
-    }
-
+    object_child_foreach(obj, query_memdev, &list);
     return list;
-
-error:
-    qapi_free_MemdevList(list);
-    return NULL;
 }
Signature · 1 line
2.4.3
#5 lore raw
Quoted 72 lines · Markus Armbruster
Quoted 34 lines · Eduardo Habkost
Quoted 24 lines · Markus Armbruster
Quoted 22 lines · Eduardo Habkost
Quoted 15 lines · Markus Armbruster

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Quoted 3 lines

Screwed up in commit 76b5d85 "qmp: add query-memdev".

Quoted 8 lines
Quoted 2 lines

Do you know how to trigger a query_memdev() error today, or is just theoretical?

Theoretical; I tested by injecting an error with gdb.

query_memdev() fails exactly when some object_property_get_FOO() fails. If we decide such a failure would always be a programming error, we can pass &error_abort and simplify things. Opinions?

Quoted 6 lines

The hostmem-backend property getters should never fail. Using &error_abort on query_memdev() would make everything simpler.

(I would even use the HostMemoryBackend struct fields directly, instead of QOM properties. Is there a good reason to use QOM to fetch the data that's readily available in the C struct?)

I can't see why not, sketch appended. Note that I still go through the property for host_nodes, because I couldn't see how to do that simpler. If you like this patch, I'll post it as v2.

numa: Clean up query-memdev error handling

qmp_query_memdev() has two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. It looks like it could, but that's just because its code is pointlessly complicated. Simplify it, and drop the useless error handling.

Messed up in commit 76b5d85 "qmp: add query-memdev".

---
 numa.c | 69 ++++++++++--------------------------------------------------------
 1 file changed, 10 insertions(+), 59 deletions(-)
numa.c +0 -0
index fdfe294..4e9be9f 100644
@@ -517,80 +517,31 @@ static int query_memdev(Object *obj, void *opaque)

[...]

Quoted 2 lines
numa.c +0 -2
-        m->value->prealloc = object_property_get_bool(obj,
-                                                      "prealloc", &err);

[...]

Quoted 1 line

+ m->value->prealloc = backend->prealloc;

This changes the QMP command result because the property getter currently[1] returns backend->prealloc || backend->force_prealloc.

So, I stand corrected: there are at least two cases where using the QOM property is useful at query_memdev(). Let's just use &error_abort and keep all the existing object_property_get_*() calls, then?

[1] I am not sure this is the right thing to do, but if we're going to change that, let's do it after 2.5.0.

Signature · 1 line
Eduardo
#6 lore raw
Quoted 82 lines · Eduardo Habkost
Quoted 72 lines · Markus Armbruster
Quoted 34 lines · Eduardo Habkost
Quoted 24 lines · Markus Armbruster
Quoted 22 lines · Eduardo Habkost
Quoted 15 lines · Markus Armbruster

qmp_query_memdev() doesn't fail. Instead, it returns an empty list. That's wrong.

Two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. This can happen, and the error to return is the one that query_memdev() currently leaks. Passing the error from query_memdev() to qmp_query_memdev() isn't so simple, because object_child_foreach() is in the way. Fixable, but I'd rather not try it in hard freeze. Plug the leak, make up an error, and add a FIXME for the remaining work.

Quoted 3 lines

Screwed up in commit 76b5d85 "qmp: add query-memdev".

Quoted 8 lines
Quoted 2 lines

Do you know how to trigger a query_memdev() error today, or is just theoretical?

Theoretical; I tested by injecting an error with gdb.

query_memdev() fails exactly when some object_property_get_FOO() fails. If we decide such a failure would always be a programming error, we can pass &error_abort and simplify things. Opinions?

Quoted 6 lines

The hostmem-backend property getters should never fail. Using &error_abort on query_memdev() would make everything simpler.

(I would even use the HostMemoryBackend struct fields directly, instead of QOM properties. Is there a good reason to use QOM to fetch the data that's readily available in the C struct?)

I can't see why not, sketch appended. Note that I still go through the property for host_nodes, because I couldn't see how to do that simpler. If you like this patch, I'll post it as v2.

numa: Clean up query-memdev error handling

qmp_query_memdev() has two error paths:

* When object_get_objects_root() returns null. It never does, so simply drop the useless error handling.

* When query_memdev() fails. It looks like it could, but that's just because its code is pointlessly complicated. Simplify it, and drop the useless error handling.

Messed up in commit 76b5d85 "qmp: add query-memdev".

---
 numa.c | 69 ++++++++++--------------------------------------------------------
 1 file changed, 10 insertions(+), 59 deletions(-)
numa.c +0 -0
index fdfe294..4e9be9f 100644
@@ -517,80 +517,31 @@ static int query_memdev(Object *obj, void *opaque)

[...]

Quoted 2 lines
numa.c +0 -2
-        m->value->prealloc = object_property_get_bool(obj,
-                                                      "prealloc", &err);

[...]

Quoted 1 line

+ m->value->prealloc = backend->prealloc;

This changes the QMP command result because the property getter currently[1] returns backend->prealloc || backend->force_prealloc.

Quoted 7 lines

So, I stand corrected: there are at least two cases where using the QOM property is useful at query_memdev(). Let's just use &error_abort and keep all the existing object_property_get_*() calls, then?

[1] I am not sure this is the right thing to do, but if we're going to change that, let's do it after 2.5.0.

Okay, will do that in v2.