Re: [Qemu-devel] [PATCH v2] hostmem: Validate host-nodes before setting bitmap

[Markus Armbruster]

Eduardo Habkost <address@hidden> writes:

> I don't understand yet if there's a leak at
> host_memory_backend_get_host_nodes().  Won't
> visit_type_uint16List() take ownership of the list on that case?

Nope.  I checked with valgrind:

    $ valgrind --leak-check=full upstream-qemu -nodefaults -S -display none 
-qmp stdio -object 
memory-backend-file,id=mem0,mem-path=x,size=4096,host-nodes=1,policy=bind
    [...]
    {"QMP": {"version": {"qemu": {"micro": 92, "minor": 0, "major": 3}, 
"package": "v3.1.0-rc2-48-g039d4e3df0-dirty"}, "capabilities": []}}
    {"execute": "qmp_capabilities"}
    {"return": {}}
    { "execute": "qom-get", "arguments": { "path": "mem0", "property": 
"host-nodes" {"execute": "qom-get", "arguments": {"path": "mem0", "property": 
"host-nodes"}}
    {"return": [1]}
    {"execute": "quit"}
    {"return": {}}
    {"timestamp": {"seconds": 1543592652, "microseconds": 950994}, "event": 
"SHUTDOWN", "data": {"guest": false}}
    ==4954== 
    ==4954== HEAP SUMMARY:
    ==4954==     in use at exit: 3,631,673 bytes in 14,706 blocks
    ==4954==   total heap usage: 51,347 allocs, 36,641 frees, 24,195,921 bytes 
allocated
    [...]
    ==4954== 16 bytes in 1 blocks are definitely lost in loss record 1,964 of 
5,297
    ==4954==    at 0x4C3111A: calloc (vg_replace_malloc.c:752)
    ==4954==    by 0x574948D: g_malloc0 (in /usr/lib64/libglib-2.0.so.0.5600.3)
    ==4954==    by 0x9E1CE0: opts_start_list (opts-visitor.c:228)
    ==4954==    by 0x9DAB35: visit_start_list (qapi-visit-core.c:78)
    ==4954==    by 0x99BA3A: visit_type_uint16List (qapi-builtin-visit.c:272)
    ==4954==    by 0x5F911B: host_memory_backend_set_host_nodes (hostmem.c:108)
    ==4954==    by 0x8AC7D4: object_property_set (object.c:1183)
    ==4954==    by 0x8AFC82: user_creatable_add_type (object_interfaces.c:73)
    ==4954==    by 0x8AFED2: user_creatable_add_opts (object_interfaces.c:131)
    ==4954==    by 0x8AFFCD: user_creatable_add_opts_foreach 
(object_interfaces.c:154)
    ==4954==    by 0xA0B9B9: qemu_opts_foreach (qemu-option.c:1171)
    ==4954==    by 0x5C6C44: main (vl.c:4415)
    ==4954== 
    ==4954== 16 bytes in 1 blocks are definitely lost in loss record 1,965 of 
5,297
    ==4954==    at 0x4C3111A: calloc (vg_replace_malloc.c:752)
    ==4954==    by 0x574948D: g_malloc0 (in /usr/lib64/libglib-2.0.so.0.5600.3)
    ==4954==    by 0x5F8FF5: host_memory_backend_get_host_nodes (hostmem.c:82)
    ==4954==    by 0x8AC739: object_property_get (object.c:1168)
    ==4954==    by 0x8AF910: object_property_get_qobject (qom-qobject.c:39)
    ==4954==    by 0x5E1736: qmp_qom_get (qmp.c:249)
    ==4954==    by 0x5D872F: qmp_marshal_qom_get (qapi-commands-misc.c:1284)
    ==4954==    by 0x9DF5C1: do_qmp_dispatch (qmp-dispatch.c:129)
    ==4954==    by 0x9DF788: qmp_dispatch (qmp-dispatch.c:171)
    ==4954==    by 0x42C0C1: monitor_qmp_dispatch (monitor.c:4085)
    ==4954==    by 0x42C3E1: monitor_qmp_bh_dispatcher (monitor.c:4157)
    ==4954==    by 0x9EEDB1: aio_bh_call (async.c:90)
    [...]
    ==4954== LEAK SUMMARY:
    ==4954==    definitely lost: 32 bytes in 2 blocks
    ==4954==    indirectly lost: 0 bytes in 0 blocks
    ==4954==      possibly lost: 2,504 bytes in 20 blocks
    ==4954==    still reachable: 3,629,137 bytes in 14,684 blocks
    ==4954==                       of which reachable via heuristic:
    ==4954==                         newarray           : 1,536 bytes in 16 
blocks
    ==4954==         suppressed: 0 bytes in 0 blocks
    ==4954== Reachable blocks (those to which a pointer was found) are not 
shown.
    ==4954== To see them, rerun with: --leak-check=full --show-leak-kinds=all
    ==4954== 
    ==4954== For counts of detected and suppressed errors, rerun with: -v
    ==4954== Use --track-origins=yes to see where uninitialised values come from
    ==4954== ERROR SUMMARY: 24 errors from 24 contexts (suppressed: 0 from 0)

The first block shown is leaked in host_memory_backend_set_host_nodes()
on behalf of -object, the second block in
host_memory_backend_get_host_nodes() on behalf of qom-get.

Full disclosure: I hacked host_memory_backend_complete() to skip
mbind():

diff --git a/backends/hostmem.c b/backends/hostmem.c
index 1a89342039..0e40bb1ad4 100644
--- a/backends/hostmem.c
+++ b/backends/hostmem.c
@@ -333,7 +333,7 @@ host_memory_backend_memory_complete(UserCreatable *uc, 
Error **errp)
         assert(sizeof(backend->host_nodes) >=
                BITS_TO_LONGS(MAX_NODES + 1) * sizeof(unsigned long));
         assert(maxnode <= MAX_NODES);
-        if (mbind(ptr, sz, backend->policy,
+        if (0 && mbind(ptr, sz, backend->policy,
                   maxnode ? backend->host_nodes : NULL, maxnode + 1, flags)) {
             if (backend->policy != MPOL_DEFAULT || errno != ENOSYS) {
                 error_setg_errno(errp, errno,