On Tuesday, 25 April 2017 22:25:20 CEST Matteo Cafasso wrote:
When constructing the returned objects, check the return value of Python APIs.
A RuntimeError will be raised on failure pointing to the problematic entry and the field name.
Signed-off-by: Matteo Cafasso <noxdafox@gmail.com> --- generator/python.ml | 143 +++++++++++++++++++++++++++++++++++----------------- 1 file changed, 97 insertions(+), 46 deletions(-)
diff --git a/generator/python.ml b/generator/python.ml index 11dc48102..7d86131b1 100644 --- a/generator/python.ml +++ b/generator/python.ml @@ -152,12 +152,20 @@ and generate_python_structs () = pr "PyObject *\n"; pr "guestfs_int_py_put_%s_list (struct guestfs_%s_list *%ss)\n" typ typ typ; pr "{\n"; - pr " PyObject *list;\n"; + pr " PyObject *list, *element;\n"; pr " size_t i;\n"; pr "\n"; pr " list = PyList_New (%ss->len);\n" typ; - pr " for (i = 0; i < %ss->len; ++i)\n" typ; - pr " PyList_SetItem (list, i, guestfs_int_py_put_%s (&%ss->val[i]));\n" typ typ; + pr " if (list == NULL) {\n"; + pr " PyErr_SetString (PyExc_RuntimeError, \"PyList_New\");\n";
This should most probably: return PyErr_NoMemory(); as it sets the right exception for "no memory" situations. Also all the other NULL checks in this patch for PySomething*() APIs should do the same.
+ pr " return NULL;\n"; + pr " }\n"; + pr " for (i = 0; i < %ss->len; ++i) {\n" typ; + pr " element = guestfs_int_py_put_%s (&%ss->val[i]);\n" typ typ; + pr " if (element == NULL)\n"; + pr " return NULL;\n";
This leaks 'list'. Similar situation also in other parts of this patch. Also, I guess also the other guestfs_int_* functions in python/handle.c need similar checks/changes? Moreover, can you please add a regression test for this? I.e. a new python/t/test8xxSomething.py, formatted as PEP-8. See the section "ADDING A NEW LANGUAGE BINDING" in the docs/guestfs-hacking documentation for the numbering of tests. Thanks, -- Pino Toscano