As discovered by David Disseldorp <ddiss@suse.de>, convert_string_talloc()
doesn't always return consistent results for a zero length string. The
API states an incoming string must *always* contain the terminating null,
but unfotunately too much code expects passing in a zero source length
to return a null terminated string, so at least ensure we return a
correct null string in the required character set and return the
correct length.
Also ensure we cannot return a zero length for a converted string
(we ensure that the returned buffer is always allocated and zero
terminated anyway) as calling code depends on the fact that returning
true from this function will *always* return a non-zero length (as
it must include the terminating null).
Note this is a different fix from what went into master (this is
identical to the fix I'm planning for 3.5.x) as convert_string_talloc()
has diverged between the two.
Jeremy.
(cherry picked from commit
bb3ed43584e6d2c4d64b5f7b9e70a7db7f3e859d)
errno = EINVAL;
return false;
}
+
if (srclen == 0) {
- ob = talloc_strdup(ctx, "");
+ /* We really should treat this as an error, but
+ there are too many callers that need this to
+ return a NULL terminated string in the correct
+ character set. */
+ if (to == CH_UTF16LE|| to == CH_UTF16BE || to == CH_UTF16MUNGED) {
+ destlen = 2;
+ } else {
+ destlen = 1;
+ }
+ ob = talloc_zero_array(ctx, char, destlen);
if (ob == NULL) {
errno = ENOMEM;
return false;
}
+ *converted_size = destlen;
*dest = ob;
- *converted_size = 0;
return true;
}
ob[destlen] = '\0';
ob[destlen+1] = '\0';
+ /* Ensure we can never return a *converted_size of zero. */
+ if (destlen == 0) {
+ /* This can happen from a bad iconv "use_as_is:" call. */
+ if (to == CH_UTF16LE|| to == CH_UTF16BE || to == CH_UTF16MUNGED) {
+ destlen = 2;
+ } else {
+ destlen = 1;
+ }
+ }
+
*converted_size = destlen;
return true;