On Fri, Jul 28, 2023 at 09:25:24AM +0200, Amadeusz Sławiński wrote: > On 7/27/2023 12:34 AM, Kees Cook wrote: > > On Wed, Jul 26, 2023 at 09:12:18PM +0000, justinstitt@xxxxxxxxxx wrote: > > > `strncpy` is deprecated for use on NUL-terminated destination strings [1]. > > > > > > A suitable replacement is `strscpy` [2] due to the fact that it > > > guarantees NUL-termination on its destination buffer argument which is > > > _not_ the case for `strncpy`! > > > > > > It was pretty difficult, in this case, to try and figure out whether or > > > not the destination buffer was zero-initialized. If it is and this > > > behavior is relied on then perhaps `strscpy_pad` is the preferred > > > option here. > > > > > > Kees was able to help me out and identify the following code snippet > > > which seems to show that the destination buffer is zero-initialized. > > > > > > | skl = devm_kzalloc(&pci->dev, sizeof(*skl), GFP_KERNEL); > > > > > > With this information, I opted for `strscpy` since padding is seemingly > > > not required. > > > > We did notice that str_elem->string is 44 bytes, but > > skl->lib_info[ref_count].name is 128 bytes. If str_elem->string isn't > > NUL-terminated, this can still hit an over-read condition (though > > CONFIG_FORTIFY_SOURCE would have caught it both before with strncpy() > > and now with strscpy()). So I assume it is expected to be > > NUL-terminated? > > > > Yes it is a filename of additional library which can be loaded, topology > UAPI only allows for passing 44 bytes long strings per string token (see > snd_soc_tplg_vendor_array -> union -> string flex array -> > snd_soc_tplg_vendor_string_elem -> SNDRV_CTL_ELEM_ID_NAME_MAXLEN), so we Thanks for the details! And just to confirm, these are (expected to be) NUL-terminated? > could also change length of > skl->lib_info[ref_count].name and potentially save few bytes. And looking at > it again I also think that we should not copy destination size number of > bytes, by which I mean ARRAY_SIZE(skl->lib_info[ref_count].name), which is > 128 in this case... so either need to change destination buffer size to be > same as topology field or calculate it differently. If the source is NUL-terminated, it's fine as-is. (And CONFIG_FORTIFY_SOURCE will catch problems if not.) -- Kees Cook