create a draft, so I can test the Windows builds.

with these changes - just downloading and testing on windows now...

just tested the windows VS-2019 build, and it seems to work correctly, and not manifest the issues reported earlier.

Thanks

This should probably be called only if ret == 0. Because if ret < 0 that would cause use-after-free errors, when the caller function calls dnssd_free_all_discovery_data(ddata).

I think (but am happy to be corrected), as long as d or ddata points to zero'd memory (which it does via zalloc) it should be fine to call free (null pointer). Correct?

I think (but am happy to be corrected), as long as d or ddata points to zero'd memory (which it does via zalloc) it should be fine to call free (null pointer). Correct?

It is fine to call free(NULL), but this is not what would happen here. After dnssd_free_all_discovery_data(ddata) is called, the ddata pointer is invalid and not NULL, so a second call to dnssd_free_all_discovery_data() would cause use-after-free.

dnssd_free_all_discovery_data(ddata) is called

yeah - sorry - I missed that I had not removed all the dnssd_free_all_discovery_data(ddata) calls. I was commenting on what I thought the code was - not what I had actually done... Sorry.

-Robin

Note that in your new commit (and actually in libiio master as well), if open_client_sockets() returns 0, the ddata is never freed. I think it would make sense to make open_client_sockets() return an error instead of 0 if it couldn't open any socket.

if open_client_sockets() returns 0, the ddata is never freed.

dns_sd.c:dnssd_context_scan() calls dnssd_find_hosts()

If open_client_sockets() returns 0, it drops down to out_free_buffer; which will return with a 0 code.

dns_sd.c:dnssd_context_scan() will try to iterate on things:
for (ndata = ddata; ndata->next != NULL; ndata = ndata->next)

and since it is all null, (via zalloc), it should drop down to dnssd_free_all_discovery_data(ddata);

shouldn't it?

I guess you are right, yes.