[gnutls-devel] GnuTLS | gnutls examples don't handle short writes correctly (#1484)

[Read-only notification of GnuTLS library development activities]

Alexander Sosedkin created an issue: https://gitlab.com/gnutls/gnutls/-/issues/1484



I see the following pattern in examples (copy-pasted from [doc/examples/ex-client-anon.c](https://gitlab.com/gnutls/gnutls/-/blob/master/doc/examples/ex-client-anon.c#L20)):

```c
#define LOOP_CHECK(rval, cmd) \
        do { \
                rval = cmd; \
        } while(rval == GNUTLS_E_AGAIN || rval == GNUTLS_E_INTERRUPTED); \
        assert(rval >= 0)
#define MAX_BUF 1024
#define MSG "GET / HTTP/1.0\r\n\r\n"
...
LOOP_CHECK(ret, gnutls_record_send(session, MSG, strlen(MSG)));  // no other loop, no nothing
```

If gnutls_record_send does a "short write", writing the buffer out only partially and returning less than `strlen(MSG)`, the request won't be sent in a complete form and the example would hang.
The problematic short write handling holds true for other examples, say, echo server.

I believe we should fix the examples to handle short writes correctly, ideally in a coordinated, uniform manner.

Alternative options include:
1. Modify `gnutls_record_send` to handle short writes correctly. This sounds like it even preserves the API guarantees, but I'm not sure it won't change behaviour non-trivially when non-blocking sockets are used, say, by introducing a performance regression.
2. Implement additional, simplified version of `gnutls_record_send` that'd handle short writes transparently, possibly also `E_AGAIN`/`E_INTERRUPTED`.

-- 
Reply to this email directly or view it on GitLab: https://gitlab.com/gnutls/gnutls/-/issues/1484
You're receiving this email because of your account on gitlab.com.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.gnupg.org/pipermail/gnutls-devel/attachments/20230522/8b6029ec/attachment-0001.html>