returns bytes when there is a non-application record pending.

I have no idea when/why this would even happen, but let's catch it and
make sure tor_tls_get_pending_bytes stays correct.


svn:r1727

svn:r1726

fix log messages


svn:r1725

svn:r1724

svn:r1723

svn:r1722

svn:r1721

svn:r1720

svn:r1719

svn:r1718

crypto_perror is a no-no, since an operation can set more than one
error.

Also, fix a bug in the unix crypto_seed_rng: mixing stdio with
/dev/urandom is a bad idea, since fopen can make all kinds of weird
extraneous syscalls (mmap, fcntl, stat64, etc.) and since fread tends
to buffer data in big chunks, thus depleting the entropy pool.


svn:r1717

svn:r1716

svn:r1714

svn:r1713

svn:r1712

just close the circ.

(this wasn't relevant before, because we were mis-handling destroys.)


svn:r1711

when passing back a destroy cell, we would use the wrong circ id.
how the heck did this work?


svn:r1710

Call tls_log_errors at a more appropriate location; we can remove the other calls in tor_tls_verify once we are sure they never happen.


svn:r1709

svn:r1708

svn:r1707

svn:r1706

svn:r1705

svn:r1704

svn:r1703

svn:r1702

svn:r1701

svn:r1700

svn:r1699

svn:r1698

remove obsolete BUF_OK macro


svn:r1697

but don't use tor_assert inside log.c, to avoid loops


svn:r1696

svn:r1695

svn:r1694

Remove onion_pkey from connection, since onion keys can change more often than connections. Also add more log messages


svn:r1693

svn:r1692

svn:r1691

svn:r1690

svn:r1689

svn:r1688

svn:r1687