[nozzle] cleanup and document nozzle_del_ip

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] cleanup and document nozzle_add_ip

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] re-order functions in order of importance

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] cleanup and document nozzle_set_down

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] cleanup and document nozzle_set_up

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] start to cleanup nozzle_close to be more informative and effective

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] cleanup nozzle_open coding style and return codes

also update the header file for Doxygen man pages

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] rename last tap bits to nozzle

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] rename API from tap_ to nozzle_

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] fix pkg-config to point to nozzle

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] rename struct _iface to nozzle_iface (looks better on public headers)

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] rename tap_t to nozzle_t

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] include net/if.h to define IFNAMSIZ instead of redefine

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[nozzle] add Doxygen header

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libnozzle] rename libtap to libnozzle (part 1)

Our fearless Debian maintainer wferi made us aware that TAP is an overloaded
term already and it would be very confusing to add yet another one.

nozzle is a good pick from http://www.thesaurus.com/browse/tap

I am sure lib*cock* would have cause havoc :-)

This commit only rename files and allow builds. The API needs to be renamed too,
but that will come over time with major code cleanup as documented here:
https://trello.com/c/pyEKd4UH/396-libnozzle-make-it-production-ga-ready

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libtap] BSD: switch to ioctl to destroy tap devices

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libtap] define IFNAMSIZ for applications that don't need/want to include net/if.h

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libtap] update README for BSD

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libtap] port to freebsd and fix a bunch of small bugs

- tap_open: do mutex lock only after malloc to avoid deadlock
- fix test suite to increase randomness when testing named tap devices
- remove hardcoded paths to /bin/true /bin/false and /bin/grep

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libtap] tests: add better IPv6 private address space randomization

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[libtap] tests: allow random localhost IPs for testing

- avoids clashing with machine routing
- allows parallel testing

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

UDP: Cope with EPERM returned from sendmsg/sendto

IPtables can cause -EPERM to be returned from socket sends, so
we need to be able to handle it without going into a spin.

SCTP seems to be unaffected.

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>

[misc] fix a few typos/stylistics

Signed-off-by: Jan Pokorný <jpokorny@redhat.com>

[tests] update error message in api_knet_{addrtostr,strtoaddr}.c

Signed-off-by: Chen Jingpiao <jchen@suse.com>

[tests] fix check condition in api_knet_{addrtostr,strtoaddr}.c

Signed-off-by: Chen Jingpiao <jchen@suse.com>

clear the errno in each exported API when the call success

Signed-off-by: Chen Jingpiao <jchen@suse.com>

[openssl] fix return type for pthread_self

spotted on freebsd-devel

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tx] fix tx_uncompressed_packets stats collect in _parse_recv_from_sock

Signed-off-by: Chen Jingpiao <jchen@suse.com>

[rx] fix incorrect packet length in _parse_recv_from_links

Signed-off-by: Chen Jingpiao <jchen@suse.com>

[tx] remove goto statement to improve readability

Signed-off-by: Chen Jingpiao <jchen@suse.com>

Add static qualifier to {compress, crypto, transport}_modules_cmds

Signed-off-by: Chen Jingpiao <jchen@suse.com>

[transports] fix transport_modules_cmd array out of range

Fixes: 18ab4888
Signed-off-by: Chen Jingpiao <jchen@suse.com>

[build] fix supported compiler warning detection

move from AC_PREPROC_IFELSE (strongly discouraged) to AC_COMPILE_IFELSE

our detection system was very weak and recent versions of clang did
show that PREPROC_IFELFE (cpp) would enable warning options that
the compiler does not support (clang).

use a full compilation test to detect what works and what doesn't.

Also expand the warning list to include new / renamed clang options
of equivalents already enabled for older versions of clang and gcc.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

tests: fix output message

Fixes: 04537c57 ("[transports] cleanup API calls")
Signed-off-by: Chen Jingpiao <chenjingpiao@gmail.com>

fix typo

Signed-off-by: Chen Jingpiao <chenjingpiao@gmail.com>

man: update XML directory in doxyxml.c

Fixes: 31945b18 ("[man] move libknet man pages and man pages build tools to top level dir")
Signed-off-by: Chen Jingpiao <chenjingpiao@gmail.com>

misc: keep covscan happy

covscan complains about is ignoring returns from pthread_mutex_lock()
but in those 2 circumstances we don't have much choice.

This patch just keeps coverity quiet so we don't miss any real bugs it
might find.

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>

Merge pull request #155 from kronosnet/stable1-proposed

Stable1 proposed

links: Don't close loopback link

knet_link_clear_config would incorrectly clear the local
loopback link if the last 'real' link to an external node
was closed. We didn't check the nodeid in the code, just
the link number when clearing the loopback flag.

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>

Import key with length not dividable by wrap key block size

Wrapping of the key is standard crypto operation which needs data
aligned to cipher block size, otherwise it fails.

Possible solution is to use a zero filled buffer with size aligned to
required wrap key block size. Private key is copied to the beginning of
the buffer and unwrap operation keeps using only required private key
size.

Signed-off-by: Jan Friesse <jfriesse@redhat.com>

[tests] clear errno to avoid incorrect propagation from void fn call

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] fix select(2) handling (mostly affects slow machines)

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[threads] make it easier to debug thread startup / stop operations with pretty names

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

Use gitarchive-version for spec file

Signed-off-by: Jan Friesse <jfriesse@redhat.com>

git-version-gen fail on UNKNOWN version

Signed-off-by: Jan Friesse <jfriesse@redhat.com>

Do not match lightweight tags in git-version-gen

Fix leftover from 175b517fd71d999150aa701ec847510490a18a6e which added
matching of lightweight git tags.

Signed-off-by: Jan Friesse <jfriesse@redhat.com>

Support for git archive stored tags

Attempt to solve problem with git archive generated tarballs
(used for example by github when release is downloaded) which are no
longer git tree and (in contrast to officially released tarballs) also
doesn't contain .tarball-version file so git-version-gen script simply
cannot obtain valid version info.

Solution is based on using gitattributes which is instructs git to
replace string in the .gitarchivever file by known ref names.
git-version-gen is enhanced to support this file and tries to parse
any string which looks like "tag: v[0-9][^,)]*". If such string
is found it's used as a version. This file is used as a last attempt and
other methods (.tarball-version, git abbrev) have precedence.

Based on idea stated by Jan Pokorný <jpokorny@redhat.com>.

Signed-off-by: Jan Friesse <jfriesse@redhat.com>

[tests] drop arm exceptions as they have been fixed in glibc

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] cleanup nss memcheck supports

add one more exception for newer nss (spotted on fedora rawhide)
drop FreeBSD excetions as we don't support running under valgrind on BSD

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

Merge pull request #150 from kronosnet/stable1-proposed

stable1-proposed

[threads] use internal array to track status of threads (running/stopped) vs a random sleep()

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] don't redefine symbols from the main library

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] adds more overrides for fedora28/arm/glibc

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[handle] reduce usage of handle_config_mutex to wrap only global lib init

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[logging] report error in case logging buffer is bigger than PIPE_BUF

See https://github.com/kronosnet/kronosnet/pull/148

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[handle] drop init_lock_done as it was used only by locked version of log_msg

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[logging] drop locking inside log_msg (part 1)

looking at the code, there are only few calls to log_msg happening outside
of locking context (error reports when failing to grab the lock).
Everything else appears to be safe.

drop the attempt to lock inside log_msg for now, part 2 will address those
few calls that can happen outside of locking context.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[handle] drop call to log_msg that would have no effect

logging is not setup yet at this stage and we can't log just yet.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] ignore nss internal memory leak (3.38+) non recurring

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] reference correct library in memcheck comments

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tests] add another lzma exception

spotted on Debian Experimental, the latest toolchain appears to add
full path to objects and valgrind would generate an error.

add a generic catch-all path exception.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

coding style: add parentheses after the sizeof operator

Signed-off-by: Ferenc Wágner <wferi@debian.org>

log_msg: no point in putting NL after the final NUL

Signed-off-by: Ferenc Wágner <wferi@debian.org>

knet_log_msg: fix documentation: the closing newline is added by flush_logs

Signed-off-by: Ferenc Wágner <wferi@debian.org>

_logthread: while loop instead of goto

Signed-off-by: Ferenc Wágner <wferi@debian.org>

_logthread: select() returns the number of set bits, as an int

Signed-off-by: Ferenc Wágner <wferi@debian.org>

flush_logs: don't ever print past the message buffer

Signed-off-by: Ferenc Wágner <wferi@debian.org>

flush_logs: while loop instead of goto

Signed-off-by: Ferenc Wágner <wferi@debian.org>

flush_logs: a for loop is more straightforward here

Signed-off-by: Ferenc Wágner <wferi@debian.org>

flush_logs: don't repeat the type of msg

Signed-off-by: Ferenc Wágner <wferi@debian.org>

flush_logs: the reads will overwrite the entire msg structure

Signed-off-by: Ferenc Wágner <wferi@debian.org>

flush_logs: remove redundant condition

The read loop is always entered, resetting len and returning from the
function if it isn't greater than 0.

Signed-off-by: Ferenc Wágner <wferi@debian.org>

Merge pull request #145 from kronosnet/stable1-proposed

stable1-proposed

[crypto] NSS_NoDB_Init: the parameter is reserved, must be NULL

Thanks to Feri for spotting it in corosync

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[crypto] improve checks around RX packet size

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

Merge pull request #140 from kronosnet/stable1-proposed

stable1-proposed

build: look for source_epoch in srcdir

Signed-off-by: Ferenc Wágner <wferi@debian.org>

build: check for SOURCE_DATE_EPOCH before turning to Git for an epoch

Signed-off-by: Ferenc Wágner <wferi@debian.org>

[build] fail and provide proper error if source tree version is unknown

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>
Signed-off-by: Ferenc Wágner <wferi@debian.org>

build: git -C is a relatively new option, avoid it

Signed-off-by: Ferenc Wágner <wferi@debian.org>

build: accomodate to the actual date utility

Signed-off-by: Ferenc Wágner <wferi@debian.org>

spec: use the source epoch here as well

Signed-off-by: Ferenc Wágner <wferi@debian.org>

man: use new doxyxml date options to build man pages

Based on Fabio's original patch.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>
Signed-off-by: Ferenc Wágner <wferi@debian.org>

man: Add date & year options to doxyxml

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>

build: distribute the source_epoch file

Signed-off-by: Ferenc Wágner <wferi@debian.org>

build: determine SOURCE_EPOCH during configuration

Signed-off-by: Ferenc Wágner <wferi@debian.org>

[compress] extend check to compare end buffer data

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[compress] update valgrind exception file to match new lzma entry points

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[compress] delegate compress_level validation to compress libraries

- make val_level internal api call optional
- keep lzo2 val_level around due to the specific nature of compress_level values
- add internal compress_lib_test to do a round-robin (compress/decompress) check with provided values
- drop unnecessary val_level checks around
- update internal compress API include file
- adjust api_knet_handle_compress test to use a known bad value for zlib validation

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[tx/rx] improve error checking when running external code

libknet cannot guarantee that the dst_host_filter is sane nor that it cannot
be exploited to return garbage to knet.

there is an infinitesimal possibility that, if the plugin returns total crap,
knet could crash by trying to access out-of-bound memory.

Prevent that by adding specific checks both in TX/RX.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

Merge pull request #116 from kronosnet/stable1-proposed

stable1-proposed fixes

[build] fix ChangeLog generation in release tarball

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

Indicate new API call in libtool version info

Signed-off-by: Ferenc Wágner <wferi@debian.org>

build: provide -version-info directly instead of -version-number

The info components carry the real OS-independent library version
information, from which the version number is computed differently
on different systems. According to the libtool manual "new projects
should use the -version-info flag instead".

Signed-off-by: Ferenc Wágner <wferi@debian.org>

New API function: knet_handle_new_ex() taking handle flags

For now, the only supported flag is KNET_HANDLE_FLAG_PRIVILEGED,
which is also the default in knet_handle_new().

The tests are adapted to switch this flag off, which makes them pass
in default environments without privileges.

Signed-off-by: Ferenc Wágner <wferi@debian.org>

UDP: Don't keep retrying send if we get EINVAL

UDP can return EINVAL if the route is down or not configured, this can
cause knet to 'spin' (well, ish, there is a usleep in there).

With this patch in place knet doesn't hog the CPU anywhere near so much
(and it does recover if the link returns).

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>

doxyxml: Use overridden XML_DIR for reading structures

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>

[man] fix path to stamp file

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[man] move kronosnetd man pages

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

[man] move libknet man pages and man pages build tools to top level dir

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>