Integrate it all with build

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

Simplify build of poc-code

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

Simplify even further. EncContext tracks the location of the buffer
and doesn't care about tmp_len

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

fix PoC to handle buffer data alignment

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

Add libnss iovec PoC

this is a small modified version of the code example available here:

https://developer.mozilla.org/en-US/docs/Mozilla/Projects/NSS/nss_sample_code/NSS_Sample_Code_sample2

to show that it is possible to encrypt from multiple src buffers
into a final buffer.

this approach will allow us to remove a lot of memcpy around the
recv/send thread and replace them with iovec mapping all across.

Possible side effects:
- less memory consumption (mapping vs allocating)
- more complex code / crypto (internal) API changes

Add poc-code directory and move access-list code from Chrissie there

Over time we have created a lot of small PoC code to make knet better.
Most of it is already integrated but other chunks are still floating around.

I prefer to keep them here so that:

1) they don't get lost
2) it will be easier to integrate them once we get to that point
3) we can share PoC and work on them together

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

Add notes for package maintainers

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

[build] use rpm bcond for a more elegant solution to override ./configure defaults

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

[build] fix spec/srpm/rpm build

build only packages that we are configuring for (--enable-libtap,
--enable-kronosnetd) and have spec file depends on the latest
configure run to regenerate the info

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

[build] fix make distcheck to include kronostned/libtap coverage

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

[build] fix conditional build to pass make distcheck

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

[tests] fix return code checking

spotted by gcc 5.3.1

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

Merge pull request #15 from chrissie-c/master

config: Make libtap & kronosnetd optional

Make sure we build libtap if kronosnetd is also being built

I didn't think it worth checking for and using local libtap,
if anybody else wants that I will take patches.

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

config: Make libtap & kronosnetd optional

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

Merge pull request #14 from jfriesse/include-compat-h

[libknet] Really use compat when needed

[libknet] Really use compat when needed

Previous patch added compatatibility layer for systems without
(recv|send)mmsg but sadly forgot include compat.h when needed.

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

Merge pull request #13 from jfriesse/genericmmsg

[libknet] Support systems without (send|recv)mmsg

[libknet] Support systems without (send|recv)mmsg

Implemented are generic (slightly reduced functionality) functions and
for systems with syscall available but libc prototypes unavailable
also syscall wrapper.

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

[send/recv] err is never <0 in that code path

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

[listener] fix listener sock configuration to match other sockets

add send buffer and avoid EAGAIN issues on high load by delegating
buffering to the kernel

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

Merge pull request #12 from jfriesse/lib64

[build] Make sure lib64 is used when available

Merge pull request #11 from jfriesse/donotoverridesyncindex

[libknet] Do not override symbols

Merge pull request #10 from jfriesse/removeprlimit

[libknet] Use getrlimit instead of prlimit

[build] Make sure lib64 is used when available

Default location for libraries is PREFIX/usr/lib. This doesn't work well
on systems with mixed 64 and 32-bit libs.

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

[libknet] Do not override symbols

sync and index are both stdlib symbols so it's better not to use them as
function arguments.

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

[libknet] Use getrlimit instead of prlimit

prlimit is Linux specific. Because passed pid is 0, we can replace
prlimit call by portable getrlimit without loose of functionality.

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

[tests] Enable pmtud setfreq test in helgrind context

update helgrind suppression file

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

[PMTUd] rewrite thread to follow the same footsteps of heartbeat thread

drop all those notification mechanisms that are a locking nightmare
with very little to none gain and use a loop similar to heartbeat.

Also break down the code in 3 sections:
- main thread loop
- pmtud data gathering per link
- pmtud data gathering for all hosts/links

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

[threads] move thread timer resolution to common section

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

[tests] move all API checks makefile bits to a separate file

prepare for int(ernal)_ and fun(ctional)_ tests

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

[tests] drop normal info and leave only progress/warnings/errors

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

[tests] Add knet_handle_enable_pmtud_notify tests

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

[tests] remove unnecessary host_remove

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

[tests] Add knet_handle_pmtud_get tests

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

[PMTUd] Fix knet_handle_pmtud_get API sanity checks

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

[tests] Add knet_handle_pmtud_getfreq tests

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

[PMTUd] Fix getfreq API check

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

Revert "[tests] enable memcheck and helgrind checks by default"

This reverts commit 7d3ca6729c224c6a592a71ba8813f3d5b811912c.

[tests] enable memcheck and helgrind checks by default

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

[tests] add api_knet_handle_pmtud_setfreq tests

currently not enabled for helgrind checks. PMTUd thread needs a good rewrite first
but the API is fine.

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

[tests] expand knet_send for basic send checks

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

[tests] add api_knet_send_sync tests and update helgrind suppression file

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

[send/recv] Fix checks for knet_send_sync and move sync test earlier

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

[tests] print warnings for tests with exception enabled

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

[tests] export check type to all tests

This allows to skip/warn on some tests while we fix the PMTUd thread

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

[tests] remove stray line

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

[tests] add knet_send tests

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

[handle] Allow threads to recognized shutdown before stopping them.

On busy threads and shutdown, there is a chance that the thread
didn't have time to process the shutdown requests and even tho
we are closing down, some locks can be destroyed while still locked.

In theory we could have each thread report that it has ack'ed the shutdown
and then close, but a sleep seems to be more than good enough for now.

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

[send/recv] reworking locking on send to link

protect reading of datafd when processing outgoing data
and avoid unnecessary double locking in _handle_send_to_links
(calling functions both uses locking already).

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

[handle] expand knet_send input sanity checks

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

[tests] fix ping_test to deal with knet_recv API fixes

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

[tests] fix memcpm test and typos

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

[tests] Add knet_recv tests

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

[handle] fix checks around knet_recv

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

[tests] add knet_handle_get_datafd tests

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

[tests] add knet_handle_get_channel tests

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

[tests] add knet_handle_remove_datafd tests

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

[tests] add knet_handle_add_datafd tests

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

[tests] drop suprious cleanup entry

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

[tests] add knet_handle_enable_sock_notify tests

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

[tests] add knet_handle_enable_filter tests

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

[tests] add knet_host_enable_status_change_notify tests

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

[tests] add knet_host_get_status tests

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

[tests] add knet_link_get_status tests

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

[tests] re-order test executions based on class

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

[tests] align coding style and error checking

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

[tests] add knet_handle_setfwd tests

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

[crypto] fix nss init/shutdown memory leaks

make nss_init more robust across invokations

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

[build] separate memcheck and helgrind suppressions files

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

[tests] add knet_handle_crypto tests

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

[crypto] fix knet_handle_crypto config checks

remove redundant and duplicated code from nsscrypto

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

[tests] fix out-of-tree path to valgrind suppression file

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

[build] make clang happy

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

[crypto] drop redudant checks

spotted by clang

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

[build] make clang happy

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

[handle] Fix check for malloc result

spotted by clang

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

[build] fix configure.ac to deal with clang return codes being weird

when testing for an option alone clang will return a warning and RC=0
unless -Werror is also specified, and that warning switch to RC=1.

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

[tests] fix missing check for api_knet_link_get_priority

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

[tests] add knet_link_get_link_list tests

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

[tests] add knet_link_get_enable tests and fix a potential segfault

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

[tests] fixup knet_link_set_enable tests

add disable code checking and silence a helgrind warning.

helgrind complains about the heartbeat code writing back to the link
status vs knet_host_add.

knet_host_add is using an wr lock, heartbeat code a read lock.

the heartbeat code is the only one changing two ping timers values.
nothing else in the code is using/changing them.

knet_host_add does a memset at alloc time.

tho, it's impossible to hit the HB code _before_ or _during_ knet_host_add,
therefor it's safe to silence the warning.

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

[tests] add knet_link_set_enable tests

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

[tests] add knet_link_get_priority tests

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

[tests] add knet_link_set_priority tests

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

[tests] add progress message

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

[build] chose between the lesser of 2 evils

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

[tests] rename all tests to add postfix _test

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

[link] rename knet_link_get_timeout to knet_link_get_ping_timers and add related tests

Also in the process, make sure to init timers when configuring a link
and export link defaults in libknet.h

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

[link] rename knet_link_set_timeout to knet_link_set_ping_timers and related tests

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

[tests] add knet_link_get_pong_count tests

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

[tests] fix knet_link_set_pong_count check

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

[tests] Add knet_link_set_pong_count tests

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

[tests] remove unused variable

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

[link] fix knet_link_get_config API and add related tests

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

[libtap] make testing of symbols more consistent with libknet

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

[build] be consistent in building libraries and use same export standards

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

[tests] make it consistent and easier to run check-memcheck and check-helgrind

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

[tests] add api_knet_link_set_config tests

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

[tests] don't sort output and use header order to display API calls

makes it easier to follow the logic

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