Commit Graph

1005 Commits

Author SHA1 Message Date
Joris Vink 5d16a7a123 make sure stdint.h is always included.
unbreaks NOTLS=1 and NOSENDFILE=1 builds on several platforms.

reported by Sam on irc, thanks.
2019-03-29 19:25:27 +01:00
Joris Vink a66d259271 kill debug line 2019-03-29 19:15:10 +01:00
Joris Vink d2aa64df5c add kore_proctitle().
manipulates the argv+environ pointers to get a sensible process title
under linux / darwin.
2019-03-29 16:24:14 +01:00
Joris Vink 92fb4974b1 only deschedule if not -1. 2019-03-27 22:15:35 +01:00
Joris Vink 39ffa047ca remove dh parameter check at configure time.
this is done when the workers startup if a domain with tls was used.
2019-03-26 21:44:44 +01:00
Joris Vink e4caac9e0c Be more clear when an invalid TLS setup is found 2019-03-26 21:36:45 +01:00
Joris Vink bf42e56c7d kill whitespace 2019-03-26 20:24:46 +01:00
Joris Vink 4ca7f29649 Add a concurrency parameter to kore.gather() 2019-03-25 10:13:52 +01:00
Joris Vink e1766e74ba always capture worker processes exiting.
even if they terminated normally.
2019-03-22 10:29:14 +01:00
Joris Vink 4238431b9e Add worker_death_policy setting.
By default kore will restart worker processes if they terminate
unexpected. However in certain scenarios you may want to bring down
an entire kore instance if a worker process fails.

By setting worker_death_policy to "terminate" the Kore server will
completely stop if a worker exits unexpected.
2019-03-22 09:49:50 +01:00
Joris Vink cd80685d9d get rid of pyko 2019-03-21 21:37:16 +01:00
Joris Vink 58c2739dee If fd is -1 in pysock_op_iternext just return None.
None in pysock signals EOF and if it was closed under our feet we should
relay that properly.
2019-03-21 14:44:41 +01:00
Joris Vink 370041656e Get rid of WORKER_LOCK_TIMEOUT.
Instead let the workers send a message on the msg channel to each
other when they have given up the accept lock and it is now available
to be grabbed.
2019-03-21 14:03:11 +01:00
Joris Vink 8b0279879a rework timers so they fire more predictably.
this change also stops python coroutines from waking up very
late after their timeout has expired.

in filerefs, don't prime the timer until we actually have something
to expire, and kill the timer when the last ref drops.
2019-03-21 10:17:08 +01:00
Joris Vink ce012e7bd5 set fd to -1 on close. 2019-03-21 10:09:24 +01:00
Joris Vink d371454dad do not grab the result if it was in a gatherop 2019-03-19 13:58:00 +01:00
Joris Vink 8afaac9fde log pid number of worker upon start. 2019-03-19 13:42:22 +01:00
Joris Vink 681a88d535 simplify af_unix recvfrom/sendto support. 2019-03-19 13:27:50 +01:00
Erik Karlsson d60994abb5 add python sendto recvfrom abstract socket support 2019-03-19 11:11:44 +01:00
Joris Vink 9caa45a050 Allow python validator methods to be async. 2019-03-18 09:34:31 +01:00
Joris Vink c3ab570f56 Append full module path if it is a directory. 2019-03-16 16:13:52 +01:00
Joris Vink ec7c8b8e1d fix a few minor problems with pylock.
- when pylocks are deallocated we ended up deleting the wrong object.
- do not call Py_DECREF on the op when inside the op its iternext call.
2019-03-14 17:31:57 +01:00
Joris Vink 2217c7a2c8 pysock: call disable read/write only when needed 2019-03-14 09:27:13 +01:00
Joris Vink d41054bd26 remove the socket from the event queue on dealloc. 2019-03-13 16:01:42 +01:00
Joris Vink b06160c768 Make sure coros are woken up on eof. 2019-03-13 15:45:55 +01:00
Joris Vink 3b4574d791 Rework pysocket async/await.
Attach the events directly to the pysocket data structure instead of
one event per pysocket_op.

Makes the code easier, gives us a good performance boost and reduces
the number of system calls required when doing an await on a socket.
2019-03-13 11:07:15 +01:00
Joris Vink 01f9b4fcde Check for EAGAIN when calling connect(2).
Linux does this under certain conditions.
2019-03-12 20:04:08 +01:00
Joris Vink eb9b7f7b14 explicitly include sys/types.h
some smaller libc variants do not include this from sys/param.h.
2019-03-06 09:29:46 +01:00
Joris Vink 61863bfd3a kill recv() timer if we have data.
also reset any exception that is set when we have successfully
read data from a socket.
2019-03-04 16:37:25 +01:00
Joris Vink a0c203f507 remove norwegian debug 2019-03-01 20:57:31 +01:00
Joris Vink 31a9a70d5a cleanup. 2019-03-01 20:56:03 +01:00
Erik Karlsson 01c1a8f8f8 support AF_UNIX in recvfrom()/sendto() python sockets 2019-03-01 20:13:28 +01:00
Joris Vink 2057f4db58 sprinkle more kore_quiet around 2019-02-27 19:59:31 +01:00
Joris Vink 1ebd82969c Add timeout support to proc.recv() 2019-02-26 15:22:55 +01:00
Joris Vink f4cd70956b Add an optional timeout to socketop.recv(). 2019-02-25 10:35:00 +01:00
Joris Vink bf1e8e5ffb bump copyright to 2019 2019-02-22 16:57:28 +01:00
Joris Vink 3dbb80463a allow double quotes in kore.proc argument lists. 2019-02-22 16:56:34 +01:00
Joris Vink 66305e551a sig_recv is extern'd in kore.h 2019-02-11 14:01:25 +01:00
Joris Vink 8a8b08f1e6 use %u when logging worker_rlimit_nofiles 2019-02-11 14:00:28 +01:00
Joris Vink 429159ee3c reprime idle timer if SSL_accept() fails.
reported by Daniel Chavez (@dacechavez)
2019-02-08 15:40:00 +01:00
Joris Vink 8040103458 fix for multiple consumers for the async queues. 2019-01-22 11:23:59 +01:00
Joris Vink 39467847fb remove SSL_get_verify_result() check.
If peer verification is turned on this becomes part of the handshake
process anyway and SSL_accept() will fail when appropriate.
2019-01-21 10:36:50 +01:00
Joris Vink 3f083d6126 deal with crls being expired / not-yet-valid.
if a crl is expired or not-yet-valid SSL_get_verify_result()
will return these errors too so check for them explicitly
instead of depending on X509_V_OK.

found by @dacechavez
2019-01-19 11:49:54 +01:00
Joris Vink d1e87c1a54 deal with PyObject_CallObject() returning NULL. 2019-01-15 10:20:13 +01:00
Joris Vink d6b05bcff7 always force reload cert so we get a new x509 store.
Otherwise older OpenSSL or current LibreSSL will fail to add the new
CRL as they still match on subject name rather then hash of the CRL data.
2019-01-14 20:57:40 +01:00
Joris Vink 73cdbd1a01 Let CRLs be reloadable via keymgr.
With these changes CRLs can be reloaded like certificates
by sending a SIGUSR1 to the parent process.

Track mtime on both certificate files and CRL files as well
and only submit them to the workers if this has changed.
2019-01-14 11:41:50 +01:00
Joris Vink b07cc0237c Support recvfrom()/sendto() on kore python sockets. 2019-01-08 17:49:00 +01:00
Joris Vink a868ff2b25 wrap accesslog calls in !KORE_NO_HTTP. 2018-12-22 09:41:55 +01:00
Joris Vink 9aa0e95643 Rework accesslog handling.
Move away from the parent constantly hitting the disk for every
accesslog the workers are sending.

The workers will now write their own accesslogs to shared
memory before the parent will pick those up. The parent
will flush them to disk once every second or if they grow
larger then 1MB.

This removes the heavy penalty for having access logs
turned on when you are dealing with a large volume
of requests.
2018-12-22 09:25:00 +01:00
Joris Vink 0d6a188b01 close fd if we're a directory in filemap. 2018-12-17 16:22:36 +01:00