Arran Cudbard-Bell [Wed, 7 May 2014 23:17:43 +0000 (00:17 +0100)]
Cleanup connection properly on error
Arran Cudbard-Bell [Wed, 7 May 2014 23:05:56 +0000 (00:05 +0100)]
Various minor PG fixes
Alan T. DeKok [Wed, 7 May 2014 20:24:52 +0000 (16:24 -0400)]
note recent changes
Alan T. DeKok [Wed, 7 May 2014 20:08:32 +0000 (16:08 -0400)]
Disabled debug checks
Alan T. DeKok [Wed, 7 May 2014 20:08:11 +0000 (16:08 -0400)]
Simplified sanity checks
Alan T. DeKok [Wed, 7 May 2014 20:05:45 +0000 (16:05 -0400)]
Spawn connections "in_use" or not.
No more double uses of the connections
Arran Cudbard-Bell [Wed, 7 May 2014 19:28:25 +0000 (20:28 +0100)]
Move PQ ssl init to instantiation function, just in case it needs to be done before we spawn threads
Arran Cudbard-Bell [Wed, 7 May 2014 18:58:16 +0000 (19:58 +0100)]
More SSL initialisation
Arran Cudbard-Bell [Wed, 7 May 2014 18:40:32 +0000 (19:40 +0100)]
Use talloc for row memory
Arran Cudbard-Bell [Wed, 7 May 2014 16:32:08 +0000 (17:32 +0100)]
Update ChangeLog
Arran Cudbard-Bell [Wed, 7 May 2014 15:14:40 +0000 (16:14 +0100)]
Anything which eventually has the value it got interpreted by pairparsevalue, or doesn't pass the length of the string around must not pre-unescape the string
This broke pretty much all escape sequences in preprocess, sql, ldap and radclient. Arguably as conffile.c is the only place which needs this, it should probably be the thing doing the unescaping.
No, I don't buy that things should be prefixed with \\, that's shit. The server should be able to reingest what it spits out in detail files and there's no way that worked in the current server with octal escape sequences.
Arran Cudbard-Bell [Wed, 7 May 2014 14:35:01 +0000 (15:35 +0100)]
Clear the error buffer after loading dictionaries
Arran Cudbard-Bell [Wed, 7 May 2014 14:31:22 +0000 (15:31 +0100)]
Remove duplicate test (list-delete) was duplicate test of update-remove-list
Arran Cudbard-Bell [Wed, 7 May 2014 14:28:47 +0000 (15:28 +0100)]
Typo
Alan T. DeKok [Wed, 7 May 2014 11:18:40 +0000 (07:18 -0400)]
Cap spawn at max
Alan T. DeKok [Tue, 6 May 2014 23:39:11 +0000 (19:39 -0400)]
Rework min/max/spare handling
So we don't let idle connections make us go below "min"
Alan T. DeKok [Tue, 6 May 2014 18:19:12 +0000 (14:19 -0400)]
Free handle if we can't spawn a thread. Found by codesonar
Arran Cudbard-Bell [Tue, 6 May 2014 22:41:55 +0000 (23:41 +0100)]
Merge pull request #625 from leprechau/v3.0.x
Make internal references match configuration references and update README
Aaron Hurt [Tue, 6 May 2014 15:57:12 +0000 (10:57 -0500)]
Make internal references match configuration references and update README.md to match.
Alan T. DeKok [Tue, 6 May 2014 15:55:41 +0000 (11:55 -0400)]
Close the FD if we can't fdopen it. Found by codesonar
Alan T. DeKok [Tue, 6 May 2014 15:53:29 +0000 (11:53 -0400)]
Close the FD if we can't fdopen it. Found by codesonar
Alan T. DeKok [Tue, 6 May 2014 15:24:27 +0000 (11:24 -0400)]
Tweak reconnect logic.
A reconnect means open new connection, or if that fails, try
to find an unused one. But don't spawn a new connection.
Also, we can only reconnect a used connection. It's an error
to reconnect an unused connection.
Alan T. DeKok [Tue, 6 May 2014 14:23:40 +0000 (10:23 -0400)]
You can only reconnect a handle if it's in-use by you
Alan T. DeKok [Tue, 6 May 2014 13:55:25 +0000 (09:55 -0400)]
Use pthread ID for more connection debugging. Helps debug #624
Alan T. DeKok [Tue, 6 May 2014 13:09:16 +0000 (09:09 -0400)]
Tests for list delete
Alan T. DeKok [Tue, 6 May 2014 12:30:36 +0000 (08:30 -0400)]
Query may be an empty string, too.
We should really fix that in the init function
Arran Cudbard-Bell [Tue, 6 May 2014 12:30:58 +0000 (13:30 +0100)]
Use consistent config item names in couchbase
Arran Cudbard-Bell [Tue, 6 May 2014 08:34:24 +0000 (09:34 +0100)]
Fix dereferencing NULL pointer in json_object_object_get_ex
Arran Cudbard-Bell [Tue, 6 May 2014 08:24:43 +0000 (09:24 +0100)]
Correct behaviour processing stops in rlm_couchbase (unintended fallthrough)
Arran Cudbard-Bell [Tue, 6 May 2014 08:18:18 +0000 (09:18 +0100)]
Check handle is not NULL before dereferencing it CID #1211838
Arran Cudbard-Bell [Mon, 5 May 2014 21:12:38 +0000 (22:12 +0100)]
Open query should default to NULL
Arran Cudbard-Bell [Mon, 5 May 2014 21:02:29 +0000 (22:02 +0100)]
Increase max nodes *sigh*
Arran Cudbard-Bell [Mon, 5 May 2014 20:58:23 +0000 (21:58 +0100)]
Don't continue to fail spawning connections if we failed once...
Arran Cudbard-Bell [Mon, 5 May 2014 19:09:14 +0000 (20:09 +0100)]
Don't need to check, it's marked as required
Arran Cudbard-Bell [Mon, 5 May 2014 09:58:00 +0000 (10:58 +0100)]
Ignore all.mk
Arran Cudbard-Bell [Mon, 5 May 2014 09:56:38 +0000 (10:56 +0100)]
Simpler way of redoing the delimiters (which also doesn't crash if the server item is commented out)
Arran Cudbard-Bell [Mon, 5 May 2014 09:24:38 +0000 (10:24 +0100)]
Config parsing is now handled by the server core, individual modules are no longer expected to do it, except in the case of special sections such as 'map'
Arran Cudbard-Bell [Mon, 5 May 2014 09:04:54 +0000 (10:04 +0100)]
Reduce doxy node depth further
Arran Cudbard-Bell [Mon, 5 May 2014 09:00:55 +0000 (10:00 +0100)]
Merge pull request #620 from leprechau/v3.0.x
import rlm_couchbase
Aaron Hurt [Mon, 5 May 2014 02:23:31 +0000 (21:23 -0500)]
import rlm_couchbase
Alan T. DeKok [Mon, 5 May 2014 00:01:55 +0000 (20:01 -0400)]
Document open_query
Arran Cudbard-Bell [Sun, 4 May 2014 23:54:47 +0000 (00:54 +0100)]
Remove obsolete declarations for new version of DoxyGen
Alan T. DeKok [Sun, 4 May 2014 23:54:43 +0000 (19:54 -0400)]
It's better as open_query
Alan T. DeKok [Sun, 4 May 2014 23:46:28 +0000 (19:46 -0400)]
Add "start_query" to set session-specific parameters
Alan T. DeKok [Sun, 4 May 2014 12:55:55 +0000 (08:55 -0400)]
Don't unlink on close. Fixes #621
Arran Cudbard-Bell [Sun, 4 May 2014 11:46:40 +0000 (12:46 +0100)]
Update ChangeLog
Arran Cudbard-Bell [Sun, 4 May 2014 11:42:40 +0000 (12:42 +0100)]
Update ChangeLog
Arran Cudbard-Bell [Sun, 4 May 2014 11:36:33 +0000 (12:36 +0100)]
Clients aren't talloced yet
Arran Cudbard-Bell [Sun, 4 May 2014 11:16:24 +0000 (12:16 +0100)]
Ignore more test products
Arran Cudbard-Bell [Sun, 4 May 2014 10:47:44 +0000 (11:47 +0100)]
Fix minor memory leaks in detail file reader
Arran Cudbard-Bell [Sun, 4 May 2014 08:50:12 +0000 (09:50 +0100)]
Avoid allocing *ANY* memory for backtraces.
Arran Cudbard-Bell [Sat, 3 May 2014 22:32:36 +0000 (23:32 +0100)]
Don't alloc any memory inside in fr_fault or functions called by fr_fault.
Alan T. DeKok [Sat, 3 May 2014 19:12:10 +0000 (15:12 -0400)]
Revert changes...
Network RADIUS [Sat, 3 May 2014 18:17:24 +0000 (13:17 -0500)]
Linux is crazy
Alan T. DeKok [Sat, 3 May 2014 18:00:14 +0000 (14:00 -0400)]
Apparently we need unlink?
Arran Cudbard-Bell [Sat, 3 May 2014 17:51:00 +0000 (18:51 +0100)]
Include NDEBUG state when printing out server features
Alan T. DeKok [Sat, 3 May 2014 16:38:25 +0000 (12:38 -0400)]
Final fix for previous commit
Arran Cudbard-Bell [Sat, 3 May 2014 16:35:23 +0000 (17:35 +0100)]
Doxygen
Alan T. DeKok [Sat, 3 May 2014 16:27:06 +0000 (12:27 -0400)]
Don't unlink the socket. Closes CID #720456
The pattern of unlinking before bind is so that we don't open
a socket owned by someone else. We already check that, so
unlinking serves no purpose.
Also, in close_socket(), unlink the file if it was a domain
socket. That way it doesn't stick around
Alan T. DeKok [Sat, 3 May 2014 16:13:32 +0000 (12:13 -0400)]
Parent is the special node NIL, not NULL. CID #1187991
Arran Cudbard-Bell [Sat, 3 May 2014 16:13:20 +0000 (17:13 +0100)]
Quiet gcc
Arran Cudbard-Bell [Sat, 3 May 2014 16:09:40 +0000 (17:09 +0100)]
Hack to temporarily enable dumpable flag, call panic_action, then disable it again (if required)
Arran Cudbard-Bell [Sat, 3 May 2014 11:49:30 +0000 (12:49 +0100)]
Only iterate over VPs if we have a request: CID #1211052
Arran Cudbard-Bell [Sat, 3 May 2014 11:43:27 +0000 (12:43 +0100)]
Remove assert (it was weird) - absence doesn't trigger complaint from clang scan
Arran Cudbard-Bell [Sat, 3 May 2014 11:35:18 +0000 (12:35 +0100)]
Make comparison functions use a generic comparator type
Add fr_quicksort to quicksort arrays using a comparator
Fix issue in radsniff, where the list of DICT_ATTRs needed to be sorted but weren't, meaning most of them would probably have been ignored
Arran Cudbard-Bell [Sat, 3 May 2014 10:34:53 +0000 (11:34 +0100)]
Remove unneeded checks in rlm_sql.c
Alan T. DeKok [Sat, 3 May 2014 01:12:20 +0000 (21:12 -0400)]
Some function calls are deprecated on OS X.
Arran Cudbard-Bell [Sat, 3 May 2014 00:02:38 +0000 (01:02 +0100)]
Set nonblock after calling connect, else it returns immediately
Arran Cudbard-Bell [Fri, 2 May 2014 23:21:39 +0000 (00:21 +0100)]
Should probably set non-blocking socket options before calling listen
Arran Cudbard-Bell [Fri, 2 May 2014 23:03:00 +0000 (00:03 +0100)]
Set nonblock on TCP client sockets
Arran Cudbard-Bell [Fri, 2 May 2014 18:35:53 +0000 (19:35 +0100)]
Revert "Use realloc to correct the length of char buffers in pairparsevalue"
This reverts commit
4a6511458ec66c4bb9ea14e061f37cc5cc6d2b0e.
Arran Cudbard-Bell [Fri, 2 May 2014 18:35:10 +0000 (19:35 +0100)]
It's ok for length to be <= octets buffer
Arran Cudbard-Bell [Fri, 2 May 2014 18:34:33 +0000 (19:34 +0100)]
Exit after asserting in fr_verify_vp
Arran Cudbard-Bell [Fri, 2 May 2014 18:30:33 +0000 (19:30 +0100)]
If there's no action set we should still need to check if the signal was SIGUSR1
Alan T. DeKok [Fri, 2 May 2014 18:23:27 +0000 (14:23 -0400)]
Remove useless panic action
Alan T. DeKok [Fri, 2 May 2014 18:19:22 +0000 (14:19 -0400)]
The attribute for Password-With-Header might be a string
Alan T. DeKok [Fri, 2 May 2014 18:18:38 +0000 (14:18 -0400)]
It's OK for the buffer to be too big.
Also print out the name of the offending attribute
Arran Cudbard-Bell [Fri, 2 May 2014 16:42:38 +0000 (17:42 +0100)]
sp
Arran Cudbard-Bell [Fri, 2 May 2014 16:40:26 +0000 (17:40 +0100)]
Use realloc to correct the length of char buffers in pairparsevalue
Arran Cudbard-Bell [Fri, 2 May 2014 16:39:58 +0000 (17:39 +0100)]
Formatting
Arran Cudbard-Bell [Fri, 2 May 2014 16:20:47 +0000 (17:20 +0100)]
Comments
Arran Cudbard-Bell [Fri, 2 May 2014 16:17:21 +0000 (17:17 +0100)]
turn VERIFY_VP into a function, the macro was getting very long
Alan T. DeKok [Fri, 2 May 2014 15:24:24 +0000 (11:24 -0400)]
Note recent changes
Alan T. DeKok [Fri, 2 May 2014 15:13:01 +0000 (11:13 -0400)]
A dup isn't a dup if the old request is done
Alan T. DeKok [Fri, 2 May 2014 14:58:40 +0000 (10:58 -0400)]
More tracing on dup / conflicting packet
Arran Cudbard-Bell [Fri, 2 May 2014 14:27:22 +0000 (15:27 +0100)]
Always print authenticator even if there are no vps
Arran Cudbard-Bell [Fri, 2 May 2014 14:22:00 +0000 (15:22 +0100)]
Output contents of authenticator in radsniff
Alan T. DeKok [Fri, 2 May 2014 11:20:45 +0000 (07:20 -0400)]
Use "ignoring" instead of bureaucratic sounding phrase
Arran Cudbard-Bell [Fri, 2 May 2014 14:03:43 +0000 (15:03 +0100)]
Check length of octets type attributes too
Arran Cudbard-Bell [Fri, 2 May 2014 11:19:17 +0000 (12:19 +0100)]
Always \0 terminate string type attributes in mschap_add_reply()
Arran Cudbard-Bell [Fri, 2 May 2014 11:12:25 +0000 (12:12 +0100)]
Formatting and cases of vp_strvalue being used for octet type attributes
Arran Cudbard-Bell [Fri, 2 May 2014 09:38:14 +0000 (10:38 +0100)]
Another regex memory leak...
Arran Cudbard-Bell [Fri, 2 May 2014 09:32:38 +0000 (10:32 +0100)]
Don't leak memory when pairmake is called with a regular expression comparison
Arran Cudbard-Bell [Fri, 2 May 2014 09:20:42 +0000 (10:20 +0100)]
Reorder operations in mschap_add_reply to try and unconfuse valgrind
Arran Cudbard-Bell [Fri, 2 May 2014 09:19:55 +0000 (10:19 +0100)]
Valgrind complains about sscanf in rlm_eap_mschapv2.c reading past the end of response->vp_strvalue. Use VERIFY_VP to check that this can't be happening.
Arran Cudbard-Bell [Fri, 2 May 2014 09:18:47 +0000 (10:18 +0100)]
Formatting
Arran Cudbard-Bell [Fri, 2 May 2014 09:18:06 +0000 (10:18 +0100)]
Add check to VERIFY_VP to check that the string buffer is still \0 terminated, and that the length of the of the buffer matches the length of the VALUE_PAIR.
Arran Cudbard-Bell [Fri, 2 May 2014 08:56:40 +0000 (09:56 +0100)]
Port acinclude changes from master
Arran Cudbard-Bell [Fri, 2 May 2014 08:43:45 +0000 (09:43 +0100)]
Make the server seem less alerted
Arran Cudbard-Bell [Fri, 2 May 2014 08:28:56 +0000 (09:28 +0100)]
Remove apple SDK hacks. There's now the option to install the headers in their usual location.