Whenever the daemon re-loads its config, it keeps any socket-based
(AF_INET) SOCK_DGRAM interfaces open that are matched by the new config,
but it used to close and then re-open all local (AF_UNIX) interfaces and
file-based interfaces such as SOCK_FILE dummy files and SOCK_STREAM
device files. This made it very difficult to develop the new
'routejava' test suite, because the Java API test harness always causes
a config re-load when it sets the REST API user/password in the config,
which caused the local socket interfaces to bounce, which interfered
with the reachability of nodes.
Now, local socket and file-based interfaces remain up after a config
re-load, as long as they are still matched by the new configuration.
Added INFO messages for interfaces that remain up after a config
re-load, and tweak interface up/down INFO messages to be consistent.
The new API follows REST rules by using the proper request verbs:
POST, PUT, PATCH and DELETE, instead of just GET.
The legacy GET-only API is still supported for backward compatibility,
but not longer tested or documented.
Add a new query-single-identity operation.
Implement the lock-single-identity operation, which until now had been
documented but not yet implemented. Whenever a single identity is
locked (released), any other unlocked identities with the same PIN are
flagged to indicate that the PIN is not "fully" unlocked, so that the
next time the PIN is entered, the slot decryption is re-tried for
non-loaded identities, and the locked identity will be unlocked again.
Update the 'keyring' and 'keyringrestful' test scripts:
- refactored to reduce curl command-line clutter in test cases
- now tests the redesigned request verbs and paths
- added a test for GET /restful/keyring/SID
- added a test for PUT /restful/keyring/SID/lock
The LOCK and UNLOCK by PIN requests were acting on the supplied PIN
and also the empty PIN, due to a bug in ob_get_str_ptr(), which returned
an empty string instead of NULL after reaching the end of the string
list.
- make "debug.h" a private header so that the DEBUG() macro does not
interfere with Xcode Debug builds, which set DEBUG=1
- move all #include "debug.h" from headers into .c files
- move 'struct idebug' into a new public header "idebug.h" so that
"log.h" can be public
- move HTTPD port number defs from "httpd.h" to "constants.h", so that
"httpd.h" can be private and "conf.h" can be public
- add missing nested includes so each header compiles stand-alone
without error
- #include "sodium.h" instead of <sodium.h>
- #include "sqlite3.h" instead of <sqlite3.h>
- add header guard to "fifo.h"
- fix header guard in "sync_keys.h"
Rename the logging primitive functions and utility functions, prefixing
all with 'serval_log', eg: logMessage() -> serval_logf() etc.
Add an XPRINTF xhexdump() function and use it to implement the
serval_log_hexdump() utility, renamed from dump(). Add macros
WHY_dump(), WARN_dump(), HINT_dump() and DEBUG_dump(), and use them
everywhere.
Remove the 'log.console.dump_config' and 'log.file.dump_config'
configuration options; configuration is now dumped in every log prolog.
The logging system now constructs the log prolog by invoking the new
'log_prolog' trigger, so that it no longer depends on the version string
and configuration system. Any system that wants to present a message in
the log prolog can define its own trigger, which calls standard log
primitives to print the message.
Split the logging system into a front-end (log.c) that provides the
logging primitives and is independent of the configuration system, and a
set of back-end "outputters" (log_output_console.c, log_output_file.c,
log_output_android.c) that may depend on the configuration system and
are decoupled from the front-end using the 'logoutput' link section.
These log outputters are explicitly linked into executables by the
Makefile rules, but could also be linked in using USE_FEATURE(). The
USE_FEATURE() calls have _not_ been added to servald_features.c, so that
different daemon executables can be built with the same feature set but
different log outputs.
Was failing because recvmsg(2) on OSX does not nul terminate the
pathname of local (AF_UNIX) sockets in the returned sockaddr buffer.
Zerofilling the buffer before calling recvmsg() does the trick.
In the process, improved debug and error logging, replacing
recvwithttl() with recv_message() and recv_message_frag(). The
mdp_net.c source file was retired.
Original DEBUG() and DEBUGF() macros renamed to _DEBUG() and _DEBUGF()
New DEBUG() and DEBUGF() macros, first argument is flagname
New DEBUGF2(foo, bar, ...) macro does if(config.debug.foo||config.debug.bar) test
Replace almost all config.debug.xxx references to IF_DEBUG(xxx)
fdpoll will now run a callback when about to sleep / woke up.
A new Java interface to indicate server started / sleeping / waking up.
An android implementation may allow the CPU to sleep.
If there's a java exception the server will try to shutdown.
Calling servald stop is currently undefined.
- ensure size_t is defined when compiling xprintf.h
- declare overlay_route_record as a union field in overlay_route_record to avoid alignment issues
- set the result of fwrite and fputs
- support slightly older git version
