changes.txt

Version 0.9
------------

***
Major changes
***

- Added support for Keepalive Timeout option
- Added a notification (through netevent) to the TCP layer when a path
  has changed after a failure recovery, so that TCP immmediately
  retransmits, and does not wait for the end of the backoff retransmit
  timer, which may be several seconds. This thus improves the speed of recovery.
- Added module support for Shim6. Shim6 can be compiled as a module
  and loaded at kernel run time. The Shim6 trigger heuristic is a
  separate module, so that people can disable the heuristic (and thus
  only answer to explicit requests from peers), or develop their own
  heuristic, properly interfaced with the LinShim6 daemon.
- Added an option to the shim6 console (shim6c) to dynamically set the
  Tsend timer, with automatic update of the peer (The peer's Tka is
  automatically set to Tsend/3).
  

Version 0.8.3
--------------

***
Misc. changes
***

-libcryptoshim6/cga.c : Added an error message in
 verify_cga_locators(), for improved verbosity.
-shim6d.c, shim6eval.c, main.c : Made the access to os directories
 (/var, /etc) more generic. It is now configurable through resp.
 ./configure --localstatedir=DIR and ./configure --sysconfdir=DIR

***
Bug fixes
***

- opt.c : I2/R2 were not accepted by the peer if exactly one CGA
  locator was used. Fixed by simply not sending PDS option, locator
  option nor signature option in that case (they are simply not
  needed).
- info_server.c, reapd.c: Fixed compilation errors when
  --enable-log-expl-time was set

Version 0.8.2
--------------

***
Bug fixes
***

- patch - reap.c : Fixed a soft lockup between send_start (user
  context) and ka_start (soft irq context).
- shim6.c : Added flag XFRM_TYPE_NON_FRAGMENT in struct xfrm_type
  shim6_type, since shim6 is layered below the fragmentation layer 
  (thus the shim6 header cannot be itself fragmented).

Version 0.8.1
--------------

***
Bug fixes
***

- random.c : The pseudo-random generator was seeded at every call to
  random_int(), with /dev/random. This may result in a very slow
  execution of the daemon. 0.8.1 contains the bug fix.

Version 0.8
------------

***
Major changes
***

- Locator updates (request and ack) now supported
- Use now union as argument of shim6_alloc_send. This respects the strict-
  aliasing rule of gcc, and thus allows for normal O2 optimization.

***
Misc. changes
***

- I2 timer now stopped when transitioning to established state (before it was 
  let on until expiry, and stopped itself when detecting that the state
  was established).
- shim6d.c - get_locators : We changed the behavior that happens when 
  the lp_peer is not included in the loc list. The previous behaviour was
  that we added the lp_peer in the loc list manually. But this is not
  complying with the demand of the peer. Thus now we instead keep the
  peer locator list as it is, but if the current peer locator is not 
  included in that list, we start a REAP exploration to find another 
  suitable locator, included in the list. Note that during the negotiation
  phase, REAP is not enabled because the context state is not established.
  In that particular case the context continues using the initial locator,
  and selects another one the first time a REAP exploration is triggered 
  (however, I don't see any useful situation where the peer would send
  a locator list not including the source locator).
  If the context is established, the situation may happen if the peer sends a
  locator list that does not contain anymore the current locator (in an update
  request). In that case the REAP exploration allows immediately switching
  to an alternate working locator.
- shim6d.c : del_addr may cause the size of the local locator set to reach
  zero. In that case nothing can be done anymore and the context is destroyed
  automatically. Previous versions did not take this case into
  account.
- For increased security, the random seed is now changed every 1000 numbers.
  
***
Bug fixes
***

- Fixed compilation problems in the idips part. (that part was late in 
  following the updates).
- In 0.7, if both HBA and CGA were used, all addresses were sent to the peer	
  in the loc list, but only the non-hba addresses were actually used by REAP.
  Now two functions have been defined in shim6d.c (get_nb_loc_locs and
  get_loc_locs_array()). Those functions are used by both opt.c when sending
  locator lists, and by REAP, when building exploration lists. Thus both 
  are now in sync.


Version 0.7.1
-------------

**
Misc changes
***
- info_server.c : Added support for rm *
- tqueue.c : Removed unused variable tq_elem curr, and unused function 
          task_interrupted().
- libnetlink.c : init to zero the struct req before to fill, to avoid 
	      sending uninitialized bytes to the kernel (this was pointed
	      out by valgrind).
- reapd.c : Padding is now set to 0 in keepalives. It was uninitialized before.
- src/Makefile.am : fixed the noinst_HEADERS variable. Its previous definition
  		  caused the .svn directories to be included in the 
		  distribution.
- opt.c : Added a missing ntohl conversion in opt_init. This is not really a
  	bug, since it does not prevent LinShim6 from correctly working, but
	it does prevent it from efficiently working, since it systematically
	invalidates the CGA cache.

***
Bug fixes
***

- shim6d.c: Some parts of the program were erroneously expecting the local ulid
  	    to be useable as a locator at any time, which is not true (the
	    locator may disappear while the local ulid remains used).
	    This is now fixed, by recording the necessary locator-related data
	    in the context, so that it can be accessed later, even if the real
	    locator has disappeared. Thus the struct in6_addr is replaced
	    with a shim6_loc_l type. This allowed also to remove the ulid_ifidx
	    field, previously present in the context.
- cgatool : The manually generated HBA addresses were incorrect because
    	  cgatool did not add the multiprefix extension to the CGA PDS.
	  The cgad generation was correct, though.
-opt.c : There is now a check before to add the CGA PDS option. It is almost
       always needed, but not when there is only one locator announced.
-tqueue.c: Added function del_task_and_free. It allows to carefully remove a
	   timer whose structure will be freed. If the structure has alread 
	   been pushed in the piper for execution, the handler is replaced 
	   with a destroy handler, and -1 is returned, so that the caller knows
	   that the structure must not be freed yet, and will be freed instead
	   by the handler. This removes the possibility for a segfault in case
	   of race conditions. Shim6d has also been updated to use that new
	   function.
-Compiling ./configure --disable-debug :
	   It did not work. Now it does. Here are the changes :	
	   * Added a #ifdef SHIM6_DEBUG in reapd.c - line 129
	   * Same in xfrm.c - line 681
	   * Same in cgad/dbg.c - whole file.
	   * configure.ac disables strict-aliasing (workaround suppressed in
	   version 0.8). Measurements showed that when efficiency is needed,
	   it is completely necessary to disable debugging (otoh, during the
	   development stage, it is absolutely recommended to keep debugging
	   on).
-info_server.c : replaced bad declaration of extern variable ct_hashtable.
	       This caused sometimes very strange bugs.

	   

Version 0.7
-----------

***
Major Improvements
***
-HBA addresses now supported. The implementation has been validated against the
implementation of Francis Dupont. Some changes has been brought to Francis
Dupont implementation, however, since the draft has evolved since then :
->The multiprefix extension has now 2 bytes instead of one for both the length
and type fields.
->The multiprefix extension has now 4 bytes instead of two for the 
flags/reserved fields.
->The multiprefix extension has now a length field in bytes unit, instead
of 8-bytes units.
We modified Francis Dupont implementation to reflect the changes in the draft,
then compared the results against ours. This can be verified at any moment by
'make check'.

-CGA multikey extension : It was not conformant with rfc4581 in that the 
length field was considered the whole extension length, while rfc4581 
specifies it as the extension length less the first four bytes (ext header).
Note, however, that this code is not used in the current implementation, and
is fixed for future use only. For this reason the fix has not been tested.

***
Misc changes
***
-added a test suite. Currently only CGA and HBA implementations are checked.
-cga_params.c : macro DEF_CGA_PARAMS_FILE removed. We now learn the config file
	      for CGA (and HBA) parameters from the config file cgad.conf
-cgad_config.h : Removed some options that were related to SEND, this file
	       is now in libcryptoshim6 (so that shim6d can access the config
	       file of cgad)
-shim6.c : The error message saying that the shim6d daemon is not there has now 
	 level KERN_INFO rather than KERN_ERR.
-shim6_pkt_listener.c : Added a ifidx parameter in struct shim6_ctx_count 
		      for compatibility with the code from DoCoMo.
-Added a 'disable-dropconfig' option to the ./configure script. It allows to
       keep the old configuration files when reinstalling shim6. This can prove
       useful especially when installing a new release, as it avoids loosing 
       all the customizations that have been done.
       see ./configure --help for more details on that new option.

Version 0.6.4
--------------

***
Bug fixes
***
-xfrm.c,xfrm_init() : When starting shim6d, the old kernel states/policies are 
		    removed, to avoid messing up new and old states.
-shim6d.c,shim6_new_loc_addr() : That function failed to register the 
			       valid_method field of the locator structure 
			       created for registering the new address.
			       Moreover, we have deleted the check for /64
			       prefix, because there is actually no reason
			       no to use /64 global unicasts.

Version 0.6.3
--------------

***
Bug fixes
***
-opt.c : psd_opts[PO_MAX-1] replaced by psd_opts[PO_MAX] 
  (could cause a segfault) - found with valgrind
-shim6d.c : Only call reap_release_ctx if the shim6 is in state established. 
	  (otherwise, the reap context is not initialized yet, and trying to 
	  release it could cause a segfault). - found with valgrind
-info_server.c : After terminating a thread, we ensure now that the pipe
	       carrying termination info is cleared (Without this, the info
	       server can sometimes refuse new connexions).
-xfrm.c : MAJOR BUG FIX : If an interface failed, causing a jump to another
	interface, it was not possible anymore to come back to the previous
	interface, even if it became up again. This was because after the first 
	exploration, all traffic was diverted to the new interface, including 
	Shim6 probes. This is now fixed by adding two new policies in xfrm.c,
	that instruct the xfrm engine not to use Shim6 for any ICMP or Shim6 
	packet.
	

Version 0.6.2
-------------

***
Misc. changes :
***
-cga.c : We previously did selective acceptation of the addresses received from
 the peer (that is accept the verified ones, simply ignore the invalid ones).
This was not conform to the draft : We should ignore the whole message if any
locator is found as invalid. This is the case now.
-cga.c : We previously considered as invalid any address that was not a CGA. 
This was not confrom to the draft : If the ULIDs are valid CGA, and the 
locators are part of the signature, then those locators need not be CGAs 
themselves. They can be normal addresses (section 3, draft v10). This is the 
case now.
-shim6d.c : It does not require anymore that all local addresses be CGAs. It 
	  now stores all local addresses, together with their attribute
	  (currently CGA or normal). However, if a normal address is used by
	  an application, The Shim6 context can still be negotiated, but only 
	  that address can be used as a locator.

***
Bug fixes
***
-info_server.c : Did a segfault if we connect to it and do a cat before the context got established (gforge bug #182)
-shim6d.c-send_i1 : Switched two instructions, (i1->R=0 is now before checksum
computation), this previously caused the checksum to be sometimes incorrect.
-shim6_pkt_listener.c : Replace spinlock() with spinlock_bh(). Otherwise a
		      soft lockup could happen. (if a packet arrives and
		      and the receiving procedure interrupts the sending 
		      procedure, that happened to be inside the lock)
-shim6_trigger : Replaced GFP_KERNEL with GFP_ATOMIC, because now, that
	       function may be called from a soft irq (receiving a packet).
	       This was responsible for a possible error :
	       'scheduling while atomic'
-shim6_local_out: Replaced GFP_KERNEL with GFP_ATOMIC, because we are under a
		  spinlock.

Version 0.6.1
---------------

***
Major Improvements
***
-A tool called cgatool, is now part of the tarball, it allows the manual
generation of CGAs, keys, and CGA parameters. (This tool has been integrated
from another project : DoCoMo SEND).
-A daemon called cgad, that auto-generates CGAs when Router Advertisements are
received.
-The auto-tools files (configure.ac, Makefile.am) have been improved to install
the proper configuration files. A new key and a new set of CGA parameters is
also generated when typing make install. This installation procedure to get 
things work (with a patched kernel of course) is thus now :
->./configure
->make
->sudo make install
->sudo cgad
->sudo shim6d

-As a consequence of the integration of these two tools, a new division in 
libraries has been done.


***
Misc. Changes
***

-info_server.c : Added the command "cat *" that dumps all contexts states, 
	       instead one.
-shim6_rawsocket.c : Changed the way packets are sent, so that now the 
	outgoing interface is forced to be the one that corresponds to the 
	source address of a packet. This is necessary to allow Shim6 to recover
	in some cases.
-reapd.c, send_probe : We now use the function shim6_send from shim6_rawsocket.c
-sysctl.h : Not anymore in the patch, binary sysctl numbers for Shim6 are 
	  now useless. The /proc/sys/net/ipv6/shim6/enabled option is still
	  there, however.
-shim6.c,xfrm6_mode_shim6.c : Adapted shim6_output() for kernel 2.6.24.

***
Bug fixes
***

-shim6.c and net/shim6.h : Removed all occurences of the now useless 
	 shim6_nb_glob_locs and shim6_glob_locs_lock variables. This caused
	 a compilation error if compiling for arch i386
-info_server.c : Fixed a possible seg fault if no context tag was given.
-shim6.c, shim6_add_glob_locator : The interface index of the new locator
	  is now also communicated to user space, so that the daemon maintains
	  coherent information about the addresses.

Version 0.6
-------------

***
Major improvements
***
-CGA addresses are now supported. A tool for CGA generation will be made 
     available later. In the mean time, one can use the tool provided by DoCoMo
     SEND to generate CGAs (see 
     http://www.docomolabs-usa.com/lab_opensource.html)
-A new heuristic is implemented in shim6_pkt_listener.c : A Shim6 negotiation 
is started if one of these conditions is met : the flow exceeds 2KB, or it 
does exist for more than one minute.
-IDIPS support added : it is enabled with ./configure --enable-idips.
With IDIPS support, an IDIPS request is sent right after a Shim6 context is 
created, so that the address pair array is ordered. The current locators are 
changed if the best one (in the sense of IDIPS) differs from the ULIDs.
 Requests to the IDIPS server are also performed when the lifetime of the 
IDIPS response expires, or when there is a failure.

***
Misc. changes
***

-info_server.c : new command "dump kernel contexts" and "dump kernel policies"
-net/ipv6/Kconfig : changed IPV6 config option from tristate to bool, because 
		  we do not support compiling as a module, yet.
-configure script : The configure script for the daemon has been modified to 
	   check for shim6 support in kernel headers. If Shim6 is not supported
	   or headers not found, a message is produced to tell how to specify
	   the correct header location.
-Replaced implementation of shim6_get_loc_addrs() : previously it used a custom
	  netlink message, now we use the glibc getifaddrs function. 
	  (shim6d.c)
-Added a -f option to shim6d, so that it can now be run in foreground
-Implementation of Shim6 TLV options field is now in file opt.c/opt.h
-All netlink messages daemon->kernel have been replaced by standard calls.
-netlink messages kernel->daemon have been changed from unicast to multicast.
-shim6.c : Removed now useless shim6_glob_locs_list
-shim6.c : shim6_filter() : Added missing checksum verification.
-reap.c : reap_input() : Added missing checksum verification.
-shim6_pkt_listener.c : Removed now useless full_ctx flag in struct shim6_ctx_count.
-xfrm_user.c : Removed now useless option XFRM_POLICY_UPD_CACHE. This was only
	     necessary in earlier versions of the kernel.
-opt.c : Added a check of the Critical bit when an option is not known. If the
       bit is set, we drop the whole message, if it is not, we only drop the
       option.
-shim6d.c : Added support for context confusion detection (section 7.6 and 
	  7.15)
-shim6d.c : rcv_i1() : Now fully compliant with draft version 9, section 7.9
-shim6d.c : RESP_NONCE_UPDATE set to SHIM6_VALIDATOR_MIN_LIFETIME. This way we 
	  ensure that "nonces that are no older than VALIDATOR_MIN_LIFETIME 
	  SHOULD be considered recent" (draft v9, sec. 7.13)
-reapd.c : Only receive a probe/ka on established shim6 
	 contexts. (it was not guaranteed before)
-reapd.c : rcv_probe() :  The check for context found was incorrect (rctx 
	 instead of ctx).

-xfrm.c : Upon context update : The routing table is now also updated, so that
	  we ensure that the right interface is used to send the now 
	  translated packets. This is especially useful if an interface 
	  physically goes down. The code adds a route to reach ULIDpeer from
	  ULIDlocal through the same device as the one used to send packets 
	  with the corresponding locators.
	  

***
Bug fixes
***
-gen_patch.py (only in the svn): Modified so that 
	      patch -p1 < patch_linshim6_*** now works
	      (patch -p7 was necessary in version 0.5.1).
-shim6d.c : rcv_i2() : If i2 triggered a shim6_ctx creation, but an error 
	  occured later, the shim6_ctx was still there, in broken state.
	  Now we destroy the newly created context if an error occurs.
-shim6d.c : shim6_del_ctx : memory leak if ctx->r1_vldt was defined.
-shim6d.c : rcv_r2 : The daemon stopped if two R2s were successively sent
	  (because of an incorrect ASSERT). Changed to 'silently ignore', 
	  as stated in draft proto-09, sec. 7.16
-shim6d.c : get_locators : A parsing error (with possible seg fault) happened 
	  if the source locator was not included in the loc. list.
-shim6d.c : get_locators : The nb_peer_locs field was not set to 0 if the
	  realloc failed, resulting in a possible segfault in other parts of
	  the program.
-shim6d.c : rcv_i2() : There was a missing 'return 0;' at the end of the 
	  function, resulting in the failure path always being executed.
-shim6d.c : fixed segfault in send_r1().
-xfrm.c : parse_expire : replaced lookup_ulid(daddr,daddr) with 
	lookup_ulid(daddr,saddr).
-reap.c : ka_handler() : Reset of the conceptual timer (Keepalive timeout) 
	when stopping the ka timer (Keepalive interval). Without this,
	only one burst of keepalives is sent in case of unidirectional traffic.
-shim6.c : Fixed memory leak in shim6_init_state() (useless memory allocation
	 in case of context update).

Version 0.5.1
--------------

-kernel patch 0.5.1-1 : Implemented function shim6_del_glob_locator
-daemon 0.5.1-3 : Length field for locators list option does include the
 internal padding now, to reflect the requirement from draft shim6-proto, 
 section 5.15.2
-daemon 0.5.1-3 : Added code for reception of i2 messages. It was lacking it 
 previous 0.5 code.
-daemon 0.5.1-3 : The resp. nonce secret is now changed every ten minutes. A 
 prev_resp_secret variable keeps a backup of the previous responder secret, 
 in order to avoid failing the negotiation after a secret change.
-Netlink messages sent to the kernel have now the flag 
 NLM_F_REQUEST, because without it, messages won't be read by 
 the kernel 2.6.23.
-The context tag generation is now randomized (all 47 bits are random).
-Daemon : Concurrency is managed differently : All shim6 related operations are
 done inside a single thread. If some other threads needs to run some function 
 (timer, information server), it sends a request for execution to the main 
 thread, through a pipe mechanism implemented in pipe.c. This makes concurrency
 management much cleaner, without affecting really the efficiency.
-Added a patch for Linux kernel 2.6.23


***
Bug fixes
***

-daemon 0.5.1-3 : It was not possible to find a matching context upon reception
 of an R1 message. Fixed by inserting lp_peer in ctx->ls_peerp upon context
 creation.
-daemon 0.5.1-3 : Added a missing free (ctx->ls_peerp) in shim6_del_ctx
-daemon 0.5.1-2 : Fixed compilation errors when LOG_EXPL_TIME option was 
 enabled. Thanks to Martin Kulas and Aaron Kunde for their patch !
-daemon 0.5.1-1 : Fixed a compile error :
 undefined reference to `shim6_glob_locs_lock' 
-daemon 0.5.1-1 : Fixed bug #167
-Fixed Compilation error :
 undefined reference to `shim6_glob_locs_lock' 
-Fixed a problem that occured when option CONFIG_XFRM_MIGRATE was enabled :
 replaced orig->ct with orig->shim6

***
Minor changes
***
-daemon 0.5.1-3 : Removed now useless full_ctx field in struct shim6_ctx

Version 0.5
-----------
While 0.4.x was for the Linux Kernel 2.6.17.11, the 0.5 is for version 2.6.21.

***
Fundamental changes
***

- The xfrm API is now used in the kernel part.
- A filter has been added in net/ipv6/raw.c, so that only shim6 control
  messages are sent to user space.
- shim6 control plane is now completely managed in user space.
- cur_resp_nonce is no longer incremented at each use, but rather every 60
  seconds. This is to take into account the update in the shim6-proto draft,
  and simplifies the implementation.

***
Misc. changes
***

- When the hash computing fails (get_resp_hash, shim6.c), the r1 packet is not
  sent.
- Simplified get and set context tag by definition of get_ct and set_ct in
  shim6__linux.h
- Added a configuration option for shim6 debug messages.
- Changed shim6 netlink protocol number from 19 to 20 because 19 has been
  allocated to ECRYPTFS.


Version 0.4.3 - most stable
-------------

This version has been run under valgrind and tested with a script that
continuously stops the currently used path, triggering an exploration.

- Randomized by +/- 20% the interval between sent probes.
- randomized the order of address combinations : each time a new exploration
starts, the path array (containing the address pairs) is randomized, so that
good pairs staying at the end have a chance of being selected. (next step
should be to support some intelligent ordering of address pairs).

***
Bug fixes
***

- A kernel freeze could occur when receiving a data packet with the reapd
  daemon still not running. This was due to the reapd_pid variable in the
  kernel being initialized to zero by default. But 0 is considered as being
  the pid of the kernel, so that the kernel was sending messages to itself,
  with a freeze as the result. This has been solved by adding some forgotten
  checks : 
  - shim6_del_glob_locator (shim6.c). 
  - reap_new_daemon_ctx (reap.c)

  In addition, we should refuse creating a new shim6 context if the reapd
  daemon is not running : Added a check at the top of __init_new_ctx (shim6.c)

Version 0.4.2
-------------

- Source address selection now done in user space. For the moment it still
does a simple iteration over all possible address combinations, but at least
this makes sure every possible path is tried when exploring.  
- The probe
report lists are cleared upon send timer expiry. This avoids pretending we are
reachable through a path that has recently failed (and maybe caused the send
timer expiry).

***
Bug fixes
***

- When a context was released, it was not removed from the hashtable (function
  reap_release_ctx). Fixed.

Version 0.4.1 
-------------

- In previous versions, a context was created when connecting to the loopback
  address. This is not the case anymore.  
- Changed netlink reap protocol number, NETLINK_REAP from 17 to 19.  (Because 
  17 is now reserved inside the kernel). This makes REAP version 0.4.1 and
  later incompatible with shim6 kernels 0.4 and earlier.
- The information interface with the reapd daemon is not anymore a signal
  handling procedure, but rather a telnet server. This allows test and
  measurement programs to directly connect to the daemon, and ask for
  information from it. 
- in /proc/net/shim6/*, the local and peer context tags were written in
  decimal, this could become confusing for large context tag. This is now
  given in hexadecimal.
- Added the possibility to manually remove a context. This makes 
  experimentation easier. To remove a context manually, telnet to localhost
  port 50000, then type del <context tag> to remove the corresponding
  context from both the kernel and the daemon.
- Also, when the program terminates (either by sigterm or sigsegv, which
  should of course not occur, but we are still in the development step), 
  all contexts are flushed, both in the kernel and the daemon. This is because
  anyway if the daemon is killed one part of the context is lost, so it is
  normal to force deletion of the other part as well.
- The R field in probe messages is now set to 0 for every probe message (as
  defined in the draft).
- reap_rcv_probe, case REAP_EXPLORING : moved downward the call to
  start_send_timer, to make sure stop_timer is correctly set to FALSE before
  by init_probe_sending(). Failing to do this could prevent the send timer to
  be started when an exploring probe was received while in state operational.
- reap_random() - reapd.c : Made use of /dev/random to generate nonces. The
  problem is that reading /dev/random may cause the program to wait during a
  very long time (10 seconds) because of the lack of random events. This has
  been replaced by a single read of /dev/random that determines a seed of a
  mathematical random number generator.

***
Bug fixes
***

- fixed a segfault (in reapd) that occured when a probe report Inbound_ok or
  operational was received without any received probe report.
- send_handler : A race condition could occur if the send timer were expiring
  while terminating an exploration (reap_end_explore), leading to an assertion
  'rctx->state==REAP_OPERATIONAL' failed. This is fixed.
- failed to check rctx!=NULL after a lookup_ct in reap_notify_in and .._out.
  fixed.

Version 0.4
-----------

- Added a documentation directory. It contains a latex document which will be
  continuously evolving.

***
Changes to agree with the failure detection draft, version 7
***
- Adapted delays between sent probes : 4 probes sent every 500ms, then 
exponential backoff until reaching a maximum inter-probe time of 60 seconds.
- Removed option for the keepalive message.
- state REAP_EXPLORING_OK renamed in REAP_INBOUND_OK
- format changed for the probe messages
- the new definition of keepalive timeout and keepalive interval is
implemented. (the keepalive option is not #ifdef CONFIG_IPV6_SHIM6supported, however)

***
Bug fixes
***

- shim6.c, shim6_del_glob_locator : only decrement shim6_nb_glob_locs and
increment glob_gen_nb if the locator were actually removed from the list (that
is, the locator given as an argument was part of the list before). This bug
was responsible for the sending of buggy locators list (produced the message
'bad number of locators'). In some special situations, a system crash could
even occur (because of the shim6_nb_glob_locs variable becoming negative, thus
making memory allocation less than the necessary size).
- reap.c : In addition to the revisions due to the move to userland, a bug has
been fixed : When reap provided new locators (thus triggering extension header
addition), then came back to the ulids as locators, the extension was still
present (but useless). This is fixed.

***
Misc changes.
***
- shim6.c, shim6_del_glob_locators : Changed to trigger an exploration
  immediately when an address becomes locally unavailable, for any context
  having this address as current local locator.
- Computation of next probe timeout moved from probe_handler() to send_probe();
  probe burst initialization : init_probe_sending(). Thoses changes allowed to
  reduce code repetition.
- Common sk_buff reservation function for shim6 and reap. (also to reduce code
  repetition)
- struct shim6_ctx : 'int nb_peer_locs' replaced by 'u8 nb_peer_locs', because
this is defined as an 8 bits field in the protocol message.
- shim6.c : cleaned shim6_init function. (In case of error, everything is
correctly undone now).
- link local addresses removed from the set of possible locators. This is
because there is a technical problem to solve before to use this
correctly. Also it should be discussed whether it is really useful to solve
the technical problem, or if it is better simply not to use link local
addresses. The technical problem is that link local addresses have sense only
if associated with a link. But we cannot carry link information together with
the address in the locators list option of shim6.
- Filenames in the /proc/net/shim6 directory are no longer memory addresses,
but rather local context tags. This is also locally unique, clearer, and has
the advantage to allow having the same filenames as the daemon info files 
(in /var/run/shim6).

-------------------------------------

Version 0.3
-----------

- shim6.h
  updated protocol constants to agree with draft version 6, section 14
- reap.c
  Adapted error interpretation in probe sending, 'network unreachable' : this
  can occur when there is a local link failure, and should not be considered
  an error. This is now just noticed with a printk KERN_NOTICE.
- BUG FIX:
  TCP MSS is now updated to take into account the shim6 payload extension
  header. This is done by increasing the field icsk_ext_hdr_len of struct
  inet_connection_sock by sizeof(shim6hdr_pld), which is 8 bytes.
  The increase in done when the socket is using shim6 for the first time
  (within shim6_translate_route). The
  drawback is that each icsk using shim6 has 8 bytes less for the MSS, also if
  it does not currently use the payload extension header. Nevertheless, the
  loss in MSS is less then 1%, and it appears as the cleanest way of making
  sure the MSS is not too large (which would be worse). The reason is that at
  the moment we decide to add the extension header, the TCP segment is already
  built, so we need to anticipate by reducing the MSS *before* building the
  TCP segment. One way of doing that is to reduce the MSS, immediately when we
  learn that shim6 is enabled for this socket.

-------------------------------------

Version 0.2
-----------
- added kconfig option for shim6. Shim6 code is now enclosed with #ifdef/#endif
- It was before only compilable for UML, this is fixed.
- Added a shim6 socket (similar to the icmp one).
- registered shim6 as an inet6_protocol (like other extension headers)
- added REAP timer information to /proc/net/shim6 entries