1 2006-06-22 Cluster File Systems, Inc. <info@clusterfs.com>
2 * version 1.4.11 / 1.6.1
3 * Support for networks:
4 socklnd - kernels up to 2.6.16
5 qswlnd - Qsnet kernel modules 5.20 and later
6 openiblnd - IbGold 1.8.2
7 o2iblnd - OFED 1.1 and 1.2
8 viblnd - Voltaire ibhost 3.4.5 and later
10 iiblnd - Infiniserv 3.3 + PathBits patch
11 gmlnd - GM 2.1.22 and later
12 mxlnd - MX 1.2.1 or later
13 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
18 Description: added LNET self test
19 Details : landing b_self_test
24 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
26 Details : do_div() macro is used incorrectly.
28 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
32 Description: make panic on lbug configurable
36 Description: Add OFED1.2 support to o2iblnd
37 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
38 are installed (other than kernel's in-tree infiniband), there
39 could be some problem while insmod o2iblnd (mismatch CRC of
41 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
42 this link provides solution:
43 https://bugs.openfabrics.org/show_bug.cgi?id=355
44 if extra Module.symvers is not supported in kernel, we will
45 have to run the script in bug 12316 to update
46 $LINUX/module.symvers before building o2iblnd.
47 More details about this are in bug 12316.
49 ------------------------------------------------------------------------------
51 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
52 * version 1.4.10 / 1.6.0
53 * Support for networks:
54 socklnd - kernels up to 2.6.16
55 qswlnd - Qsnet kernel modules 5.20 and later
56 openiblnd - IbGold 1.8.2
58 viblnd - Voltaire ibhost 3.4.5 and later
59 ciblnd - Topspin 3.2.0
60 iiblnd - Infiniserv 3.3 + PathBits patch
61 gmlnd - GM 2.1.22 and later
62 mxlnd - MX 1.2.1 or later
63 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
68 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
69 possibly accessed in kptllnd_shutdown. Ptllnd should init
70 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
74 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
78 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
82 Description: the_lnet.ln_finalizing was not set when the current thread is
83 about to complete messages. It only affects multi-threaded
89 Description: Changed the default kqswlnd ntxmsg=512
94 Description: Assertion failure in kernel ptllnd caused by posting passive
95 bulk buffers before connection establishment complete.
100 Description: A race in kernel ptllnd between deleting a peer and posting
101 new communications for it could hang communications -
102 manifesting as "Unexpectedly long timeout" messages.
107 Description: Kernel ptllnd lock ordering issue could hang a node.
112 Description: node crash on socket teardown race
115 Frequency : 'lctl peer_list' issued on a mx net
117 Description: Enable lctl's peer_list for MXLND
120 Frequency : after Ptllnd timeouts and portals congestion
122 Description: Credit overflows
123 Details : This was a bug in ptllnd connection establishment. The fix
124 implements better peer stamps to disambiguate connection
125 establishment and ensure both peers enter the credit flow
126 state machine consistently.
131 Description: kptllnd didn't propagate some network errors up to LNET
132 Details : This bug was spotted while investigating 11394. The fix
133 ensures network errors on sends and bulk transfers are
134 propagated to LNET/lustre correctly.
136 Severity : enhancement
138 Description: Fixed console chatter in case of -ETIMEDOUT.
140 Severity : enhancement
142 Description: Added D_NETTRACE for recording network packet history
143 (initially only for ptllnd). Also a separate userspace
144 ptllnd facility to gather history which should really be
145 covered by D_NETTRACE too, if only CDEBUG recorded history in
151 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
152 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
153 callback can occur before a connection has actually been
154 established. This caused an assertion failure previously.
156 Severity : enhancement
158 Description: Multiple instances for o2iblnd
159 Details : Allow multiple instances of o2iblnd to enable networking over
160 multiple HCAs and routing between them.
164 Description: lnet deadlock in router_checker
165 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
166 into BH locks to eliminate potential deadlock caused by
167 ksocknal_data_ready() preempting code holding these locks.
171 Description: Millions of failed socklnd connection attempts cause a very slow FS
172 Details : added a new route flag ksnr_scheduled to distinguish from
173 ksnr_connecting, so that a peer connection request is only turned
174 down for race concerns when an active connection to the same peer
175 is under progress (instead of just being scheduled).
177 ------------------------------------------------------------------------------
179 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
181 * Support for networks:
182 socklnd - kernels up to 2.6.16
183 qswlnd - Qsnet kernel modules 5.20 and later
184 openiblnd - IbGold 1.8.2
186 viblnd - Voltaire ibhost 3.4.5 and later
187 ciblnd - Topspin 3.2.0
188 iiblnd - Infiniserv 3.3 + PathBits patch
189 gmlnd - GM 2.1.22 and later
190 mxlnd - MX 1.2.1 or later
191 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
194 Severity : major on XT3
196 Description: libcfs overwrites /proc/sys/portals
197 Details : libcfs created a symlink from /proc/sys/portals to
198 /proc/sys/lnet for backwards compatibility. This is no
199 longer required and makes the Cray portals /proc variables
204 Description: OFED FMR API change
205 Details : This changes parameter usage to reflect a change in
206 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
207 that FMR support is only used in experimental versions of the
208 o2iblnd - this change does not affect standard usage at all.
210 Severity : enhancement
212 Description: new ko2iblnd module parameter: ib_mtu
213 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
214 HCAs. You can avoid this problem by setting the MTU to 1024
215 using this module parameter.
217 Severity : enhancement
218 Bugzilla : 11118/11620
219 Description: ptllnd small request message buffer alignment fix
220 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
221 Round up small message size on sends in case this option
222 is not supported. 11620 was a defect in the initial
223 implementation which effectively asserted all peers had to be
224 running the correct protocol version which was fixed by always
225 NAK-ing such requests and handling any misalignments they
230 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
231 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
232 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
234 Severity : enhancement
236 Description: Patchless ZC(zero copy) socklnd
237 Details : New protocol for socklnd, socklnd can support zero copy without
238 kernel patch, it's compatible with old socklnd. Checksum is
239 moved from tunables to modparams.
243 Description: When ksocknal_del_peer() is called upon a peer whose
244 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
245 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
248 Frequency : when ptlrpc is under heavy use and runs out of request buffer
250 Description: In lnet_match_blocked_msg(), md can be used without holding a
254 Frequency : very rarely
256 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
257 If connd connects a route which has been closed by
258 ksocknal_shutdown(), ksocknal_create_routes() may create new
259 routes which hold references on the peer, causing shutdown
260 process to wait for peer to disappear forever.
262 Severity : enhancement
264 Description: Dump XT3 portals traces on kptllnd timeout
265 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
266 dump Cray portals debug traces to a file. The kptllnd module
267 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
268 is the basename of the dump file.
271 Frequency : infrequent
273 Description: kernel ptllnd fix bug in connection re-establishment
274 Details : Kernel ptllnd could produce protocol errors e.g. illegal
275 matchbits and/or violate the credit flow protocol when trying
276 to re-establish a connection with a peer after an error or
279 Severity : enhancement
281 Description: Allow /proc/sys/lnet/debug to be set symbolically
282 Details : Allow debug and subsystem debug values to be read/set by name
283 in addition to numerically, for ease of use.
286 Frequency : only in configurations with LNET routers
288 Description: routes automatically marked down and recovered
289 Details : In configurations with LNET routers if a router fails routers
290 now actively try to recover routes that are down, unless they
291 are marked down by an administrator.
293 ------------------------------------------------------------------------------
295 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
298 Frequency : very rarely, in configurations with LNET routers and TCP
300 Description: incorrect data written to files on OSTs
301 Details : In certain high-load conditions incorrect data may be written
302 to files on the OST when using TCP networks.
304 ------------------------------------------------------------------------------
306 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
308 - rework CDEBUG messages rate-limiting mechanism b=10375
309 - add per-socket tunables for socklnd if the kernel is patched b=10327
311 ------------------------------------------------------------------------------
313 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
315 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
316 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
317 - quiet LNET startup LNI message for liblustre b=10128
318 - Better console error messages if 'ip2nets' can't match an IP address
319 - Fixed overflow/use-before-set bugs in linux-time.h
320 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
321 - LNET teardown failed an assertion about the route table being empty
322 - Fixed a crash in LNetEQPoll(<invalid handle>)
323 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
324 - improve debug message for liblustre/Catamount nodes (b=10116)
326 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
327 * Configuration change for the XT3
328 The PTLLND is now used to run Lustre over Portals on the XT3.
329 The configure option(s) --with-cray-portals are no longer
330 used. Rather --with-portals=<path-to-portals-includes> is
331 used to enable building on the XT3. In addition to enable
332 XT3 specific features the option --enable-cray-xt3 must be
335 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
336 * Portals has been removed, replaced by LNET.
337 LNET is new networking infrastructure for Lustre, it includes a
338 reorganized network configuration mode (see the user
339 documentation for full details) as well as support for routing
340 between different network fabrics. Lustre Networking Devices
341 (LNDS) for the supported network fabrics have also been created
342 for this new infrastructure.
344 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
349 Frequency : rare (large Voltaire clusters only)
351 Description: the default number of reserved transmit descriptors was too low
352 for some large clusters
353 Details : As a workaround, the number was increased. A proper fix includes
356 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
361 Frequency : occasional (large-scale events, cluster reboot, network failure)
363 Description: too many error messages on console obscure actual problem and
364 can slow down/panic server, or cause recovery to fail repeatedly
365 Details : enable rate-limiting of console error messages, and some messages
366 that were console errors now only go to the kernel log
368 Severity : enhancement
370 Description: add /proc/sys/portals/catastrophe entry which will report if
371 that node has previously LBUGged
373 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
375 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
377 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
379 - handle error return code in kranal_check_fma_rx() (5915,6054)
381 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
383 - update vibnal (Voltaire IB NAL)
384 - update gmnal (Myrinet NAL), gmnalid
386 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
388 * Landed portals:b_port_step as follows...
390 - removed CFS_DECL_SPIN*
391 just use 'spinlock_t' and initialise with spin_lock_init()
393 - removed CFS_DECL_MUTEX*
394 just use 'struct semaphore' and initialise with init_mutex()
396 - removed CFS_DECL_RWSEM*
397 just use 'struct rw_semaphore' and initialise with init_rwsem()
399 - renamed cfs_sleep_chan -> cfs_waitq
400 cfs_sleep_link -> cfs_waitlink
402 - fixed race in linux version of arch-independent socknal
403 (the ENOMEM/EAGAIN decision).
405 - Didn't fix problems in Darwin version of arch-independent socknal
406 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
408 - removed libcfs types from non-socknal header files (only some types
409 in the header files had been changed; the .c files hadn't been