1 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
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
15 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
16 * version 1.4.11 / 1.6.1
17 * Support for networks:
18 socklnd - kernels up to 2.6.16
19 qswlnd - Qsnet kernel modules 5.20 and later
20 openiblnd - IbGold 1.8.2
21 o2iblnd - OFED 1.1 and 1.2
22 viblnd - Voltaire ibhost 3.4.5 and later
23 ciblnd - Topspin 3.2.0
24 iiblnd - Infiniserv 3.3 + PathBits patch
25 gmlnd - GM 2.1.22 and later
26 mxlnd - MX 1.2.1 or later
27 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
31 Details : lnet_clear_peer_table can wait forever if user forgets to
35 Details : libcfs_id2str should check pid against LNET_PID_ANY.
39 Description: added LNET self test
40 Details : landing b_self_test
45 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
47 Details : do_div() macro is used incorrectly.
49 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
53 Description: make panic on lbug configurable
57 Description: Add OFED1.2 support to o2iblnd
58 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
59 are installed (other than kernel's in-tree infiniband), there
60 could be some problem while insmod o2iblnd (mismatch CRC of
62 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
63 this link provides solution:
64 https://bugs.openfabrics.org/show_bug.cgi?id=355
65 if extra Module.symvers is not supported in kernel, we will
66 have to run the script in bug 12316 to update
67 $LINUX/module.symvers before building o2iblnd.
68 More details about this are in bug 12316.
70 ------------------------------------------------------------------------------
72 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
73 * version 1.4.10 / 1.6.0
74 * Support for networks:
75 socklnd - kernels up to 2.6.16
76 qswlnd - Qsnet kernel modules 5.20 and later
77 openiblnd - IbGold 1.8.2
79 viblnd - Voltaire ibhost 3.4.5 and later
80 ciblnd - Topspin 3.2.0
81 iiblnd - Infiniserv 3.3 + PathBits patch
82 gmlnd - GM 2.1.22 and later
83 mxlnd - MX 1.2.1 or later
84 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
89 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
90 possibly accessed in kptllnd_shutdown. Ptllnd should init
91 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
95 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
99 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
103 Description: the_lnet.ln_finalizing was not set when the current thread is
104 about to complete messages. It only affects multi-threaded
110 Description: Changed the default kqswlnd ntxmsg=512
115 Description: Assertion failure in kernel ptllnd caused by posting passive
116 bulk buffers before connection establishment complete.
121 Description: A race in kernel ptllnd between deleting a peer and posting
122 new communications for it could hang communications -
123 manifesting as "Unexpectedly long timeout" messages.
128 Description: Kernel ptllnd lock ordering issue could hang a node.
133 Description: node crash on socket teardown race
136 Frequency : 'lctl peer_list' issued on a mx net
138 Description: Enable lctl's peer_list for MXLND
141 Frequency : after Ptllnd timeouts and portals congestion
143 Description: Credit overflows
144 Details : This was a bug in ptllnd connection establishment. The fix
145 implements better peer stamps to disambiguate connection
146 establishment and ensure both peers enter the credit flow
147 state machine consistently.
152 Description: kptllnd didn't propagate some network errors up to LNET
153 Details : This bug was spotted while investigating 11394. The fix
154 ensures network errors on sends and bulk transfers are
155 propagated to LNET/lustre correctly.
157 Severity : enhancement
159 Description: Fixed console chatter in case of -ETIMEDOUT.
161 Severity : enhancement
163 Description: Added D_NETTRACE for recording network packet history
164 (initially only for ptllnd). Also a separate userspace
165 ptllnd facility to gather history which should really be
166 covered by D_NETTRACE too, if only CDEBUG recorded history in
172 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
173 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
174 callback can occur before a connection has actually been
175 established. This caused an assertion failure previously.
177 Severity : enhancement
179 Description: Multiple instances for o2iblnd
180 Details : Allow multiple instances of o2iblnd to enable networking over
181 multiple HCAs and routing between them.
185 Description: lnet deadlock in router_checker
186 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
187 into BH locks to eliminate potential deadlock caused by
188 ksocknal_data_ready() preempting code holding these locks.
192 Description: Millions of failed socklnd connection attempts cause a very slow FS
193 Details : added a new route flag ksnr_scheduled to distinguish from
194 ksnr_connecting, so that a peer connection request is only turned
195 down for race concerns when an active connection to the same peer
196 is under progress (instead of just being scheduled).
198 ------------------------------------------------------------------------------
200 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
202 * Support for networks:
203 socklnd - kernels up to 2.6.16
204 qswlnd - Qsnet kernel modules 5.20 and later
205 openiblnd - IbGold 1.8.2
207 viblnd - Voltaire ibhost 3.4.5 and later
208 ciblnd - Topspin 3.2.0
209 iiblnd - Infiniserv 3.3 + PathBits patch
210 gmlnd - GM 2.1.22 and later
211 mxlnd - MX 1.2.1 or later
212 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
215 Severity : major on XT3
217 Description: libcfs overwrites /proc/sys/portals
218 Details : libcfs created a symlink from /proc/sys/portals to
219 /proc/sys/lnet for backwards compatibility. This is no
220 longer required and makes the Cray portals /proc variables
225 Description: OFED FMR API change
226 Details : This changes parameter usage to reflect a change in
227 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
228 that FMR support is only used in experimental versions of the
229 o2iblnd - this change does not affect standard usage at all.
231 Severity : enhancement
233 Description: new ko2iblnd module parameter: ib_mtu
234 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
235 HCAs. You can avoid this problem by setting the MTU to 1024
236 using this module parameter.
238 Severity : enhancement
239 Bugzilla : 11118/11620
240 Description: ptllnd small request message buffer alignment fix
241 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
242 Round up small message size on sends in case this option
243 is not supported. 11620 was a defect in the initial
244 implementation which effectively asserted all peers had to be
245 running the correct protocol version which was fixed by always
246 NAK-ing such requests and handling any misalignments they
251 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
252 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
253 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
255 Severity : enhancement
257 Description: Patchless ZC(zero copy) socklnd
258 Details : New protocol for socklnd, socklnd can support zero copy without
259 kernel patch, it's compatible with old socklnd. Checksum is
260 moved from tunables to modparams.
264 Description: When ksocknal_del_peer() is called upon a peer whose
265 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
266 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
269 Frequency : when ptlrpc is under heavy use and runs out of request buffer
271 Description: In lnet_match_blocked_msg(), md can be used without holding a
275 Frequency : very rarely
277 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
278 If connd connects a route which has been closed by
279 ksocknal_shutdown(), ksocknal_create_routes() may create new
280 routes which hold references on the peer, causing shutdown
281 process to wait for peer to disappear forever.
283 Severity : enhancement
285 Description: Dump XT3 portals traces on kptllnd timeout
286 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
287 dump Cray portals debug traces to a file. The kptllnd module
288 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
289 is the basename of the dump file.
292 Frequency : infrequent
294 Description: kernel ptllnd fix bug in connection re-establishment
295 Details : Kernel ptllnd could produce protocol errors e.g. illegal
296 matchbits and/or violate the credit flow protocol when trying
297 to re-establish a connection with a peer after an error or
300 Severity : enhancement
302 Description: Allow /proc/sys/lnet/debug to be set symbolically
303 Details : Allow debug and subsystem debug values to be read/set by name
304 in addition to numerically, for ease of use.
307 Frequency : only in configurations with LNET routers
309 Description: routes automatically marked down and recovered
310 Details : In configurations with LNET routers if a router fails routers
311 now actively try to recover routes that are down, unless they
312 are marked down by an administrator.
314 ------------------------------------------------------------------------------
316 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
319 Frequency : very rarely, in configurations with LNET routers and TCP
321 Description: incorrect data written to files on OSTs
322 Details : In certain high-load conditions incorrect data may be written
323 to files on the OST when using TCP networks.
325 ------------------------------------------------------------------------------
327 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
329 - rework CDEBUG messages rate-limiting mechanism b=10375
330 - add per-socket tunables for socklnd if the kernel is patched b=10327
332 ------------------------------------------------------------------------------
334 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
336 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
337 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
338 - quiet LNET startup LNI message for liblustre b=10128
339 - Better console error messages if 'ip2nets' can't match an IP address
340 - Fixed overflow/use-before-set bugs in linux-time.h
341 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
342 - LNET teardown failed an assertion about the route table being empty
343 - Fixed a crash in LNetEQPoll(<invalid handle>)
344 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
345 - improve debug message for liblustre/Catamount nodes (b=10116)
347 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
348 * Configuration change for the XT3
349 The PTLLND is now used to run Lustre over Portals on the XT3.
350 The configure option(s) --with-cray-portals are no longer
351 used. Rather --with-portals=<path-to-portals-includes> is
352 used to enable building on the XT3. In addition to enable
353 XT3 specific features the option --enable-cray-xt3 must be
356 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
357 * Portals has been removed, replaced by LNET.
358 LNET is new networking infrastructure for Lustre, it includes a
359 reorganized network configuration mode (see the user
360 documentation for full details) as well as support for routing
361 between different network fabrics. Lustre Networking Devices
362 (LNDS) for the supported network fabrics have also been created
363 for this new infrastructure.
365 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
370 Frequency : rare (large Voltaire clusters only)
372 Description: the default number of reserved transmit descriptors was too low
373 for some large clusters
374 Details : As a workaround, the number was increased. A proper fix includes
377 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
382 Frequency : occasional (large-scale events, cluster reboot, network failure)
384 Description: too many error messages on console obscure actual problem and
385 can slow down/panic server, or cause recovery to fail repeatedly
386 Details : enable rate-limiting of console error messages, and some messages
387 that were console errors now only go to the kernel log
389 Severity : enhancement
391 Description: add /proc/sys/portals/catastrophe entry which will report if
392 that node has previously LBUGged
394 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
396 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
398 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
400 - handle error return code in kranal_check_fma_rx() (5915,6054)
402 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
404 - update vibnal (Voltaire IB NAL)
405 - update gmnal (Myrinet NAL), gmnalid
407 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
409 * Landed portals:b_port_step as follows...
411 - removed CFS_DECL_SPIN*
412 just use 'spinlock_t' and initialise with spin_lock_init()
414 - removed CFS_DECL_MUTEX*
415 just use 'struct semaphore' and initialise with init_mutex()
417 - removed CFS_DECL_RWSEM*
418 just use 'struct rw_semaphore' and initialise with init_rwsem()
420 - renamed cfs_sleep_chan -> cfs_waitq
421 cfs_sleep_link -> cfs_waitlink
423 - fixed race in linux version of arch-independent socknal
424 (the ENOMEM/EAGAIN decision).
426 - Didn't fix problems in Darwin version of arch-independent socknal
427 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
429 - removed libcfs types from non-socknal header files (only some types
430 in the header files had been changed; the .c files hadn't been