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
32 Description: Initialize cpumask before use
36 Description: ASSERTION failures when upgrading to the patchless zero-copy
38 Details : This bug affects "rolling upgrades", causing an inconsistent
39 protocol version negotiation and subsequent assertion failure
40 during rolling upgrades after the first wave of upgrades.
44 Details : Change "dropped message" CERRORs to D_NETERROR so they are
45 logged instead of creating "console chatter" when a lustre
46 timeout races with normal RPC completion.
49 Details : lnet_clear_peer_table can wait forever if user forgets to
53 Details : libcfs_id2str should check pid against LNET_PID_ANY.
57 Description: added LNET self test
58 Details : landing b_self_test
63 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
65 Details : do_div() macro is used incorrectly.
67 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
71 Description: make panic on lbug configurable
75 Description: Add OFED1.2 support to o2iblnd
76 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
77 are installed (other than kernel's in-tree infiniband), there
78 could be some problem while insmod o2iblnd (mismatch CRC of
80 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
81 this link provides solution:
82 https://bugs.openfabrics.org/show_bug.cgi?id=355
83 if extra Module.symvers is not supported in kernel, we will
84 have to run the script in bug 12316 to update
85 $LINUX/module.symvers before building o2iblnd.
86 More details about this are in bug 12316.
88 ------------------------------------------------------------------------------
90 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
91 * version 1.4.10 / 1.6.0
92 * Support for networks:
93 socklnd - kernels up to 2.6.16
94 qswlnd - Qsnet kernel modules 5.20 and later
95 openiblnd - IbGold 1.8.2
97 viblnd - Voltaire ibhost 3.4.5 and later
98 ciblnd - Topspin 3.2.0
99 iiblnd - Infiniserv 3.3 + PathBits patch
100 gmlnd - GM 2.1.22 and later
101 mxlnd - MX 1.2.1 or later
102 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
107 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
108 possibly accessed in kptllnd_shutdown. Ptllnd should init
109 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
113 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
117 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
121 Description: the_lnet.ln_finalizing was not set when the current thread is
122 about to complete messages. It only affects multi-threaded
128 Description: Changed the default kqswlnd ntxmsg=512
133 Description: Assertion failure in kernel ptllnd caused by posting passive
134 bulk buffers before connection establishment complete.
139 Description: A race in kernel ptllnd between deleting a peer and posting
140 new communications for it could hang communications -
141 manifesting as "Unexpectedly long timeout" messages.
146 Description: Kernel ptllnd lock ordering issue could hang a node.
151 Description: node crash on socket teardown race
154 Frequency : 'lctl peer_list' issued on a mx net
156 Description: Enable lctl's peer_list for MXLND
159 Frequency : after Ptllnd timeouts and portals congestion
161 Description: Credit overflows
162 Details : This was a bug in ptllnd connection establishment. The fix
163 implements better peer stamps to disambiguate connection
164 establishment and ensure both peers enter the credit flow
165 state machine consistently.
170 Description: kptllnd didn't propagate some network errors up to LNET
171 Details : This bug was spotted while investigating 11394. The fix
172 ensures network errors on sends and bulk transfers are
173 propagated to LNET/lustre correctly.
175 Severity : enhancement
177 Description: Fixed console chatter in case of -ETIMEDOUT.
179 Severity : enhancement
181 Description: Added D_NETTRACE for recording network packet history
182 (initially only for ptllnd). Also a separate userspace
183 ptllnd facility to gather history which should really be
184 covered by D_NETTRACE too, if only CDEBUG recorded history in
190 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
191 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
192 callback can occur before a connection has actually been
193 established. This caused an assertion failure previously.
195 Severity : enhancement
197 Description: Multiple instances for o2iblnd
198 Details : Allow multiple instances of o2iblnd to enable networking over
199 multiple HCAs and routing between them.
203 Description: lnet deadlock in router_checker
204 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
205 into BH locks to eliminate potential deadlock caused by
206 ksocknal_data_ready() preempting code holding these locks.
210 Description: Millions of failed socklnd connection attempts cause a very slow FS
211 Details : added a new route flag ksnr_scheduled to distinguish from
212 ksnr_connecting, so that a peer connection request is only turned
213 down for race concerns when an active connection to the same peer
214 is under progress (instead of just being scheduled).
216 ------------------------------------------------------------------------------
218 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
220 * Support for networks:
221 socklnd - kernels up to 2.6.16
222 qswlnd - Qsnet kernel modules 5.20 and later
223 openiblnd - IbGold 1.8.2
225 viblnd - Voltaire ibhost 3.4.5 and later
226 ciblnd - Topspin 3.2.0
227 iiblnd - Infiniserv 3.3 + PathBits patch
228 gmlnd - GM 2.1.22 and later
229 mxlnd - MX 1.2.1 or later
230 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
233 Severity : major on XT3
235 Description: libcfs overwrites /proc/sys/portals
236 Details : libcfs created a symlink from /proc/sys/portals to
237 /proc/sys/lnet for backwards compatibility. This is no
238 longer required and makes the Cray portals /proc variables
243 Description: OFED FMR API change
244 Details : This changes parameter usage to reflect a change in
245 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
246 that FMR support is only used in experimental versions of the
247 o2iblnd - this change does not affect standard usage at all.
249 Severity : enhancement
251 Description: new ko2iblnd module parameter: ib_mtu
252 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
253 HCAs. You can avoid this problem by setting the MTU to 1024
254 using this module parameter.
256 Severity : enhancement
257 Bugzilla : 11118/11620
258 Description: ptllnd small request message buffer alignment fix
259 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
260 Round up small message size on sends in case this option
261 is not supported. 11620 was a defect in the initial
262 implementation which effectively asserted all peers had to be
263 running the correct protocol version which was fixed by always
264 NAK-ing such requests and handling any misalignments they
269 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
270 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
271 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
273 Severity : enhancement
275 Description: Patchless ZC(zero copy) socklnd
276 Details : New protocol for socklnd, socklnd can support zero copy without
277 kernel patch, it's compatible with old socklnd. Checksum is
278 moved from tunables to modparams.
282 Description: When ksocknal_del_peer() is called upon a peer whose
283 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
284 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
287 Frequency : when ptlrpc is under heavy use and runs out of request buffer
289 Description: In lnet_match_blocked_msg(), md can be used without holding a
293 Frequency : very rarely
295 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
296 If connd connects a route which has been closed by
297 ksocknal_shutdown(), ksocknal_create_routes() may create new
298 routes which hold references on the peer, causing shutdown
299 process to wait for peer to disappear forever.
301 Severity : enhancement
303 Description: Dump XT3 portals traces on kptllnd timeout
304 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
305 dump Cray portals debug traces to a file. The kptllnd module
306 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
307 is the basename of the dump file.
310 Frequency : infrequent
312 Description: kernel ptllnd fix bug in connection re-establishment
313 Details : Kernel ptllnd could produce protocol errors e.g. illegal
314 matchbits and/or violate the credit flow protocol when trying
315 to re-establish a connection with a peer after an error or
318 Severity : enhancement
320 Description: Allow /proc/sys/lnet/debug to be set symbolically
321 Details : Allow debug and subsystem debug values to be read/set by name
322 in addition to numerically, for ease of use.
325 Frequency : only in configurations with LNET routers
327 Description: routes automatically marked down and recovered
328 Details : In configurations with LNET routers if a router fails routers
329 now actively try to recover routes that are down, unless they
330 are marked down by an administrator.
332 ------------------------------------------------------------------------------
334 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
337 Frequency : very rarely, in configurations with LNET routers and TCP
339 Description: incorrect data written to files on OSTs
340 Details : In certain high-load conditions incorrect data may be written
341 to files on the OST when using TCP networks.
343 ------------------------------------------------------------------------------
345 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
347 - rework CDEBUG messages rate-limiting mechanism b=10375
348 - add per-socket tunables for socklnd if the kernel is patched b=10327
350 ------------------------------------------------------------------------------
352 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
354 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
355 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
356 - quiet LNET startup LNI message for liblustre b=10128
357 - Better console error messages if 'ip2nets' can't match an IP address
358 - Fixed overflow/use-before-set bugs in linux-time.h
359 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
360 - LNET teardown failed an assertion about the route table being empty
361 - Fixed a crash in LNetEQPoll(<invalid handle>)
362 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
363 - improve debug message for liblustre/Catamount nodes (b=10116)
365 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
366 * Configuration change for the XT3
367 The PTLLND is now used to run Lustre over Portals on the XT3.
368 The configure option(s) --with-cray-portals are no longer
369 used. Rather --with-portals=<path-to-portals-includes> is
370 used to enable building on the XT3. In addition to enable
371 XT3 specific features the option --enable-cray-xt3 must be
374 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
375 * Portals has been removed, replaced by LNET.
376 LNET is new networking infrastructure for Lustre, it includes a
377 reorganized network configuration mode (see the user
378 documentation for full details) as well as support for routing
379 between different network fabrics. Lustre Networking Devices
380 (LNDS) for the supported network fabrics have also been created
381 for this new infrastructure.
383 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
388 Frequency : rare (large Voltaire clusters only)
390 Description: the default number of reserved transmit descriptors was too low
391 for some large clusters
392 Details : As a workaround, the number was increased. A proper fix includes
395 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
400 Frequency : occasional (large-scale events, cluster reboot, network failure)
402 Description: too many error messages on console obscure actual problem and
403 can slow down/panic server, or cause recovery to fail repeatedly
404 Details : enable rate-limiting of console error messages, and some messages
405 that were console errors now only go to the kernel log
407 Severity : enhancement
409 Description: add /proc/sys/portals/catastrophe entry which will report if
410 that node has previously LBUGged
412 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
414 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
416 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
418 - handle error return code in kranal_check_fma_rx() (5915,6054)
420 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
422 - update vibnal (Voltaire IB NAL)
423 - update gmnal (Myrinet NAL), gmnalid
425 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
427 * Landed portals:b_port_step as follows...
429 - removed CFS_DECL_SPIN*
430 just use 'spinlock_t' and initialise with spin_lock_init()
432 - removed CFS_DECL_MUTEX*
433 just use 'struct semaphore' and initialise with init_mutex()
435 - removed CFS_DECL_RWSEM*
436 just use 'struct rw_semaphore' and initialise with init_rwsem()
438 - renamed cfs_sleep_chan -> cfs_waitq
439 cfs_sleep_link -> cfs_waitlink
441 - fixed race in linux version of arch-independent socknal
442 (the ENOMEM/EAGAIN decision).
444 - Didn't fix problems in Darwin version of arch-independent socknal
445 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
447 - removed libcfs types from non-socknal header files (only some types
448 in the header files had been changed; the .c files hadn't been