1 tbd Sun Microsystems, Inc.
3 * Support for networks:
4 socklnd - any kernel supported by Lustre,
5 qswlnd - Qsnet kernel modules 5.20 and later,
6 openiblnd - IbGold 1.8.2,
7 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
8 viblnd - Voltaire ibhost 3.4.5 and later,
9 ciblnd - Topspin 3.2.0,
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
22 Description: workaround for OOM from o2iblnd
23 Details : OFED needs allocate big chunk of memory for QP while creating
24 connection for o2iblnd, OOM can happen if no such a contiguous
26 QP size is decided by concurrent_sends and max_fragments of
27 o2iblnd, now we permit user to specify smaller value for
28 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
29 will decrease memory block size required by creating QP.
33 Description: Support Zerocopy receive of Chelsio device
34 Details : Chelsio driver can support zerocopy for iov[1] if it's
35 contiguous and large enough.
39 Description: fix credit flow deadlock in uptllnd
43 Description: finalize network operation in reasonable time
44 Details : conf-sanity test_32a couldn't stop ost and mds because it
45 tried to access non-existent peer and tcp connect took
46 quite long before timing out.
50 Description: Remove portals compatibility
51 Details : Remove portals compatibility, not interoperable with releases
56 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
57 Details : Lost reference on conn prevents peer from being destroyed, which
58 could prevent new peer creation if peer count has reached upper
63 Description: LNET Selftest results in Soft lockup on OSS CPU
64 Details : only hits when 8 or more o2ib clients involved and a session is
65 torn down with 'lst end_session' without preceeding 'lst stop'.
69 Description: concurrent_sends in IB LNDs should not be changeable at run time
70 Details : concurrent_sends in IB LNDs should not be changeable at run time
74 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
75 Details : only hits under out-of-memory situations
78 -------------------------------------------------------------------------------
81 04-26-2008 Sun Microsystems, Inc.
83 * Support for networks:
84 socklnd - any kernel supported by Lustre,
85 qswlnd - Qsnet kernel modules 5.20 and later,
86 openiblnd - IbGold 1.8.2,
87 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
88 viblnd - Voltaire ibhost 3.4.5 and later,
89 ciblnd - Topspin 3.2.0,
90 iiblnd - Infiniserv 3.3 + PathBits patch,
91 gmlnd - GM 2.1.22 and later,
92 mxlnd - MX 1.2.1 or later,
93 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
97 Description: excessive debug information removed
98 Details : excessive debug information removed
102 Description: ksocknal_create_conn() hit ASSERTION during connection race
103 Details : ksocknal_create_conn() hit ASSERTION during connection race
107 Description: ksocknal_send_hello() hit ASSERTION while connecting race
108 Details : ksocknal_send_hello() hit ASSERTION while connecting race
112 Description: o2iblnd/ptllnd credit deadlock in a routed config.
113 Details : o2iblnd/ptllnd credit deadlock in a routed config.
117 Description: High load after starting lnet
118 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
119 uptime utility reports high load that looks confusingly.
123 Description: ksocklnd fails to establish connection if accept_port is high
124 Details : PID remapping must not be done for active (outgoing) connections
126 --------------------------------------------------------------------------------
128 2008-01-11 Sun Microsystems, Inc.
130 * Support for networks:
131 socklnd - any kernel supported by Lustre,
132 qswlnd - Qsnet kernel modules 5.20 and later,
133 openiblnd - IbGold 1.8.2,
134 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
135 viblnd - Voltaire ibhost 3.4.5 and later,
136 ciblnd - Topspin 3.2.0,
137 iiblnd - Infiniserv 3.3 + PathBits patch,
138 gmlnd - GM 2.1.22 and later,
139 mxlnd - MX 1.2.1 or later,
140 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
143 Description: liblustre network error
144 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
145 variable even if they don't start lnet acceptor.
149 Description: Strange message from lnet (Ignoring prediction from the future)
150 Details : Incorrect calculation of peer's last_alive value in ksocklnd
152 --------------------------------------------------------------------------------
154 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
156 * Support for networks:
157 socklnd - any kernel supported by Lustre,
158 qswlnd - Qsnet kernel modules 5.20 and later,
159 openiblnd - IbGold 1.8.2,
160 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
161 viblnd - Voltaire ibhost 3.4.5 and later,
162 ciblnd - Topspin 3.2.0,
163 iiblnd - Infiniserv 3.3 + PathBits patch,
164 gmlnd - GM 2.1.22 and later,
165 mxlnd - MX 1.2.1 or later,
166 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
170 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
174 Description: increase send queue size for ciblnd/openiblnd
178 Description: new userspace socklnd
179 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
180 with new one - usocklnd.
182 Severity : enhancement
184 Description: Console message flood
185 Details : Make cdls ratelimiting more tunable by adding several tunable in
186 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
187 /proc/sys/lnet/console_backoff.
189 --------------------------------------------------------------------------------
191 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
193 * Support for networks:
194 socklnd - any kernel supported by Lustre,
195 qswlnd - Qsnet kernel modules 5.20 and later,
196 openiblnd - IbGold 1.8.2,
197 o2iblnd - OFED 1.1 and 1.2,
198 viblnd - Voltaire ibhost 3.4.5 and later,
199 ciblnd - Topspin 3.2.0,
200 iiblnd - Infiniserv 3.3 + PathBits patch,
201 gmlnd - GM 2.1.22 and later,
202 mxlnd - MX 1.2.1 or later,
203 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
207 Description: /proc/sys/lnet has non-sysctl entries
208 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
212 Description: TOE Kernel panic by ksocklnd
213 Details : offloaded sockets provide their own implementation of sendpage,
214 can't call tcp_sendpage() directly
218 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
219 Details : races between lnd_shutdown and peer creation prevent
220 lnd_shutdown from finishing.
224 Description: open files rlimit 1024 reached while liblustre testing
225 Details : ulnds/socklnd must close open socket after unsuccessful
230 Description: build error
231 Details : fix typos in gmlnd, ptllnd and viblnd
233 ------------------------------------------------------------------------------
235 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
237 * Support for networks:
238 socklnd - kernels up to 2.6.16,
239 qswlnd - Qsnet kernel modules 5.20 and later,
240 openiblnd - IbGold 1.8.2,
241 o2iblnd - OFED 1.1 and 1.2
242 viblnd - Voltaire ibhost 3.4.5 and later,
243 ciblnd - Topspin 3.2.0,
244 iiblnd - Infiniserv 3.3 + PathBits patch,
245 gmlnd - GM 2.1.22 and later,
246 mxlnd - MX 1.2.1 or later,
247 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
249 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
251 * Support for networks:
252 socklnd - kernels up to 2.6.16,
253 qswlnd - Qsnet kernel modules 5.20 and later,
254 openiblnd - IbGold 1.8.2,
256 viblnd - Voltaire ibhost 3.4.5 and later,
257 ciblnd - Topspin 3.2.0,
258 iiblnd - Infiniserv 3.3 + PathBits patch,
259 gmlnd - GM 2.1.22 and later,
260 mxlnd - MX 1.2.1 or later,
261 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
265 Description: Initialize cpumask before use
269 Description: ASSERTION failures when upgrading to the patchless zero-copy
271 Details : This bug affects "rolling upgrades", causing an inconsistent
272 protocol version negotiation and subsequent assertion failure
273 during rolling upgrades after the first wave of upgrades.
277 Details : Change "dropped message" CERRORs to D_NETERROR so they are
278 logged instead of creating "console chatter" when a lustre
279 timeout races with normal RPC completion.
282 Details : lnet_clear_peer_table can wait forever if user forgets to
286 Details : libcfs_id2str should check pid against LNET_PID_ANY.
290 Description: added LNET self test
291 Details : landing b_self_test
296 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
298 Details : do_div() macro is used incorrectly.
300 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
304 Description: make panic on lbug configurable
308 Description: Add OFED1.2 support to o2iblnd
309 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
310 are installed (other than kernel's in-tree infiniband), there
311 could be some problem while insmod o2iblnd (mismatch CRC of
313 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
314 this link provides solution:
315 https://bugs.openfabrics.org/show_bug.cgi?id=355
316 if extra Module.symvers is not supported in kernel, we will
317 have to run the script in bug 12316 to update
318 $LINUX/module.symvers before building o2iblnd.
319 More details about this are in bug 12316.
321 ------------------------------------------------------------------------------
323 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
324 * version 1.4.10 / 1.6.0
325 * Support for networks:
326 socklnd - kernels up to 2.6.16,
327 qswlnd - Qsnet kernel modules 5.20 and later,
328 openiblnd - IbGold 1.8.2,
330 viblnd - Voltaire ibhost 3.4.5 and later,
331 ciblnd - Topspin 3.2.0,
332 iiblnd - Infiniserv 3.3 + PathBits patch,
333 gmlnd - GM 2.1.22 and later,
334 mxlnd - MX 1.2.1 or later,
335 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
339 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
340 possibly accessed in kptllnd_shutdown. Ptllnd should init
341 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
345 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
349 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
353 Description: the_lnet.ln_finalizing was not set when the current thread is
354 about to complete messages. It only affects multi-threaded
360 Description: Changed the default kqswlnd ntxmsg=512
365 Description: Assertion failure in kernel ptllnd caused by posting passive
366 bulk buffers before connection establishment complete.
371 Description: A race in kernel ptllnd between deleting a peer and posting
372 new communications for it could hang communications -
373 manifesting as "Unexpectedly long timeout" messages.
378 Description: Kernel ptllnd lock ordering issue could hang a node.
383 Description: node crash on socket teardown race
386 Frequency : 'lctl peer_list' issued on a mx net
388 Description: Enable lctl's peer_list for MXLND
391 Frequency : after Ptllnd timeouts and portals congestion
393 Description: Credit overflows
394 Details : This was a bug in ptllnd connection establishment. The fix
395 implements better peer stamps to disambiguate connection
396 establishment and ensure both peers enter the credit flow
397 state machine consistently.
402 Description: kptllnd didn't propagate some network errors up to LNET
403 Details : This bug was spotted while investigating 11394. The fix
404 ensures network errors on sends and bulk transfers are
405 propagated to LNET/lustre correctly.
407 Severity : enhancement
409 Description: Fixed console chatter in case of -ETIMEDOUT.
411 Severity : enhancement
413 Description: Added D_NETTRACE for recording network packet history
414 (initially only for ptllnd). Also a separate userspace
415 ptllnd facility to gather history which should really be
416 covered by D_NETTRACE too, if only CDEBUG recorded history in
422 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
423 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
424 callback can occur before a connection has actually been
425 established. This caused an assertion failure previously.
427 Severity : enhancement
429 Description: Multiple instances for o2iblnd
430 Details : Allow multiple instances of o2iblnd to enable networking over
431 multiple HCAs and routing between them.
435 Description: lnet deadlock in router_checker
436 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
437 into BH locks to eliminate potential deadlock caused by
438 ksocknal_data_ready() preempting code holding these locks.
442 Description: Millions of failed socklnd connection attempts cause a very slow FS
443 Details : added a new route flag ksnr_scheduled to distinguish from
444 ksnr_connecting, so that a peer connection request is only turned
445 down for race concerns when an active connection to the same peer
446 is under progress (instead of just being scheduled).
448 ------------------------------------------------------------------------------
450 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
452 * Support for networks:
453 socklnd - kernels up to 2.6.16
454 qswlnd - Qsnet kernel modules 5.20 and later
455 openiblnd - IbGold 1.8.2
457 viblnd - Voltaire ibhost 3.4.5 and later
458 ciblnd - Topspin 3.2.0
459 iiblnd - Infiniserv 3.3 + PathBits patch
460 gmlnd - GM 2.1.22 and later
461 mxlnd - MX 1.2.1 or later
462 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
465 Severity : major on XT3
467 Description: libcfs overwrites /proc/sys/portals
468 Details : libcfs created a symlink from /proc/sys/portals to
469 /proc/sys/lnet for backwards compatibility. This is no
470 longer required and makes the Cray portals /proc variables
475 Description: OFED FMR API change
476 Details : This changes parameter usage to reflect a change in
477 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
478 that FMR support is only used in experimental versions of the
479 o2iblnd - this change does not affect standard usage at all.
481 Severity : enhancement
483 Description: new ko2iblnd module parameter: ib_mtu
484 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
485 HCAs. You can avoid this problem by setting the MTU to 1024
486 using this module parameter.
488 Severity : enhancement
489 Bugzilla : 11118/11620
490 Description: ptllnd small request message buffer alignment fix
491 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
492 Round up small message size on sends in case this option
493 is not supported. 11620 was a defect in the initial
494 implementation which effectively asserted all peers had to be
495 running the correct protocol version which was fixed by always
496 NAK-ing such requests and handling any misalignments they
501 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
502 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
503 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
505 Severity : enhancement
507 Description: Patchless ZC(zero copy) socklnd
508 Details : New protocol for socklnd, socklnd can support zero copy without
509 kernel patch, it's compatible with old socklnd. Checksum is
510 moved from tunables to modparams.
514 Description: When ksocknal_del_peer() is called upon a peer whose
515 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
516 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
519 Frequency : when ptlrpc is under heavy use and runs out of request buffer
521 Description: In lnet_match_blocked_msg(), md can be used without holding a
525 Frequency : very rarely
527 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
528 If connd connects a route which has been closed by
529 ksocknal_shutdown(), ksocknal_create_routes() may create new
530 routes which hold references on the peer, causing shutdown
531 process to wait for peer to disappear forever.
533 Severity : enhancement
535 Description: Dump XT3 portals traces on kptllnd timeout
536 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
537 dump Cray portals debug traces to a file. The kptllnd module
538 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
539 is the basename of the dump file.
542 Frequency : infrequent
544 Description: kernel ptllnd fix bug in connection re-establishment
545 Details : Kernel ptllnd could produce protocol errors e.g. illegal
546 matchbits and/or violate the credit flow protocol when trying
547 to re-establish a connection with a peer after an error or
550 Severity : enhancement
552 Description: Allow /proc/sys/lnet/debug to be set symbolically
553 Details : Allow debug and subsystem debug values to be read/set by name
554 in addition to numerically, for ease of use.
557 Frequency : only in configurations with LNET routers
559 Description: routes automatically marked down and recovered
560 Details : In configurations with LNET routers if a router fails routers
561 now actively try to recover routes that are down, unless they
562 are marked down by an administrator.
564 ------------------------------------------------------------------------------
566 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
569 Frequency : very rarely, in configurations with LNET routers and TCP
571 Description: incorrect data written to files on OSTs
572 Details : In certain high-load conditions incorrect data may be written
573 to files on the OST when using TCP networks.
575 ------------------------------------------------------------------------------
577 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
579 - rework CDEBUG messages rate-limiting mechanism b=10375
580 - add per-socket tunables for socklnd if the kernel is patched b=10327
582 ------------------------------------------------------------------------------
584 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
586 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
587 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
588 - quiet LNET startup LNI message for liblustre b=10128
589 - Better console error messages if 'ip2nets' can't match an IP address
590 - Fixed overflow/use-before-set bugs in linux-time.h
591 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
592 - LNET teardown failed an assertion about the route table being empty
593 - Fixed a crash in LNetEQPoll(<invalid handle>)
594 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
595 - improve debug message for liblustre/Catamount nodes (b=10116)
597 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
598 * Configuration change for the XT3
599 The PTLLND is now used to run Lustre over Portals on the XT3.
600 The configure option(s) --with-cray-portals are no longer
601 used. Rather --with-portals=<path-to-portals-includes> is
602 used to enable building on the XT3. In addition to enable
603 XT3 specific features the option --enable-cray-xt3 must be
606 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
607 * Portals has been removed, replaced by LNET.
608 LNET is new networking infrastructure for Lustre, it includes a
609 reorganized network configuration mode (see the user
610 documentation for full details) as well as support for routing
611 between different network fabrics. Lustre Networking Devices
612 (LNDS) for the supported network fabrics have also been created
613 for this new infrastructure.
615 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
620 Frequency : rare (large Voltaire clusters only)
622 Description: the default number of reserved transmit descriptors was too low
623 for some large clusters
624 Details : As a workaround, the number was increased. A proper fix includes
627 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
632 Frequency : occasional (large-scale events, cluster reboot, network failure)
634 Description: too many error messages on console obscure actual problem and
635 can slow down/panic server, or cause recovery to fail repeatedly
636 Details : enable rate-limiting of console error messages, and some messages
637 that were console errors now only go to the kernel log
639 Severity : enhancement
641 Description: add /proc/sys/portals/catastrophe entry which will report if
642 that node has previously LBUGged
644 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
646 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
648 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
650 - handle error return code in kranal_check_fma_rx() (5915,6054)
652 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
654 - update vibnal (Voltaire IB NAL)
655 - update gmnal (Myrinet NAL), gmnalid
657 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
659 * Landed portals:b_port_step as follows...
661 - removed CFS_DECL_SPIN*
662 just use 'spinlock_t' and initialise with spin_lock_init()
664 - removed CFS_DECL_MUTEX*
665 just use 'struct semaphore' and initialise with init_mutex()
667 - removed CFS_DECL_RWSEM*
668 just use 'struct rw_semaphore' and initialise with init_rwsem()
670 - renamed cfs_sleep_chan -> cfs_waitq
671 cfs_sleep_link -> cfs_waitlink
673 - fixed race in linux version of arch-independent socknal
674 (the ENOMEM/EAGAIN decision).
676 - Didn't fix problems in Darwin version of arch-independent socknal
677 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
679 - removed libcfs types from non-socknal header files (only some types
680 in the header files had been changed; the .c files hadn't been