1 11-05-2018 Whamcloud, Inc.
3 * Support for networks:
4 socklnd - any kernel supported by Lustre
5 o2iblnd - OFED from any kernels supported by Lustre
6 o2ilbnd - MLNX_OFED 3.4-2.0.0.0, MLNX_OFED 4.0-1.0.1.0
7 MLNX_OFED 4.0-2.0.0.1, MLNX_OFED 4.1-1.0.2.0
8 MLNX_OFED 4.2-1.0.0.0, MLNX_OFED 4.2-1.2.0.0
9 MLNX_OFED 4.3-1.0.1.0. MLNX_OFED 4.4-1.0.0.0
12 LNet Health: Keep track of network interface health and select
17 * Support for networks:
18 socklnd - any kernel supported by Lustre,
19 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
20 mxlnd - MX 1.2.10 or later
21 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
23 -------------------------------------------------------------------------------
25 09-30-2011 Whamcloud, Inc.
27 * Support for networks:
28 socklnd - any kernel supported by Lustre,
29 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
30 * Available but unsupported:
31 mxlnd - MX 1.2.10 or later
32 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
34 -------------------------------------------------------------------------------
36 2010-07-15 Oracle, Inc.
38 * Support for networks:
39 socklnd - any kernel supported by Lustre,
40 qswlnd - Qsnet kernel modules 5.20 and later,
41 openiblnd - IbGold 1.8.2,
42 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
43 viblnd - Voltaire ibhost 3.4.5 and later,
44 ciblnd - Topspin 3.2.0,
45 iiblnd - Infiniserv 3.3 + PathBits patch,
46 gmlnd - GM 2.1.22 and later,
47 mxlnd - MX 1.2.10 or later,
48 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
52 Description: should update lp_alive for non-router peers
54 Severity : enhancement
56 Description: LNet router shuffler.
58 Severity : enhancement
60 Description: LNet fine grain routing support.
64 Description: router checker stops working when system wall clock goes backward
65 Details : use monotonic timing source instead of system wall clock time.
67 Severity : enhancement
69 Description: avoid asymmetrical router failures
71 Severity : enhancement
73 Description: multiple-instance support for kptllnd
77 Description: ksocknal_close_conn_locked connection race
78 Details : A race was possible when ksocknal_create_conn calls
79 ksocknal_close_conn_locked for already closed conn.
83 Description: router_proc.c is rewritten to use sysctl-interface for parameters
84 residing in /proc/sys/lnet
86 Severity : enhancement
88 Description: port router pinger to userspace
92 Description: kptllnd HELLO protocol deadlock
93 Details : kptllnd HELLO protocol doesn't run to completion in finite time
97 Description: LNet selftest fixes and enhancements
99 Severity : enhancement
101 Description: allow a test node to be a member of multiple test groups
103 Severity : enhancement
105 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
106 Details : an update from the upstream developer Scott Atchley.
108 Severity : enhancement
110 Description: add a new LND optiion to control peer buffer credits on routers
114 Description: Fixing deadlock in usocklnd
115 Details : A deadlock was possible in usocklnd due to race condition while
116 tearing connection down. The problem resulted from erroneous
117 assumption that lnet_finalize() could have been called holding
118 some lnd-level locks.
121 Bugzilla : 13621, 15983
122 Description: Protocol V2 of o2iblnd
123 Details : o2iblnd V2 has several new features:
124 . map-on-demand: map-on-demand is disabled by default, it can
125 be enabled by using modparam "map_on_demand=@value@", @value@
126 should >= 0 and < 256, 0 will disable map-on-demand, any other
127 valid value will enable map-on-demand.
128 Oi2blnd will create FMR or physical MR for RDMA if fragments of
130 Enable map-on-demand will take less memory for new connection,
131 but a little more CPU for RDMA.
132 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
133 are recommanded value. iWARP will probably fail for value >=128.
134 . OOB NOOP message: to resolve deadlock on router.
135 . tunable peer_credits_hiw: (high water to return credits),
136 default value of peer_credits_hiw equals to (peer_credits -1),
137 user can change it between peer_credits/2 and (peer_credits - 1).
138 Lower value is recommended for high latency network.
139 . tunable message queue size: it always equals to peer_credits,
140 higher value is recommended for high latency network.
141 . It's compatible with earlier version of o2iblnd
145 Description: Fixing 'running out of ports' issue
146 Details : Add a delay before next reconnect attempt in ksocklnd in
147 the case of lost race. Limit the frequency of query-requests
148 in lnet. Improved handling of 'dead peer' notifications in
153 Description: Change ptllnd timeout and watchdog timers
154 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
155 Portals wire timeout.
159 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
160 Details : Shared routing enhancements - peer health detection.
162 Severity : enhancement
164 Description: acceptor.c cleanup
165 Details : Code duplication in acceptor.c for the cases of kernel and
166 user-space removed. User-space libcfs tcpip primitives
167 uniformed to have prototypes similar to kernel ones. Minor
168 cosmetic changes in usocklnd to use cfs_socket_t as
169 representation of socket.
173 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
174 Details : See comment 46 in bug 11245 for details - it's indeed a bug
175 introduced by the original 11245 fix.
179 Description: uptllnd credit overflow fix
180 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
185 Description: socklnd protocol version 3
186 Details : With current protocol V2, connections on router can be
187 blocked and can't receive any incoming messages when there is no
188 more router buffer, so ZC-ACK can't be handled (LNet message
189 can't be finalized) and will cause deadlock on router.
190 Protocol V3 has a dedicated connection for emergency messages
191 like ZC-ACK to router, messages on this dedicated connection
192 don't need any credit so will never be blocked. Also, V3 can send
193 keepalive ping in specified period for router healthy checking.
195 -------------------------------------------------------------------------------
197 12-31-2008 Sun Microsystems, Inc.
199 * Support for networks:
200 socklnd - any kernel supported by Lustre,
201 qswlnd - Qsnet kernel modules 5.20 and later,
202 openiblnd - IbGold 1.8.2,
203 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
204 viblnd - Voltaire ibhost 3.4.5 and later,
205 ciblnd - Topspin 3.2.0,
206 iiblnd - Infiniserv 3.3 + PathBits patch,
207 gmlnd - GM 2.1.22 and later,
208 mxlnd - MX 1.2.1 or later,
209 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
213 Description: workaround for OOM from o2iblnd
214 Details : OFED needs allocate big chunk of memory for QP while creating
215 connection for o2iblnd, OOM can happen if no such a contiguous
217 QP size is decided by concurrent_sends and max_fragments of
218 o2iblnd, now we permit user to specify smaller value for
219 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
220 will decrease memory block size required by creating QP.
224 Description: Support Zerocopy receive of Chelsio device
225 Details : Chelsio driver can support zerocopy for iov[1] if it's
226 contiguous and large enough.
230 Description: fix credit flow deadlock in uptllnd
234 Description: finalize network operation in reasonable time
235 Details : conf-sanity test_32a couldn't stop ost and mds because it
236 tried to access non-existent peer and tcp connect took
237 quite long before timing out.
241 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
242 Details : Lost reference on conn prevents peer from being destroyed, which
243 could prevent new peer creation if peer count has reached upper
248 Description: LNET Selftest results in Soft lockup on OSS CPU
249 Details : only hits when 8 or more o2ib clients involved and a session is
250 torn down with 'lst end_session' without preceeding 'lst stop'.
254 Description: concurrent_sends in IB LNDs should not be changeable at run time
255 Details : concurrent_sends in IB LNDs should not be changeable at run time
259 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
260 Details : only hits under out-of-memory situations
263 -------------------------------------------------------------------------------
265 2009-02-07 Sun Microsystems, Inc.
267 * Support for networks:
268 socklnd - any kernel supported by Lustre,
269 qswlnd - Qsnet kernel modules 5.20 and later,
270 openiblnd - IbGold 1.8.2,
271 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
272 viblnd - Voltaire ibhost 3.4.5 and later,
273 ciblnd - Topspin 3.2.0,
274 iiblnd - Infiniserv 3.3 + PathBits patch,
275 gmlnd - GM 2.1.22 and later,
276 mxlnd - MX 1.2.1 or later,
277 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
280 Description: workaround for OOM from o2iblnd
281 Details : OFED needs allocate big chunk of memory for QP while creating
282 connection for o2iblnd, OOM can happen if no such a contiguous
284 QP size is decided by concurrent_sends and max_fragments of
285 o2iblnd, now we permit user to specify smaller value for
286 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
287 will decrease memory block size required by creating QP.
291 Description: Support Zerocopy receive of Chelsio device
292 Details : Chelsio driver can support zerocopy for iov[1] if it's
293 contiguous and large enough.
296 Description: fix credit flow deadlock in uptllnd
300 Description: finalize network operation in reasonable time
301 Details : conf-sanity test_32a couldn't stop ost and mds because it
302 tried to access non-existent peer and tcp connect took
303 quite long before timing out.
307 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
308 Details : Lost reference on conn prevents peer from being destroyed, which
309 could prevent new peer creation if peer count has reached upper
314 Description: LNET Selftest results in Soft lockup on OSS CPU
315 Details : only hits when 8 or more o2ib clients involved and a session is
316 torn down with 'lst end_session' without preceeding 'lst stop'.
320 Description: concurrent_sends in IB LNDs should not be changeable at run time
321 Details : concurrent_sends in IB LNDs should not be changeable at run time
323 -------------------------------------------------------------------------------
325 11-03-2008 Sun Microsystems, Inc.
327 * Support for networks:
328 socklnd - any kernel supported by Lustre,
329 qswlnd - Qsnet kernel modules 5.20 and later,
330 openiblnd - IbGold 1.8.2,
331 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
332 viblnd - Voltaire ibhost 3.4.5 and later,
333 ciblnd - Topspin 3.2.0,
334 iiblnd - Infiniserv 3.3 + PathBits patch,
335 gmlnd - GM 2.1.22 and later,
336 mxlnd - MX 1.2.1 or later,
337 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
341 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
342 Details : only hits under out-of-memory situations
344 -------------------------------------------------------------------------------
346 04-26-2008 Sun Microsystems, Inc.
348 * Support for networks:
349 socklnd - any kernel supported by Lustre,
350 qswlnd - Qsnet kernel modules 5.20 and later,
351 openiblnd - IbGold 1.8.2,
352 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
353 viblnd - Voltaire ibhost 3.4.5 and later,
354 ciblnd - Topspin 3.2.0,
355 iiblnd - Infiniserv 3.3 + PathBits patch,
356 gmlnd - GM 2.1.22 and later,
357 mxlnd - MX 1.2.1 or later,
358 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
362 Description: excessive debug information removed
363 Details : excessive debug information removed
367 Description: ksocknal_create_conn() hit ASSERTION during connection race
368 Details : ksocknal_create_conn() hit ASSERTION during connection race
372 Description: ksocknal_send_hello() hit ASSERTION while connecting race
373 Details : ksocknal_send_hello() hit ASSERTION while connecting race
377 Description: o2iblnd/ptllnd credit deadlock in a routed config.
378 Details : o2iblnd/ptllnd credit deadlock in a routed config.
382 Description: High load after starting lnet
383 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
384 uptime utility reports high load that looks confusingly.
388 Description: ksocklnd fails to establish connection if accept_port is high
389 Details : PID remapping must not be done for active (outgoing) connections
392 --------------------------------------------------------------------------------
394 2008-01-11 Sun Microsystems, Inc.
396 * Support for networks:
397 socklnd - any kernel supported by Lustre,
398 qswlnd - Qsnet kernel modules 5.20 and later,
399 openiblnd - IbGold 1.8.2,
400 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
401 viblnd - Voltaire ibhost 3.4.5 and later,
402 ciblnd - Topspin 3.2.0,
403 iiblnd - Infiniserv 3.3 + PathBits patch,
404 gmlnd - GM 2.1.22 and later,
405 mxlnd - MX 1.2.1 or later,
406 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
410 Description: liblustre network error
411 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
412 variable even if they don't start lnet acceptor.
416 Description: Strange message from lnet (Ignoring prediction from the future)
417 Details : Incorrect calculation of peer's last_alive value in ksocklnd
419 --------------------------------------------------------------------------------
421 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
423 * Support for networks:
424 socklnd - any kernel supported by Lustre,
425 qswlnd - Qsnet kernel modules 5.20 and later,
426 openiblnd - IbGold 1.8.2,
427 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
428 viblnd - Voltaire ibhost 3.4.5 and later,
429 ciblnd - Topspin 3.2.0,
430 iiblnd - Infiniserv 3.3 + PathBits patch,
431 gmlnd - GM 2.1.22 and later,
432 mxlnd - MX 1.2.1 or later,
433 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
437 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
441 Description: increase send queue size for ciblnd/openiblnd
445 Description: new userspace socklnd
446 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
447 with new one - usocklnd.
449 Severity : enhancement
451 Description: Console message flood
452 Details : Make cdls ratelimiting more tunable by adding several tunable in
453 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
454 /proc/sys/lnet/console_backoff.
456 --------------------------------------------------------------------------------
458 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
460 * Support for networks:
461 socklnd - any kernel supported by Lustre,
462 qswlnd - Qsnet kernel modules 5.20 and later,
463 openiblnd - IbGold 1.8.2,
464 o2iblnd - OFED 1.1 and 1.2,
465 viblnd - Voltaire ibhost 3.4.5 and later,
466 ciblnd - Topspin 3.2.0,
467 iiblnd - Infiniserv 3.3 + PathBits patch,
468 gmlnd - GM 2.1.22 and later,
469 mxlnd - MX 1.2.1 or later,
470 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
474 Description: /proc/sys/lnet has non-sysctl entries
475 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
479 Description: TOE Kernel panic by ksocklnd
480 Details : offloaded sockets provide their own implementation of sendpage,
481 can't call tcp_sendpage() directly
485 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
486 Details : races between lnd_shutdown and peer creation prevent
487 lnd_shutdown from finishing.
491 Description: open files rlimit 1024 reached while liblustre testing
492 Details : ulnds/socklnd must close open socket after unsuccessful
497 Description: build error
498 Details : fix typos in gmlnd, ptllnd and viblnd
500 --------------------------------------------------------------------------------
502 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
504 * Support for networks:
505 socklnd - kernels up to 2.6.16,
506 qswlnd - Qsnet kernel modules 5.20 and later,
507 openiblnd - IbGold 1.8.2,
508 o2iblnd - OFED 1.1 and 1.2
509 viblnd - Voltaire ibhost 3.4.5 and later,
510 ciblnd - Topspin 3.2.0,
511 iiblnd - Infiniserv 3.3 + PathBits patch,
512 gmlnd - GM 2.1.22 and later,
513 mxlnd - MX 1.2.1 or later,
514 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
516 --------------------------------------------------------------------------------
518 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
520 * Support for networks:
521 socklnd - kernels up to 2.6.16,
522 qswlnd - Qsnet kernel modules 5.20 and later,
523 openiblnd - IbGold 1.8.2,
525 viblnd - Voltaire ibhost 3.4.5 and later,
526 ciblnd - Topspin 3.2.0,
527 iiblnd - Infiniserv 3.3 + PathBits patch,
528 gmlnd - GM 2.1.22 and later,
529 mxlnd - MX 1.2.1 or later,
530 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
534 Description: Initialize cpumask before use
538 Description: ASSERTION failures when upgrading to the patchless zero-copy
540 Details : This bug affects "rolling upgrades", causing an inconsistent
541 protocol version negotiation and subsequent assertion failure
542 during rolling upgrades after the first wave of upgrades.
546 Details : Change "dropped message" CERRORs to D_NETERROR so they are
547 logged instead of creating "console chatter" when a lustre
548 timeout races with normal RPC completion.
551 Details : lnet_clear_peer_table can wait forever if user forgets to
555 Details : libcfs_id2str should check pid against LNET_PID_ANY.
559 Description: added LNET self test
560 Details : landing b_self_test
565 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
567 Details : do_div() macro is used incorrectly.
569 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
573 Description: make panic on lbug configurable
577 Description: Add OFED1.2 support to o2iblnd
578 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
579 are installed (other than kernel's in-tree infiniband), there
580 could be some problem while insmod o2iblnd (mismatch CRC of
582 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
583 this link provides solution:
584 https://bugs.openfabrics.org/show_bug.cgi?id=355
585 if extra Module.symvers is not supported in kernel, we will
586 have to run the script in bug 12316 to update
587 $LINUX/module.symvers before building o2iblnd.
588 More details about this are in bug 12316.
590 ------------------------------------------------------------------------------
592 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
593 * version 1.4.10 / 1.6.0
594 * Support for networks:
595 socklnd - kernels up to 2.6.16,
596 qswlnd - Qsnet kernel modules 5.20 and later,
597 openiblnd - IbGold 1.8.2,
599 viblnd - Voltaire ibhost 3.4.5 and later,
600 ciblnd - Topspin 3.2.0,
601 iiblnd - Infiniserv 3.3 + PathBits patch,
602 gmlnd - GM 2.1.22 and later,
603 mxlnd - MX 1.2.1 or later,
604 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
608 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
609 possibly accessed in kptllnd_shutdown. Ptllnd should init
610 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
614 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
618 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
622 Description: the_lnet.ln_finalizing was not set when the current thread is
623 about to complete messages. It only affects multi-threaded
629 Description: Changed the default kqswlnd ntxmsg=512
634 Description: Assertion failure in kernel ptllnd caused by posting passive
635 bulk buffers before connection establishment complete.
640 Description: A race in kernel ptllnd between deleting a peer and posting
641 new communications for it could hang communications -
642 manifesting as "Unexpectedly long timeout" messages.
647 Description: Kernel ptllnd lock ordering issue could hang a node.
652 Description: node crash on socket teardown race
655 Frequency : 'lctl peer_list' issued on a mx net
657 Description: Enable lctl's peer_list for MXLND
660 Frequency : after Ptllnd timeouts and portals congestion
662 Description: Credit overflows
663 Details : This was a bug in ptllnd connection establishment. The fix
664 implements better peer stamps to disambiguate connection
665 establishment and ensure both peers enter the credit flow
666 state machine consistently.
671 Description: kptllnd didn't propagate some network errors up to LNET
672 Details : This bug was spotted while investigating 11394. The fix
673 ensures network errors on sends and bulk transfers are
674 propagated to LNET/lustre correctly.
676 Severity : enhancement
678 Description: Fixed console chatter in case of -ETIMEDOUT.
680 Severity : enhancement
682 Description: Added D_NETTRACE for recording network packet history
683 (initially only for ptllnd). Also a separate userspace
684 ptllnd facility to gather history which should really be
685 covered by D_NETTRACE too, if only CDEBUG recorded history in
691 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
692 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
693 callback can occur before a connection has actually been
694 established. This caused an assertion failure previously.
696 Severity : enhancement
698 Description: Multiple instances for o2iblnd
699 Details : Allow multiple instances of o2iblnd to enable networking over
700 multiple HCAs and routing between them.
704 Description: lnet deadlock in router_checker
705 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
706 into BH locks to eliminate potential deadlock caused by
707 ksocknal_data_ready() preempting code holding these locks.
711 Description: Millions of failed socklnd connection attempts cause a very slow FS
712 Details : added a new route flag ksnr_scheduled to distinguish from
713 ksnr_connecting, so that a peer connection request is only turned
714 down for race concerns when an active connection to the same peer
715 is under progress (instead of just being scheduled).
717 ------------------------------------------------------------------------------
719 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
721 * Support for networks:
722 socklnd - kernels up to 2.6.16
723 qswlnd - Qsnet kernel modules 5.20 and later
724 openiblnd - IbGold 1.8.2
726 viblnd - Voltaire ibhost 3.4.5 and later
727 ciblnd - Topspin 3.2.0
728 iiblnd - Infiniserv 3.3 + PathBits patch
729 gmlnd - GM 2.1.22 and later
730 mxlnd - MX 1.2.1 or later
731 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
734 Severity : major on XT3
736 Description: libcfs overwrites /proc/sys/portals
737 Details : libcfs created a symlink from /proc/sys/portals to
738 /proc/sys/lnet for backwards compatibility. This is no
739 longer required and makes the Cray portals /proc variables
744 Description: OFED FMR API change
745 Details : This changes parameter usage to reflect a change in
746 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
747 that FMR support is only used in experimental versions of the
748 o2iblnd - this change does not affect standard usage at all.
750 Severity : enhancement
752 Description: new ko2iblnd module parameter: ib_mtu
753 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
754 HCAs. You can avoid this problem by setting the MTU to 1024
755 using this module parameter.
757 Severity : enhancement
758 Bugzilla : 11118/11620
759 Description: ptllnd small request message buffer alignment fix
760 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
761 Round up small message size on sends in case this option
762 is not supported. 11620 was a defect in the initial
763 implementation which effectively asserted all peers had to be
764 running the correct protocol version which was fixed by always
765 NAK-ing such requests and handling any misalignments they
770 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
771 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
772 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
774 Severity : enhancement
776 Description: Patchless ZC(zero copy) socklnd
777 Details : New protocol for socklnd, socklnd can support zero copy without
778 kernel patch, it's compatible with old socklnd. Checksum is
779 moved from tunables to modparams.
783 Description: When ksocknal_del_peer() is called upon a peer whose
784 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
785 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
788 Frequency : when ptlrpc is under heavy use and runs out of request buffer
790 Description: In lnet_match_blocked_msg(), md can be used without holding a
794 Frequency : very rarely
796 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
797 If connd connects a route which has been closed by
798 ksocknal_shutdown(), ksocknal_create_routes() may create new
799 routes which hold references on the peer, causing shutdown
800 process to wait for peer to disappear forever.
802 Severity : enhancement
804 Description: Dump XT3 portals traces on kptllnd timeout
805 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
806 dump Cray portals debug traces to a file. The kptllnd module
807 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
808 is the basename of the dump file.
811 Frequency : infrequent
813 Description: kernel ptllnd fix bug in connection re-establishment
814 Details : Kernel ptllnd could produce protocol errors e.g. illegal
815 matchbits and/or violate the credit flow protocol when trying
816 to re-establish a connection with a peer after an error or
819 Severity : enhancement
821 Description: Allow /proc/sys/lnet/debug to be set symbolically
822 Details : Allow debug and subsystem debug values to be read/set by name
823 in addition to numerically, for ease of use.
826 Frequency : only in configurations with LNET routers
828 Description: routes automatically marked down and recovered
829 Details : In configurations with LNET routers if a router fails routers
830 now actively try to recover routes that are down, unless they
831 are marked down by an administrator.
833 ------------------------------------------------------------------------------
835 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
838 Frequency : very rarely, in configurations with LNET routers and TCP
840 Description: incorrect data written to files on OSTs
841 Details : In certain high-load conditions incorrect data may be written
842 to files on the OST when using TCP networks.
844 ------------------------------------------------------------------------------
846 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
848 - rework CDEBUG messages rate-limiting mechanism b=10375
849 - add per-socket tunables for socklnd if the kernel is patched b=10327
851 ------------------------------------------------------------------------------
853 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
855 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
856 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
857 - quiet LNET startup LNI message for liblustre b=10128
858 - Better console error messages if 'ip2nets' can't match an IP address
859 - Fixed overflow/use-before-set bugs in linux-time.h
860 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
861 - LNET teardown failed an assertion about the route table being empty
862 - Fixed a crash in LNetEQPoll(<invalid handle>)
863 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
864 - improve debug message for liblustre/Catamount nodes (b=10116)
866 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
867 * Configuration change for the XT3
868 The PTLLND is now used to run Lustre over Portals on the XT3.
869 The configure option(s) --with-cray-portals are no longer
870 used. Rather --with-portals=<path-to-portals-includes> is
871 used to enable building on the XT3. In addition to enable
872 XT3 specific features the option --enable-cray-xt3 must be
875 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
876 * Portals has been removed, replaced by LNET.
877 LNET is new networking infrastructure for Lustre, it includes a
878 reorganized network configuration mode (see the user
879 documentation for full details) as well as support for routing
880 between different network fabrics. Lustre Networking Devices
881 (LNDS) for the supported network fabrics have also been created
882 for this new infrastructure.
884 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
889 Frequency : rare (large Voltaire clusters only)
891 Description: the default number of reserved transmit descriptors was too low
892 for some large clusters
893 Details : As a workaround, the number was increased. A proper fix includes
896 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
901 Frequency : occasional (large-scale events, cluster reboot, network failure)
903 Description: too many error messages on console obscure actual problem and
904 can slow down/panic server, or cause recovery to fail repeatedly
905 Details : enable rate-limiting of console error messages, and some messages
906 that were console errors now only go to the kernel log
908 Severity : enhancement
910 Description: add /proc/sys/portals/catastrophe entry which will report if
911 that node has previously LBUGged
913 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
915 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
917 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
919 - handle error return code in kranal_check_fma_rx() (5915,6054)
921 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
923 - update vibnal (Voltaire IB NAL)
924 - update gmnal (Myrinet NAL), gmnalid
926 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
928 * Landed portals:b_port_step as follows...
930 - removed CFS_DECL_SPIN*
931 just use 'spinlock_t' and initialise with spin_lock_init()
933 - removed CFS_DECL_MUTEX*
934 just use 'struct semaphore' and initialise with init_mutex()
936 - removed CFS_DECL_RWSEM*
937 just use 'struct rw_semaphore' and initialise with init_rwsem()
939 - renamed cfs_sleep_chan -> cfs_waitq
940 cfs_sleep_link -> cfs_waitlink
942 - fixed race in linux version of arch-independent socknal
943 (the ENOMEM/EAGAIN decision).
945 - Didn't fix problems in Darwin version of arch-independent socknal
946 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
948 - removed libcfs types from non-socknal header files (only some types
949 in the header files had been changed; the .c files hadn't been