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, 1.3, and 1.4.1
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: kptllnd HELLO protocol deadlock
23 Details : kptllnd HELLO protocol doesn't run to completion in finite time
27 Description: LNet selftest fixes and enhancements
29 Severity : enhancement
31 Description: allow a test node to be a member of multiple test groups
33 Severity : enhancement
35 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
36 Details : an update from the upstream developer Scott Atchley.
38 Severity : enhancement
40 Description: add a new LND optiion to control peer buffer credits on routers
44 Description: Fixing deadlock in usocklnd
45 Details : A deadlock was possible in usocklnd due to race condition while
46 tearing connection down. The problem resulted from erroneous
47 assumption that lnet_finalize() could have been called holding
51 Bugzilla : 13621, 15983
52 Description: Protocol V2 of o2iblnd
53 Details : o2iblnd V2 has several new features:
54 . map-on-demand: map-on-demand is disabled by default, it can
55 be enabled by using modparam "map_on_demand=@value@", @value@
56 should >= 0 and < 256, 0 will disable map-on-demand, any other
57 valid value will enable map-on-demand.
58 Oi2blnd will create FMR or physical MR for RDMA if fragments of
60 Enable map-on-demand will take less memory for new connection,
61 but a little more CPU for RDMA.
62 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
63 are recommanded value. iWARP will probably fail for value >=128.
64 . OOB NOOP message: to resolve deadlock on router.
65 . tunable peer_credits_hiw: (high water to return credits),
66 default value of peer_credits_hiw equals to (peer_credits -1),
67 user can change it between peer_credits/2 and (peer_credits - 1).
68 Lower value is recommended for high latency network.
69 . tunable message queue size: it always equals to peer_credits,
70 higher value is recommended for high latency network.
71 . It's compatible with earlier version of o2iblnd
75 Description: Fixing 'running out of ports' issue
76 Details : Add a delay before next reconnect attempt in ksocklnd in
77 the case of lost race. Limit the frequency of query-requests
78 in lnet. Improved handling of 'dead peer' notifications in
83 Description: Change ptllnd timeout and watchdog timers
84 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
89 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
90 Details : Shared routing enhancements - peer health detection.
92 Severity : enhancement
94 Description: acceptor.c cleanup
95 Details : Code duplication in acceptor.c for the cases of kernel and
96 user-space removed. User-space libcfs tcpip primitives
97 uniformed to have prototypes similar to kernel ones. Minor
98 cosmetic changes in usocklnd to use cfs_socket_t as
99 representation of socket.
103 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
104 Details : See comment 46 in bug 11245 for details - it's indeed a bug
105 introduced by the original 11245 fix.
109 Description: uptllnd credit overflow fix
110 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
115 Description: socklnd protocol version 3
116 Details : With current protocol V2, connections on router can be
117 blocked and can't receive any incoming messages when there is no
118 more router buffer, so ZC-ACK can't be handled (LNet message
119 can't be finalized) and will cause deadlock on router.
120 Protocol V3 has a dedicated connection for emergency messages
121 like ZC-ACK to router, messages on this dedicated connection
122 don't need any credit so will never be blocked. Also, V3 can send
123 keepalive ping in specified period for router healthy checking.
125 -------------------------------------------------------------------------------
127 12-31-2008 Sun Microsystems, Inc.
129 * Support for networks:
130 socklnd - any kernel supported by Lustre,
131 qswlnd - Qsnet kernel modules 5.20 and later,
132 openiblnd - IbGold 1.8.2,
133 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
134 viblnd - Voltaire ibhost 3.4.5 and later,
135 ciblnd - Topspin 3.2.0,
136 iiblnd - Infiniserv 3.3 + PathBits patch,
137 gmlnd - GM 2.1.22 and later,
138 mxlnd - MX 1.2.1 or later,
139 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
143 Description: workaround for OOM from o2iblnd
144 Details : OFED needs allocate big chunk of memory for QP while creating
145 connection for o2iblnd, OOM can happen if no such a contiguous
147 QP size is decided by concurrent_sends and max_fragments of
148 o2iblnd, now we permit user to specify smaller value for
149 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
150 will decrease memory block size required by creating QP.
154 Description: Support Zerocopy receive of Chelsio device
155 Details : Chelsio driver can support zerocopy for iov[1] if it's
156 contiguous and large enough.
160 Description: fix credit flow deadlock in uptllnd
164 Description: finalize network operation in reasonable time
165 Details : conf-sanity test_32a couldn't stop ost and mds because it
166 tried to access non-existent peer and tcp connect took
167 quite long before timing out.
171 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
172 Details : Lost reference on conn prevents peer from being destroyed, which
173 could prevent new peer creation if peer count has reached upper
178 Description: LNET Selftest results in Soft lockup on OSS CPU
179 Details : only hits when 8 or more o2ib clients involved and a session is
180 torn down with 'lst end_session' without preceeding 'lst stop'.
184 Description: concurrent_sends in IB LNDs should not be changeable at run time
185 Details : concurrent_sends in IB LNDs should not be changeable at run time
189 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
190 Details : only hits under out-of-memory situations
193 -------------------------------------------------------------------------------
195 2009-02-07 Sun Microsystems, Inc.
197 * Support for networks:
198 socklnd - any kernel supported by Lustre,
199 qswlnd - Qsnet kernel modules 5.20 and later,
200 openiblnd - IbGold 1.8.2,
201 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
202 viblnd - Voltaire ibhost 3.4.5 and later,
203 ciblnd - Topspin 3.2.0,
204 iiblnd - Infiniserv 3.3 + PathBits patch,
205 gmlnd - GM 2.1.22 and later,
206 mxlnd - MX 1.2.1 or later,
207 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
210 Description: workaround for OOM from o2iblnd
211 Details : OFED needs allocate big chunk of memory for QP while creating
212 connection for o2iblnd, OOM can happen if no such a contiguous
214 QP size is decided by concurrent_sends and max_fragments of
215 o2iblnd, now we permit user to specify smaller value for
216 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
217 will decrease memory block size required by creating QP.
221 Description: Support Zerocopy receive of Chelsio device
222 Details : Chelsio driver can support zerocopy for iov[1] if it's
223 contiguous and large enough.
226 Description: fix credit flow deadlock in uptllnd
230 Description: finalize network operation in reasonable time
231 Details : conf-sanity test_32a couldn't stop ost and mds because it
232 tried to access non-existent peer and tcp connect took
233 quite long before timing out.
237 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
238 Details : Lost reference on conn prevents peer from being destroyed, which
239 could prevent new peer creation if peer count has reached upper
244 Description: LNET Selftest results in Soft lockup on OSS CPU
245 Details : only hits when 8 or more o2ib clients involved and a session is
246 torn down with 'lst end_session' without preceeding 'lst stop'.
250 Description: concurrent_sends in IB LNDs should not be changeable at run time
251 Details : concurrent_sends in IB LNDs should not be changeable at run time
253 -------------------------------------------------------------------------------
255 11-03-2008 Sun Microsystems, Inc.
257 * Support for networks:
258 socklnd - any kernel supported by Lustre,
259 qswlnd - Qsnet kernel modules 5.20 and later,
260 openiblnd - IbGold 1.8.2,
261 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
262 viblnd - Voltaire ibhost 3.4.5 and later,
263 ciblnd - Topspin 3.2.0,
264 iiblnd - Infiniserv 3.3 + PathBits patch,
265 gmlnd - GM 2.1.22 and later,
266 mxlnd - MX 1.2.1 or later,
267 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
271 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
272 Details : only hits under out-of-memory situations
274 -------------------------------------------------------------------------------
276 04-26-2008 Sun Microsystems, Inc.
278 * Support for networks:
279 socklnd - any kernel supported by Lustre,
280 qswlnd - Qsnet kernel modules 5.20 and later,
281 openiblnd - IbGold 1.8.2,
282 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
283 viblnd - Voltaire ibhost 3.4.5 and later,
284 ciblnd - Topspin 3.2.0,
285 iiblnd - Infiniserv 3.3 + PathBits patch,
286 gmlnd - GM 2.1.22 and later,
287 mxlnd - MX 1.2.1 or later,
288 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
292 Description: excessive debug information removed
293 Details : excessive debug information removed
297 Description: ksocknal_create_conn() hit ASSERTION during connection race
298 Details : ksocknal_create_conn() hit ASSERTION during connection race
302 Description: ksocknal_send_hello() hit ASSERTION while connecting race
303 Details : ksocknal_send_hello() hit ASSERTION while connecting race
307 Description: o2iblnd/ptllnd credit deadlock in a routed config.
308 Details : o2iblnd/ptllnd credit deadlock in a routed config.
312 Description: High load after starting lnet
313 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
314 uptime utility reports high load that looks confusingly.
318 Description: ksocklnd fails to establish connection if accept_port is high
319 Details : PID remapping must not be done for active (outgoing) connections
322 --------------------------------------------------------------------------------
324 2008-01-11 Sun Microsystems, Inc.
326 * Support for networks:
327 socklnd - any kernel supported by Lustre,
328 qswlnd - Qsnet kernel modules 5.20 and later,
329 openiblnd - IbGold 1.8.2,
330 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
331 viblnd - Voltaire ibhost 3.4.5 and later,
332 ciblnd - Topspin 3.2.0,
333 iiblnd - Infiniserv 3.3 + PathBits patch,
334 gmlnd - GM 2.1.22 and later,
335 mxlnd - MX 1.2.1 or later,
336 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
340 Description: liblustre network error
341 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
342 variable even if they don't start lnet acceptor.
346 Description: Strange message from lnet (Ignoring prediction from the future)
347 Details : Incorrect calculation of peer's last_alive value in ksocklnd
349 --------------------------------------------------------------------------------
351 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
353 * Support for networks:
354 socklnd - any kernel supported by Lustre,
355 qswlnd - Qsnet kernel modules 5.20 and later,
356 openiblnd - IbGold 1.8.2,
357 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
358 viblnd - Voltaire ibhost 3.4.5 and later,
359 ciblnd - Topspin 3.2.0,
360 iiblnd - Infiniserv 3.3 + PathBits patch,
361 gmlnd - GM 2.1.22 and later,
362 mxlnd - MX 1.2.1 or later,
363 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
367 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
371 Description: increase send queue size for ciblnd/openiblnd
375 Description: new userspace socklnd
376 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
377 with new one - usocklnd.
379 Severity : enhancement
381 Description: Console message flood
382 Details : Make cdls ratelimiting more tunable by adding several tunable in
383 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
384 /proc/sys/lnet/console_backoff.
386 --------------------------------------------------------------------------------
388 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
390 * Support for networks:
391 socklnd - any kernel supported by Lustre,
392 qswlnd - Qsnet kernel modules 5.20 and later,
393 openiblnd - IbGold 1.8.2,
394 o2iblnd - OFED 1.1 and 1.2,
395 viblnd - Voltaire ibhost 3.4.5 and later,
396 ciblnd - Topspin 3.2.0,
397 iiblnd - Infiniserv 3.3 + PathBits patch,
398 gmlnd - GM 2.1.22 and later,
399 mxlnd - MX 1.2.1 or later,
400 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
404 Description: /proc/sys/lnet has non-sysctl entries
405 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
409 Description: TOE Kernel panic by ksocklnd
410 Details : offloaded sockets provide their own implementation of sendpage,
411 can't call tcp_sendpage() directly
415 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
416 Details : races between lnd_shutdown and peer creation prevent
417 lnd_shutdown from finishing.
421 Description: open files rlimit 1024 reached while liblustre testing
422 Details : ulnds/socklnd must close open socket after unsuccessful
427 Description: build error
428 Details : fix typos in gmlnd, ptllnd and viblnd
430 --------------------------------------------------------------------------------
432 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
434 * Support for networks:
435 socklnd - kernels up to 2.6.16,
436 qswlnd - Qsnet kernel modules 5.20 and later,
437 openiblnd - IbGold 1.8.2,
438 o2iblnd - OFED 1.1 and 1.2
439 viblnd - Voltaire ibhost 3.4.5 and later,
440 ciblnd - Topspin 3.2.0,
441 iiblnd - Infiniserv 3.3 + PathBits patch,
442 gmlnd - GM 2.1.22 and later,
443 mxlnd - MX 1.2.1 or later,
444 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
446 --------------------------------------------------------------------------------
448 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
450 * Support for networks:
451 socklnd - kernels up to 2.6.16,
452 qswlnd - Qsnet kernel modules 5.20 and later,
453 openiblnd - IbGold 1.8.2,
455 viblnd - Voltaire ibhost 3.4.5 and later,
456 ciblnd - Topspin 3.2.0,
457 iiblnd - Infiniserv 3.3 + PathBits patch,
458 gmlnd - GM 2.1.22 and later,
459 mxlnd - MX 1.2.1 or later,
460 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
464 Description: Initialize cpumask before use
468 Description: ASSERTION failures when upgrading to the patchless zero-copy
470 Details : This bug affects "rolling upgrades", causing an inconsistent
471 protocol version negotiation and subsequent assertion failure
472 during rolling upgrades after the first wave of upgrades.
476 Details : Change "dropped message" CERRORs to D_NETERROR so they are
477 logged instead of creating "console chatter" when a lustre
478 timeout races with normal RPC completion.
481 Details : lnet_clear_peer_table can wait forever if user forgets to
485 Details : libcfs_id2str should check pid against LNET_PID_ANY.
489 Description: added LNET self test
490 Details : landing b_self_test
495 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
497 Details : do_div() macro is used incorrectly.
499 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
503 Description: make panic on lbug configurable
507 Description: Add OFED1.2 support to o2iblnd
508 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
509 are installed (other than kernel's in-tree infiniband), there
510 could be some problem while insmod o2iblnd (mismatch CRC of
512 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
513 this link provides solution:
514 https://bugs.openfabrics.org/show_bug.cgi?id=355
515 if extra Module.symvers is not supported in kernel, we will
516 have to run the script in bug 12316 to update
517 $LINUX/module.symvers before building o2iblnd.
518 More details about this are in bug 12316.
520 ------------------------------------------------------------------------------
522 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
523 * version 1.4.10 / 1.6.0
524 * Support for networks:
525 socklnd - kernels up to 2.6.16,
526 qswlnd - Qsnet kernel modules 5.20 and later,
527 openiblnd - IbGold 1.8.2,
529 viblnd - Voltaire ibhost 3.4.5 and later,
530 ciblnd - Topspin 3.2.0,
531 iiblnd - Infiniserv 3.3 + PathBits patch,
532 gmlnd - GM 2.1.22 and later,
533 mxlnd - MX 1.2.1 or later,
534 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
538 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
539 possibly accessed in kptllnd_shutdown. Ptllnd should init
540 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
544 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
548 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
552 Description: the_lnet.ln_finalizing was not set when the current thread is
553 about to complete messages. It only affects multi-threaded
559 Description: Changed the default kqswlnd ntxmsg=512
564 Description: Assertion failure in kernel ptllnd caused by posting passive
565 bulk buffers before connection establishment complete.
570 Description: A race in kernel ptllnd between deleting a peer and posting
571 new communications for it could hang communications -
572 manifesting as "Unexpectedly long timeout" messages.
577 Description: Kernel ptllnd lock ordering issue could hang a node.
582 Description: node crash on socket teardown race
585 Frequency : 'lctl peer_list' issued on a mx net
587 Description: Enable lctl's peer_list for MXLND
590 Frequency : after Ptllnd timeouts and portals congestion
592 Description: Credit overflows
593 Details : This was a bug in ptllnd connection establishment. The fix
594 implements better peer stamps to disambiguate connection
595 establishment and ensure both peers enter the credit flow
596 state machine consistently.
601 Description: kptllnd didn't propagate some network errors up to LNET
602 Details : This bug was spotted while investigating 11394. The fix
603 ensures network errors on sends and bulk transfers are
604 propagated to LNET/lustre correctly.
606 Severity : enhancement
608 Description: Fixed console chatter in case of -ETIMEDOUT.
610 Severity : enhancement
612 Description: Added D_NETTRACE for recording network packet history
613 (initially only for ptllnd). Also a separate userspace
614 ptllnd facility to gather history which should really be
615 covered by D_NETTRACE too, if only CDEBUG recorded history in
621 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
622 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
623 callback can occur before a connection has actually been
624 established. This caused an assertion failure previously.
626 Severity : enhancement
628 Description: Multiple instances for o2iblnd
629 Details : Allow multiple instances of o2iblnd to enable networking over
630 multiple HCAs and routing between them.
634 Description: lnet deadlock in router_checker
635 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
636 into BH locks to eliminate potential deadlock caused by
637 ksocknal_data_ready() preempting code holding these locks.
641 Description: Millions of failed socklnd connection attempts cause a very slow FS
642 Details : added a new route flag ksnr_scheduled to distinguish from
643 ksnr_connecting, so that a peer connection request is only turned
644 down for race concerns when an active connection to the same peer
645 is under progress (instead of just being scheduled).
647 ------------------------------------------------------------------------------
649 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
651 * Support for networks:
652 socklnd - kernels up to 2.6.16
653 qswlnd - Qsnet kernel modules 5.20 and later
654 openiblnd - IbGold 1.8.2
656 viblnd - Voltaire ibhost 3.4.5 and later
657 ciblnd - Topspin 3.2.0
658 iiblnd - Infiniserv 3.3 + PathBits patch
659 gmlnd - GM 2.1.22 and later
660 mxlnd - MX 1.2.1 or later
661 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
664 Severity : major on XT3
666 Description: libcfs overwrites /proc/sys/portals
667 Details : libcfs created a symlink from /proc/sys/portals to
668 /proc/sys/lnet for backwards compatibility. This is no
669 longer required and makes the Cray portals /proc variables
674 Description: OFED FMR API change
675 Details : This changes parameter usage to reflect a change in
676 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
677 that FMR support is only used in experimental versions of the
678 o2iblnd - this change does not affect standard usage at all.
680 Severity : enhancement
682 Description: new ko2iblnd module parameter: ib_mtu
683 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
684 HCAs. You can avoid this problem by setting the MTU to 1024
685 using this module parameter.
687 Severity : enhancement
688 Bugzilla : 11118/11620
689 Description: ptllnd small request message buffer alignment fix
690 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
691 Round up small message size on sends in case this option
692 is not supported. 11620 was a defect in the initial
693 implementation which effectively asserted all peers had to be
694 running the correct protocol version which was fixed by always
695 NAK-ing such requests and handling any misalignments they
700 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
701 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
702 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
704 Severity : enhancement
706 Description: Patchless ZC(zero copy) socklnd
707 Details : New protocol for socklnd, socklnd can support zero copy without
708 kernel patch, it's compatible with old socklnd. Checksum is
709 moved from tunables to modparams.
713 Description: When ksocknal_del_peer() is called upon a peer whose
714 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
715 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
718 Frequency : when ptlrpc is under heavy use and runs out of request buffer
720 Description: In lnet_match_blocked_msg(), md can be used without holding a
724 Frequency : very rarely
726 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
727 If connd connects a route which has been closed by
728 ksocknal_shutdown(), ksocknal_create_routes() may create new
729 routes which hold references on the peer, causing shutdown
730 process to wait for peer to disappear forever.
732 Severity : enhancement
734 Description: Dump XT3 portals traces on kptllnd timeout
735 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
736 dump Cray portals debug traces to a file. The kptllnd module
737 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
738 is the basename of the dump file.
741 Frequency : infrequent
743 Description: kernel ptllnd fix bug in connection re-establishment
744 Details : Kernel ptllnd could produce protocol errors e.g. illegal
745 matchbits and/or violate the credit flow protocol when trying
746 to re-establish a connection with a peer after an error or
749 Severity : enhancement
751 Description: Allow /proc/sys/lnet/debug to be set symbolically
752 Details : Allow debug and subsystem debug values to be read/set by name
753 in addition to numerically, for ease of use.
756 Frequency : only in configurations with LNET routers
758 Description: routes automatically marked down and recovered
759 Details : In configurations with LNET routers if a router fails routers
760 now actively try to recover routes that are down, unless they
761 are marked down by an administrator.
763 ------------------------------------------------------------------------------
765 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
768 Frequency : very rarely, in configurations with LNET routers and TCP
770 Description: incorrect data written to files on OSTs
771 Details : In certain high-load conditions incorrect data may be written
772 to files on the OST when using TCP networks.
774 ------------------------------------------------------------------------------
776 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
778 - rework CDEBUG messages rate-limiting mechanism b=10375
779 - add per-socket tunables for socklnd if the kernel is patched b=10327
781 ------------------------------------------------------------------------------
783 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
785 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
786 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
787 - quiet LNET startup LNI message for liblustre b=10128
788 - Better console error messages if 'ip2nets' can't match an IP address
789 - Fixed overflow/use-before-set bugs in linux-time.h
790 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
791 - LNET teardown failed an assertion about the route table being empty
792 - Fixed a crash in LNetEQPoll(<invalid handle>)
793 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
794 - improve debug message for liblustre/Catamount nodes (b=10116)
796 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
797 * Configuration change for the XT3
798 The PTLLND is now used to run Lustre over Portals on the XT3.
799 The configure option(s) --with-cray-portals are no longer
800 used. Rather --with-portals=<path-to-portals-includes> is
801 used to enable building on the XT3. In addition to enable
802 XT3 specific features the option --enable-cray-xt3 must be
805 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
806 * Portals has been removed, replaced by LNET.
807 LNET is new networking infrastructure for Lustre, it includes a
808 reorganized network configuration mode (see the user
809 documentation for full details) as well as support for routing
810 between different network fabrics. Lustre Networking Devices
811 (LNDS) for the supported network fabrics have also been created
812 for this new infrastructure.
814 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
819 Frequency : rare (large Voltaire clusters only)
821 Description: the default number of reserved transmit descriptors was too low
822 for some large clusters
823 Details : As a workaround, the number was increased. A proper fix includes
826 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
831 Frequency : occasional (large-scale events, cluster reboot, network failure)
833 Description: too many error messages on console obscure actual problem and
834 can slow down/panic server, or cause recovery to fail repeatedly
835 Details : enable rate-limiting of console error messages, and some messages
836 that were console errors now only go to the kernel log
838 Severity : enhancement
840 Description: add /proc/sys/portals/catastrophe entry which will report if
841 that node has previously LBUGged
843 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
845 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
847 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
849 - handle error return code in kranal_check_fma_rx() (5915,6054)
851 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
853 - update vibnal (Voltaire IB NAL)
854 - update gmnal (Myrinet NAL), gmnalid
856 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
858 * Landed portals:b_port_step as follows...
860 - removed CFS_DECL_SPIN*
861 just use 'spinlock_t' and initialise with spin_lock_init()
863 - removed CFS_DECL_MUTEX*
864 just use 'struct semaphore' and initialise with init_mutex()
866 - removed CFS_DECL_RWSEM*
867 just use 'struct rw_semaphore' and initialise with init_rwsem()
869 - renamed cfs_sleep_chan -> cfs_waitq
870 cfs_sleep_link -> cfs_waitlink
872 - fixed race in linux version of arch-independent socknal
873 (the ENOMEM/EAGAIN decision).
875 - Didn't fix problems in Darwin version of arch-independent socknal
876 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
878 - removed libcfs types from non-socknal header files (only some types
879 in the header files had been changed; the .c files hadn't been