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: ksocknal_close_conn_locked connection race
23 Details : A race was possible when ksocknal_create_conn calls
24 ksocknal_close_conn_locked for already closed conn.
28 Description: router_proc.c is rewritten to use sysctl-interface for parameters
29 residing in /proc/sys/lnet
31 Severity : enhancement
33 Description: port router pinger to userspace
37 Description: kptllnd HELLO protocol deadlock
38 Details : kptllnd HELLO protocol doesn't run to completion in finite time
42 Description: LNet selftest fixes and enhancements
44 Severity : enhancement
46 Description: allow a test node to be a member of multiple test groups
48 Severity : enhancement
50 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
51 Details : an update from the upstream developer Scott Atchley.
53 Severity : enhancement
55 Description: add a new LND optiion to control peer buffer credits on routers
59 Description: Fixing deadlock in usocklnd
60 Details : A deadlock was possible in usocklnd due to race condition while
61 tearing connection down. The problem resulted from erroneous
62 assumption that lnet_finalize() could have been called holding
66 Bugzilla : 13621, 15983
67 Description: Protocol V2 of o2iblnd
68 Details : o2iblnd V2 has several new features:
69 . map-on-demand: map-on-demand is disabled by default, it can
70 be enabled by using modparam "map_on_demand=@value@", @value@
71 should >= 0 and < 256, 0 will disable map-on-demand, any other
72 valid value will enable map-on-demand.
73 Oi2blnd will create FMR or physical MR for RDMA if fragments of
75 Enable map-on-demand will take less memory for new connection,
76 but a little more CPU for RDMA.
77 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
78 are recommanded value. iWARP will probably fail for value >=128.
79 . OOB NOOP message: to resolve deadlock on router.
80 . tunable peer_credits_hiw: (high water to return credits),
81 default value of peer_credits_hiw equals to (peer_credits -1),
82 user can change it between peer_credits/2 and (peer_credits - 1).
83 Lower value is recommended for high latency network.
84 . tunable message queue size: it always equals to peer_credits,
85 higher value is recommended for high latency network.
86 . It's compatible with earlier version of o2iblnd
90 Description: Fixing 'running out of ports' issue
91 Details : Add a delay before next reconnect attempt in ksocklnd in
92 the case of lost race. Limit the frequency of query-requests
93 in lnet. Improved handling of 'dead peer' notifications in
98 Description: Change ptllnd timeout and watchdog timers
99 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
100 Portals wire timeout.
104 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
105 Details : Shared routing enhancements - peer health detection.
107 Severity : enhancement
109 Description: acceptor.c cleanup
110 Details : Code duplication in acceptor.c for the cases of kernel and
111 user-space removed. User-space libcfs tcpip primitives
112 uniformed to have prototypes similar to kernel ones. Minor
113 cosmetic changes in usocklnd to use cfs_socket_t as
114 representation of socket.
118 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
119 Details : See comment 46 in bug 11245 for details - it's indeed a bug
120 introduced by the original 11245 fix.
124 Description: uptllnd credit overflow fix
125 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
130 Description: socklnd protocol version 3
131 Details : With current protocol V2, connections on router can be
132 blocked and can't receive any incoming messages when there is no
133 more router buffer, so ZC-ACK can't be handled (LNet message
134 can't be finalized) and will cause deadlock on router.
135 Protocol V3 has a dedicated connection for emergency messages
136 like ZC-ACK to router, messages on this dedicated connection
137 don't need any credit so will never be blocked. Also, V3 can send
138 keepalive ping in specified period for router healthy checking.
140 -------------------------------------------------------------------------------
142 12-31-2008 Sun Microsystems, Inc.
144 * Support for networks:
145 socklnd - any kernel supported by Lustre,
146 qswlnd - Qsnet kernel modules 5.20 and later,
147 openiblnd - IbGold 1.8.2,
148 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
149 viblnd - Voltaire ibhost 3.4.5 and later,
150 ciblnd - Topspin 3.2.0,
151 iiblnd - Infiniserv 3.3 + PathBits patch,
152 gmlnd - GM 2.1.22 and later,
153 mxlnd - MX 1.2.1 or later,
154 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
158 Description: workaround for OOM from o2iblnd
159 Details : OFED needs allocate big chunk of memory for QP while creating
160 connection for o2iblnd, OOM can happen if no such a contiguous
162 QP size is decided by concurrent_sends and max_fragments of
163 o2iblnd, now we permit user to specify smaller value for
164 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
165 will decrease memory block size required by creating QP.
169 Description: Support Zerocopy receive of Chelsio device
170 Details : Chelsio driver can support zerocopy for iov[1] if it's
171 contiguous and large enough.
175 Description: fix credit flow deadlock in uptllnd
179 Description: finalize network operation in reasonable time
180 Details : conf-sanity test_32a couldn't stop ost and mds because it
181 tried to access non-existent peer and tcp connect took
182 quite long before timing out.
186 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
187 Details : Lost reference on conn prevents peer from being destroyed, which
188 could prevent new peer creation if peer count has reached upper
193 Description: LNET Selftest results in Soft lockup on OSS CPU
194 Details : only hits when 8 or more o2ib clients involved and a session is
195 torn down with 'lst end_session' without preceeding 'lst stop'.
199 Description: concurrent_sends in IB LNDs should not be changeable at run time
200 Details : concurrent_sends in IB LNDs should not be changeable at run time
204 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
205 Details : only hits under out-of-memory situations
208 -------------------------------------------------------------------------------
210 2009-02-07 Sun Microsystems, Inc.
212 * Support for networks:
213 socklnd - any kernel supported by Lustre,
214 qswlnd - Qsnet kernel modules 5.20 and later,
215 openiblnd - IbGold 1.8.2,
216 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
217 viblnd - Voltaire ibhost 3.4.5 and later,
218 ciblnd - Topspin 3.2.0,
219 iiblnd - Infiniserv 3.3 + PathBits patch,
220 gmlnd - GM 2.1.22 and later,
221 mxlnd - MX 1.2.1 or later,
222 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
225 Description: workaround for OOM from o2iblnd
226 Details : OFED needs allocate big chunk of memory for QP while creating
227 connection for o2iblnd, OOM can happen if no such a contiguous
229 QP size is decided by concurrent_sends and max_fragments of
230 o2iblnd, now we permit user to specify smaller value for
231 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
232 will decrease memory block size required by creating QP.
236 Description: Support Zerocopy receive of Chelsio device
237 Details : Chelsio driver can support zerocopy for iov[1] if it's
238 contiguous and large enough.
241 Description: fix credit flow deadlock in uptllnd
245 Description: finalize network operation in reasonable time
246 Details : conf-sanity test_32a couldn't stop ost and mds because it
247 tried to access non-existent peer and tcp connect took
248 quite long before timing out.
252 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
253 Details : Lost reference on conn prevents peer from being destroyed, which
254 could prevent new peer creation if peer count has reached upper
259 Description: LNET Selftest results in Soft lockup on OSS CPU
260 Details : only hits when 8 or more o2ib clients involved and a session is
261 torn down with 'lst end_session' without preceeding 'lst stop'.
265 Description: concurrent_sends in IB LNDs should not be changeable at run time
266 Details : concurrent_sends in IB LNDs should not be changeable at run time
268 -------------------------------------------------------------------------------
270 11-03-2008 Sun Microsystems, Inc.
272 * Support for networks:
273 socklnd - any kernel supported by Lustre,
274 qswlnd - Qsnet kernel modules 5.20 and later,
275 openiblnd - IbGold 1.8.2,
276 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
277 viblnd - Voltaire ibhost 3.4.5 and later,
278 ciblnd - Topspin 3.2.0,
279 iiblnd - Infiniserv 3.3 + PathBits patch,
280 gmlnd - GM 2.1.22 and later,
281 mxlnd - MX 1.2.1 or later,
282 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
286 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
287 Details : only hits under out-of-memory situations
289 -------------------------------------------------------------------------------
291 04-26-2008 Sun Microsystems, Inc.
293 * Support for networks:
294 socklnd - any kernel supported by Lustre,
295 qswlnd - Qsnet kernel modules 5.20 and later,
296 openiblnd - IbGold 1.8.2,
297 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
298 viblnd - Voltaire ibhost 3.4.5 and later,
299 ciblnd - Topspin 3.2.0,
300 iiblnd - Infiniserv 3.3 + PathBits patch,
301 gmlnd - GM 2.1.22 and later,
302 mxlnd - MX 1.2.1 or later,
303 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
307 Description: excessive debug information removed
308 Details : excessive debug information removed
312 Description: ksocknal_create_conn() hit ASSERTION during connection race
313 Details : ksocknal_create_conn() hit ASSERTION during connection race
317 Description: ksocknal_send_hello() hit ASSERTION while connecting race
318 Details : ksocknal_send_hello() hit ASSERTION while connecting race
322 Description: o2iblnd/ptllnd credit deadlock in a routed config.
323 Details : o2iblnd/ptllnd credit deadlock in a routed config.
327 Description: High load after starting lnet
328 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
329 uptime utility reports high load that looks confusingly.
333 Description: ksocklnd fails to establish connection if accept_port is high
334 Details : PID remapping must not be done for active (outgoing) connections
337 --------------------------------------------------------------------------------
339 2008-01-11 Sun Microsystems, Inc.
341 * Support for networks:
342 socklnd - any kernel supported by Lustre,
343 qswlnd - Qsnet kernel modules 5.20 and later,
344 openiblnd - IbGold 1.8.2,
345 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
346 viblnd - Voltaire ibhost 3.4.5 and later,
347 ciblnd - Topspin 3.2.0,
348 iiblnd - Infiniserv 3.3 + PathBits patch,
349 gmlnd - GM 2.1.22 and later,
350 mxlnd - MX 1.2.1 or later,
351 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
355 Description: liblustre network error
356 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
357 variable even if they don't start lnet acceptor.
361 Description: Strange message from lnet (Ignoring prediction from the future)
362 Details : Incorrect calculation of peer's last_alive value in ksocklnd
364 --------------------------------------------------------------------------------
366 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
368 * Support for networks:
369 socklnd - any kernel supported by Lustre,
370 qswlnd - Qsnet kernel modules 5.20 and later,
371 openiblnd - IbGold 1.8.2,
372 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
373 viblnd - Voltaire ibhost 3.4.5 and later,
374 ciblnd - Topspin 3.2.0,
375 iiblnd - Infiniserv 3.3 + PathBits patch,
376 gmlnd - GM 2.1.22 and later,
377 mxlnd - MX 1.2.1 or later,
378 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
382 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
386 Description: increase send queue size for ciblnd/openiblnd
390 Description: new userspace socklnd
391 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
392 with new one - usocklnd.
394 Severity : enhancement
396 Description: Console message flood
397 Details : Make cdls ratelimiting more tunable by adding several tunable in
398 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
399 /proc/sys/lnet/console_backoff.
401 --------------------------------------------------------------------------------
403 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
405 * Support for networks:
406 socklnd - any kernel supported by Lustre,
407 qswlnd - Qsnet kernel modules 5.20 and later,
408 openiblnd - IbGold 1.8.2,
409 o2iblnd - OFED 1.1 and 1.2,
410 viblnd - Voltaire ibhost 3.4.5 and later,
411 ciblnd - Topspin 3.2.0,
412 iiblnd - Infiniserv 3.3 + PathBits patch,
413 gmlnd - GM 2.1.22 and later,
414 mxlnd - MX 1.2.1 or later,
415 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
419 Description: /proc/sys/lnet has non-sysctl entries
420 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
424 Description: TOE Kernel panic by ksocklnd
425 Details : offloaded sockets provide their own implementation of sendpage,
426 can't call tcp_sendpage() directly
430 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
431 Details : races between lnd_shutdown and peer creation prevent
432 lnd_shutdown from finishing.
436 Description: open files rlimit 1024 reached while liblustre testing
437 Details : ulnds/socklnd must close open socket after unsuccessful
442 Description: build error
443 Details : fix typos in gmlnd, ptllnd and viblnd
445 --------------------------------------------------------------------------------
447 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
449 * Support for networks:
450 socklnd - kernels up to 2.6.16,
451 qswlnd - Qsnet kernel modules 5.20 and later,
452 openiblnd - IbGold 1.8.2,
453 o2iblnd - OFED 1.1 and 1.2
454 viblnd - Voltaire ibhost 3.4.5 and later,
455 ciblnd - Topspin 3.2.0,
456 iiblnd - Infiniserv 3.3 + PathBits patch,
457 gmlnd - GM 2.1.22 and later,
458 mxlnd - MX 1.2.1 or later,
459 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
461 --------------------------------------------------------------------------------
463 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
465 * Support for networks:
466 socklnd - kernels up to 2.6.16,
467 qswlnd - Qsnet kernel modules 5.20 and later,
468 openiblnd - IbGold 1.8.2,
470 viblnd - Voltaire ibhost 3.4.5 and later,
471 ciblnd - Topspin 3.2.0,
472 iiblnd - Infiniserv 3.3 + PathBits patch,
473 gmlnd - GM 2.1.22 and later,
474 mxlnd - MX 1.2.1 or later,
475 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
479 Description: Initialize cpumask before use
483 Description: ASSERTION failures when upgrading to the patchless zero-copy
485 Details : This bug affects "rolling upgrades", causing an inconsistent
486 protocol version negotiation and subsequent assertion failure
487 during rolling upgrades after the first wave of upgrades.
491 Details : Change "dropped message" CERRORs to D_NETERROR so they are
492 logged instead of creating "console chatter" when a lustre
493 timeout races with normal RPC completion.
496 Details : lnet_clear_peer_table can wait forever if user forgets to
500 Details : libcfs_id2str should check pid against LNET_PID_ANY.
504 Description: added LNET self test
505 Details : landing b_self_test
510 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
512 Details : do_div() macro is used incorrectly.
514 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
518 Description: make panic on lbug configurable
522 Description: Add OFED1.2 support to o2iblnd
523 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
524 are installed (other than kernel's in-tree infiniband), there
525 could be some problem while insmod o2iblnd (mismatch CRC of
527 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
528 this link provides solution:
529 https://bugs.openfabrics.org/show_bug.cgi?id=355
530 if extra Module.symvers is not supported in kernel, we will
531 have to run the script in bug 12316 to update
532 $LINUX/module.symvers before building o2iblnd.
533 More details about this are in bug 12316.
535 ------------------------------------------------------------------------------
537 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
538 * version 1.4.10 / 1.6.0
539 * Support for networks:
540 socklnd - kernels up to 2.6.16,
541 qswlnd - Qsnet kernel modules 5.20 and later,
542 openiblnd - IbGold 1.8.2,
544 viblnd - Voltaire ibhost 3.4.5 and later,
545 ciblnd - Topspin 3.2.0,
546 iiblnd - Infiniserv 3.3 + PathBits patch,
547 gmlnd - GM 2.1.22 and later,
548 mxlnd - MX 1.2.1 or later,
549 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
553 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
554 possibly accessed in kptllnd_shutdown. Ptllnd should init
555 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
559 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
563 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
567 Description: the_lnet.ln_finalizing was not set when the current thread is
568 about to complete messages. It only affects multi-threaded
574 Description: Changed the default kqswlnd ntxmsg=512
579 Description: Assertion failure in kernel ptllnd caused by posting passive
580 bulk buffers before connection establishment complete.
585 Description: A race in kernel ptllnd between deleting a peer and posting
586 new communications for it could hang communications -
587 manifesting as "Unexpectedly long timeout" messages.
592 Description: Kernel ptllnd lock ordering issue could hang a node.
597 Description: node crash on socket teardown race
600 Frequency : 'lctl peer_list' issued on a mx net
602 Description: Enable lctl's peer_list for MXLND
605 Frequency : after Ptllnd timeouts and portals congestion
607 Description: Credit overflows
608 Details : This was a bug in ptllnd connection establishment. The fix
609 implements better peer stamps to disambiguate connection
610 establishment and ensure both peers enter the credit flow
611 state machine consistently.
616 Description: kptllnd didn't propagate some network errors up to LNET
617 Details : This bug was spotted while investigating 11394. The fix
618 ensures network errors on sends and bulk transfers are
619 propagated to LNET/lustre correctly.
621 Severity : enhancement
623 Description: Fixed console chatter in case of -ETIMEDOUT.
625 Severity : enhancement
627 Description: Added D_NETTRACE for recording network packet history
628 (initially only for ptllnd). Also a separate userspace
629 ptllnd facility to gather history which should really be
630 covered by D_NETTRACE too, if only CDEBUG recorded history in
636 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
637 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
638 callback can occur before a connection has actually been
639 established. This caused an assertion failure previously.
641 Severity : enhancement
643 Description: Multiple instances for o2iblnd
644 Details : Allow multiple instances of o2iblnd to enable networking over
645 multiple HCAs and routing between them.
649 Description: lnet deadlock in router_checker
650 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
651 into BH locks to eliminate potential deadlock caused by
652 ksocknal_data_ready() preempting code holding these locks.
656 Description: Millions of failed socklnd connection attempts cause a very slow FS
657 Details : added a new route flag ksnr_scheduled to distinguish from
658 ksnr_connecting, so that a peer connection request is only turned
659 down for race concerns when an active connection to the same peer
660 is under progress (instead of just being scheduled).
662 ------------------------------------------------------------------------------
664 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
666 * Support for networks:
667 socklnd - kernels up to 2.6.16
668 qswlnd - Qsnet kernel modules 5.20 and later
669 openiblnd - IbGold 1.8.2
671 viblnd - Voltaire ibhost 3.4.5 and later
672 ciblnd - Topspin 3.2.0
673 iiblnd - Infiniserv 3.3 + PathBits patch
674 gmlnd - GM 2.1.22 and later
675 mxlnd - MX 1.2.1 or later
676 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
679 Severity : major on XT3
681 Description: libcfs overwrites /proc/sys/portals
682 Details : libcfs created a symlink from /proc/sys/portals to
683 /proc/sys/lnet for backwards compatibility. This is no
684 longer required and makes the Cray portals /proc variables
689 Description: OFED FMR API change
690 Details : This changes parameter usage to reflect a change in
691 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
692 that FMR support is only used in experimental versions of the
693 o2iblnd - this change does not affect standard usage at all.
695 Severity : enhancement
697 Description: new ko2iblnd module parameter: ib_mtu
698 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
699 HCAs. You can avoid this problem by setting the MTU to 1024
700 using this module parameter.
702 Severity : enhancement
703 Bugzilla : 11118/11620
704 Description: ptllnd small request message buffer alignment fix
705 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
706 Round up small message size on sends in case this option
707 is not supported. 11620 was a defect in the initial
708 implementation which effectively asserted all peers had to be
709 running the correct protocol version which was fixed by always
710 NAK-ing such requests and handling any misalignments they
715 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
716 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
717 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
719 Severity : enhancement
721 Description: Patchless ZC(zero copy) socklnd
722 Details : New protocol for socklnd, socklnd can support zero copy without
723 kernel patch, it's compatible with old socklnd. Checksum is
724 moved from tunables to modparams.
728 Description: When ksocknal_del_peer() is called upon a peer whose
729 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
730 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
733 Frequency : when ptlrpc is under heavy use and runs out of request buffer
735 Description: In lnet_match_blocked_msg(), md can be used without holding a
739 Frequency : very rarely
741 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
742 If connd connects a route which has been closed by
743 ksocknal_shutdown(), ksocknal_create_routes() may create new
744 routes which hold references on the peer, causing shutdown
745 process to wait for peer to disappear forever.
747 Severity : enhancement
749 Description: Dump XT3 portals traces on kptllnd timeout
750 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
751 dump Cray portals debug traces to a file. The kptllnd module
752 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
753 is the basename of the dump file.
756 Frequency : infrequent
758 Description: kernel ptllnd fix bug in connection re-establishment
759 Details : Kernel ptllnd could produce protocol errors e.g. illegal
760 matchbits and/or violate the credit flow protocol when trying
761 to re-establish a connection with a peer after an error or
764 Severity : enhancement
766 Description: Allow /proc/sys/lnet/debug to be set symbolically
767 Details : Allow debug and subsystem debug values to be read/set by name
768 in addition to numerically, for ease of use.
771 Frequency : only in configurations with LNET routers
773 Description: routes automatically marked down and recovered
774 Details : In configurations with LNET routers if a router fails routers
775 now actively try to recover routes that are down, unless they
776 are marked down by an administrator.
778 ------------------------------------------------------------------------------
780 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
783 Frequency : very rarely, in configurations with LNET routers and TCP
785 Description: incorrect data written to files on OSTs
786 Details : In certain high-load conditions incorrect data may be written
787 to files on the OST when using TCP networks.
789 ------------------------------------------------------------------------------
791 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
793 - rework CDEBUG messages rate-limiting mechanism b=10375
794 - add per-socket tunables for socklnd if the kernel is patched b=10327
796 ------------------------------------------------------------------------------
798 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
800 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
801 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
802 - quiet LNET startup LNI message for liblustre b=10128
803 - Better console error messages if 'ip2nets' can't match an IP address
804 - Fixed overflow/use-before-set bugs in linux-time.h
805 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
806 - LNET teardown failed an assertion about the route table being empty
807 - Fixed a crash in LNetEQPoll(<invalid handle>)
808 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
809 - improve debug message for liblustre/Catamount nodes (b=10116)
811 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
812 * Configuration change for the XT3
813 The PTLLND is now used to run Lustre over Portals on the XT3.
814 The configure option(s) --with-cray-portals are no longer
815 used. Rather --with-portals=<path-to-portals-includes> is
816 used to enable building on the XT3. In addition to enable
817 XT3 specific features the option --enable-cray-xt3 must be
820 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
821 * Portals has been removed, replaced by LNET.
822 LNET is new networking infrastructure for Lustre, it includes a
823 reorganized network configuration mode (see the user
824 documentation for full details) as well as support for routing
825 between different network fabrics. Lustre Networking Devices
826 (LNDS) for the supported network fabrics have also been created
827 for this new infrastructure.
829 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
834 Frequency : rare (large Voltaire clusters only)
836 Description: the default number of reserved transmit descriptors was too low
837 for some large clusters
838 Details : As a workaround, the number was increased. A proper fix includes
841 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
846 Frequency : occasional (large-scale events, cluster reboot, network failure)
848 Description: too many error messages on console obscure actual problem and
849 can slow down/panic server, or cause recovery to fail repeatedly
850 Details : enable rate-limiting of console error messages, and some messages
851 that were console errors now only go to the kernel log
853 Severity : enhancement
855 Description: add /proc/sys/portals/catastrophe entry which will report if
856 that node has previously LBUGged
858 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
860 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
862 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
864 - handle error return code in kranal_check_fma_rx() (5915,6054)
866 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
868 - update vibnal (Voltaire IB NAL)
869 - update gmnal (Myrinet NAL), gmnalid
871 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
873 * Landed portals:b_port_step as follows...
875 - removed CFS_DECL_SPIN*
876 just use 'spinlock_t' and initialise with spin_lock_init()
878 - removed CFS_DECL_MUTEX*
879 just use 'struct semaphore' and initialise with init_mutex()
881 - removed CFS_DECL_RWSEM*
882 just use 'struct rw_semaphore' and initialise with init_rwsem()
884 - renamed cfs_sleep_chan -> cfs_waitq
885 cfs_sleep_link -> cfs_waitlink
887 - fixed race in linux version of arch-independent socknal
888 (the ENOMEM/EAGAIN decision).
890 - Didn't fix problems in Darwin version of arch-independent socknal
891 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
893 - removed libcfs types from non-socknal header files (only some types
894 in the header files had been changed; the .c files hadn't been