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 and 1.2.0, 1.2.5
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
16 Description: High load after starting lnet
17 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
18 uptime utility reports high load that looks confusingly.
22 Description: ksocklnd fails to establish connection if accept_port is high
23 Details : PID remapping must not be done for active (outgoing) connections
25 --------------------------------------------------------------------------------
27 2008-01-11 Sun Microsystems, Inc.
29 * Support for networks:
30 socklnd - any kernel supported by Lustre,
31 qswlnd - Qsnet kernel modules 5.20 and later,
32 openiblnd - IbGold 1.8.2,
33 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
34 viblnd - Voltaire ibhost 3.4.5 and later,
35 ciblnd - Topspin 3.2.0,
36 iiblnd - Infiniserv 3.3 + PathBits patch,
37 gmlnd - GM 2.1.22 and later,
38 mxlnd - MX 1.2.1 or later,
39 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
42 Description: liblustre network error
43 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
44 variable even if they don't start lnet acceptor.
48 Description: Strange message from lnet (Ignoring prediction from the future)
49 Details : Incorrect calculation of peer's last_alive value in ksocklnd
51 --------------------------------------------------------------------------------
53 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
55 * Support for networks:
56 socklnd - any kernel supported by Lustre,
57 qswlnd - Qsnet kernel modules 5.20 and later,
58 openiblnd - IbGold 1.8.2,
59 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
60 viblnd - Voltaire ibhost 3.4.5 and later,
61 ciblnd - Topspin 3.2.0,
62 iiblnd - Infiniserv 3.3 + PathBits patch,
63 gmlnd - GM 2.1.22 and later,
64 mxlnd - MX 1.2.1 or later,
65 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
69 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
73 Description: increase send queue size for ciblnd/openiblnd
77 Description: new userspace socklnd
78 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
79 with new one - usocklnd.
81 Severity : enhancement
83 Description: Console message flood
84 Details : Make cdls ratelimiting more tunable by adding several tunable in
85 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
86 /proc/sys/lnet/console_backoff.
88 --------------------------------------------------------------------------------
90 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
92 * Support for networks:
93 socklnd - any kernel supported by Lustre,
94 qswlnd - Qsnet kernel modules 5.20 and later,
95 openiblnd - IbGold 1.8.2,
96 o2iblnd - OFED 1.1 and 1.2,
97 viblnd - Voltaire ibhost 3.4.5 and later,
98 ciblnd - Topspin 3.2.0,
99 iiblnd - Infiniserv 3.3 + PathBits patch,
100 gmlnd - GM 2.1.22 and later,
101 mxlnd - MX 1.2.1 or later,
102 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
106 Description: /proc/sys/lnet has non-sysctl entries
107 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
111 Description: TOE Kernel panic by ksocklnd
112 Details : offloaded sockets provide their own implementation of sendpage,
113 can't call tcp_sendpage() directly
117 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
118 Details : races between lnd_shutdown and peer creation prevent
119 lnd_shutdown from finishing.
123 Description: open files rlimit 1024 reached while liblustre testing
124 Details : ulnds/socklnd must close open socket after unsuccessful
129 Description: build error
130 Details : fix typos in gmlnd, ptllnd and viblnd
132 ------------------------------------------------------------------------------
134 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
136 * Support for networks:
137 socklnd - kernels up to 2.6.16,
138 qswlnd - Qsnet kernel modules 5.20 and later,
139 openiblnd - IbGold 1.8.2,
140 o2iblnd - OFED 1.1 and 1.2
141 viblnd - Voltaire ibhost 3.4.5 and later,
142 ciblnd - Topspin 3.2.0,
143 iiblnd - Infiniserv 3.3 + PathBits patch,
144 gmlnd - GM 2.1.22 and later,
145 mxlnd - MX 1.2.1 or later,
146 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
148 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
150 * Support for networks:
151 socklnd - kernels up to 2.6.16,
152 qswlnd - Qsnet kernel modules 5.20 and later,
153 openiblnd - IbGold 1.8.2,
155 viblnd - Voltaire ibhost 3.4.5 and later,
156 ciblnd - Topspin 3.2.0,
157 iiblnd - Infiniserv 3.3 + PathBits patch,
158 gmlnd - GM 2.1.22 and later,
159 mxlnd - MX 1.2.1 or later,
160 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
164 Description: Initialize cpumask before use
168 Description: ASSERTION failures when upgrading to the patchless zero-copy
170 Details : This bug affects "rolling upgrades", causing an inconsistent
171 protocol version negotiation and subsequent assertion failure
172 during rolling upgrades after the first wave of upgrades.
176 Details : Change "dropped message" CERRORs to D_NETERROR so they are
177 logged instead of creating "console chatter" when a lustre
178 timeout races with normal RPC completion.
181 Details : lnet_clear_peer_table can wait forever if user forgets to
185 Details : libcfs_id2str should check pid against LNET_PID_ANY.
189 Description: added LNET self test
190 Details : landing b_self_test
195 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
197 Details : do_div() macro is used incorrectly.
199 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
203 Description: make panic on lbug configurable
207 Description: Add OFED1.2 support to o2iblnd
208 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
209 are installed (other than kernel's in-tree infiniband), there
210 could be some problem while insmod o2iblnd (mismatch CRC of
212 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
213 this link provides solution:
214 https://bugs.openfabrics.org/show_bug.cgi?id=355
215 if extra Module.symvers is not supported in kernel, we will
216 have to run the script in bug 12316 to update
217 $LINUX/module.symvers before building o2iblnd.
218 More details about this are in bug 12316.
220 ------------------------------------------------------------------------------
222 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
223 * version 1.4.10 / 1.6.0
224 * Support for networks:
225 socklnd - kernels up to 2.6.16,
226 qswlnd - Qsnet kernel modules 5.20 and later,
227 openiblnd - IbGold 1.8.2,
229 viblnd - Voltaire ibhost 3.4.5 and later,
230 ciblnd - Topspin 3.2.0,
231 iiblnd - Infiniserv 3.3 + PathBits patch,
232 gmlnd - GM 2.1.22 and later,
233 mxlnd - MX 1.2.1 or later,
234 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
238 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
239 possibly accessed in kptllnd_shutdown. Ptllnd should init
240 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
244 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
248 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
252 Description: the_lnet.ln_finalizing was not set when the current thread is
253 about to complete messages. It only affects multi-threaded
259 Description: Changed the default kqswlnd ntxmsg=512
264 Description: Assertion failure in kernel ptllnd caused by posting passive
265 bulk buffers before connection establishment complete.
270 Description: A race in kernel ptllnd between deleting a peer and posting
271 new communications for it could hang communications -
272 manifesting as "Unexpectedly long timeout" messages.
277 Description: Kernel ptllnd lock ordering issue could hang a node.
282 Description: node crash on socket teardown race
285 Frequency : 'lctl peer_list' issued on a mx net
287 Description: Enable lctl's peer_list for MXLND
290 Frequency : after Ptllnd timeouts and portals congestion
292 Description: Credit overflows
293 Details : This was a bug in ptllnd connection establishment. The fix
294 implements better peer stamps to disambiguate connection
295 establishment and ensure both peers enter the credit flow
296 state machine consistently.
301 Description: kptllnd didn't propagate some network errors up to LNET
302 Details : This bug was spotted while investigating 11394. The fix
303 ensures network errors on sends and bulk transfers are
304 propagated to LNET/lustre correctly.
306 Severity : enhancement
308 Description: Fixed console chatter in case of -ETIMEDOUT.
310 Severity : enhancement
312 Description: Added D_NETTRACE for recording network packet history
313 (initially only for ptllnd). Also a separate userspace
314 ptllnd facility to gather history which should really be
315 covered by D_NETTRACE too, if only CDEBUG recorded history in
321 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
322 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
323 callback can occur before a connection has actually been
324 established. This caused an assertion failure previously.
326 Severity : enhancement
328 Description: Multiple instances for o2iblnd
329 Details : Allow multiple instances of o2iblnd to enable networking over
330 multiple HCAs and routing between them.
334 Description: lnet deadlock in router_checker
335 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
336 into BH locks to eliminate potential deadlock caused by
337 ksocknal_data_ready() preempting code holding these locks.
341 Description: Millions of failed socklnd connection attempts cause a very slow FS
342 Details : added a new route flag ksnr_scheduled to distinguish from
343 ksnr_connecting, so that a peer connection request is only turned
344 down for race concerns when an active connection to the same peer
345 is under progress (instead of just being scheduled).
347 ------------------------------------------------------------------------------
349 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
351 * Support for networks:
352 socklnd - kernels up to 2.6.16
353 qswlnd - Qsnet kernel modules 5.20 and later
354 openiblnd - IbGold 1.8.2
356 viblnd - Voltaire ibhost 3.4.5 and later
357 ciblnd - Topspin 3.2.0
358 iiblnd - Infiniserv 3.3 + PathBits patch
359 gmlnd - GM 2.1.22 and later
360 mxlnd - MX 1.2.1 or later
361 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
364 Severity : major on XT3
366 Description: libcfs overwrites /proc/sys/portals
367 Details : libcfs created a symlink from /proc/sys/portals to
368 /proc/sys/lnet for backwards compatibility. This is no
369 longer required and makes the Cray portals /proc variables
374 Description: OFED FMR API change
375 Details : This changes parameter usage to reflect a change in
376 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
377 that FMR support is only used in experimental versions of the
378 o2iblnd - this change does not affect standard usage at all.
380 Severity : enhancement
382 Description: new ko2iblnd module parameter: ib_mtu
383 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
384 HCAs. You can avoid this problem by setting the MTU to 1024
385 using this module parameter.
387 Severity : enhancement
388 Bugzilla : 11118/11620
389 Description: ptllnd small request message buffer alignment fix
390 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
391 Round up small message size on sends in case this option
392 is not supported. 11620 was a defect in the initial
393 implementation which effectively asserted all peers had to be
394 running the correct protocol version which was fixed by always
395 NAK-ing such requests and handling any misalignments they
400 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
401 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
402 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
404 Severity : enhancement
406 Description: Patchless ZC(zero copy) socklnd
407 Details : New protocol for socklnd, socklnd can support zero copy without
408 kernel patch, it's compatible with old socklnd. Checksum is
409 moved from tunables to modparams.
413 Description: When ksocknal_del_peer() is called upon a peer whose
414 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
415 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
418 Frequency : when ptlrpc is under heavy use and runs out of request buffer
420 Description: In lnet_match_blocked_msg(), md can be used without holding a
424 Frequency : very rarely
426 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
427 If connd connects a route which has been closed by
428 ksocknal_shutdown(), ksocknal_create_routes() may create new
429 routes which hold references on the peer, causing shutdown
430 process to wait for peer to disappear forever.
432 Severity : enhancement
434 Description: Dump XT3 portals traces on kptllnd timeout
435 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
436 dump Cray portals debug traces to a file. The kptllnd module
437 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
438 is the basename of the dump file.
441 Frequency : infrequent
443 Description: kernel ptllnd fix bug in connection re-establishment
444 Details : Kernel ptllnd could produce protocol errors e.g. illegal
445 matchbits and/or violate the credit flow protocol when trying
446 to re-establish a connection with a peer after an error or
449 Severity : enhancement
451 Description: Allow /proc/sys/lnet/debug to be set symbolically
452 Details : Allow debug and subsystem debug values to be read/set by name
453 in addition to numerically, for ease of use.
456 Frequency : only in configurations with LNET routers
458 Description: routes automatically marked down and recovered
459 Details : In configurations with LNET routers if a router fails routers
460 now actively try to recover routes that are down, unless they
461 are marked down by an administrator.
463 ------------------------------------------------------------------------------
465 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
468 Frequency : very rarely, in configurations with LNET routers and TCP
470 Description: incorrect data written to files on OSTs
471 Details : In certain high-load conditions incorrect data may be written
472 to files on the OST when using TCP networks.
474 ------------------------------------------------------------------------------
476 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
478 - rework CDEBUG messages rate-limiting mechanism b=10375
479 - add per-socket tunables for socklnd if the kernel is patched b=10327
481 ------------------------------------------------------------------------------
483 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
485 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
486 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
487 - quiet LNET startup LNI message for liblustre b=10128
488 - Better console error messages if 'ip2nets' can't match an IP address
489 - Fixed overflow/use-before-set bugs in linux-time.h
490 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
491 - LNET teardown failed an assertion about the route table being empty
492 - Fixed a crash in LNetEQPoll(<invalid handle>)
493 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
494 - improve debug message for liblustre/Catamount nodes (b=10116)
496 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
497 * Configuration change for the XT3
498 The PTLLND is now used to run Lustre over Portals on the XT3.
499 The configure option(s) --with-cray-portals are no longer
500 used. Rather --with-portals=<path-to-portals-includes> is
501 used to enable building on the XT3. In addition to enable
502 XT3 specific features the option --enable-cray-xt3 must be
505 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
506 * Portals has been removed, replaced by LNET.
507 LNET is new networking infrastructure for Lustre, it includes a
508 reorganized network configuration mode (see the user
509 documentation for full details) as well as support for routing
510 between different network fabrics. Lustre Networking Devices
511 (LNDS) for the supported network fabrics have also been created
512 for this new infrastructure.
514 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
519 Frequency : rare (large Voltaire clusters only)
521 Description: the default number of reserved transmit descriptors was too low
522 for some large clusters
523 Details : As a workaround, the number was increased. A proper fix includes
526 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
531 Frequency : occasional (large-scale events, cluster reboot, network failure)
533 Description: too many error messages on console obscure actual problem and
534 can slow down/panic server, or cause recovery to fail repeatedly
535 Details : enable rate-limiting of console error messages, and some messages
536 that were console errors now only go to the kernel log
538 Severity : enhancement
540 Description: add /proc/sys/portals/catastrophe entry which will report if
541 that node has previously LBUGged
543 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
545 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
547 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
549 - handle error return code in kranal_check_fma_rx() (5915,6054)
551 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
553 - update vibnal (Voltaire IB NAL)
554 - update gmnal (Myrinet NAL), gmnalid
556 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
558 * Landed portals:b_port_step as follows...
560 - removed CFS_DECL_SPIN*
561 just use 'spinlock_t' and initialise with spin_lock_init()
563 - removed CFS_DECL_MUTEX*
564 just use 'struct semaphore' and initialise with init_mutex()
566 - removed CFS_DECL_RWSEM*
567 just use 'struct rw_semaphore' and initialise with init_rwsem()
569 - renamed cfs_sleep_chan -> cfs_waitq
570 cfs_sleep_link -> cfs_waitlink
572 - fixed race in linux version of arch-independent socknal
573 (the ENOMEM/EAGAIN decision).
575 - Didn't fix problems in Darwin version of arch-independent socknal
576 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
578 - removed libcfs types from non-socknal header files (only some types
579 in the header files had been changed; the .c files hadn't been