1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
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,
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
15 --------------------------------------------------------------------------------
17 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
19 * Support for networks:
20 socklnd - any kernel supported by Lustre,
21 qswlnd - Qsnet kernel modules 5.20 and later,
22 openiblnd - IbGold 1.8.2,
23 o2iblnd - OFED 1.1 and 1.2,
24 viblnd - Voltaire ibhost 3.4.5 and later,
25 ciblnd - Topspin 3.2.0,
26 iiblnd - Infiniserv 3.3 + PathBits patch,
27 gmlnd - GM 2.1.22 and later,
28 mxlnd - MX 1.2.1 or later,
29 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
33 Description: TOE Kernel panic by ksocklnd
34 Details : offloaded sockets provide their own implementation of sendpage,
35 can't call tcp_sendpage() directly
39 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
40 Details : races between lnd_shutdown and peer creation prevent
41 lnd_shutdown from finishing.
45 Description: open files rlimit 1024 reached while liblustre testing
46 Details : ulnds/socklnd must close open socket after unsuccessful
49 ------------------------------------------------------------------------------
51 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
53 * Support for networks:
54 socklnd - kernels up to 2.6.16,
55 qswlnd - Qsnet kernel modules 5.20 and later,
56 openiblnd - IbGold 1.8.2,
57 o2iblnd - OFED 1.1 and 1.2
58 viblnd - Voltaire ibhost 3.4.5 and later,
59 ciblnd - Topspin 3.2.0,
60 iiblnd - Infiniserv 3.3 + PathBits patch,
61 gmlnd - GM 2.1.22 and later,
62 mxlnd - MX 1.2.1 or later,
63 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
65 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
67 * Support for networks:
68 socklnd - kernels up to 2.6.16,
69 qswlnd - Qsnet kernel modules 5.20 and later,
70 openiblnd - IbGold 1.8.2,
72 viblnd - Voltaire ibhost 3.4.5 and later,
73 ciblnd - Topspin 3.2.0,
74 iiblnd - Infiniserv 3.3 + PathBits patch,
75 gmlnd - GM 2.1.22 and later,
76 mxlnd - MX 1.2.1 or later,
77 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
81 Description: Initialize cpumask before use
85 Description: ASSERTION failures when upgrading to the patchless zero-copy
87 Details : This bug affects "rolling upgrades", causing an inconsistent
88 protocol version negotiation and subsequent assertion failure
89 during rolling upgrades after the first wave of upgrades.
93 Details : Change "dropped message" CERRORs to D_NETERROR so they are
94 logged instead of creating "console chatter" when a lustre
95 timeout races with normal RPC completion.
98 Details : lnet_clear_peer_table can wait forever if user forgets to
102 Details : libcfs_id2str should check pid against LNET_PID_ANY.
106 Description: added LNET self test
107 Details : landing b_self_test
112 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
114 Details : do_div() macro is used incorrectly.
116 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
120 Description: make panic on lbug configurable
124 Description: Add OFED1.2 support to o2iblnd
125 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
126 are installed (other than kernel's in-tree infiniband), there
127 could be some problem while insmod o2iblnd (mismatch CRC of
129 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
130 this link provides solution:
131 https://bugs.openfabrics.org/show_bug.cgi?id=355
132 if extra Module.symvers is not supported in kernel, we will
133 have to run the script in bug 12316 to update
134 $LINUX/module.symvers before building o2iblnd.
135 More details about this are in bug 12316.
137 ------------------------------------------------------------------------------
139 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
140 * version 1.4.10 / 1.6.0
141 * Support for networks:
142 socklnd - kernels up to 2.6.16,
143 qswlnd - Qsnet kernel modules 5.20 and later,
144 openiblnd - IbGold 1.8.2,
146 viblnd - Voltaire ibhost 3.4.5 and later,
147 ciblnd - Topspin 3.2.0,
148 iiblnd - Infiniserv 3.3 + PathBits patch,
149 gmlnd - GM 2.1.22 and later,
150 mxlnd - MX 1.2.1 or later,
151 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
155 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
156 possibly accessed in kptllnd_shutdown. Ptllnd should init
157 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
161 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
165 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
169 Description: the_lnet.ln_finalizing was not set when the current thread is
170 about to complete messages. It only affects multi-threaded
176 Description: Changed the default kqswlnd ntxmsg=512
181 Description: Assertion failure in kernel ptllnd caused by posting passive
182 bulk buffers before connection establishment complete.
187 Description: A race in kernel ptllnd between deleting a peer and posting
188 new communications for it could hang communications -
189 manifesting as "Unexpectedly long timeout" messages.
194 Description: Kernel ptllnd lock ordering issue could hang a node.
199 Description: node crash on socket teardown race
202 Frequency : 'lctl peer_list' issued on a mx net
204 Description: Enable lctl's peer_list for MXLND
207 Frequency : after Ptllnd timeouts and portals congestion
209 Description: Credit overflows
210 Details : This was a bug in ptllnd connection establishment. The fix
211 implements better peer stamps to disambiguate connection
212 establishment and ensure both peers enter the credit flow
213 state machine consistently.
218 Description: kptllnd didn't propagate some network errors up to LNET
219 Details : This bug was spotted while investigating 11394. The fix
220 ensures network errors on sends and bulk transfers are
221 propagated to LNET/lustre correctly.
223 Severity : enhancement
225 Description: Fixed console chatter in case of -ETIMEDOUT.
227 Severity : enhancement
229 Description: Added D_NETTRACE for recording network packet history
230 (initially only for ptllnd). Also a separate userspace
231 ptllnd facility to gather history which should really be
232 covered by D_NETTRACE too, if only CDEBUG recorded history in
238 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
239 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
240 callback can occur before a connection has actually been
241 established. This caused an assertion failure previously.
243 Severity : enhancement
245 Description: Multiple instances for o2iblnd
246 Details : Allow multiple instances of o2iblnd to enable networking over
247 multiple HCAs and routing between them.
251 Description: lnet deadlock in router_checker
252 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
253 into BH locks to eliminate potential deadlock caused by
254 ksocknal_data_ready() preempting code holding these locks.
258 Description: Millions of failed socklnd connection attempts cause a very slow FS
259 Details : added a new route flag ksnr_scheduled to distinguish from
260 ksnr_connecting, so that a peer connection request is only turned
261 down for race concerns when an active connection to the same peer
262 is under progress (instead of just being scheduled).
264 ------------------------------------------------------------------------------
266 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
268 * Support for networks:
269 socklnd - kernels up to 2.6.16
270 qswlnd - Qsnet kernel modules 5.20 and later
271 openiblnd - IbGold 1.8.2
273 viblnd - Voltaire ibhost 3.4.5 and later
274 ciblnd - Topspin 3.2.0
275 iiblnd - Infiniserv 3.3 + PathBits patch
276 gmlnd - GM 2.1.22 and later
277 mxlnd - MX 1.2.1 or later
278 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
281 Severity : major on XT3
283 Description: libcfs overwrites /proc/sys/portals
284 Details : libcfs created a symlink from /proc/sys/portals to
285 /proc/sys/lnet for backwards compatibility. This is no
286 longer required and makes the Cray portals /proc variables
291 Description: OFED FMR API change
292 Details : This changes parameter usage to reflect a change in
293 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
294 that FMR support is only used in experimental versions of the
295 o2iblnd - this change does not affect standard usage at all.
297 Severity : enhancement
299 Description: new ko2iblnd module parameter: ib_mtu
300 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
301 HCAs. You can avoid this problem by setting the MTU to 1024
302 using this module parameter.
304 Severity : enhancement
305 Bugzilla : 11118/11620
306 Description: ptllnd small request message buffer alignment fix
307 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
308 Round up small message size on sends in case this option
309 is not supported. 11620 was a defect in the initial
310 implementation which effectively asserted all peers had to be
311 running the correct protocol version which was fixed by always
312 NAK-ing such requests and handling any misalignments they
317 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
318 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
319 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
321 Severity : enhancement
323 Description: Patchless ZC(zero copy) socklnd
324 Details : New protocol for socklnd, socklnd can support zero copy without
325 kernel patch, it's compatible with old socklnd. Checksum is
326 moved from tunables to modparams.
330 Description: When ksocknal_del_peer() is called upon a peer whose
331 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
332 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
335 Frequency : when ptlrpc is under heavy use and runs out of request buffer
337 Description: In lnet_match_blocked_msg(), md can be used without holding a
341 Frequency : very rarely
343 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
344 If connd connects a route which has been closed by
345 ksocknal_shutdown(), ksocknal_create_routes() may create new
346 routes which hold references on the peer, causing shutdown
347 process to wait for peer to disappear forever.
349 Severity : enhancement
351 Description: Dump XT3 portals traces on kptllnd timeout
352 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
353 dump Cray portals debug traces to a file. The kptllnd module
354 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
355 is the basename of the dump file.
358 Frequency : infrequent
360 Description: kernel ptllnd fix bug in connection re-establishment
361 Details : Kernel ptllnd could produce protocol errors e.g. illegal
362 matchbits and/or violate the credit flow protocol when trying
363 to re-establish a connection with a peer after an error or
366 Severity : enhancement
368 Description: Allow /proc/sys/lnet/debug to be set symbolically
369 Details : Allow debug and subsystem debug values to be read/set by name
370 in addition to numerically, for ease of use.
373 Frequency : only in configurations with LNET routers
375 Description: routes automatically marked down and recovered
376 Details : In configurations with LNET routers if a router fails routers
377 now actively try to recover routes that are down, unless they
378 are marked down by an administrator.
380 ------------------------------------------------------------------------------
382 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
385 Frequency : very rarely, in configurations with LNET routers and TCP
387 Description: incorrect data written to files on OSTs
388 Details : In certain high-load conditions incorrect data may be written
389 to files on the OST when using TCP networks.
391 ------------------------------------------------------------------------------
393 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
395 - rework CDEBUG messages rate-limiting mechanism b=10375
396 - add per-socket tunables for socklnd if the kernel is patched b=10327
398 ------------------------------------------------------------------------------
400 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
402 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
403 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
404 - quiet LNET startup LNI message for liblustre b=10128
405 - Better console error messages if 'ip2nets' can't match an IP address
406 - Fixed overflow/use-before-set bugs in linux-time.h
407 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
408 - LNET teardown failed an assertion about the route table being empty
409 - Fixed a crash in LNetEQPoll(<invalid handle>)
410 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
411 - improve debug message for liblustre/Catamount nodes (b=10116)
413 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
414 * Configuration change for the XT3
415 The PTLLND is now used to run Lustre over Portals on the XT3.
416 The configure option(s) --with-cray-portals are no longer
417 used. Rather --with-portals=<path-to-portals-includes> is
418 used to enable building on the XT3. In addition to enable
419 XT3 specific features the option --enable-cray-xt3 must be
422 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
423 * Portals has been removed, replaced by LNET.
424 LNET is new networking infrastructure for Lustre, it includes a
425 reorganized network configuration mode (see the user
426 documentation for full details) as well as support for routing
427 between different network fabrics. Lustre Networking Devices
428 (LNDS) for the supported network fabrics have also been created
429 for this new infrastructure.
431 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
436 Frequency : rare (large Voltaire clusters only)
438 Description: the default number of reserved transmit descriptors was too low
439 for some large clusters
440 Details : As a workaround, the number was increased. A proper fix includes
443 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
448 Frequency : occasional (large-scale events, cluster reboot, network failure)
450 Description: too many error messages on console obscure actual problem and
451 can slow down/panic server, or cause recovery to fail repeatedly
452 Details : enable rate-limiting of console error messages, and some messages
453 that were console errors now only go to the kernel log
455 Severity : enhancement
457 Description: add /proc/sys/portals/catastrophe entry which will report if
458 that node has previously LBUGged
460 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
462 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
464 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
466 - handle error return code in kranal_check_fma_rx() (5915,6054)
468 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
470 - update vibnal (Voltaire IB NAL)
471 - update gmnal (Myrinet NAL), gmnalid
473 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
475 * Landed portals:b_port_step as follows...
477 - removed CFS_DECL_SPIN*
478 just use 'spinlock_t' and initialise with spin_lock_init()
480 - removed CFS_DECL_MUTEX*
481 just use 'struct semaphore' and initialise with init_mutex()
483 - removed CFS_DECL_RWSEM*
484 just use 'struct rw_semaphore' and initialise with init_rwsem()
486 - renamed cfs_sleep_chan -> cfs_waitq
487 cfs_sleep_link -> cfs_waitlink
489 - fixed race in linux version of arch-independent socknal
490 (the ENOMEM/EAGAIN decision).
492 - Didn't fix problems in Darwin version of arch-independent socknal
493 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
495 - removed libcfs types from non-socknal header files (only some types
496 in the header files had been changed; the .c files hadn't been