1 2007-09-24 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
17 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
18 Details : races between lnd_shutdown and peer creation prevent
19 lnd_shutdown from finishing.
23 Description: open files rlimit 1024 reached while liblustre testing
24 Details : ulnds/socklnd must close open socket after unsuccessful
27 ------------------------------------------------------------------------------
29 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
30 * version 1.4.11 / 1.6.1
31 * Support for networks:
32 socklnd - kernels up to 2.6.16,
33 qswlnd - Qsnet kernel modules 5.20 and later,
34 openiblnd - IbGold 1.8.2,
35 o2iblnd - OFED 1.1 and 1.2,
36 viblnd - Voltaire ibhost 3.4.5 and later,
37 ciblnd - Topspin 3.2.0,
38 iiblnd - Infiniserv 3.3 + PathBits patch,
39 gmlnd - GM 2.1.22 and later,
40 mxlnd - MX 1.2.1 or later,
41 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
45 Description: Initialize cpumask before use
49 Description: ASSERTION failures when upgrading to the patchless zero-copy
51 Details : This bug affects "rolling upgrades", causing an inconsistent
52 protocol version negotiation and subsequent assertion failure
53 during rolling upgrades after the first wave of upgrades.
57 Details : Change "dropped message" CERRORs to D_NETERROR so they are
58 logged instead of creating "console chatter" when a lustre
59 timeout races with normal RPC completion.
62 Details : lnet_clear_peer_table can wait forever if user forgets to
66 Details : libcfs_id2str should check pid against LNET_PID_ANY.
70 Description: added LNET self test
71 Details : landing b_self_test
76 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
78 Details : do_div() macro is used incorrectly.
80 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
84 Description: make panic on lbug configurable
88 Description: Add OFED1.2 support to o2iblnd
89 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
90 are installed (other than kernel's in-tree infiniband), there
91 could be some problem while insmod o2iblnd (mismatch CRC of
93 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
94 this link provides solution:
95 https://bugs.openfabrics.org/show_bug.cgi?id=355
96 if extra Module.symvers is not supported in kernel, we will
97 have to run the script in bug 12316 to update
98 $LINUX/module.symvers before building o2iblnd.
99 More details about this are in bug 12316.
101 ------------------------------------------------------------------------------
103 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
104 * version 1.4.10 / 1.6.0
105 * Support for networks:
106 socklnd - kernels up to 2.6.16,
107 qswlnd - Qsnet kernel modules 5.20 and later,
108 openiblnd - IbGold 1.8.2,
110 viblnd - Voltaire ibhost 3.4.5 and later,
111 ciblnd - Topspin 3.2.0,
112 iiblnd - Infiniserv 3.3 + PathBits patch,
113 gmlnd - GM 2.1.22 and later,
114 mxlnd - MX 1.2.1 or later,
115 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
119 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
120 possibly accessed in kptllnd_shutdown. Ptllnd should init
121 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
125 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
129 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
133 Description: the_lnet.ln_finalizing was not set when the current thread is
134 about to complete messages. It only affects multi-threaded
140 Description: Changed the default kqswlnd ntxmsg=512
145 Description: Assertion failure in kernel ptllnd caused by posting passive
146 bulk buffers before connection establishment complete.
151 Description: A race in kernel ptllnd between deleting a peer and posting
152 new communications for it could hang communications -
153 manifesting as "Unexpectedly long timeout" messages.
158 Description: Kernel ptllnd lock ordering issue could hang a node.
163 Description: node crash on socket teardown race
166 Frequency : 'lctl peer_list' issued on a mx net
168 Description: Enable lctl's peer_list for MXLND
171 Frequency : after Ptllnd timeouts and portals congestion
173 Description: Credit overflows
174 Details : This was a bug in ptllnd connection establishment. The fix
175 implements better peer stamps to disambiguate connection
176 establishment and ensure both peers enter the credit flow
177 state machine consistently.
182 Description: kptllnd didn't propagate some network errors up to LNET
183 Details : This bug was spotted while investigating 11394. The fix
184 ensures network errors on sends and bulk transfers are
185 propagated to LNET/lustre correctly.
187 Severity : enhancement
189 Description: Fixed console chatter in case of -ETIMEDOUT.
191 Severity : enhancement
193 Description: Added D_NETTRACE for recording network packet history
194 (initially only for ptllnd). Also a separate userspace
195 ptllnd facility to gather history which should really be
196 covered by D_NETTRACE too, if only CDEBUG recorded history in
202 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
203 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
204 callback can occur before a connection has actually been
205 established. This caused an assertion failure previously.
207 Severity : enhancement
209 Description: Multiple instances for o2iblnd
210 Details : Allow multiple instances of o2iblnd to enable networking over
211 multiple HCAs and routing between them.
215 Description: lnet deadlock in router_checker
216 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
217 into BH locks to eliminate potential deadlock caused by
218 ksocknal_data_ready() preempting code holding these locks.
222 Description: Millions of failed socklnd connection attempts cause a very slow FS
223 Details : added a new route flag ksnr_scheduled to distinguish from
224 ksnr_connecting, so that a peer connection request is only turned
225 down for race concerns when an active connection to the same peer
226 is under progress (instead of just being scheduled).
228 ------------------------------------------------------------------------------
230 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
232 * Support for networks:
233 socklnd - kernels up to 2.6.16
234 qswlnd - Qsnet kernel modules 5.20 and later
235 openiblnd - IbGold 1.8.2
237 viblnd - Voltaire ibhost 3.4.5 and later
238 ciblnd - Topspin 3.2.0
239 iiblnd - Infiniserv 3.3 + PathBits patch
240 gmlnd - GM 2.1.22 and later
241 mxlnd - MX 1.2.1 or later
242 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
245 Severity : major on XT3
247 Description: libcfs overwrites /proc/sys/portals
248 Details : libcfs created a symlink from /proc/sys/portals to
249 /proc/sys/lnet for backwards compatibility. This is no
250 longer required and makes the Cray portals /proc variables
255 Description: OFED FMR API change
256 Details : This changes parameter usage to reflect a change in
257 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
258 that FMR support is only used in experimental versions of the
259 o2iblnd - this change does not affect standard usage at all.
261 Severity : enhancement
263 Description: new ko2iblnd module parameter: ib_mtu
264 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
265 HCAs. You can avoid this problem by setting the MTU to 1024
266 using this module parameter.
268 Severity : enhancement
269 Bugzilla : 11118/11620
270 Description: ptllnd small request message buffer alignment fix
271 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
272 Round up small message size on sends in case this option
273 is not supported. 11620 was a defect in the initial
274 implementation which effectively asserted all peers had to be
275 running the correct protocol version which was fixed by always
276 NAK-ing such requests and handling any misalignments they
281 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
282 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
283 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
285 Severity : enhancement
287 Description: Patchless ZC(zero copy) socklnd
288 Details : New protocol for socklnd, socklnd can support zero copy without
289 kernel patch, it's compatible with old socklnd. Checksum is
290 moved from tunables to modparams.
294 Description: When ksocknal_del_peer() is called upon a peer whose
295 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
296 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
299 Frequency : when ptlrpc is under heavy use and runs out of request buffer
301 Description: In lnet_match_blocked_msg(), md can be used without holding a
305 Frequency : very rarely
307 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
308 If connd connects a route which has been closed by
309 ksocknal_shutdown(), ksocknal_create_routes() may create new
310 routes which hold references on the peer, causing shutdown
311 process to wait for peer to disappear forever.
313 Severity : enhancement
315 Description: Dump XT3 portals traces on kptllnd timeout
316 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
317 dump Cray portals debug traces to a file. The kptllnd module
318 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
319 is the basename of the dump file.
322 Frequency : infrequent
324 Description: kernel ptllnd fix bug in connection re-establishment
325 Details : Kernel ptllnd could produce protocol errors e.g. illegal
326 matchbits and/or violate the credit flow protocol when trying
327 to re-establish a connection with a peer after an error or
330 Severity : enhancement
332 Description: Allow /proc/sys/lnet/debug to be set symbolically
333 Details : Allow debug and subsystem debug values to be read/set by name
334 in addition to numerically, for ease of use.
337 Frequency : only in configurations with LNET routers
339 Description: routes automatically marked down and recovered
340 Details : In configurations with LNET routers if a router fails routers
341 now actively try to recover routes that are down, unless they
342 are marked down by an administrator.
344 ------------------------------------------------------------------------------
346 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
349 Frequency : very rarely, in configurations with LNET routers and TCP
351 Description: incorrect data written to files on OSTs
352 Details : In certain high-load conditions incorrect data may be written
353 to files on the OST when using TCP networks.
355 ------------------------------------------------------------------------------
357 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
359 - rework CDEBUG messages rate-limiting mechanism b=10375
360 - add per-socket tunables for socklnd if the kernel is patched b=10327
362 ------------------------------------------------------------------------------
364 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
366 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
367 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
368 - quiet LNET startup LNI message for liblustre b=10128
369 - Better console error messages if 'ip2nets' can't match an IP address
370 - Fixed overflow/use-before-set bugs in linux-time.h
371 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
372 - LNET teardown failed an assertion about the route table being empty
373 - Fixed a crash in LNetEQPoll(<invalid handle>)
374 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
375 - improve debug message for liblustre/Catamount nodes (b=10116)
377 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
378 * Configuration change for the XT3
379 The PTLLND is now used to run Lustre over Portals on the XT3.
380 The configure option(s) --with-cray-portals are no longer
381 used. Rather --with-portals=<path-to-portals-includes> is
382 used to enable building on the XT3. In addition to enable
383 XT3 specific features the option --enable-cray-xt3 must be
386 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
387 * Portals has been removed, replaced by LNET.
388 LNET is new networking infrastructure for Lustre, it includes a
389 reorganized network configuration mode (see the user
390 documentation for full details) as well as support for routing
391 between different network fabrics. Lustre Networking Devices
392 (LNDS) for the supported network fabrics have also been created
393 for this new infrastructure.
395 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
400 Frequency : rare (large Voltaire clusters only)
402 Description: the default number of reserved transmit descriptors was too low
403 for some large clusters
404 Details : As a workaround, the number was increased. A proper fix includes
407 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
412 Frequency : occasional (large-scale events, cluster reboot, network failure)
414 Description: too many error messages on console obscure actual problem and
415 can slow down/panic server, or cause recovery to fail repeatedly
416 Details : enable rate-limiting of console error messages, and some messages
417 that were console errors now only go to the kernel log
419 Severity : enhancement
421 Description: add /proc/sys/portals/catastrophe entry which will report if
422 that node has previously LBUGged
424 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
426 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
428 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
430 - handle error return code in kranal_check_fma_rx() (5915,6054)
432 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
434 - update vibnal (Voltaire IB NAL)
435 - update gmnal (Myrinet NAL), gmnalid
437 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
439 * Landed portals:b_port_step as follows...
441 - removed CFS_DECL_SPIN*
442 just use 'spinlock_t' and initialise with spin_lock_init()
444 - removed CFS_DECL_MUTEX*
445 just use 'struct semaphore' and initialise with init_mutex()
447 - removed CFS_DECL_RWSEM*
448 just use 'struct rw_semaphore' and initialise with init_rwsem()
450 - renamed cfs_sleep_chan -> cfs_waitq
451 cfs_sleep_link -> cfs_waitlink
453 - fixed race in linux version of arch-independent socknal
454 (the ENOMEM/EAGAIN decision).
456 - Didn't fix problems in Darwin version of arch-independent socknal
457 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
459 - removed libcfs types from non-socknal header files (only some types
460 in the header files had been changed; the .c files hadn't been