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: TOE Kernel panic by ksocklnd
18 Details : offloaded sockets provide their own implementation of sendpage,
19 can't call tcp_sendpage() directly
23 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
24 Details : races between lnd_shutdown and peer creation prevent
25 lnd_shutdown from finishing.
29 Description: open files rlimit 1024 reached while liblustre testing
30 Details : ulnds/socklnd must close open socket after unsuccessful
33 ------------------------------------------------------------------------------
35 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
37 * Support for networks:
38 socklnd - kernels up to 2.6.16,
39 qswlnd - Qsnet kernel modules 5.20 and later,
40 openiblnd - IbGold 1.8.2,
41 o2iblnd - OFED 1.1 and 1.2
42 viblnd - Voltaire ibhost 3.4.5 and later,
43 ciblnd - Topspin 3.2.0,
44 iiblnd - Infiniserv 3.3 + PathBits patch,
45 gmlnd - GM 2.1.22 and later,
46 mxlnd - MX 1.2.1 or later,
47 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
49 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
51 * Support for networks:
52 socklnd - kernels up to 2.6.16,
53 qswlnd - Qsnet kernel modules 5.20 and later,
54 openiblnd - IbGold 1.8.2,
56 viblnd - Voltaire ibhost 3.4.5 and later,
57 ciblnd - Topspin 3.2.0,
58 iiblnd - Infiniserv 3.3 + PathBits patch,
59 gmlnd - GM 2.1.22 and later,
60 mxlnd - MX 1.2.1 or later,
61 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
65 Description: Initialize cpumask before use
69 Description: ASSERTION failures when upgrading to the patchless zero-copy
71 Details : This bug affects "rolling upgrades", causing an inconsistent
72 protocol version negotiation and subsequent assertion failure
73 during rolling upgrades after the first wave of upgrades.
77 Details : Change "dropped message" CERRORs to D_NETERROR so they are
78 logged instead of creating "console chatter" when a lustre
79 timeout races with normal RPC completion.
82 Details : lnet_clear_peer_table can wait forever if user forgets to
86 Details : libcfs_id2str should check pid against LNET_PID_ANY.
90 Description: added LNET self test
91 Details : landing b_self_test
96 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
98 Details : do_div() macro is used incorrectly.
100 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
104 Description: make panic on lbug configurable
108 Description: Add OFED1.2 support to o2iblnd
109 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
110 are installed (other than kernel's in-tree infiniband), there
111 could be some problem while insmod o2iblnd (mismatch CRC of
113 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
114 this link provides solution:
115 https://bugs.openfabrics.org/show_bug.cgi?id=355
116 if extra Module.symvers is not supported in kernel, we will
117 have to run the script in bug 12316 to update
118 $LINUX/module.symvers before building o2iblnd.
119 More details about this are in bug 12316.
121 ------------------------------------------------------------------------------
123 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
124 * version 1.4.10 / 1.6.0
125 * Support for networks:
126 socklnd - kernels up to 2.6.16,
127 qswlnd - Qsnet kernel modules 5.20 and later,
128 openiblnd - IbGold 1.8.2,
130 viblnd - Voltaire ibhost 3.4.5 and later,
131 ciblnd - Topspin 3.2.0,
132 iiblnd - Infiniserv 3.3 + PathBits patch,
133 gmlnd - GM 2.1.22 and later,
134 mxlnd - MX 1.2.1 or later,
135 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
139 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
140 possibly accessed in kptllnd_shutdown. Ptllnd should init
141 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
145 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
149 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
153 Description: the_lnet.ln_finalizing was not set when the current thread is
154 about to complete messages. It only affects multi-threaded
160 Description: Changed the default kqswlnd ntxmsg=512
165 Description: Assertion failure in kernel ptllnd caused by posting passive
166 bulk buffers before connection establishment complete.
171 Description: A race in kernel ptllnd between deleting a peer and posting
172 new communications for it could hang communications -
173 manifesting as "Unexpectedly long timeout" messages.
178 Description: Kernel ptllnd lock ordering issue could hang a node.
183 Description: node crash on socket teardown race
186 Frequency : 'lctl peer_list' issued on a mx net
188 Description: Enable lctl's peer_list for MXLND
191 Frequency : after Ptllnd timeouts and portals congestion
193 Description: Credit overflows
194 Details : This was a bug in ptllnd connection establishment. The fix
195 implements better peer stamps to disambiguate connection
196 establishment and ensure both peers enter the credit flow
197 state machine consistently.
202 Description: kptllnd didn't propagate some network errors up to LNET
203 Details : This bug was spotted while investigating 11394. The fix
204 ensures network errors on sends and bulk transfers are
205 propagated to LNET/lustre correctly.
207 Severity : enhancement
209 Description: Fixed console chatter in case of -ETIMEDOUT.
211 Severity : enhancement
213 Description: Added D_NETTRACE for recording network packet history
214 (initially only for ptllnd). Also a separate userspace
215 ptllnd facility to gather history which should really be
216 covered by D_NETTRACE too, if only CDEBUG recorded history in
222 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
223 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
224 callback can occur before a connection has actually been
225 established. This caused an assertion failure previously.
227 Severity : enhancement
229 Description: Multiple instances for o2iblnd
230 Details : Allow multiple instances of o2iblnd to enable networking over
231 multiple HCAs and routing between them.
235 Description: lnet deadlock in router_checker
236 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
237 into BH locks to eliminate potential deadlock caused by
238 ksocknal_data_ready() preempting code holding these locks.
242 Description: Millions of failed socklnd connection attempts cause a very slow FS
243 Details : added a new route flag ksnr_scheduled to distinguish from
244 ksnr_connecting, so that a peer connection request is only turned
245 down for race concerns when an active connection to the same peer
246 is under progress (instead of just being scheduled).
248 ------------------------------------------------------------------------------
250 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
252 * Support for networks:
253 socklnd - kernels up to 2.6.16
254 qswlnd - Qsnet kernel modules 5.20 and later
255 openiblnd - IbGold 1.8.2
257 viblnd - Voltaire ibhost 3.4.5 and later
258 ciblnd - Topspin 3.2.0
259 iiblnd - Infiniserv 3.3 + PathBits patch
260 gmlnd - GM 2.1.22 and later
261 mxlnd - MX 1.2.1 or later
262 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
265 Severity : major on XT3
267 Description: libcfs overwrites /proc/sys/portals
268 Details : libcfs created a symlink from /proc/sys/portals to
269 /proc/sys/lnet for backwards compatibility. This is no
270 longer required and makes the Cray portals /proc variables
275 Description: OFED FMR API change
276 Details : This changes parameter usage to reflect a change in
277 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
278 that FMR support is only used in experimental versions of the
279 o2iblnd - this change does not affect standard usage at all.
281 Severity : enhancement
283 Description: new ko2iblnd module parameter: ib_mtu
284 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
285 HCAs. You can avoid this problem by setting the MTU to 1024
286 using this module parameter.
288 Severity : enhancement
289 Bugzilla : 11118/11620
290 Description: ptllnd small request message buffer alignment fix
291 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
292 Round up small message size on sends in case this option
293 is not supported. 11620 was a defect in the initial
294 implementation which effectively asserted all peers had to be
295 running the correct protocol version which was fixed by always
296 NAK-ing such requests and handling any misalignments they
301 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
302 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
303 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
305 Severity : enhancement
307 Description: Patchless ZC(zero copy) socklnd
308 Details : New protocol for socklnd, socklnd can support zero copy without
309 kernel patch, it's compatible with old socklnd. Checksum is
310 moved from tunables to modparams.
314 Description: When ksocknal_del_peer() is called upon a peer whose
315 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
316 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
319 Frequency : when ptlrpc is under heavy use and runs out of request buffer
321 Description: In lnet_match_blocked_msg(), md can be used without holding a
325 Frequency : very rarely
327 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
328 If connd connects a route which has been closed by
329 ksocknal_shutdown(), ksocknal_create_routes() may create new
330 routes which hold references on the peer, causing shutdown
331 process to wait for peer to disappear forever.
333 Severity : enhancement
335 Description: Dump XT3 portals traces on kptllnd timeout
336 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
337 dump Cray portals debug traces to a file. The kptllnd module
338 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
339 is the basename of the dump file.
342 Frequency : infrequent
344 Description: kernel ptllnd fix bug in connection re-establishment
345 Details : Kernel ptllnd could produce protocol errors e.g. illegal
346 matchbits and/or violate the credit flow protocol when trying
347 to re-establish a connection with a peer after an error or
350 Severity : enhancement
352 Description: Allow /proc/sys/lnet/debug to be set symbolically
353 Details : Allow debug and subsystem debug values to be read/set by name
354 in addition to numerically, for ease of use.
357 Frequency : only in configurations with LNET routers
359 Description: routes automatically marked down and recovered
360 Details : In configurations with LNET routers if a router fails routers
361 now actively try to recover routes that are down, unless they
362 are marked down by an administrator.
364 ------------------------------------------------------------------------------
366 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
369 Frequency : very rarely, in configurations with LNET routers and TCP
371 Description: incorrect data written to files on OSTs
372 Details : In certain high-load conditions incorrect data may be written
373 to files on the OST when using TCP networks.
375 ------------------------------------------------------------------------------
377 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
379 - rework CDEBUG messages rate-limiting mechanism b=10375
380 - add per-socket tunables for socklnd if the kernel is patched b=10327
382 ------------------------------------------------------------------------------
384 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
386 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
387 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
388 - quiet LNET startup LNI message for liblustre b=10128
389 - Better console error messages if 'ip2nets' can't match an IP address
390 - Fixed overflow/use-before-set bugs in linux-time.h
391 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
392 - LNET teardown failed an assertion about the route table being empty
393 - Fixed a crash in LNetEQPoll(<invalid handle>)
394 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
395 - improve debug message for liblustre/Catamount nodes (b=10116)
397 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
398 * Configuration change for the XT3
399 The PTLLND is now used to run Lustre over Portals on the XT3.
400 The configure option(s) --with-cray-portals are no longer
401 used. Rather --with-portals=<path-to-portals-includes> is
402 used to enable building on the XT3. In addition to enable
403 XT3 specific features the option --enable-cray-xt3 must be
406 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
407 * Portals has been removed, replaced by LNET.
408 LNET is new networking infrastructure for Lustre, it includes a
409 reorganized network configuration mode (see the user
410 documentation for full details) as well as support for routing
411 between different network fabrics. Lustre Networking Devices
412 (LNDS) for the supported network fabrics have also been created
413 for this new infrastructure.
415 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
420 Frequency : rare (large Voltaire clusters only)
422 Description: the default number of reserved transmit descriptors was too low
423 for some large clusters
424 Details : As a workaround, the number was increased. A proper fix includes
427 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
432 Frequency : occasional (large-scale events, cluster reboot, network failure)
434 Description: too many error messages on console obscure actual problem and
435 can slow down/panic server, or cause recovery to fail repeatedly
436 Details : enable rate-limiting of console error messages, and some messages
437 that were console errors now only go to the kernel log
439 Severity : enhancement
441 Description: add /proc/sys/portals/catastrophe entry which will report if
442 that node has previously LBUGged
444 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
446 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
448 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
450 - handle error return code in kranal_check_fma_rx() (5915,6054)
452 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
454 - update vibnal (Voltaire IB NAL)
455 - update gmnal (Myrinet NAL), gmnalid
457 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
459 * Landed portals:b_port_step as follows...
461 - removed CFS_DECL_SPIN*
462 just use 'spinlock_t' and initialise with spin_lock_init()
464 - removed CFS_DECL_MUTEX*
465 just use 'struct semaphore' and initialise with init_mutex()
467 - removed CFS_DECL_RWSEM*
468 just use 'struct rw_semaphore' and initialise with init_rwsem()
470 - renamed cfs_sleep_chan -> cfs_waitq
471 cfs_sleep_link -> cfs_waitlink
473 - fixed race in linux version of arch-independent socknal
474 (the ENOMEM/EAGAIN decision).
476 - Didn't fix problems in Darwin version of arch-independent socknal
477 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
479 - removed libcfs types from non-socknal header files (only some types
480 in the header files had been changed; the .c files hadn't been