1 2007-10-26 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: new userspace socklnd
18 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
19 with new one - usocklnd.
21 Severity : enhancement
23 Description: Console message flood
24 Details : Make cdls ratelimiting more tunable by adding several tunable in
25 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
26 /proc/sys/lnet/console_backoff.
28 --------------------------------------------------------------------------------
30 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
32 * Support for networks:
33 socklnd - any kernel supported by Lustre,
34 qswlnd - Qsnet kernel modules 5.20 and later,
35 openiblnd - IbGold 1.8.2,
36 o2iblnd - OFED 1.1 and 1.2,
37 viblnd - Voltaire ibhost 3.4.5 and later,
38 ciblnd - Topspin 3.2.0,
39 iiblnd - Infiniserv 3.3 + PathBits patch,
40 gmlnd - GM 2.1.22 and later,
41 mxlnd - MX 1.2.1 or later,
42 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
46 Description: /proc/sys/lnet has non-sysctl entries
47 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
51 Description: TOE Kernel panic by ksocklnd
52 Details : offloaded sockets provide their own implementation of sendpage,
53 can't call tcp_sendpage() directly
57 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
58 Details : races between lnd_shutdown and peer creation prevent
59 lnd_shutdown from finishing.
63 Description: open files rlimit 1024 reached while liblustre testing
64 Details : ulnds/socklnd must close open socket after unsuccessful
69 Description: build error
70 Details : fix typos in gmlnd, ptllnd and viblnd
72 ------------------------------------------------------------------------------
74 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
76 * Support for networks:
77 socklnd - kernels up to 2.6.16,
78 qswlnd - Qsnet kernel modules 5.20 and later,
79 openiblnd - IbGold 1.8.2,
80 o2iblnd - OFED 1.1 and 1.2
81 viblnd - Voltaire ibhost 3.4.5 and later,
82 ciblnd - Topspin 3.2.0,
83 iiblnd - Infiniserv 3.3 + PathBits patch,
84 gmlnd - GM 2.1.22 and later,
85 mxlnd - MX 1.2.1 or later,
86 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
88 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
90 * Support for networks:
91 socklnd - kernels up to 2.6.16,
92 qswlnd - Qsnet kernel modules 5.20 and later,
93 openiblnd - IbGold 1.8.2,
95 viblnd - Voltaire ibhost 3.4.5 and later,
96 ciblnd - Topspin 3.2.0,
97 iiblnd - Infiniserv 3.3 + PathBits patch,
98 gmlnd - GM 2.1.22 and later,
99 mxlnd - MX 1.2.1 or later,
100 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
104 Description: Initialize cpumask before use
108 Description: ASSERTION failures when upgrading to the patchless zero-copy
110 Details : This bug affects "rolling upgrades", causing an inconsistent
111 protocol version negotiation and subsequent assertion failure
112 during rolling upgrades after the first wave of upgrades.
116 Details : Change "dropped message" CERRORs to D_NETERROR so they are
117 logged instead of creating "console chatter" when a lustre
118 timeout races with normal RPC completion.
121 Details : lnet_clear_peer_table can wait forever if user forgets to
125 Details : libcfs_id2str should check pid against LNET_PID_ANY.
129 Description: added LNET self test
130 Details : landing b_self_test
135 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
137 Details : do_div() macro is used incorrectly.
139 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
143 Description: make panic on lbug configurable
147 Description: Add OFED1.2 support to o2iblnd
148 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
149 are installed (other than kernel's in-tree infiniband), there
150 could be some problem while insmod o2iblnd (mismatch CRC of
152 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
153 this link provides solution:
154 https://bugs.openfabrics.org/show_bug.cgi?id=355
155 if extra Module.symvers is not supported in kernel, we will
156 have to run the script in bug 12316 to update
157 $LINUX/module.symvers before building o2iblnd.
158 More details about this are in bug 12316.
160 ------------------------------------------------------------------------------
162 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
163 * version 1.4.10 / 1.6.0
164 * Support for networks:
165 socklnd - kernels up to 2.6.16,
166 qswlnd - Qsnet kernel modules 5.20 and later,
167 openiblnd - IbGold 1.8.2,
169 viblnd - Voltaire ibhost 3.4.5 and later,
170 ciblnd - Topspin 3.2.0,
171 iiblnd - Infiniserv 3.3 + PathBits patch,
172 gmlnd - GM 2.1.22 and later,
173 mxlnd - MX 1.2.1 or later,
174 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
178 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
179 possibly accessed in kptllnd_shutdown. Ptllnd should init
180 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
184 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
188 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
192 Description: the_lnet.ln_finalizing was not set when the current thread is
193 about to complete messages. It only affects multi-threaded
199 Description: Changed the default kqswlnd ntxmsg=512
204 Description: Assertion failure in kernel ptllnd caused by posting passive
205 bulk buffers before connection establishment complete.
210 Description: A race in kernel ptllnd between deleting a peer and posting
211 new communications for it could hang communications -
212 manifesting as "Unexpectedly long timeout" messages.
217 Description: Kernel ptllnd lock ordering issue could hang a node.
222 Description: node crash on socket teardown race
225 Frequency : 'lctl peer_list' issued on a mx net
227 Description: Enable lctl's peer_list for MXLND
230 Frequency : after Ptllnd timeouts and portals congestion
232 Description: Credit overflows
233 Details : This was a bug in ptllnd connection establishment. The fix
234 implements better peer stamps to disambiguate connection
235 establishment and ensure both peers enter the credit flow
236 state machine consistently.
241 Description: kptllnd didn't propagate some network errors up to LNET
242 Details : This bug was spotted while investigating 11394. The fix
243 ensures network errors on sends and bulk transfers are
244 propagated to LNET/lustre correctly.
246 Severity : enhancement
248 Description: Fixed console chatter in case of -ETIMEDOUT.
250 Severity : enhancement
252 Description: Added D_NETTRACE for recording network packet history
253 (initially only for ptllnd). Also a separate userspace
254 ptllnd facility to gather history which should really be
255 covered by D_NETTRACE too, if only CDEBUG recorded history in
261 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
262 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
263 callback can occur before a connection has actually been
264 established. This caused an assertion failure previously.
266 Severity : enhancement
268 Description: Multiple instances for o2iblnd
269 Details : Allow multiple instances of o2iblnd to enable networking over
270 multiple HCAs and routing between them.
274 Description: lnet deadlock in router_checker
275 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
276 into BH locks to eliminate potential deadlock caused by
277 ksocknal_data_ready() preempting code holding these locks.
281 Description: Millions of failed socklnd connection attempts cause a very slow FS
282 Details : added a new route flag ksnr_scheduled to distinguish from
283 ksnr_connecting, so that a peer connection request is only turned
284 down for race concerns when an active connection to the same peer
285 is under progress (instead of just being scheduled).
287 ------------------------------------------------------------------------------
289 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
291 * Support for networks:
292 socklnd - kernels up to 2.6.16
293 qswlnd - Qsnet kernel modules 5.20 and later
294 openiblnd - IbGold 1.8.2
296 viblnd - Voltaire ibhost 3.4.5 and later
297 ciblnd - Topspin 3.2.0
298 iiblnd - Infiniserv 3.3 + PathBits patch
299 gmlnd - GM 2.1.22 and later
300 mxlnd - MX 1.2.1 or later
301 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
304 Severity : major on XT3
306 Description: libcfs overwrites /proc/sys/portals
307 Details : libcfs created a symlink from /proc/sys/portals to
308 /proc/sys/lnet for backwards compatibility. This is no
309 longer required and makes the Cray portals /proc variables
314 Description: OFED FMR API change
315 Details : This changes parameter usage to reflect a change in
316 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
317 that FMR support is only used in experimental versions of the
318 o2iblnd - this change does not affect standard usage at all.
320 Severity : enhancement
322 Description: new ko2iblnd module parameter: ib_mtu
323 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
324 HCAs. You can avoid this problem by setting the MTU to 1024
325 using this module parameter.
327 Severity : enhancement
328 Bugzilla : 11118/11620
329 Description: ptllnd small request message buffer alignment fix
330 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
331 Round up small message size on sends in case this option
332 is not supported. 11620 was a defect in the initial
333 implementation which effectively asserted all peers had to be
334 running the correct protocol version which was fixed by always
335 NAK-ing such requests and handling any misalignments they
340 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
341 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
342 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
344 Severity : enhancement
346 Description: Patchless ZC(zero copy) socklnd
347 Details : New protocol for socklnd, socklnd can support zero copy without
348 kernel patch, it's compatible with old socklnd. Checksum is
349 moved from tunables to modparams.
353 Description: When ksocknal_del_peer() is called upon a peer whose
354 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
355 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
358 Frequency : when ptlrpc is under heavy use and runs out of request buffer
360 Description: In lnet_match_blocked_msg(), md can be used without holding a
364 Frequency : very rarely
366 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
367 If connd connects a route which has been closed by
368 ksocknal_shutdown(), ksocknal_create_routes() may create new
369 routes which hold references on the peer, causing shutdown
370 process to wait for peer to disappear forever.
372 Severity : enhancement
374 Description: Dump XT3 portals traces on kptllnd timeout
375 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
376 dump Cray portals debug traces to a file. The kptllnd module
377 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
378 is the basename of the dump file.
381 Frequency : infrequent
383 Description: kernel ptllnd fix bug in connection re-establishment
384 Details : Kernel ptllnd could produce protocol errors e.g. illegal
385 matchbits and/or violate the credit flow protocol when trying
386 to re-establish a connection with a peer after an error or
389 Severity : enhancement
391 Description: Allow /proc/sys/lnet/debug to be set symbolically
392 Details : Allow debug and subsystem debug values to be read/set by name
393 in addition to numerically, for ease of use.
396 Frequency : only in configurations with LNET routers
398 Description: routes automatically marked down and recovered
399 Details : In configurations with LNET routers if a router fails routers
400 now actively try to recover routes that are down, unless they
401 are marked down by an administrator.
403 ------------------------------------------------------------------------------
405 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
408 Frequency : very rarely, in configurations with LNET routers and TCP
410 Description: incorrect data written to files on OSTs
411 Details : In certain high-load conditions incorrect data may be written
412 to files on the OST when using TCP networks.
414 ------------------------------------------------------------------------------
416 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
418 - rework CDEBUG messages rate-limiting mechanism b=10375
419 - add per-socket tunables for socklnd if the kernel is patched b=10327
421 ------------------------------------------------------------------------------
423 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
425 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
426 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
427 - quiet LNET startup LNI message for liblustre b=10128
428 - Better console error messages if 'ip2nets' can't match an IP address
429 - Fixed overflow/use-before-set bugs in linux-time.h
430 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
431 - LNET teardown failed an assertion about the route table being empty
432 - Fixed a crash in LNetEQPoll(<invalid handle>)
433 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
434 - improve debug message for liblustre/Catamount nodes (b=10116)
436 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
437 * Configuration change for the XT3
438 The PTLLND is now used to run Lustre over Portals on the XT3.
439 The configure option(s) --with-cray-portals are no longer
440 used. Rather --with-portals=<path-to-portals-includes> is
441 used to enable building on the XT3. In addition to enable
442 XT3 specific features the option --enable-cray-xt3 must be
445 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
446 * Portals has been removed, replaced by LNET.
447 LNET is new networking infrastructure for Lustre, it includes a
448 reorganized network configuration mode (see the user
449 documentation for full details) as well as support for routing
450 between different network fabrics. Lustre Networking Devices
451 (LNDS) for the supported network fabrics have also been created
452 for this new infrastructure.
454 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
459 Frequency : rare (large Voltaire clusters only)
461 Description: the default number of reserved transmit descriptors was too low
462 for some large clusters
463 Details : As a workaround, the number was increased. A proper fix includes
466 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
471 Frequency : occasional (large-scale events, cluster reboot, network failure)
473 Description: too many error messages on console obscure actual problem and
474 can slow down/panic server, or cause recovery to fail repeatedly
475 Details : enable rate-limiting of console error messages, and some messages
476 that were console errors now only go to the kernel log
478 Severity : enhancement
480 Description: add /proc/sys/portals/catastrophe entry which will report if
481 that node has previously LBUGged
483 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
485 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
487 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
489 - handle error return code in kranal_check_fma_rx() (5915,6054)
491 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
493 - update vibnal (Voltaire IB NAL)
494 - update gmnal (Myrinet NAL), gmnalid
496 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
498 * Landed portals:b_port_step as follows...
500 - removed CFS_DECL_SPIN*
501 just use 'spinlock_t' and initialise with spin_lock_init()
503 - removed CFS_DECL_MUTEX*
504 just use 'struct semaphore' and initialise with init_mutex()
506 - removed CFS_DECL_RWSEM*
507 just use 'struct rw_semaphore' and initialise with init_rwsem()
509 - renamed cfs_sleep_chan -> cfs_waitq
510 cfs_sleep_link -> cfs_waitlink
512 - fixed race in linux version of arch-independent socknal
513 (the ENOMEM/EAGAIN decision).
515 - Didn't fix problems in Darwin version of arch-independent socknal
516 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
518 - removed libcfs types from non-socknal header files (only some types
519 in the header files had been changed; the .c files hadn't been