Newer
Older
* version 1.6.5
* Support for networks:
socklnd - any kernel supported by Lustre,
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
Severity : major
Bugzilla : 13983
Description: ksocknal_send_hello() hit ASSERTION while connecting race
Details : ksocknal_send_hello() hit ASSERTION while connecting race
Severity : major
Bugzilla : 14425
Description: o2iblnd/ptllnd credit deadlock in a routed config.
Details : o2iblnd/ptllnd credit deadlock in a routed config.
Bugzilla : 14956
Description: High load after starting lnet
Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
uptime utility reports high load that looks confusingly.
Severity : normal
Bugzilla : 14838
Description: ksocklnd fails to establish connection if accept_port is high
Details : PID remapping must not be done for active (outgoing) connections
--------------------------------------------------------------------------------
* Support for networks:
socklnd - any kernel supported by Lustre,
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
Severity : normal
Bugzilla : 14387
Description: liblustre network error
Details : liblustre clients should understand LNET_ACCEPT_PORT environment
variable even if they don't start lnet acceptor.
Severity : normal
Bugzilla : 14300
Description: Strange message from lnet (Ignoring prediction from the future)
Details : Incorrect calculation of peer's last_alive value in ksocklnd
--------------------------------------------------------------------------------
2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.6.4
* Support for networks:
socklnd - any kernel supported by Lustre,
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
Severity : normal
Bugzilla : 14238
Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
Severity : normal
Bugzilla : 12494
Description: increase send queue size for ciblnd/openiblnd
Severity : normal
Bugzilla : 12302
Description: new userspace socklnd
Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
with new one - usocklnd.
Severity : enhancement
Bugzilla : 11686
Description: Console message flood
Details : Make cdls ratelimiting more tunable by adding several tunable in
procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
/proc/sys/lnet/console_backoff.
--------------------------------------------------------------------------------
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
o2iblnd - OFED 1.1 and 1.2,
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
Severity : normal
Bugzilla : 12782
Description: /proc/sys/lnet has non-sysctl entries
Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
Severity : major
Bugzilla : 13236
Description: TOE Kernel panic by ksocklnd
Details : offloaded sockets provide their own implementation of sendpage,
can't call tcp_sendpage() directly
Severity : normal
Bugzilla : 10778
Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
Details : races between lnd_shutdown and peer creation prevent
lnd_shutdown from finishing.
Severity : normal
Bugzilla : 13279
Description: open files rlimit 1024 reached while liblustre testing
Details : ulnds/socklnd must close open socket after unsuccessful
'say hello' attempt.
Severity : major
Bugzilla : 13482
Description: build error
Details : fix typos in gmlnd, ptllnd and viblnd
------------------------------------------------------------------------------
socklnd - kernels up to 2.6.16,
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
o2iblnd - OFED 1.1 and 1.2
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.11
* Support for networks:
socklnd - kernels up to 2.6.16,
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
o2iblnd - OFED 1.1
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
Severity : minor
Bugzilla : 13288
Description: Initialize cpumask before use
Severity : major
Bugzilla : 12014
Description: ASSERTION failures when upgrading to the patchless zero-copy
socklnd
Details : This bug affects "rolling upgrades", causing an inconsistent
protocol version negotiation and subsequent assertion failure
during rolling upgrades after the first wave of upgrades.
Bugzilla : 11223
Details : Change "dropped message" CERRORs to D_NETERROR so they are
logged instead of creating "console chatter" when a lustre
timeout races with normal RPC completion.
Severity : minor
Details : lnet_clear_peer_table can wait forever if user forgets to
clear a lazy portal.
Severity : minor
Details : libcfs_id2str should check pid against LNET_PID_ANY.
Severity : major
Bugzilla : 10916
Description: added LNET self test
Details : landing b_self_test
Severity : minor
Frequency : rare
Bugzilla : 12227
Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
struct timeval.
Details : do_div() macro is used incorrectly.
2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
Severity : normal
Bugzilla : 11680
Description: make panic on lbug configurable
Severity : major
Bugzilla : 12316
Description: Add OFED1.2 support to o2iblnd
Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
are installed (other than kernel's in-tree infiniband), there
could be some problem while insmod o2iblnd (mismatch CRC of
ib_* symbols).
If extra Module.symvers is supported in kernel (i.e, 2.6.17),
this link provides solution:
https://bugs.openfabrics.org/show_bug.cgi?id=355
if extra Module.symvers is not supported in kernel, we will
have to run the script in bug 12316 to update
$LINUX/module.symvers before building o2iblnd.
More details about this are in bug 12316.
------------------------------------------------------------------------------
2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.10 / 1.6.0
socklnd - kernels up to 2.6.16,
qswlnd - Qsnet kernel modules 5.20 and later,
openiblnd - IbGold 1.8.2,
o2iblnd - OFED 1.1,
viblnd - Voltaire ibhost 3.4.5 and later,
ciblnd - Topspin 3.2.0,
iiblnd - Infiniserv 3.3 + PathBits patch,
gmlnd - GM 2.1.22 and later,
mxlnd - MX 1.2.1 or later,
ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
Severity : minor
Frequency : rare
Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
possibly accessed in kptllnd_shutdown. Ptllnd should init
kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
Severity : normal
Frequency : rare
Description: gmlnd ignored some transmit errors when finalizing lnet messages.
Severity : minor
Frequency : rare
Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
Severity : minor
Frequency : rare
Description: the_lnet.ln_finalizing was not set when the current thread is
about to complete messages. It only affects multi-threaded
user space LNet.
Severity : normal
Frequency : rare
Bugzilla : 11472
Description: Changed the default kqswlnd ntxmsg=512
Severity : major
Frequency : rare
Bugzilla : 12458
Description: Assertion failure in kernel ptllnd caused by posting passive
bulk buffers before connection establishment complete.
Description: A race in kernel ptllnd between deleting a peer and posting
new communications for it could hang communications -
manifesting as "Unexpectedly long timeout" messages.
Severity : major
Frequency : rare
Bugzilla : 12432
Description: Kernel ptllnd lock ordering issue could hang a node.
Severity : major
Frequency : rare
Bugzilla : 12016
Description: node crash on socket teardown race
Severity : minor
Frequency : 'lctl peer_list' issued on a mx net
Bugzilla : 12237
Description: Enable lctl's peer_list for MXLND
Severity : major
Frequency : after Ptllnd timeouts and portals congestion
Bugzilla : 11659
Description: Credit overflows
Details : This was a bug in ptllnd connection establishment. The fix
implements better peer stamps to disambiguate connection
establishment and ensure both peers enter the credit flow
state machine consistently.
Severity : major
Frequency : rare
Bugzilla : 11394
Description: kptllnd didn't propagate some network errors up to LNET
Details : This bug was spotted while investigating 11394. The fix
ensures network errors on sends and bulk transfers are
propagated to LNET/lustre correctly.
Severity : enhancement
Bugzilla : 10316
Description: Fixed console chatter in case of -ETIMEDOUT.
Severity : enhancement
Bugzilla : 11684
Description: Added D_NETTRACE for recording network packet history
(initially only for ptllnd). Also a separate userspace
ptllnd facility to gather history which should really be
covered by D_NETTRACE too, if only CDEBUG recorded history in
userspace.
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
Severity : major
Frequency : rare
Bugzilla : 11616
Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
callback can occur before a connection has actually been
established. This caused an assertion failure previously.
Severity : enhancement
Bugzilla : 11094
Description: Multiple instances for o2iblnd
Details : Allow multiple instances of o2iblnd to enable networking over
multiple HCAs and routing between them.
Severity : major
Bugzilla : 11201
Description: lnet deadlock in router_checker
Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
into BH locks to eliminate potential deadlock caused by
ksocknal_data_ready() preempting code holding these locks.
Severity : major
Bugzilla : 11126
Description: Millions of failed socklnd connection attempts cause a very slow FS
Details : added a new route flag ksnr_scheduled to distinguish from
ksnr_connecting, so that a peer connection request is only turned
down for race concerns when an active connection to the same peer
is under progress (instead of just being scheduled).
------------------------------------------------------------------------------
2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.9
* Support for networks:
qswlnd - Qsnet kernel modules 5.20 and later
openiblnd - IbGold 1.8.2
o2iblnd - OFED 1.1
viblnd - Voltaire ibhost 3.4.5 and later
ciblnd - Topspin 3.2.0
iiblnd - Infiniserv 3.3 + PathBits patch
gmlnd - GM 2.1.22 and later
mxlnd - MX 1.2.1 or later
ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
* bug fixes
Severity : major on XT3
Bugzilla : none
Description: libcfs overwrites /proc/sys/portals
Details : libcfs created a symlink from /proc/sys/portals to
/proc/sys/lnet for backwards compatibility. This is no
longer required and makes the Cray portals /proc variables
inaccessible.
Severity : minor
Bugzilla : 11312
Description: OFED FMR API change
Details : This changes parameter usage to reflect a change in
ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
that FMR support is only used in experimental versions of the
o2iblnd - this change does not affect standard usage at all.
Severity : enhancement
Bugzilla : 11245
Description: new ko2iblnd module parameter: ib_mtu
Details : the default IB MTU of 2048 performs badly on 23108 Tavor
HCAs. You can avoid this problem by setting the MTU to 1024
using this module parameter.
Severity : enhancement
Bugzilla : 11118/11620
Description: ptllnd small request message buffer alignment fix
Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
Round up small message size on sends in case this option
is not supported. 11620 was a defect in the initial
implementation which effectively asserted all peers had to be
running the correct protocol version which was fixed by always
NAK-ing such requests and handling any misalignments they
introduce.
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
Severity : minor
Frequency : rarely
Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
Severity : enhancement
Bugzilla : 11250
Description: Patchless ZC(zero copy) socklnd
Details : New protocol for socklnd, socklnd can support zero copy without
kernel patch, it's compatible with old socklnd. Checksum is
moved from tunables to modparams.
Severity : minor
Frequency : rarely
Description: When ksocknal_del_peer() is called upon a peer whose
ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
Severity : normal
Frequency : when ptlrpc is under heavy use and runs out of request buffer
Bugzilla : 11318
Description: In lnet_match_blocked_msg(), md can be used without holding a
ref on it.
Severity : minor
Frequency : very rarely
Bugzilla : 10727
Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
If connd connects a route which has been closed by
ksocknal_shutdown(), ksocknal_create_routes() may create new
routes which hold references on the peer, causing shutdown
process to wait for peer to disappear forever.
Severity : enhancement
Bugzilla : 11234
Description: Dump XT3 portals traces on kptllnd timeout
Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
dump Cray portals debug traces to a file. The kptllnd module
parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
is the basename of the dump file.
Severity : major
Frequency : infrequent
Bugzilla : 11308
Description: kernel ptllnd fix bug in connection re-establishment
Details : Kernel ptllnd could produce protocol errors e.g. illegal
matchbits and/or violate the credit flow protocol when trying
to re-establish a connection with a peer after an error or
timeout.
Severity : enhancement
Bugzilla : 10316
Description: Allow /proc/sys/lnet/debug to be set symbolically
Details : Allow debug and subsystem debug values to be read/set by name
in addition to numerically, for ease of use.
Severity : normal
Frequency : only in configurations with LNET routers
Bugzilla : 10316
Description: routes automatically marked down and recovered
Details : In configurations with LNET routers if a router fails routers
now actively try to recover routes that are down, unless they
are marked down by an administrator.
------------------------------------------------------------------------------
2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
Severity : critical
Frequency : very rarely, in configurations with LNET routers and TCP
Bugzilla : 10889
Description: incorrect data written to files on OSTs
Details : In certain high-load conditions incorrect data may be written
to files on the OST when using TCP networks.
------------------------------------------------------------------------------
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.7
- rework CDEBUG messages rate-limiting mechanism b=10375
- add per-socket tunables for socklnd if the kernel is patched b=10327
------------------------------------------------------------------------------
2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.6
- fix use of portals/lnet pid to avoid dropping RPCs b=10074
- iiblnd wasn't mapping all memory, resulting in comms errors b=9776
- quiet LNET startup LNI message for liblustre b=10128
- Better console error messages if 'ip2nets' can't match an IP address
- Fixed overflow/use-before-set bugs in linux-time.h
- Fixed ptllnd bug that wasn't initialising rx descriptors completely
- LNET teardown failed an assertion about the route table being empty
- Fixed a crash in LNetEQPoll(<invalid handle>)
- Future protocol compatibility work (b_rls146_lnetprotovrsn)
- improve debug message for liblustre/Catamount nodes (b=10116)
2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
* Configuration change for the XT3
The PTLLND is now used to run Lustre over Portals on the XT3.
The configure option(s) --with-cray-portals are no longer
used. Rather --with-portals=<path-to-portals-includes> is
used to enable building on the XT3. In addition to enable
XT3 specific features the option --enable-cray-xt3 must be
used.
2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
* Portals has been removed, replaced by LNET.
LNET is new networking infrastructure for Lustre, it includes a
reorganized network configuration mode (see the user
documentation for full details) as well as support for routing
between different network fabrics. Lustre Networking Devices
(LNDS) for the supported network fabrics have also been created
for this new infrastructure.
2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.4
* bug fixes
Severity : major
Frequency : rare (large Voltaire clusters only)
Bugzilla : 6993
Description: the default number of reserved transmit descriptors was too low
Details : As a workaround, the number was increased. A proper fix includes
2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
* version 1.4.3
* bug fixes
Severity : major
Frequency : occasional (large-scale events, cluster reboot, network failure)
Bugzilla : 6411
Description: too many error messages on console obscure actual problem and
can slow down/panic server, or cause recovery to fail repeatedly
Details : enable rate-limiting of console error messages, and some messages
that were console errors now only go to the kernel log
Severity : enhancement
Bugzilla : 1693
Description: add /proc/sys/portals/catastrophe entry which will report if
2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
* bugs
- update gmnal to use PTL_MTU, fix module refcounting (b=5786)
2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
* bugs
- handle error return code in kranal_check_fma_rx() (5915,6054)
2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
* miscellania
- update vibnal (Voltaire IB NAL)
- update gmnal (Myrinet NAL), gmnalid
2005-02-04 Eric Barton <eeb@bartonsoftware.com>
* Landed portals:b_port_step as follows...
- removed CFS_DECL_SPIN*
just use 'spinlock_t' and initialise with spin_lock_init()
- removed CFS_DECL_MUTEX*
just use 'struct semaphore' and initialise with init_mutex()
- removed CFS_DECL_RWSEM*
just use 'struct rw_semaphore' and initialise with init_rwsem()
- renamed cfs_sleep_chan -> cfs_waitq
cfs_sleep_link -> cfs_waitlink
- fixed race in linux version of arch-independent socknal
(the ENOMEM/EAGAIN decision).
- Didn't fix problems in Darwin version of arch-independent socknal
(resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
- removed libcfs types from non-socknal header files (only some types
in the header files had been changed; the .c files hadn't been
updated at all).