1 2006-06-22 Cluster File Systems, Inc. <info@clusterfs.com>
2 * version 1.4.11 / 1.6.1
3 * Support for networks:
4 socklnd - kernels up to 2.6.16
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
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
19 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
21 Details : do_div() macro is used incorrectly.
23 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
27 Description: make panic on lbug configurable
31 Description: Add OFED1.2 support to o2iblnd
32 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
33 are installed (other than kernel's in-tree infiniband), there
34 could be some problem while insmod o2iblnd (mismatch CRC of
36 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
37 this link provides solution:
38 https://bugs.openfabrics.org/show_bug.cgi?id=355
39 if extra Module.symvers is not supported in kernel, we will
40 have to run the script in bug 12316 to update
41 $LINUX/module.symvers before building o2iblnd.
42 More details about this are in bug 12316.
44 ------------------------------------------------------------------------------
46 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
47 * version 1.4.10 / 1.6.0
48 * Support for networks:
49 socklnd - kernels up to 2.6.16
50 qswlnd - Qsnet kernel modules 5.20 and later
51 openiblnd - IbGold 1.8.2
53 viblnd - Voltaire ibhost 3.4.5 and later
54 ciblnd - Topspin 3.2.0
55 iiblnd - Infiniserv 3.3 + PathBits patch
56 gmlnd - GM 2.1.22 and later
57 mxlnd - MX 1.2.1 or later
58 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
63 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
64 possibly accessed in kptllnd_shutdown. Ptllnd should init
65 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
69 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
73 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
77 Description: the_lnet.ln_finalizing was not set when the current thread is
78 about to complete messages. It only affects multi-threaded
84 Description: Changed the default kqswlnd ntxmsg=512
89 Description: Assertion failure in kernel ptllnd caused by posting passive
90 bulk buffers before connection establishment complete.
95 Description: A race in kernel ptllnd between deleting a peer and posting
96 new communications for it could hang communications -
97 manifesting as "Unexpectedly long timeout" messages.
102 Description: Kernel ptllnd lock ordering issue could hang a node.
107 Description: node crash on socket teardown race
110 Frequency : 'lctl peer_list' issued on a mx net
112 Description: Enable lctl's peer_list for MXLND
115 Frequency : after Ptllnd timeouts and portals congestion
117 Description: Credit overflows
118 Details : This was a bug in ptllnd connection establishment. The fix
119 implements better peer stamps to disambiguate connection
120 establishment and ensure both peers enter the credit flow
121 state machine consistently.
126 Description: kptllnd didn't propagate some network errors up to LNET
127 Details : This bug was spotted while investigating 11394. The fix
128 ensures network errors on sends and bulk transfers are
129 propagated to LNET/lustre correctly.
131 Severity : enhancement
133 Description: Fixed console chatter in case of -ETIMEDOUT.
135 Severity : enhancement
137 Description: Added D_NETTRACE for recording network packet history
138 (initially only for ptllnd). Also a separate userspace
139 ptllnd facility to gather history which should really be
140 covered by D_NETTRACE too, if only CDEBUG recorded history in
146 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
147 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
148 callback can occur before a connection has actually been
149 established. This caused an assertion failure previously.
151 Severity : enhancement
153 Description: Multiple instances for o2iblnd
154 Details : Allow multiple instances of o2iblnd to enable networking over
155 multiple HCAs and routing between them.
159 Description: lnet deadlock in router_checker
160 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
161 into BH locks to eliminate potential deadlock caused by
162 ksocknal_data_ready() preempting code holding these locks.
166 Description: Millions of failed socklnd connection attempts cause a very slow FS
167 Details : added a new route flag ksnr_scheduled to distinguish from
168 ksnr_connecting, so that a peer connection request is only turned
169 down for race concerns when an active connection to the same peer
170 is under progress (instead of just being scheduled).
172 ------------------------------------------------------------------------------
174 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
176 * Support for networks:
177 socklnd - kernels up to 2.6.16
178 qswlnd - Qsnet kernel modules 5.20 and later
179 openiblnd - IbGold 1.8.2
181 viblnd - Voltaire ibhost 3.4.5 and later
182 ciblnd - Topspin 3.2.0
183 iiblnd - Infiniserv 3.3 + PathBits patch
184 gmlnd - GM 2.1.22 and later
185 mxlnd - MX 1.2.1 or later
186 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
189 Severity : major on XT3
191 Description: libcfs overwrites /proc/sys/portals
192 Details : libcfs created a symlink from /proc/sys/portals to
193 /proc/sys/lnet for backwards compatibility. This is no
194 longer required and makes the Cray portals /proc variables
199 Description: OFED FMR API change
200 Details : This changes parameter usage to reflect a change in
201 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
202 that FMR support is only used in experimental versions of the
203 o2iblnd - this change does not affect standard usage at all.
205 Severity : enhancement
207 Description: new ko2iblnd module parameter: ib_mtu
208 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
209 HCAs. You can avoid this problem by setting the MTU to 1024
210 using this module parameter.
212 Severity : enhancement
213 Bugzilla : 11118/11620
214 Description: ptllnd small request message buffer alignment fix
215 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
216 Round up small message size on sends in case this option
217 is not supported. 11620 was a defect in the initial
218 implementation which effectively asserted all peers had to be
219 running the correct protocol version which was fixed by always
220 NAK-ing such requests and handling any misalignments they
225 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
226 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
227 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
229 Severity : enhancement
231 Description: Patchless ZC(zero copy) socklnd
232 Details : New protocol for socklnd, socklnd can support zero copy without
233 kernel patch, it's compatible with old socklnd. Checksum is
234 moved from tunables to modparams.
238 Description: When ksocknal_del_peer() is called upon a peer whose
239 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
240 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
243 Frequency : when ptlrpc is under heavy use and runs out of request buffer
245 Description: In lnet_match_blocked_msg(), md can be used without holding a
249 Frequency : very rarely
251 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
252 If connd connects a route which has been closed by
253 ksocknal_shutdown(), ksocknal_create_routes() may create new
254 routes which hold references on the peer, causing shutdown
255 process to wait for peer to disappear forever.
257 Severity : enhancement
259 Description: Dump XT3 portals traces on kptllnd timeout
260 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
261 dump Cray portals debug traces to a file. The kptllnd module
262 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
263 is the basename of the dump file.
266 Frequency : infrequent
268 Description: kernel ptllnd fix bug in connection re-establishment
269 Details : Kernel ptllnd could produce protocol errors e.g. illegal
270 matchbits and/or violate the credit flow protocol when trying
271 to re-establish a connection with a peer after an error or
274 Severity : enhancement
276 Description: Allow /proc/sys/lnet/debug to be set symbolically
277 Details : Allow debug and subsystem debug values to be read/set by name
278 in addition to numerically, for ease of use.
281 Frequency : only in configurations with LNET routers
283 Description: routes automatically marked down and recovered
284 Details : In configurations with LNET routers if a router fails routers
285 now actively try to recover routes that are down, unless they
286 are marked down by an administrator.
288 ------------------------------------------------------------------------------
290 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
293 Frequency : very rarely, in configurations with LNET routers and TCP
295 Description: incorrect data written to files on OSTs
296 Details : In certain high-load conditions incorrect data may be written
297 to files on the OST when using TCP networks.
299 ------------------------------------------------------------------------------
301 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
303 - rework CDEBUG messages rate-limiting mechanism b=10375
304 - add per-socket tunables for socklnd if the kernel is patched b=10327
306 ------------------------------------------------------------------------------
308 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
310 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
311 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
312 - quiet LNET startup LNI message for liblustre b=10128
313 - Better console error messages if 'ip2nets' can't match an IP address
314 - Fixed overflow/use-before-set bugs in linux-time.h
315 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
316 - LNET teardown failed an assertion about the route table being empty
317 - Fixed a crash in LNetEQPoll(<invalid handle>)
318 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
319 - improve debug message for liblustre/Catamount nodes (b=10116)
321 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
322 * Configuration change for the XT3
323 The PTLLND is now used to run Lustre over Portals on the XT3.
324 The configure option(s) --with-cray-portals are no longer
325 used. Rather --with-portals=<path-to-portals-includes> is
326 used to enable building on the XT3. In addition to enable
327 XT3 specific features the option --enable-cray-xt3 must be
330 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
331 * Portals has been removed, replaced by LNET.
332 LNET is new networking infrastructure for Lustre, it includes a
333 reorganized network configuration mode (see the user
334 documentation for full details) as well as support for routing
335 between different network fabrics. Lustre Networking Devices
336 (LNDS) for the supported network fabrics have also been created
337 for this new infrastructure.
339 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
344 Frequency : rare (large Voltaire clusters only)
346 Description: the default number of reserved transmit descriptors was too low
347 for some large clusters
348 Details : As a workaround, the number was increased. A proper fix includes
351 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
356 Frequency : occasional (large-scale events, cluster reboot, network failure)
358 Description: too many error messages on console obscure actual problem and
359 can slow down/panic server, or cause recovery to fail repeatedly
360 Details : enable rate-limiting of console error messages, and some messages
361 that were console errors now only go to the kernel log
363 Severity : enhancement
365 Description: add /proc/sys/portals/catastrophe entry which will report if
366 that node has previously LBUGged
368 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
370 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
372 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
374 - handle error return code in kranal_check_fma_rx() (5915,6054)
376 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
378 - update vibnal (Voltaire IB NAL)
379 - update gmnal (Myrinet NAL), gmnalid
381 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
383 * Landed portals:b_port_step as follows...
385 - removed CFS_DECL_SPIN*
386 just use 'spinlock_t' and initialise with spin_lock_init()
388 - removed CFS_DECL_MUTEX*
389 just use 'struct semaphore' and initialise with init_mutex()
391 - removed CFS_DECL_RWSEM*
392 just use 'struct rw_semaphore' and initialise with init_rwsem()
394 - renamed cfs_sleep_chan -> cfs_waitq
395 cfs_sleep_link -> cfs_waitlink
397 - fixed race in linux version of arch-independent socknal
398 (the ENOMEM/EAGAIN decision).
400 - Didn't fix problems in Darwin version of arch-independent socknal
401 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
403 - removed libcfs types from non-socknal header files (only some types
404 in the header files had been changed; the .c files hadn't been