1 2007-04-23 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
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
16 ------------------------------------------------------------------------------
18 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
19 * version 1.4.10 / 1.6.0
20 * Support for networks:
21 socklnd - kernels up to 2.6.16
22 qswlnd - Qsnet kernel modules 5.20 and later
23 openiblnd - IbGold 1.8.2
25 viblnd - Voltaire ibhost 3.4.5 and later
26 ciblnd - Topspin 3.2.0
27 iiblnd - Infiniserv 3.3 + PathBits patch
28 gmlnd - GM 2.1.22 and later
29 mxlnd - MX 1.2.1 or later
30 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
35 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
36 possibly accessed in kptllnd_shutdown. Ptllnd should init
37 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
41 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
45 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
49 Description: the_lnet.ln_finalizing was not set when the current thread is
50 about to complete messages. It only affects multi-threaded
56 Description: Changed the default kqswlnd ntxmsg=512
61 Description: Assertion failure in kernel ptllnd caused by posting passive
62 bulk buffers before connection establishment complete.
67 Description: A race in kernel ptllnd between deleting a peer and posting
68 new communications for it could hang communications -
69 manifesting as "Unexpectedly long timeout" messages.
74 Description: Kernel ptllnd lock ordering issue could hang a node.
79 Description: node crash on socket teardown race
82 Frequency : 'lctl peer_list' issued on a mx net
84 Description: Enable lctl's peer_list for MXLND
87 Frequency : after Ptllnd timeouts and portals congestion
89 Description: Credit overflows
90 Details : This was a bug in ptllnd connection establishment. The fix
91 implements better peer stamps to disambiguate connection
92 establishment and ensure both peers enter the credit flow
93 state machine consistently.
98 Description: kptllnd didn't propagate some network errors up to LNET
99 Details : This bug was spotted while investigating 11394. The fix
100 ensures network errors on sends and bulk transfers are
101 propagated to LNET/lustre correctly.
103 Severity : enhancement
105 Description: Fixed console chatter in case of -ETIMEDOUT.
107 Severity : enhancement
109 Description: Added D_NETTRACE for recording network packet history
110 (initially only for ptllnd). Also a separate userspace
111 ptllnd facility to gather history which should really be
112 covered by D_NETTRACE too, if only CDEBUG recorded history in
118 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
119 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
120 callback can occur before a connection has actually been
121 established. This caused an assertion failure previously.
123 Severity : enhancement
125 Description: Multiple instances for o2iblnd
126 Details : Allow multiple instances of o2iblnd to enable networking over
127 multiple HCAs and routing between them.
131 Description: lnet deadlock in router_checker
132 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
133 into BH locks to eliminate potential deadlock caused by
134 ksocknal_data_ready() preempting code holding these locks.
138 Description: Millions of failed socklnd connection attempts cause a very slow FS
139 Details : added a new route flag ksnr_scheduled to distinguish from
140 ksnr_connecting, so that a peer connection request is only turned
141 down for race concerns when an active connection to the same peer
142 is under progress (instead of just being scheduled).
144 ------------------------------------------------------------------------------
146 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
148 * Support for networks:
149 socklnd - kernels up to 2.6.16
150 qswlnd - Qsnet kernel modules 5.20 and later
151 openiblnd - IbGold 1.8.2
153 viblnd - Voltaire ibhost 3.4.5 and later
154 ciblnd - Topspin 3.2.0
155 iiblnd - Infiniserv 3.3 + PathBits patch
156 gmlnd - GM 2.1.22 and later
157 mxlnd - MX 1.2.1 or later
158 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
161 Severity : major on XT3
163 Description: libcfs overwrites /proc/sys/portals
164 Details : libcfs created a symlink from /proc/sys/portals to
165 /proc/sys/lnet for backwards compatibility. This is no
166 longer required and makes the Cray portals /proc variables
171 Description: OFED FMR API change
172 Details : This changes parameter usage to reflect a change in
173 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
174 that FMR support is only used in experimental versions of the
175 o2iblnd - this change does not affect standard usage at all.
177 Severity : enhancement
179 Description: new ko2iblnd module parameter: ib_mtu
180 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
181 HCAs. You can avoid this problem by setting the MTU to 1024
182 using this module parameter.
184 Severity : enhancement
185 Bugzilla : 11118/11620
186 Description: ptllnd small request message buffer alignment fix
187 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
188 Round up small message size on sends in case this option
189 is not supported. 11620 was a defect in the initial
190 implementation which effectively asserted all peers had to be
191 running the correct protocol version which was fixed by always
192 NAK-ing such requests and handling any misalignments they
197 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
198 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
199 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
201 Severity : enhancement
203 Description: Patchless ZC(zero copy) socklnd
204 Details : New protocol for socklnd, socklnd can support zero copy without
205 kernel patch, it's compatible with old socklnd. Checksum is
206 moved from tunables to modparams.
210 Description: When ksocknal_del_peer() is called upon a peer whose
211 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
212 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
215 Frequency : when ptlrpc is under heavy use and runs out of request buffer
217 Description: In lnet_match_blocked_msg(), md can be used without holding a
221 Frequency : very rarely
223 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
224 If connd connects a route which has been closed by
225 ksocknal_shutdown(), ksocknal_create_routes() may create new
226 routes which hold references on the peer, causing shutdown
227 process to wait for peer to disappear forever.
229 Severity : enhancement
231 Description: Dump XT3 portals traces on kptllnd timeout
232 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
233 dump Cray portals debug traces to a file. The kptllnd module
234 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
235 is the basename of the dump file.
238 Frequency : infrequent
240 Description: kernel ptllnd fix bug in connection re-establishment
241 Details : Kernel ptllnd could produce protocol errors e.g. illegal
242 matchbits and/or violate the credit flow protocol when trying
243 to re-establish a connection with a peer after an error or
246 Severity : enhancement
248 Description: Allow /proc/sys/lnet/debug to be set symbolically
249 Details : Allow debug and subsystem debug values to be read/set by name
250 in addition to numerically, for ease of use.
253 Frequency : only in configurations with LNET routers
255 Description: routes automatically marked down and recovered
256 Details : In configurations with LNET routers if a router fails routers
257 now actively try to recover routes that are down, unless they
258 are marked down by an administrator.
260 ------------------------------------------------------------------------------
262 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
265 Frequency : very rarely, in configurations with LNET routers and TCP
267 Description: incorrect data written to files on OSTs
268 Details : In certain high-load conditions incorrect data may be written
269 to files on the OST when using TCP networks.
271 ------------------------------------------------------------------------------
273 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
275 - rework CDEBUG messages rate-limiting mechanism b=10375
276 - add per-socket tunables for socklnd if the kernel is patched b=10327
278 ------------------------------------------------------------------------------
280 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
282 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
283 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
284 - quiet LNET startup LNI message for liblustre b=10128
285 - Better console error messages if 'ip2nets' can't match an IP address
286 - Fixed overflow/use-before-set bugs in linux-time.h
287 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
288 - LNET teardown failed an assertion about the route table being empty
289 - Fixed a crash in LNetEQPoll(<invalid handle>)
290 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
291 - improve debug message for liblustre/Catamount nodes (b=10116)
293 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
294 * Configuration change for the XT3
295 The PTLLND is now used to run Lustre over Portals on the XT3.
296 The configure option(s) --with-cray-portals are no longer
297 used. Rather --with-portals=<path-to-portals-includes> is
298 used to enable building on the XT3. In addition to enable
299 XT3 specific features the option --enable-cray-xt3 must be
302 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
303 * Portals has been removed, replaced by LNET.
304 LNET is new networking infrastructure for Lustre, it includes a
305 reorganized network configuration mode (see the user
306 documentation for full details) as well as support for routing
307 between different network fabrics. Lustre Networking Devices
308 (LNDS) for the supported network fabrics have also been created
309 for this new infrastructure.
311 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
316 Frequency : rare (large Voltaire clusters only)
318 Description: the default number of reserved transmit descriptors was too low
319 for some large clusters
320 Details : As a workaround, the number was increased. A proper fix includes
323 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
328 Frequency : occasional (large-scale events, cluster reboot, network failure)
330 Description: too many error messages on console obscure actual problem and
331 can slow down/panic server, or cause recovery to fail repeatedly
332 Details : enable rate-limiting of console error messages, and some messages
333 that were console errors now only go to the kernel log
335 Severity : enhancement
337 Description: add /proc/sys/portals/catastrophe entry which will report if
338 that node has previously LBUGged
340 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
342 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
344 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
346 - handle error return code in kranal_check_fma_rx() (5915,6054)
348 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
350 - update vibnal (Voltaire IB NAL)
351 - update gmnal (Myrinet NAL), gmnalid
353 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
355 * Landed portals:b_port_step as follows...
357 - removed CFS_DECL_SPIN*
358 just use 'spinlock_t' and initialise with spin_lock_init()
360 - removed CFS_DECL_MUTEX*
361 just use 'struct semaphore' and initialise with init_mutex()
363 - removed CFS_DECL_RWSEM*
364 just use 'struct rw_semaphore' and initialise with init_rwsem()
366 - renamed cfs_sleep_chan -> cfs_waitq
367 cfs_sleep_link -> cfs_waitlink
369 - fixed race in linux version of arch-independent socknal
370 (the ENOMEM/EAGAIN decision).
372 - Didn't fix problems in Darwin version of arch-independent socknal
373 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
375 - removed libcfs types from non-socknal header files (only some types
376 in the header files had been changed; the .c files hadn't been