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
36 Description: Assertion failure in kernel ptllnd caused by posting passive
37 bulk buffers before connection establishment complete.
42 Description: A race in kernel ptllnd between deleting a peer and posting
43 new communications for it could hang communications -
44 manifesting as "Unexpectedly long timeout" messages.
49 Description: Kernel ptllnd lock ordering issue could hang a node.
54 Description: node crash on socket teardown race
57 Frequency : 'lctl peer_list' issued on a mx net
59 Description: Enable lctl's peer_list for MXLND
62 Frequency : after Ptllnd timeouts and portals congestion
64 Description: Credit overflows
65 Details : This was a bug in ptllnd connection establishment. The fix
66 implements better peer stamps to disambiguate connection
67 establishment and ensure both peers enter the credit flow
68 state machine consistently.
73 Description: kptllnd didn't propagate some network errors up to LNET
74 Details : This bug was spotted while investigating 11394. The fix
75 ensures network errors on sends and bulk transfers are
76 propagated to LNET/lustre correctly.
78 Severity : enhancement
80 Description: Fixed console chatter in case of -ETIMEDOUT.
82 Severity : enhancement
84 Description: Added D_NETTRACE for recording network packet history
85 (initially only for ptllnd). Also a separate userspace
86 ptllnd facility to gather history which should really be
87 covered by D_NETTRACE too, if only CDEBUG recorded history in
93 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
94 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
95 callback can occur before a connection has actually been
96 established. This caused an assertion failure previously.
98 Severity : enhancement
100 Description: Multiple instances for o2iblnd
101 Details : Allow multiple instances of o2iblnd to enable networking over
102 multiple HCAs and routing between them.
106 Description: lnet deadlock in router_checker
107 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
108 into BH locks to eliminate potential deadlock caused by
109 ksocknal_data_ready() preempting code holding these locks.
113 Description: Millions of failed socklnd connection attempts cause a very slow FS
114 Details : added a new route flag ksnr_scheduled to distinguish from
115 ksnr_connecting, so that a peer connection request is only turned
116 down for race concerns when an active connection to the same peer
117 is under progress (instead of just being scheduled).
119 ------------------------------------------------------------------------------
121 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
123 * Support for networks:
124 socklnd - kernels up to 2.6.16
125 qswlnd - Qsnet kernel modules 5.20 and later
126 openiblnd - IbGold 1.8.2
128 viblnd - Voltaire ibhost 3.4.5 and later
129 ciblnd - Topspin 3.2.0
130 iiblnd - Infiniserv 3.3 + PathBits patch
131 gmlnd - GM 2.1.22 and later
132 mxlnd - MX 1.2.1 or later
133 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
136 Severity : major on XT3
138 Description: libcfs overwrites /proc/sys/portals
139 Details : libcfs created a symlink from /proc/sys/portals to
140 /proc/sys/lnet for backwards compatibility. This is no
141 longer required and makes the Cray portals /proc variables
146 Description: OFED FMR API change
147 Details : This changes parameter usage to reflect a change in
148 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
149 that FMR support is only used in experimental versions of the
150 o2iblnd - this change does not affect standard usage at all.
152 Severity : enhancement
154 Description: new ko2iblnd module parameter: ib_mtu
155 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
156 HCAs. You can avoid this problem by setting the MTU to 1024
157 using this module parameter.
159 Severity : enhancement
160 Bugzilla : 11118/11620
161 Description: ptllnd small request message buffer alignment fix
162 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
163 Round up small message size on sends in case this option
164 is not supported. 11620 was a defect in the initial
165 implementation which effectively asserted all peers had to be
166 running the correct protocol version which was fixed by always
167 NAK-ing such requests and handling any misalignments they
172 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
173 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
174 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
176 Severity : enhancement
178 Description: Patchless ZC(zero copy) socklnd
179 Details : New protocol for socklnd, socklnd can support zero copy without
180 kernel patch, it's compatible with old socklnd. Checksum is
181 moved from tunables to modparams.
185 Description: When ksocknal_del_peer() is called upon a peer whose
186 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
187 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
190 Frequency : when ptlrpc is under heavy use and runs out of request buffer
192 Description: In lnet_match_blocked_msg(), md can be used without holding a
196 Frequency : very rarely
198 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
199 If connd connects a route which has been closed by
200 ksocknal_shutdown(), ksocknal_create_routes() may create new
201 routes which hold references on the peer, causing shutdown
202 process to wait for peer to disappear forever.
204 Severity : enhancement
206 Description: Dump XT3 portals traces on kptllnd timeout
207 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
208 dump Cray portals debug traces to a file. The kptllnd module
209 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
210 is the basename of the dump file.
213 Frequency : infrequent
215 Description: kernel ptllnd fix bug in connection re-establishment
216 Details : Kernel ptllnd could produce protocol errors e.g. illegal
217 matchbits and/or violate the credit flow protocol when trying
218 to re-establish a connection with a peer after an error or
221 Severity : enhancement
223 Description: Allow /proc/sys/lnet/debug to be set symbolically
224 Details : Allow debug and subsystem debug values to be read/set by name
225 in addition to numerically, for ease of use.
228 Frequency : only in configurations with LNET routers
230 Description: routes automatically marked down and recovered
231 Details : In configurations with LNET routers if a router fails routers
232 now actively try to recover routes that are down, unless they
233 are marked down by an administrator.
235 ------------------------------------------------------------------------------
237 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
240 Frequency : very rarely, in configurations with LNET routers and TCP
242 Description: incorrect data written to files on OSTs
243 Details : In certain high-load conditions incorrect data may be written
244 to files on the OST when using TCP networks.
246 ------------------------------------------------------------------------------
248 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
250 - rework CDEBUG messages rate-limiting mechanism b=10375
251 - add per-socket tunables for socklnd if the kernel is patched b=10327
253 ------------------------------------------------------------------------------
255 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
257 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
258 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
259 - quiet LNET startup LNI message for liblustre b=10128
260 - Better console error messages if 'ip2nets' can't match an IP address
261 - Fixed overflow/use-before-set bugs in linux-time.h
262 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
263 - LNET teardown failed an assertion about the route table being empty
264 - Fixed a crash in LNetEQPoll(<invalid handle>)
265 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
266 - improve debug message for liblustre/Catamount nodes (b=10116)
268 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
269 * Configuration change for the XT3
270 The PTLLND is now used to run Lustre over Portals on the XT3.
271 The configure option(s) --with-cray-portals are no longer
272 used. Rather --with-portals=<path-to-portals-includes> is
273 used to enable building on the XT3. In addition to enable
274 XT3 specific features the option --enable-cray-xt3 must be
277 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
278 * Portals has been removed, replaced by LNET.
279 LNET is new networking infrastructure for Lustre, it includes a
280 reorganized network configuration mode (see the user
281 documentation for full details) as well as support for routing
282 between different network fabrics. Lustre Networking Devices
283 (LNDS) for the supported network fabrics have also been created
284 for this new infrastructure.
286 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
291 Frequency : rare (large Voltaire clusters only)
293 Description: the default number of reserved transmit descriptors was too low
294 for some large clusters
295 Details : As a workaround, the number was increased. A proper fix includes
298 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
303 Frequency : occasional (large-scale events, cluster reboot, network failure)
305 Description: too many error messages on console obscure actual problem and
306 can slow down/panic server, or cause recovery to fail repeatedly
307 Details : enable rate-limiting of console error messages, and some messages
308 that were console errors now only go to the kernel log
310 Severity : enhancement
312 Description: add /proc/sys/portals/catastrophe entry which will report if
313 that node has previously LBUGged
315 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
317 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
319 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
321 - handle error return code in kranal_check_fma_rx() (5915,6054)
323 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
325 - update vibnal (Voltaire IB NAL)
326 - update gmnal (Myrinet NAL), gmnalid
328 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
330 * Landed portals:b_port_step as follows...
332 - removed CFS_DECL_SPIN*
333 just use 'spinlock_t' and initialise with spin_lock_init()
335 - removed CFS_DECL_MUTEX*
336 just use 'struct semaphore' and initialise with init_mutex()
338 - removed CFS_DECL_RWSEM*
339 just use 'struct rw_semaphore' and initialise with init_rwsem()
341 - renamed cfs_sleep_chan -> cfs_waitq
342 cfs_sleep_link -> cfs_waitlink
344 - fixed race in linux version of arch-independent socknal
345 (the ENOMEM/EAGAIN decision).
347 - Didn't fix problems in Darwin version of arch-independent socknal
348 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
350 - removed libcfs types from non-socknal header files (only some types
351 in the header files had been changed; the .c files hadn't been