1 TBD 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 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
17 * version 1.4.10 / 1.6.0
18 * Support for networks:
19 socklnd - kernels up to 2.6.16
20 qswlnd - Qsnet kernel modules 5.20 and later
21 openiblnd - IbGold 1.8.2
23 viblnd - Voltaire ibhost 3.4.5 and later
24 ciblnd - Topspin 3.2.0
25 iiblnd - Infiniserv 3.3 + PathBits patch
26 gmlnd - GM 2.1.22 and later
27 mxlnd - MX 1.2.1 or later
28 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
32 Frequency : after Ptllnd timeouts and portals congestion
34 Description: Credit overflows
35 Details : This was a bug in ptllnd connection establishment. The fix
36 implements better peer stamps to disambiguate connection
37 establishment and ensure both peers enter the credit flow
38 state machine consistently.
43 Description: kptllnd didn't propagate some network errors up to LNET
44 Details : This bug was spotted while investigating 11394. The fix
45 ensures network errors on sends and bulk transfers are
46 propagated to LNET/lustre correctly.
48 Severity : enhancement
50 Description: Fixed console chatter in case of -ETIMEDOUT.
52 Severity : enhancement
54 Description: Added D_NETTRACE for recording network packet history
55 (initially only for ptllnd). Also a separate userspace
56 ptllnd facility to gather history which should really be
57 covered by D_NETTRACE too, if only CDEBUG recorded history in
63 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
64 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
65 callback can occur before a connection has actually been
66 established. This caused an assertion failure previously.
68 Severity : enhancement
70 Description: Multiple instances for o2iblnd
71 Details : Allow multiple instances of o2iblnd to enable networking over
72 multiple HCAs and routing between them.
76 Description: lnet deadlock in router_checker
77 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
78 into BH locks to eliminate potential deadlock caused by
79 ksocknal_data_ready() preempting code holding these locks.
83 Description: Millions of failed socklnd connection attempts cause a very slow FS
84 Details : added a new route flag ksnr_scheduled to distinguish from
85 ksnr_connecting, so that a peer connection request is only turned
86 down for race concerns when an active connection to the same peer
87 is under progress (instead of just being scheduled).
89 ------------------------------------------------------------------------------
91 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
93 * Support for networks:
94 socklnd - kernels up to 2.6.16
95 qswlnd - Qsnet kernel modules 5.20 and later
96 openiblnd - IbGold 1.8.2
98 viblnd - Voltaire ibhost 3.4.5 and later
99 ciblnd - Topspin 3.2.0
100 iiblnd - Infiniserv 3.3 + PathBits patch
101 gmlnd - GM 2.1.22 and later
102 mxlnd - MX 1.2.1 or later
103 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
106 Severity : major on XT3
108 Description: libcfs overwrites /proc/sys/portals
109 Details : libcfs created a symlink from /proc/sys/portals to
110 /proc/sys/lnet for backwards compatibility. This is no
111 longer required and makes the Cray portals /proc variables
116 Description: OFED FMR API change
117 Details : This changes parameter usage to reflect a change in
118 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
119 that FMR support is only used in experimental versions of the
120 o2iblnd - this change does not affect standard usage at all.
122 Severity : enhancement
124 Description: new ko2iblnd module parameter: ib_mtu
125 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
126 HCAs. You can avoid this problem by setting the MTU to 1024
127 using this module parameter.
129 Severity : enhancement
130 Bugzilla : 11118/11620
131 Description: ptllnd small request message buffer alignment fix
132 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
133 Round up small message size on sends in case this option
134 is not supported. 11620 was a defect in the initial
135 implementation which effectively asserted all peers had to be
136 running the correct protocol version which was fixed by always
137 NAK-ing such requests and handling any misalignments they
142 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
143 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
144 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
146 Severity : enhancement
148 Description: Patchless ZC(zero copy) socklnd
149 Details : New protocol for socklnd, socklnd can support zero copy without
150 kernel patch, it's compatible with old socklnd. Checksum is
151 moved from tunables to modparams.
155 Description: When ksocknal_del_peer() is called upon a peer whose
156 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
157 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
160 Frequency : when ptlrpc is under heavy use and runs out of request buffer
162 Description: In lnet_match_blocked_msg(), md can be used without holding a
166 Frequency : very rarely
168 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
169 If connd connects a route which has been closed by
170 ksocknal_shutdown(), ksocknal_create_routes() may create new
171 routes which hold references on the peer, causing shutdown
172 process to wait for peer to disappear forever.
174 Severity : enhancement
176 Description: Dump XT3 portals traces on kptllnd timeout
177 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
178 dump Cray portals debug traces to a file. The kptllnd module
179 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
180 is the basename of the dump file.
183 Frequency : infrequent
185 Description: kernel ptllnd fix bug in connection re-establishment
186 Details : Kernel ptllnd could produce protocol errors e.g. illegal
187 matchbits and/or violate the credit flow protocol when trying
188 to re-establish a connection with a peer after an error or
191 Severity : enhancement
193 Description: Allow /proc/sys/lnet/debug to be set symbolically
194 Details : Allow debug and subsystem debug values to be read/set by name
195 in addition to numerically, for ease of use.
198 Frequency : only in configurations with LNET routers
200 Description: routes automatically marked down and recovered
201 Details : In configurations with LNET routers if a router fails routers
202 now actively try to recover routes that are down, unless they
203 are marked down by an administrator.
205 ------------------------------------------------------------------------------
207 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
210 Frequency : very rarely, in configurations with LNET routers and TCP
212 Description: incorrect data written to files on OSTs
213 Details : In certain high-load conditions incorrect data may be written
214 to files on the OST when using TCP networks.
216 ------------------------------------------------------------------------------
218 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
220 - rework CDEBUG messages rate-limiting mechanism b=10375
221 - add per-socket tunables for socklnd if the kernel is patched b=10327
223 ------------------------------------------------------------------------------
225 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
227 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
228 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
229 - quiet LNET startup LNI message for liblustre b=10128
230 - Better console error messages if 'ip2nets' can't match an IP address
231 - Fixed overflow/use-before-set bugs in linux-time.h
232 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
233 - LNET teardown failed an assertion about the route table being empty
234 - Fixed a crash in LNetEQPoll(<invalid handle>)
235 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
236 - improve debug message for liblustre/Catamount nodes (b=10116)
238 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
239 * Configuration change for the XT3
240 The PTLLND is now used to run Lustre over Portals on the XT3.
241 The configure option(s) --with-cray-portals are no longer
242 used. Rather --with-portals=<path-to-portals-includes> is
243 used to enable building on the XT3. In addition to enable
244 XT3 specific features the option --enable-cray-xt3 must be
247 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
248 * Portals has been removed, replaced by LNET.
249 LNET is new networking infrastructure for Lustre, it includes a
250 reorganized network configuration mode (see the user
251 documentation for full details) as well as support for routing
252 between different network fabrics. Lustre Networking Devices
253 (LNDS) for the supported network fabrics have also been created
254 for this new infrastructure.
256 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
261 Frequency : rare (large Voltaire clusters only)
263 Description: the default number of reserved transmit descriptors was too low
264 for some large clusters
265 Details : As a workaround, the number was increased. A proper fix includes
268 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
273 Frequency : occasional (large-scale events, cluster reboot, network failure)
275 Description: too many error messages on console obscure actual problem and
276 can slow down/panic server, or cause recovery to fail repeatedly
277 Details : enable rate-limiting of console error messages, and some messages
278 that were console errors now only go to the kernel log
280 Severity : enhancement
282 Description: add /proc/sys/portals/catastrophe entry which will report if
283 that node has previously LBUGged
285 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
287 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
289 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
291 - handle error return code in kranal_check_fma_rx() (5915,6054)
293 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
295 - update vibnal (Voltaire IB NAL)
296 - update gmnal (Myrinet NAL), gmnalid
298 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
300 * Landed portals:b_port_step as follows...
302 - removed CFS_DECL_SPIN*
303 just use 'spinlock_t' and initialise with spin_lock_init()
305 - removed CFS_DECL_MUTEX*
306 just use 'struct semaphore' and initialise with init_mutex()
308 - removed CFS_DECL_RWSEM*
309 just use 'struct rw_semaphore' and initialise with init_rwsem()
311 - renamed cfs_sleep_chan -> cfs_waitq
312 cfs_sleep_link -> cfs_waitlink
314 - fixed race in linux version of arch-independent socknal
315 (the ENOMEM/EAGAIN decision).
317 - Didn't fix problems in Darwin version of arch-independent socknal
318 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
320 - removed libcfs types from non-socknal header files (only some types
321 in the header files had been changed; the .c files hadn't been