1 TBD Cluster File Systems, Inc. <info@clusterfs.com>
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 Severity : enhancement
18 Description: Fixed console chatter in case of -ETIMEDOUT.
20 Severity : enhancement
22 Description: Added D_NETTRACE for recording network packet history
23 (initially only for ptllnd). Also a separate userspace
24 ptllnd facility to gather history which should really be
25 covered by D_NETTRACE too, if only CDEBUG recorded history in
31 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
32 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
33 callback can occur before a connection has actually been
34 established. This caused an assertion failure previously.
36 Severity : enhancement
38 Description: Multiple instances for o2iblnd
39 Details : Allow multiple instances of o2iblnd to enable networking over
40 multiple HCAs and routing between them.
44 Description: lnet deadlock in router_checker
45 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
46 into BH locks to eliminate potential deadlock caused by
47 ksocknal_data_ready() preempting code holding these locks.
51 Description: Millions of failed socklnd connection attempts cause a very slow FS
52 Details : added a new route flag ksnr_scheduled to distinguish from
53 ksnr_connecting, so that a peer connection request is only turned
54 down for race concerns when an active connection to the same peer
55 is under progress (instead of just being scheduled).
57 ------------------------------------------------------------------------------
59 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
61 * Support for networks:
62 socklnd - kernels up to 2.6.16
63 qswlnd - Qsnet kernel modules 5.20 and later
64 openiblnd - IbGold 1.8.2
66 viblnd - Voltaire ibhost 3.4.5 and later
67 ciblnd - Topspin 3.2.0
68 iiblnd - Infiniserv 3.3 + PathBits patch
69 gmlnd - GM 2.1.22 and later
70 mxlnd - MX 1.2.1 or later
71 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
74 Severity : major on XT3
76 Description: libcfs overwrites /proc/sys/portals
77 Details : libcfs created a symlink from /proc/sys/portals to
78 /proc/sys/lnet for backwards compatibility. This is no
79 longer required and makes the Cray portals /proc variables
84 Description: OFED FMR API change
85 Details : This changes parameter usage to reflect a change in
86 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
87 that FMR support is only used in experimental versions of the
88 o2iblnd - this change does not affect standard usage at all.
90 Severity : enhancement
92 Description: new ko2iblnd module parameter: ib_mtu
93 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
94 HCAs. You can avoid this problem by setting the MTU to 1024
95 using this module parameter.
97 Severity : enhancement
98 Bugzilla : 11118/11620
99 Description: ptllnd small request message buffer alignment fix
100 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
101 Round up small message size on sends in case this option
102 is not supported. 11620 was a defect in the initial
103 implementation which effectively asserted all peers had to be
104 running the correct protocol version which was fixed by always
105 NAK-ing such requests and handling any misalignments they
110 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
111 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
112 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
114 Severity : enhancement
116 Description: Patchless ZC(zero copy) socklnd
117 Details : New protocol for socklnd, socklnd can support zero copy without
118 kernel patch, it's compatible with old socklnd. Checksum is
119 moved from tunables to modparams.
123 Description: When ksocknal_del_peer() is called upon a peer whose
124 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
125 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
128 Frequency : when ptlrpc is under heavy use and runs out of request buffer
130 Description: In lnet_match_blocked_msg(), md can be used without holding a
134 Frequency : very rarely
136 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
137 If connd connects a route which has been closed by
138 ksocknal_shutdown(), ksocknal_create_routes() may create new
139 routes which hold references on the peer, causing shutdown
140 process to wait for peer to disappear forever.
142 Severity : enhancement
144 Description: Dump XT3 portals traces on kptllnd timeout
145 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
146 dump Cray portals debug traces to a file. The kptllnd module
147 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
148 is the basename of the dump file.
151 Frequency : infrequent
153 Description: kernel ptllnd fix bug in connection re-establishment
154 Details : Kernel ptllnd could produce protocol errors e.g. illegal
155 matchbits and/or violate the credit flow protocol when trying
156 to re-establish a connection with a peer after an error or
159 Severity : enhancement
161 Description: Allow /proc/sys/lnet/debug to be set symbolically
162 Details : Allow debug and subsystem debug values to be read/set by name
163 in addition to numerically, for ease of use.
166 Frequency : only in configurations with LNET routers
168 Description: routes automatically marked down and recovered
169 Details : In configurations with LNET routers if a router fails routers
170 now actively try to recover routes that are down, unless they
171 are marked down by an administrator.
173 ------------------------------------------------------------------------------
175 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
178 Frequency : very rarely, in configurations with LNET routers and TCP
180 Description: incorrect data written to files on OSTs
181 Details : In certain high-load conditions incorrect data may be written
182 to files on the OST when using TCP networks.
184 ------------------------------------------------------------------------------
186 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
188 - rework CDEBUG messages rate-limiting mechanism b=10375
189 - add per-socket tunables for socklnd if the kernel is patched b=10327
191 ------------------------------------------------------------------------------
193 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
195 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
196 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
197 - quiet LNET startup LNI message for liblustre b=10128
198 - Better console error messages if 'ip2nets' can't match an IP address
199 - Fixed overflow/use-before-set bugs in linux-time.h
200 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
201 - LNET teardown failed an assertion about the route table being empty
202 - Fixed a crash in LNetEQPoll(<invalid handle>)
203 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
204 - improve debug message for liblustre/Catamount nodes (b=10116)
206 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
207 * Configuration change for the XT3
208 The PTLLND is now used to run Lustre over Portals on the XT3.
209 The configure option(s) --with-cray-portals are no longer
210 used. Rather --with-portals=<path-to-portals-includes> is
211 used to enable building on the XT3. In addition to enable
212 XT3 specific features the option --enable-cray-xt3 must be
215 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
216 * Portals has been removed, replaced by LNET.
217 LNET is new networking infrastructure for Lustre, it includes a
218 reorganized network configuration mode (see the user
219 documentation for full details) as well as support for routing
220 between different network fabrics. Lustre Networking Devices
221 (LNDS) for the supported network fabrics have also been created
222 for this new infrastructure.
224 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
229 Frequency : rare (large Voltaire clusters only)
231 Description: the default number of reserved transmit descriptors was too low
232 for some large clusters
233 Details : As a workaround, the number was increased. A proper fix includes
236 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
241 Frequency : occasional (large-scale events, cluster reboot, network failure)
243 Description: too many error messages on console obscure actual problem and
244 can slow down/panic server, or cause recovery to fail repeatedly
245 Details : enable rate-limiting of console error messages, and some messages
246 that were console errors now only go to the kernel log
248 Severity : enhancement
250 Description: add /proc/sys/portals/catastrophe entry which will report if
251 that node has previously LBUGged
253 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
255 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
257 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
259 - handle error return code in kranal_check_fma_rx() (5915,6054)
261 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
263 - update vibnal (Voltaire IB NAL)
264 - update gmnal (Myrinet NAL), gmnalid
266 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
268 * Landed portals:b_port_step as follows...
270 - removed CFS_DECL_SPIN*
271 just use 'spinlock_t' and initialise with spin_lock_init()
273 - removed CFS_DECL_MUTEX*
274 just use 'struct semaphore' and initialise with init_mutex()
276 - removed CFS_DECL_RWSEM*
277 just use 'struct rw_semaphore' and initialise with init_rwsem()
279 - renamed cfs_sleep_chan -> cfs_waitq
280 cfs_sleep_link -> cfs_waitlink
282 - fixed race in linux version of arch-independent socknal
283 (the ENOMEM/EAGAIN decision).
285 - Didn't fix problems in Darwin version of arch-independent socknal
286 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
288 - removed libcfs types from non-socknal header files (only some types
289 in the header files had been changed; the .c files hadn't been