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
17 Frequency : after Ptllnd timeouts and portals congestion
19 Description: Credit overflows
20 Details : This was a bug in ptllnd connection establishment. The fix
21 implements better peer stamps to disambiguate connection
22 establishment and ensure both peers enter the credit flow
23 state machine consistently.
28 Description: kptllnd didn't propagate some network errors up to LNET
29 Details : This bug was spotted while investigating 11394. The fix
30 ensures network errors on sends and bulk transfers are
31 propagated to LNET/lustre correctly.
33 Severity : enhancement
35 Description: Fixed console chatter in case of -ETIMEDOUT.
37 Severity : enhancement
39 Description: Added D_NETTRACE for recording network packet history
40 (initially only for ptllnd). Also a separate userspace
41 ptllnd facility to gather history which should really be
42 covered by D_NETTRACE too, if only CDEBUG recorded history in
48 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
49 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
50 callback can occur before a connection has actually been
51 established. This caused an assertion failure previously.
53 Severity : enhancement
55 Description: Multiple instances for o2iblnd
56 Details : Allow multiple instances of o2iblnd to enable networking over
57 multiple HCAs and routing between them.
61 Description: lnet deadlock in router_checker
62 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
63 into BH locks to eliminate potential deadlock caused by
64 ksocknal_data_ready() preempting code holding these locks.
68 Description: Millions of failed socklnd connection attempts cause a very slow FS
69 Details : added a new route flag ksnr_scheduled to distinguish from
70 ksnr_connecting, so that a peer connection request is only turned
71 down for race concerns when an active connection to the same peer
72 is under progress (instead of just being scheduled).
74 ------------------------------------------------------------------------------
76 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
78 * Support for networks:
79 socklnd - kernels up to 2.6.16
80 qswlnd - Qsnet kernel modules 5.20 and later
81 openiblnd - IbGold 1.8.2
83 viblnd - Voltaire ibhost 3.4.5 and later
84 ciblnd - Topspin 3.2.0
85 iiblnd - Infiniserv 3.3 + PathBits patch
86 gmlnd - GM 2.1.22 and later
87 mxlnd - MX 1.2.1 or later
88 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
91 Severity : major on XT3
93 Description: libcfs overwrites /proc/sys/portals
94 Details : libcfs created a symlink from /proc/sys/portals to
95 /proc/sys/lnet for backwards compatibility. This is no
96 longer required and makes the Cray portals /proc variables
101 Description: OFED FMR API change
102 Details : This changes parameter usage to reflect a change in
103 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
104 that FMR support is only used in experimental versions of the
105 o2iblnd - this change does not affect standard usage at all.
107 Severity : enhancement
109 Description: new ko2iblnd module parameter: ib_mtu
110 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
111 HCAs. You can avoid this problem by setting the MTU to 1024
112 using this module parameter.
114 Severity : enhancement
115 Bugzilla : 11118/11620
116 Description: ptllnd small request message buffer alignment fix
117 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
118 Round up small message size on sends in case this option
119 is not supported. 11620 was a defect in the initial
120 implementation which effectively asserted all peers had to be
121 running the correct protocol version which was fixed by always
122 NAK-ing such requests and handling any misalignments they
127 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
128 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
129 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
131 Severity : enhancement
133 Description: Patchless ZC(zero copy) socklnd
134 Details : New protocol for socklnd, socklnd can support zero copy without
135 kernel patch, it's compatible with old socklnd. Checksum is
136 moved from tunables to modparams.
140 Description: When ksocknal_del_peer() is called upon a peer whose
141 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
142 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
145 Frequency : when ptlrpc is under heavy use and runs out of request buffer
147 Description: In lnet_match_blocked_msg(), md can be used without holding a
151 Frequency : very rarely
153 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
154 If connd connects a route which has been closed by
155 ksocknal_shutdown(), ksocknal_create_routes() may create new
156 routes which hold references on the peer, causing shutdown
157 process to wait for peer to disappear forever.
159 Severity : enhancement
161 Description: Dump XT3 portals traces on kptllnd timeout
162 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
163 dump Cray portals debug traces to a file. The kptllnd module
164 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
165 is the basename of the dump file.
168 Frequency : infrequent
170 Description: kernel ptllnd fix bug in connection re-establishment
171 Details : Kernel ptllnd could produce protocol errors e.g. illegal
172 matchbits and/or violate the credit flow protocol when trying
173 to re-establish a connection with a peer after an error or
176 Severity : enhancement
178 Description: Allow /proc/sys/lnet/debug to be set symbolically
179 Details : Allow debug and subsystem debug values to be read/set by name
180 in addition to numerically, for ease of use.
183 Frequency : only in configurations with LNET routers
185 Description: routes automatically marked down and recovered
186 Details : In configurations with LNET routers if a router fails routers
187 now actively try to recover routes that are down, unless they
188 are marked down by an administrator.
190 ------------------------------------------------------------------------------
192 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
195 Frequency : very rarely, in configurations with LNET routers and TCP
197 Description: incorrect data written to files on OSTs
198 Details : In certain high-load conditions incorrect data may be written
199 to files on the OST when using TCP networks.
201 ------------------------------------------------------------------------------
203 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
205 - rework CDEBUG messages rate-limiting mechanism b=10375
206 - add per-socket tunables for socklnd if the kernel is patched b=10327
208 ------------------------------------------------------------------------------
210 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
212 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
213 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
214 - quiet LNET startup LNI message for liblustre b=10128
215 - Better console error messages if 'ip2nets' can't match an IP address
216 - Fixed overflow/use-before-set bugs in linux-time.h
217 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
218 - LNET teardown failed an assertion about the route table being empty
219 - Fixed a crash in LNetEQPoll(<invalid handle>)
220 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
221 - improve debug message for liblustre/Catamount nodes (b=10116)
223 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
224 * Configuration change for the XT3
225 The PTLLND is now used to run Lustre over Portals on the XT3.
226 The configure option(s) --with-cray-portals are no longer
227 used. Rather --with-portals=<path-to-portals-includes> is
228 used to enable building on the XT3. In addition to enable
229 XT3 specific features the option --enable-cray-xt3 must be
232 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
233 * Portals has been removed, replaced by LNET.
234 LNET is new networking infrastructure for Lustre, it includes a
235 reorganized network configuration mode (see the user
236 documentation for full details) as well as support for routing
237 between different network fabrics. Lustre Networking Devices
238 (LNDS) for the supported network fabrics have also been created
239 for this new infrastructure.
241 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
246 Frequency : rare (large Voltaire clusters only)
248 Description: the default number of reserved transmit descriptors was too low
249 for some large clusters
250 Details : As a workaround, the number was increased. A proper fix includes
253 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
258 Frequency : occasional (large-scale events, cluster reboot, network failure)
260 Description: too many error messages on console obscure actual problem and
261 can slow down/panic server, or cause recovery to fail repeatedly
262 Details : enable rate-limiting of console error messages, and some messages
263 that were console errors now only go to the kernel log
265 Severity : enhancement
267 Description: add /proc/sys/portals/catastrophe entry which will report if
268 that node has previously LBUGged
270 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
272 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
274 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
276 - handle error return code in kranal_check_fma_rx() (5915,6054)
278 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
280 - update vibnal (Voltaire IB NAL)
281 - update gmnal (Myrinet NAL), gmnalid
283 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
285 * Landed portals:b_port_step as follows...
287 - removed CFS_DECL_SPIN*
288 just use 'spinlock_t' and initialise with spin_lock_init()
290 - removed CFS_DECL_MUTEX*
291 just use 'struct semaphore' and initialise with init_mutex()
293 - removed CFS_DECL_RWSEM*
294 just use 'struct rw_semaphore' and initialise with init_rwsem()
296 - renamed cfs_sleep_chan -> cfs_waitq
297 cfs_sleep_link -> cfs_waitlink
299 - fixed race in linux version of arch-independent socknal
300 (the ENOMEM/EAGAIN decision).
302 - Didn't fix problems in Darwin version of arch-independent socknal
303 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
305 - removed libcfs types from non-socknal header files (only some types
306 in the header files had been changed; the .c files hadn't been