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: Added D_NETTRACE for recording network packet history
19 (initially only for ptllnd). Also a separate userspace
20 ptllnd facility to gather history which should really be
21 covered by D_NETTRACE too, if only CDEBUG recorded history in
27 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
28 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
29 callback can occur before a connection has actually been
30 established. This caused an assertion failure previously.
32 Severity : enhancement
34 Description: Multiple instances for o2iblnd
35 Details : Allow multiple instances of o2iblnd to enable networking over
36 multiple HCAs and routing between them.
40 Description: lnet deadlock in router_checker
41 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
42 into BH locks to eliminate potential deadlock caused by
43 ksocknal_data_ready() preempting code holding these locks.
47 Description: Millions of failed socklnd connection attempts cause a very slow FS
48 Details : added a new route flag ksnr_scheduled to distinguish from
49 ksnr_connecting, so that a peer connection request is only turned
50 down for race concerns when an active connection to the same peer
51 is under progress (instead of just being scheduled).
53 ------------------------------------------------------------------------------
55 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
57 * Support for networks:
58 socklnd - kernels up to 2.6.16
59 qswlnd - Qsnet kernel modules 5.20 and later
60 openiblnd - IbGold 1.8.2
62 viblnd - Voltaire ibhost 3.4.5 and later
63 ciblnd - Topspin 3.2.0
64 iiblnd - Infiniserv 3.3 + PathBits patch
65 gmlnd - GM 2.1.22 and later
66 mxlnd - MX 1.2.1 or later
67 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
70 Severity : major on XT3
72 Description: libcfs overwrites /proc/sys/portals
73 Details : libcfs created a symlink from /proc/sys/portals to
74 /proc/sys/lnet for backwards compatibility. This is no
75 longer required and makes the Cray portals /proc variables
80 Description: OFED FMR API change
81 Details : This changes parameter usage to reflect a change in
82 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
83 that FMR support is only used in experimental versions of the
84 o2iblnd - this change does not affect standard usage at all.
86 Severity : enhancement
88 Description: new ko2iblnd module parameter: ib_mtu
89 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
90 HCAs. You can avoid this problem by setting the MTU to 1024
91 using this module parameter.
93 Severity : enhancement
94 Bugzilla : 11118/11620
95 Description: ptllnd small request message buffer alignment fix
96 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
97 Round up small message size on sends in case this option
98 is not supported. 11620 was a defect in the initial
99 implementation which effectively asserted all peers had to be
100 running the correct protocol version which was fixed by always
101 NAK-ing such requests and handling any misalignments they
106 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
107 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
108 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
110 Severity : enhancement
112 Description: Patchless ZC(zero copy) socklnd
113 Details : New protocol for socklnd, socklnd can support zero copy without
114 kernel patch, it's compatible with old socklnd. Checksum is
115 moved from tunables to modparams.
119 Description: When ksocknal_del_peer() is called upon a peer whose
120 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
121 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
124 Frequency : when ptlrpc is under heavy use and runs out of request buffer
126 Description: In lnet_match_blocked_msg(), md can be used without holding a
130 Frequency : very rarely
132 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
133 If connd connects a route which has been closed by
134 ksocknal_shutdown(), ksocknal_create_routes() may create new
135 routes which hold references on the peer, causing shutdown
136 process to wait for peer to disappear forever.
138 Severity : enhancement
140 Description: Dump XT3 portals traces on kptllnd timeout
141 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
142 dump Cray portals debug traces to a file. The kptllnd module
143 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
144 is the basename of the dump file.
147 Frequency : infrequent
149 Description: kernel ptllnd fix bug in connection re-establishment
150 Details : Kernel ptllnd could produce protocol errors e.g. illegal
151 matchbits and/or violate the credit flow protocol when trying
152 to re-establish a connection with a peer after an error or
155 Severity : enhancement
157 Description: Allow /proc/sys/lnet/debug to be set symbolically
158 Details : Allow debug and subsystem debug values to be read/set by name
159 in addition to numerically, for ease of use.
162 Frequency : only in configurations with LNET routers
164 Description: routes automatically marked down and recovered
165 Details : In configurations with LNET routers if a router fails routers
166 now actively try to recover routes that are down, unless they
167 are marked down by an administrator.
169 ------------------------------------------------------------------------------
171 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
173 - rework CDEBUG messages rate-limiting mechanism b=10375
174 - add per-socket tunables for socklnd if the kernel is patched b=10327
176 ------------------------------------------------------------------------------
178 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
180 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
181 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
182 - quiet LNET startup LNI message for liblustre b=10128
183 - Better console error messages if 'ip2nets' can't match an IP address
184 - Fixed overflow/use-before-set bugs in linux-time.h
185 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
186 - LNET teardown failed an assertion about the route table being empty
187 - Fixed a crash in LNetEQPoll(<invalid handle>)
188 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
189 - improve debug message for liblustre/Catamount nodes (b=10116)
191 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
192 * Configuration change for the XT3
193 The PTLLND is now used to run Lustre over Portals on the XT3.
194 The configure option(s) --with-cray-portals are no longer
195 used. Rather --with-portals=<path-to-portals-includes> is
196 used to enable building on the XT3. In addition to enable
197 XT3 specific features the option --enable-cray-xt3 must be
200 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
201 * Portals has been removed, replaced by LNET.
202 LNET is new networking infrastructure for Lustre, it includes a
203 reorganized network configuration mode (see the user
204 documentation for full details) as well as support for routing
205 between different network fabrics. Lustre Networking Devices
206 (LNDS) for the supported network fabrics have also been created
207 for this new infrastructure.
209 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
214 Frequency : rare (large Voltaire clusters only)
216 Description: the default number of reserved transmit descriptors was too low
217 for some large clusters
218 Details : As a workaround, the number was increased. A proper fix includes
221 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
226 Frequency : occasional (large-scale events, cluster reboot, network failure)
228 Description: too many error messages on console obscure actual problem and
229 can slow down/panic server, or cause recovery to fail repeatedly
230 Details : enable rate-limiting of console error messages, and some messages
231 that were console errors now only go to the kernel log
233 Severity : enhancement
235 Description: add /proc/sys/portals/catastrophe entry which will report if
236 that node has previously LBUGged
238 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
240 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
242 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
244 - handle error return code in kranal_check_fma_rx() (5915,6054)
246 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
248 - update vibnal (Voltaire IB NAL)
249 - update gmnal (Myrinet NAL), gmnalid
251 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
253 * Landed portals:b_port_step as follows...
255 - removed CFS_DECL_SPIN*
256 just use 'spinlock_t' and initialise with spin_lock_init()
258 - removed CFS_DECL_MUTEX*
259 just use 'struct semaphore' and initialise with init_mutex()
261 - removed CFS_DECL_RWSEM*
262 just use 'struct rw_semaphore' and initialise with init_rwsem()
264 - renamed cfs_sleep_chan -> cfs_waitq
265 cfs_sleep_link -> cfs_waitlink
267 - fixed race in linux version of arch-independent socknal
268 (the ENOMEM/EAGAIN decision).
270 - Didn't fix problems in Darwin version of arch-independent socknal
271 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
273 - removed libcfs types from non-socknal header files (only some types
274 in the header files had been changed; the .c files hadn't been