1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
7 Bugzilla : 5719, 9635, 9792, 9684,
8 Description: OST (or MDS) trips assertions in (re)connection under heavy load
9 Details : If a server is under heavy load and cannot reply to new
10 connection requests before the client resends the (re)connect,
11 the connection handling code can behave badly if two service
12 threads are concurrently handing separate (re)connections from
13 the same client. Add better locking to the connection handling
14 code, and ensure that only a single connection will be processed
15 for a given client UUID, even if the lock is dropped.
17 Severity : enhancement
19 Description: add TCP zero-copy support to kernel
20 Details : Add support to the kernel TCP stack to allow zero-copy bulk
21 sends if the hardware supports scatter-gather and checksumming.
22 This allows socklnd to do client-write and server-read more
23 efficiently and reduce CPU utilization from skbuf copying.
26 Frequency : only if NFS exporting from client
28 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
29 Details : If setting attributes on a file created under NFS that had
30 never been opened it would be possible to oops the client
31 if the file had no objects.
34 Frequency : always for liblustre
36 Description: liblustre client does MDS+OSTs setattr RPC for each write
37 Details : When doing a write from a liblustre client, the client
38 incorrectly issued an RPC to the MDS and each OST the file was
39 striped over in order to update the timestamps. When writing
40 with small chunks and many clients this could overwhelm the MDS
41 with RPCs. In all cases it would slow down the write because
42 these RPCs are unnecessary.
44 Severity : enhancement
46 Description: allow number of MDS service threads to be changed at module load
47 Details : It is now possible to change the number of MDS service threads
48 running. Adding "options mds mds_num_threads=N" will set the
49 number of threads for the next time Lustre is restarted (assuming
50 the "mds" module is also reloaded at that time). The default
51 number of threads will stay the same, 32 for most systems.
56 Description: OST crash if filesystem is unformatted or corrupt
57 Details : If an OST is started on a device that has never been formatted
58 or if the filesystem is corrupt and cannot even mount then the
59 error handling cleanup routines would dereference a NULL pointer.
64 Description: NULL pointer deref in llap_from_page.
65 Details : get_cache_page_nowait can return a page with NULL (or otherwise
66 incorrect) mapping if the page was truncated/reclaimed while it was
67 searched for. Check for this condition and skip such pages when
68 doing readahead. Introduce extra check to llap_from_page() to
69 verify page->mapping->host is non-NULL (so page is not anonymous).
72 Frequency : Sometimes when using sys_sendfile
74 Description: "page not covered by a lock" warnings from ll_readpage
75 Details : sendfile called ll_readpage without right page locks present.
76 Now we introduced ll_file_sendfile that does necessary locking
77 around call to generic_file_sendfile() much like we do in
81 Frequency : with certain MDS communication failures at client mount time
83 Description: NULL pointer deref after failed client mount
84 Details : a client connection request may delayed by the network layer
85 and not be sent until after the PTLRPC layer has timed out the
86 request. If the client fails the mount immediately it will try
87 to clean up before the network times out the request. Add a
88 reference from the request import to the obd device and delay
89 the cleanup until the network drops the request.
92 Frequency : occasionally during client (re)connect
94 Description: assertion failure during client (re)connect
95 Details : processing a client connection request may be delayed by the
96 client or server longer than the client connect timeout. This
97 causes the client to resend the connection request. If the
98 original connection request is replied in this interval, the
99 client may trip an assertion failure in ptlrpc_connect_interpret()
100 which thought it would be the only running connect process.
103 Frequency : only with obd_echo servers and clients that are rebooted
105 Description: kernel BUG accessing uninitialized data structure
106 Details : When running an obd_echo server it did not start the ping_evictor
107 thread, and when a client was evicted an uninitialized data
108 structure was accessed. Start the ping_evictor in the RPC
109 service startup instead of the OBD startup.
111 Severity : enhancement
112 Bugzilla : 10393 (patchless)
113 Description: Remove dependency on various unexported kernel interfaces.
114 Details : No longer need reparent_to_init, exit_mm, exit_files,
115 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
118 Frequency : rare (only users of deprecated and unsupported LDAP config)
120 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
121 Details : LDAP apparently contains 'lustreName' attributes instead of
122 'name'. A simple remapping of the name is sufficient.
125 Frequency : rare (only with non-default dump_on_timeout debug enabled)
127 Description: waiting_locks_callback trips kernel BUG if client is evicted
128 Details : Running with the dump_on_timeout debug flag turned on makes
129 it possible that the waiting_locks_callback() can try to dump
130 the Lustre kernel debug logs from an interrupt handler. Defer
131 this log dumping to the expired_lock_main() thread.
133 Severity : enhancement
135 Description: Support NFS exporting on 2.6 kernels.
136 Details : Implement non-rawops metadata methods for NFS server to use without
137 changing NFS server code.
140 Frequency : very rare (synthetic metadata workload only)
142 Description: two racing renames might cause an MDS thread to deadlock
143 Details : Running the "racer" program may cause one MDS thread to rename
144 a file from being the source of a rename to being the target of
145 a rename at exactly the same time that another thread is doing
146 so, and the second thread has already enqueued these locks after
147 doing a lookup of the target and is trying to relock them in
148 order. Ensure that we don't try to re-lock the same resource.
151 Frequency : only very large systems with liblustre clients
153 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
154 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
155 all OSTs in parallel. This allows the eviction of stale liblustre
156 clients to proceed much faster than if they were done in series,
157 and also offers similar improvements for other set_info RPCs.
161 Description: excessive CPU usage during initial read phase on client
162 Details : During the initial read phase on a client, it would agressively
163 retry readahead on the file, consuming too much CPU and impacting
164 performance (since 1.4.5.8). Improve the readahead algorithm
165 to avoid this, and also improve some other common cases (read
166 of small files in particular, where "small" is files smaller than
167 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
171 Description: MDS crash when receiving packet with unknown intent.
172 Details : Do not LBUG in unknown intent case, just return -EFAULT
175 ------------------------------------------------------------------------------
177 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
179 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
180 INTEROPERATE with older versions automatically. Please read the
181 user documentation before upgrading any part of a live system.
182 * WARNING: Lustre networking configuration changes are required with
183 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
186 * Support for newer kernels:
187 2.6.9-22.0.2.EL (RHEL 4),
188 2.6.5-7.244 (SLES 9) - same as 1.4.5.2.
189 2.6.12.6 vanilla (kernel.org)
192 Severity : enhancement
194 Description: Introduced Lustre Networking (LNET)
195 Details : LNET is new networking infrastructure for Lustre, it includes
196 a reorganized network configuration mode (see the user
197 documentation for full details) as well as support for routing
198 between different network fabrics. Lustre Networking Devices
199 (LNDs) for the supported network fabrics have also been
200 created for this new infrastructure.
202 Severity : enhancement
203 Description: Introduced Access control lists
204 Details : clients can set ACLs on files and directories in order to have
205 more fine-grained permissions than the standard Unix UGO+RWX.
206 The MDS must be started with the "-o acl" mount option.
208 Severity : enhancement
209 Description: Introduced filesystem quotas
210 Details : Administrators may now establish per-user quotas on the
213 Severity : enhancement
215 Description: Configuration change for the XT3
216 The PTLLND is now used to run Lustre over Portals on the XT3
217 The configure option(s) --with-cray-portals are no longer used.
218 Rather --with-portals=<path-to-portals-includes> is used to
219 enable building on the XT3. In addition to enable XT3 specific
220 features the option --enable-cray-xt3 must be used.
225 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
226 Details : A race between allocating a new llcd and re-getting the llcd_lock
227 allowed another thread to grab newly-allocated llcd.
229 Severity : enhancement
231 Description: 2.6 OST async journal commit and locking fix to improve performance
232 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
233 2.6 kernels are now async as they already were in 2.4 kernels so
234 that they can commit concurrently with the network bulk transfer.
235 For block-allocated files the filter allocation semaphore is held
236 to avoid filesystem fragmentation during allocation. BKL lock
237 removed for 2.6 xattr operations where it is no longer needed.
242 Description: lconf incorrectly determined whether two IP networks could talk
243 Details : In some more complicated routing and multiple-network
244 configurations, lconf will avoid trying to make a network
245 connection to a disjoint part of the IP space. It was doing the
246 math incorrectly for one set of cases.
251 Description: Fix for potential infinite loop processing records in an llog.
252 Details : If an llog record is corrupted/zeroed, it is possible to loop
253 forever in llog_process(). Validate the llog record length
254 and skip the remainder of the block on error.
257 Frequency : occasional (liblustre only)
259 Description: liblustre could not open files whose last component is a symlink
260 Details : sysio_path_walk() would incorrectly pass the open intent to
261 intermediate path components.
264 Frequency : rare (liblustre only with non-standard tuning)
265 Bugzilla : 7201 (7350)
266 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
267 Details : llu_lookup_finish_locks() tries to set lock data on a lock
268 after it has been released, only do this for referenced locks
270 Severity : enhancement
272 Description: specifying an (invalid) directory default stripe_size of -1
273 would reset the directory default striping
274 Details : stripe_size -1 was used internally to signal directory stripe
275 removal, now use "all default" to signal dir stripe removal
276 as a directory striping of "all default" is not useful
279 Frequency : common for large clusters running liblustre clients
281 Description: doing an ls when liblustre clients are running is slow
282 Details : sending a glimpse AST to a liblustre client waits for every AST
283 to time out, as liblustre clients will not respond. Since they
284 cannot cache data we refresh the OST lock LVB from disk instead.
286 Severity : enhancement
288 Description: doing an ls at the same time as file IO can be slow
289 Details : enqueue and other "small" requests can be blocked behind many
290 large IO requests. Create a new OST IO portal for non-IO
291 requests so they can be processed faster.
294 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
296 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
297 created file with mode 000
298 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
299 allow an ftruncate() as a non-root user to a file with mode 000.
300 Lustre now respects this flag to disable mode checks when
301 truncating a file owned by the user
304 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
307 Description: Revoking locks from clients that went dead or catatonic might take
309 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
310 cancellation of such locks instant on servers without waiting for
311 any reply from clients. Clients drops these locks when cancel
312 notification from server is received without replying.
315 Frequency : liblustre-only, when liblustre client dies or becomes busy
317 Description: Doing ls on Linux clients can take a long time with active
319 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
320 granting such locks to it in the first place if possible. Locks
321 are taken by proxy on the OST during the read or write and
322 dropped immediately afterward. Add connect flags handling, do
323 not grant locks to liblustre clients for glimpse ASTs.
325 Severity : enhancement
327 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
328 Details : Existing read-ahead algorithm is tuned for the case of streamlined
329 sequential reads and behaves badly with applications doing random
330 reads. Improve it by reading ahead at least read region, and
331 avoiding excessive large RPC for small reads.
333 Severity : enhancement
335 Description: Creating more than 1000 files for a single job may cause a load
336 imbalance on the OSTs if there are also a large number of OSTs.
337 Details : qos_prep_create() uses an OST index reseed value that is an
338 even multiple of the number of available OSTs so that if the
339 reseed happens in the middle of the object allocation it will
340 still utilize the OSTs as uniformly as possible.
345 Description: OST or MDS may oops in ping_evictor_main()
346 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
347 but doesn't restart at beginning of obd_exports_timed list
350 Severity : enhancement
352 Description: improve by-nid export eviction on the MDS and OST
353 Details : allow multiple exports with the same NID to be evicted at one
354 time without re-searching the exports list.
357 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
359 Description: MDS may oops in groups_free()
360 Details : in rare race conditions a newly allocated group_info struct is
361 freed again, and this can be NULL. The 2.4 compatibility code
362 for groups_free() checked for a NULL pointer, but 2.6 did not.
365 Frequency : common for liblustre clients doing little filesystem IO
366 Bugzilla : 9352, 7313
367 Description: server may evict liblustre clients accessing contended locks
368 Details : if a client is granted a lock or receives a completion AST
369 with a blocking AST already set it would not reply to the AST
370 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
373 Frequency : lfs setstripe, only systems with more than 160 OSTs
375 Description: unable to set striping with a starting offset beyond OST 160
376 Details : llapi_create_file() incorrectly limited the starting stripe
377 index to the maximum single-file stripe count.
380 Frequency : LDAP users only
382 Description: lconf did not handle in-kernel recovery with LDAP properly
383 Details : lconf/LustreDB get_refs() is searching the wrong namespace
385 Severity : enhancement
387 Description: bind OST threads to NUMA nodes to improve performance
388 Details : all OST threads are uniformly bound to CPUs on a single NUMA
389 node and do their allocations there to localize memory access
391 Severity : enhancement
393 Description: llmount can determine client NID directly from Myrinet (GM)
394 Details : the client NID code from gmnalnid was moved directly into
395 llmount, removing the need to use this or specifying the
396 client NID explicitly when mounting GM clients with zeroconf
399 Frequency : if client is started with down MDS
401 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
402 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
403 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
404 ptlrpc_import_delayed_req() because we didn't check if the
405 request was interrupted, and we also didn't break out of the
406 event loop if there was no timeout
411 Description: data loss during non-page-aligned writes to a single file from
412 both multiple nodes and multiple threads on one node at same time
413 Details : updates to KMS and lsm weren't protected by common lock. Resulting
414 inconsistency led to false short-reads, that were cached and later
415 used by ->prepare_write() to fill in partially written page,
416 leading to data loss.
419 Frequency : always, if lconf --abort_recovery used
421 Description: lconf --abort_recovery fails with 'Operation not supported'
422 Details : lconf was attempting to abort recovery on the MDT device and not
425 Severity : enhancement
427 Description: remove cleanup logs
428 Details : replace lconf-generated cleanup logs with lustre internal
429 cleanup routines. Eliminates the need for client-cleanup and
432 Severity : enhancement
434 Description: add support for EAs (user and system) on lustre filesystems
435 Details : it is now possible to store extended attributes in the Lustre
436 client filesystem, and with the user_xattr mount option it
437 is possible to allow users to store EAs on their files also
439 Severity : enhancement
441 Description: Add possibility (config option) to show minimal available OST free
443 Details : When compiled with --enable-mindf configure option, statfs(2)
444 (and so, df) will return least minimal free space available from
445 all OSTs as amount of free space on FS, instead of summary of
446 free spaces of all OSTs.
448 Severity : enhancement
450 Description: do not expand extent locks acquired on OST-side
451 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
452 by server: they are not cached anyway.
455 Frequency : when mmap is used/binaries executed from Lustre
457 Description: Unmmap pages before throwing them away from read cache.
458 Details : llap_shrink cache now attempts to unmap pages before discarding
459 them (if unmapping failed - do not discard). SLES9 kernel has
460 extra checks that trigger if this unmapping is not done first.
465 Description: lconf didn't resolve symlinks before checking to see whether a
466 given mountpoint was already in use
469 Frequency : when migrating failover services
470 Bugzilla : 6395, 9514
471 Description: When migrating a subset of services from a node (e.g. failback
472 from a failover service node) the remaining services would
473 time out and evict clients.
474 Details : lconf --force (implied by --failover) sets the global obd_timeout
475 to 5 seconds in order to quickly disconnect, but this caused
476 other RPCs to time out too quickly. Do not change the global
477 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
479 Severity : enhancement
480 Frequency : if MDS is started with down OST
482 Description: Allow startup/shutdown of an MDS without depending on the
483 availability of the OSTs.
484 Details : Asynchronously call mds_lov_synchronize during MDS startup.
485 Add appropriate locking and lov-osc refcounts for safe
486 cleaning. Add osc abort_inflight calls in case the
487 synchronize never started.
490 Frequency : occasional (Cray XT3 only)
492 Description: root not authorized to access files in CRAY_PORTALS environment
493 Details : The client process capabilities were not honoured on the MDS in
494 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
495 been accessed by an authorized user then root was able to access
496 the file on the local client also. The root user capabilities
497 are now allowed on the MDS, as this environment has secure UID.
500 Frequency : occasional
502 Description: ldiskfs "too long searching" message happens too often
503 Details : A debugging message (otherwise harmless) prints too often on
504 the OST console. This has been reduced to only happen when
505 there are fragmentation problems on the filesystem.
510 Description: Division by zero in statfs when all OSCs are inactive
511 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
512 lov_statfs divides by value returned by lov_get_stripecnt().
516 Bugzilla : 9489, 3273
517 Description: First write from each client to each OST was only 4kB in size,
518 to initialize client writeback cache, which caused sub-optimal
519 RPCs and poor layout on disk for the first writen file.
520 Details : Clients now request an initial cache grant at (re)connect time
521 and so that they can start streaming writes to the cache right
522 away and always do full-sized RPCs if there is enough data.
523 If the OST is rebooted the client also re-establishes its grant
524 so that client cached writes will be honoured under the grant.
529 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
530 Details : Now I/O RPCs go to different portal number and (presumably) fast
531 lock requests (and glimses) and other RPCs get their own service
532 threads pool that should be able to service those RPCs
535 Severity : enhancement
537 Description: Ability to exchange lustre version between client and servers and
538 issue warnings at client side if client is too old. Also for
539 liblustre clients there is ability to refuse connection of too old
541 Details : New 'version' field is added to connect data structure that is
542 filled with version info. That info is later checked by server and
546 Frequency : rare, liblustre only.
547 Bugzilla : 9296, 9581
548 Description: Two simultaneous writes from liblustre at offset within same page
549 might proceed at the same time overwriting eachother with stale
551 Details : I/O lock withing llu_file_prwv was released too early, before data
552 actually was hitting the wire. Extended lock-holding time until
553 server acknowledges receiving data.
556 Frequency : extremely rare. Never observed in practice.
558 Description: avoid generating lustre_handle cookie of 0.
559 Details : class_handle_hash() generates handle cookies by incrementing
560 global counter, and can hit 0 occasionaly (this is unlikely, but
561 not impossible, because initial value of cookie counter is
562 selected randonly). Value of 0 is used as a sentinel meaning
563 "unassigned handle" --- avoid it. Also coalesce two critical
564 sections in this function into one.
566 Severity : enhancement
568 Description: allow liblustre clients to delegate truncate locking to OST
569 Details : To avoid overhead of locking, liblustre client instructs OST to
570 take extent lock in ost_punch() on client's behalf. New connection
571 flag is added to handle backward compatibility.
573 Severity : enhancement
574 Bugzilla : 4928, 7341, 9758
575 Description: allow number of OST service threads to be specified
576 Details : a module parameter allows the number of OST service threads
577 to be specified via "options ost ost_num_threads=X" in
578 /etc/modules.conf or /etc/modutils.conf.
582 Bugzilla : 6146, 9635, 9895
583 Description: servers crash with bad pointer in target_handle_connect()
584 Details : In rare cases when a client is reconnecting it was possible that
585 the connection request was the last reference for that export.
586 We would temporarily drop the export reference and get a new
587 one, but this may have been the last reference and the export
588 was just destroyed. Get new reference before dropping old one.
590 Severity : enhancement
591 Frequency : if client is started with failover MDS
593 Description: Allow multiple MDS hostnames in the mount command
594 Details : Try to read the configuration from all specified MDS
595 hostnames during a client mount in case the "primary"
598 Severity : enhancement
600 Description: Stop sending data to evicted clients as soon as possible.
601 Details : Check if the client we are about to send or are sending data to
602 was evicted already. (Check is done every second of waiting,
603 for which l_wait_event interface was extended to allow checking
604 of exit condition at specified intervals).
607 Frequency : rare, normally only when NFS exporting is done from client
609 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
611 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
614 Frequency : always, if srand() is called before liblustre initialization
616 Description: Liblustre uses system PRNG disturbing its usage by user application
617 Details : Introduce internal to lustre fast and high-quality PRNG for
618 lustre usage and make liblustre and some other places in generic
619 lustre code to use it.
621 Severity : enhancement
622 Bugzilla : 9477, 9557, 9870
623 Description: Verify that the MDS configuration logs are updated when xml is
624 Details : Check if the .xml configuration logs are newer than the config
625 logs stored on the MDS and report an error if this is the case.
626 Request --write-conf, or allow starting with --old_conf.
628 Severity : enhancement
630 Description: Handle symlinks in the path when checking if Lustre is mounted.
631 Details : Resolve intermediate symlinks when checking if a client has
632 mounted a filesystem to avoid duplicate client mounts.
637 Description: lconf can hit an error exception but still return success.
638 Details : The lconf command catches the Command error exception at the top
639 level script context and will exit with the associated exit
640 status, but doesn't ensure that this exit status is non-zero.
645 Description: failure of ptlrpc thread startup can cause oops
646 Details : Starting a ptlrpc service thread can fail if there are a large
647 number of threads or the server memory is very fragmented.
648 Handle this without oopsing.
651 Frequency : always, only if liblustre and non-default acceptor port was used
653 Description: liblustre cannot connect to servers with non-default acceptor port
654 Details : tcpnal_set_default_params() was not called and was therefore
655 ignoring the environment varaible TCPNAL_PORT, as well as other
656 TCPNAL_ environment variables
661 Description: two objects could be created on the same OST for a single file
662 Details : If an OST is down, in some cases it was possible to create two
663 objects on a single OST for a single file. No problems other
664 than potential performance impact and spurious error messages.
668 Bugzilla : 5681, 9562
669 Description: Client may oops in ll_unhash_aliases
670 Details : Client dcache may become inconsistent in race condition.
671 In some cases "getcwd" can fail if the current directory is
677 Description: Inode refcounting problems in NFS export code
678 Details : link_raw functions used to call d_instantiate without obtaining
679 extra inode reference first.
683 Bugzilla : 9942, 9903
684 Description: Referencing freed requests leading to crash, memleaks with NFS.
685 Details : We used to require that call to ll_revalidate_it was always
686 followed by ll_lookup_it. Also with revalidate_special() it is
687 possible to call ll_revalidate_it() twice for the same dentry
688 even if first occurence returned success. This fix changes semantic
689 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
690 reference on a request referred from the intent.
691 ll_intent_release() then releases such a request.
694 Frequency : rare, normally benchmark loads only
696 Description: unlinked inodes were kept in memory on the client
697 Details : If a client is repeatedly creating and unlinking files it
698 can accumulate a lot of stale inodes in the inode slab cache.
699 If there is no other client load running this can cause the
700 client node to run out of memory. Instead flush old inodes
701 from client cache that have the same inode number as a new inode.
704 Frequency : rare, unless heavy write-truncate concurrency is continuous
705 Bugzilla : 4180, 6984, 7171, 9963, 9331
706 Description: OST becomes very slow and/or deadlocked during object unlink
707 Details : filter_destroy() was holding onto the parent directory lock
708 while truncating+unlinking objects. For very large objects this
709 may block other threads for a long time and slow overall OST
710 responsiveness. It may also be possible to get a lock ordering
711 deadlock in this case, or run out of journal credits because of
712 the combined truncate+unlink. Solution is to do object truncate
713 first in one transaction without parent lock, and then do the
714 final unlink in a new transaction with the parent lock. This
715 reduces the lock hold time dramatically.
718 Frequency : rare, 2.4 kernels only
720 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
721 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
722 before unmounting their filesystem in order to avoid deadlock.
723 The kernel_locked() function in 2.4 kernels only checks whether
724 the kernel lock is held, not whether it is this process that is
725 holding it as 2.6 kernels do.
730 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
731 Details : The client ptlrpc code may be trying to reconnect to a down
732 server before a previous connection attempt has timed out.
733 Increase the reconnect interval to be longer than the connection
734 timeout interval to avoid sending duplicate connections to
738 Frequency : echo_client brw_test command
740 Description: fix echo_client to work with OST preallocated code
741 Details : OST preallocation code (5137) didn't take echo_client IO path
742 into account: echo_client calls filter methods outside of any
743 OST thread and, hence, there is no per-thread preallocated
744 pages and buffers to use. Solution: hijack pga pages for IO. As
745 a byproduct, this avoids unnecessary data copying.
749 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
750 Description: Client can oops in mdc_commit_close() after open replay
751 Details : It was possible for the MDS to return an open request with no
752 transaction number in mds_finish_transno() if the client was
753 evicted, but without actually returning an error. Clients
754 would later try to replay that open and may trip an assertion
755 Simplify the client close codepath, and always return an error
756 from the MDS in case the open is not successful.
759 Frequency : rare, 2.6 OSTs only
761 Description: OST may deadlock under high load on fragmented files
762 Details : If there was a heavy load and highly-fragmented OST filesystems
763 it was possible to have all the OST threads deadlock waiting on
764 allocation of biovecs, because the biovecs were not released
765 until the entire RPC IO was completed. Instead, release biovecs
766 as soon as they are complete to ensure forward IO progress.
768 Severity : enhancement
770 Description: Support for specifying external journal device at mount
771 Details : If an OST or MDS device is formatted with an external journal
772 device, this device major/minor is stored in the ext3 superblock
773 and may not be valid for failover. Allow detecting and
774 specifying the external journal at mount time.
779 Description: Mounting an MDS with pending unlinked files may cause oops
780 Details : target_finish_recovery() calls mds_postrecov() which returned
781 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
782 considers this an error and immediately begins cleaning up the
783 lov, just after starting the mds_lov process
785 Severity : enhancement
787 Description: Implement 'lfs df' to report actual free space on per-OST basis
788 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
789 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
790 usage of objects; '-h' to report in human readable format.
792 ------------------------------------------------------------------------------
794 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
801 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
802 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
803 at startup during orphan unlinking. Instead, skip buddy update
804 before setup, as it will be regenerated after recovery is complete.
809 Description: If an OST is inactive, its locks might reference stale inodes.
810 Details : lov_change_cbdata() must iterate over all namespaces, even if
811 they are inactive to clear inode references from the lock.
813 Severity : enhancement
814 Frequency : occasional, if non-standard max_dirty_mb used
816 Description: Client will block write RPCs if not enough grant
817 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
818 then the client will block writes while waiting for another RPC
819 to complete instead of consuming its dirty limit. With change
820 we get improved performance when max_dirty_mb is small.
822 Severity : enhancement
823 Bugzilla : 3389, 6253
824 Description: Add support for supplementary groups on the MDS.
825 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
826 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
827 lookups for user supplementary groups into a cache.
831 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
832 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
833 for liblustre O_EXCL is enforced by MDS.
837 Description: readdir never returns NULL in liblustre.
838 Details : Corrected llu_iop_getdirentries logic, to return offset of next
839 dentry in struct dirent.
843 Frequency : liblustre only, depends on application IO pattern
844 Description: liblustre clients evicted if not contacting servers
845 Details : Don't put liblustre clients into the ping_evictor list, so
846 they will not be evicted by the pinger ever.
848 Severity : enhancement
850 Description: Add ability to evict clients by NID from MDS.
851 Details : By echoing "nid:$NID" string into
852 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
853 $NID would be instantly evicted from this MDS and from all active
854 OSTs connected to it.
858 Description: Do not query file size twice, somewhat slowing stat(2) calls.
859 Details : lookup_it_finish() used to query file size from OSTs that was not
864 Description: service threads change working directory to that of init
865 Details : Starting lustre service threads may pin the working directory
866 of the parent thread, making that filesystem busy. Threads
867 now change to the working directory of init to avoid this.
871 Frequency : during shutdown only
872 Description: shutdown with a failed MDS or OST can cause unmount to hang
873 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
879 Description: chmod/chown may include an extra supplementary group
880 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
881 supplementary group and if none is specified this is used.
884 Bugzilla : 5479 (6816)
886 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
887 Details : If lookup is in progress on a file that is unlinked we might try
888 to revalidate the inode and fail in revalidate after lookup is
889 complete and ll_file_open() enqueues the open again but
890 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
893 Frequency : always, if lconf --abort_recovery used
895 Description: lconf --abort_recovery fails with 'Operation not supported'
896 Details : lconf was attempting to abort recovery on the MDT device and not
899 ------------------------------------------------------------------------------
901 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
906 Frequency : rare (only unsupported configurations with a node running as an
908 Bugzilla : 6514, 5137
909 Description: Mounting a Lustre file system on a node running as an OST could
911 Details : OSTs now preallocates memory needed to write out data at
912 startup, instead of when needed, to avoid having to
913 allocate memory in possibly low memory situations.
914 Specifically, if the file system is mounted on on OST,
915 memory pressure could force it to try to write out data,
916 which it needed to allocate memory to do. Due to the low
917 memory, it would be unable to do so and the node would
920 Severity : enhancement
922 Description: Addition of lconf --service command line option
923 Details : lconf now accepts a '--service <arg>' option, which is
924 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
926 Severity : enhancement
928 Description: Failover mode is now the default for OSTs.
929 Details : By default, OSTs will now run in failover mode. To return to
930 the old behaviour, add '--failout' to the lmc line for OSTs.
932 Severity : enhancement
934 Description: Health checks are now provided for MDS and OSTs
935 Details : Additional detailed health check information on MSD and OSTs
936 is now provided through the procfs health_check value.
939 Frequency : occasional, depends on IO load
941 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
942 after numerous create/delete cycles
943 Details : The ext3 inode allocation policy would not allocate new inodes
944 very well on the OSTs because there are no new directories
945 being created. Instead we look for groups with free space if
946 the parent directories are nearly full.
951 Description: Network or server problems during mount may cause partially
952 mounted clients instead of returning an error.
953 Details : The config llog parsing code may overwrite the error return
954 code during mount error handling, returning success instead
960 Description: MDS can fail to allocate large reply buffers
961 Details : After long uptimes the MDS can fail to allocate large reply
962 buffers (e.g. zconf client mount config records) due to memory
963 fragmentation or consumption by the buffer cache. Preallocate
964 some large reply buffers so that these replies can be sent even
965 under memory pressure.
969 Frequency : rare (liblustre)
970 Description: fsx running with liblustre complained that using truncate() to
971 extend the file doesn't work. This patch corrects that issue.
972 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
973 ATTR_SIZE and lsm use in llu_setattr_raw.
977 Frequency : rare, only 2.6 kernels
978 Description: Unusual file access patterns on the MDS may result in inode
979 data being lost in very rare circumstances.
980 Details : Bad interaction between the ea-in-inode patch and the "no-read"
981 code in the 2.6 kernel caused the inode and/or EA data not to
982 be read from disk, causing single-file corruption.
986 Frequency : rare, only 2.6 filesystems using extents
987 Description: Heavy concurrent write and delete load may cause data corruption.
988 Details : It was possible under high-load situations to have an extent
989 metadata block in the block device cache from a just-unlinked
990 file overwrite a newly-allocated data block. We now unmap any
991 metadata buffers that alias just-allocated data blocks.
995 Frequency : filesystems with default stripe_count larger than 77
996 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
997 Details : lconf specifies an inode size of 4096 bytes when the default
998 stripe_count is larger than 77. This conflicts with the default
999 inode density of 1 per 4096 bytes. Allocate smaller inodes in
1000 this case to avoid pinning too much memory for large EAs.
1002 ------------------------------------------------------------------------------
1004 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
1009 Frequency : rare (extremely heavy IO load with hundreds of clients)
1011 Description: Client is evicted, gets IO error writing to file
1012 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
1013 caused clients to be evicted for AST timeouts. The fixes in
1014 bug 5192 mean we no longer need to have such short AST timeouts
1015 so ldlm_timeout has been increased.
1018 Frequency : occasional during --force or --failover shutdown under load
1019 Bugzilla : 5949, 4834
1020 Description: Server oops/LBUG if stopped with --force or --failover under load
1021 Details : a collection of import/export refcount and cleanup ordering
1022 issues fixed for safer force cleanup
1025 Frequency : only filesystems larger than 120 OSTs
1026 Bugzilla : 5990, 6223
1027 Description: lfs getstripe would oops on a very large filesystem
1028 Details : lov_getconfig used kfree on vmalloc'd memory
1031 Frequency : only filesystems exporting via NFS to Solaris 10 clients
1032 Bugzilla : 6242, 6243
1033 Description: reading from files that had been truncated to a non-zero size
1034 but never opened returned no data
1035 Details : ll_file_read() reads zeros from no-object files to EOF
1040 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
1041 Details : There's little harm in aborting MDS/OSS recovery and letting it
1042 try again, so I removed the LASSERT and return an error instead.
1044 Severity : enhancement
1046 Description: New debugging infrastructure for tracking down data corruption
1047 Details : The I/O checksum code was replaced to: (a) control it at runtime,
1048 (b) cover more of the client-side code path, and (c) try to narrow
1049 down where problems occurred
1053 Bugzilla : 3819, 4364, 4397, 6313
1054 Description: Racing close and eviction MDS could cause assertion in mds_close
1055 Details : It was possible to get multiple mfd references during close and
1056 client eviction, leading to one thread referencing a freed mfd.
1058 Severity: : enhancement
1059 Bugzilla : 3262, 6359
1060 Description: Attempts to reconnect to servers are now more aggressive.
1061 Details : This builds on the enhanced upcall-less recovery that was added
1062 in 1.4.2. When trying to reconnect to servers, clients will
1063 now try each server in the failover group every 10 seconds. By
1064 default, clients would previously try one server every 25 seconds.
1069 Description: After recovery, certain operations trigger a failed
1070 assertion on a client.
1071 Details : Failing over an mds, using lconf -d --failover, while a
1072 client was doing a readdir() call would cause the client to
1073 LBUG after recovery completed and the readdir() was resent.
1075 Severity : enhancement
1077 Description: Default groups are now added by lconf
1078 Details : You can now run lconf --group <servicename> without having to
1079 manually add groups with lmc.
1082 Frequency : occasional
1084 Description: Nodes with an elan id of 0 trigger a failed assertion
1087 Frequency : always when accessing e.g. tty/console device nodes
1089 Description: tty and some other devices nodes cannot be used on lustre
1090 Details : file's private_data field is used by device data and lustre
1091 values in there got lost. New field was added to struct file to
1092 store fs-specific private data.
1095 Frequency : when exporting Lustre via NFS
1097 Description: NFSD failed occasionally when looking up a path component
1098 Details : NFSD is looking up ".." which was broken in ext3 directories
1099 that had grown large enough to become hashed.
1102 Frequency : Clusters with multiple interfaces not on the same subnet
1104 Description: Nodes will repeatedly try to reconnect to an interface which it
1105 cannot reach and report an error to the log.
1106 Details : Extra peer list entries will be created by lconf with some peers
1107 unreachable. lconf now validates the peer before adding it.
1110 Frequency : Only if a default stripe is set on the filesystem root.
1112 Description: Setting a default stripe on the filesystem root prevented the
1113 filesystem from being remounted.
1114 Details : The client was sending extra request flags in the root getattr
1115 request and did not allocate a reply buffer for the dir EA.
1118 Frequency : occasional, higher if lots of files are accessed by one client
1119 Bugzilla : 6159, 6097
1120 Description: Client trips assertion regarding lsm mismatch/magic
1121 Details : While revalidating inodes the VFS looks up inodes with ifind()
1122 and in rare cases can find an inode that is being freed.
1123 The ll_test_inode() code will free the lsm during ifind()
1124 when it finds an existing inode and then the VFS later attaches
1125 this free lsm to a new inode.
1129 Bugzilla : 6422, 7030
1130 Description: MDS deadlock between mkdir and client eviction
1131 Details : Creating a new file via mkdir or mknod (starting a transaction
1132 and getting the ns lock) can deadlock with client eviction
1133 (gets ns lock and trying to finish a synchronous transaction).
1136 Frequency : occasional
1137 Description: While starting a server, the fsfilt_ext3 module could not be
1139 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
1140 kernels. Previously, lconf would still use the ext3 name
1141 when trying to load modules. Now, it will correctly use
1142 ext3 on 2.4 and ldiskfs on 2.6.
1144 Severity : enhancement
1145 Description: The default stripe count has been changed to 1
1146 Details : The interpretation of the default stripe count (0, to lfs
1147 or lmc) has been changed to mean striping across a single
1148 OST, rather than all available. For general usage we have
1149 found a stripe count of 1 or 2 works best.
1151 Severity : enhancement
1152 Description: Add support for compiling against Cray portals.
1153 Details : Conditional compiling for some areas that are different
1157 Frequency : occasional
1158 Bugzilla : 6409, 6834
1159 Description: Creating files with an explicit stripe count may lead to
1160 a failed assertion on the MDS
1161 Details : If some OSTs are full or unavailable, creating files may
1162 trigger a failed assertion on the MDS. Now, Lustre will
1163 try to use other servers or return an error to the
1167 Frequency : occasional
1169 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
1170 caused lock timeout during readahead (since 1.4.2).
1171 Details : Processes doing readahead might match a lock that hasn't been
1172 granted yet if there are overlapping and conflicting lock
1173 requests. The readahead process waits on ungranted lock
1174 (original lock is CBPENDING), while OST waits for that process
1175 to cancel CBPENDING read lock and eventually evicts client.
1177 Severity : enhancement
1179 Description: Initial enabling of flock support for clients
1180 Details : Implements fcntl advisory locking and file status functions.
1181 This feature is provided as an optional mount flag (default
1182 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
1183 locking are implemented yet, and those that are are not guaranteed
1184 to be completely correct in production environments.
1185 mount -t lustre -o [flock|noflock] ...
1188 Frequency : occasional
1190 Description: OSTs running 2.4 kernels but with extents enabled might trip an
1191 assertion in the ext3 JBD (journaling) layer.
1192 Details : The b_committed_data struct is protected by the big kernel lock
1193 in 2.4 kernels, serializing journal_commit_transaction() and
1194 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
1195 access to this struct. In 2.6 kernels there is finer grained
1196 locking to improve SMP performance of the JBD layer.
1200 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
1202 -----------------------------------------------------------------------------
1204 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
1206 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
1207 versions of Lustre. Please update all servers and clients to
1208 version 1.4.2 or later at the same time. You must also run
1209 "lconf --write-conf {config}.xml" on the MDS while it is stopped
1210 to update the configuration logs.
1212 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
1213 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
1214 - swab llog records of type '0' so we get proper header size/idx (5861)
1215 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
1216 - fix rename of one directory over another leaking an inode (5953)
1217 - avoid SetPageDirty on 2.6 (5981)
1218 - don't re-add just-being-destroyed locks to the waiting list (5653)
1219 - when creating new directories, inherit the parent's custom
1220 striping settings if present parent (3048)
1221 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
1222 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
1223 - don't hold client locks on temporary worklist from l_lru (5666)
1224 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
1225 - automatically evict dead clients (5921)
1226 - Update file size properly in create+truncate+fstat case (6196)
1227 - Do not unhash mountpoint dentries, do not allow removal of
1229 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
1230 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
1231 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
1232 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
1233 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
1234 - Correctly handle reads of files with no objects (6243)
1235 - lctl recover will also mark a device active if deactivate used (5933)
1237 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
1238 - allow --write-conf on an MDS with different nettype than client (5619)
1239 - don't write config llogs to MDS for mounts not from that MDS (5617)
1240 - lconf should create multiple TCP connections from a client (5201)
1241 - init scripts are now turned off by default; run chkconfig --on
1242 lustre and chkconfig --on lustrefs to use them
1243 - upcalls are no longer needed for clients to recover to failover
1245 - add --abort-recovery option to lconf to abort recovery on device
1247 - add support for an arbitrary number of OSTs (3026)
1248 - Quota support protocol changes.
1249 - forward compatibility changes to wire structs (6007)
1250 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
1251 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
1252 - improved "lustre status" script
1253 - initialize blocksize for non-regular files (6062)
1254 - added --disable-server and --disable-client configure options (5782)
1255 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
1256 - Vanilla 2.4.29 support
1257 - increase maximum number of obd devices to 520 (6242)
1258 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
1259 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
1260 SLES 9 2.6.5 kernels
1261 - limit stripes per file to 160 (the maximum EA size) (6093)
1263 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
1266 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
1267 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
1268 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
1269 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
1270 - small ext3 extents cleanups and fixes (5733)
1271 - improved mballoc code, several small races and bugs fixed (5733, 5638)
1272 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
1273 - avoid needless client->OST connect, fix handle mismatch (5317)
1274 - fix DLM error path that led to out-of-sync client, long delays (5779)
1275 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
1276 - fix several locking issues related to i_size (5492,5624,5654,5672)
1277 - don't move pending lock onto export if it is already evicted (5683)
1278 - fix kernel oops when creating .foo in unlinked directory (5548)
1279 - fix deadlock in obdfilter statistics vs. object create (5811)
1280 - use time_{before,after} to avoid timer jiffies wrap (5882)
1281 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
1282 - Do not leak request if server was not able to process it (5154)
1283 - If mds_open unable to find parent dir, make that negative lookup(5154)
1284 - don't create new directories with extent-mapping (5909, 5936)
1286 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
1287 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
1288 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
1289 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
1290 - support for RHEL4 kernel on i686 (5773)
1291 - provide error messages when incompatible logs are encountered (5898)
1293 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
1294 * version 1.4.0.10 (1.4.1 release candidate 1)
1296 - don't keep a lock reference when lock is not granted (4238)
1297 - unsafe list practices (rarely) led to infinite eviction loop (4908)
1298 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
1299 - drop import inflight refcount on signal_completed_replay error (5255)
1300 - unlock page after async write error during send (3677)
1301 - handle missing objects in filter_preprw_read properly (5265)
1302 - no transno return for symlink open, don't save no-trasno open (3440)
1303 - don't try to complete elan receive that already failed (4012)
1304 - free RPC server reply state on error (5406)
1305 - clean up thread from ptlrpc_start_thread() on error (5160)
1306 - readahead could read extra page into cache that wasn't ejected (5388)
1307 - prevent races in class_attach/setup/cleanup/detach (5260)
1308 - don't dereference de->d_inode after l_dput of de (5458)
1309 - use "int" for stripe value returned from lock_to_stripe (5544)
1310 - mballoc allocation and error-checking fixes in 2.6 (5504)
1311 - block device patches to fix I/O request sizes in 2.6 (5482)
1312 - look up hostnames for IB nals (5602)
1313 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
1314 - don't start multiple acceptors for the same port (5277)
1315 - fix incorrect LASSERT in mds_getattr_name (5635)
1316 - export a proc file for general "ping" checking (5628)
1317 - fix "lfs check" to not block when the MDS is down (5628)
1319 - service request history (4965)
1320 - put {ll,lov,osc}_async_page structs in a single slab (4699)
1321 - create an "evict_client" /proc entry on OSTs, like the MDS has
1322 - fix mount usage message, return errors per mount(8) (5168)
1323 - change grep [] to grep "[]" in tests so they work in more UMLs
1324 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
1325 - remove ancient LOV_MAGIC_V0 EA support (5047)
1326 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
1327 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
1328 - allow readahead allocations to fail when low on memory (5383)
1329 - mmap locking landed again, after considerable improvement (2828)
1330 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
1332 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
1335 - send OST transaction number in read/write reply to free req (4966)
1336 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
1337 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
1338 - the watchdog thread now runs as interruptible (5246)
1339 - flock/lockf fixes (but it's still disabled, pending 5135)
1340 - don't use EXT3 constants in llite code (5094)
1341 - memory shortage at startup could cause assertion (5176)
1343 - reorganization of lov code
1344 - single portals codebase
1346 - add extents/mballoc support (5025)
1347 - direct I/O reads in the obdfilter (4048)
1348 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
1350 tbd Cluster File Systems, Inc. <info@clusterfs.com>
1353 - send OST transaction number in read/write reply to free req (4966)
1354 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
1355 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
1356 - the watchdog thread now runs as interruptible (5246)
1357 - handle missing objects in filter_preprw_read properly (5265)
1358 - unsafe list practices (rarely) led to infinite eviction loop (4908)
1359 - drop import inflight refcount on signal_completed_replay error (5255)
1360 - unlock page after async write error during send (3677)
1361 - return original error code on reconstructed replies (3761)
1362 - no transno return for symlink open, don't save no-trasno open (3440)
1364 - add pid to ldlm debugging output (4922)
1365 - bump the watchdog timeouts -- we can't handle 30sec yet
1366 - extra debugging for orphan dentry/inode bug (5259)
1368 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
1371 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
1372 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
1373 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
1374 - fix NULL dereference in /proc/sys/portals/routes (4827)
1375 - allow failed mdc_close() operations to be interrupted (4561)
1376 - stop precreate on OST before MDS would time out on it (4778)
1377 - don't send partial-page writes before EOF from client (4410)
1378 - discard client grant for sub-page writes on large-page clients (4520)
1379 - don't free dentries not owned by NFS code, check generation (4806)
1380 - fix lsm leak if mds_create_objects() fails (4801)
1381 - limit debug_daemon file size, always print CERROR messages (4789)
1382 - use transno after validating reply (3892)
1383 - process timed out requests if import state changes (3754)
1384 - update mtime on OST during writes, return in glimpse (4829)
1385 - add mkfsoptions to LDAP (4679)
1386 - use ->max_readahead method instead of zapping global ra (5039)
1387 - don't interrupt __l_wait_event() during strace
1389 - add software watchdogs to catch hung threads quickly (4941)
1390 - make lustrefs init script start after nfs is mounted
1391 - fix CWARN/ERROR duplication (4930)
1392 - return async write errors to application if possible (2248)
1393 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
1394 - print NAL number in %x format (4645)
1395 - update barely-supported suse-2.4.21-171 series (4842)
1396 - support for sles 9 %post scripts
1397 - support for building 2.6 kernel-source packages
1398 - support for sles km_* packages
1400 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
1403 - ignore -ENOENT errors in osc_destroy (3639)
1404 - notify osc create thread that OSC is being cleaned up (4600)
1405 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
1406 - reconstruct ost_handle() like mds_handle() (4657)
1407 - create a new thread to do import eviction to avoid deadlock (3969)
1408 - let lconf resolve symlinked-to devices (4629)
1409 - don't unlink "objects" from directory with default EA (4554)
1410 - hold socknal file ref over connect in case target is down (4394)
1411 - allow more than 32000 subdirectories in a single directory (3244)
1412 - fix blocks count for O_DIRECT writes (3751)
1413 - OST returns ENOSPC from object create when no space left (4539)
1414 - don't send truncate RPC if file size isn't changing (4410)
1415 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
1416 - bind to privileged port in socknal and tcpnal (3689)
1418 - rate limit CERROR/CWARN console message to avoid overload (4519)
1419 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
1420 - basic mmap support (3918)
1421 - kernel patch series update from b1_4 (4711)
1423 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
1426 - avoid crash during MDS cleanup with OST shut down (2775)
1427 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
1428 - don't use bad inodes on the MDS (3744)
1429 - dynamic object preallocation to improve recovery speed (4236)
1430 - don't hold spinlock over lock dumping or change debug flags (4401)
1431 - don't zero obd_dev when it is force cleaned (3651)
1432 - print grants to console if they go negative (4431)
1433 - "lctl deactivate" will stop automatic recovery attempts (3406)
1434 - look for existing locks in ldlm_handle_enqueue() (3764)
1435 - don't resolve lock handle twice in recovery avoiding race (4401)
1436 - revalidate should check working dir is a directory (4134)
1438 - don't always mark "slow" obdfilter messages as errors (4418)
1440 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
1443 - don't close LustreDB during write_conf until it is done (3860)
1444 - fix typo in lconf for_each_profile (3821)
1445 - allow dumping logs from multiple threads at one time (3820)
1446 - don't allow multiple threads in OSC recovery (3812)
1447 - fix debug_size parameters (3864)
1448 - fix mds_postrecov to initialize import for llog ctxt (3121)
1449 - replace config semaphore with spinlock (3306)
1450 - be sure to send a reply for a CANCEL rpc with bad export (3863)
1451 - don't allow enqueue to complete on a destroyed export (3822)
1452 - down write_lock before checking llog header bitmap (3825)
1453 - recover from lock replay timeout (3764)
1454 - up llog sem before sending rpc (3652)
1455 - reduce ns lock hold times when setting kms (3267)
1456 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
1457 - fix NULL deref and obd_dev leak on setup error (3312)
1458 - replace some LBUG about llog ops with error handling (3841)
1459 - don't match INVALID dentries from d_lookup and spin (3784)
1460 - hold dcache_lock while marking dentries INVALID and hashing (4255)
1461 - fix invalid assertion in ptlrpc_set_wait (3880)
1463 - add libwrap support for the TCP acceptor (3996)
1464 - add /proc/sys/portals/routes for non-root route listing (3994)
1465 - allow setting MDS UUID in .xml (2580)
1466 - print the stack of a process that LBUGs (4228)
1468 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
1471 - don't cleanup request in ll_file_open() on failed MDS open (3430)
1472 - make sure to unset replay flag from failed open requests (3440)
1473 - if default stripe count is 0, use OST count for inode size (3636)
1474 - update parent mtime/ctime on client for create/unlink (2611)
1475 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
1476 - free recovery state on server during a forced cleanup (3571)
1477 - unregister_reply for resent reqs (3063)
1478 - loop back devices mounting and status check on 2.6 (3563)
1479 - fix resource-creation race that can provoke i_size == 0 (3513)
1480 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
1481 - more debugging for page-accounting assertion (3746)
1482 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
1483 - avoid deadlock after precreation failure (3758)
1484 - fix race and lock order deadlock in orphan handling (3450, 3750)
1485 - add validity checks when grabbing inodes from l_ast_data (3599)
1487 - add /proc/.../recovery_status to obdfilter (3428)
1488 - lightweight CDEBUG infrastructure, debug daemon (3668)
1489 - change default OSC RPC parameters to be better on small clusters
1490 - turn off OST read cache for files smaller than 32MB
1491 - install man pages and include them in rpms (3100)
1492 - add new init script for (un)mounting lustre filesystems (2593)
1493 - run chkconfig in %post for init scripts (3701)
1494 - drop scimac NAL (unmaintained)
1496 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
1499 - clean kiobufs before and after use (3485)
1500 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
1501 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
1502 - change init script to fail more clearly if not run as root (1528)
1503 - allow clients to reconnect during replay (1742)
1504 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
1505 - don't do DNS lookups on NIDs too small for IP addresses (3442)
1506 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
1507 - fix cond_resched (3554)
1508 - only evict unfinished clients after recovery (3515)
1509 - allow bulk resend, prevent data loss (3570)
1510 - dynamic ptlrpc request buffer allocation (2102)
1511 - don't allow unlinking open directory if it isn't empty (2904)
1512 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
1513 - remove extraneous obd dereference causing LASSERT failure (3334)
1514 - don't use get_cycles() when creating temp. files on the mds (3156)
1515 - hold i_sem when setting i_size in ll_extent_lock() (3564)
1516 - handle EEXIST for set-stripe, set proper directory name (3336)
1518 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
1519 - fix ksocknal_fmb_callback() error messages (2918)
1521 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
1524 - don't copy lvb into (possibly NULL) reply on error (2983)
1525 - don't deref dentry after dput, don't free lvb on error (2922)
1526 - use the kms to determine writeback rpc length (2947)
1527 - increment oti_logcookies when osc is inactive (2948)
1528 - update client's i_blocks count via lvb messages (2543)
1529 - handle intent open/close of special files properly (1557)
1530 - mount MDS with errors=remount-ro, like obdfilter (2009)
1531 - initialize lock handle to avoid ASSERT on error cleanup (3057)
1532 - don't use cancelling-locks' kms values (2947)
1533 - use highest lock extent for kms, not last one (2925)
1534 - don't dereference ERR_PTR() dentry in error handling path (3107)
1535 - fix thread race in portals_debug_dumplog() (3122)
1536 - create lprocfs device entries at setup instead of at attach (1519)
1537 - common AST error handler, don't evict client on completion race (3145)
1538 - zero nameidata in detach_mnt in 2.6 (3118)
1539 - verify d_inode after revalidate_special is valid in 2.6 (3116)
1540 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
1541 - initialize RPC timeout timer earlier for 2.6 (3219)
1542 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
1543 - print nal/nid for unknown nid (3258)
1544 - additional checks for oscc recovery before doing precreate (3284)
1545 - fix ll_extent_lock() error return code for 64-bit systems (3043)
1546 - don't crash in mdc_close for bad permissions on open (3285)
1547 - zero i_rdev for non-device files (3147)
1548 - clear page->private before handing to FS, better assertion (3119)
1549 - tune the read pipeline (3236)
1550 - fix incorrect decref of invalidated dentry (2350)
1551 - provide read-ahead stats and refine rpc in flight stats (3328)
1552 - don't hold journal transaction open across create RPC (3313)
1553 - update atime on MDS at close time (3265)
1554 - close LDAP connection when recovering to avoid server load (3315)
1555 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
1556 - don't leak open file on MDS after open resend (3325)
1557 - serialize filter_precreate and filter_destroy_precreated (3329)
1558 - loop device shouldn't call sync_dev() for nul device (3092)
1559 - clear page cache after eviction (2766)
1560 - resynchronize MDS->OST in background (2824)
1561 - refuse to mount the same filesystem twice on same mountpoint (3394)
1562 - allow llmount to create routes for mounting behind routers (3320)
1563 - push lock cancellation to blocking thread for glimpse ASTs (3409)
1564 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
1565 - fix rare problem with rename on htree directories (3417)
1567 - allow default OST striping configuration per directory (1414)
1568 - fix compilation for qswnal for 2.6 kernels (3125)
1569 - increase maximum number of MDS request buffers for large systems
1570 - change liblustreapi to be useful for external progs like lfsck (3098)
1571 - increase local configuration timeout for slow disks (3353)
1572 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
1574 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
1577 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
1578 - don't overwrite extent policy data in reply if lock was blocked (2901)
1579 - drop filter export grants atomically with removal from device (2663)
1580 - del obd_self_export from work_list in class_disconnect_exports (2908)
1581 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
1582 - swab reply message in mdc_close, other PPC fixes (2464)
1583 - fix destroying of named logs (2325)
1584 - overwrite old logs when running lconf --write_conf (2264)
1585 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
1586 - fix race in target_handle_connect (2898)
1587 - mds_reint_create() should take same inode create lock (2926)
1588 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
1589 - don't close files for self_export to avoid uninitialized obd (2936)
1590 - allow MDS with the same name as client node (2939)
1591 - hold dentry reference for closed log files for unlink (2325)
1592 - reserve space for all logs during transactions (2059)
1593 - don't evict page beyond end of stripe extent (2925)
1594 - don't oops on a deleted current working directory (2399)
1595 - handle hard links to targets without a parent properly (2517)
1596 - don't dereference NULL lock when racing during eviction (2867)
1597 - don't grow lock extents when lots of conflicting locks (2919)
1599 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
1602 - account for cache space usage on clients to avoid data loss (974)
1603 - lfsck support in lustre kernel code (2349)
1604 - reduce journal credits needed for BRW writes (2370)
1605 - orphan handling to avoid losing space on client/server crashes
1606 - ptlrpcd can be blocked, stopping ALL progress (2477)
1607 - use lock value blocks to assist in proper KMS, faster stat (1021)
1608 - takes i_sem instead of DLM locks internally on obdfilter (2720)
1609 - recovery for initial connections (2355)
1610 - fixes for mds_cleanup_orphans (1934)
1611 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
1612 - block all file creations until orphan recovery completes (1901)
1613 - client remove rq_connection from request struct (2423)
1614 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
1615 - recovery timer race (2670)
1616 - mdc_close recovey bug (2532)
1617 - ptlrpc cleanup bug (2710)
1618 - mds timeout on local locks (2588)
1619 - namespace lock held during RPCs (2431)
1620 - handle interrupted sync write properly (2503)
1621 - don't try to handle a message that hasn't been replied to (2699)
1622 - client assert failure during cleanup after abort recovery (2701)
1623 - leak mdc device after failed mount (2712)
1624 - ptlrpc_check_set allows timedout requests to complete (2714)
1625 - wait for inflight reqs when ptlrpcd finishes (2710)
1626 - make sure unregistered services are removed from the srv_list
1627 - reset bulk XID's when resending them (caught by 1138 test)
1628 - unregister_bulk after timeout
1629 - fix lconf error (2694)
1630 - handle write after unfinished setstripe, stripe-only getstripe (2388)
1631 - readahead locks pages, leaves pending causing memory pressure (2673)
1632 - increase OST request buffers to 4096 on large machines (2729)
1633 - fix up permission of existing directories in simple_mkdir (2661)
1634 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
1635 - don't assign transno to errored transactions (2742)
1636 - don't delete objects on OST if given a bogus objid from MDS (2751)
1637 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
1638 - if rq_no_resend, then timeout request after recovery (2432)
1639 - fix MDS llog_logid record size, 64-bit array alignment (2733)
1640 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
1641 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
1642 - MDS recovery shouldn't delete objects that it has given out (2730)
1643 - if enqueue arrives after completion, don't clobber LVB (2819)
1644 - don't unlock pages twice when trigger_group_io returns error (2814)
1645 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
1646 - don't write pages to disk if there was an error (1450)
1647 - don't ping imports that have recovery disabled (2676)
1648 - take buffered bytes into account when balancing socknal conn (2817)
1649 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
1650 - reconnect unlink llog connection after MDS reconnects to OST (2816)
1651 - remove little-endian swabbing of llog records (1987)
1652 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
1653 - retry reposting request buffers if they fail (1191)
1654 - grow extent at grant time to avoid granting a revoked lock (2809)
1655 - lock revoke doesn't evict page if covered by a second lock (2765)
1656 - disable VM readahead to avoid reading outside lock extents (2805)
1658 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
1659 - updated kernel patches for hp-2.4.20 kernel (2681)
1661 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
1664 - fix truncated write corruption (2366)
1665 - fix for failed assertion in iopen_connect_dentry (1792,2517)
1667 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
1668 - ensure len(uuid) < 37 in lmc (1171)
1669 - fix ia64 OOPS in llog_test (2255)
1670 - zero end of page at obdfilter for partial page writes (2648)
1671 - don't leave stale dentries around after renames (bug 2428)
1672 - fix timeouts when evicting a client with a single lock held (2642)
1673 - set deadline for the initial HELLO message to drain (2634)
1674 - print out dotted-quad IP addresses in the socknal (2302)
1676 - additional debugging for MDS client eviction problem (2443)
1677 - fix mkfsoptions support for osts (2603, 2604)
1679 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
1682 - add series for the vanilla 2.6.0 kernel
1683 - add series for the vanilla 2.4.24 kernel
1684 - add series for a cray x86/64 UL kernel drop
1685 - fix xattr patches for the vanilla 2.4.19 series
1687 - generate true UUIDs in lmc (1171)
1688 - have portals stack dumping break in UML (2466)
1689 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
1690 - allocate with _NOFS in ldlm to avoid deadlock (1933)
1691 - wake callback waiting threads on client eviction (2460)
1692 - Add --ptldebug and --subsystem to lmc (1719)
1693 - update assertion to allow safe interrupt allocation
1694 - set rq_no_resend for cancel requests (2432)
1695 - recalculate ptlrpcd timeout after resend (2494)
1696 - call vfs_rmdir when removing pending directories (2368)
1697 - fix renaming a file to itself (2429)
1698 - lmc creates a default one-stripe lov (2454)
1699 - expand procfs space to handle large clusters (2326)
1700 - increase UML stack to avoid overflow
1701 - update lconf's list of debug and subsystem masks
1702 - fix lfs find --obd (2510)
1703 - /proc tunable for disabling filter read caching (2591)
1704 - stop rpm packages from altering slapd.conf (2301)
1705 - disable nagle in the socknal under 0conf (2578)
1706 - choose mds inode size based on stripe count (2572)
1707 - fix kernel-source rpm problems (2516)
1709 - add --disable-doc to avoid pdf generation (2421)
1710 - update documentation, tests, type-os, comments
1711 - avoid format warnings on ia64
1712 - remove the TOE NAL
1713 - tiny code cleanups by removing unused fields
1715 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
1718 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
1719 - remove the most blatant lies from BUILDING file (2371)
1720 - change default debug level to reasonable production setting
1721 - reduce client side cache size to reduce cache flush time
1722 - reduce max RPCs in flight to avoid unnecessary file fragmentation
1723 - make TCP zerocopy and pinger support enabled by default (2476)
1724 - sync writes completed after process exits caused crashes (2319)
1725 - maintain correct mount count on the MDS (2356)
1726 - backout 1557, because 2316 wasn't really fixed
1727 - better file I/O statistics gathering in /proc
1728 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
1729 - another kernel patch to fix zero-copy TCP function export
1730 - don't take duplicate lock when processing re-sent getattr (2420)
1731 - lctl uses obd_self_export instead of creating new conn (2353)
1732 - MDS/OST recovery case which requires object creation asserted (2425)
1733 - move lfs from /usr/sbin to /usr/bin in packages
1734 - fix race between mds_client_add and mds_client_free (2417)
1735 - use kmalloc instead of slabs in portals (2430)
1736 - don't create duplicate records when a failover MDS is present (2442)
1737 - remove unnecessary mount age check (2332)
1738 - don't remove directory inodes from locks prematurely (2451)
1739 - don't break if MDS service name is the same as hostname (2103)
1740 - fix races in client write RPC generation when cache full (2482)
1742 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
1745 - remove now-unused request->rq_obd (278)
1746 - if an allocation fails, print out how much memory we've used (1933)
1747 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
1748 - add the "configurable stack size" patch to most series files (1256)
1749 - ability to write large log records, for 100+ OST configs (2306)
1750 - fix NULL deref when filter_prep fails (2314)
1751 - fix operator precedence error in filter_sync
1752 - dynamic allocation of socknal TX descriptors (2315)
1753 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
1754 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
1755 - ensure that CWARN messages in Portals always get to the syslog
1756 - __init/__exit are not for prototype decls (ldlm_init/exit)
1757 - x86-64 compile warning fixes
1758 - fix gateway LMC keyword conflict (2318)
1759 - fix MDS lock inversions in getattr/reint paths (1844)
1760 - fix a rare lock re-ordering bug, which caused deadlock (2322)
1761 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
1762 - DLM race condition prevented some lock evictions (2328)
1763 - ENOMEM detection and retry on socknal sends (2230)
1764 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
1765 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
1766 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
1767 - better error messages when a client is rejected during recovery (1505)
1768 - avoid cancelling locks which were never granted, after failure (2330)
1769 - fix i_sem/journal inversion in mds_client_add (2333)
1770 - fix truncate/getattr lock cycle deadlock (2334)
1771 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
1772 - fix two rare exit paths which could leak an l_lock() ref (2321)
1773 - fencepost error in MDS/OST orphan recovery (2226)
1774 - make log record alignment 8 bytes (1988)
1775 - lstripe now fails when requested offset > ost_count (2237)
1776 - ensure that all kernel series have a complete list.h (1607)
1777 - fix crashes in special-file operations (2316)
1778 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
1780 - allow configurable automake binary, for testing new versions
1781 - small update to the lfs documentation
1783 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
1785 * fix negative export reference count in fsfilt_sync (2312)
1787 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
1788 * release candidate 0.9.1
1790 - orphans are moved into the PENDING directory for possible recovery
1791 - replayed opens now open by fid for orphan/rename safety (1042)
1792 - last close of an orphan inode generates a transno (683)
1793 - chdir() and mount() now pin the directory entry (1020)
1794 - avoid CERROR in normal ll_setattr_raw() error case (1500)
1795 - discard very old requests without processing them (1502)
1796 - remove some common, well-understood CERRORs (1505)
1797 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
1798 - clear "grant" flags in OST replies until OST grant code lands (1644)
1799 - fix read performance by not clobbering i_blksize on client (1598)
1800 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
1801 - make LRU size a /proc tunable, clears locks when reduced (707)
1802 - fix some lprocfs rot that prevented ptlbd from loading (1732)
1803 - server locks take references on exports now (1558)
1804 - build fixes for 2.4.20-rh trees (1663)
1805 - return an error from lov_create if all OSCs are inactive (1751)
1806 - fix import levels when a reconnect happens without a timeout (1597)
1807 - exit early from mds_open if we get a lookup error (1749)
1808 - partial page read at EOF wouldn't wait for disk before sending (1642)
1809 - avoid NULL deref in obdfilter when reading page past EOF (1592)
1810 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
1811 - fix LBUG in ll_it_open_error with rc = -2 (1861)
1812 - write/truncate lock inversion (1639)
1813 - Don't auto-load obdclass, portals modules during cleanup (1495)
1814 - fix timestamps from jumping to "now" (1763)
1815 - extra journal assertions (1648)
1816 - add an extra multiunlink test (1771)
1817 - fix read_record/write_record API (1776)
1818 - fix leak of offset_extent, possible incorrect i_size later (1772)
1819 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
1820 - Debugging for the kqswnal_get_idle_tx problems (1820)
1821 - Allow recovery to be attempted multiple times (1536)
1822 - Write out MDS last_rcvd file after it is first created (1600)
1823 - Fix tx_descriptor leak in failed transmit situations (1827)
1824 - ext3 journaling fixes for assertion failure after IO error (1871)
1825 - class_export_put() on freed export after completion AST error (1896)
1826 - Fix revalidate looping in VFS (1322)
1827 - Don't access a freed export during MDS_REINT timeout (1521)
1828 - Add open-unlink recovery support on the MDS (1673,1764)
1829 - Return an error if no MDS data was read from last_rcvd (1946)
1830 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
1831 - Don't setup a disk device that doesn't match exported UUID (317)
1832 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
1833 - avoid committing NULL handle in force close
1834 - local.sh is now a one-stripe LOV configuration
1835 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
1836 - fix ext3 htree duplicate directory entry corruption (1516)
1837 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
1838 - update to new LOV EA format (2097)
1839 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
1841 2003-06-15 Phil Schwan <phil@clusterfs.com>
1844 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
1845 - per-import/export recovery handling (958, 931, 959)
1846 - multiple last-rcvd slots, for serving multiple FSes (949)
1847 - connections are again shared between multiple imp/exports (963, 964)
1848 - "umount -f" would hang if any requests needed to be sent (393, 978)
1849 - avoid pinning large req buffer by copying for queued messages (989)
1850 - add "uuid" to "lctl device" command to help upcalls (991)
1851 - "open" RPCs with transnos would confuse recovery counters (1037)
1852 - do proper endian conversion of all wire messages (288, 340, 891)
1853 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
1854 - call waiting locks callback from LDLM recovery thread (1127, 1151)
1855 - fix ptlrpc_connection leak in target_handle_connect (1174)
1856 - fix import refcounting bug in OST and MDS cleanup (1134)
1857 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
1858 - fix very unlikely obd_types race condition (501)
1859 - remove osc_open hack for echo_client (1187)
1860 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
1861 - a failure in read_inode2 leads to deadlock (1139)
1862 - cancel ack-locks as soon as transaction is committed (1072)
1863 - fix major leaks and crashes in the bulk I/O path (937, 1057)
1864 - make sure to commitrw after any preprw to avoid deadlock (1162)
1865 - failing to execute a file in a lustre FS would lock inode (1203)
1866 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
1867 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
1868 - obd_uuid2tgt would walk off the end of the list (1255)
1869 - on IA64 the osc would give portals incorrect bulk size (1258)
1870 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
1871 - fix lock inversion caused by new llite matching code (1282)
1872 - limit the number of dirty pages on a client to 10MB (1286)
1873 - timed out locks were not being corrected cancelled (1289)
1874 - fix O_DIRECT above 4GB on IA-32 (1292)
1875 * major user-visible changes
1876 - fail out/fail over policy now controlled by the upcall (993)
1878 - add OBD_PING to check server availability and failure (954)
1879 - lustre messages are now sent in sending host order (288, 340, 891)
1880 - add eadatalen to MDS getattr reply (340)
1881 - OST read replies may contain second buffer, with per-page status (593)
1883 2003-03-11 Phil Schwan <phil@clusterfs.com>
1886 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
1887 - failed open()s could cause deadlock; fixed (867, 869)
1888 - stop cancelling OST locks when files are closed (481)
1889 - overlapping XID spaces caused network corruption (851, 853)
1890 - fix unsafe fsfilt counter arithmetic; change to atomic_t
1891 - setattr_raw added, to do single-RPC, server-side setattrs
1892 - lmc/lconf syntax change for OST UUIDs
1893 - fix crashy race condition between ptlrpc_free_req and osc_close
1894 - don't use request in mdc_enqueue if we hit a timeout (889)
1895 - don't set the inode i_size for regular files from the MDS (896)
1896 - handle out of order completion AST (842)
1897 - don't LBUG if a lock request times out after receiving AST (913)
1898 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
1899 - if a bad lock AST arrives, send an error instead of dropping entirely
1900 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
1901 - fix leak in bulk IO when only partially completed (899, 900, 926)
1902 - fix O_DIRECT for ia64 (55)
1903 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
1904 - C-z after timeout could hang a process forever; fixed (977)
1906 - client-side I/O cache (678, 924, 929, 941, 970)
1908 - READPAGE and SETATTRs which don't take server-side locks get
1911 2003-02-11 Phil Schwan <phil@clusterfs.com>
1914 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
1915 - Fix fsfilt_extN_readpage() to read a full page of directory
1916 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
1917 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
1918 - Fix imbalanced LOV object allocation and out-of-bound access (469)
1919 - Most intent operations were removed, in favour of a new RPC mode
1920 that does a single RPC to the server and bypasses most of the VFS
1921 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
1922 - Aggressively cancel local locks on DLM servers
1923 - mds_reint_unlink sends EA to the client if it's the last nlink.
1924 client uses that EA to unlink OST objects.
1925 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
1926 - recursive symlinks were fixed (439)
1927 - fixed NULL deref in DEBUG_REQ
1928 - filter_update_lastobjid no longer calls sync, which annoyed extN
1929 - fixed multi-client small-writes to a single file problem (445)
1930 - fixed mtime updates during file writes (607)
1931 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
1932 - fixed bug in obd_brw_read/write() (under guise of testing 367)
1933 - fixed Linux OST size reporting problem (444, 656)
1934 - OST now updates object mtime with writes or setattr (607, 619)
1935 - client verifies file size before zeroing page past EOF (445)
1936 - OST now writes last allocated objid to disk with allocation (108)
1937 - LOV on echo now works (409)
1939 - mds_reint_unlink sends a new buffer, with the EA included. this
1940 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
1941 set if a regular file was being unlinked, and it was the last link
1942 - use PtlGet from the target for bulk writes (315)
1943 - OST now updates object mtime with writes or setattr (607, 619)
1944 - LDLM now has a grant-time callback to revalidate locked items, if
1946 - Many MDS operations were reorganized to combat race conditions
1948 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
1949 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
1951 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
1954 - Fully reactivate OST imports after reconnection (512, others)
1955 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
1956 - More graceful error handling for truncating on dead OST (515)
1957 - Don't error out unless we're actually accessing dead stripes (474)
1958 - Fix garbage sizes when stripes are missing (410)
1959 - LRU counters were broken, causing constant lock purge (433, 432)
1960 - garbage on read from stripes with failed OSTs (441)
1961 - mark OSCs as active before reconnecting during recovery (438)
1962 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
1963 - lfind did not preserve OST order in output (443)
1964 - symlinks cause hung clients, incorrect data (439)
1965 - stop dereferencing request after dropping refcount (457)
1966 - don't LASSERT(spin_is_locked) on non-SMP (455)
1967 - fixes for many rename() bugs
1968 - fstat didn't correctly synchronize attributes (399)
1969 - server must handle lock cancellation during blocking AST prep (487)
1970 - bulk descriptors were free()d too soon (511)
1971 - fix paths in lconf, which would load incorrect modules (451, 507)
1972 - fix confusing lconf 'host not found' error message (386)
1973 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
1974 - fix race condition in mdc_blocking_ast() for inode access (526)
1975 - fix lov_unpackmd() unpacking wrong number of stripes (537)
1976 - fix lov_set_osc_active() marking wrong OSC inactive (440)
1977 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
1978 - fix multiple writes of stripe MD to MDS (358, maybe 519)
1979 - fix lstripe in several ways (kernel side) (527)
1980 - fix request leak in ldlm_cli_enqueue (262)
1981 - incorrect OSC was marked inactive after OST failure
1982 - call mds_fs_cleanup before unmounting filesystem (524)
1983 - fix races between taking ns_lock and ldlm_lock_change_resource
1984 - fix races updating LOV export open file list
1985 - fix lov_enqueue error path, avoid decref-ing bad lock handle
1986 - fix recovery NULL deref in ldlm_cli_cancel_unused
1987 - fix some DLM races by using new hash table for lock handles (419)
1988 - permit the client to specify desired inodes, at replay
1989 - duplicate requests when we queue them for replay reintegration
1990 - fix last_rcvd offset calculation
1991 - sync after each recovered transaction, so we always make progress
1992 - never, not always, ERESTART requests without transnos
1993 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
1995 - skip replay if the MDS says that the client is already connected
1996 - don't check for a recovery-enabled export to match lctl's UUID
1997 - don't INC_USE_COUNT for phantom exports
1998 - don't crash when cleaning up phantom exports (567)
1999 - don't double-finish or set replay data for errored mdc_open requests
2000 - abort requests when they time out, so we don't get old replies
2001 - send/receive replies for AST messages again
2002 - if the client says that it doesn't have the lock, cancel it on the
2004 - if we timeout during I/O, don't try to cancel an in-use lock; instead
2005 mark it as destroyed, it will all work out when decref is called
2006 - fix module use counts (22, 581)
2008 - ASTs now expect a reply (server cancels lock on error reply)
2010 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
2013 - fix many simultaneous client startup (392)
2014 - fix dentry->d_it clobbering
2015 - credentials weren't being shipped for readdir/getattr operations
2016 - remove invalid assertions triggered during some concurrent MD
2018 - proper Lustre versions added (336, 389)
2019 - fix memory leak for create error case (398)
2020 - fix LOV locking bug that would get cli/srv out of sync
2021 - fix echo client over LOV (409)
2022 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
2023 - fix double-O_EXCL intent crash (424)
2024 - avoid sending multiple lock CANCELs (352)
2026 - MDS can do multi-client recovery (modulo bugs in new code)
2028 - many updates, edits, cleanups
2030 2002-11-18 Phil Schwan <phil@clusterfs.com>
2033 - fix null d_it dereference (346)
2034 - fix full OST/dbench hang (333)
2035 - fix permission problem with file removal (286)
2036 - fix removal of OSCs from LOV when they fail
2037 - fix NULL deref during bulk timeout (214)
2038 - fix problems related to multiple filesystems on one MDS (241)
2039 - fixed serious subtle metadata locking bugs
2040 - free locks on clients when inodes are removed due to memory
2042 - fix inode pointer in lock data (285)
2043 - partial support for multiple MDS on a single host (241)
2044 - data locks weren't cancelled at clear_inode time (290, 311)
2045 - intent locks could lead to unbounded lock growth (205)
2046 - added a maximum lock count, an LRU list, and a flusher
2047 - fix multiple rename (365)
2048 - properly abstracted the echo client
2049 - OSC locked 1 byte too many; fixed
2050 - rewrote brw callback code:
2051 - fixed recovery bugs related to LOVs (306)
2052 - fixed too-many-pages-in-one-write crash (191)
2053 - fixed (again) crash in sync_io_timeout (214)
2054 - probably fixed callback-related race (385)
2056 - Add capability to MDS protocol
2057 - LDLM cancellations and callbacks on different portals
2059 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
2062 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
2063 - defer open object destruction to close time (601981, 138)
2064 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
2065 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
2066 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
2067 - don't call obd_setattr in ll_file_release for destroyed objects
2068 * protocol change to lustre_msg: move |version| and add |flags|
2069 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
2070 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
2071 * added replay of create, unlink, link and rename operations during
2072 MDS failover; recovery should be much more robust now
2073 * remove failed OSCs from LOVs (only lov_create uses this so far)
2074 * the lustre-HOWTO was brought (more) up to date (582544)
2076 2002-10-23 Phil Schwan <phil@clusterfs.com>
2079 - in-use dentries weren't being reused properly (617851)
2080 - prevent multiple LDLM setup (599178)
2081 - fix LOV size calculations for truncate (617853)
2082 - fix client handling of MDS intent errors (POSIX)
2083 - fix permission bug in lovstripe.c test (624321)
2084 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
2085 - truncate past end of file could corrupt data
2086 - proper cleanup after timeouts, crashes, etc (592524, 550815)
2087 - a race in recovery could return ETIMEDOUT to apps (623947)
2088 - building outside the source directory was fixed
2089 * the lustre-HOWTO was brought (more) up to date (582544)
2090 * major progress was made on recovery functionality
2092 2002-10-10 Phil Schwan <phil@clusterfs.com>
2095 - recovery deadlock fix
2096 - rm -rf causes LBUG fix (617817)
2097 - file open by multiple tasks fix (618962)
2098 - directory permissions bugs (602707 and 620007)
2099 - journal_stop fixed with locking (611313)
2100 - O_APPEND failures resolved (618273, perhaps 614459)
2101 - lconf PATH fix (619770)
2102 - IA64 build fix (621450)
2103 - RPC buffer sizes scale with amount of memory
2105 2002-10-01 Phil Schwan <phil@clusterfs.com>
2108 - locks would be cancelled without throwing away data pages,
2109 resulting in inconsistent data (605627)
2110 - inode attributes were not always being refreshed (605627, 612449)
2111 - lconf now continues to cleanup after lctl reports an error
2112 - MDS now enforces user permissions (602707)
2113 - lprocfs cleanup fixed, but not yet enabled (614157)
2114 - fixed infinite server hang, should a client not respond to an AST
2115 - avoid going into recovery if user calls readlink() with a buffer
2116 that's too small (613941)
2117 - AST RPCs no longer require replies (614867) -- this may be changed
2118 - don't crash server if client sends an IOV that's too big (611336)
2119 - fixed lock conversion deadlock (611892)
2120 - fixed the following of symlinks (614622)
2121 * recovery: the server can remove locks from a client that dies, other
2122 clients can make progress
2123 * more extN patch fixes
2124 * compile-time configurable ptlrpc buffer allocations
2126 - collaborative read cache document
2127 - Lustre Lite Performance CDR document-in-progress
2129 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
2132 - fix typo in patch-2.4.18
2134 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
2137 - clear ptlrpc request each time in handle_incoming_request()
2138 - unlink of files now destroys the object on the OST
2140 2002-09-19 Peter Braam <braam@clusterfs.com>
2142 * add hard link support
2143 * change obdfile creation method
2144 * kernel patch changed
2146 2002-09-19 Peter Braam <braam@clusterfs.com>
2149 - stack overflow bug in extN fixed
2151 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
2153 * documentation updates
2154 - add man pages for config tools
2155 - update tests/README to describe testing with new config tools
2156 - finish metadata API descriptions
2157 * bug fixes and cleanups
2158 - statfs workaround for 16TB limit
2159 - LOV stripe allocation improved, can stripe on subset of OSTs
2160 - LOV file size/IO offset was wrong for files > 4GB in size
2161 - object EA data was being dropped, caused files to be unreadable
2162 - memory overflow with non-LOV OST caused memory corruption
2163 - fixed regression tests to work with new config tools, obdfilter
2164 - fixed bug when directory size became larger than 1 block
2165 - fixed bug (for single client case) when PWD was deleted
2166 - invalidate local directory pages when doing intent-based ops
2167 - avoid LDLM oops when lock callback contained bad data
2169 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
2171 * documentation updates
2172 * bug fixes and cleanups
2173 - configuration tools
2176 - 64-bit compile warnings
2177 - 64-bit internal statfs data
2179 * test_brw on persistent OST devices
2181 * lprocfs (disabled)
2183 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
2185 * documentation updates
2186 * bug fixes and cleanups
2187 * configuration tools
2189 2002-08-30 Peter J. Braam <braam@clusterfs.com>
2192 * many small fixes to 0_5_4
2193 * io/network handling
2194 * thinkos in MDS operations
2196 2002-08-24 Peter J. Braam <braam@clusterfs.com>
2199 * crucial basic fixes to 0.5.3
2200 * IOR, Iozone work over Elan
2203 2002-08-07 Phil Schwan <phil@clusterfs.com>
2204 * version 0_5_3, our first alpha
2205 * we use the new Portals iovs
2206 * documentation updates
2207 * bug fixes and cleanups
2208 * small changes in the DLM wire protocol
2210 2002-07-25 Peter J. Braam <braam@clusterfs.com>
2211 * version 0_5_1 with some initial stability,
2212 * locking on MD and file I/O.
2213 * documentation updates
2214 * several bug fixes since 0.5.0
2215 * small changes in wire protocol
2217 2002-07-18 Phil Schwan <phil@clusterfs.com>
2219 * delivered as Lustre Light Alpha
2220 * fixed a crash after handling invalid MDS requests
2221 * fixed directory pages for architectures with non-4k pages sizes
2223 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
2224 * release version v0_4_4
2225 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
2227 * Fixes a number of interruption problems with OST operations.
2228 * Update documentation for portals header changes
2229 * Move all wire protocol structs/defines to lustre_idl.h
2230 * Fixes symlink length bug.
2231 * Add tcpdump to repository.
2233 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
2234 * release version v0_4_3
2235 * Fixes statfs for inodes on extN.
2236 * Fixes bug in runtests which would delete /etc/hosts.
2237 * Use 64-bit object IDs wherever possible (not into VFS though)
2238 Remove ost_get_info, which is unused by lustre, and out of date.
2240 2002-07-03 Peter Braam <braam@clusterfs.com>
2241 * release version v0_4_2 Fixes a lookup error (type not passed)
2242 * move forward to head of Portals
2243 * move forward to latest Lustre kernel
2245 2002-06-25 Peter Braam <braam@clusterfs.com>
2246 * release version v0_4_1. Hopefully stable on single node use.