1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4 INTEROPERATE with older versions automatically. In many cases a
5 special upgrade step is needed. Please read the
6 user documentation before upgrading any part of a live system.
7 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
8 version will not interoperate with 1.6 betas before beta5 (1.5.95).
9 * WARNING: Lustre configuration and startup changes are required with
10 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
15 Severity : enhancement
17 Description: MountConf
18 Details : Lustre configuration is now managed via mkfs and mount
19 commands instead of lmc and lconf. New obd types (MGS, MGC)
20 are added for dynamic configuration management. See
21 https://mail.clusterfs.com/wikis/lustre/MountConf for
24 Severity : enhancement
26 Description: dynamic OST addition
27 Details : OSTs can now be added to a live filesystem
29 Severity : enhancement
31 Description: startup order invariance
32 Details : MDTs and OSTs can be started in any order. Clients only
33 require the MDT to complete startup.
35 Severity : enhancement
37 Description: parallel, asynchronous orphan cleanup
38 Details : orphan cleanup is now performed in separate threads for each
39 OST, allowing parallel non-blocking operation.
41 Severity : enhancement
43 Description: optimized stripe assignment
44 Details : stripe assignments are now made based on ost space available,
45 ost previous usage, and OSS previous usage, in order to try
46 to optimize storage space and networking resources.
48 Severity : enhancement
50 Description: Permanently set tunables
51 Details : All writable /proc/fs/lustre tunables can now be permanently
52 set on a per-server basis, at mkfs time or on a live
55 Severity : enhancement
57 Description: Lustre message v2
58 Details : Add lustre message format v2.
60 Severity : enhancement
62 Description: client OST exclusion list
63 Details : Clients can be started with a list of OSTs that should be
64 declared "inactive" for known non-responsive OSTs.
68 Description: SPEC SFS validation failure on NFS v2 over lustre.
69 Details : Changes the blocksize for regular files to be 2x RPC size,
70 and not depend on stripe size.
72 Severity : enhancement
74 Description: Multiple MD RPCs in flight.
75 Details : Further unserialise some read-only MDS RPCs - learn about intents.
76 To avoid overly-overloading MDS, introduce a limit on number of
77 MDS RPCs in flight for a single client and add /proc controls
80 Severity : enhancement
82 Description: client read/write statistics
83 Details : Add client read/write call usage stats for performance
84 analysis of user processes.
85 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
86 file access. extents_stats shows chunk size distribution.
87 extents_stats_per_process show chunk size distribution per
90 Severity : enhancement
92 Description: mds statistics
93 Details : Add detailed mds operations statistics in
94 /proc/fs/lustre/mds/*/stats.
98 Description: Failure of copying files with lustre special EAs.
99 Details : Client side always return success for setxattr call for lustre
100 special xattr (currently only "trusted.lov").
102 ------------------------------------------------------------------------------
104 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
106 * Support for kernels:
109 2.4.21-40.EL (RHEL 3)
110 2.6.12.6 vanilla (kernel.org)
115 Bugzilla : 5719, 9635, 9792, 9684
116 Description: OST (or MDS) trips assertions in (re)connection under heavy load
117 Details : If a server is under heavy load and cannot reply to new
118 connection requests before the client resends the (re)connect,
119 the connection handling code can behave badly if two service
120 threads are concurrently handing separate (re)connections from
121 the same client. Add better locking to the connection handling
122 code, and ensure that only a single connection will be processed
123 for a given client UUID, even if the lock is dropped.
125 Severity : enhancement
127 Description: add TCP zero-copy support to kernel
128 Details : Add support to the kernel TCP stack to allow zero-copy bulk
129 sends if the hardware supports scatter-gather and checksumming.
130 This allows socklnd to do client-write and server-read more
131 efficiently and reduce CPU utilization from skbuf copying.
134 Frequency : only if NFS exporting from client
136 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
137 Details : If setting attributes on a file created under NFS that had
138 never been opened it would be possible to oops the client
139 if the file had no objects.
142 Frequency : always for liblustre
144 Description: liblustre client does MDS+OSTs setattr RPC for each write
145 Details : When doing a write from a liblustre client, the client
146 incorrectly issued an RPC to the MDS and each OST the file was
147 striped over in order to update the timestamps. When writing
148 with small chunks and many clients this could overwhelm the MDS
149 with RPCs. In all cases it would slow down the write because
150 these RPCs are unnecessary.
152 Severity : enhancement
154 Description: allow number of MDS service threads to be changed at module load
155 Details : It is now possible to change the number of MDS service threads
156 running. Adding "options mds mds_num_threads={N}" to the MDS's
157 /etc/modprobe.conf will set the number of threads for the next
158 time Lustre is restarted (assuming the "mds" module is also
159 reloaded at that time). The default number of threads will
160 stay the same, 32 for most systems.
165 Description: OST crash if filesystem is unformatted or corrupt
166 Details : If an OST is started on a device that has never been formatted
167 or if the filesystem is corrupt and cannot even mount then the
168 error handling cleanup routines would dereference a NULL pointer.
173 Description: NULL pointer deref in llap_from_page.
174 Details : get_cache_page_nowait can return a page with NULL (or otherwise
175 incorrect) mapping if the page was truncated/reclaimed while it was
176 searched for. Check for this condition and skip such pages when
177 doing readahead. Introduce extra check to llap_from_page() to
178 verify page->mapping->host is non-NULL (so page is not anonymous).
181 Frequency : Sometimes when using sys_sendfile
183 Description: "page not covered by a lock" warnings from ll_readpage
184 Details : sendfile called ll_readpage without right page locks present.
185 Now we introduced ll_file_sendfile that does necessary locking
186 around call to generic_file_sendfile() much like we do in
190 Frequency : with certain MDS communication failures at client mount time
192 Description: NULL pointer deref after failed client mount
193 Details : a client connection request may delayed by the network layer
194 and not be sent until after the PTLRPC layer has timed out the
195 request. If the client fails the mount immediately it will try
196 to clean up before the network times out the request. Add a
197 reference from the request import to the obd device and delay
198 the cleanup until the network drops the request.
201 Frequency : occasionally during client (re)connect
203 Description: assertion failure during client (re)connect
204 Details : processing a client connection request may be delayed by the
205 client or server longer than the client connect timeout. This
206 causes the client to resend the connection request. If the
207 original connection request is replied in this interval, the
208 client may trip an assertion failure in ptlrpc_connect_interpret()
209 which thought it would be the only running connect process.
212 Frequency : only with obd_echo servers and clients that are rebooted
214 Description: kernel BUG accessing uninitialized data structure
215 Details : When running an obd_echo server it did not start the ping_evictor
216 thread, and when a client was evicted an uninitialized data
217 structure was accessed. Start the ping_evictor in the RPC
218 service startup instead of the OBD startup.
220 Severity : enhancement
221 Bugzilla : 10193 (patchless)
222 Description: Remove dependency on various unexported kernel interfaces.
223 Details : No longer need reparent_to_init, exit_mm, exit_files,
224 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
227 Frequency : rare (only users of deprecated and unsupported LDAP config)
229 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
230 Details : LDAP apparently contains 'lustreName' attributes instead of
231 'name'. A simple remapping of the name is sufficient.
234 Frequency : rare (only with non-default dump_on_timeout debug enabled)
236 Description: waiting_locks_callback trips kernel BUG if client is evicted
237 Details : Running with the dump_on_timeout debug flag turned on makes
238 it possible that the waiting_locks_callback() can try to dump
239 the Lustre kernel debug logs from an interrupt handler. Defer
240 this log dumping to the expired_lock_main() thread.
242 Severity : enhancement
244 Description: Support NFS exporting on 2.6 kernels.
245 Details : Implement non-rawops metadata methods for NFS server to use without
246 changing NFS server code.
249 Frequency : very rare (synthetic metadata workload only)
251 Description: two racing renames might cause an MDS thread to deadlock
252 Details : Running the "racer" program may cause one MDS thread to rename
253 a file from being the source of a rename to being the target of
254 a rename at exactly the same time that another thread is doing
255 so, and the second thread has already enqueued these locks after
256 doing a lookup of the target and is trying to relock them in
257 order. Ensure that we don't try to re-lock the same resource.
260 Frequency : only very large systems with liblustre clients
262 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
263 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
264 all OSTs in parallel. This allows the eviction of stale liblustre
265 clients to proceed much faster than if they were done in series,
266 and also offers similar improvements for other set_info RPCs.
271 Description: excessive CPU usage during initial read phase on client
272 Details : During the initial read phase on a client, it would agressively
273 retry readahead on the file, consuming too much CPU and impacting
274 performance (since 1.4.5.8). Improve the readahead algorithm
275 to avoid this, and also improve some other common cases (read
276 of small files in particular, where "small" is files smaller than
277 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
282 Description: MDS crash when receiving packet with unknown intent.
283 Details : Do not LBUG in unknown intent case, just return -EFAULT
285 Severity : enhancement
286 Bugzilla : 9293, 9385
287 Description: MDS RPCs are serialised on client. This is unnecessary for some.
288 Details : Do not serialize getattr (non-intent version) and statfs.
291 Frequency : occasional, when OST network is overloaded/intermittent
293 Description: client evicted by OST after bulk IO timeout
294 Details : If a client sends a bulk IO request (read or write) the OST
295 may evict the client if it is unresposive to its data GET/PUT
296 request. This is incorrect if the network is overloaded (takes
297 too long to transfer the RPC data) or dropped the OST GET/PUT
298 request. There is no need to evict the client at all, since
299 the pinger and/or lock callbacks will handle this, and the
300 client can restart the bulk request.
303 Frequency : Always when mmapping file with no objects
305 Description: client crashes when mmapping file with no objects
306 Details : Check that we actually have objects in a file before doing any
307 operations on objects in ll_vm_open, ll_vm_close and
313 Description: Request leak when working with deleted CWD
314 Details : Introduce advanced request refcount tracking for requests
315 referenced from lustre intent.
317 Severity : Enhancement
319 Description: Cache open file handles on client.
320 Details : MDS now will return special lock along with openhandle, if
321 requested and client is allowed to hold openhandle, even if unused,
322 until such a lock is revoked. Helps NFS a lot, since NFS is opening
323 closing files for every read/write openration.
325 Severity : Enhancement
327 Description: Cache open negative dentries on client when possible.
328 Details : Guard negative dentries with UPDATE lock on parent dir, drop
329 negative dentries on lock revocation.
334 Description: Remounting a client read-only wasn't possible with a zconf mount
335 Details : It wasn't possible to remount a client read-only with llmount.
337 Severity : enhancement
338 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
339 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
340 MPICH is included to add Lustre-specific ADIO interfaces.
341 This is based closely on the UFS ADIO layer and only differs
342 in file creation, in order to allow the OST striping to be set.
343 This is user-contributed code and not supported by CFS.
348 Description: extended inode attributes (immutable, append-only) work improperly
349 when 2.4 and 2.6 kernels are used on client/server or vice versa
350 Details : Introduce kernel-independent values for these flags.
352 Severity : enhancement
355 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
356 Details : Many of the /proc/ tunables can only be tuned at a megabyte
357 granularity. Now, Fractional MB granularity is be supported,
358 this is very useful for low memory system.
360 Severity : enhancement
362 Description: Getattr by fid
363 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
364 locks, avoids extra getattr rpc requests to MDS, allows '/' to
365 have locks and avoids getattr rpc requests for it on every stat.
368 Frequency : Always, for filesystems larger than 2TB
370 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
371 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
372 filesystems larger than 2TB to be mounted with mballoc enabled.
375 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
377 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
378 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
379 filesystems larger than 2TB is mounted then the kernel will
380 silently corrupt the start of the filesystem. CONFIG_LBD is
381 enabled for all CFS-supported kernels, but the possibility of
382 this happening with a modified kernel config exists.
384 Severity : enhancement
386 Description: add client O_DIRECT support for 2.6 kernels
387 Details : It is now possible to do O_DIRECT reads and writes to files
388 in the Lustre client mountpoint on 2.6 kernel clients.
390 Severity : enhancement
392 Description: parallel glimpse, setattr, statfs, punch, destroy requests
393 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
394 parallel, not waiting for response from every OST before sending
395 a rpc to the next OST.
400 Description: setattr vs write race when updating file timestamps
401 Details : Client processes that update a file timestamp into the past
402 right after writing to the file (e.g. tar) it is possible that
403 the updated file modification time can be reset to the current
404 time due to a race between processing the setattr and write RPC.
406 Severity : enhancement
408 Description: Bring 'lfs find' closer in line with regular Linux find.
409 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
410 -print0 options and obtains all the needed info through the lustre
413 Severity : enhancement
415 Description: support up to 1024 configured devices on one node
416 Details : change obd_dev array from statically allocated to dynamically
417 allocated structs as they are first used to reduce memory usage
422 Description: Flush dirty partially truncated pages during truncate
423 Details : Immediatelly flush partially truncated pages in filter_setattr,
424 this way we completely avoid having any pages in page cache on OST
425 and can retire ugly workarounds during writes to flush such pages.
430 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
431 Details : protect inode from truncation within vfs_unlink() context
432 just take a reference before calling vfs_unlink() and release it
433 when parent's i_sem is free.
438 Description: last_id value checked outside lock on OST caused LASSERT failure
439 Details : If there were multiple MDS->OST object precreate requests in
440 flight, it was possible that the OST's last object id was checked
441 outside a lock and incorrectly tripped an assertion. Move checks
442 inside locks, and discard old precreate requests.
445 Frequency : always, if extents are used on OSTs
447 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
448 Details : OSTs using the extents format would not zero the high 16 bits of
449 the index physical block number. This is not a problem for any
450 OST filesystems smaller than 16TB, and no kernels support ext3
451 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
452 new/modified files) and can be fixed for existing filesystems
453 with e2fsprogs-1.39-cfs1.
458 Description: import connection selection may be incorrect if timer wraps
459 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
460 import connections to be ignored if the 32-bit jiffies counter
461 wraps. Use a 64-bit jiffies counter.
464 Frequency : very large clusters immediately after boot
466 Description: LNET request buffers exhausted under heavy short-term load
467 Details : If a large number of client requests are generated on a service
468 that has previously never seen so many requests it is possible
469 that the request buffer growth cannot keep up with the spike in
470 demand. Instead of dropping incoming requests, they are held in
471 the LND until the RPC service can accept more requests.
474 Frequency : Sometimes during replay
476 Description: Assertion failure in ll_local_open after replay.
477 Details : If replay happened on an open request reply before we were able
478 to set replay handler, reply will become not swabbed tripping the
479 assertion in ll_local_open. Now we set the handler right after
480 recognising of open request
483 Frequency : very rare
485 Description: kernel reports "badness in vsnprintf"
486 Details : Reading from the "recovery_status" /proc file in small chunks
487 may cause a negative length in lprocfs_obd_rd_recovery_status()
488 call to vsnprintf() (which is otherwise harmless). Exit early
489 if there is no more space in the output buffer.
491 Severity : enhancement
493 Description: clear OBD RPC statistics by writing to them
494 Details : It is now possible to clear the OBD RPC statistics by writing
500 Description: Client mtime is not the same on different clients after utimes
501 Details : In some cases, the client was using the utimes() syscall on
502 a file cached on another node. The clients now validate the
503 ctime from the MDS + OSTs to determine which one is right.
508 Description: Inability to activate failout mode
509 Details : lconf script incorrectly assumed that in pythong string's numeric
510 value is used in comparisons.
513 Frequency : always with multiple stripes per file
515 Description: Inefficient object allocation for mutli-stripe files
516 Details : When selecting which OSTs to stripe files over, for files with
517 a stripe count that divides evenly into the number of OSTs,
518 the MDS is always picking the same starting OST for each file.
519 Return the OST selection heuristic to the original design.
524 Description: mount failures may take full timeout to return an error
525 Details : Under some heavy load conditions it is possible that a
526 failed mount can wait for the full obd_timeout interval,
527 possibly several minutes, before reporting an error.
528 Instead return an error as soon as the status is known.
530 Frequency : quota enabled and large files being deleted
532 Description: releasing more than 4GB of quota at once hangs OST
533 Details : If a user deletes more than 4GB of files on a single OST it
534 will cause the OST to spin in an infinite loop. Release
535 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
540 Description: statfs data retrieved from /proc may be stale or zero
541 Details : When reading per-device statfs data from /proc, in the
542 {kbytes,files}_{total,free,avail} files, it may appear
543 as zero or be out of date.
546 Frequency : systems with MD RAID1 external journal devices
548 Description: lconf's call to blkid is confused by RAID1 journal devices
549 Details : Use the "blkid -l" flag to locate the MD RAID device instead
550 of returning all block devices that match the journal UUID.
553 Frequency : always, for aggregate stripe size over 4GB
555 Description: assertion fails when trying to use 4GB stripe size
556 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
557 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
560 Frequency : always on ppc64
562 Description: the first write on an ext3 filesystem with mballoc got stuck
563 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
564 perform endianness conversion.
566 ------------------------------------------------------------------------------
568 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
570 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
571 INTEROPERATE with older versions automatically. Please read the
572 user documentation before upgrading any part of a live system.
573 * WARNING: Lustre networking configuration changes are required with
574 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
577 * Support for kernels:
578 2.6.9-22.0.2.EL (RHEL 4)
580 2.6.12.6 vanilla (kernel.org)
583 Severity : enhancement
585 Description: Introduced Lustre Networking (LNET)
586 Details : LNET is new networking infrastructure for Lustre, it includes
587 a reorganized network configuration mode (see the user
588 documentation for full details) as well as support for routing
589 between different network fabrics. Lustre Networking Devices
590 (LNDs) for the supported network fabrics have also been
591 created for this new infrastructure.
593 Severity : enhancement
594 Description: Introduced Access control lists
595 Details : clients can set ACLs on files and directories in order to have
596 more fine-grained permissions than the standard Unix UGO+RWX.
597 The MDS must be started with the "-o acl" mount option.
599 Severity : enhancement
600 Description: Introduced filesystem quotas
601 Details : Administrators may now establish per-user quotas on the
604 Severity : enhancement
606 Description: Configuration change for the XT3
607 The PTLLND is now used to run Lustre over Portals on the XT3
608 The configure option(s) --with-cray-portals are no longer used.
609 Rather --with-portals=<path-to-portals-includes> is used to
610 enable building on the XT3. In addition to enable XT3 specific
611 features the option --enable-cray-xt3 must be used.
616 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
617 Details : A race between allocating a new llcd and re-getting the llcd_lock
618 allowed another thread to grab newly-allocated llcd.
620 Severity : enhancement
622 Description: 2.6 OST async journal commit and locking fix to improve performance
623 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
624 2.6 kernels are now async as they already were in 2.4 kernels so
625 that they can commit concurrently with the network bulk transfer.
626 For block-allocated files the filter allocation semaphore is held
627 to avoid filesystem fragmentation during allocation. BKL lock
628 removed for 2.6 xattr operations where it is no longer needed.
633 Description: lconf incorrectly determined whether two IP networks could talk
634 Details : In some more complicated routing and multiple-network
635 configurations, lconf will avoid trying to make a network
636 connection to a disjoint part of the IP space. It was doing the
637 math incorrectly for one set of cases.
642 Description: Fix for potential infinite loop processing records in an llog.
643 Details : If an llog record is corrupted/zeroed, it is possible to loop
644 forever in llog_process(). Validate the llog record length
645 and skip the remainder of the block on error.
648 Frequency : occasional (liblustre only)
650 Description: liblustre could not open files whose last component is a symlink
651 Details : sysio_path_walk() would incorrectly pass the open intent to
652 intermediate path components.
655 Frequency : rare (liblustre only with non-standard tuning)
656 Bugzilla : 7201 (7350)
657 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
658 Details : llu_lookup_finish_locks() tries to set lock data on a lock
659 after it has been released, only do this for referenced locks
661 Severity : enhancement
663 Description: specifying an (invalid) directory default stripe_size of -1
664 would reset the directory default striping
665 Details : stripe_size -1 was used internally to signal directory stripe
666 removal, now use "all default" to signal dir stripe removal
667 as a directory striping of "all default" is not useful
670 Frequency : common for large clusters running liblustre clients
672 Description: doing an ls when liblustre clients are running is slow
673 Details : sending a glimpse AST to a liblustre client waits for every AST
674 to time out, as liblustre clients will not respond. Since they
675 cannot cache data we refresh the OST lock LVB from disk instead.
677 Severity : enhancement
679 Description: doing an ls at the same time as file IO can be slow
680 Details : enqueue and other "small" requests can be blocked behind many
681 large IO requests. Create a new OST IO portal for non-IO
682 requests so they can be processed faster.
685 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
687 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
688 created file with mode 000
689 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
690 allow an ftruncate() as a non-root user to a file with mode 000.
691 Lustre now respects this flag to disable mode checks when
692 truncating a file owned by the user
695 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
698 Description: Revoking locks from clients that went dead or catatonic might take
700 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
701 cancellation of such locks instant on servers without waiting for
702 any reply from clients. Clients drops these locks when cancel
703 notification from server is received without replying.
706 Frequency : liblustre-only, when liblustre client dies or becomes busy
708 Description: Doing ls on Linux clients can take a long time with active
710 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
711 granting such locks to it in the first place if possible. Locks
712 are taken by proxy on the OST during the read or write and
713 dropped immediately afterward. Add connect flags handling, do
714 not grant locks to liblustre clients for glimpse ASTs.
716 Severity : enhancement
718 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
719 Details : Existing read-ahead algorithm is tuned for the case of streamlined
720 sequential reads and behaves badly with applications doing random
721 reads. Improve it by reading ahead at least read region, and
722 avoiding excessive large RPC for small reads.
724 Severity : enhancement
726 Description: Creating more than 1000 files for a single job may cause a load
727 imbalance on the OSTs if there are also a large number of OSTs.
728 Details : qos_prep_create() uses an OST index reseed value that is an
729 even multiple of the number of available OSTs so that if the
730 reseed happens in the middle of the object allocation it will
731 still utilize the OSTs as uniformly as possible.
736 Description: OST or MDS may oops in ping_evictor_main()
737 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
738 but doesn't restart at beginning of obd_exports_timed list
741 Severity : enhancement
743 Description: improve by-nid export eviction on the MDS and OST
744 Details : allow multiple exports with the same NID to be evicted at one
745 time without re-searching the exports list.
748 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
750 Description: MDS may oops in groups_free()
751 Details : in rare race conditions a newly allocated group_info struct is
752 freed again, and this can be NULL. The 2.4 compatibility code
753 for groups_free() checked for a NULL pointer, but 2.6 did not.
756 Frequency : common for liblustre clients doing little filesystem IO
757 Bugzilla : 9352, 7313
758 Description: server may evict liblustre clients accessing contended locks
759 Details : if a client is granted a lock or receives a completion AST
760 with a blocking AST already set it would not reply to the AST
761 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
764 Frequency : lfs setstripe, only systems with more than 160 OSTs
766 Description: unable to set striping with a starting offset beyond OST 160
767 Details : llapi_create_file() incorrectly limited the starting stripe
768 index to the maximum single-file stripe count.
771 Frequency : LDAP users only
773 Description: lconf did not handle in-kernel recovery with LDAP properly
774 Details : lconf/LustreDB get_refs() is searching the wrong namespace
776 Severity : enhancement
778 Description: bind OST threads to NUMA nodes to improve performance
779 Details : all OST threads are uniformly bound to CPUs on a single NUMA
780 node and do their allocations there to localize memory access
782 Severity : enhancement
784 Description: llmount can determine client NID directly from Myrinet (GM)
785 Details : the client NID code from gmnalnid was moved directly into
786 llmount, removing the need to use this or specifying the
787 client NID explicitly when mounting GM clients with zeroconf
790 Frequency : if client is started with down MDS
792 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
793 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
794 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
795 ptlrpc_import_delayed_req() because we didn't check if the
796 request was interrupted, and we also didn't break out of the
797 event loop if there was no timeout
802 Description: data loss during non-page-aligned writes to a single file from
803 both multiple nodes and multiple threads on one node at same time
804 Details : updates to KMS and lsm weren't protected by common lock. Resulting
805 inconsistency led to false short-reads, that were cached and later
806 used by ->prepare_write() to fill in partially written page,
807 leading to data loss.
810 Frequency : always, if lconf --abort_recovery used
812 Description: lconf --abort_recovery fails with 'Operation not supported'
813 Details : lconf was attempting to abort recovery on the MDT device and not
816 Severity : enhancement
818 Description: remove cleanup logs
819 Details : replace lconf-generated cleanup logs with lustre internal
820 cleanup routines. Eliminates the need for client-cleanup and
823 Severity : enhancement
825 Description: add support for EAs (user and system) on lustre filesystems
826 Details : it is now possible to store extended attributes in the Lustre
827 client filesystem, and with the user_xattr mount option it
828 is possible to allow users to store EAs on their files also
830 Severity : enhancement
832 Description: Add possibility (config option) to show minimal available OST free
834 Details : When compiled with --enable-mindf configure option, statfs(2)
835 (and so, df) will return least minimal free space available from
836 all OSTs as amount of free space on FS, instead of summary of
837 free spaces of all OSTs.
839 Severity : enhancement
841 Description: do not expand extent locks acquired on OST-side
842 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
843 by server: they are not cached anyway.
846 Frequency : when mmap is used/binaries executed from Lustre
848 Description: Unmmap pages before throwing them away from read cache.
849 Details : llap_shrink cache now attempts to unmap pages before discarding
850 them (if unmapping failed - do not discard). SLES9 kernel has
851 extra checks that trigger if this unmapping is not done first.
856 Description: lconf didn't resolve symlinks before checking to see whether a
857 given mountpoint was already in use
860 Frequency : when migrating failover services
861 Bugzilla : 6395, 9514
862 Description: When migrating a subset of services from a node (e.g. failback
863 from a failover service node) the remaining services would
864 time out and evict clients.
865 Details : lconf --force (implied by --failover) sets the global obd_timeout
866 to 5 seconds in order to quickly disconnect, but this caused
867 other RPCs to time out too quickly. Do not change the global
868 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
870 Severity : enhancement
871 Frequency : if MDS is started with down OST
873 Description: Allow startup/shutdown of an MDS without depending on the
874 availability of the OSTs.
875 Details : Asynchronously call mds_lov_synchronize during MDS startup.
876 Add appropriate locking and lov-osc refcounts for safe
877 cleaning. Add osc abort_inflight calls in case the
878 synchronize never started.
881 Frequency : occasional (Cray XT3 only)
883 Description: root not authorized to access files in CRAY_PORTALS environment
884 Details : The client process capabilities were not honoured on the MDS in
885 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
886 been accessed by an authorized user then root was able to access
887 the file on the local client also. The root user capabilities
888 are now allowed on the MDS, as this environment has secure UID.
891 Frequency : occasional
893 Description: ldiskfs "too long searching" message happens too often
894 Details : A debugging message (otherwise harmless) prints too often on
895 the OST console. This has been reduced to only happen when
896 there are fragmentation problems on the filesystem.
901 Description: Division by zero in statfs when all OSCs are inactive
902 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
903 lov_statfs divides by value returned by lov_get_stripecnt().
907 Bugzilla : 9489, 3273
908 Description: First write from each client to each OST was only 4kB in size,
909 to initialize client writeback cache, which caused sub-optimal
910 RPCs and poor layout on disk for the first writen file.
911 Details : Clients now request an initial cache grant at (re)connect time
912 and so that they can start streaming writes to the cache right
913 away and always do full-sized RPCs if there is enough data.
914 If the OST is rebooted the client also re-establishes its grant
915 so that client cached writes will be honoured under the grant.
920 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
921 Details : Now I/O RPCs go to different portal number and (presumably) fast
922 lock requests (and glimses) and other RPCs get their own service
923 threads pool that should be able to service those RPCs
926 Severity : enhancement
928 Description: Ability to exchange lustre version between client and servers and
929 issue warnings at client side if client is too old. Also for
930 liblustre clients there is ability to refuse connection of too old
932 Details : New 'version' field is added to connect data structure that is
933 filled with version info. That info is later checked by server and
937 Frequency : rare, liblustre only.
938 Bugzilla : 9296, 9581
939 Description: Two simultaneous writes from liblustre at offset within same page
940 might proceed at the same time overwriting eachother with stale
942 Details : I/O lock withing llu_file_prwv was released too early, before data
943 actually was hitting the wire. Extended lock-holding time until
944 server acknowledges receiving data.
947 Frequency : extremely rare. Never observed in practice.
949 Description: avoid generating lustre_handle cookie of 0.
950 Details : class_handle_hash() generates handle cookies by incrementing
951 global counter, and can hit 0 occasionaly (this is unlikely, but
952 not impossible, because initial value of cookie counter is
953 selected randonly). Value of 0 is used as a sentinel meaning
954 "unassigned handle" --- avoid it. Also coalesce two critical
955 sections in this function into one.
957 Severity : enhancement
959 Description: allow liblustre clients to delegate truncate locking to OST
960 Details : To avoid overhead of locking, liblustre client instructs OST to
961 take extent lock in ost_punch() on client's behalf. New connection
962 flag is added to handle backward compatibility.
964 Severity : enhancement
965 Bugzilla : 4928, 7341, 9758
966 Description: allow number of OST service threads to be specified
967 Details : a module parameter allows the number of OST service threads
968 to be specified via "options ost ost_num_threads={N}" in the
969 OSS's /etc/modules.conf or /etc/modprobe.conf.
973 Bugzilla : 6146, 9635, 9895
974 Description: servers crash with bad pointer in target_handle_connect()
975 Details : In rare cases when a client is reconnecting it was possible that
976 the connection request was the last reference for that export.
977 We would temporarily drop the export reference and get a new
978 one, but this may have been the last reference and the export
979 was just destroyed. Get new reference before dropping old one.
981 Severity : enhancement
982 Frequency : if client is started with failover MDS
984 Description: Allow multiple MDS hostnames in the mount command
985 Details : Try to read the configuration from all specified MDS
986 hostnames during a client mount in case the "primary"
989 Severity : enhancement
991 Description: Stop sending data to evicted clients as soon as possible.
992 Details : Check if the client we are about to send or are sending data to
993 was evicted already. (Check is done every second of waiting,
994 for which l_wait_event interface was extended to allow checking
995 of exit condition at specified intervals).
998 Frequency : rare, normally only when NFS exporting is done from client
1000 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1002 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1005 Frequency : always, if srand() is called before liblustre initialization
1007 Description: Liblustre uses system PRNG disturbing its usage by user application
1008 Details : Introduce internal to lustre fast and high-quality PRNG for
1009 lustre usage and make liblustre and some other places in generic
1010 lustre code to use it.
1012 Severity : enhancement
1013 Bugzilla : 9477, 9557, 9870
1014 Description: Verify that the MDS configuration logs are updated when xml is
1015 Details : Check if the .xml configuration logs are newer than the config
1016 logs stored on the MDS and report an error if this is the case.
1017 Request --write-conf, or allow starting with --old_conf.
1019 Severity : enhancement
1021 Description: Handle symlinks in the path when checking if Lustre is mounted.
1022 Details : Resolve intermediate symlinks when checking if a client has
1023 mounted a filesystem to avoid duplicate client mounts.
1028 Description: lconf can hit an error exception but still return success.
1029 Details : The lconf command catches the Command error exception at the top
1030 level script context and will exit with the associated exit
1031 status, but doesn't ensure that this exit status is non-zero.
1036 Description: failure of ptlrpc thread startup can cause oops
1037 Details : Starting a ptlrpc service thread can fail if there are a large
1038 number of threads or the server memory is very fragmented.
1039 Handle this without oopsing.
1042 Frequency : always, only if liblustre and non-default acceptor port was used
1044 Description: liblustre cannot connect to servers with non-default acceptor port
1045 Details : tcpnal_set_default_params() was not called and was therefore
1046 ignoring the environment varaible TCPNAL_PORT, as well as other
1047 TCPNAL_ environment variables
1052 Description: two objects could be created on the same OST for a single file
1053 Details : If an OST is down, in some cases it was possible to create two
1054 objects on a single OST for a single file. No problems other
1055 than potential performance impact and spurious error messages.
1059 Bugzilla : 5681, 9562
1060 Description: Client may oops in ll_unhash_aliases
1061 Details : Client dcache may become inconsistent in race condition.
1062 In some cases "getcwd" can fail if the current directory is
1068 Description: Inode refcounting problems in NFS export code
1069 Details : link_raw functions used to call d_instantiate without obtaining
1070 extra inode reference first.
1074 Bugzilla : 9942, 9903
1075 Description: Referencing freed requests leading to crash, memleaks with NFS.
1076 Details : We used to require that call to ll_revalidate_it was always
1077 followed by ll_lookup_it. Also with revalidate_special() it is
1078 possible to call ll_revalidate_it() twice for the same dentry
1079 even if first occurence returned success. This fix changes semantic
1080 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1081 reference on a request referred from the intent.
1082 ll_intent_release() then releases such a request.
1085 Frequency : rare, normally benchmark loads only
1087 Description: unlinked inodes were kept in memory on the client
1088 Details : If a client is repeatedly creating and unlinking files it
1089 can accumulate a lot of stale inodes in the inode slab cache.
1090 If there is no other client load running this can cause the
1091 client node to run out of memory. Instead flush old inodes
1092 from client cache that have the same inode number as a new inode.
1095 Frequency : SLES9 2.6.5 kernel and long filenames only
1096 Bugzilla : 9969, 10379
1097 Description: utime reports stale NFS file handle
1098 Details : SLES9 uses out-of-dentry names in some cases, which confused
1099 the lustre dentry revalidation. Change it to always use the
1103 Frequency : rare, unless heavy write-truncate concurrency is continuous
1104 Bugzilla : 4180, 6984, 7171, 9963, 9331
1105 Description: OST becomes very slow and/or deadlocked during object unlink
1106 Details : filter_destroy() was holding onto the parent directory lock
1107 while truncating+unlinking objects. For very large objects this
1108 may block other threads for a long time and slow overall OST
1109 responsiveness. It may also be possible to get a lock ordering
1110 deadlock in this case, or run out of journal credits because of
1111 the combined truncate+unlink. Solution is to do object truncate
1112 first in one transaction without parent lock, and then do the
1113 final unlink in a new transaction with the parent lock. This
1114 reduces the lock hold time dramatically.
1117 Frequency : rare, 2.4 kernels only
1119 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1120 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1121 before unmounting their filesystem in order to avoid deadlock.
1122 The kernel_locked() function in 2.4 kernels only checks whether
1123 the kernel lock is held, not whether it is this process that is
1124 holding it as 2.6 kernels do.
1129 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1130 Details : The client ptlrpc code may be trying to reconnect to a down
1131 server before a previous connection attempt has timed out.
1132 Increase the reconnect interval to be longer than the connection
1133 timeout interval to avoid sending duplicate connections to
1137 Frequency : echo_client brw_test command
1139 Description: fix echo_client to work with OST preallocated code
1140 Details : OST preallocation code (5137) didn't take echo_client IO path
1141 into account: echo_client calls filter methods outside of any
1142 OST thread and, hence, there is no per-thread preallocated
1143 pages and buffers to use. Solution: hijack pga pages for IO. As
1144 a byproduct, this avoids unnecessary data copying.
1148 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1149 Description: Client can oops in mdc_commit_close() after open replay
1150 Details : It was possible for the MDS to return an open request with no
1151 transaction number in mds_finish_transno() if the client was
1152 evicted, but without actually returning an error. Clients
1153 would later try to replay that open and may trip an assertion
1154 Simplify the client close codepath, and always return an error
1155 from the MDS in case the open is not successful.
1158 Frequency : rare, 2.6 OSTs only
1160 Description: OST may deadlock under high load on fragmented files
1161 Details : If there was a heavy load and highly-fragmented OST filesystems
1162 it was possible to have all the OST threads deadlock waiting on
1163 allocation of biovecs, because the biovecs were not released
1164 until the entire RPC IO was completed. Instead, release biovecs
1165 as soon as they are complete to ensure forward IO progress.
1167 Severity : enhancement
1169 Description: Support for specifying external journal device at mount
1170 Details : If an OST or MDS device is formatted with an external journal
1171 device, this device major/minor is stored in the ext3 superblock
1172 and may not be valid for failover. Allow detecting and
1173 specifying the external journal at mount time.
1178 Description: Mounting an MDS with pending unlinked files may cause oops
1179 Details : target_finish_recovery() calls mds_postrecov() which returned
1180 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1181 considers this an error and immediately begins cleaning up the
1182 lov, just after starting the mds_lov process
1184 Severity : enhancement
1186 Description: Implement 'lfs df' to report actual free space on per-OST basis
1187 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1188 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1189 usage of objects; '-h' to report in human readable format.
1191 ------------------------------------------------------------------------------
1193 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
1200 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1201 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1202 at startup during orphan unlinking. Instead, skip buddy update
1203 before setup, as it will be regenerated after recovery is complete.
1208 Description: If an OST is inactive, its locks might reference stale inodes.
1209 Details : lov_change_cbdata() must iterate over all namespaces, even if
1210 they are inactive to clear inode references from the lock.
1212 Severity : enhancement
1213 Frequency : occasional, if non-standard max_dirty_mb used
1215 Description: Client will block write RPCs if not enough grant
1216 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1217 then the client will block writes while waiting for another RPC
1218 to complete instead of consuming its dirty limit. With change
1219 we get improved performance when max_dirty_mb is small.
1221 Severity : enhancement
1222 Bugzilla : 3389, 6253
1223 Description: Add support for supplementary groups on the MDS.
1224 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1225 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1226 lookups for user supplementary groups into a cache.
1230 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1231 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1232 for liblustre O_EXCL is enforced by MDS.
1236 Description: readdir never returns NULL in liblustre.
1237 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1238 dentry in struct dirent.
1242 Frequency : liblustre only, depends on application IO pattern
1243 Description: liblustre clients evicted if not contacting servers
1244 Details : Don't put liblustre clients into the ping_evictor list, so
1245 they will not be evicted by the pinger ever.
1247 Severity : enhancement
1249 Description: Add ability to evict clients by NID from MDS.
1250 Details : By echoing "nid:$NID" string into
1251 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1252 $NID would be instantly evicted from this MDS and from all active
1253 OSTs connected to it.
1257 Description: Do not query file size twice, somewhat slowing stat(2) calls.
1258 Details : lookup_it_finish() used to query file size from OSTs that was not
1263 Description: service threads change working directory to that of init
1264 Details : Starting lustre service threads may pin the working directory
1265 of the parent thread, making that filesystem busy. Threads
1266 now change to the working directory of init to avoid this.
1270 Frequency : during shutdown only
1271 Description: shutdown with a failed MDS or OST can cause unmount to hang
1272 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
1277 Frequency : 2.6 only
1278 Description: chmod/chown may include an extra supplementary group
1279 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
1280 supplementary group and if none is specified this is used.
1283 Bugzilla : 5479 (6816)
1285 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
1286 Details : If lookup is in progress on a file that is unlinked we might try
1287 to revalidate the inode and fail in revalidate after lookup is
1288 complete and ll_file_open() enqueues the open again but
1289 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
1292 Frequency : always, if lconf --abort_recovery used
1294 Description: lconf --abort_recovery fails with 'Operation not supported'
1295 Details : lconf was attempting to abort recovery on the MDT device and not
1298 ------------------------------------------------------------------------------
1300 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1305 Frequency : rare (only unsupported configurations with a node running as an
1307 Bugzilla : 6514, 5137
1308 Description: Mounting a Lustre file system on a node running as an OST could
1310 Details : OSTs now preallocates memory needed to write out data at
1311 startup, instead of when needed, to avoid having to
1312 allocate memory in possibly low memory situations.
1313 Specifically, if the file system is mounted on on OST,
1314 memory pressure could force it to try to write out data,
1315 which it needed to allocate memory to do. Due to the low
1316 memory, it would be unable to do so and the node would
1317 become unresponsive.
1319 Severity : enhancement
1321 Description: Addition of lconf --service command line option
1322 Details : lconf now accepts a '--service <arg>' option, which is
1323 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
1325 Severity : enhancement
1327 Description: Failover mode is now the default for OSTs.
1328 Details : By default, OSTs will now run in failover mode. To return to
1329 the old behaviour, add '--failout' to the lmc line for OSTs.
1331 Severity : enhancement
1333 Description: Health checks are now provided for MDS and OSTs
1334 Details : Additional detailed health check information on MSD and OSTs
1335 is now provided through the procfs health_check value.
1338 Frequency : occasional, depends on IO load
1340 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
1341 after numerous create/delete cycles
1342 Details : The ext3 inode allocation policy would not allocate new inodes
1343 very well on the OSTs because there are no new directories
1344 being created. Instead we look for groups with free space if
1345 the parent directories are nearly full.
1350 Description: Network or server problems during mount may cause partially
1351 mounted clients instead of returning an error.
1352 Details : The config llog parsing code may overwrite the error return
1353 code during mount error handling, returning success instead
1359 Description: MDS can fail to allocate large reply buffers
1360 Details : After long uptimes the MDS can fail to allocate large reply
1361 buffers (e.g. zconf client mount config records) due to memory
1362 fragmentation or consumption by the buffer cache. Preallocate
1363 some large reply buffers so that these replies can be sent even
1364 under memory pressure.
1368 Frequency : rare (liblustre)
1369 Description: fsx running with liblustre complained that using truncate() to
1370 extend the file doesn't work. This patch corrects that issue.
1371 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
1372 ATTR_SIZE and lsm use in llu_setattr_raw.
1376 Frequency : rare, only 2.6 kernels
1377 Description: Unusual file access patterns on the MDS may result in inode
1378 data being lost in very rare circumstances.
1379 Details : Bad interaction between the ea-in-inode patch and the "no-read"
1380 code in the 2.6 kernel caused the inode and/or EA data not to
1381 be read from disk, causing single-file corruption.
1385 Frequency : rare, only 2.6 filesystems using extents
1386 Description: Heavy concurrent write and delete load may cause data corruption.
1387 Details : It was possible under high-load situations to have an extent
1388 metadata block in the block device cache from a just-unlinked
1389 file overwrite a newly-allocated data block. We now unmap any
1390 metadata buffers that alias just-allocated data blocks.
1394 Frequency : filesystems with default stripe_count larger than 77
1395 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
1396 Details : lconf specifies an inode size of 4096 bytes when the default
1397 stripe_count is larger than 77. This conflicts with the default
1398 inode density of 1 per 4096 bytes. Allocate smaller inodes in
1399 this case to avoid pinning too much memory for large EAs.
1401 ------------------------------------------------------------------------------
1403 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
1408 Frequency : rare (extremely heavy IO load with hundreds of clients)
1410 Description: Client is evicted, gets IO error writing to file
1411 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
1412 caused clients to be evicted for AST timeouts. The fixes in
1413 bug 5192 mean we no longer need to have such short AST timeouts
1414 so ldlm_timeout has been increased.
1417 Frequency : occasional during --force or --failover shutdown under load
1418 Bugzilla : 5949, 4834
1419 Description: Server oops/LBUG if stopped with --force or --failover under load
1420 Details : a collection of import/export refcount and cleanup ordering
1421 issues fixed for safer force cleanup
1424 Frequency : only filesystems larger than 120 OSTs
1425 Bugzilla : 5990, 6223
1426 Description: lfs getstripe would oops on a very large filesystem
1427 Details : lov_getconfig used kfree on vmalloc'd memory
1430 Frequency : only filesystems exporting via NFS to Solaris 10 clients
1431 Bugzilla : 6242, 6243
1432 Description: reading from files that had been truncated to a non-zero size
1433 but never opened returned no data
1434 Details : ll_file_read() reads zeros from no-object files to EOF
1439 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
1440 Details : There's little harm in aborting MDS/OSS recovery and letting it
1441 try again, so I removed the LASSERT and return an error instead.
1443 Severity : enhancement
1445 Description: New debugging infrastructure for tracking down data corruption
1446 Details : The I/O checksum code was replaced to: (a) control it at runtime,
1447 (b) cover more of the client-side code path, and (c) try to narrow
1448 down where problems occurred
1452 Bugzilla : 3819, 4364, 4397, 6313
1453 Description: Racing close and eviction MDS could cause assertion in mds_close
1454 Details : It was possible to get multiple mfd references during close and
1455 client eviction, leading to one thread referencing a freed mfd.
1457 Severity: : enhancement
1458 Bugzilla : 3262, 6359
1459 Description: Attempts to reconnect to servers are now more aggressive.
1460 Details : This builds on the enhanced upcall-less recovery that was added
1461 in 1.4.2. When trying to reconnect to servers, clients will
1462 now try each server in the failover group every 10 seconds. By
1463 default, clients would previously try one server every 25 seconds.
1468 Description: After recovery, certain operations trigger a failed
1469 assertion on a client.
1470 Details : Failing over an mds, using lconf -d --failover, while a
1471 client was doing a readdir() call would cause the client to
1472 LBUG after recovery completed and the readdir() was resent.
1474 Severity : enhancement
1476 Description: Default groups are now added by lconf
1477 Details : You can now run lconf --group <servicename> without having to
1478 manually add groups with lmc.
1481 Frequency : occasional
1483 Description: Nodes with an elan id of 0 trigger a failed assertion
1486 Frequency : always when accessing e.g. tty/console device nodes
1488 Description: tty and some other devices nodes cannot be used on lustre
1489 Details : file's private_data field is used by device data and lustre
1490 values in there got lost. New field was added to struct file to
1491 store fs-specific private data.
1494 Frequency : when exporting Lustre via NFS
1496 Description: NFSD failed occasionally when looking up a path component
1497 Details : NFSD is looking up ".." which was broken in ext3 directories
1498 that had grown large enough to become hashed.
1501 Frequency : Clusters with multiple interfaces not on the same subnet
1503 Description: Nodes will repeatedly try to reconnect to an interface which it
1504 cannot reach and report an error to the log.
1505 Details : Extra peer list entries will be created by lconf with some peers
1506 unreachable. lconf now validates the peer before adding it.
1509 Frequency : Only if a default stripe is set on the filesystem root.
1511 Description: Setting a default stripe on the filesystem root prevented the
1512 filesystem from being remounted.
1513 Details : The client was sending extra request flags in the root getattr
1514 request and did not allocate a reply buffer for the dir EA.
1517 Frequency : occasional, higher if lots of files are accessed by one client
1518 Bugzilla : 6159, 6097
1519 Description: Client trips assertion regarding lsm mismatch/magic
1520 Details : While revalidating inodes the VFS looks up inodes with ifind()
1521 and in rare cases can find an inode that is being freed.
1522 The ll_test_inode() code will free the lsm during ifind()
1523 when it finds an existing inode and then the VFS later attaches
1524 this free lsm to a new inode.
1528 Bugzilla : 6422, 7030
1529 Description: MDS deadlock between mkdir and client eviction
1530 Details : Creating a new file via mkdir or mknod (starting a transaction
1531 and getting the ns lock) can deadlock with client eviction
1532 (gets ns lock and trying to finish a synchronous transaction).
1535 Frequency : occasional
1536 Description: While starting a server, the fsfilt_ext3 module could not be
1538 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
1539 kernels. Previously, lconf would still use the ext3 name
1540 when trying to load modules. Now, it will correctly use
1541 ext3 on 2.4 and ldiskfs on 2.6.
1543 Severity : enhancement
1544 Description: The default stripe count has been changed to 1
1545 Details : The interpretation of the default stripe count (0, to lfs
1546 or lmc) has been changed to mean striping across a single
1547 OST, rather than all available. For general usage we have
1548 found a stripe count of 1 or 2 works best.
1550 Severity : enhancement
1551 Description: Add support for compiling against Cray portals.
1552 Details : Conditional compiling for some areas that are different
1556 Frequency : occasional
1557 Bugzilla : 6409, 6834
1558 Description: Creating files with an explicit stripe count may lead to
1559 a failed assertion on the MDS
1560 Details : If some OSTs are full or unavailable, creating files may
1561 trigger a failed assertion on the MDS. Now, Lustre will
1562 try to use other servers or return an error to the
1566 Frequency : occasional
1568 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
1569 caused lock timeout during readahead (since 1.4.2).
1570 Details : Processes doing readahead might match a lock that hasn't been
1571 granted yet if there are overlapping and conflicting lock
1572 requests. The readahead process waits on ungranted lock
1573 (original lock is CBPENDING), while OST waits for that process
1574 to cancel CBPENDING read lock and eventually evicts client.
1576 Severity : enhancement
1578 Description: Initial enabling of flock support for clients
1579 Details : Implements fcntl advisory locking and file status functions.
1580 This feature is provided as an optional mount flag (default
1581 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
1582 locking are implemented yet, and those that are are not guaranteed
1583 to be completely correct in production environments.
1584 mount -t lustre -o [flock|noflock] ...
1587 Frequency : occasional
1589 Description: OSTs running 2.4 kernels but with extents enabled might trip an
1590 assertion in the ext3 JBD (journaling) layer.
1591 Details : The b_committed_data struct is protected by the big kernel lock
1592 in 2.4 kernels, serializing journal_commit_transaction() and
1593 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
1594 access to this struct. In 2.6 kernels there is finer grained
1595 locking to improve SMP performance of the JBD layer.
1599 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
1601 -----------------------------------------------------------------------------
1603 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
1605 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
1606 versions of Lustre. Please update all servers and clients to
1607 version 1.4.2 or later at the same time. You must also run
1608 "lconf --write-conf {config}.xml" on the MDS while it is stopped
1609 to update the configuration logs.
1611 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
1612 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
1613 - swab llog records of type '0' so we get proper header size/idx (5861)
1614 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
1615 - fix rename of one directory over another leaking an inode (5953)
1616 - avoid SetPageDirty on 2.6 (5981)
1617 - don't re-add just-being-destroyed locks to the waiting list (5653)
1618 - when creating new directories, inherit the parent's custom
1619 striping settings if present parent (3048)
1620 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
1621 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
1622 - don't hold client locks on temporary worklist from l_lru (5666)
1623 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
1624 - automatically evict dead clients (5921)
1625 - Update file size properly in create+truncate+fstat case (6196)
1626 - Do not unhash mountpoint dentries, do not allow removal of
1628 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
1629 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
1630 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
1631 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
1632 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
1633 - Correctly handle reads of files with no objects (6243)
1634 - lctl recover will also mark a device active if deactivate used (5933)
1636 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
1637 - allow --write-conf on an MDS with different nettype than client (5619)
1638 - don't write config llogs to MDS for mounts not from that MDS (5617)
1639 - lconf should create multiple TCP connections from a client (5201)
1640 - init scripts are now turned off by default; run chkconfig --on
1641 lustre and chkconfig --on lustrefs to use them
1642 - upcalls are no longer needed for clients to recover to failover
1644 - add --abort-recovery option to lconf to abort recovery on device
1646 - add support for an arbitrary number of OSTs (3026)
1647 - Quota support protocol changes.
1648 - forward compatibility changes to wire structs (6007)
1649 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
1650 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
1651 - improved "lustre status" script
1652 - initialize blocksize for non-regular files (6062)
1653 - added --disable-server and --disable-client configure options (5782)
1654 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
1655 - Vanilla 2.4.29 support
1656 - increase maximum number of obd devices to 520 (6242)
1657 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
1658 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
1659 SLES 9 2.6.5 kernels
1660 - limit stripes per file to 160 (the maximum EA size) (6093)
1662 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
1665 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
1666 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
1667 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
1668 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
1669 - small ext3 extents cleanups and fixes (5733)
1670 - improved mballoc code, several small races and bugs fixed (5733, 5638)
1671 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
1672 - avoid needless client->OST connect, fix handle mismatch (5317)
1673 - fix DLM error path that led to out-of-sync client, long delays (5779)
1674 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
1675 - fix several locking issues related to i_size (5492,5624,5654,5672)
1676 - don't move pending lock onto export if it is already evicted (5683)
1677 - fix kernel oops when creating .foo in unlinked directory (5548)
1678 - fix deadlock in obdfilter statistics vs. object create (5811)
1679 - use time_{before,after} to avoid timer jiffies wrap (5882)
1680 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
1681 - Do not leak request if server was not able to process it (5154)
1682 - If mds_open unable to find parent dir, make that negative lookup(5154)
1683 - don't create new directories with extent-mapping (5909, 5936)
1685 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
1686 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
1687 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
1688 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
1689 - support for RHEL4 kernel on i686 (5773)
1690 - provide error messages when incompatible logs are encountered (5898)
1692 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
1693 * version 1.4.0.10 (1.4.1 release candidate 1)
1695 - don't keep a lock reference when lock is not granted (4238)
1696 - unsafe list practices (rarely) led to infinite eviction loop (4908)
1697 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
1698 - drop import inflight refcount on signal_completed_replay error (5255)
1699 - unlock page after async write error during send (3677)
1700 - handle missing objects in filter_preprw_read properly (5265)
1701 - no transno return for symlink open, don't save no-trasno open (3440)
1702 - don't try to complete elan receive that already failed (4012)
1703 - free RPC server reply state on error (5406)
1704 - clean up thread from ptlrpc_start_thread() on error (5160)
1705 - readahead could read extra page into cache that wasn't ejected (5388)
1706 - prevent races in class_attach/setup/cleanup/detach (5260)
1707 - don't dereference de->d_inode after l_dput of de (5458)
1708 - use "int" for stripe value returned from lock_to_stripe (5544)
1709 - mballoc allocation and error-checking fixes in 2.6 (5504)
1710 - block device patches to fix I/O request sizes in 2.6 (5482)
1711 - look up hostnames for IB nals (5602)
1712 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
1713 - don't start multiple acceptors for the same port (5277)
1714 - fix incorrect LASSERT in mds_getattr_name (5635)
1715 - export a proc file for general "ping" checking (5628)
1716 - fix "lfs check" to not block when the MDS is down (5628)
1718 - service request history (4965)
1719 - put {ll,lov,osc}_async_page structs in a single slab (4699)
1720 - create an "evict_client" /proc entry on OSTs, like the MDS has
1721 - fix mount usage message, return errors per mount(8) (5168)
1722 - change grep [] to grep "[]" in tests so they work in more UMLs
1723 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
1724 - remove ancient LOV_MAGIC_V0 EA support (5047)
1725 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
1726 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
1727 - allow readahead allocations to fail when low on memory (5383)
1728 - mmap locking landed again, after considerable improvement (2828)
1729 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
1731 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
1734 - send OST transaction number in read/write reply to free req (4966)
1735 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
1736 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
1737 - the watchdog thread now runs as interruptible (5246)
1738 - flock/lockf fixes (but it's still disabled, pending 5135)
1739 - don't use EXT3 constants in llite code (5094)
1740 - memory shortage at startup could cause assertion (5176)
1742 - reorganization of lov code
1743 - single portals codebase
1745 - add extents/mballoc support (5025)
1746 - direct I/O reads in the obdfilter (4048)
1747 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
1749 tbd Cluster File Systems, Inc. <info@clusterfs.com>
1752 - send OST transaction number in read/write reply to free req (4966)
1753 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
1754 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
1755 - the watchdog thread now runs as interruptible (5246)
1756 - handle missing objects in filter_preprw_read properly (5265)
1757 - unsafe list practices (rarely) led to infinite eviction loop (4908)
1758 - drop import inflight refcount on signal_completed_replay error (5255)
1759 - unlock page after async write error during send (3677)
1760 - return original error code on reconstructed replies (3761)
1761 - no transno return for symlink open, don't save no-trasno open (3440)
1763 - add pid to ldlm debugging output (4922)
1764 - bump the watchdog timeouts -- we can't handle 30sec yet
1765 - extra debugging for orphan dentry/inode bug (5259)
1767 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
1770 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
1771 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
1772 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
1773 - fix NULL dereference in /proc/sys/portals/routes (4827)
1774 - allow failed mdc_close() operations to be interrupted (4561)
1775 - stop precreate on OST before MDS would time out on it (4778)
1776 - don't send partial-page writes before EOF from client (4410)
1777 - discard client grant for sub-page writes on large-page clients (4520)
1778 - don't free dentries not owned by NFS code, check generation (4806)
1779 - fix lsm leak if mds_create_objects() fails (4801)
1780 - limit debug_daemon file size, always print CERROR messages (4789)
1781 - use transno after validating reply (3892)
1782 - process timed out requests if import state changes (3754)
1783 - update mtime on OST during writes, return in glimpse (4829)
1784 - add mkfsoptions to LDAP (4679)
1785 - use ->max_readahead method instead of zapping global ra (5039)
1786 - don't interrupt __l_wait_event() during strace
1788 - add software watchdogs to catch hung threads quickly (4941)
1789 - make lustrefs init script start after nfs is mounted
1790 - fix CWARN/ERROR duplication (4930)
1791 - return async write errors to application if possible (2248)
1792 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
1793 - print NAL number in %x format (4645)
1794 - update barely-supported suse-2.4.21-171 series (4842)
1795 - support for sles 9 %post scripts
1796 - support for building 2.6 kernel-source packages
1797 - support for sles km_* packages
1799 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
1802 - ignore -ENOENT errors in osc_destroy (3639)
1803 - notify osc create thread that OSC is being cleaned up (4600)
1804 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
1805 - reconstruct ost_handle() like mds_handle() (4657)
1806 - create a new thread to do import eviction to avoid deadlock (3969)
1807 - let lconf resolve symlinked-to devices (4629)
1808 - don't unlink "objects" from directory with default EA (4554)
1809 - hold socknal file ref over connect in case target is down (4394)
1810 - allow more than 32000 subdirectories in a single directory (3244)
1811 - fix blocks count for O_DIRECT writes (3751)
1812 - OST returns ENOSPC from object create when no space left (4539)
1813 - don't send truncate RPC if file size isn't changing (4410)
1814 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
1815 - bind to privileged port in socknal and tcpnal (3689)
1817 - rate limit CERROR/CWARN console message to avoid overload (4519)
1818 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
1819 - basic mmap support (3918)
1820 - kernel patch series update from b1_4 (4711)
1822 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
1825 - avoid crash during MDS cleanup with OST shut down (2775)
1826 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
1827 - don't use bad inodes on the MDS (3744)
1828 - dynamic object preallocation to improve recovery speed (4236)
1829 - don't hold spinlock over lock dumping or change debug flags (4401)
1830 - don't zero obd_dev when it is force cleaned (3651)
1831 - print grants to console if they go negative (4431)
1832 - "lctl deactivate" will stop automatic recovery attempts (3406)
1833 - look for existing locks in ldlm_handle_enqueue() (3764)
1834 - don't resolve lock handle twice in recovery avoiding race (4401)
1835 - revalidate should check working dir is a directory (4134)
1837 - don't always mark "slow" obdfilter messages as errors (4418)
1839 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
1842 - don't close LustreDB during write_conf until it is done (3860)
1843 - fix typo in lconf for_each_profile (3821)
1844 - allow dumping logs from multiple threads at one time (3820)
1845 - don't allow multiple threads in OSC recovery (3812)
1846 - fix debug_size parameters (3864)
1847 - fix mds_postrecov to initialize import for llog ctxt (3121)
1848 - replace config semaphore with spinlock (3306)
1849 - be sure to send a reply for a CANCEL rpc with bad export (3863)
1850 - don't allow enqueue to complete on a destroyed export (3822)
1851 - down write_lock before checking llog header bitmap (3825)
1852 - recover from lock replay timeout (3764)
1853 - up llog sem before sending rpc (3652)
1854 - reduce ns lock hold times when setting kms (3267)
1855 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
1856 - fix NULL deref and obd_dev leak on setup error (3312)
1857 - replace some LBUG about llog ops with error handling (3841)
1858 - don't match INVALID dentries from d_lookup and spin (3784)
1859 - hold dcache_lock while marking dentries INVALID and hashing (4255)
1860 - fix invalid assertion in ptlrpc_set_wait (3880)
1862 - add libwrap support for the TCP acceptor (3996)
1863 - add /proc/sys/portals/routes for non-root route listing (3994)
1864 - allow setting MDS UUID in .xml (2580)
1865 - print the stack of a process that LBUGs (4228)
1867 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
1870 - don't cleanup request in ll_file_open() on failed MDS open (3430)
1871 - make sure to unset replay flag from failed open requests (3440)
1872 - if default stripe count is 0, use OST count for inode size (3636)
1873 - update parent mtime/ctime on client for create/unlink (2611)
1874 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
1875 - free recovery state on server during a forced cleanup (3571)
1876 - unregister_reply for resent reqs (3063)
1877 - loop back devices mounting and status check on 2.6 (3563)
1878 - fix resource-creation race that can provoke i_size == 0 (3513)
1879 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
1880 - more debugging for page-accounting assertion (3746)
1881 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
1882 - avoid deadlock after precreation failure (3758)
1883 - fix race and lock order deadlock in orphan handling (3450, 3750)
1884 - add validity checks when grabbing inodes from l_ast_data (3599)
1886 - add /proc/.../recovery_status to obdfilter (3428)
1887 - lightweight CDEBUG infrastructure, debug daemon (3668)
1888 - change default OSC RPC parameters to be better on small clusters
1889 - turn off OST read cache for files smaller than 32MB
1890 - install man pages and include them in rpms (3100)
1891 - add new init script for (un)mounting lustre filesystems (2593)
1892 - run chkconfig in %post for init scripts (3701)
1893 - drop scimac NAL (unmaintained)
1895 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
1898 - clean kiobufs before and after use (3485)
1899 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
1900 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
1901 - change init script to fail more clearly if not run as root (1528)
1902 - allow clients to reconnect during replay (1742)
1903 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
1904 - don't do DNS lookups on NIDs too small for IP addresses (3442)
1905 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
1906 - fix cond_resched (3554)
1907 - only evict unfinished clients after recovery (3515)
1908 - allow bulk resend, prevent data loss (3570)
1909 - dynamic ptlrpc request buffer allocation (2102)
1910 - don't allow unlinking open directory if it isn't empty (2904)
1911 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
1912 - remove extraneous obd dereference causing LASSERT failure (3334)
1913 - don't use get_cycles() when creating temp. files on the mds (3156)
1914 - hold i_sem when setting i_size in ll_extent_lock() (3564)
1915 - handle EEXIST for set-stripe, set proper directory name (3336)
1917 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
1918 - fix ksocknal_fmb_callback() error messages (2918)
1920 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
1923 - don't copy lvb into (possibly NULL) reply on error (2983)
1924 - don't deref dentry after dput, don't free lvb on error (2922)
1925 - use the kms to determine writeback rpc length (2947)
1926 - increment oti_logcookies when osc is inactive (2948)
1927 - update client's i_blocks count via lvb messages (2543)
1928 - handle intent open/close of special files properly (1557)
1929 - mount MDS with errors=remount-ro, like obdfilter (2009)
1930 - initialize lock handle to avoid ASSERT on error cleanup (3057)
1931 - don't use cancelling-locks' kms values (2947)
1932 - use highest lock extent for kms, not last one (2925)
1933 - don't dereference ERR_PTR() dentry in error handling path (3107)
1934 - fix thread race in portals_debug_dumplog() (3122)
1935 - create lprocfs device entries at setup instead of at attach (1519)
1936 - common AST error handler, don't evict client on completion race (3145)
1937 - zero nameidata in detach_mnt in 2.6 (3118)
1938 - verify d_inode after revalidate_special is valid in 2.6 (3116)
1939 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
1940 - initialize RPC timeout timer earlier for 2.6 (3219)
1941 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
1942 - print nal/nid for unknown nid (3258)
1943 - additional checks for oscc recovery before doing precreate (3284)
1944 - fix ll_extent_lock() error return code for 64-bit systems (3043)
1945 - don't crash in mdc_close for bad permissions on open (3285)
1946 - zero i_rdev for non-device files (3147)
1947 - clear page->private before handing to FS, better assertion (3119)
1948 - tune the read pipeline (3236)
1949 - fix incorrect decref of invalidated dentry (2350)
1950 - provide read-ahead stats and refine rpc in flight stats (3328)
1951 - don't hold journal transaction open across create RPC (3313)
1952 - update atime on MDS at close time (3265)
1953 - close LDAP connection when recovering to avoid server load (3315)
1954 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
1955 - don't leak open file on MDS after open resend (3325)
1956 - serialize filter_precreate and filter_destroy_precreated (3329)
1957 - loop device shouldn't call sync_dev() for nul device (3092)
1958 - clear page cache after eviction (2766)
1959 - resynchronize MDS->OST in background (2824)
1960 - refuse to mount the same filesystem twice on same mountpoint (3394)
1961 - allow llmount to create routes for mounting behind routers (3320)
1962 - push lock cancellation to blocking thread for glimpse ASTs (3409)
1963 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
1964 - fix rare problem with rename on htree directories (3417)
1966 - allow default OST striping configuration per directory (1414)
1967 - fix compilation for qswnal for 2.6 kernels (3125)
1968 - increase maximum number of MDS request buffers for large systems
1969 - change liblustreapi to be useful for external progs like lfsck (3098)
1970 - increase local configuration timeout for slow disks (3353)
1971 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
1973 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
1976 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
1977 - don't overwrite extent policy data in reply if lock was blocked (2901)
1978 - drop filter export grants atomically with removal from device (2663)
1979 - del obd_self_export from work_list in class_disconnect_exports (2908)
1980 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
1981 - swab reply message in mdc_close, other PPC fixes (2464)
1982 - fix destroying of named logs (2325)
1983 - overwrite old logs when running lconf --write_conf (2264)
1984 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
1985 - fix race in target_handle_connect (2898)
1986 - mds_reint_create() should take same inode create lock (2926)
1987 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
1988 - don't close files for self_export to avoid uninitialized obd (2936)
1989 - allow MDS with the same name as client node (2939)
1990 - hold dentry reference for closed log files for unlink (2325)
1991 - reserve space for all logs during transactions (2059)
1992 - don't evict page beyond end of stripe extent (2925)
1993 - don't oops on a deleted current working directory (2399)
1994 - handle hard links to targets without a parent properly (2517)
1995 - don't dereference NULL lock when racing during eviction (2867)
1996 - don't grow lock extents when lots of conflicting locks (2919)
1998 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2001 - account for cache space usage on clients to avoid data loss (974)
2002 - lfsck support in lustre kernel code (2349)
2003 - reduce journal credits needed for BRW writes (2370)
2004 - orphan handling to avoid losing space on client/server crashes
2005 - ptlrpcd can be blocked, stopping ALL progress (2477)
2006 - use lock value blocks to assist in proper KMS, faster stat (1021)
2007 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2008 - recovery for initial connections (2355)
2009 - fixes for mds_cleanup_orphans (1934)
2010 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2011 - block all file creations until orphan recovery completes (1901)
2012 - client remove rq_connection from request struct (2423)
2013 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2014 - recovery timer race (2670)
2015 - mdc_close recovey bug (2532)
2016 - ptlrpc cleanup bug (2710)
2017 - mds timeout on local locks (2588)
2018 - namespace lock held during RPCs (2431)
2019 - handle interrupted sync write properly (2503)
2020 - don't try to handle a message that hasn't been replied to (2699)
2021 - client assert failure during cleanup after abort recovery (2701)
2022 - leak mdc device after failed mount (2712)
2023 - ptlrpc_check_set allows timedout requests to complete (2714)
2024 - wait for inflight reqs when ptlrpcd finishes (2710)
2025 - make sure unregistered services are removed from the srv_list
2026 - reset bulk XID's when resending them (caught by 1138 test)
2027 - unregister_bulk after timeout
2028 - fix lconf error (2694)
2029 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2030 - readahead locks pages, leaves pending causing memory pressure (2673)
2031 - increase OST request buffers to 4096 on large machines (2729)
2032 - fix up permission of existing directories in simple_mkdir (2661)
2033 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2034 - don't assign transno to errored transactions (2742)
2035 - don't delete objects on OST if given a bogus objid from MDS (2751)
2036 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2037 - if rq_no_resend, then timeout request after recovery (2432)
2038 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2039 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2040 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2041 - MDS recovery shouldn't delete objects that it has given out (2730)
2042 - if enqueue arrives after completion, don't clobber LVB (2819)
2043 - don't unlock pages twice when trigger_group_io returns error (2814)
2044 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2045 - don't write pages to disk if there was an error (1450)
2046 - don't ping imports that have recovery disabled (2676)
2047 - take buffered bytes into account when balancing socknal conn (2817)
2048 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2049 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2050 - remove little-endian swabbing of llog records (1987)
2051 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2052 - retry reposting request buffers if they fail (1191)
2053 - grow extent at grant time to avoid granting a revoked lock (2809)
2054 - lock revoke doesn't evict page if covered by a second lock (2765)
2055 - disable VM readahead to avoid reading outside lock extents (2805)
2057 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2058 - updated kernel patches for hp-2.4.20 kernel (2681)
2060 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2063 - fix truncated write corruption (2366)
2064 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2066 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2067 - ensure len(uuid) < 37 in lmc (1171)
2068 - fix ia64 OOPS in llog_test (2255)
2069 - zero end of page at obdfilter for partial page writes (2648)
2070 - don't leave stale dentries around after renames (bug 2428)
2071 - fix timeouts when evicting a client with a single lock held (2642)
2072 - set deadline for the initial HELLO message to drain (2634)
2073 - print out dotted-quad IP addresses in the socknal (2302)
2075 - additional debugging for MDS client eviction problem (2443)
2076 - fix mkfsoptions support for osts (2603, 2604)
2078 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2081 - add series for the vanilla 2.6.0 kernel
2082 - add series for the vanilla 2.4.24 kernel
2083 - add series for a cray x86/64 UL kernel drop
2084 - fix xattr patches for the vanilla 2.4.19 series
2086 - generate true UUIDs in lmc (1171)
2087 - have portals stack dumping break in UML (2466)
2088 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2089 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2090 - wake callback waiting threads on client eviction (2460)
2091 - Add --ptldebug and --subsystem to lmc (1719)
2092 - update assertion to allow safe interrupt allocation
2093 - set rq_no_resend for cancel requests (2432)
2094 - recalculate ptlrpcd timeout after resend (2494)
2095 - call vfs_rmdir when removing pending directories (2368)
2096 - fix renaming a file to itself (2429)
2097 - lmc creates a default one-stripe lov (2454)
2098 - expand procfs space to handle large clusters (2326)
2099 - increase UML stack to avoid overflow
2100 - update lconf's list of debug and subsystem masks
2101 - fix lfs find --obd (2510)
2102 - /proc tunable for disabling filter read caching (2591)
2103 - stop rpm packages from altering slapd.conf (2301)
2104 - disable nagle in the socknal under 0conf (2578)
2105 - choose mds inode size based on stripe count (2572)
2106 - fix kernel-source rpm problems (2516)
2108 - add --disable-doc to avoid pdf generation (2421)
2109 - update documentation, tests, type-os, comments
2110 - avoid format warnings on ia64
2111 - remove the TOE NAL
2112 - tiny code cleanups by removing unused fields
2114 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2117 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2118 - remove the most blatant lies from BUILDING file (2371)
2119 - change default debug level to reasonable production setting
2120 - reduce client side cache size to reduce cache flush time
2121 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2122 - make TCP zerocopy and pinger support enabled by default (2476)
2123 - sync writes completed after process exits caused crashes (2319)
2124 - maintain correct mount count on the MDS (2356)
2125 - backout 1557, because 2316 wasn't really fixed
2126 - better file I/O statistics gathering in /proc
2127 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2128 - another kernel patch to fix zero-copy TCP function export
2129 - don't take duplicate lock when processing re-sent getattr (2420)
2130 - lctl uses obd_self_export instead of creating new conn (2353)
2131 - MDS/OST recovery case which requires object creation asserted (2425)
2132 - move lfs from /usr/sbin to /usr/bin in packages
2133 - fix race between mds_client_add and mds_client_free (2417)
2134 - use kmalloc instead of slabs in portals (2430)
2135 - don't create duplicate records when a failover MDS is present (2442)
2136 - remove unnecessary mount age check (2332)
2137 - don't remove directory inodes from locks prematurely (2451)
2138 - don't break if MDS service name is the same as hostname (2103)
2139 - fix races in client write RPC generation when cache full (2482)
2141 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2144 - remove now-unused request->rq_obd (278)
2145 - if an allocation fails, print out how much memory we've used (1933)
2146 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2147 - add the "configurable stack size" patch to most series files (1256)
2148 - ability to write large log records, for 100+ OST configs (2306)
2149 - fix NULL deref when filter_prep fails (2314)
2150 - fix operator precedence error in filter_sync
2151 - dynamic allocation of socknal TX descriptors (2315)
2152 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2153 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2154 - ensure that CWARN messages in Portals always get to the syslog
2155 - __init/__exit are not for prototype decls (ldlm_init/exit)
2156 - x86-64 compile warning fixes
2157 - fix gateway LMC keyword conflict (2318)
2158 - fix MDS lock inversions in getattr/reint paths (1844)
2159 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2160 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2161 - DLM race condition prevented some lock evictions (2328)
2162 - ENOMEM detection and retry on socknal sends (2230)
2163 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2164 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2165 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2166 - better error messages when a client is rejected during recovery (1505)
2167 - avoid cancelling locks which were never granted, after failure (2330)
2168 - fix i_sem/journal inversion in mds_client_add (2333)
2169 - fix truncate/getattr lock cycle deadlock (2334)
2170 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2171 - fix two rare exit paths which could leak an l_lock() ref (2321)
2172 - fencepost error in MDS/OST orphan recovery (2226)
2173 - make log record alignment 8 bytes (1988)
2174 - lstripe now fails when requested offset > ost_count (2237)
2175 - ensure that all kernel series have a complete list.h (1607)
2176 - fix crashes in special-file operations (2316)
2177 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2179 - allow configurable automake binary, for testing new versions
2180 - small update to the lfs documentation
2182 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2184 * fix negative export reference count in fsfilt_sync (2312)
2186 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2187 * release candidate 0.9.1
2189 - orphans are moved into the PENDING directory for possible recovery
2190 - replayed opens now open by fid for orphan/rename safety (1042)
2191 - last close of an orphan inode generates a transno (683)
2192 - chdir() and mount() now pin the directory entry (1020)
2193 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2194 - discard very old requests without processing them (1502)
2195 - remove some common, well-understood CERRORs (1505)
2196 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2197 - clear "grant" flags in OST replies until OST grant code lands (1644)
2198 - fix read performance by not clobbering i_blksize on client (1598)
2199 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2200 - make LRU size a /proc tunable, clears locks when reduced (707)
2201 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2202 - server locks take references on exports now (1558)
2203 - build fixes for 2.4.20-rh trees (1663)
2204 - return an error from lov_create if all OSCs are inactive (1751)
2205 - fix import levels when a reconnect happens without a timeout (1597)
2206 - exit early from mds_open if we get a lookup error (1749)
2207 - partial page read at EOF wouldn't wait for disk before sending (1642)
2208 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2209 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2210 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2211 - write/truncate lock inversion (1639)
2212 - Don't auto-load obdclass, portals modules during cleanup (1495)
2213 - fix timestamps from jumping to "now" (1763)
2214 - extra journal assertions (1648)
2215 - add an extra multiunlink test (1771)
2216 - fix read_record/write_record API (1776)
2217 - fix leak of offset_extent, possible incorrect i_size later (1772)
2218 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2219 - Debugging for the kqswnal_get_idle_tx problems (1820)
2220 - Allow recovery to be attempted multiple times (1536)
2221 - Write out MDS last_rcvd file after it is first created (1600)
2222 - Fix tx_descriptor leak in failed transmit situations (1827)
2223 - ext3 journaling fixes for assertion failure after IO error (1871)
2224 - class_export_put() on freed export after completion AST error (1896)
2225 - Fix revalidate looping in VFS (1322)
2226 - Don't access a freed export during MDS_REINT timeout (1521)
2227 - Add open-unlink recovery support on the MDS (1673,1764)
2228 - Return an error if no MDS data was read from last_rcvd (1946)
2229 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2230 - Don't setup a disk device that doesn't match exported UUID (317)
2231 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2232 - avoid committing NULL handle in force close
2233 - local.sh is now a one-stripe LOV configuration
2234 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2235 - fix ext3 htree duplicate directory entry corruption (1516)
2236 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2237 - update to new LOV EA format (2097)
2238 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2240 2003-06-15 Phil Schwan <phil@clusterfs.com>
2243 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2244 - per-import/export recovery handling (958, 931, 959)
2245 - multiple last-rcvd slots, for serving multiple FSes (949)
2246 - connections are again shared between multiple imp/exports (963, 964)
2247 - "umount -f" would hang if any requests needed to be sent (393, 978)
2248 - avoid pinning large req buffer by copying for queued messages (989)
2249 - add "uuid" to "lctl device" command to help upcalls (991)
2250 - "open" RPCs with transnos would confuse recovery counters (1037)
2251 - do proper endian conversion of all wire messages (288, 340, 891)
2252 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
2253 - call waiting locks callback from LDLM recovery thread (1127, 1151)
2254 - fix ptlrpc_connection leak in target_handle_connect (1174)
2255 - fix import refcounting bug in OST and MDS cleanup (1134)
2256 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
2257 - fix very unlikely obd_types race condition (501)
2258 - remove osc_open hack for echo_client (1187)
2259 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
2260 - a failure in read_inode2 leads to deadlock (1139)
2261 - cancel ack-locks as soon as transaction is committed (1072)
2262 - fix major leaks and crashes in the bulk I/O path (937, 1057)
2263 - make sure to commitrw after any preprw to avoid deadlock (1162)
2264 - failing to execute a file in a lustre FS would lock inode (1203)
2265 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
2266 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
2267 - obd_uuid2tgt would walk off the end of the list (1255)
2268 - on IA64 the osc would give portals incorrect bulk size (1258)
2269 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
2270 - fix lock inversion caused by new llite matching code (1282)
2271 - limit the number of dirty pages on a client to 10MB (1286)
2272 - timed out locks were not being corrected cancelled (1289)
2273 - fix O_DIRECT above 4GB on IA-32 (1292)
2274 * major user-visible changes
2275 - fail out/fail over policy now controlled by the upcall (993)
2277 - add OBD_PING to check server availability and failure (954)
2278 - lustre messages are now sent in sending host order (288, 340, 891)
2279 - add eadatalen to MDS getattr reply (340)
2280 - OST read replies may contain second buffer, with per-page status (593)
2282 2003-03-11 Phil Schwan <phil@clusterfs.com>
2285 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
2286 - failed open()s could cause deadlock; fixed (867, 869)
2287 - stop cancelling OST locks when files are closed (481)
2288 - overlapping XID spaces caused network corruption (851, 853)
2289 - fix unsafe fsfilt counter arithmetic; change to atomic_t
2290 - setattr_raw added, to do single-RPC, server-side setattrs
2291 - lmc/lconf syntax change for OST UUIDs
2292 - fix crashy race condition between ptlrpc_free_req and osc_close
2293 - don't use request in mdc_enqueue if we hit a timeout (889)
2294 - don't set the inode i_size for regular files from the MDS (896)
2295 - handle out of order completion AST (842)
2296 - don't LBUG if a lock request times out after receiving AST (913)
2297 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
2298 - if a bad lock AST arrives, send an error instead of dropping entirely
2299 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
2300 - fix leak in bulk IO when only partially completed (899, 900, 926)
2301 - fix O_DIRECT for ia64 (55)
2302 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
2303 - C-z after timeout could hang a process forever; fixed (977)
2305 - client-side I/O cache (678, 924, 929, 941, 970)
2307 - READPAGE and SETATTRs which don't take server-side locks get
2310 2003-02-11 Phil Schwan <phil@clusterfs.com>
2313 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
2314 - Fix fsfilt_extN_readpage() to read a full page of directory
2315 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
2316 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
2317 - Fix imbalanced LOV object allocation and out-of-bound access (469)
2318 - Most intent operations were removed, in favour of a new RPC mode
2319 that does a single RPC to the server and bypasses most of the VFS
2320 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
2321 - Aggressively cancel local locks on DLM servers
2322 - mds_reint_unlink sends EA to the client if it's the last nlink.
2323 client uses that EA to unlink OST objects.
2324 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
2325 - recursive symlinks were fixed (439)
2326 - fixed NULL deref in DEBUG_REQ
2327 - filter_update_lastobjid no longer calls sync, which annoyed extN
2328 - fixed multi-client small-writes to a single file problem (445)
2329 - fixed mtime updates during file writes (607)
2330 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
2331 - fixed bug in obd_brw_read/write() (under guise of testing 367)
2332 - fixed Linux OST size reporting problem (444, 656)
2333 - OST now updates object mtime with writes or setattr (607, 619)
2334 - client verifies file size before zeroing page past EOF (445)
2335 - OST now writes last allocated objid to disk with allocation (108)
2336 - LOV on echo now works (409)
2338 - mds_reint_unlink sends a new buffer, with the EA included. this
2339 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
2340 set if a regular file was being unlinked, and it was the last link
2341 - use PtlGet from the target for bulk writes (315)
2342 - OST now updates object mtime with writes or setattr (607, 619)
2343 - LDLM now has a grant-time callback to revalidate locked items, if
2345 - Many MDS operations were reorganized to combat race conditions
2347 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
2348 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
2350 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
2353 - Fully reactivate OST imports after reconnection (512, others)
2354 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
2355 - More graceful error handling for truncating on dead OST (515)
2356 - Don't error out unless we're actually accessing dead stripes (474)
2357 - Fix garbage sizes when stripes are missing (410)
2358 - LRU counters were broken, causing constant lock purge (433, 432)
2359 - garbage on read from stripes with failed OSTs (441)
2360 - mark OSCs as active before reconnecting during recovery (438)
2361 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
2362 - lfind did not preserve OST order in output (443)
2363 - symlinks cause hung clients, incorrect data (439)
2364 - stop dereferencing request after dropping refcount (457)
2365 - don't LASSERT(spin_is_locked) on non-SMP (455)
2366 - fixes for many rename() bugs
2367 - fstat didn't correctly synchronize attributes (399)
2368 - server must handle lock cancellation during blocking AST prep (487)
2369 - bulk descriptors were free()d too soon (511)
2370 - fix paths in lconf, which would load incorrect modules (451, 507)
2371 - fix confusing lconf 'host not found' error message (386)
2372 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
2373 - fix race condition in mdc_blocking_ast() for inode access (526)
2374 - fix lov_unpackmd() unpacking wrong number of stripes (537)
2375 - fix lov_set_osc_active() marking wrong OSC inactive (440)
2376 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
2377 - fix multiple writes of stripe MD to MDS (358, maybe 519)
2378 - fix lstripe in several ways (kernel side) (527)
2379 - fix request leak in ldlm_cli_enqueue (262)
2380 - incorrect OSC was marked inactive after OST failure
2381 - call mds_fs_cleanup before unmounting filesystem (524)
2382 - fix races between taking ns_lock and ldlm_lock_change_resource
2383 - fix races updating LOV export open file list
2384 - fix lov_enqueue error path, avoid decref-ing bad lock handle
2385 - fix recovery NULL deref in ldlm_cli_cancel_unused
2386 - fix some DLM races by using new hash table for lock handles (419)
2387 - permit the client to specify desired inodes, at replay
2388 - duplicate requests when we queue them for replay reintegration
2389 - fix last_rcvd offset calculation
2390 - sync after each recovered transaction, so we always make progress
2391 - never, not always, ERESTART requests without transnos
2392 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
2394 - skip replay if the MDS says that the client is already connected
2395 - don't check for a recovery-enabled export to match lctl's UUID
2396 - don't INC_USE_COUNT for phantom exports
2397 - don't crash when cleaning up phantom exports (567)
2398 - don't double-finish or set replay data for errored mdc_open requests
2399 - abort requests when they time out, so we don't get old replies
2400 - send/receive replies for AST messages again
2401 - if the client says that it doesn't have the lock, cancel it on the
2403 - if we timeout during I/O, don't try to cancel an in-use lock; instead
2404 mark it as destroyed, it will all work out when decref is called
2405 - fix module use counts (22, 581)
2407 - ASTs now expect a reply (server cancels lock on error reply)
2409 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
2412 - fix many simultaneous client startup (392)
2413 - fix dentry->d_it clobbering
2414 - credentials weren't being shipped for readdir/getattr operations
2415 - remove invalid assertions triggered during some concurrent MD
2417 - proper Lustre versions added (336, 389)
2418 - fix memory leak for create error case (398)
2419 - fix LOV locking bug that would get cli/srv out of sync
2420 - fix echo client over LOV (409)
2421 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
2422 - fix double-O_EXCL intent crash (424)
2423 - avoid sending multiple lock CANCELs (352)
2425 - MDS can do multi-client recovery (modulo bugs in new code)
2427 - many updates, edits, cleanups
2429 2002-11-18 Phil Schwan <phil@clusterfs.com>
2432 - fix null d_it dereference (346)
2433 - fix full OST/dbench hang (333)
2434 - fix permission problem with file removal (286)
2435 - fix removal of OSCs from LOV when they fail
2436 - fix NULL deref during bulk timeout (214)
2437 - fix problems related to multiple filesystems on one MDS (241)
2438 - fixed serious subtle metadata locking bugs
2439 - free locks on clients when inodes are removed due to memory
2441 - fix inode pointer in lock data (285)
2442 - partial support for multiple MDS on a single host (241)
2443 - data locks weren't cancelled at clear_inode time (290, 311)
2444 - intent locks could lead to unbounded lock growth (205)
2445 - added a maximum lock count, an LRU list, and a flusher
2446 - fix multiple rename (365)
2447 - properly abstracted the echo client
2448 - OSC locked 1 byte too many; fixed
2449 - rewrote brw callback code:
2450 - fixed recovery bugs related to LOVs (306)
2451 - fixed too-many-pages-in-one-write crash (191)
2452 - fixed (again) crash in sync_io_timeout (214)
2453 - probably fixed callback-related race (385)
2455 - Add capability to MDS protocol
2456 - LDLM cancellations and callbacks on different portals
2458 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
2461 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
2462 - defer open object destruction to close time (601981, 138)
2463 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
2464 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
2465 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
2466 - don't call obd_setattr in ll_file_release for destroyed objects
2467 * protocol change to lustre_msg: move |version| and add |flags|
2468 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
2469 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
2470 * added replay of create, unlink, link and rename operations during
2471 MDS failover; recovery should be much more robust now
2472 * remove failed OSCs from LOVs (only lov_create uses this so far)
2473 * the lustre-HOWTO was brought (more) up to date (582544)
2475 2002-10-23 Phil Schwan <phil@clusterfs.com>
2478 - in-use dentries weren't being reused properly (617851)
2479 - prevent multiple LDLM setup (599178)
2480 - fix LOV size calculations for truncate (617853)
2481 - fix client handling of MDS intent errors (POSIX)
2482 - fix permission bug in lovstripe.c test (624321)
2483 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
2484 - truncate past end of file could corrupt data
2485 - proper cleanup after timeouts, crashes, etc (592524, 550815)
2486 - a race in recovery could return ETIMEDOUT to apps (623947)
2487 - building outside the source directory was fixed
2488 * the lustre-HOWTO was brought (more) up to date (582544)
2489 * major progress was made on recovery functionality
2491 2002-10-10 Phil Schwan <phil@clusterfs.com>
2494 - recovery deadlock fix
2495 - rm -rf causes LBUG fix (617817)
2496 - file open by multiple tasks fix (618962)
2497 - directory permissions bugs (602707 and 620007)
2498 - journal_stop fixed with locking (611313)
2499 - O_APPEND failures resolved (618273, perhaps 614459)
2500 - lconf PATH fix (619770)
2501 - IA64 build fix (621450)
2502 - RPC buffer sizes scale with amount of memory
2504 2002-10-01 Phil Schwan <phil@clusterfs.com>
2507 - locks would be cancelled without throwing away data pages,
2508 resulting in inconsistent data (605627)
2509 - inode attributes were not always being refreshed (605627, 612449)
2510 - lconf now continues to cleanup after lctl reports an error
2511 - MDS now enforces user permissions (602707)
2512 - lprocfs cleanup fixed, but not yet enabled (614157)
2513 - fixed infinite server hang, should a client not respond to an AST
2514 - avoid going into recovery if user calls readlink() with a buffer
2515 that's too small (613941)
2516 - AST RPCs no longer require replies (614867) -- this may be changed
2517 - don't crash server if client sends an IOV that's too big (611336)
2518 - fixed lock conversion deadlock (611892)
2519 - fixed the following of symlinks (614622)
2520 * recovery: the server can remove locks from a client that dies, other
2521 clients can make progress
2522 * more extN patch fixes
2523 * compile-time configurable ptlrpc buffer allocations
2525 - collaborative read cache document
2526 - Lustre Lite Performance CDR document-in-progress
2528 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
2531 - fix typo in patch-2.4.18
2533 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
2536 - clear ptlrpc request each time in handle_incoming_request()
2537 - unlink of files now destroys the object on the OST
2539 2002-09-19 Peter Braam <braam@clusterfs.com>
2541 * add hard link support
2542 * change obdfile creation method
2543 * kernel patch changed
2545 2002-09-19 Peter Braam <braam@clusterfs.com>
2548 - stack overflow bug in extN fixed
2550 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
2552 * documentation updates
2553 - add man pages for config tools
2554 - update tests/README to describe testing with new config tools
2555 - finish metadata API descriptions
2556 * bug fixes and cleanups
2557 - statfs workaround for 16TB limit
2558 - LOV stripe allocation improved, can stripe on subset of OSTs
2559 - LOV file size/IO offset was wrong for files > 4GB in size
2560 - object EA data was being dropped, caused files to be unreadable
2561 - memory overflow with non-LOV OST caused memory corruption
2562 - fixed regression tests to work with new config tools, obdfilter
2563 - fixed bug when directory size became larger than 1 block
2564 - fixed bug (for single client case) when PWD was deleted
2565 - invalidate local directory pages when doing intent-based ops
2566 - avoid LDLM oops when lock callback contained bad data
2568 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
2570 * documentation updates
2571 * bug fixes and cleanups
2572 - configuration tools
2575 - 64-bit compile warnings
2576 - 64-bit internal statfs data
2578 * test_brw on persistent OST devices
2580 * lprocfs (disabled)
2582 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
2584 * documentation updates
2585 * bug fixes and cleanups
2586 * configuration tools
2588 2002-08-30 Peter J. Braam <braam@clusterfs.com>
2591 * many small fixes to 0_5_4
2592 * io/network handling
2593 * thinkos in MDS operations
2595 2002-08-24 Peter J. Braam <braam@clusterfs.com>
2598 * crucial basic fixes to 0.5.3
2599 * IOR, Iozone work over Elan
2602 2002-08-07 Phil Schwan <phil@clusterfs.com>
2603 * version 0_5_3, our first alpha
2604 * we use the new Portals iovs
2605 * documentation updates
2606 * bug fixes and cleanups
2607 * small changes in the DLM wire protocol
2609 2002-07-25 Peter J. Braam <braam@clusterfs.com>
2610 * version 0_5_1 with some initial stability,
2611 * locking on MD and file I/O.
2612 * documentation updates
2613 * several bug fixes since 0.5.0
2614 * small changes in wire protocol
2616 2002-07-18 Phil Schwan <phil@clusterfs.com>
2618 * delivered as Lustre Light Alpha
2619 * fixed a crash after handling invalid MDS requests
2620 * fixed directory pages for architectures with non-4k pages sizes
2622 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
2623 * release version v0_4_4
2624 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
2626 * Fixes a number of interruption problems with OST operations.
2627 * Update documentation for portals header changes
2628 * Move all wire protocol structs/defines to lustre_idl.h
2629 * Fixes symlink length bug.
2630 * Add tcpdump to repository.
2632 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
2633 * release version v0_4_3
2634 * Fixes statfs for inodes on extN.
2635 * Fixes bug in runtests which would delete /etc/hosts.
2636 * Use 64-bit object IDs wherever possible (not into VFS though)
2637 Remove ost_get_info, which is unused by lustre, and out of date.
2639 2002-07-03 Peter Braam <braam@clusterfs.com>
2640 * release version v0_4_2 Fixes a lookup error (type not passed)
2641 * move forward to head of Portals
2642 * move forward to latest Lustre kernel
2644 2002-06-25 Peter Braam <braam@clusterfs.com>
2645 * release version v0_4_1. Hopefully stable on single node use.