4 2.6.9-55.0.9.EL (RHEL 4),
5 2.6.16.46-0.14 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
16 Description: Remove client patches from SLES 10 kernel.
17 Details : This causes SLES 10 clients to behave as patchless clients
18 even on a Lustre-patched (server) kernel.
20 Severity : enhancement
22 Description: self-adjustable client's lru lists
23 Details : use adaptive algorithm for managing client cached locks lru
24 lists according to current server load, other client's work
25 pattern, memory activities, etc. Both, server and client
26 side namespaces provide number of proc tunables for controlling
29 Severity : enhancement
31 Description: light-weight GSS support
32 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
33 and reduce performance overhead.
35 Severity : enhancement
37 Description: Linux keyring support
38 Details : Support using service of Linux keyring for Lustre GSS internal
39 context refresh/cache mechanism.
43 Description: Fix errors in lfs documentation
44 Details : Fixes man pages
48 Description: don't use GFP_* in generic Lustre code.
49 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
53 Description: obdclass is limited by single OBD_ALLOC(idarray)
54 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
58 Description: warnings with build patchless client with vanila 2.6.19 and up
59 Details : change old ctl_table style and replace ctl_table/ctl_table_header
60 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
64 Frequency : Only for SLES
65 Description: sanity_quota fail test_1
66 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
67 file, which makes TSTID[2] inunique.
71 Frequency : Only for SLES9
72 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
73 Details : Update target/ChangeLog/which_patch .
77 Frequency : Only for test_10 in sanity-quota.sh
78 Description: a bug in quota test code
79 Details : checking a test flag in a wrong place
83 Frequency : Only for test_13 in sanity-quota.sh
84 Description: a bug in quota test script
85 Details : shouldn't check overfull allocated quota
89 Description: jbd statistics
90 Details : Port older jbd statistics patch for sles10
92 Severity : enhancement
94 Description: Add printing inode info into message about error in writepage.
99 Description: reply_lock_interpret crash due to race with it and lock cancel.
100 Details : Do not replay locks that are being cancelled. Do not reference
101 locks by their address during replay, just by their handle.
105 Frequency : When flocks are used.
106 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
107 Details : Flock locks might destroy just granted lock if it could be merged
108 with another existing flock, this is done in completion handler,
109 so teach ldlm_cli_enquque_fini that this is a valid case for
115 Description: Oops in read and write path when failing to allocate lock.
116 Details : Check if lock allocation failed and return error back.
120 Description: lstripe command fails for valid OST index
121 Details : The stripe offset is compared to lov->desc.ld_tgt_count
122 instead of lov->desc.ld_active_tgt_count.
126 Description: sanity.sh failed test 103
127 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
132 Description: df doesn't work properly if diskfs blocksize != 4K
133 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
137 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
138 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
142 Description: obd_health_check_timeout too short
143 Details : set obd_health_check_timeout as 1.5x of obd_timeout
147 Description: llapi_file_create() does not allow some changes
148 Details : add llapi_file_open() that allows specifying the mode and
149 open flags, and also returns an open file handle.
153 Description: merge and cleanup kernel patches.
154 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
158 Description: Add journal checksum support.(Kernel part)
159 Details : The journal checksum feature adds two new flags i.e
160 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
161 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
162 indicates that the commit block contains the checksum for
163 the blocks described by the descriptor blocks. Now commit
164 record can be sent to disk without waiting for descriptor
165 blocks to be written to disk. This behavior is controlled
166 using JBD2_FEATURE_ASYNC_COMMIT flag.
170 Description: OSS needs mutliple precreate threads
171 Details : Add ability to start more than one create thread per OSS.
175 Description: Sanity.sh test_65k failed: setstripe should have succeeded
176 Details : execute lfs setstripe on client
180 Description: mds_obd_create error creating tmp object
181 Details : When the user sets quota on root, llog will be affected and can't
182 create files and write files.
185 Frequency : Always on ia64 patchless client, and possibly others.
187 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
188 Details : This allows the patchless client to be loaded on architectures
193 Description: RedHat Update kernel for RHEL5
194 Details : Add the kernel config file for RHEL5.
198 Description: RedHat Update kernel for RHEL5
199 Details : Modify the kernel config file more closer RHEL5.
203 Description: Build failure against Centos5 (RHEL5)
204 Details : Define PAGE_SIZE when it isn't present.
208 Description: test_7 on sanity-quota.sh can't be used on separate machines
209 Details : add facet to handle it
213 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
214 Details : with NFS, the anon dentry's parent was set to itself in
215 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
216 corresponding dentry other than use rec->ur_name.
218 Severity : enhancement
220 Description: lfs setstripe enhancement
221 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
225 Description: testing performance impact of enabling checksumming
226 Details : enable checksum by default, allow --disable-checksum
227 configure option and "-o nochecksum" mount option
231 Description: lustre support for RHEL5
232 Details : Add support for RHEL5.
236 Description: Client eviction due to failover config
237 Details : after a connection loss, the lustre client should attempt to
238 reconnect to the last active server first before trying the
239 other potential connections.
243 Description: when mds and osts use different quota unit(32bit and 64bit),
244 quota will be released repeatly.
245 Details : void sending multiple quota reqs to mds, which will keep the
246 status between the reqs.
250 Description: osts not allocated evenly to files
251 Details : change the condition to increase offset_idx
254 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
256 Description: Data corruption for OSTs that are formatted larger than 2TB
258 Details : When generating the bio request for lustre file writes the
259 sector number would overflow a temporary variable before being
260 used for the IO. The data reads correctly from Lustre (which
261 will overflow in a similar manner) but other file data or
262 filesystem metadata may be corrupted in some cases.
266 Description: Tune the kernel for good SCSI performance.
267 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
268 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
273 Description: rewrite ext2-derived code in obdclass/uuid.c
274 Details : rewrite inherited code (uuid parsing code from ext2 utils)
275 from scratch preserving functionality.
279 Description: "lfs find -obd UUID" prints directories
280 Details : "lfs find -obd UUID" will return all directory names instead
281 of just file names. It is incorrect because the directories
282 do not reside on the OSTs.
286 Description: use i_size_read and i_size_write in 2.6 port
287 Details : replace inode->i_size access with i_size_read/write()
290 Frequency : when removing large files
292 Description: scheduling issue during removal of large Lustre files
293 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
294 It causes scheduling issues when removing large files (17TB in the
298 Frequency : only with liblustre clients on XT3
300 Description: evictions taking too long
301 Details : allow llrd to evict clients directly on OSTs
304 Frequency : only on ppc
306 Description: /proc/fs/lustre/devices broken on ppc
307 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
308 We should make sure the type of 'index' is loff_t and then cast
309 explicitly as needed below. Do not assign an explicitly cast
313 Frequency : only for rhel5
315 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
316 Details : Modify the target file & which_kernel.
318 Severity : enhancement
320 Description: omit set fsid for export NFS
321 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
322 as FS_REQUIRES_DEV for avoid problems with generate fsid.
326 Frequency : Always, for kernels after 2.6.16
327 Description: Fix warning idr_remove called for id=.. which is not
329 Details : Last kernels save old s_dev before kill super and not allow
330 to restore from callback - restore it before call kill_anon_super.
334 Description: buffer overruns could theoretically occur
335 Details : llapi_semantic_traverse() modifies the "path" argument by
336 appending values to the end of the origin string, and a
337 overrun may occur. Adding buffer overrun check in liblustreapi.
341 Description: client sometimes tried revalidate root dentry.
342 Details : not revalidate dentry if it root dentry.
346 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
347 Details : Modify vm-tunables-rhel4.patch.
351 Description: Kernel config for 2.6.18-vanilla.
352 Details : Modify targets/2.6-vanilla.target.in.
353 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
354 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
355 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
356 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
360 Description: improve handling recoverable errors
361 Details : if request processig with error which can be recoverable on server
362 request should be resend, otherwise page released from cache and
367 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
368 Details : Modify target file & which_patch.
369 A flaw was found in the IA32 system call emulation provided
370 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
371 value could be stored in the %RAX register, which could trigger an
372 out-of-bounds system call table access. An untrusted local user
373 could exploit this flaw to run code in the kernel
374 (ie a root privilege escalation). (CVE-2007-4573).
378 Description: Remove 2.4 kernel checks lustre 1.8
379 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
380 quota_interface.c in quota subsystem.
381 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
383 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
384 lvfs_linux.c upcall_cache.c in lvfs subsystem.
388 Description: O_DIRECT bypasses client statistics.
389 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
390 read_ahead_stats were updated. copied the stats section from
391 osc_send_oap_rpc() into async_internal().
395 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
399 Description: change order of libsysio includes
400 Details : '#include sysio.h' should always come before '#include xtio.h'
404 Description: handle "serious error: objid * already exists" more gracefully
405 Details : If LAST_ID value on disk is smaller than the objects existing in
406 the O/0/d* directories, it indicates disk corruption and causes an
407 LBUG(). If the object is 0-length, then we should use the existing
408 object. This will help to avoid a full fsck in most cases.
410 Severity : enhancement
412 Description: adapt the lustre_config script to support the upgrade case
413 Details : Add "-u" option for lustre_config script to support upgrading 1.4
414 server targets to 1.6 in parallel.
418 Description: grant fixes to avoid grant_space > avail_space
419 Details : In filter_check_grant, for non_grant cache write, we should
420 check the left space by if (*left > ungranted + bytes), instead
421 of (*left > ungranted), because only we are sure the left space
422 is enough for another "bytes", then the ungrant space should be
423 increase. In client, we should update cl_avail_grant only there
424 is OBD_MD_FLGRANT in the reply.
428 Description: Update RHEL 4 kernel to fix local root privilege escalation.
429 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
430 described in CVE-2007-4573. This problem could allow untrusted
431 local users to gain root access.
435 Description: Remove 2.4 kernel checks lustre 1.8
436 Details : Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
437 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
440 --------------------------------------------------------------------------------
442 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
444 * Support for kernels:
445 2.6.5-7.283 (SLES 9),
446 2.6.9-55.EL (RHEL 4),
447 2.6.16.46-0.14 (SLES 10),
448 2.6.18.8 vanilla (kernel.org)
449 * Client support for unpatched kernels:
450 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
451 2.6.16 - 2.6.22 vanilla (kernel.org)
452 * Due to recently discovered recovery problems, we do not recommend
453 using patchless RHEL 4 clients with this or any earlier release.
454 * Recommended e2fsprogs version: 1.39.cfs8
455 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
456 * Starting with this release, the ldiskfs backing filesystem required
457 by Lustre is now in its own package, lustre-ldiskfs. This package
458 should be installed. It is versioned separately from Lustre and
459 may be released separately in future.
463 Description: block reactivating mgc import until all deactivates complete
464 Details : Fix race when failing back MDT/MGS to itself (testing)
466 Severity : enhancement
468 Description: add optional extra BUILD_VERSION info
469 Details : add a new environment variable (namely LUSTRE_VERS) which allows
470 to override the lustre version.
473 Frequency : 2.6.18 servers only
475 Description: ll_kern_mount() doesn't release the module reference
476 Details : The ldiskfs module reference count never drops down to 0
477 because ll_kern_mount() doesn't release the module reference.
482 Description: server LBUG when using old ost_num_threads parameter
483 Details : Accept the old ost_num_threads parameter but warn that it
484 is deprecated, and fix an off-by-one error that caused an LBUG.
489 Description: Transient SCSI error results in persistent IO issue
490 Details : iobuf->dr_error is not reinitialized to 0 between two
494 Frequency : sometimes when underlying device returns I/O errors
496 Description: OSTs not going read-only during write failures
497 Details : OSTs are not remounted read-only when the journal commit threads
498 get I/O errors because fsfilt_ext3 calls journal_start/stop()
499 instead of the ext3 wrappers.
503 Description: poor connect scaling with increasing client count
504 Details : Don't run filter_grant_sanity_check for more than 100 exports
505 to improve scaling for large numbers of clients.
508 Frequency : SLES10 only
510 Description: sanity-quota.sh quotacheck failed: rc = -22
511 Details : Quotas cannot be enabled on SLES10.
514 Frequency : liblustre clients only
516 Description: getdirentries does not give error when run on compute nodes
517 Details : getdirentries does not fail when the size specified as an argument
518 is too small to contain at least one entry
520 Severity : enhancement
522 Description: Add LNET router traceability for debug purposes
523 Details : If a checksum failure occurs with a router as part of the
524 IO path, the NID of the last router that forwarded the bulk data
525 is printed so it can be identified.
530 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
531 Details : Due to a race condition, liblustre clients were occasionally
534 Severity : enhancement
536 Description: lfs setstripe use optional parameters instead of postional
539 Severity : enhancement
541 Description: Nanosecond timestamp support for ldiskfs
542 Details : The on-disk ldiskfs filesystem has added support for nanosecond
543 resolution timestamps. There is not yet support for this at
544 the Lustre filesystem level.
547 Frequency : during server recovery
549 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
550 Details : request with rq_no_resend flag not awake l_wait_event if they get a
554 Frequency : nfs export on patchless client
556 Description: connectathon hang when test nfs export over patchless client
557 Details : Disconnected dentry cannot be found with lookup, so we do not need
558 to unhash it or make it invalid
561 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
562 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
563 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
568 Description: open req refcounting wrong on reconnect
569 Details : If reconnect happened between getting open reply from server and
570 call to mdc_set_replay_data in ll_file_open, we will schedule
571 replay for unreferenced request that we are about to free.
572 Subsequent close will crash in variety of ways.
573 Check that request is still eligible for replay in
574 mdc_set_replay_data().
579 Description: disable writes to filesystem when reading health_check file
580 Details : the default for reading the health_check proc file has changed
581 to NOT do a journal transaction and write to disk, because this
582 can cause reads of the /proc file to hang and block HA state
583 checking on a healthy but otherwise heavily loaded system. It
584 is possible to return to the previous behaviour during configure
585 with --enable-health-write.
587 Severity : enhancement
589 Description: 64-bit inode version
590 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
591 to the inode. This will be required for version-based recovery.
596 Description: MDS fails to start if a duplicate client export is detected
597 Details : in some rare cases it was possible for a client to connect to
598 an MDS multiple times. Upon recovery the MDS would detect this
599 and fail during startup. Handle this more gracefully.
601 Severity : enhancement
603 Description: Add -o localflock option to simulate old noflock
605 Details : This will achieve local-only flock/fcntl locks
611 Description: log_commit_thread vs filter_destroy race leads to crash
612 Details : Take import reference before releasing llog record semaphore
617 Description: Wrong request locking in request set processing
618 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
619 imp_delayed_list, in this place should be used imp_lock.
622 Frequency : when reconnection
624 Description: Grant Leak when osc reconnect to OST
625 Details : When osc reconnect ost, OST(filter) should check whether it
626 should grant more space to client by comparing fed_grant and
627 cl_avail_grant, and return the granted space to client instead
628 of "new granted" space, because client will call osc_init_grant
629 to update the client grant space info.
632 Frequency : when client reconnect to OST
634 Description: Grant Leak when osc do resend and replay bulk write
635 Details : When osc reconnect to OST, OST(filter)should clear grant info of
636 bulk write request, because the grant info while be sync between
637 OSC and OST when reconnect, and we should ignore the grant info
638 these of resend/replay write req.
643 Description: Grant space more than avaiable left space sometimes.
644 Details : When then OST is about to be full, if two bulk writing from
645 different clients came to OST. Accord the avaliable space of the
646 OST, the first req should be permitted, and the second one
647 should be denied by ENOSPC. But if the seconde arrived before
648 the first one is commited. The OST might wrongly permit second
649 writing, which will cause grant space > avaiable space.
652 Frequency : when client is evicted
654 Description: Grant might be wrongly erased when osc is evicted by OST
655 Details : when the import is evicted by server, it will fork another
656 thread ptlrpc_invalidate_import_thread to invalidate the
657 import, where the grant will be set to 0. While the original
658 thread will update the grant it got when connecting. So if
659 the former happened latter, the grant will be wrongly errased
660 because of this race.
665 Description: Checking Stale with correct fid
666 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
667 to check whether it is stale, because sometimes, we want the
668 enqueue happened anyway, and op_data.fid2 will not be initialized.
670 Severity : enhancement
672 Description: update patchless client
673 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
676 Frequency : only with 2.4 kernel
678 Description: random memory corruption
679 Details : size of struct ll_inode_info is to big for union inode.u and this
680 can be cause of random memory corruption.
685 Description: Memory leak in recovery
686 Details : Lov_mds_md was not free in an error handler in mds_create_object.
687 It should also check obd_fail before fsfilt_start, otherwise if
688 fsfilt_start return -EROFS,(failover mds during mds recovery).
689 then the req will return with repmsg->transno = 0 and rc = EROFS.
690 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
691 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
692 be freed no matter whether fsfilt_commit success or not.
695 Frequency : only with huge count clients
697 Description: Prevents from taking the superblock lock in llap_from_page for
699 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
700 for llap_from_page call in ll_removepage prevents from taking the
701 superblock lock for a soon died page.
706 Description: Not check open intent error before release open handle
707 Details : in some rare cases, the open intent error is not checked before
708 release open handle, which may cause
709 ASSERTION(open_req->rq_transno != 0), because it tries to release
710 the failed open handle.
715 Description: Set cat log bitmap only after create log success.
716 Details : in some rare cases, the cat log bitmap is set too early. and it
717 should be set only after create log success.
721 Description: Accessing a block bevice can re-enable I/O when Lustre is
722 tearing down a device.
723 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
728 Description: service threads may hog cpus when there are a lot of requests
730 Details : Insert cond_resched to give other threads a chance to use some of
736 Description: the cat log was not initialized in recovery
737 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
738 unlink log on mds will not be initialized until mds post
739 recovery. And also in mds post recovery, the unlink log will
740 initialization will be done asynchronausly, so there will be race
741 between add unlink log and unlink log initialization.
745 Description: brw_stats were being printed incorrectly
746 Details : brw_stats were being printed as log2 but all of them were not
747 recorded as log2. Also remove some code duplication arising from
748 filter_tally_{read,write}.
752 Frequency : rare, only in recovery.
753 Description: ASSERTION(req->rq_type != LI_POISON) failed
754 Details : imp_lock should be held while iterating over imp_sending_list for
755 prevent destroy request after get timeout in ptlrpc_queue_wait.
759 Description: replay-single.sh test 52 fails
760 Details : A lock's skiplist need to be cleanup when it being unlinked
761 from its resource list.
765 Description: Short directio read returns full requested size rather than
767 Details : Direct I/O operations should return actual amount of bytes
768 transferred rather than requested size.
770 Severity : enhancement
772 Description: metadata RPC reduction (e.g. for rm performance)
773 Details : decrease the amount of synchronous RPC between clients and servers
774 by canceling conflicing lock before the operation on the client side
775 and packing thier handles into the main operation RPC to server.
777 Severity : enhancement
779 Description: Async OSC create to avoid the blocking unnecessarily.
780 Details : If a OST has no remain object, system will block on the creating
781 when need to create a new object on this OST. Now, ways use
782 pre-created objects when available, instead of blocking on an
783 empty osc while others are not empty. If we must block, we block
784 for the shortest possible period of time.
788 Description: improve handling recoverable errors
789 Details : if request processig with error which can be recoverable on server
790 request should be resend, otherwise page released from cache and
793 Severity : enhancement
795 Description: refine locking for avoid write wrong info into lov_objid file
796 Details : fix possible races with add new target and write/update data in
799 --------------------------------------------------------------------------------
801 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
806 Frequency : on some architectures
808 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
809 Details : Uninitialized flags sometimes cause configuration commands to
813 Frequency : patchless clients only
815 Description: missing __iget() symbol export
816 Details : The __iget() symbol export is missing. To avoid the need for
817 this on patchless clients the deathrow inode reaper is turned
818 off, and we depend on the VM to clean up old inodes. This
819 dependency was during via the fix for bug 12181.
821 --------------------------------------------------------------------------------
823 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
825 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
826 INTEROPERATE with older versions automatically. In many cases a
827 special upgrade step is needed. Please read the
828 user documentation before upgrading any part of a live system.
829 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
830 version will not interoperate with 1.6 betas before beta5 (1.5.95).
831 * WARNING: Lustre configuration and startup changes are required with
832 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
838 Severity : enhancement
840 Description: MountConf
841 Details : Lustre configuration is now managed via mkfs and mount
842 commands instead of lmc and lconf. New obd types (MGS, MGC)
843 are added for dynamic configuration management. See
844 https://mail.clusterfs.com/wikis/lustre/MountConf for
847 Severity : enhancement
849 Description: dynamic OST addition
850 Details : OSTs can now be added to a live filesystem
852 Severity : enhancement
854 Description: startup order invariance
855 Details : MDTs and OSTs can be started in any order. Clients only
856 require the MDT to complete startup.
858 Severity : enhancement
860 Description: parallel, asynchronous orphan cleanup
861 Details : orphan cleanup is now performed in separate threads for each
862 OST, allowing parallel non-blocking operation.
864 Severity : enhancement
866 Description: optimized stripe assignment
867 Details : stripe assignments are now made based on ost space available,
868 ost previous usage, and OSS previous usage, in order to try
869 to optimize storage space and networking resources.
871 Severity : enhancement
873 Description: Permanently set tunables
874 Details : All writable /proc/fs/lustre tunables can now be permanently
875 set on a per-server basis, at mkfs time or on a live
878 Severity : enhancement
880 Description: Lustre message v2
881 Details : Add lustre message format v2.
883 Severity : enhancement
885 Description: client OST exclusion list
886 Details : Clients can be started with a list of OSTs that should be
887 declared "inactive" for known non-responsive OSTs.
891 Description: SPEC SFS validation failure on NFS v2 over lustre.
892 Details : Changes the blocksize for regular files to be 2x RPC size,
893 and not depend on stripe size.
895 Severity : enhancement
897 Description: Multiple MD RPCs in flight.
898 Details : Further unserialise some read-only MDS RPCs - learn about intents.
899 To avoid overly-overloading MDS, introduce a limit on number of
900 MDS RPCs in flight for a single client and add /proc controls
901 to adjust this limit.
903 Severity : enhancement
905 Description: client read/write statistics
906 Details : Add client read/write call usage stats for performance
907 analysis of user processes.
908 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
909 file access. extents_stats shows chunk size distribution.
910 extents_stats_per_process show chunk size distribution per
913 Severity : enhancement
915 Description: mds statistics
916 Details : Add detailed mds operations statistics in
917 /proc/fs/lustre/mds/*/stats.
921 Description: Failure of copying files with lustre special EAs.
922 Details : Client side always return success for setxattr call for lustre
923 special xattr (currently only "trusted.lov").
928 Description: Refcount LNET uuids
929 Details : The global LNET uuid list grew linearly with every startup;
930 refcount repeated list entries instead of always adding to
933 Severity : enhancement
935 Description: Dynamic service threads
936 Details : Within a small range, start extra service threads
937 automatically when the request queue builds up.
940 Frequency : mixed-endian client/server environments
942 Description: mixed-endian crashes
943 Details : The new msg_v2 system had some failures in mixed-endian
946 Severity : enhancement
948 Description: Easy OST removal
949 Details : OSTs can be permanently deactivated with e.g. 'lctl
950 conf_param lustre-OST0001.osc.active=0'
952 Severity : enhancement
954 Description: MGS proc entries
955 Details : Added basic proc entries for the MGS showing what filesystems
958 Severity : enhancement
960 Description: provide MGS failover
961 Details : Added config lock reacquisition after MGS server failover.
963 Severity : enhancement
965 Description: add Linux 2.4 support
966 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
970 Description: a large application tries to do I/O to the same resource and dies
972 Details : Check the req->rq_arrival time after the call to
973 ost_brw_lock_get(), but before we do anything about
974 processing it & sending the BULK transfer request. This
975 should help move old stale pending locks off the queue as
976 quickly as obd_timeout.
979 Frequency : when an incorrect nid is specified during startup
981 Description: ptlrpc connect to non-existant node causes kernel crash
982 Details : LNET can't be re-entered from an event callback, which
983 happened when we expire a message after the export has been
984 cleaned up. Instead, hand the zombie cleanup off to another
987 Severity : enhancement
989 Description: plain/inodebits lock performance improvement
990 Details : Grouping plain/inodebits in granted list by their request modes
991 and bits policy, thus improving the performance of search through
995 Frequency : only if OST filesystem is corrupted
997 Description: client incorrectly hits assertion in ptlrpc_replay_req()
998 Details : for a short time RPCs with bulk IO are in the replay list,
999 but replay of bulk IOs is unimplemented. If the OST filesystem
1000 is corrupted due to disk cache incoherency and then replay is
1001 started it is possible to trip an assertion. Avoid putting
1002 committed RPCs into the replay list at all to avoid this issue.
1005 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1007 Description: System hang on startup
1008 Details : This bug allowed the liblustre (e.g. catamount) client to
1009 return to the app before handling all startup RPCs. This
1010 could leave the node unresponsive to lustre network traffic
1011 and manifested as a server ptllnd timeout.
1013 Severity : enhancement
1015 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1016 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1017 boolean to control whether to print peer debug info when a
1018 client's RPC times out.
1021 Frequency : only for kernels with patches from Lustre below 1.4.3
1023 Description: Remove old rdonly API
1024 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1027 Frequency : only for devices with external journals
1029 Description: Set external device read-only also
1030 Details : During a commanded failover stop, we set the disk device
1031 read-only while the server shuts down. We now also set any
1032 external journal device read-only at the same time.
1035 Frequency : when upgrading from 1.4 while trying to change parameters
1037 Description: The wrong (new) MDC name was used when setting parameters for
1038 upgraded MDT's. Also allows changing of OSC (and MDC)
1039 parameters if --writeconf is specified at tunefs upgrade time.
1042 Frequency : when setting specific ost indicies
1044 Description: QOS code breaks on skipped indicies
1045 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1046 created with --index need not be sequential.
1048 Severity : enhancement
1050 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1051 Details : The uninit_groups feature works in conjunction with the kernel
1052 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1053 up the pass1 processing of e2fsck. This is a read-only feature
1054 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1055 mount filesystems that have had this feature enabled.
1057 Severity : enhancement
1059 Description: Improve multi-block allocation algorithm to avoid fragmentation
1060 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1061 allocation locality and avoid filesystem fragmentation.
1063 ------------------------------------------------------------------------------
1065 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1067 * Support for kernels:
1068 2.4.21-47.0.1.EL (RHEL 3)
1069 2.6.5-7.283 (SLES 9)
1070 2.6.9-42.0.10.EL (RHEL 4)
1071 2.6.12.6 vanilla (kernel.org)
1072 2.6.16.27-0.9 (SLES 10)
1073 * Recommended e2fsprogs version: 1.39.cfs5
1075 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1079 Frequency : occasional, depends on client load and configuration
1080 Bugzilla : 12181, 12203
1081 Description: data loss for recently-modified files
1083 Details : In some cases it is possible that recently written or created
1084 files may not be written to disk in a timely manner (this should
1085 normally be within 30s unless client IO load is very high).
1086 The problem appears as zero-length files or files that are a
1087 multiple of 1MB in size after a client crash or client eviction
1088 that are missing data at the end of the file.
1090 This problem is more likely to be hit on clients where files are
1091 repeatedly created and unlinked in the same directory, clients
1092 have a large amount of RAM, have many CPUs, the filesystem has
1093 many OSTs, the clients are rebooted frequently, and/or the files
1094 are not accessed by other nodes after being written.
1096 The presence of the problem can be detected by looking at
1097 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1098 smaller than the second (nr_unused) then dirty files will not
1099 be flushed automatically to disk. "sync; sleep 10" should be
1100 run several times on the node before unmounting it to update
1101 Lustre (this is also safe to run on nodes without this problem).
1103 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1104 that can cause this same problem, so customers using that kernel
1105 also need to update the kernel in addition to Lustre. In order
1106 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1108 It is normal that files written just before a client crash (less
1109 than 30s) may not yet have been flushed to disk, even for local
1113 Frequency : frequent on thin XT3 nodes
1115 Description: UUID collision on thin XT3 Linux nodes
1116 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1117 randomly, since we relied on an insufficiently-seeded PRNG.
1122 Description: OSS hangs after "All ost request buffers busy"
1123 Details : A deadlock between quota and journal operations caused OSS
1124 hangs after printing "All ost request buffers busy."
1127 Frequency : always on liblustre builds
1129 Description: Cleanup compiler warnings on liblustre
1132 Frequency : always on liblustre builds on XT3
1134 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1138 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1139 > 32000 subdirectories
1140 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1141 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1142 subdirectory count crosses 32000. This will aid e2fsck to
1143 correctly handle more than 32000 subdirectories.
1146 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1148 Description: System hang on startup
1149 Details : This bug allowed the liblustre (e.g. catamount) client to
1150 return to the app before handling all startup RPCs. This
1151 could leave the node unresponsive to lustre network traffic
1152 and manifested as a server ptllnd timeout.
1154 Severity : enhancement
1156 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1157 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1158 boolean to control whether to print peer debug info when a
1159 client's RPC times out.
1164 Description: make O_SYNC working on 2.6 kernels
1165 Details : 2.6 kernels use different method for mark pages for write,
1166 so need add a code to lustre for O_SYNC work.
1171 Description: Failure to close file and release space on NFS
1172 Details : Put inode details into lock acquired in ll_intent_file_open.
1173 Use mdc_intent_lock in ll_intent_open to properly
1174 detect all kind of errors unhandled by mdc_enqueue.
1179 Description: proc file read during shutdown sometimes raced obd removal,
1181 Details : Add lock to prevent obd access after proc file removal.
1184 Frequency : Only for files larger than 4GB on 32-bit clients.
1186 Description: improperly doing page alignment of locks
1187 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1188 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1193 Description: RPCs being resent when they shouldn't be
1194 Details : Some RPCs that should not be resent are being resent. This
1195 can cause inconsistencies in the RPC state machine. Do not
1196 resend such requests.
1199 Frequency : rare, only with NFS export
1201 Description: Crash on NFS re-export node
1202 Details : under very unusual load conditions an assertion is hit in
1203 ll_intent_file_open()
1206 Frequency : only if OST filesystem is corrupted
1208 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1209 Details : for a short time RPCs with bulk IO are in the replay list,
1210 but replay of bulk IOs is unimplemented. If the OST filesystem
1211 is corrupted due to disk cache incoherency and then replay is
1212 started it is possible to trip an assertion. Avoid putting
1213 committed RPCs into the replay list at all to avoid this issue.
1218 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1219 Details : Large single O_DIRECT read and write calls can fail to allocate
1220 a sufficiently large buffer to process the request. In case of
1221 allocation failure the allocation is retried with a smaller
1222 buffer and broken into smaller requests.
1224 Severity : enhancement
1226 Description: Add -o localflock option to simulate old noflock behaviour.
1227 Details : This will achieve local-only flock/fcntl locks coherentness.
1232 Description: versioning check is incomplete
1233 Details : Checking the version difference of client vs. server, report
1234 error if the gap is too big.
1239 Description: add support PG_writeback bit
1240 Details : add support for PG_writeback bit for Lustre, for more carefull
1241 work with page cache in 2.6 kernel. This also fix some deadlocks
1242 and remove hack for work O_SYNC with 2.6 kernel.
1244 Severity : enhancement
1246 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1247 Details : The uninit_groups feature works in conjunction with the kernel
1248 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1249 up the pass1 processing of e2fsck. This is a read-only feature
1250 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1251 mount filesystems that have had this feature enabled.
1253 Severity : enhancement
1255 Description: Improve multi-block allocation algorithm to avoid fragmentation
1256 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1257 allocation locality and avoid filesystem fragmentation.
1259 ------------------------------------------------------------------------------
1261 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1263 * Support for kernels:
1264 2.6.9-42.0.3.EL (RHEL 4)
1265 2.6.5-7.276 (SLES 9)
1266 2.4.21-47.0.1.EL (RHEL 3)
1267 2.6.12.6 vanilla (kernel.org)
1268 2.6.16.21-0.8 (SLES10)
1269 * Recommended e2fsprogs version: 1.39.cfs2-0
1271 * The backwards-compatible /proc/sys/portals symlink has been removed
1272 in this release. Before upgrading, please ensure that you change
1273 any configuration scripts or /etc/sysctl.conf files that access
1274 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1275 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1276 in advance of the upgrade on any system running Lustre 1.4.6 or
1277 newer, since /proc/sys/lnet was added in that version.
1278 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1282 Frequency : only when quota is used
1284 Description: avoid scanning export list for quota master
1285 Details : Change the algorithms to avoid scanning export list in order
1286 to improve the efficiency.
1289 Frequency : MDS failover only, very rarely
1291 Description: "went back in time" messages on mds failover
1292 Details : The greatest transno may be lost when the current operation
1293 finishes with an error (transno==0) and the client's last_rcvd
1294 record is over-written. Save the greatest transno in the
1295 mds_last_transno for this case.
1298 Frequency : always for specific kernels and striping counts
1300 Description: client may get "Matching packet too big" without ACL support
1301 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1302 when trying to access files in certain configurations. The
1303 clients should in fact be denied when mounting because they do
1304 not understand ACLs.
1307 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1309 Description: many clients connecting with IO in progress causes connect timeouts
1310 Details : Avoid synchronous journal commits to avoid delays caused by many
1311 clients connecting/disconnecting when bulk IO is in progress.
1312 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1313 OST_IO_PORTAL to avoid delays behind potentially many pending
1317 Frequency : occasionally with multiple writers to a single file
1319 Description: shared writes to file may result in wrong size reported by stat()
1320 Details : Allow growing of kms when extent lock is cancelled
1323 Frequency : always with random mmap IO to multi-striped file
1325 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1326 Details : Only if the hole is at the end of OST object so that kms is too
1327 small. Fix is to increase kms accordingly in ll_nopage.
1330 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1332 Description: writes to a missing object would leak memory on the OST
1333 Details : If there is an inconsistency between the MDS and OST filesystems,
1334 such that the MDS references an object that doesn't exist, writes
1335 to that object will leak memory due to incorrect cleanup in the
1336 error handling path, eventually running out of memory on the OST.
1341 Description: Creating too long symlink causes lustre errors
1342 Details : Check symlink and name lengths before sending requests to MDS.
1345 Frequency : only if flock is enabled (not on by default)
1347 Description: posix locks not released on fd closure on 2.6.9+
1348 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1349 kernels, this caused such locks not to be released on fd close and
1350 then assertions on fs unmount about still used locks.
1353 Frequency : MDS failover only, very rarely
1355 Description: clients may get ASSERTION(granted_lock != NULL)
1356 Details : When request was taking a long time, and a client was resending
1357 a getattr by name lock request. The were multiple lock requests
1358 with the same client lock handle and
1359 mds_getattr_name->fixup_handle_for_resent_request found one of the
1360 lock handles but later failed with ASSERTION(granted_lock != NULL).
1365 Description: handle->h_buffer_credits > 0, assertion failure
1366 Details : h_buffer_credits is zero after truncate, causing assertion
1367 failure. This patch extends the transaction or creates a new
1371 Frequency : NFS re-export or patchless client
1372 Bugzilla : 11179, 10796
1373 Description: Crash on NFS re-export node (__d_move)
1374 Details : We do not want to hash the dentry if we don't have a lock.
1375 But if this dentry is later used in d_move, we'd hit uninitialised
1376 list head d_hash, so we just do this to init d_hash field but
1377 leave dentry unhashed.
1380 Frequency : NFS re-export or patchless client
1382 Description: NFS exports has problem with symbolic link
1383 Details : lustre client didn't properly install dentry when re-exported
1384 to NFS or running patchless client.
1387 Frequency : NFS re-export or patchless client
1389 Description: Various nfs/patchless fixes.
1390 Details : fixes reuse disconected alias for lookup process - this fixes
1391 warning "find_exported_dentry: npd != pd",
1392 fix permission error with open files at nfs.
1393 fix apply umask when do revalidate.
1396 Frequency : occasional
1398 Description: Crash on NFS re-export node
1399 Details : calling clear_page() on the wrong pointer triggered oops in
1400 generic_mapping_read().
1403 Frequency : rarely, using O_DIRECT IO
1405 Description: unaligned directio crashes client with LASSERT
1406 Details : check for unaligned buffers before trying any requests.
1409 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1411 Description: stale data returned from RAID cache
1412 Details : If only a small amount of IO is done to the RAID device before
1413 reading it again it is possible to get stale data from the RAID
1414 cache instead of reading it from disk.
1417 Frequency : always for sles10 kernel
1419 Description: sles10 support
1420 Details : ll_follow_link: compile fixes and using of nd_set_link
1421 under newer kernels.
1424 Frequency : depends on arch, kernel and compiler version, always on sles10
1427 Description: recursive or deep enough symlinks cause stack overflow
1428 Details : getting rid of large stack-allocated variable in
1432 Frequency : depends on hardware
1434 Description: lustre write performance loss in the SLES10 kernel
1435 Details : the performance loss is caused by using of write barriers in the
1436 ext3 code. The SLES10 kernel turns barrier support on by
1437 default. The fix is to undo that change for ldiskfs.
1439 ------------------------------------------------------------------------------
1441 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1443 * Support for kernels:
1444 2.6.9-42.0.3EL (RHEL 4)
1445 2.6.5-7.276 (SLES 9)
1446 2.4.21-47.0.1.EL (RHEL 3)
1447 2.6.12.6 vanilla (kernel.org)
1451 Frequency : quota enabled and large files being deleted
1453 Description: releasing more than 4GB of quota at once hangs OST
1454 Details : If a user deletes more than 4GB of files on a single OST it
1455 will cause the OST to spin in an infinite loop. Release
1456 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1461 Description: statfs data retrieved from /proc may be stale or zero
1462 Details : When reading per-device statfs data from /proc, in the
1463 {kbytes,files}_{total,free,avail} files, it may appear
1464 as zero or be out of date.
1467 Frequency : systems with MD RAID1 external journal devices
1469 Description: lconf's call to blkid is confused by RAID1 journal devices
1470 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1471 of returning all block devices that match the journal UUID.
1474 Frequency : always, for aggregate stripe size over 4GB
1476 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1477 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1478 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1481 Frequency : always if "lfs find" used on a local file/directory
1483 Description: "lfs find" segfaults if used on a local file/directory
1484 Details : The case where a directory component was not specified wasn't
1485 handled correctly. Handle this properly.
1488 Frequency : always on ppc64
1490 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1491 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1492 perform endianness conversion.
1495 Frequency : rarely (truncate to non-zero file size after write under load)
1496 Bugzilla : 10730, 10687
1497 Description: Files padded with zeros to next 4K multiple
1498 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1499 it is possible that files that are truncated to a non-zero size
1500 immediately after being written are filled with zero bytes beyond
1501 the truncated size. No file data is lost.
1503 Severity : enhancement
1505 Description: Allow recovery/failover for liblustre clients.
1506 Details : liblustre clients were unaware of failover configurations until
1509 Severity : enhancement
1511 Description: user file locks should fail when not mounting with flock option
1512 Details : Set up an error-returning stub in ll_file_operations.lock field
1513 to prevent incorrect behaviour when client is mounted without
1514 flock option. Also, set up properly f_op->flock field for
1518 Frequency : always on ia64
1520 Description: "lfs df" loops on printing out MDS statfs information
1521 Details : The obd_ioctl_data was not initialized and in some systems
1522 this caused a failure during the ioctl that did not return
1523 an error. Initialize the struct and return an error on failure.
1526 Frequency : SLES 9 only
1528 Description: Error of copying files with lustre special EAs as root
1529 Details : Client side always return success for setxattr call for lustre
1530 special xattr (currently only "trusted.lov").
1533 Frequency : rarely on clusters with both ia64+i386 clients
1535 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1536 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1537 clients are doing concurrent writes to the same file it is
1538 possible that the ia64 clients may LASSERT because the OST
1539 extent locks are not PAGE_SIZE aligned. Ensure that grown
1540 locks are always aligned on the request boundary.
1543 Frequency : specific use, occasional
1545 Description: Overwriting in use executable truncates on-disk binary image
1546 Details : If one node attempts to overwrite an executable in use by
1547 another node, we now correctly return ETXTBSY instead of
1548 truncating the file.
1550 Severity : enhancement
1552 Description: Async OSC create to avoid the blocking unnecessarily.
1553 Details : If a OST has no remain object, system will block on the creating
1554 when need to create a new object on this OST. Now, ways use
1555 pre-created objects when available, instead of blocking on an
1556 empty osc while others are not empty. If we must block, we block
1557 for the shortest possible period of time.
1562 Description: chmod on Lustre root is propagated to other clients
1563 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1564 invalid by the follow_mount time.
1567 Frequency : liblustre clients only
1569 Description: Race in 'instant cancel' lock handling could lead to such locks
1570 never to be granted in case of SMP MDS
1571 Details : Do not destroy not yet granted but cbpending locks in
1575 Frequency : replay/resend of open
1577 Description: non null lock assetion failure in mds_intent_policy
1578 Details : Trying to replay/resend lockless open requests resulted in
1579 mds_open() returning 0 with no lock. Now it sets a flag if
1580 a lock is going to be returned.
1582 Severity : enhancement
1584 Description: Checksum enhancements
1585 Details : New checksum enhancements allow for resending RPCs that failed
1588 Severity : enhancement
1590 Description: Tunables on number of dirty pages in cacche
1591 Details : Allow to set limit on number of dirty pages cached.
1596 Description: client crash on unmount - lock still has references
1597 Details : In some error handling cases it was possible to leak a lock
1598 reference on a client while accessing a file. This was not
1599 harmful to the client during operation, but would cause the
1600 client to crash when the filesystem is unmounted.
1603 Frequency : specific case, rare
1605 Description: ETXTBSY on mds though file not in use by client
1606 Details : ETXTBSY is no longer incorrectly returned when attempting to
1607 chmod or chown a directory that the user previously tried to
1608 execute or a currently-executing binary.
1611 Frequency : extremely rare except on liblustre-based clients
1613 Description: Lustre space not freed when files are deleted
1614 Details : Clean up open-unlinked files after client eviction. Previously
1615 the unlink was skipped and the files remained as orphans.
1620 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1621 Details : In some cases it was possible to send a blocking callback to
1622 a client doing a glimpse, even though that client didn't get
1623 a lock granted. When the glimpse lock is cancelled on the OST
1624 the freed lock is left on the waiting list and corrupted the list.
1627 Frequency : all core dumps
1629 Description: Broke core dumps to lustre
1630 Details : Negative dentry may be unhashed if parent does not have UPDATE
1631 lock, but some callers, e.g. do_coredump, expect dentry to be
1632 hashed after successful create, hash it in ll_create_it.
1634 ------------------------------------------------------------------------------
1636 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1638 * Support for kernels:
1639 2.6.9-42.0.2.EL (RHEL 4)
1640 2.6.5-7.276 (SLES 9)
1641 2.4.21-40.EL (RHEL 3)
1642 2.6.12.6 vanilla (kernel.org)
1646 Frequency : always on RHEL 3
1648 Description: Number of open files grows over time
1649 Details : The number of open files grows over time, whether or not
1650 Lustre is started. This was due to a filp leak introduced
1651 by one of our kernel patches.
1653 ------------------------------------------------------------------------------
1655 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1657 * Support for kernels:
1658 2.6.9-42.EL (RHEL 4)
1659 2.6.5-7.276 (SLES 9)
1660 2.4.21-40.EL (RHEL 3)
1661 2.6.12.6 vanilla (kernel.org)
1666 Bugzilla : 5719, 9635, 9792, 9684
1667 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1668 Details : If a server is under heavy load and cannot reply to new
1669 connection requests before the client resends the (re)connect,
1670 the connection handling code can behave badly if two service
1671 threads are concurrently handing separate (re)connections from
1672 the same client. Add better locking to the connection handling
1673 code, and ensure that only a single connection will be processed
1674 for a given client UUID, even if the lock is dropped.
1676 Severity : enhancement
1678 Description: add TCP zero-copy support to kernel
1679 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1680 sends if the hardware supports scatter-gather and checksumming.
1681 This allows socklnd to do client-write and server-read more
1682 efficiently and reduce CPU utilization from skbuf copying.
1685 Frequency : only if NFS exporting from client
1687 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1688 Details : If setting attributes on a file created under NFS that had
1689 never been opened it would be possible to oops the client
1690 if the file had no objects.
1693 Frequency : always for liblustre
1695 Description: liblustre client does MDS+OSTs setattr RPC for each write
1696 Details : When doing a write from a liblustre client, the client
1697 incorrectly issued an RPC to the MDS and each OST the file was
1698 striped over in order to update the timestamps. When writing
1699 with small chunks and many clients this could overwhelm the MDS
1700 with RPCs. In all cases it would slow down the write because
1701 these RPCs are unnecessary.
1703 Severity : enhancement
1705 Description: allow number of MDS service threads to be changed at module load
1706 Details : It is now possible to change the number of MDS service threads
1707 running. Adding "options mds mds_num_threads={N}" to the MDS's
1708 /etc/modprobe.conf will set the number of threads for the next
1709 time Lustre is restarted (assuming the "mds" module is also
1710 reloaded at that time). The default number of threads will
1711 stay the same, 32 for most systems.
1716 Description: OST crash if filesystem is unformatted or corrupt
1717 Details : If an OST is started on a device that has never been formatted
1718 or if the filesystem is corrupt and cannot even mount then the
1719 error handling cleanup routines would dereference a NULL pointer.
1724 Description: NULL pointer deref in llap_from_page.
1725 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1726 incorrect) mapping if the page was truncated/reclaimed while it was
1727 searched for. Check for this condition and skip such pages when
1728 doing readahead. Introduce extra check to llap_from_page() to
1729 verify page->mapping->host is non-NULL (so page is not anonymous).
1732 Frequency : Sometimes when using sys_sendfile
1734 Description: "page not covered by a lock" warnings from ll_readpage
1735 Details : sendfile called ll_readpage without right page locks present.
1736 Now we introduced ll_file_sendfile that does necessary locking
1737 around call to generic_file_sendfile() much like we do in
1741 Frequency : with certain MDS communication failures at client mount time
1743 Description: NULL pointer deref after failed client mount
1744 Details : a client connection request may delayed by the network layer
1745 and not be sent until after the PTLRPC layer has timed out the
1746 request. If the client fails the mount immediately it will try
1747 to clean up before the network times out the request. Add a
1748 reference from the request import to the obd device and delay
1749 the cleanup until the network drops the request.
1752 Frequency : occasionally during client (re)connect
1754 Description: assertion failure during client (re)connect
1755 Details : processing a client connection request may be delayed by the
1756 client or server longer than the client connect timeout. This
1757 causes the client to resend the connection request. If the
1758 original connection request is replied in this interval, the
1759 client may trip an assertion failure in ptlrpc_connect_interpret()
1760 which thought it would be the only running connect process.
1763 Frequency : only with obd_echo servers and clients that are rebooted
1765 Description: kernel BUG accessing uninitialized data structure
1766 Details : When running an obd_echo server it did not start the ping_evictor
1767 thread, and when a client was evicted an uninitialized data
1768 structure was accessed. Start the ping_evictor in the RPC
1769 service startup instead of the OBD startup.
1771 Severity : enhancement
1772 Bugzilla : 10193 (patchless)
1773 Description: Remove dependency on various unexported kernel interfaces.
1774 Details : No longer need reparent_to_init, exit_mm, exit_files,
1775 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1778 Frequency : rare (only users of deprecated and unsupported LDAP config)
1780 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1781 Details : LDAP apparently contains 'lustreName' attributes instead of
1782 'name'. A simple remapping of the name is sufficient.
1785 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1787 Description: waiting_locks_callback trips kernel BUG if client is evicted
1788 Details : Running with the dump_on_timeout debug flag turned on makes
1789 it possible that the waiting_locks_callback() can try to dump
1790 the Lustre kernel debug logs from an interrupt handler. Defer
1791 this log dumping to the expired_lock_main() thread.
1793 Severity : enhancement
1795 Description: Support NFS exporting on 2.6 kernels.
1796 Details : Implement non-rawops metadata methods for NFS server to use without
1797 changing NFS server code.
1800 Frequency : very rare (synthetic metadata workload only)
1802 Description: two racing renames might cause an MDS thread to deadlock
1803 Details : Running the "racer" program may cause one MDS thread to rename
1804 a file from being the source of a rename to being the target of
1805 a rename at exactly the same time that another thread is doing
1806 so, and the second thread has already enqueued these locks after
1807 doing a lookup of the target and is trying to relock them in
1808 order. Ensure that we don't try to re-lock the same resource.
1811 Frequency : only very large systems with liblustre clients
1813 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1814 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1815 all OSTs in parallel. This allows the eviction of stale liblustre
1816 clients to proceed much faster than if they were done in series,
1817 and also offers similar improvements for other set_info RPCs.
1822 Description: excessive CPU usage during initial read phase on client
1823 Details : During the initial read phase on a client, it would agressively
1824 retry readahead on the file, consuming too much CPU and impacting
1825 performance (since 1.4.5.8). Improve the readahead algorithm
1826 to avoid this, and also improve some other common cases (read
1827 of small files in particular, where "small" is files smaller than
1828 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1833 Description: MDS crash when receiving packet with unknown intent.
1834 Details : Do not LBUG in unknown intent case, just return -EFAULT
1836 Severity : enhancement
1837 Bugzilla : 9293, 9385
1838 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1839 Details : Do not serialize getattr (non-intent version) and statfs.
1842 Frequency : occasional, when OST network is overloaded/intermittent
1844 Description: client evicted by OST after bulk IO timeout
1845 Details : If a client sends a bulk IO request (read or write) the OST
1846 may evict the client if it is unresposive to its data GET/PUT
1847 request. This is incorrect if the network is overloaded (takes
1848 too long to transfer the RPC data) or dropped the OST GET/PUT
1849 request. There is no need to evict the client at all, since
1850 the pinger and/or lock callbacks will handle this, and the
1851 client can restart the bulk request.
1854 Frequency : Always when mmapping file with no objects
1856 Description: client crashes when mmapping file with no objects
1857 Details : Check that we actually have objects in a file before doing any
1858 operations on objects in ll_vm_open, ll_vm_close and
1864 Description: Request leak when working with deleted CWD
1865 Details : Introduce advanced request refcount tracking for requests
1866 referenced from lustre intent.
1868 Severity : Enhancement
1870 Description: Cache open file handles on client.
1871 Details : MDS now will return special lock along with openhandle, if
1872 requested and client is allowed to hold openhandle, even if unused,
1873 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1874 closing files for every read/write openration.
1876 Severity : Enhancement
1878 Description: Cache open negative dentries on client when possible.
1879 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1880 negative dentries on lock revocation.
1885 Description: Remounting a client read-only wasn't possible with a zconf mount
1886 Details : It wasn't possible to remount a client read-only with llmount.
1888 Severity : enhancement
1889 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1890 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1891 MPICH is included to add Lustre-specific ADIO interfaces.
1892 This is based closely on the UFS ADIO layer and only differs
1893 in file creation, in order to allow the OST striping to be set.
1894 This is user-contributed code and not supported by CFS.
1899 Description: extended inode attributes (immutable, append-only) work improperly
1900 when 2.4 and 2.6 kernels are used on client/server or vice versa
1901 Details : Introduce kernel-independent values for these flags.
1903 Severity : enhancement
1906 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1907 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1908 granularity. Now, Fractional MB granularity is be supported,
1909 this is very useful for low memory system.
1911 Severity : enhancement
1913 Description: Getattr by fid
1914 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1915 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1916 have locks and avoids getattr rpc requests for it on every stat.
1919 Frequency : Always, for filesystems larger than 2TB
1921 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1922 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1923 filesystems larger than 2TB to be mounted with mballoc enabled.
1926 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1928 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1929 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1930 filesystems larger than 2TB is mounted then the kernel will
1931 silently corrupt the start of the filesystem. CONFIG_LBD is
1932 enabled for all CFS-supported kernels, but the possibility of
1933 this happening with a modified kernel config exists.
1935 Severity : enhancement
1937 Description: add client O_DIRECT support for 2.6 kernels
1938 Details : It is now possible to do O_DIRECT reads and writes to files
1939 in the Lustre client mountpoint on 2.6 kernel clients.
1941 Severity : enhancement
1943 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1944 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1945 parallel, not waiting for response from every OST before sending
1946 a rpc to the next OST.
1951 Description: setattr vs write race when updating file timestamps
1952 Details : Client processes that update a file timestamp into the past
1953 right after writing to the file (e.g. tar) it is possible that
1954 the updated file modification time can be reset to the current
1955 time due to a race between processing the setattr and write RPC.
1957 Severity : enhancement
1959 Description: Bring 'lfs find' closer in line with regular Linux find.
1960 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1961 -print0 options and obtains all the needed info through the lustre
1964 Severity : enhancement
1966 Description: support up to 1024 configured devices on one node
1967 Details : change obd_dev array from statically allocated to dynamically
1968 allocated structs as they are first used to reduce memory usage
1973 Description: Flush dirty partially truncated pages during truncate
1974 Details : Immediatelly flush partially truncated pages in filter_setattr,
1975 this way we completely avoid having any pages in page cache on OST
1976 and can retire ugly workarounds during writes to flush such pages.
1981 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1982 Details : protect inode from truncation within vfs_unlink() context
1983 just take a reference before calling vfs_unlink() and release it
1984 when parent's i_sem is free.
1989 Description: last_id value checked outside lock on OST caused LASSERT failure
1990 Details : If there were multiple MDS->OST object precreate requests in
1991 flight, it was possible that the OST's last object id was checked
1992 outside a lock and incorrectly tripped an assertion. Move checks
1993 inside locks, and discard old precreate requests.
1996 Frequency : always, if extents are used on OSTs
1998 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1999 Details : OSTs using the extents format would not zero the high 16 bits of
2000 the index physical block number. This is not a problem for any
2001 OST filesystems smaller than 16TB, and no kernels support ext3
2002 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2003 new/modified files) and can be fixed for existing filesystems
2004 with e2fsprogs-1.39-cfs1.
2009 Description: import connection selection may be incorrect if timer wraps
2010 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2011 import connections to be ignored if the 32-bit jiffies counter
2012 wraps. Use a 64-bit jiffies counter.
2015 Frequency : very large clusters immediately after boot
2017 Description: LNET request buffers exhausted under heavy short-term load
2018 Details : If a large number of client requests are generated on a service
2019 that has previously never seen so many requests it is possible
2020 that the request buffer growth cannot keep up with the spike in
2021 demand. Instead of dropping incoming requests, they are held in
2022 the LND until the RPC service can accept more requests.
2025 Frequency : Sometimes during replay
2027 Description: Assertion failure in ll_local_open after replay.
2028 Details : If replay happened on an open request reply before we were able
2029 to set replay handler, reply will become not swabbed tripping the
2030 assertion in ll_local_open. Now we set the handler right after
2031 recognising of open request
2034 Frequency : very rare
2036 Description: kernel reports "badness in vsnprintf"
2037 Details : Reading from the "recovery_status" /proc file in small chunks
2038 may cause a negative length in lprocfs_obd_rd_recovery_status()
2039 call to vsnprintf() (which is otherwise harmless). Exit early
2040 if there is no more space in the output buffer.
2042 Severity : enhancement
2044 Description: clear OBD RPC statistics by writing to them
2045 Details : It is now possible to clear the OBD RPC statistics by writing
2046 to the "stats" file.
2051 Description: Client mtime is not the same on different clients after utimes
2052 Details : In some cases, the client was using the utimes() syscall on
2053 a file cached on another node. The clients now validate the
2054 ctime from the MDS + OSTs to determine which one is right.
2059 Description: Inability to activate failout mode
2060 Details : lconf script incorrectly assumed that in pythong string's numeric
2061 value is used in comparisons.
2064 Frequency : always with multiple stripes per file
2066 Description: Inefficient object allocation for mutli-stripe files
2067 Details : When selecting which OSTs to stripe files over, for files with
2068 a stripe count that divides evenly into the number of OSTs,
2069 the MDS is always picking the same starting OST for each file.
2070 Return the OST selection heuristic to the original design.
2075 Description: mount failures may take full timeout to return an error
2076 Details : Under some heavy load conditions it is possible that a
2077 failed mount can wait for the full obd_timeout interval,
2078 possibly several minutes, before reporting an error.
2079 Instead return an error as soon as the status is known.
2081 Frequency : quota enabled and large files being deleted
2083 Description: releasing more than 4GB of quota at once hangs OST
2084 Details : If a user deletes more than 4GB of files on a single OST it
2085 will cause the OST to spin in an infinite loop. Release
2086 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2091 Description: statfs data retrieved from /proc may be stale or zero
2092 Details : When reading per-device statfs data from /proc, in the
2093 {kbytes,files}_{total,free,avail} files, it may appear
2094 as zero or be out of date.
2097 Frequency : systems with MD RAID1 external journal devices
2099 Description: lconf's call to blkid is confused by RAID1 journal devices
2100 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2101 of returning all block devices that match the journal UUID.
2104 Frequency : always, for aggregate stripe size over 4GB
2106 Description: assertion fails when trying to use 4GB stripe size
2107 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2108 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2111 Frequency : always on ppc64
2113 Description: the first write on an ext3 filesystem with mballoc got stuck
2114 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2115 perform endianness conversion.
2117 ------------------------------------------------------------------------------
2119 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2121 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2122 INTEROPERATE with older versions automatically. Please read the
2123 user documentation before upgrading any part of a live system.
2124 * WARNING: Lustre networking configuration changes are required with
2125 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2128 * Support for kernels:
2129 2.6.9-22.0.2.EL (RHEL 4)
2130 2.6.5-7.244 (SLES 9)
2131 2.6.12.6 vanilla (kernel.org)
2134 Severity : enhancement
2135 Bugzilla : 7981/8208
2136 Description: Introduced Lustre Networking (LNET)
2137 Details : LNET is new networking infrastructure for Lustre, it includes
2138 a reorganized network configuration mode (see the user
2139 documentation for full details) as well as support for routing
2140 between different network fabrics. Lustre Networking Devices
2141 (LNDs) for the supported network fabrics have also been
2142 created for this new infrastructure.
2144 Severity : enhancement
2145 Description: Introduced Access control lists
2146 Details : clients can set ACLs on files and directories in order to have
2147 more fine-grained permissions than the standard Unix UGO+RWX.
2148 The MDS must be started with the "-o acl" mount option.
2150 Severity : enhancement
2151 Description: Introduced filesystem quotas
2152 Details : Administrators may now establish per-user quotas on the
2155 Severity : enhancement
2157 Description: Configuration change for the XT3
2158 The PTLLND is now used to run Lustre over Portals on the XT3
2159 The configure option(s) --with-cray-portals are no longer used.
2160 Rather --with-portals=<path-to-portals-includes> is used to
2161 enable building on the XT3. In addition to enable XT3 specific
2162 features the option --enable-cray-xt3 must be used.
2167 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2168 Details : A race between allocating a new llcd and re-getting the llcd_lock
2169 allowed another thread to grab newly-allocated llcd.
2171 Severity : enhancement
2173 Description: 2.6 OST async journal commit and locking fix to improve performance
2174 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2175 2.6 kernels are now async as they already were in 2.4 kernels so
2176 that they can commit concurrently with the network bulk transfer.
2177 For block-allocated files the filter allocation semaphore is held
2178 to avoid filesystem fragmentation during allocation. BKL lock
2179 removed for 2.6 xattr operations where it is no longer needed.
2184 Description: lconf incorrectly determined whether two IP networks could talk
2185 Details : In some more complicated routing and multiple-network
2186 configurations, lconf will avoid trying to make a network
2187 connection to a disjoint part of the IP space. It was doing the
2188 math incorrectly for one set of cases.
2193 Description: Fix for potential infinite loop processing records in an llog.
2194 Details : If an llog record is corrupted/zeroed, it is possible to loop
2195 forever in llog_process(). Validate the llog record length
2196 and skip the remainder of the block on error.
2199 Frequency : occasional (liblustre only)
2201 Description: liblustre could not open files whose last component is a symlink
2202 Details : sysio_path_walk() would incorrectly pass the open intent to
2203 intermediate path components.
2206 Frequency : rare (liblustre only with non-standard tuning)
2207 Bugzilla : 7201 (7350)
2208 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2209 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2210 after it has been released, only do this for referenced locks
2212 Severity : enhancement
2214 Description: specifying an (invalid) directory default stripe_size of -1
2215 would reset the directory default striping
2216 Details : stripe_size -1 was used internally to signal directory stripe
2217 removal, now use "all default" to signal dir stripe removal
2218 as a directory striping of "all default" is not useful
2221 Frequency : common for large clusters running liblustre clients
2223 Description: doing an ls when liblustre clients are running is slow
2224 Details : sending a glimpse AST to a liblustre client waits for every AST
2225 to time out, as liblustre clients will not respond. Since they
2226 cannot cache data we refresh the OST lock LVB from disk instead.
2228 Severity : enhancement
2230 Description: doing an ls at the same time as file IO can be slow
2231 Details : enqueue and other "small" requests can be blocked behind many
2232 large IO requests. Create a new OST IO portal for non-IO
2233 requests so they can be processed faster.
2236 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2238 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2239 created file with mode 000
2240 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2241 allow an ftruncate() as a non-root user to a file with mode 000.
2242 Lustre now respects this flag to disable mode checks when
2243 truncating a file owned by the user
2246 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2249 Description: Revoking locks from clients that went dead or catatonic might take
2251 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2252 cancellation of such locks instant on servers without waiting for
2253 any reply from clients. Clients drops these locks when cancel
2254 notification from server is received without replying.
2257 Frequency : liblustre-only, when liblustre client dies or becomes busy
2259 Description: Doing ls on Linux clients can take a long time with active
2261 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2262 granting such locks to it in the first place if possible. Locks
2263 are taken by proxy on the OST during the read or write and
2264 dropped immediately afterward. Add connect flags handling, do
2265 not grant locks to liblustre clients for glimpse ASTs.
2267 Severity : enhancement
2269 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2270 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2271 sequential reads and behaves badly with applications doing random
2272 reads. Improve it by reading ahead at least read region, and
2273 avoiding excessive large RPC for small reads.
2275 Severity : enhancement
2277 Description: Creating more than 1000 files for a single job may cause a load
2278 imbalance on the OSTs if there are also a large number of OSTs.
2279 Details : qos_prep_create() uses an OST index reseed value that is an
2280 even multiple of the number of available OSTs so that if the
2281 reseed happens in the middle of the object allocation it will
2282 still utilize the OSTs as uniformly as possible.
2287 Description: OST or MDS may oops in ping_evictor_main()
2288 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2289 but doesn't restart at beginning of obd_exports_timed list
2292 Severity : enhancement
2294 Description: improve by-nid export eviction on the MDS and OST
2295 Details : allow multiple exports with the same NID to be evicted at one
2296 time without re-searching the exports list.
2299 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2301 Description: MDS may oops in groups_free()
2302 Details : in rare race conditions a newly allocated group_info struct is
2303 freed again, and this can be NULL. The 2.4 compatibility code
2304 for groups_free() checked for a NULL pointer, but 2.6 did not.
2307 Frequency : common for liblustre clients doing little filesystem IO
2308 Bugzilla : 9352, 7313
2309 Description: server may evict liblustre clients accessing contended locks
2310 Details : if a client is granted a lock or receives a completion AST
2311 with a blocking AST already set it would not reply to the AST
2312 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2315 Frequency : lfs setstripe, only systems with more than 160 OSTs
2317 Description: unable to set striping with a starting offset beyond OST 160
2318 Details : llapi_create_file() incorrectly limited the starting stripe
2319 index to the maximum single-file stripe count.
2322 Frequency : LDAP users only
2324 Description: lconf did not handle in-kernel recovery with LDAP properly
2325 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2327 Severity : enhancement
2329 Description: bind OST threads to NUMA nodes to improve performance
2330 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2331 node and do their allocations there to localize memory access
2333 Severity : enhancement
2335 Description: llmount can determine client NID directly from Myrinet (GM)
2336 Details : the client NID code from gmnalnid was moved directly into
2337 llmount, removing the need to use this or specifying the
2338 client NID explicitly when mounting GM clients with zeroconf
2341 Frequency : if client is started with down MDS
2343 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2344 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2345 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2346 ptlrpc_import_delayed_req() because we didn't check if the
2347 request was interrupted, and we also didn't break out of the
2348 event loop if there was no timeout
2353 Description: data loss during non-page-aligned writes to a single file from
2354 both multiple nodes and multiple threads on one node at same time
2355 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2356 inconsistency led to false short-reads, that were cached and later
2357 used by ->prepare_write() to fill in partially written page,
2358 leading to data loss.
2361 Frequency : always, if lconf --abort_recovery used
2363 Description: lconf --abort_recovery fails with 'Operation not supported'
2364 Details : lconf was attempting to abort recovery on the MDT device and not
2367 Severity : enhancement
2369 Description: remove cleanup logs
2370 Details : replace lconf-generated cleanup logs with lustre internal
2371 cleanup routines. Eliminates the need for client-cleanup and
2374 Severity : enhancement
2376 Description: add support for EAs (user and system) on lustre filesystems
2377 Details : it is now possible to store extended attributes in the Lustre
2378 client filesystem, and with the user_xattr mount option it
2379 is possible to allow users to store EAs on their files also
2381 Severity : enhancement
2383 Description: Add possibility (config option) to show minimal available OST free
2385 Details : When compiled with --enable-mindf configure option, statfs(2)
2386 (and so, df) will return least minimal free space available from
2387 all OSTs as amount of free space on FS, instead of summary of
2388 free spaces of all OSTs.
2390 Severity : enhancement
2392 Description: do not expand extent locks acquired on OST-side
2393 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2394 by server: they are not cached anyway.
2397 Frequency : when mmap is used/binaries executed from Lustre
2399 Description: Unmmap pages before throwing them away from read cache.
2400 Details : llap_shrink cache now attempts to unmap pages before discarding
2401 them (if unmapping failed - do not discard). SLES9 kernel has
2402 extra checks that trigger if this unmapping is not done first.
2407 Description: lconf didn't resolve symlinks before checking to see whether a
2408 given mountpoint was already in use
2411 Frequency : when migrating failover services
2412 Bugzilla : 6395, 9514
2413 Description: When migrating a subset of services from a node (e.g. failback
2414 from a failover service node) the remaining services would
2415 time out and evict clients.
2416 Details : lconf --force (implied by --failover) sets the global obd_timeout
2417 to 5 seconds in order to quickly disconnect, but this caused
2418 other RPCs to time out too quickly. Do not change the global
2419 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2421 Severity : enhancement
2422 Frequency : if MDS is started with down OST
2423 Bugzilla : 9439,5706
2424 Description: Allow startup/shutdown of an MDS without depending on the
2425 availability of the OSTs.
2426 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2427 Add appropriate locking and lov-osc refcounts for safe
2428 cleaning. Add osc abort_inflight calls in case the
2429 synchronize never started.
2432 Frequency : occasional (Cray XT3 only)
2434 Description: root not authorized to access files in CRAY_PORTALS environment
2435 Details : The client process capabilities were not honoured on the MDS in
2436 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2437 been accessed by an authorized user then root was able to access
2438 the file on the local client also. The root user capabilities
2439 are now allowed on the MDS, as this environment has secure UID.
2442 Frequency : occasional
2444 Description: ldiskfs "too long searching" message happens too often
2445 Details : A debugging message (otherwise harmless) prints too often on
2446 the OST console. This has been reduced to only happen when
2447 there are fragmentation problems on the filesystem.
2452 Description: Division by zero in statfs when all OSCs are inactive
2453 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2454 lov_statfs divides by value returned by lov_get_stripecnt().
2458 Bugzilla : 9489, 3273
2459 Description: First write from each client to each OST was only 4kB in size,
2460 to initialize client writeback cache, which caused sub-optimal
2461 RPCs and poor layout on disk for the first writen file.
2462 Details : Clients now request an initial cache grant at (re)connect time
2463 and so that they can start streaming writes to the cache right
2464 away and always do full-sized RPCs if there is enough data.
2465 If the OST is rebooted the client also re-establishes its grant
2466 so that client cached writes will be honoured under the grant.
2471 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2472 Details : Now I/O RPCs go to different portal number and (presumably) fast
2473 lock requests (and glimses) and other RPCs get their own service
2474 threads pool that should be able to service those RPCs
2477 Severity : enhancement
2479 Description: Ability to exchange lustre version between client and servers and
2480 issue warnings at client side if client is too old. Also for
2481 liblustre clients there is ability to refuse connection of too old
2483 Details : New 'version' field is added to connect data structure that is
2484 filled with version info. That info is later checked by server and
2488 Frequency : rare, liblustre only.
2489 Bugzilla : 9296, 9581
2490 Description: Two simultaneous writes from liblustre at offset within same page
2491 might proceed at the same time overwriting eachother with stale
2493 Details : I/O lock withing llu_file_prwv was released too early, before data
2494 actually was hitting the wire. Extended lock-holding time until
2495 server acknowledges receiving data.
2498 Frequency : extremely rare. Never observed in practice.
2500 Description: avoid generating lustre_handle cookie of 0.
2501 Details : class_handle_hash() generates handle cookies by incrementing
2502 global counter, and can hit 0 occasionaly (this is unlikely, but
2503 not impossible, because initial value of cookie counter is
2504 selected randonly). Value of 0 is used as a sentinel meaning
2505 "unassigned handle" --- avoid it. Also coalesce two critical
2506 sections in this function into one.
2508 Severity : enhancement
2510 Description: allow liblustre clients to delegate truncate locking to OST
2511 Details : To avoid overhead of locking, liblustre client instructs OST to
2512 take extent lock in ost_punch() on client's behalf. New connection
2513 flag is added to handle backward compatibility.
2515 Severity : enhancement
2516 Bugzilla : 4928, 7341, 9758
2517 Description: allow number of OST service threads to be specified
2518 Details : a module parameter allows the number of OST service threads
2519 to be specified via "options ost ost_num_threads={N}" in the
2520 OSS's /etc/modules.conf or /etc/modprobe.conf.
2524 Bugzilla : 6146, 9635, 9895
2525 Description: servers crash with bad pointer in target_handle_connect()
2526 Details : In rare cases when a client is reconnecting it was possible that
2527 the connection request was the last reference for that export.
2528 We would temporarily drop the export reference and get a new
2529 one, but this may have been the last reference and the export
2530 was just destroyed. Get new reference before dropping old one.
2532 Severity : enhancement
2533 Frequency : if client is started with failover MDS
2535 Description: Allow multiple MDS hostnames in the mount command
2536 Details : Try to read the configuration from all specified MDS
2537 hostnames during a client mount in case the "primary"
2540 Severity : enhancement
2542 Description: Stop sending data to evicted clients as soon as possible.
2543 Details : Check if the client we are about to send or are sending data to
2544 was evicted already. (Check is done every second of waiting,
2545 for which l_wait_event interface was extended to allow checking
2546 of exit condition at specified intervals).
2549 Frequency : rare, normally only when NFS exporting is done from client
2551 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2553 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2556 Frequency : always, if srand() is called before liblustre initialization
2558 Description: Liblustre uses system PRNG disturbing its usage by user application
2559 Details : Introduce internal to lustre fast and high-quality PRNG for
2560 lustre usage and make liblustre and some other places in generic
2561 lustre code to use it.
2563 Severity : enhancement
2564 Bugzilla : 9477, 9557, 9870
2565 Description: Verify that the MDS configuration logs are updated when xml is
2566 Details : Check if the .xml configuration logs are newer than the config
2567 logs stored on the MDS and report an error if this is the case.
2568 Request --write-conf, or allow starting with --old_conf.
2570 Severity : enhancement
2572 Description: Handle symlinks in the path when checking if Lustre is mounted.
2573 Details : Resolve intermediate symlinks when checking if a client has
2574 mounted a filesystem to avoid duplicate client mounts.
2579 Description: lconf can hit an error exception but still return success.
2580 Details : The lconf command catches the Command error exception at the top
2581 level script context and will exit with the associated exit
2582 status, but doesn't ensure that this exit status is non-zero.
2587 Description: failure of ptlrpc thread startup can cause oops
2588 Details : Starting a ptlrpc service thread can fail if there are a large
2589 number of threads or the server memory is very fragmented.
2590 Handle this without oopsing.
2593 Frequency : always, only if liblustre and non-default acceptor port was used
2595 Description: liblustre cannot connect to servers with non-default acceptor port
2596 Details : tcpnal_set_default_params() was not called and was therefore
2597 ignoring the environment varaible TCPNAL_PORT, as well as other
2598 TCPNAL_ environment variables
2603 Description: two objects could be created on the same OST for a single file
2604 Details : If an OST is down, in some cases it was possible to create two
2605 objects on a single OST for a single file. No problems other
2606 than potential performance impact and spurious error messages.
2610 Bugzilla : 5681, 9562
2611 Description: Client may oops in ll_unhash_aliases
2612 Details : Client dcache may become inconsistent in race condition.
2613 In some cases "getcwd" can fail if the current directory is
2619 Description: Inode refcounting problems in NFS export code
2620 Details : link_raw functions used to call d_instantiate without obtaining
2621 extra inode reference first.
2625 Bugzilla : 9942, 9903
2626 Description: Referencing freed requests leading to crash, memleaks with NFS.
2627 Details : We used to require that call to ll_revalidate_it was always
2628 followed by ll_lookup_it. Also with revalidate_special() it is
2629 possible to call ll_revalidate_it() twice for the same dentry
2630 even if first occurence returned success. This fix changes semantic
2631 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2632 reference on a request referred from the intent.
2633 ll_intent_release() then releases such a request.
2636 Frequency : rare, normally benchmark loads only
2638 Description: unlinked inodes were kept in memory on the client
2639 Details : If a client is repeatedly creating and unlinking files it
2640 can accumulate a lot of stale inodes in the inode slab cache.
2641 If there is no other client load running this can cause the
2642 client node to run out of memory. Instead flush old inodes
2643 from client cache that have the same inode number as a new inode.
2646 Frequency : SLES9 2.6.5 kernel and long filenames only
2647 Bugzilla : 9969, 10379
2648 Description: utime reports stale NFS file handle
2649 Details : SLES9 uses out-of-dentry names in some cases, which confused
2650 the lustre dentry revalidation. Change it to always use the
2654 Frequency : rare, unless heavy write-truncate concurrency is continuous
2655 Bugzilla : 4180, 6984, 7171, 9963, 9331
2656 Description: OST becomes very slow and/or deadlocked during object unlink
2657 Details : filter_destroy() was holding onto the parent directory lock
2658 while truncating+unlinking objects. For very large objects this
2659 may block other threads for a long time and slow overall OST
2660 responsiveness. It may also be possible to get a lock ordering
2661 deadlock in this case, or run out of journal credits because of
2662 the combined truncate+unlink. Solution is to do object truncate
2663 first in one transaction without parent lock, and then do the
2664 final unlink in a new transaction with the parent lock. This
2665 reduces the lock hold time dramatically.
2668 Frequency : rare, 2.4 kernels only
2670 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2671 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2672 before unmounting their filesystem in order to avoid deadlock.
2673 The kernel_locked() function in 2.4 kernels only checks whether
2674 the kernel lock is held, not whether it is this process that is
2675 holding it as 2.6 kernels do.
2680 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2681 Details : The client ptlrpc code may be trying to reconnect to a down
2682 server before a previous connection attempt has timed out.
2683 Increase the reconnect interval to be longer than the connection
2684 timeout interval to avoid sending duplicate connections to
2688 Frequency : echo_client brw_test command
2690 Description: fix echo_client to work with OST preallocated code
2691 Details : OST preallocation code (5137) didn't take echo_client IO path
2692 into account: echo_client calls filter methods outside of any
2693 OST thread and, hence, there is no per-thread preallocated
2694 pages and buffers to use. Solution: hijack pga pages for IO. As
2695 a byproduct, this avoids unnecessary data copying.
2699 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2700 Description: Client can oops in mdc_commit_close() after open replay
2701 Details : It was possible for the MDS to return an open request with no
2702 transaction number in mds_finish_transno() if the client was
2703 evicted, but without actually returning an error. Clients
2704 would later try to replay that open and may trip an assertion
2705 Simplify the client close codepath, and always return an error
2706 from the MDS in case the open is not successful.
2709 Frequency : rare, 2.6 OSTs only
2711 Description: OST may deadlock under high load on fragmented files
2712 Details : If there was a heavy load and highly-fragmented OST filesystems
2713 it was possible to have all the OST threads deadlock waiting on
2714 allocation of biovecs, because the biovecs were not released
2715 until the entire RPC IO was completed. Instead, release biovecs
2716 as soon as they are complete to ensure forward IO progress.
2718 Severity : enhancement
2720 Description: Support for specifying external journal device at mount
2721 Details : If an OST or MDS device is formatted with an external journal
2722 device, this device major/minor is stored in the ext3 superblock
2723 and may not be valid for failover. Allow detecting and
2724 specifying the external journal at mount time.
2729 Description: Mounting an MDS with pending unlinked files may cause oops
2730 Details : target_finish_recovery() calls mds_postrecov() which returned
2731 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2732 considers this an error and immediately begins cleaning up the
2733 lov, just after starting the mds_lov process
2735 Severity : enhancement
2737 Description: Implement 'lfs df' to report actual free space on per-OST basis
2738 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2739 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2740 usage of objects; '-h' to report in human readable format.
2742 ------------------------------------------------------------------------------
2744 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2751 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2752 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2753 at startup during orphan unlinking. Instead, skip buddy update
2754 before setup, as it will be regenerated after recovery is complete.
2759 Description: If an OST is inactive, its locks might reference stale inodes.
2760 Details : lov_change_cbdata() must iterate over all namespaces, even if
2761 they are inactive to clear inode references from the lock.
2763 Severity : enhancement
2764 Frequency : occasional, if non-standard max_dirty_mb used
2766 Description: Client will block write RPCs if not enough grant
2767 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2768 then the client will block writes while waiting for another RPC
2769 to complete instead of consuming its dirty limit. With change
2770 we get improved performance when max_dirty_mb is small.
2772 Severity : enhancement
2773 Bugzilla : 3389, 6253
2774 Description: Add support for supplementary groups on the MDS.
2775 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2776 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2777 lookups for user supplementary groups into a cache.
2781 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2782 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2783 for liblustre O_EXCL is enforced by MDS.
2787 Description: readdir never returns NULL in liblustre.
2788 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2789 dentry in struct dirent.
2793 Frequency : liblustre only, depends on application IO pattern
2794 Description: liblustre clients evicted if not contacting servers
2795 Details : Don't put liblustre clients into the ping_evictor list, so
2796 they will not be evicted by the pinger ever.
2798 Severity : enhancement
2800 Description: Add ability to evict clients by NID from MDS.
2801 Details : By echoing "nid:$NID" string into
2802 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2803 $NID would be instantly evicted from this MDS and from all active
2804 OSTs connected to it.
2808 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2809 Details : lookup_it_finish() used to query file size from OSTs that was not
2814 Description: service threads change working directory to that of init
2815 Details : Starting lustre service threads may pin the working directory
2816 of the parent thread, making that filesystem busy. Threads
2817 now change to the working directory of init to avoid this.
2821 Frequency : during shutdown only
2822 Description: shutdown with a failed MDS or OST can cause unmount to hang
2823 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2828 Frequency : 2.6 only
2829 Description: chmod/chown may include an extra supplementary group
2830 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2831 supplementary group and if none is specified this is used.
2834 Bugzilla : 5479 (6816)
2836 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2837 Details : If lookup is in progress on a file that is unlinked we might try
2838 to revalidate the inode and fail in revalidate after lookup is
2839 complete and ll_file_open() enqueues the open again but
2840 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2843 Frequency : always, if lconf --abort_recovery used
2845 Description: lconf --abort_recovery fails with 'Operation not supported'
2846 Details : lconf was attempting to abort recovery on the MDT device and not
2849 ------------------------------------------------------------------------------
2851 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2856 Frequency : rare (only unsupported configurations with a node running as an
2858 Bugzilla : 6514, 5137
2859 Description: Mounting a Lustre file system on a node running as an OST could
2861 Details : OSTs now preallocates memory needed to write out data at
2862 startup, instead of when needed, to avoid having to
2863 allocate memory in possibly low memory situations.
2864 Specifically, if the file system is mounted on on OST,
2865 memory pressure could force it to try to write out data,
2866 which it needed to allocate memory to do. Due to the low
2867 memory, it would be unable to do so and the node would
2868 become unresponsive.
2870 Severity : enhancement
2872 Description: Addition of lconf --service command line option
2873 Details : lconf now accepts a '--service <arg>' option, which is
2874 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2876 Severity : enhancement
2878 Description: Failover mode is now the default for OSTs.
2879 Details : By default, OSTs will now run in failover mode. To return to
2880 the old behaviour, add '--failout' to the lmc line for OSTs.
2882 Severity : enhancement
2884 Description: Health checks are now provided for MDS and OSTs
2885 Details : Additional detailed health check information on MSD and OSTs
2886 is now provided through the procfs health_check value.
2889 Frequency : occasional, depends on IO load
2891 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2892 after numerous create/delete cycles
2893 Details : The ext3 inode allocation policy would not allocate new inodes
2894 very well on the OSTs because there are no new directories
2895 being created. Instead we look for groups with free space if
2896 the parent directories are nearly full.
2901 Description: Network or server problems during mount may cause partially
2902 mounted clients instead of returning an error.
2903 Details : The config llog parsing code may overwrite the error return
2904 code during mount error handling, returning success instead
2910 Description: MDS can fail to allocate large reply buffers
2911 Details : After long uptimes the MDS can fail to allocate large reply
2912 buffers (e.g. zconf client mount config records) due to memory
2913 fragmentation or consumption by the buffer cache. Preallocate
2914 some large reply buffers so that these replies can be sent even
2915 under memory pressure.
2919 Frequency : rare (liblustre)
2920 Description: fsx running with liblustre complained that using truncate() to
2921 extend the file doesn't work. This patch corrects that issue.
2922 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2923 ATTR_SIZE and lsm use in llu_setattr_raw.
2927 Frequency : rare, only 2.6 kernels
2928 Description: Unusual file access patterns on the MDS may result in inode
2929 data being lost in very rare circumstances.
2930 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2931 code in the 2.6 kernel caused the inode and/or EA data not to
2932 be read from disk, causing single-file corruption.
2936 Frequency : rare, only 2.6 filesystems using extents
2937 Description: Heavy concurrent write and delete load may cause data corruption.
2938 Details : It was possible under high-load situations to have an extent
2939 metadata block in the block device cache from a just-unlinked
2940 file overwrite a newly-allocated data block. We now unmap any
2941 metadata buffers that alias just-allocated data blocks.
2945 Frequency : filesystems with default stripe_count larger than 77
2946 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2947 Details : lconf specifies an inode size of 4096 bytes when the default
2948 stripe_count is larger than 77. This conflicts with the default
2949 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2950 this case to avoid pinning too much memory for large EAs.
2952 ------------------------------------------------------------------------------
2954 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2959 Frequency : rare (extremely heavy IO load with hundreds of clients)
2961 Description: Client is evicted, gets IO error writing to file
2962 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2963 caused clients to be evicted for AST timeouts. The fixes in
2964 bug 5192 mean we no longer need to have such short AST timeouts
2965 so ldlm_timeout has been increased.
2968 Frequency : occasional during --force or --failover shutdown under load
2969 Bugzilla : 5949, 4834
2970 Description: Server oops/LBUG if stopped with --force or --failover under load
2971 Details : a collection of import/export refcount and cleanup ordering
2972 issues fixed for safer force cleanup
2975 Frequency : only filesystems larger than 120 OSTs
2976 Bugzilla : 5990, 6223
2977 Description: lfs getstripe would oops on a very large filesystem
2978 Details : lov_getconfig used kfree on vmalloc'd memory
2981 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2982 Bugzilla : 6242, 6243
2983 Description: reading from files that had been truncated to a non-zero size
2984 but never opened returned no data
2985 Details : ll_file_read() reads zeros from no-object files to EOF
2990 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2991 Details : There's little harm in aborting MDS/OSS recovery and letting it
2992 try again, so I removed the LASSERT and return an error instead.
2994 Severity : enhancement
2996 Description: New debugging infrastructure for tracking down data corruption
2997 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2998 (b) cover more of the client-side code path, and (c) try to narrow
2999 down where problems occurred
3003 Bugzilla : 3819, 4364, 4397, 6313
3004 Description: Racing close and eviction MDS could cause assertion in mds_close
3005 Details : It was possible to get multiple mfd references during close and
3006 client eviction, leading to one thread referencing a freed mfd.
3008 Severity: : enhancement
3009 Bugzilla : 3262, 6359
3010 Description: Attempts to reconnect to servers are now more aggressive.
3011 Details : This builds on the enhanced upcall-less recovery that was added
3012 in 1.4.2. When trying to reconnect to servers, clients will
3013 now try each server in the failover group every 10 seconds. By
3014 default, clients would previously try one server every 25 seconds.
3019 Description: After recovery, certain operations trigger a failed
3020 assertion on a client.
3021 Details : Failing over an mds, using lconf -d --failover, while a
3022 client was doing a readdir() call would cause the client to
3023 LBUG after recovery completed and the readdir() was resent.
3025 Severity : enhancement
3027 Description: Default groups are now added by lconf
3028 Details : You can now run lconf --group <servicename> without having to
3029 manually add groups with lmc.
3032 Frequency : occasional
3034 Description: Nodes with an elan id of 0 trigger a failed assertion
3037 Frequency : always when accessing e.g. tty/console device nodes
3039 Description: tty and some other devices nodes cannot be used on lustre
3040 Details : file's private_data field is used by device data and lustre
3041 values in there got lost. New field was added to struct file to
3042 store fs-specific private data.
3045 Frequency : when exporting Lustre via NFS
3047 Description: NFSD failed occasionally when looking up a path component
3048 Details : NFSD is looking up ".." which was broken in ext3 directories
3049 that had grown large enough to become hashed.
3052 Frequency : Clusters with multiple interfaces not on the same subnet
3054 Description: Nodes will repeatedly try to reconnect to an interface which it
3055 cannot reach and report an error to the log.
3056 Details : Extra peer list entries will be created by lconf with some peers
3057 unreachable. lconf now validates the peer before adding it.
3060 Frequency : Only if a default stripe is set on the filesystem root.
3062 Description: Setting a default stripe on the filesystem root prevented the
3063 filesystem from being remounted.
3064 Details : The client was sending extra request flags in the root getattr
3065 request and did not allocate a reply buffer for the dir EA.
3068 Frequency : occasional, higher if lots of files are accessed by one client
3069 Bugzilla : 6159, 6097
3070 Description: Client trips assertion regarding lsm mismatch/magic
3071 Details : While revalidating inodes the VFS looks up inodes with ifind()
3072 and in rare cases can find an inode that is being freed.
3073 The ll_test_inode() code will free the lsm during ifind()
3074 when it finds an existing inode and then the VFS later attaches
3075 this free lsm to a new inode.
3079 Bugzilla : 6422, 7030
3080 Description: MDS deadlock between mkdir and client eviction
3081 Details : Creating a new file via mkdir or mknod (starting a transaction
3082 and getting the ns lock) can deadlock with client eviction
3083 (gets ns lock and trying to finish a synchronous transaction).
3086 Frequency : occasional
3087 Description: While starting a server, the fsfilt_ext3 module could not be
3089 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3090 kernels. Previously, lconf would still use the ext3 name
3091 when trying to load modules. Now, it will correctly use
3092 ext3 on 2.4 and ldiskfs on 2.6.
3094 Severity : enhancement
3095 Description: The default stripe count has been changed to 1
3096 Details : The interpretation of the default stripe count (0, to lfs
3097 or lmc) has been changed to mean striping across a single
3098 OST, rather than all available. For general usage we have
3099 found a stripe count of 1 or 2 works best.
3101 Severity : enhancement
3102 Description: Add support for compiling against Cray portals.
3103 Details : Conditional compiling for some areas that are different
3107 Frequency : occasional
3108 Bugzilla : 6409, 6834
3109 Description: Creating files with an explicit stripe count may lead to
3110 a failed assertion on the MDS
3111 Details : If some OSTs are full or unavailable, creating files may
3112 trigger a failed assertion on the MDS. Now, Lustre will
3113 try to use other servers or return an error to the
3117 Frequency : occasional
3119 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3120 caused lock timeout during readahead (since 1.4.2).
3121 Details : Processes doing readahead might match a lock that hasn't been
3122 granted yet if there are overlapping and conflicting lock
3123 requests. The readahead process waits on ungranted lock
3124 (original lock is CBPENDING), while OST waits for that process
3125 to cancel CBPENDING read lock and eventually evicts client.
3127 Severity : enhancement
3129 Description: Initial enabling of flock support for clients
3130 Details : Implements fcntl advisory locking and file status functions.
3131 This feature is provided as an optional mount flag (default
3132 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3133 locking are implemented yet, and those that are are not guaranteed
3134 to be completely correct in production environments.
3135 mount -t lustre -o [flock|noflock] ...
3138 Frequency : occasional
3140 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3141 assertion in the ext3 JBD (journaling) layer.
3142 Details : The b_committed_data struct is protected by the big kernel lock
3143 in 2.4 kernels, serializing journal_commit_transaction() and
3144 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3145 access to this struct. In 2.6 kernels there is finer grained
3146 locking to improve SMP performance of the JBD layer.
3150 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3155 Description: merge and cleanup kernel patches.
3158 -----------------------------------------------------------------------------
3160 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3162 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3163 versions of Lustre. Please update all servers and clients to
3164 version 1.4.2 or later at the same time. You must also run
3165 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3166 to update the configuration logs.
3168 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3169 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3170 - swab llog records of type '0' so we get proper header size/idx (5861)
3171 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3172 - fix rename of one directory over another leaking an inode (5953)
3173 - avoid SetPageDirty on 2.6 (5981)
3174 - don't re-add just-being-destroyed locks to the waiting list (5653)
3175 - when creating new directories, inherit the parent's custom
3176 striping settings if present parent (3048)
3177 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3178 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3179 - don't hold client locks on temporary worklist from l_lru (5666)
3180 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3181 - automatically evict dead clients (5921)
3182 - Update file size properly in create+truncate+fstat case (6196)
3183 - Do not unhash mountpoint dentries, do not allow removal of
3185 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3186 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3187 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3188 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3189 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3190 - Correctly handle reads of files with no objects (6243)
3191 - lctl recover will also mark a device active if deactivate used (5933)
3193 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3194 - allow --write-conf on an MDS with different nettype than client (5619)
3195 - don't write config llogs to MDS for mounts not from that MDS (5617)
3196 - lconf should create multiple TCP connections from a client (5201)
3197 - init scripts are now turned off by default; run chkconfig --on
3198 lustre and chkconfig --on lustrefs to use them
3199 - upcalls are no longer needed for clients to recover to failover
3201 - add --abort-recovery option to lconf to abort recovery on device
3203 - add support for an arbitrary number of OSTs (3026)
3204 - Quota support protocol changes.
3205 - forward compatibility changes to wire structs (6007)
3206 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3207 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3208 - improved "lustre status" script
3209 - initialize blocksize for non-regular files (6062)
3210 - added --disable-server and --disable-client configure options (5782)
3211 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3212 - Vanilla 2.4.29 support
3213 - increase maximum number of obd devices to 520 (6242)
3214 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3215 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3216 SLES 9 2.6.5 kernels
3217 - limit stripes per file to 160 (the maximum EA size) (6093)
3219 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3222 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3223 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3224 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3225 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3226 - small ext3 extents cleanups and fixes (5733)
3227 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3228 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3229 - avoid needless client->OST connect, fix handle mismatch (5317)
3230 - fix DLM error path that led to out-of-sync client, long delays (5779)
3231 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3232 - fix several locking issues related to i_size (5492,5624,5654,5672)
3233 - don't move pending lock onto export if it is already evicted (5683)
3234 - fix kernel oops when creating .foo in unlinked directory (5548)
3235 - fix deadlock in obdfilter statistics vs. object create (5811)
3236 - use time_{before,after} to avoid timer jiffies wrap (5882)
3237 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3238 - Do not leak request if server was not able to process it (5154)
3239 - If mds_open unable to find parent dir, make that negative lookup(5154)
3240 - don't create new directories with extent-mapping (5909, 5936)
3242 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3243 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3244 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3245 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3246 - support for RHEL4 kernel on i686 (5773)
3247 - provide error messages when incompatible logs are encountered (5898)
3249 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3250 * version 1.4.0.10 (1.4.1 release candidate 1)
3252 - don't keep a lock reference when lock is not granted (4238)
3253 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3254 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3255 - drop import inflight refcount on signal_completed_replay error (5255)
3256 - unlock page after async write error during send (3677)
3257 - handle missing objects in filter_preprw_read properly (5265)
3258 - no transno return for symlink open, don't save no-trasno open (3440)
3259 - don't try to complete elan receive that already failed (4012)
3260 - free RPC server reply state on error (5406)
3261 - clean up thread from ptlrpc_start_thread() on error (5160)
3262 - readahead could read extra page into cache that wasn't ejected (5388)
3263 - prevent races in class_attach/setup/cleanup/detach (5260)
3264 - don't dereference de->d_inode after l_dput of de (5458)
3265 - use "int" for stripe value returned from lock_to_stripe (5544)
3266 - mballoc allocation and error-checking fixes in 2.6 (5504)
3267 - block device patches to fix I/O request sizes in 2.6 (5482)
3268 - look up hostnames for IB nals (5602)
3269 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3270 - don't start multiple acceptors for the same port (5277)
3271 - fix incorrect LASSERT in mds_getattr_name (5635)
3272 - export a proc file for general "ping" checking (5628)
3273 - fix "lfs check" to not block when the MDS is down (5628)
3275 - service request history (4965)
3276 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3277 - create an "evict_client" /proc entry on OSTs, like the MDS has
3278 - fix mount usage message, return errors per mount(8) (5168)
3279 - change grep [] to grep "[]" in tests so they work in more UMLs
3280 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3281 - remove ancient LOV_MAGIC_V0 EA support (5047)
3282 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3283 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3284 - allow readahead allocations to fail when low on memory (5383)
3285 - mmap locking landed again, after considerable improvement (2828)
3286 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3288 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3291 - send OST transaction number in read/write reply to free req (4966)
3292 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3293 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3294 - the watchdog thread now runs as interruptible (5246)
3295 - flock/lockf fixes (but it's still disabled, pending 5135)
3296 - don't use EXT3 constants in llite code (5094)
3297 - memory shortage at startup could cause assertion (5176)
3299 - reorganization of lov code
3300 - single portals codebase
3302 - add extents/mballoc support (5025)
3303 - direct I/O reads in the obdfilter (4048)
3304 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3306 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3309 - send OST transaction number in read/write reply to free req (4966)
3310 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3311 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3312 - the watchdog thread now runs as interruptible (5246)
3313 - handle missing objects in filter_preprw_read properly (5265)
3314 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3315 - drop import inflight refcount on signal_completed_replay error (5255)
3316 - unlock page after async write error during send (3677)
3317 - return original error code on reconstructed replies (3761)
3318 - no transno return for symlink open, don't save no-trasno open (3440)
3320 - add pid to ldlm debugging output (4922)
3321 - bump the watchdog timeouts -- we can't handle 30sec yet
3322 - extra debugging for orphan dentry/inode bug (5259)
3324 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3327 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3328 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3329 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3330 - fix NULL dereference in /proc/sys/portals/routes (4827)
3331 - allow failed mdc_close() operations to be interrupted (4561)
3332 - stop precreate on OST before MDS would time out on it (4778)
3333 - don't send partial-page writes before EOF from client (4410)
3334 - discard client grant for sub-page writes on large-page clients (4520)
3335 - don't free dentries not owned by NFS code, check generation (4806)
3336 - fix lsm leak if mds_create_objects() fails (4801)
3337 - limit debug_daemon file size, always print CERROR messages (4789)
3338 - use transno after validating reply (3892)
3339 - process timed out requests if import state changes (3754)
3340 - update mtime on OST during writes, return in glimpse (4829)
3341 - add mkfsoptions to LDAP (4679)
3342 - use ->max_readahead method instead of zapping global ra (5039)
3343 - don't interrupt __l_wait_event() during strace
3345 - add software watchdogs to catch hung threads quickly (4941)
3346 - make lustrefs init script start after nfs is mounted
3347 - fix CWARN/ERROR duplication (4930)
3348 - return async write errors to application if possible (2248)
3349 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3350 - print NAL number in %x format (4645)
3351 - update barely-supported suse-2.4.21-171 series (4842)
3352 - support for sles 9 %post scripts
3353 - support for building 2.6 kernel-source packages
3354 - support for sles km_* packages
3356 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3359 - ignore -ENOENT errors in osc_destroy (3639)
3360 - notify osc create thread that OSC is being cleaned up (4600)
3361 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3362 - reconstruct ost_handle() like mds_handle() (4657)
3363 - create a new thread to do import eviction to avoid deadlock (3969)
3364 - let lconf resolve symlinked-to devices (4629)
3365 - don't unlink "objects" from directory with default EA (4554)
3366 - hold socknal file ref over connect in case target is down (4394)
3367 - allow more than 32000 subdirectories in a single directory (3244)
3368 - fix blocks count for O_DIRECT writes (3751)
3369 - OST returns ENOSPC from object create when no space left (4539)
3370 - don't send truncate RPC if file size isn't changing (4410)
3371 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3372 - bind to privileged port in socknal and tcpnal (3689)
3374 - rate limit CERROR/CWARN console message to avoid overload (4519)
3375 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3376 - basic mmap support (3918)
3377 - kernel patch series update from b1_4 (4711)
3379 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3382 - avoid crash during MDS cleanup with OST shut down (2775)
3383 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3384 - don't use bad inodes on the MDS (3744)
3385 - dynamic object preallocation to improve recovery speed (4236)
3386 - don't hold spinlock over lock dumping or change debug flags (4401)
3387 - don't zero obd_dev when it is force cleaned (3651)
3388 - print grants to console if they go negative (4431)
3389 - "lctl deactivate" will stop automatic recovery attempts (3406)
3390 - look for existing locks in ldlm_handle_enqueue() (3764)
3391 - don't resolve lock handle twice in recovery avoiding race (4401)
3392 - revalidate should check working dir is a directory (4134)
3394 - don't always mark "slow" obdfilter messages as errors (4418)
3396 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3399 - don't close LustreDB during write_conf until it is done (3860)
3400 - fix typo in lconf for_each_profile (3821)
3401 - allow dumping logs from multiple threads at one time (3820)
3402 - don't allow multiple threads in OSC recovery (3812)
3403 - fix debug_size parameters (3864)
3404 - fix mds_postrecov to initialize import for llog ctxt (3121)
3405 - replace config semaphore with spinlock (3306)
3406 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3407 - don't allow enqueue to complete on a destroyed export (3822)
3408 - down write_lock before checking llog header bitmap (3825)
3409 - recover from lock replay timeout (3764)
3410 - up llog sem before sending rpc (3652)
3411 - reduce ns lock hold times when setting kms (3267)
3412 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3413 - fix NULL deref and obd_dev leak on setup error (3312)
3414 - replace some LBUG about llog ops with error handling (3841)
3415 - don't match INVALID dentries from d_lookup and spin (3784)
3416 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3417 - fix invalid assertion in ptlrpc_set_wait (3880)
3419 - add libwrap support for the TCP acceptor (3996)
3420 - add /proc/sys/portals/routes for non-root route listing (3994)
3421 - allow setting MDS UUID in .xml (2580)
3422 - print the stack of a process that LBUGs (4228)
3424 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3427 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3428 - make sure to unset replay flag from failed open requests (3440)
3429 - if default stripe count is 0, use OST count for inode size (3636)
3430 - update parent mtime/ctime on client for create/unlink (2611)
3431 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3432 - free recovery state on server during a forced cleanup (3571)
3433 - unregister_reply for resent reqs (3063)
3434 - loop back devices mounting and status check on 2.6 (3563)
3435 - fix resource-creation race that can provoke i_size == 0 (3513)
3436 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3437 - more debugging for page-accounting assertion (3746)
3438 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3439 - avoid deadlock after precreation failure (3758)
3440 - fix race and lock order deadlock in orphan handling (3450, 3750)
3441 - add validity checks when grabbing inodes from l_ast_data (3599)
3443 - add /proc/.../recovery_status to obdfilter (3428)
3444 - lightweight CDEBUG infrastructure, debug daemon (3668)
3445 - change default OSC RPC parameters to be better on small clusters
3446 - turn off OST read cache for files smaller than 32MB
3447 - install man pages and include them in rpms (3100)
3448 - add new init script for (un)mounting lustre filesystems (2593)
3449 - run chkconfig in %post for init scripts (3701)
3450 - drop scimac NAL (unmaintained)
3452 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3455 - clean kiobufs before and after use (3485)
3456 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3457 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3458 - change init script to fail more clearly if not run as root (1528)
3459 - allow clients to reconnect during replay (1742)
3460 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3461 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3462 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3463 - fix cond_resched (3554)
3464 - only evict unfinished clients after recovery (3515)
3465 - allow bulk resend, prevent data loss (3570)
3466 - dynamic ptlrpc request buffer allocation (2102)
3467 - don't allow unlinking open directory if it isn't empty (2904)
3468 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3469 - remove extraneous obd dereference causing LASSERT failure (3334)
3470 - don't use get_cycles() when creating temp. files on the mds (3156)
3471 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3472 - handle EEXIST for set-stripe, set proper directory name (3336)
3474 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3475 - fix ksocknal_fmb_callback() error messages (2918)
3477 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3480 - don't copy lvb into (possibly NULL) reply on error (2983)
3481 - don't deref dentry after dput, don't free lvb on error (2922)
3482 - use the kms to determine writeback rpc length (2947)
3483 - increment oti_logcookies when osc is inactive (2948)
3484 - update client's i_blocks count via lvb messages (2543)
3485 - handle intent open/close of special files properly (1557)
3486 - mount MDS with errors=remount-ro, like obdfilter (2009)
3487 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3488 - don't use cancelling-locks' kms values (2947)
3489 - use highest lock extent for kms, not last one (2925)
3490 - don't dereference ERR_PTR() dentry in error handling path (3107)
3491 - fix thread race in portals_debug_dumplog() (3122)
3492 - create lprocfs device entries at setup instead of at attach (1519)
3493 - common AST error handler, don't evict client on completion race (3145)
3494 - zero nameidata in detach_mnt in 2.6 (3118)
3495 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3496 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3497 - initialize RPC timeout timer earlier for 2.6 (3219)
3498 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3499 - print nal/nid for unknown nid (3258)
3500 - additional checks for oscc recovery before doing precreate (3284)
3501 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3502 - don't crash in mdc_close for bad permissions on open (3285)
3503 - zero i_rdev for non-device files (3147)
3504 - clear page->private before handing to FS, better assertion (3119)
3505 - tune the read pipeline (3236)
3506 - fix incorrect decref of invalidated dentry (2350)
3507 - provide read-ahead stats and refine rpc in flight stats (3328)
3508 - don't hold journal transaction open across create RPC (3313)
3509 - update atime on MDS at close time (3265)
3510 - close LDAP connection when recovering to avoid server load (3315)
3511 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3512 - don't leak open file on MDS after open resend (3325)
3513 - serialize filter_precreate and filter_destroy_precreated (3329)
3514 - loop device shouldn't call sync_dev() for nul device (3092)
3515 - clear page cache after eviction (2766)
3516 - resynchronize MDS->OST in background (2824)
3517 - refuse to mount the same filesystem twice on same mountpoint (3394)
3518 - allow llmount to create routes for mounting behind routers (3320)
3519 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3520 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3521 - fix rare problem with rename on htree directories (3417)
3523 - allow default OST striping configuration per directory (1414)
3524 - fix compilation for qswnal for 2.6 kernels (3125)
3525 - increase maximum number of MDS request buffers for large systems
3526 - change liblustreapi to be useful for external progs like lfsck (3098)
3527 - increase local configuration timeout for slow disks (3353)
3528 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3530 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3533 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3534 - don't overwrite extent policy data in reply if lock was blocked (2901)
3535 - drop filter export grants atomically with removal from device (2663)
3536 - del obd_self_export from work_list in class_disconnect_exports (2908)
3537 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3538 - swab reply message in mdc_close, other PPC fixes (2464)
3539 - fix destroying of named logs (2325)
3540 - overwrite old logs when running lconf --write_conf (2264)
3541 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3542 - fix race in target_handle_connect (2898)
3543 - mds_reint_create() should take same inode create lock (2926)
3544 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3545 - don't close files for self_export to avoid uninitialized obd (2936)
3546 - allow MDS with the same name as client node (2939)
3547 - hold dentry reference for closed log files for unlink (2325)
3548 - reserve space for all logs during transactions (2059)
3549 - don't evict page beyond end of stripe extent (2925)
3550 - don't oops on a deleted current working directory (2399)
3551 - handle hard links to targets without a parent properly (2517)
3552 - don't dereference NULL lock when racing during eviction (2867)
3553 - don't grow lock extents when lots of conflicting locks (2919)
3555 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3558 - account for cache space usage on clients to avoid data loss (974)
3559 - lfsck support in lustre kernel code (2349)
3560 - reduce journal credits needed for BRW writes (2370)
3561 - orphan handling to avoid losing space on client/server crashes
3562 - ptlrpcd can be blocked, stopping ALL progress (2477)
3563 - use lock value blocks to assist in proper KMS, faster stat (1021)
3564 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3565 - recovery for initial connections (2355)
3566 - fixes for mds_cleanup_orphans (1934)
3567 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3568 - block all file creations until orphan recovery completes (1901)
3569 - client remove rq_connection from request struct (2423)
3570 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3571 - recovery timer race (2670)
3572 - mdc_close recovey bug (2532)
3573 - ptlrpc cleanup bug (2710)
3574 - mds timeout on local locks (2588)
3575 - namespace lock held during RPCs (2431)
3576 - handle interrupted sync write properly (2503)
3577 - don't try to handle a message that hasn't been replied to (2699)
3578 - client assert failure during cleanup after abort recovery (2701)
3579 - leak mdc device after failed mount (2712)
3580 - ptlrpc_check_set allows timedout requests to complete (2714)
3581 - wait for inflight reqs when ptlrpcd finishes (2710)
3582 - make sure unregistered services are removed from the srv_list
3583 - reset bulk XID's when resending them (caught by 1138 test)
3584 - unregister_bulk after timeout
3585 - fix lconf error (2694)
3586 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3587 - readahead locks pages, leaves pending causing memory pressure (2673)
3588 - increase OST request buffers to 4096 on large machines (2729)
3589 - fix up permission of existing directories in simple_mkdir (2661)
3590 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3591 - don't assign transno to errored transactions (2742)
3592 - don't delete objects on OST if given a bogus objid from MDS (2751)
3593 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3594 - if rq_no_resend, then timeout request after recovery (2432)
3595 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3596 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3597 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3598 - MDS recovery shouldn't delete objects that it has given out (2730)
3599 - if enqueue arrives after completion, don't clobber LVB (2819)
3600 - don't unlock pages twice when trigger_group_io returns error (2814)
3601 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3602 - don't write pages to disk if there was an error (1450)
3603 - don't ping imports that have recovery disabled (2676)
3604 - take buffered bytes into account when balancing socknal conn (2817)
3605 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3606 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3607 - remove little-endian swabbing of llog records (1987)
3608 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3609 - retry reposting request buffers if they fail (1191)
3610 - grow extent at grant time to avoid granting a revoked lock (2809)
3611 - lock revoke doesn't evict page if covered by a second lock (2765)
3612 - disable VM readahead to avoid reading outside lock extents (2805)
3614 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3615 - updated kernel patches for hp-2.4.20 kernel (2681)
3617 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3620 - fix truncated write corruption (2366)
3621 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3623 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3624 - ensure len(uuid) < 37 in lmc (1171)
3625 - fix ia64 OOPS in llog_test (2255)
3626 - zero end of page at obdfilter for partial page writes (2648)
3627 - don't leave stale dentries around after renames (bug 2428)
3628 - fix timeouts when evicting a client with a single lock held (2642)
3629 - set deadline for the initial HELLO message to drain (2634)
3630 - print out dotted-quad IP addresses in the socknal (2302)
3632 - additional debugging for MDS client eviction problem (2443)
3633 - fix mkfsoptions support for osts (2603, 2604)
3635 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3638 - add series for the vanilla 2.6.0 kernel
3639 - add series for the vanilla 2.4.24 kernel
3640 - add series for a cray x86/64 UL kernel drop
3641 - fix xattr patches for the vanilla 2.4.19 series
3643 - generate true UUIDs in lmc (1171)
3644 - have portals stack dumping break in UML (2466)
3645 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3646 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3647 - wake callback waiting threads on client eviction (2460)
3648 - Add --ptldebug and --subsystem to lmc (1719)
3649 - update assertion to allow safe interrupt allocation
3650 - set rq_no_resend for cancel requests (2432)
3651 - recalculate ptlrpcd timeout after resend (2494)
3652 - call vfs_rmdir when removing pending directories (2368)
3653 - fix renaming a file to itself (2429)
3654 - lmc creates a default one-stripe lov (2454)
3655 - expand procfs space to handle large clusters (2326)
3656 - increase UML stack to avoid overflow
3657 - update lconf's list of debug and subsystem masks
3658 - fix lfs find --obd (2510)
3659 - /proc tunable for disabling filter read caching (2591)
3660 - stop rpm packages from altering slapd.conf (2301)
3661 - disable nagle in the socknal under 0conf (2578)
3662 - choose mds inode size based on stripe count (2572)
3663 - fix kernel-source rpm problems (2516)
3665 - add --disable-doc to avoid pdf generation (2421)
3666 - update documentation, tests, type-os, comments
3667 - avoid format warnings on ia64
3668 - remove the TOE NAL
3669 - tiny code cleanups by removing unused fields
3671 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3674 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3675 - remove the most blatant lies from BUILDING file (2371)
3676 - change default debug level to reasonable production setting
3677 - reduce client side cache size to reduce cache flush time
3678 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3679 - make TCP zerocopy and pinger support enabled by default (2476)
3680 - sync writes completed after process exits caused crashes (2319)
3681 - maintain correct mount count on the MDS (2356)
3682 - backout 1557, because 2316 wasn't really fixed
3683 - better file I/O statistics gathering in /proc
3684 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3685 - another kernel patch to fix zero-copy TCP function export
3686 - don't take duplicate lock when processing re-sent getattr (2420)
3687 - lctl uses obd_self_export instead of creating new conn (2353)
3688 - MDS/OST recovery case which requires object creation asserted (2425)
3689 - move lfs from /usr/sbin to /usr/bin in packages
3690 - fix race between mds_client_add and mds_client_free (2417)
3691 - use kmalloc instead of slabs in portals (2430)
3692 - don't create duplicate records when a failover MDS is present (2442)
3693 - remove unnecessary mount age check (2332)
3694 - don't remove directory inodes from locks prematurely (2451)
3695 - don't break if MDS service name is the same as hostname (2103)
3696 - fix races in client write RPC generation when cache full (2482)
3698 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3701 - remove now-unused request->rq_obd (278)
3702 - if an allocation fails, print out how much memory we've used (1933)
3703 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3704 - add the "configurable stack size" patch to most series files (1256)
3705 - ability to write large log records, for 100+ OST configs (2306)
3706 - fix NULL deref when filter_prep fails (2314)
3707 - fix operator precedence error in filter_sync
3708 - dynamic allocation of socknal TX descriptors (2315)
3709 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3710 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3711 - ensure that CWARN messages in Portals always get to the syslog
3712 - __init/__exit are not for prototype decls (ldlm_init/exit)
3713 - x86-64 compile warning fixes
3714 - fix gateway LMC keyword conflict (2318)
3715 - fix MDS lock inversions in getattr/reint paths (1844)
3716 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3717 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3718 - DLM race condition prevented some lock evictions (2328)
3719 - ENOMEM detection and retry on socknal sends (2230)
3720 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3721 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3722 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3723 - better error messages when a client is rejected during recovery (1505)
3724 - avoid cancelling locks which were never granted, after failure (2330)
3725 - fix i_sem/journal inversion in mds_client_add (2333)
3726 - fix truncate/getattr lock cycle deadlock (2334)
3727 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3728 - fix two rare exit paths which could leak an l_lock() ref (2321)
3729 - fencepost error in MDS/OST orphan recovery (2226)
3730 - make log record alignment 8 bytes (1988)
3731 - lstripe now fails when requested offset > ost_count (2237)
3732 - ensure that all kernel series have a complete list.h (1607)
3733 - fix crashes in special-file operations (2316)
3734 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3736 - allow configurable automake binary, for testing new versions
3737 - small update to the lfs documentation
3739 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3741 * fix negative export reference count in fsfilt_sync (2312)
3743 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3744 * release candidate 0.9.1
3746 - orphans are moved into the PENDING directory for possible recovery
3747 - replayed opens now open by fid for orphan/rename safety (1042)
3748 - last close of an orphan inode generates a transno (683)
3749 - chdir() and mount() now pin the directory entry (1020)
3750 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3751 - discard very old requests without processing them (1502)
3752 - remove some common, well-understood CERRORs (1505)
3753 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3754 - clear "grant" flags in OST replies until OST grant code lands (1644)
3755 - fix read performance by not clobbering i_blksize on client (1598)
3756 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3757 - make LRU size a /proc tunable, clears locks when reduced (707)
3758 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3759 - server locks take references on exports now (1558)
3760 - build fixes for 2.4.20-rh trees (1663)
3761 - return an error from lov_create if all OSCs are inactive (1751)
3762 - fix import levels when a reconnect happens without a timeout (1597)
3763 - exit early from mds_open if we get a lookup error (1749)
3764 - partial page read at EOF wouldn't wait for disk before sending (1642)
3765 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3766 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3767 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3768 - write/truncate lock inversion (1639)
3769 - Don't auto-load obdclass, portals modules during cleanup (1495)
3770 - fix timestamps from jumping to "now" (1763)
3771 - extra journal assertions (1648)
3772 - add an extra multiunlink test (1771)
3773 - fix read_record/write_record API (1776)
3774 - fix leak of offset_extent, possible incorrect i_size later (1772)
3775 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3776 - Debugging for the kqswnal_get_idle_tx problems (1820)
3777 - Allow recovery to be attempted multiple times (1536)
3778 - Write out MDS last_rcvd file after it is first created (1600)
3779 - Fix tx_descriptor leak in failed transmit situations (1827)
3780 - ext3 journaling fixes for assertion failure after IO error (1871)
3781 - class_export_put() on freed export after completion AST error (1896)
3782 - Fix revalidate looping in VFS (1322)
3783 - Don't access a freed export during MDS_REINT timeout (1521)
3784 - Add open-unlink recovery support on the MDS (1673,1764)
3785 - Return an error if no MDS data was read from last_rcvd (1946)
3786 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3787 - Don't setup a disk device that doesn't match exported UUID (317)
3788 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3789 - avoid committing NULL handle in force close
3790 - local.sh is now a one-stripe LOV configuration
3791 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3792 - fix ext3 htree duplicate directory entry corruption (1516)
3793 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3794 - update to new LOV EA format (2097)
3795 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3797 2003-06-15 Phil Schwan <phil@clusterfs.com>
3800 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3801 - per-import/export recovery handling (958, 931, 959)
3802 - multiple last-rcvd slots, for serving multiple FSes (949)
3803 - connections are again shared between multiple imp/exports (963, 964)
3804 - "umount -f" would hang if any requests needed to be sent (393, 978)
3805 - avoid pinning large req buffer by copying for queued messages (989)
3806 - add "uuid" to "lctl device" command to help upcalls (991)
3807 - "open" RPCs with transnos would confuse recovery counters (1037)
3808 - do proper endian conversion of all wire messages (288, 340, 891)
3809 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3810 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3811 - fix ptlrpc_connection leak in target_handle_connect (1174)
3812 - fix import refcounting bug in OST and MDS cleanup (1134)
3813 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3814 - fix very unlikely obd_types race condition (501)
3815 - remove osc_open hack for echo_client (1187)
3816 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3817 - a failure in read_inode2 leads to deadlock (1139)
3818 - cancel ack-locks as soon as transaction is committed (1072)
3819 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3820 - make sure to commitrw after any preprw to avoid deadlock (1162)
3821 - failing to execute a file in a lustre FS would lock inode (1203)
3822 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3823 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3824 - obd_uuid2tgt would walk off the end of the list (1255)
3825 - on IA64 the osc would give portals incorrect bulk size (1258)
3826 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3827 - fix lock inversion caused by new llite matching code (1282)
3828 - limit the number of dirty pages on a client to 10MB (1286)
3829 - timed out locks were not being corrected cancelled (1289)
3830 - fix O_DIRECT above 4GB on IA-32 (1292)
3831 * major user-visible changes
3832 - fail out/fail over policy now controlled by the upcall (993)
3834 - add OBD_PING to check server availability and failure (954)
3835 - lustre messages are now sent in sending host order (288, 340, 891)
3836 - add eadatalen to MDS getattr reply (340)
3837 - OST read replies may contain second buffer, with per-page status (593)
3839 2003-03-11 Phil Schwan <phil@clusterfs.com>
3842 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3843 - failed open()s could cause deadlock; fixed (867, 869)
3844 - stop cancelling OST locks when files are closed (481)
3845 - overlapping XID spaces caused network corruption (851, 853)
3846 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3847 - setattr_raw added, to do single-RPC, server-side setattrs
3848 - lmc/lconf syntax change for OST UUIDs
3849 - fix crashy race condition between ptlrpc_free_req and osc_close
3850 - don't use request in mdc_enqueue if we hit a timeout (889)
3851 - don't set the inode i_size for regular files from the MDS (896)
3852 - handle out of order completion AST (842)
3853 - don't LBUG if a lock request times out after receiving AST (913)
3854 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3855 - if a bad lock AST arrives, send an error instead of dropping entirely
3856 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3857 - fix leak in bulk IO when only partially completed (899, 900, 926)
3858 - fix O_DIRECT for ia64 (55)
3859 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3860 - C-z after timeout could hang a process forever; fixed (977)
3862 - client-side I/O cache (678, 924, 929, 941, 970)
3864 - READPAGE and SETATTRs which don't take server-side locks get
3867 2003-02-11 Phil Schwan <phil@clusterfs.com>
3870 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3871 - Fix fsfilt_extN_readpage() to read a full page of directory
3872 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3873 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3874 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3875 - Most intent operations were removed, in favour of a new RPC mode
3876 that does a single RPC to the server and bypasses most of the VFS
3877 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3878 - Aggressively cancel local locks on DLM servers
3879 - mds_reint_unlink sends EA to the client if it's the last nlink.
3880 client uses that EA to unlink OST objects.
3881 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3882 - recursive symlinks were fixed (439)
3883 - fixed NULL deref in DEBUG_REQ
3884 - filter_update_lastobjid no longer calls sync, which annoyed extN
3885 - fixed multi-client small-writes to a single file problem (445)
3886 - fixed mtime updates during file writes (607)
3887 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3888 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3889 - fixed Linux OST size reporting problem (444, 656)
3890 - OST now updates object mtime with writes or setattr (607, 619)
3891 - client verifies file size before zeroing page past EOF (445)
3892 - OST now writes last allocated objid to disk with allocation (108)
3893 - LOV on echo now works (409)
3895 - mds_reint_unlink sends a new buffer, with the EA included. this
3896 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3897 set if a regular file was being unlinked, and it was the last link
3898 - use PtlGet from the target for bulk writes (315)
3899 - OST now updates object mtime with writes or setattr (607, 619)
3900 - LDLM now has a grant-time callback to revalidate locked items, if
3902 - Many MDS operations were reorganized to combat race conditions
3904 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3905 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3907 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3910 - Fully reactivate OST imports after reconnection (512, others)
3911 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3912 - More graceful error handling for truncating on dead OST (515)
3913 - Don't error out unless we're actually accessing dead stripes (474)
3914 - Fix garbage sizes when stripes are missing (410)
3915 - LRU counters were broken, causing constant lock purge (433, 432)
3916 - garbage on read from stripes with failed OSTs (441)
3917 - mark OSCs as active before reconnecting during recovery (438)
3918 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3919 - lfind did not preserve OST order in output (443)
3920 - symlinks cause hung clients, incorrect data (439)
3921 - stop dereferencing request after dropping refcount (457)
3922 - don't LASSERT(spin_is_locked) on non-SMP (455)
3923 - fixes for many rename() bugs
3924 - fstat didn't correctly synchronize attributes (399)
3925 - server must handle lock cancellation during blocking AST prep (487)
3926 - bulk descriptors were free()d too soon (511)
3927 - fix paths in lconf, which would load incorrect modules (451, 507)
3928 - fix confusing lconf 'host not found' error message (386)
3929 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3930 - fix race condition in mdc_blocking_ast() for inode access (526)
3931 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3932 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3933 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3934 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3935 - fix lstripe in several ways (kernel side) (527)
3936 - fix request leak in ldlm_cli_enqueue (262)
3937 - incorrect OSC was marked inactive after OST failure
3938 - call mds_fs_cleanup before unmounting filesystem (524)
3939 - fix races between taking ns_lock and ldlm_lock_change_resource
3940 - fix races updating LOV export open file list
3941 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3942 - fix recovery NULL deref in ldlm_cli_cancel_unused
3943 - fix some DLM races by using new hash table for lock handles (419)
3944 - permit the client to specify desired inodes, at replay
3945 - duplicate requests when we queue them for replay reintegration
3946 - fix last_rcvd offset calculation
3947 - sync after each recovered transaction, so we always make progress
3948 - never, not always, ERESTART requests without transnos
3949 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3951 - skip replay if the MDS says that the client is already connected
3952 - don't check for a recovery-enabled export to match lctl's UUID
3953 - don't INC_USE_COUNT for phantom exports
3954 - don't crash when cleaning up phantom exports (567)
3955 - don't double-finish or set replay data for errored mdc_open requests
3956 - abort requests when they time out, so we don't get old replies
3957 - send/receive replies for AST messages again
3958 - if the client says that it doesn't have the lock, cancel it on the
3960 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3961 mark it as destroyed, it will all work out when decref is called
3962 - fix module use counts (22, 581)
3964 - ASTs now expect a reply (server cancels lock on error reply)
3966 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3969 - fix many simultaneous client startup (392)
3970 - fix dentry->d_it clobbering
3971 - credentials weren't being shipped for readdir/getattr operations
3972 - remove invalid assertions triggered during some concurrent MD
3974 - proper Lustre versions added (336, 389)
3975 - fix memory leak for create error case (398)
3976 - fix LOV locking bug that would get cli/srv out of sync
3977 - fix echo client over LOV (409)
3978 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3979 - fix double-O_EXCL intent crash (424)
3980 - avoid sending multiple lock CANCELs (352)
3982 - MDS can do multi-client recovery (modulo bugs in new code)
3984 - many updates, edits, cleanups
3986 2002-11-18 Phil Schwan <phil@clusterfs.com>
3989 - fix null d_it dereference (346)
3990 - fix full OST/dbench hang (333)
3991 - fix permission problem with file removal (286)
3992 - fix removal of OSCs from LOV when they fail
3993 - fix NULL deref during bulk timeout (214)
3994 - fix problems related to multiple filesystems on one MDS (241)
3995 - fixed serious subtle metadata locking bugs
3996 - free locks on clients when inodes are removed due to memory
3998 - fix inode pointer in lock data (285)
3999 - partial support for multiple MDS on a single host (241)
4000 - data locks weren't cancelled at clear_inode time (290, 311)
4001 - intent locks could lead to unbounded lock growth (205)
4002 - added a maximum lock count, an LRU list, and a flusher
4003 - fix multiple rename (365)
4004 - properly abstracted the echo client
4005 - OSC locked 1 byte too many; fixed
4006 - rewrote brw callback code:
4007 - fixed recovery bugs related to LOVs (306)
4008 - fixed too-many-pages-in-one-write crash (191)
4009 - fixed (again) crash in sync_io_timeout (214)
4010 - probably fixed callback-related race (385)
4012 - Add capability to MDS protocol
4013 - LDLM cancellations and callbacks on different portals
4015 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4018 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4019 - defer open object destruction to close time (601981, 138)
4020 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4021 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4022 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4023 - don't call obd_setattr in ll_file_release for destroyed objects
4024 * protocol change to lustre_msg: move |version| and add |flags|
4025 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4026 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4027 * added replay of create, unlink, link and rename operations during
4028 MDS failover; recovery should be much more robust now
4029 * remove failed OSCs from LOVs (only lov_create uses this so far)
4030 * the lustre-HOWTO was brought (more) up to date (582544)
4032 2002-10-23 Phil Schwan <phil@clusterfs.com>
4035 - in-use dentries weren't being reused properly (617851)
4036 - prevent multiple LDLM setup (599178)
4037 - fix LOV size calculations for truncate (617853)
4038 - fix client handling of MDS intent errors (POSIX)
4039 - fix permission bug in lovstripe.c test (624321)
4040 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4041 - truncate past end of file could corrupt data
4042 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4043 - a race in recovery could return ETIMEDOUT to apps (623947)
4044 - building outside the source directory was fixed
4045 * the lustre-HOWTO was brought (more) up to date (582544)
4046 * major progress was made on recovery functionality
4048 2002-10-10 Phil Schwan <phil@clusterfs.com>
4051 - recovery deadlock fix
4052 - rm -rf causes LBUG fix (617817)
4053 - file open by multiple tasks fix (618962)
4054 - directory permissions bugs (602707 and 620007)
4055 - journal_stop fixed with locking (611313)
4056 - O_APPEND failures resolved (618273, perhaps 614459)
4057 - lconf PATH fix (619770)
4058 - IA64 build fix (621450)
4059 - RPC buffer sizes scale with amount of memory
4061 2002-10-01 Phil Schwan <phil@clusterfs.com>
4064 - locks would be cancelled without throwing away data pages,
4065 resulting in inconsistent data (605627)
4066 - inode attributes were not always being refreshed (605627, 612449)
4067 - lconf now continues to cleanup after lctl reports an error
4068 - MDS now enforces user permissions (602707)
4069 - lprocfs cleanup fixed, but not yet enabled (614157)
4070 - fixed infinite server hang, should a client not respond to an AST
4071 - avoid going into recovery if user calls readlink() with a buffer
4072 that's too small (613941)
4073 - AST RPCs no longer require replies (614867) -- this may be changed
4074 - don't crash server if client sends an IOV that's too big (611336)
4075 - fixed lock conversion deadlock (611892)
4076 - fixed the following of symlinks (614622)
4077 * recovery: the server can remove locks from a client that dies, other
4078 clients can make progress
4079 * more extN patch fixes
4080 * compile-time configurable ptlrpc buffer allocations
4082 - collaborative read cache document
4083 - Lustre Lite Performance CDR document-in-progress
4085 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4088 - fix typo in patch-2.4.18
4090 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4093 - clear ptlrpc request each time in handle_incoming_request()
4094 - unlink of files now destroys the object on the OST
4096 2002-09-19 Peter Braam <braam@clusterfs.com>
4098 * add hard link support
4099 * change obdfile creation method
4100 * kernel patch changed
4102 2002-09-19 Peter Braam <braam@clusterfs.com>
4105 - stack overflow bug in extN fixed
4107 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4109 * documentation updates
4110 - add man pages for config tools
4111 - update tests/README to describe testing with new config tools
4112 - finish metadata API descriptions
4113 * bug fixes and cleanups
4114 - statfs workaround for 16TB limit
4115 - LOV stripe allocation improved, can stripe on subset of OSTs
4116 - LOV file size/IO offset was wrong for files > 4GB in size
4117 - object EA data was being dropped, caused files to be unreadable
4118 - memory overflow with non-LOV OST caused memory corruption
4119 - fixed regression tests to work with new config tools, obdfilter
4120 - fixed bug when directory size became larger than 1 block
4121 - fixed bug (for single client case) when PWD was deleted
4122 - invalidate local directory pages when doing intent-based ops
4123 - avoid LDLM oops when lock callback contained bad data
4125 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4127 * documentation updates
4128 * bug fixes and cleanups
4129 - configuration tools
4132 - 64-bit compile warnings
4133 - 64-bit internal statfs data
4135 * test_brw on persistent OST devices
4137 * lprocfs (disabled)
4139 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4141 * documentation updates
4142 * bug fixes and cleanups
4143 * configuration tools
4145 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4148 * many small fixes to 0_5_4
4149 * io/network handling
4150 * thinkos in MDS operations
4152 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4155 * crucial basic fixes to 0.5.3
4156 * IOR, Iozone work over Elan
4159 2002-08-07 Phil Schwan <phil@clusterfs.com>
4160 * version 0_5_3, our first alpha
4161 * we use the new Portals iovs
4162 * documentation updates
4163 * bug fixes and cleanups
4164 * small changes in the DLM wire protocol
4166 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4167 * version 0_5_1 with some initial stability,
4168 * locking on MD and file I/O.
4169 * documentation updates
4170 * several bug fixes since 0.5.0
4171 * small changes in wire protocol
4173 2002-07-18 Phil Schwan <phil@clusterfs.com>
4175 * delivered as Lustre Light Alpha
4176 * fixed a crash after handling invalid MDS requests
4177 * fixed directory pages for architectures with non-4k pages sizes
4179 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4180 * release version v0_4_4
4181 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4183 * Fixes a number of interruption problems with OST operations.
4184 * Update documentation for portals header changes
4185 * Move all wire protocol structs/defines to lustre_idl.h
4186 * Fixes symlink length bug.
4187 * Add tcpdump to repository.
4189 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4190 * release version v0_4_3
4191 * Fixes statfs for inodes on extN.
4192 * Fixes bug in runtests which would delete /etc/hosts.
4193 * Use 64-bit object IDs wherever possible (not into VFS though)
4194 Remove ost_get_info, which is unused by lustre, and out of date.
4196 2002-07-03 Peter Braam <braam@clusterfs.com>
4197 * release version v0_4_2 Fixes a lookup error (type not passed)
4198 * move forward to head of Portals
4199 * move forward to latest Lustre kernel
4201 2002-06-25 Peter Braam <braam@clusterfs.com>
4202 * release version v0_4_1. Hopefully stable on single node use.