4 2.6.9-55.0.9.EL (RHEL 4),
5 2.6.16.53-0.8 (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.
17 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
18 Details : Update which_patch & target file for SLES10 latest kernel.
21 Description: Remove client patches from SLES 10 kernel.
22 Details : This causes SLES 10 clients to behave as patchless clients
23 even on a Lustre-patched (server) kernel.
25 Severity : enhancement
27 Description: self-adjustable client's lru lists
28 Details : use adaptive algorithm for managing client cached locks lru
29 lists according to current server load, other client's work
30 pattern, memory activities, etc. Both, server and client
31 side namespaces provide number of proc tunables for controlling
34 Severity : enhancement
36 Description: light-weight GSS support
37 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
38 and reduce performance overhead.
40 Severity : enhancement
42 Description: Linux keyring support
43 Details : Support using service of Linux keyring for Lustre GSS internal
44 context refresh/cache mechanism.
48 Description: Fix errors in lfs documentation
49 Details : Fixes man pages
53 Description: don't use GFP_* in generic Lustre code.
54 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
58 Description: obdclass is limited by single OBD_ALLOC(idarray)
59 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
63 Description: warnings with build patchless client with vanila 2.6.19 and up
64 Details : change old ctl_table style and replace ctl_table/ctl_table_header
65 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
69 Frequency : Only for SLES
70 Description: sanity_quota fail test_1
71 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
72 file, which makes TSTID[2] inunique.
76 Frequency : Only for SLES9
77 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
78 Details : Update target/ChangeLog/which_patch .
82 Frequency : Only for test_10 in sanity-quota.sh
83 Description: a bug in quota test code
84 Details : checking a test flag in a wrong place
88 Frequency : Only for test_13 in sanity-quota.sh
89 Description: a bug in quota test script
90 Details : shouldn't check overfull allocated quota
94 Description: jbd statistics
95 Details : Port older jbd statistics patch for sles10
97 Severity : enhancement
99 Description: Add printing inode info into message about error in writepage.
104 Description: reply_lock_interpret crash due to race with it and lock cancel.
105 Details : Do not replay locks that are being cancelled. Do not reference
106 locks by their address during replay, just by their handle.
110 Frequency : When flocks are used.
111 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
112 Details : Flock locks might destroy just granted lock if it could be merged
113 with another existing flock, this is done in completion handler,
114 so teach ldlm_cli_enquque_fini that this is a valid case for
120 Description: Oops in read and write path when failing to allocate lock.
121 Details : Check if lock allocation failed and return error back.
125 Description: lstripe command fails for valid OST index
126 Details : The stripe offset is compared to lov->desc.ld_tgt_count
127 instead of lov->desc.ld_active_tgt_count.
131 Description: sanity.sh failed test 103
132 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
137 Description: df doesn't work properly if diskfs blocksize != 4K
138 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
142 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
143 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
147 Description: obd_health_check_timeout too short
148 Details : set obd_health_check_timeout as 1.5x of obd_timeout
152 Description: llapi_file_create() does not allow some changes
153 Details : add llapi_file_open() that allows specifying the mode and
154 open flags, and also returns an open file handle.
158 Description: merge and cleanup kernel patches.
159 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
163 Description: Add journal checksum support.(Kernel part)
164 Details : The journal checksum feature adds two new flags i.e
165 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
166 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
167 indicates that the commit block contains the checksum for
168 the blocks described by the descriptor blocks. Now commit
169 record can be sent to disk without waiting for descriptor
170 blocks to be written to disk. This behavior is controlled
171 using JBD2_FEATURE_ASYNC_COMMIT flag.
175 Description: OSS needs mutliple precreate threads
176 Details : Add ability to start more than one create thread per OSS.
180 Description: Sanity.sh test_65k failed: setstripe should have succeeded
181 Details : execute lfs setstripe on client
185 Description: mds_obd_create error creating tmp object
186 Details : When the user sets quota on root, llog will be affected and can't
187 create files and write files.
190 Frequency : Always on ia64 patchless client, and possibly others.
192 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
193 Details : This allows the patchless client to be loaded on architectures
198 Description: RedHat Update kernel for RHEL5
199 Details : Add the kernel config file for RHEL5.
203 Description: RedHat Update kernel for RHEL5
204 Details : Modify the kernel config file more closer RHEL5.
208 Description: Build failure against Centos5 (RHEL5)
209 Details : Define PAGE_SIZE when it isn't present.
213 Description: test_7 on sanity-quota.sh can't be used on separate machines
214 Details : add facet to handle it
218 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
219 Details : with NFS, the anon dentry's parent was set to itself in
220 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
221 corresponding dentry other than use rec->ur_name.
223 Severity : enhancement
225 Description: lfs setstripe enhancement
226 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
230 Description: testing performance impact of enabling checksumming
231 Details : enable checksum by default, allow --disable-checksum
232 configure option and "-o nochecksum" mount option
236 Description: lustre support for RHEL5
237 Details : Add support for RHEL5.
241 Description: Client eviction due to failover config
242 Details : after a connection loss, the lustre client should attempt to
243 reconnect to the last active server first before trying the
244 other potential connections.
248 Description: when mds and osts use different quota unit(32bit and 64bit),
249 quota will be released repeatly.
250 Details : void sending multiple quota reqs to mds, which will keep the
251 status between the reqs.
255 Description: osts not allocated evenly to files
256 Details : change the condition to increase offset_idx
259 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
261 Description: Data corruption for OSTs that are formatted larger than 2TB
263 Details : When generating the bio request for lustre file writes the
264 sector number would overflow a temporary variable before being
265 used for the IO. The data reads correctly from Lustre (which
266 will overflow in a similar manner) but other file data or
267 filesystem metadata may be corrupted in some cases.
271 Description: Tune the kernel for good SCSI performance.
272 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
273 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
278 Description: rewrite ext2-derived code in obdclass/uuid.c
279 Details : rewrite inherited code (uuid parsing code from ext2 utils)
280 from scratch preserving functionality.
284 Description: "lfs find -obd UUID" prints directories
285 Details : "lfs find -obd UUID" will return all directory names instead
286 of just file names. It is incorrect because the directories
287 do not reside on the OSTs.
291 Description: use i_size_read and i_size_write in 2.6 port
292 Details : replace inode->i_size access with i_size_read/write()
295 Frequency : when removing large files
297 Description: scheduling issue during removal of large Lustre files
298 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
299 It causes scheduling issues when removing large files (17TB in the
303 Frequency : only with liblustre clients on XT3
305 Description: evictions taking too long
306 Details : allow llrd to evict clients directly on OSTs
309 Frequency : only on ppc
311 Description: /proc/fs/lustre/devices broken on ppc
312 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
313 We should make sure the type of 'index' is loff_t and then cast
314 explicitly as needed below. Do not assign an explicitly cast
318 Frequency : only for rhel5
320 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
321 Details : Modify the target file & which_kernel.
323 Severity : enhancement
325 Description: omit set fsid for export NFS
326 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
327 as FS_REQUIRES_DEV for avoid problems with generate fsid.
331 Frequency : Always, for kernels after 2.6.16
332 Description: Fix warning idr_remove called for id=.. which is not
334 Details : Last kernels save old s_dev before kill super and not allow
335 to restore from callback - restore it before call kill_anon_super.
339 Description: buffer overruns could theoretically occur
340 Details : llapi_semantic_traverse() modifies the "path" argument by
341 appending values to the end of the origin string, and a
342 overrun may occur. Adding buffer overrun check in liblustreapi.
346 Description: Fix error on 'ls .' at the top of the Lustre mount.
347 Details : Don't revalidate dentry if it is a root dentry.
351 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
352 Details : Modify vm-tunables-rhel4.patch.
356 Description: Kernel config for 2.6.18-vanilla.
357 Details : Modify targets/2.6-vanilla.target.in.
358 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
359 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
360 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
361 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
365 Description: improve handling recoverable errors
366 Details : if request processig with error which can be recoverable on server
367 request should be resend, otherwise page released from cache and
372 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
373 Details : Modify target file & which_patch.
374 A flaw was found in the IA32 system call emulation provided
375 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
376 value could be stored in the %RAX register, which could trigger an
377 out-of-bounds system call table access. An untrusted local user
378 could exploit this flaw to run code in the kernel
379 (ie a root privilege escalation). (CVE-2007-4573).
383 Description: Remove 2.4 kernel checks lustre 1.8
384 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
385 quota_interface.c in quota subsystem.
386 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
388 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
389 lvfs_linux.c upcall_cache.c in lvfs subsystem.
393 Description: O_DIRECT bypasses client statistics.
394 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
395 read_ahead_stats were updated. copied the stats section from
396 osc_send_oap_rpc() into async_internal().
400 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
404 Description: change order of libsysio includes
405 Details : '#include sysio.h' should always come before '#include xtio.h'
409 Description: handle "serious error: objid * already exists" more gracefully
410 Details : If LAST_ID value on disk is smaller than the objects existing in
411 the O/0/d* directories, it indicates disk corruption and causes an
412 LBUG(). If the object is 0-length, then we should use the existing
413 object. This will help to avoid a full fsck in most cases.
415 Severity : enhancement
417 Description: adapt the lustre_config script to support the upgrade case
418 Details : Add "-u" option for lustre_config script to support upgrading 1.4
419 server targets to 1.6 in parallel.
423 Description: To avoid grant space > avaible space when the disk is almost
424 full. Without this patch you might see the error "grant XXXX >
425 available" or some LBUG about grant, when the disk is almost
427 Details : In filter_check_grant, for non_grant cache write, we should
428 check the left space by if (*left > ungranted + bytes), instead
429 of (*left > ungranted), because only we are sure the left space
430 is enough for another "bytes", then the ungrant space should be
431 increase. In client, we should update cl_avail_grant only there
432 is OBD_MD_FLGRANT in the reply.
436 Description: Update RHEL 4 kernel to fix local root privilege escalation.
437 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
438 described in CVE-2007-4573. This problem could allow untrusted
439 local users to gain root access.
443 Description: Remove 2.4 kernel checks lustre 1.8
444 Details : Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
445 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
447 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
448 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
451 --------------------------------------------------------------------------------
453 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
455 * Support for kernels:
456 2.6.5-7.283 (SLES 9),
457 2.6.9-55.EL (RHEL 4),
458 2.6.16.46-0.14 (SLES 10),
459 2.6.18.8 vanilla (kernel.org)
460 * Client support for unpatched kernels:
461 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
462 2.6.16 - 2.6.22 vanilla (kernel.org)
463 * Due to recently discovered recovery problems, we do not recommend
464 using patchless RHEL 4 clients with this or any earlier release.
465 * Recommended e2fsprogs version: 1.39.cfs8
466 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
467 * Starting with this release, the ldiskfs backing filesystem required
468 by Lustre is now in its own package, lustre-ldiskfs. This package
469 should be installed. It is versioned separately from Lustre and
470 may be released separately in future.
474 Description: block reactivating mgc import until all deactivates complete
475 Details : Fix race when failing back MDT/MGS to itself (testing)
477 Severity : enhancement
479 Description: add optional extra BUILD_VERSION info
480 Details : add a new environment variable (namely LUSTRE_VERS) which allows
481 to override the lustre version.
484 Frequency : 2.6.18 servers only
486 Description: ll_kern_mount() doesn't release the module reference
487 Details : The ldiskfs module reference count never drops down to 0
488 because ll_kern_mount() doesn't release the module reference.
493 Description: server LBUG when using old ost_num_threads parameter
494 Details : Accept the old ost_num_threads parameter but warn that it
495 is deprecated, and fix an off-by-one error that caused an LBUG.
500 Description: Transient SCSI error results in persistent IO issue
501 Details : iobuf->dr_error is not reinitialized to 0 between two
505 Frequency : sometimes when underlying device returns I/O errors
507 Description: OSTs not going read-only during write failures
508 Details : OSTs are not remounted read-only when the journal commit threads
509 get I/O errors because fsfilt_ext3 calls journal_start/stop()
510 instead of the ext3 wrappers.
514 Description: poor connect scaling with increasing client count
515 Details : Don't run filter_grant_sanity_check for more than 100 exports
516 to improve scaling for large numbers of clients.
519 Frequency : SLES10 only
521 Description: sanity-quota.sh quotacheck failed: rc = -22
522 Details : Quotas cannot be enabled on SLES10.
525 Frequency : liblustre clients only
527 Description: getdirentries does not give error when run on compute nodes
528 Details : getdirentries does not fail when the size specified as an argument
529 is too small to contain at least one entry
531 Severity : enhancement
533 Description: Add LNET router traceability for debug purposes
534 Details : If a checksum failure occurs with a router as part of the
535 IO path, the NID of the last router that forwarded the bulk data
536 is printed so it can be identified.
541 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
542 Details : Due to a race condition, liblustre clients were occasionally
545 Severity : enhancement
547 Description: lfs setstripe use optional parameters instead of postional
550 Severity : enhancement
552 Description: Nanosecond timestamp support for ldiskfs
553 Details : The on-disk ldiskfs filesystem has added support for nanosecond
554 resolution timestamps. There is not yet support for this at
555 the Lustre filesystem level.
558 Frequency : during server recovery
560 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
561 Details : request with rq_no_resend flag not awake l_wait_event if they get a
565 Frequency : nfs export on patchless client
567 Description: connectathon hang when test nfs export over patchless client
568 Details : Disconnected dentry cannot be found with lookup, so we do not need
569 to unhash it or make it invalid
572 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
573 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
574 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
579 Description: open req refcounting wrong on reconnect
580 Details : If reconnect happened between getting open reply from server and
581 call to mdc_set_replay_data in ll_file_open, we will schedule
582 replay for unreferenced request that we are about to free.
583 Subsequent close will crash in variety of ways.
584 Check that request is still eligible for replay in
585 mdc_set_replay_data().
590 Description: disable writes to filesystem when reading health_check file
591 Details : the default for reading the health_check proc file has changed
592 to NOT do a journal transaction and write to disk, because this
593 can cause reads of the /proc file to hang and block HA state
594 checking on a healthy but otherwise heavily loaded system. It
595 is possible to return to the previous behaviour during configure
596 with --enable-health-write.
598 Severity : enhancement
600 Description: 64-bit inode version
601 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
602 to the inode. This will be required for version-based recovery.
607 Description: MDS fails to start if a duplicate client export is detected
608 Details : in some rare cases it was possible for a client to connect to
609 an MDS multiple times. Upon recovery the MDS would detect this
610 and fail during startup. Handle this more gracefully.
612 Severity : enhancement
614 Description: Add -o localflock option to simulate old noflock
616 Details : This will achieve local-only flock/fcntl locks
622 Description: log_commit_thread vs filter_destroy race leads to crash
623 Details : Take import reference before releasing llog record semaphore
628 Description: Wrong request locking in request set processing
629 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
630 imp_delayed_list, in this place should be used imp_lock.
633 Frequency : when reconnection
635 Description: Grant Leak when osc reconnect to OST
636 Details : When osc reconnect ost, OST(filter) should check whether it
637 should grant more space to client by comparing fed_grant and
638 cl_avail_grant, and return the granted space to client instead
639 of "new granted" space, because client will call osc_init_grant
640 to update the client grant space info.
643 Frequency : when client reconnect to OST
645 Description: Grant Leak when osc do resend and replay bulk write
646 Details : When osc reconnect to OST, OST(filter)should clear grant info of
647 bulk write request, because the grant info while be sync between
648 OSC and OST when reconnect, and we should ignore the grant info
649 these of resend/replay write req.
654 Description: Grant space more than avaiable left space sometimes.
655 Details : When then OST is about to be full, if two bulk writing from
656 different clients came to OST. Accord the avaliable space of the
657 OST, the first req should be permitted, and the second one
658 should be denied by ENOSPC. But if the seconde arrived before
659 the first one is commited. The OST might wrongly permit second
660 writing, which will cause grant space > avaiable space.
663 Frequency : when client is evicted
665 Description: Grant might be wrongly erased when osc is evicted by OST
666 Details : when the import is evicted by server, it will fork another
667 thread ptlrpc_invalidate_import_thread to invalidate the
668 import, where the grant will be set to 0. While the original
669 thread will update the grant it got when connecting. So if
670 the former happened latter, the grant will be wrongly errased
671 because of this race.
676 Description: Checking Stale with correct fid
677 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
678 to check whether it is stale, because sometimes, we want the
679 enqueue happened anyway, and op_data.fid2 will not be initialized.
681 Severity : enhancement
683 Description: update patchless client
684 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
687 Frequency : only with 2.4 kernel
689 Description: random memory corruption
690 Details : size of struct ll_inode_info is to big for union inode.u and this
691 can be cause of random memory corruption.
696 Description: Memory leak in recovery
697 Details : Lov_mds_md was not free in an error handler in mds_create_object.
698 It should also check obd_fail before fsfilt_start, otherwise if
699 fsfilt_start return -EROFS,(failover mds during mds recovery).
700 then the req will return with repmsg->transno = 0 and rc = EROFS.
701 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
702 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
703 be freed no matter whether fsfilt_commit success or not.
706 Frequency : only with huge count clients
708 Description: Prevents from taking the superblock lock in llap_from_page for
710 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
711 for llap_from_page call in ll_removepage prevents from taking the
712 superblock lock for a soon died page.
717 Description: Not check open intent error before release open handle
718 Details : in some rare cases, the open intent error is not checked before
719 release open handle, which may cause
720 ASSERTION(open_req->rq_transno != 0), because it tries to release
721 the failed open handle.
726 Description: Set cat log bitmap only after create log success.
727 Details : in some rare cases, the cat log bitmap is set too early. and it
728 should be set only after create log success.
732 Description: Accessing a block bevice can re-enable I/O when Lustre is
733 tearing down a device.
734 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
739 Description: service threads may hog cpus when there are a lot of requests
741 Details : Insert cond_resched to give other threads a chance to use some of
747 Description: the cat log was not initialized in recovery
748 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
749 unlink log on mds will not be initialized until mds post
750 recovery. And also in mds post recovery, the unlink log will
751 initialization will be done asynchronausly, so there will be race
752 between add unlink log and unlink log initialization.
756 Description: brw_stats were being printed incorrectly
757 Details : brw_stats were being printed as log2 but all of them were not
758 recorded as log2. Also remove some code duplication arising from
759 filter_tally_{read,write}.
763 Frequency : rare, only in recovery.
764 Description: ASSERTION(req->rq_type != LI_POISON) failed
765 Details : imp_lock should be held while iterating over imp_sending_list for
766 prevent destroy request after get timeout in ptlrpc_queue_wait.
770 Description: replay-single.sh test 52 fails
771 Details : A lock's skiplist need to be cleanup when it being unlinked
772 from its resource list.
776 Description: Short directio read returns full requested size rather than
778 Details : Direct I/O operations should return actual amount of bytes
779 transferred rather than requested size.
781 Severity : enhancement
783 Description: metadata RPC reduction (e.g. for rm performance)
784 Details : decrease the amount of synchronous RPC between clients and servers
785 by canceling conflicing lock before the operation on the client side
786 and packing thier handles into the main operation RPC to server.
788 Severity : enhancement
790 Description: Async OSC create to avoid the blocking unnecessarily.
791 Details : If a OST has no remain object, system will block on the creating
792 when need to create a new object on this OST. Now, ways use
793 pre-created objects when available, instead of blocking on an
794 empty osc while others are not empty. If we must block, we block
795 for the shortest possible period of time.
797 --------------------------------------------------------------------------------
799 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
804 Frequency : on some architectures
806 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
807 Details : Uninitialized flags sometimes cause configuration commands to
811 Frequency : patchless clients only
813 Description: missing __iget() symbol export
814 Details : The __iget() symbol export is missing. To avoid the need for
815 this on patchless clients the deathrow inode reaper is turned
816 off, and we depend on the VM to clean up old inodes. This
817 dependency was during via the fix for bug 12181.
819 --------------------------------------------------------------------------------
821 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
823 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
824 INTEROPERATE with older versions automatically. In many cases a
825 special upgrade step is needed. Please read the
826 user documentation before upgrading any part of a live system.
827 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
828 version will not interoperate with 1.6 betas before beta5 (1.5.95).
829 * WARNING: Lustre configuration and startup changes are required with
830 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
836 Severity : enhancement
838 Description: MountConf
839 Details : Lustre configuration is now managed via mkfs and mount
840 commands instead of lmc and lconf. New obd types (MGS, MGC)
841 are added for dynamic configuration management. See
842 https://mail.clusterfs.com/wikis/lustre/MountConf for
845 Severity : enhancement
847 Description: dynamic OST addition
848 Details : OSTs can now be added to a live filesystem
850 Severity : enhancement
852 Description: startup order invariance
853 Details : MDTs and OSTs can be started in any order. Clients only
854 require the MDT to complete startup.
856 Severity : enhancement
858 Description: parallel, asynchronous orphan cleanup
859 Details : orphan cleanup is now performed in separate threads for each
860 OST, allowing parallel non-blocking operation.
862 Severity : enhancement
864 Description: optimized stripe assignment
865 Details : stripe assignments are now made based on ost space available,
866 ost previous usage, and OSS previous usage, in order to try
867 to optimize storage space and networking resources.
869 Severity : enhancement
871 Description: Permanently set tunables
872 Details : All writable /proc/fs/lustre tunables can now be permanently
873 set on a per-server basis, at mkfs time or on a live
876 Severity : enhancement
878 Description: Lustre message v2
879 Details : Add lustre message format v2.
881 Severity : enhancement
883 Description: client OST exclusion list
884 Details : Clients can be started with a list of OSTs that should be
885 declared "inactive" for known non-responsive OSTs.
889 Description: SPEC SFS validation failure on NFS v2 over lustre.
890 Details : Changes the blocksize for regular files to be 2x RPC size,
891 and not depend on stripe size.
893 Severity : enhancement
895 Description: Multiple MD RPCs in flight.
896 Details : Further unserialise some read-only MDS RPCs - learn about intents.
897 To avoid overly-overloading MDS, introduce a limit on number of
898 MDS RPCs in flight for a single client and add /proc controls
899 to adjust this limit.
901 Severity : enhancement
903 Description: client read/write statistics
904 Details : Add client read/write call usage stats for performance
905 analysis of user processes.
906 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
907 file access. extents_stats shows chunk size distribution.
908 extents_stats_per_process show chunk size distribution per
911 Severity : enhancement
913 Description: mds statistics
914 Details : Add detailed mds operations statistics in
915 /proc/fs/lustre/mds/*/stats.
919 Description: Failure of copying files with lustre special EAs.
920 Details : Client side always return success for setxattr call for lustre
921 special xattr (currently only "trusted.lov").
926 Description: Refcount LNET uuids
927 Details : The global LNET uuid list grew linearly with every startup;
928 refcount repeated list entries instead of always adding to
931 Severity : enhancement
933 Description: Dynamic service threads
934 Details : Within a small range, start extra service threads
935 automatically when the request queue builds up.
938 Frequency : mixed-endian client/server environments
940 Description: mixed-endian crashes
941 Details : The new msg_v2 system had some failures in mixed-endian
944 Severity : enhancement
946 Description: Easy OST removal
947 Details : OSTs can be permanently deactivated with e.g. 'lctl
948 conf_param lustre-OST0001.osc.active=0'
950 Severity : enhancement
952 Description: MGS proc entries
953 Details : Added basic proc entries for the MGS showing what filesystems
956 Severity : enhancement
958 Description: provide MGS failover
959 Details : Added config lock reacquisition after MGS server failover.
961 Severity : enhancement
963 Description: add Linux 2.4 support
964 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
968 Description: a large application tries to do I/O to the same resource and dies
970 Details : Check the req->rq_arrival time after the call to
971 ost_brw_lock_get(), but before we do anything about
972 processing it & sending the BULK transfer request. This
973 should help move old stale pending locks off the queue as
974 quickly as obd_timeout.
977 Frequency : when an incorrect nid is specified during startup
979 Description: ptlrpc connect to non-existant node causes kernel crash
980 Details : LNET can't be re-entered from an event callback, which
981 happened when we expire a message after the export has been
982 cleaned up. Instead, hand the zombie cleanup off to another
985 Severity : enhancement
987 Description: plain/inodebits lock performance improvement
988 Details : Grouping plain/inodebits in granted list by their request modes
989 and bits policy, thus improving the performance of search through
993 Frequency : only if OST filesystem is corrupted
995 Description: client incorrectly hits assertion in ptlrpc_replay_req()
996 Details : for a short time RPCs with bulk IO are in the replay list,
997 but replay of bulk IOs is unimplemented. If the OST filesystem
998 is corrupted due to disk cache incoherency and then replay is
999 started it is possible to trip an assertion. Avoid putting
1000 committed RPCs into the replay list at all to avoid this issue.
1003 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1005 Description: System hang on startup
1006 Details : This bug allowed the liblustre (e.g. catamount) client to
1007 return to the app before handling all startup RPCs. This
1008 could leave the node unresponsive to lustre network traffic
1009 and manifested as a server ptllnd timeout.
1011 Severity : enhancement
1013 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1014 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1015 boolean to control whether to print peer debug info when a
1016 client's RPC times out.
1019 Frequency : only for kernels with patches from Lustre below 1.4.3
1021 Description: Remove old rdonly API
1022 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1025 Frequency : only for devices with external journals
1027 Description: Set external device read-only also
1028 Details : During a commanded failover stop, we set the disk device
1029 read-only while the server shuts down. We now also set any
1030 external journal device read-only at the same time.
1033 Frequency : when upgrading from 1.4 while trying to change parameters
1035 Description: The wrong (new) MDC name was used when setting parameters for
1036 upgraded MDT's. Also allows changing of OSC (and MDC)
1037 parameters if --writeconf is specified at tunefs upgrade time.
1040 Frequency : when setting specific ost indicies
1042 Description: QOS code breaks on skipped indicies
1043 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1044 created with --index need not be sequential.
1046 Severity : enhancement
1048 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1049 Details : The uninit_groups feature works in conjunction with the kernel
1050 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1051 up the pass1 processing of e2fsck. This is a read-only feature
1052 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1053 mount filesystems that have had this feature enabled.
1055 Severity : enhancement
1057 Description: Improve multi-block allocation algorithm to avoid fragmentation
1058 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1059 allocation locality and avoid filesystem fragmentation.
1061 ------------------------------------------------------------------------------
1063 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1065 * Support for kernels:
1066 2.4.21-47.0.1.EL (RHEL 3)
1067 2.6.5-7.283 (SLES 9)
1068 2.6.9-42.0.10.EL (RHEL 4)
1069 2.6.12.6 vanilla (kernel.org)
1070 2.6.16.27-0.9 (SLES 10)
1071 * Recommended e2fsprogs version: 1.39.cfs5
1073 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1077 Frequency : occasional, depends on client load and configuration
1078 Bugzilla : 12181, 12203
1079 Description: data loss for recently-modified files
1081 Details : In some cases it is possible that recently written or created
1082 files may not be written to disk in a timely manner (this should
1083 normally be within 30s unless client IO load is very high).
1084 The problem appears as zero-length files or files that are a
1085 multiple of 1MB in size after a client crash or client eviction
1086 that are missing data at the end of the file.
1088 This problem is more likely to be hit on clients where files are
1089 repeatedly created and unlinked in the same directory, clients
1090 have a large amount of RAM, have many CPUs, the filesystem has
1091 many OSTs, the clients are rebooted frequently, and/or the files
1092 are not accessed by other nodes after being written.
1094 The presence of the problem can be detected by looking at
1095 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1096 smaller than the second (nr_unused) then dirty files will not
1097 be flushed automatically to disk. "sync; sleep 10" should be
1098 run several times on the node before unmounting it to update
1099 Lustre (this is also safe to run on nodes without this problem).
1101 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1102 that can cause this same problem, so customers using that kernel
1103 also need to update the kernel in addition to Lustre. In order
1104 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1106 It is normal that files written just before a client crash (less
1107 than 30s) may not yet have been flushed to disk, even for local
1111 Frequency : frequent on thin XT3 nodes
1113 Description: UUID collision on thin XT3 Linux nodes
1114 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1115 randomly, since we relied on an insufficiently-seeded PRNG.
1120 Description: OSS hangs after "All ost request buffers busy"
1121 Details : A deadlock between quota and journal operations caused OSS
1122 hangs after printing "All ost request buffers busy."
1125 Frequency : always on liblustre builds
1127 Description: Cleanup compiler warnings on liblustre
1130 Frequency : always on liblustre builds on XT3
1132 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1136 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1137 > 32000 subdirectories
1138 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1139 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1140 subdirectory count crosses 32000. This will aid e2fsck to
1141 correctly handle more than 32000 subdirectories.
1144 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1146 Description: System hang on startup
1147 Details : This bug allowed the liblustre (e.g. catamount) client to
1148 return to the app before handling all startup RPCs. This
1149 could leave the node unresponsive to lustre network traffic
1150 and manifested as a server ptllnd timeout.
1152 Severity : enhancement
1154 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1155 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1156 boolean to control whether to print peer debug info when a
1157 client's RPC times out.
1162 Description: make O_SYNC working on 2.6 kernels
1163 Details : 2.6 kernels use different method for mark pages for write,
1164 so need add a code to lustre for O_SYNC work.
1169 Description: Failure to close file and release space on NFS
1170 Details : Put inode details into lock acquired in ll_intent_file_open.
1171 Use mdc_intent_lock in ll_intent_open to properly
1172 detect all kind of errors unhandled by mdc_enqueue.
1177 Description: proc file read during shutdown sometimes raced obd removal,
1179 Details : Add lock to prevent obd access after proc file removal.
1182 Frequency : Only for files larger than 4GB on 32-bit clients.
1184 Description: improperly doing page alignment of locks
1185 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1186 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1191 Description: RPCs being resent when they shouldn't be
1192 Details : Some RPCs that should not be resent are being resent. This
1193 can cause inconsistencies in the RPC state machine. Do not
1194 resend such requests.
1197 Frequency : rare, only with NFS export
1199 Description: Crash on NFS re-export node
1200 Details : under very unusual load conditions an assertion is hit in
1201 ll_intent_file_open()
1204 Frequency : only if OST filesystem is corrupted
1206 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1207 Details : for a short time RPCs with bulk IO are in the replay list,
1208 but replay of bulk IOs is unimplemented. If the OST filesystem
1209 is corrupted due to disk cache incoherency and then replay is
1210 started it is possible to trip an assertion. Avoid putting
1211 committed RPCs into the replay list at all to avoid this issue.
1216 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1217 Details : Large single O_DIRECT read and write calls can fail to allocate
1218 a sufficiently large buffer to process the request. In case of
1219 allocation failure the allocation is retried with a smaller
1220 buffer and broken into smaller requests.
1222 Severity : enhancement
1224 Description: Add -o localflock option to simulate old noflock behaviour.
1225 Details : This will achieve local-only flock/fcntl locks coherentness.
1230 Description: versioning check is incomplete
1231 Details : Checking the version difference of client vs. server, report
1232 error if the gap is too big.
1237 Description: add support PG_writeback bit
1238 Details : add support for PG_writeback bit for Lustre, for more carefull
1239 work with page cache in 2.6 kernel. This also fix some deadlocks
1240 and remove hack for work O_SYNC with 2.6 kernel.
1242 Severity : enhancement
1244 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1245 Details : The uninit_groups feature works in conjunction with the kernel
1246 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1247 up the pass1 processing of e2fsck. This is a read-only feature
1248 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1249 mount filesystems that have had this feature enabled.
1251 Severity : enhancement
1253 Description: Improve multi-block allocation algorithm to avoid fragmentation
1254 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1255 allocation locality and avoid filesystem fragmentation.
1257 ------------------------------------------------------------------------------
1259 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1261 * Support for kernels:
1262 2.6.9-42.0.3.EL (RHEL 4)
1263 2.6.5-7.276 (SLES 9)
1264 2.4.21-47.0.1.EL (RHEL 3)
1265 2.6.12.6 vanilla (kernel.org)
1266 2.6.16.21-0.8 (SLES10)
1267 * Recommended e2fsprogs version: 1.39.cfs2-0
1269 * The backwards-compatible /proc/sys/portals symlink has been removed
1270 in this release. Before upgrading, please ensure that you change
1271 any configuration scripts or /etc/sysctl.conf files that access
1272 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1273 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1274 in advance of the upgrade on any system running Lustre 1.4.6 or
1275 newer, since /proc/sys/lnet was added in that version.
1276 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1280 Frequency : only when quota is used
1282 Description: avoid scanning export list for quota master
1283 Details : Change the algorithms to avoid scanning export list in order
1284 to improve the efficiency.
1287 Frequency : MDS failover only, very rarely
1289 Description: "went back in time" messages on mds failover
1290 Details : The greatest transno may be lost when the current operation
1291 finishes with an error (transno==0) and the client's last_rcvd
1292 record is over-written. Save the greatest transno in the
1293 mds_last_transno for this case.
1296 Frequency : always for specific kernels and striping counts
1298 Description: client may get "Matching packet too big" without ACL support
1299 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1300 when trying to access files in certain configurations. The
1301 clients should in fact be denied when mounting because they do
1302 not understand ACLs.
1305 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1307 Description: many clients connecting with IO in progress causes connect timeouts
1308 Details : Avoid synchronous journal commits to avoid delays caused by many
1309 clients connecting/disconnecting when bulk IO is in progress.
1310 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1311 OST_IO_PORTAL to avoid delays behind potentially many pending
1315 Frequency : occasionally with multiple writers to a single file
1317 Description: shared writes to file may result in wrong size reported by stat()
1318 Details : Allow growing of kms when extent lock is cancelled
1321 Frequency : always with random mmap IO to multi-striped file
1323 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1324 Details : Only if the hole is at the end of OST object so that kms is too
1325 small. Fix is to increase kms accordingly in ll_nopage.
1328 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1330 Description: writes to a missing object would leak memory on the OST
1331 Details : If there is an inconsistency between the MDS and OST filesystems,
1332 such that the MDS references an object that doesn't exist, writes
1333 to that object will leak memory due to incorrect cleanup in the
1334 error handling path, eventually running out of memory on the OST.
1339 Description: Creating too long symlink causes lustre errors
1340 Details : Check symlink and name lengths before sending requests to MDS.
1343 Frequency : only if flock is enabled (not on by default)
1345 Description: posix locks not released on fd closure on 2.6.9+
1346 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1347 kernels, this caused such locks not to be released on fd close and
1348 then assertions on fs unmount about still used locks.
1351 Frequency : MDS failover only, very rarely
1353 Description: clients may get ASSERTION(granted_lock != NULL)
1354 Details : When request was taking a long time, and a client was resending
1355 a getattr by name lock request. The were multiple lock requests
1356 with the same client lock handle and
1357 mds_getattr_name->fixup_handle_for_resent_request found one of the
1358 lock handles but later failed with ASSERTION(granted_lock != NULL).
1363 Description: handle->h_buffer_credits > 0, assertion failure
1364 Details : h_buffer_credits is zero after truncate, causing assertion
1365 failure. This patch extends the transaction or creates a new
1369 Frequency : NFS re-export or patchless client
1370 Bugzilla : 11179, 10796
1371 Description: Crash on NFS re-export node (__d_move)
1372 Details : We do not want to hash the dentry if we don't have a lock.
1373 But if this dentry is later used in d_move, we'd hit uninitialised
1374 list head d_hash, so we just do this to init d_hash field but
1375 leave dentry unhashed.
1378 Frequency : NFS re-export or patchless client
1380 Description: NFS exports has problem with symbolic link
1381 Details : lustre client didn't properly install dentry when re-exported
1382 to NFS or running patchless client.
1385 Frequency : NFS re-export or patchless client
1387 Description: Various nfs/patchless fixes.
1388 Details : fixes reuse disconected alias for lookup process - this fixes
1389 warning "find_exported_dentry: npd != pd",
1390 fix permission error with open files at nfs.
1391 fix apply umask when do revalidate.
1394 Frequency : occasional
1396 Description: Crash on NFS re-export node
1397 Details : calling clear_page() on the wrong pointer triggered oops in
1398 generic_mapping_read().
1401 Frequency : rarely, using O_DIRECT IO
1403 Description: unaligned directio crashes client with LASSERT
1404 Details : check for unaligned buffers before trying any requests.
1407 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1409 Description: stale data returned from RAID cache
1410 Details : If only a small amount of IO is done to the RAID device before
1411 reading it again it is possible to get stale data from the RAID
1412 cache instead of reading it from disk.
1415 Frequency : always for sles10 kernel
1417 Description: sles10 support
1418 Details : ll_follow_link: compile fixes and using of nd_set_link
1419 under newer kernels.
1422 Frequency : depends on arch, kernel and compiler version, always on sles10
1425 Description: recursive or deep enough symlinks cause stack overflow
1426 Details : getting rid of large stack-allocated variable in
1430 Frequency : depends on hardware
1432 Description: lustre write performance loss in the SLES10 kernel
1433 Details : the performance loss is caused by using of write barriers in the
1434 ext3 code. The SLES10 kernel turns barrier support on by
1435 default. The fix is to undo that change for ldiskfs.
1437 ------------------------------------------------------------------------------
1439 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1441 * Support for kernels:
1442 2.6.9-42.0.3EL (RHEL 4)
1443 2.6.5-7.276 (SLES 9)
1444 2.4.21-47.0.1.EL (RHEL 3)
1445 2.6.12.6 vanilla (kernel.org)
1449 Frequency : quota enabled and large files being deleted
1451 Description: releasing more than 4GB of quota at once hangs OST
1452 Details : If a user deletes more than 4GB of files on a single OST it
1453 will cause the OST to spin in an infinite loop. Release
1454 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1459 Description: statfs data retrieved from /proc may be stale or zero
1460 Details : When reading per-device statfs data from /proc, in the
1461 {kbytes,files}_{total,free,avail} files, it may appear
1462 as zero or be out of date.
1465 Frequency : systems with MD RAID1 external journal devices
1467 Description: lconf's call to blkid is confused by RAID1 journal devices
1468 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1469 of returning all block devices that match the journal UUID.
1472 Frequency : always, for aggregate stripe size over 4GB
1474 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1475 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1476 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1479 Frequency : always if "lfs find" used on a local file/directory
1481 Description: "lfs find" segfaults if used on a local file/directory
1482 Details : The case where a directory component was not specified wasn't
1483 handled correctly. Handle this properly.
1486 Frequency : always on ppc64
1488 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1489 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1490 perform endianness conversion.
1493 Frequency : rarely (truncate to non-zero file size after write under load)
1494 Bugzilla : 10730, 10687
1495 Description: Files padded with zeros to next 4K multiple
1496 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1497 it is possible that files that are truncated to a non-zero size
1498 immediately after being written are filled with zero bytes beyond
1499 the truncated size. No file data is lost.
1501 Severity : enhancement
1503 Description: Allow recovery/failover for liblustre clients.
1504 Details : liblustre clients were unaware of failover configurations until
1507 Severity : enhancement
1509 Description: user file locks should fail when not mounting with flock option
1510 Details : Set up an error-returning stub in ll_file_operations.lock field
1511 to prevent incorrect behaviour when client is mounted without
1512 flock option. Also, set up properly f_op->flock field for
1516 Frequency : always on ia64
1518 Description: "lfs df" loops on printing out MDS statfs information
1519 Details : The obd_ioctl_data was not initialized and in some systems
1520 this caused a failure during the ioctl that did not return
1521 an error. Initialize the struct and return an error on failure.
1524 Frequency : SLES 9 only
1526 Description: Error of copying files with lustre special EAs as root
1527 Details : Client side always return success for setxattr call for lustre
1528 special xattr (currently only "trusted.lov").
1531 Frequency : rarely on clusters with both ia64+i386 clients
1533 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1534 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1535 clients are doing concurrent writes to the same file it is
1536 possible that the ia64 clients may LASSERT because the OST
1537 extent locks are not PAGE_SIZE aligned. Ensure that grown
1538 locks are always aligned on the request boundary.
1541 Frequency : specific use, occasional
1543 Description: Overwriting in use executable truncates on-disk binary image
1544 Details : If one node attempts to overwrite an executable in use by
1545 another node, we now correctly return ETXTBSY instead of
1546 truncating the file.
1548 Severity : enhancement
1550 Description: Async OSC create to avoid the blocking unnecessarily.
1551 Details : If a OST has no remain object, system will block on the creating
1552 when need to create a new object on this OST. Now, ways use
1553 pre-created objects when available, instead of blocking on an
1554 empty osc while others are not empty. If we must block, we block
1555 for the shortest possible period of time.
1560 Description: chmod on Lustre root is propagated to other clients
1561 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1562 invalid by the follow_mount time.
1565 Frequency : liblustre clients only
1567 Description: Race in 'instant cancel' lock handling could lead to such locks
1568 never to be granted in case of SMP MDS
1569 Details : Do not destroy not yet granted but cbpending locks in
1573 Frequency : replay/resend of open
1575 Description: non null lock assetion failure in mds_intent_policy
1576 Details : Trying to replay/resend lockless open requests resulted in
1577 mds_open() returning 0 with no lock. Now it sets a flag if
1578 a lock is going to be returned.
1580 Severity : enhancement
1582 Description: Checksum enhancements
1583 Details : New checksum enhancements allow for resending RPCs that failed
1586 Severity : enhancement
1588 Description: Tunables on number of dirty pages in cacche
1589 Details : Allow to set limit on number of dirty pages cached.
1594 Description: client crash on unmount - lock still has references
1595 Details : In some error handling cases it was possible to leak a lock
1596 reference on a client while accessing a file. This was not
1597 harmful to the client during operation, but would cause the
1598 client to crash when the filesystem is unmounted.
1601 Frequency : specific case, rare
1603 Description: ETXTBSY on mds though file not in use by client
1604 Details : ETXTBSY is no longer incorrectly returned when attempting to
1605 chmod or chown a directory that the user previously tried to
1606 execute or a currently-executing binary.
1609 Frequency : extremely rare except on liblustre-based clients
1611 Description: Lustre space not freed when files are deleted
1612 Details : Clean up open-unlinked files after client eviction. Previously
1613 the unlink was skipped and the files remained as orphans.
1618 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1619 Details : In some cases it was possible to send a blocking callback to
1620 a client doing a glimpse, even though that client didn't get
1621 a lock granted. When the glimpse lock is cancelled on the OST
1622 the freed lock is left on the waiting list and corrupted the list.
1625 Frequency : all core dumps
1627 Description: Broke core dumps to lustre
1628 Details : Negative dentry may be unhashed if parent does not have UPDATE
1629 lock, but some callers, e.g. do_coredump, expect dentry to be
1630 hashed after successful create, hash it in ll_create_it.
1632 ------------------------------------------------------------------------------
1634 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1636 * Support for kernels:
1637 2.6.9-42.0.2.EL (RHEL 4)
1638 2.6.5-7.276 (SLES 9)
1639 2.4.21-40.EL (RHEL 3)
1640 2.6.12.6 vanilla (kernel.org)
1644 Frequency : always on RHEL 3
1646 Description: Number of open files grows over time
1647 Details : The number of open files grows over time, whether or not
1648 Lustre is started. This was due to a filp leak introduced
1649 by one of our kernel patches.
1651 ------------------------------------------------------------------------------
1653 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1655 * Support for kernels:
1656 2.6.9-42.EL (RHEL 4)
1657 2.6.5-7.276 (SLES 9)
1658 2.4.21-40.EL (RHEL 3)
1659 2.6.12.6 vanilla (kernel.org)
1664 Bugzilla : 5719, 9635, 9792, 9684
1665 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1666 Details : If a server is under heavy load and cannot reply to new
1667 connection requests before the client resends the (re)connect,
1668 the connection handling code can behave badly if two service
1669 threads are concurrently handing separate (re)connections from
1670 the same client. Add better locking to the connection handling
1671 code, and ensure that only a single connection will be processed
1672 for a given client UUID, even if the lock is dropped.
1674 Severity : enhancement
1676 Description: add TCP zero-copy support to kernel
1677 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1678 sends if the hardware supports scatter-gather and checksumming.
1679 This allows socklnd to do client-write and server-read more
1680 efficiently and reduce CPU utilization from skbuf copying.
1683 Frequency : only if NFS exporting from client
1685 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1686 Details : If setting attributes on a file created under NFS that had
1687 never been opened it would be possible to oops the client
1688 if the file had no objects.
1691 Frequency : always for liblustre
1693 Description: liblustre client does MDS+OSTs setattr RPC for each write
1694 Details : When doing a write from a liblustre client, the client
1695 incorrectly issued an RPC to the MDS and each OST the file was
1696 striped over in order to update the timestamps. When writing
1697 with small chunks and many clients this could overwhelm the MDS
1698 with RPCs. In all cases it would slow down the write because
1699 these RPCs are unnecessary.
1701 Severity : enhancement
1703 Description: allow number of MDS service threads to be changed at module load
1704 Details : It is now possible to change the number of MDS service threads
1705 running. Adding "options mds mds_num_threads={N}" to the MDS's
1706 /etc/modprobe.conf will set the number of threads for the next
1707 time Lustre is restarted (assuming the "mds" module is also
1708 reloaded at that time). The default number of threads will
1709 stay the same, 32 for most systems.
1714 Description: OST crash if filesystem is unformatted or corrupt
1715 Details : If an OST is started on a device that has never been formatted
1716 or if the filesystem is corrupt and cannot even mount then the
1717 error handling cleanup routines would dereference a NULL pointer.
1722 Description: NULL pointer deref in llap_from_page.
1723 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1724 incorrect) mapping if the page was truncated/reclaimed while it was
1725 searched for. Check for this condition and skip such pages when
1726 doing readahead. Introduce extra check to llap_from_page() to
1727 verify page->mapping->host is non-NULL (so page is not anonymous).
1730 Frequency : Sometimes when using sys_sendfile
1732 Description: "page not covered by a lock" warnings from ll_readpage
1733 Details : sendfile called ll_readpage without right page locks present.
1734 Now we introduced ll_file_sendfile that does necessary locking
1735 around call to generic_file_sendfile() much like we do in
1739 Frequency : with certain MDS communication failures at client mount time
1741 Description: NULL pointer deref after failed client mount
1742 Details : a client connection request may delayed by the network layer
1743 and not be sent until after the PTLRPC layer has timed out the
1744 request. If the client fails the mount immediately it will try
1745 to clean up before the network times out the request. Add a
1746 reference from the request import to the obd device and delay
1747 the cleanup until the network drops the request.
1750 Frequency : occasionally during client (re)connect
1752 Description: assertion failure during client (re)connect
1753 Details : processing a client connection request may be delayed by the
1754 client or server longer than the client connect timeout. This
1755 causes the client to resend the connection request. If the
1756 original connection request is replied in this interval, the
1757 client may trip an assertion failure in ptlrpc_connect_interpret()
1758 which thought it would be the only running connect process.
1761 Frequency : only with obd_echo servers and clients that are rebooted
1763 Description: kernel BUG accessing uninitialized data structure
1764 Details : When running an obd_echo server it did not start the ping_evictor
1765 thread, and when a client was evicted an uninitialized data
1766 structure was accessed. Start the ping_evictor in the RPC
1767 service startup instead of the OBD startup.
1769 Severity : enhancement
1770 Bugzilla : 10193 (patchless)
1771 Description: Remove dependency on various unexported kernel interfaces.
1772 Details : No longer need reparent_to_init, exit_mm, exit_files,
1773 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1776 Frequency : rare (only users of deprecated and unsupported LDAP config)
1778 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1779 Details : LDAP apparently contains 'lustreName' attributes instead of
1780 'name'. A simple remapping of the name is sufficient.
1783 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1785 Description: waiting_locks_callback trips kernel BUG if client is evicted
1786 Details : Running with the dump_on_timeout debug flag turned on makes
1787 it possible that the waiting_locks_callback() can try to dump
1788 the Lustre kernel debug logs from an interrupt handler. Defer
1789 this log dumping to the expired_lock_main() thread.
1791 Severity : enhancement
1793 Description: Support NFS exporting on 2.6 kernels.
1794 Details : Implement non-rawops metadata methods for NFS server to use without
1795 changing NFS server code.
1798 Frequency : very rare (synthetic metadata workload only)
1800 Description: two racing renames might cause an MDS thread to deadlock
1801 Details : Running the "racer" program may cause one MDS thread to rename
1802 a file from being the source of a rename to being the target of
1803 a rename at exactly the same time that another thread is doing
1804 so, and the second thread has already enqueued these locks after
1805 doing a lookup of the target and is trying to relock them in
1806 order. Ensure that we don't try to re-lock the same resource.
1809 Frequency : only very large systems with liblustre clients
1811 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1812 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1813 all OSTs in parallel. This allows the eviction of stale liblustre
1814 clients to proceed much faster than if they were done in series,
1815 and also offers similar improvements for other set_info RPCs.
1820 Description: excessive CPU usage during initial read phase on client
1821 Details : During the initial read phase on a client, it would agressively
1822 retry readahead on the file, consuming too much CPU and impacting
1823 performance (since 1.4.5.8). Improve the readahead algorithm
1824 to avoid this, and also improve some other common cases (read
1825 of small files in particular, where "small" is files smaller than
1826 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1831 Description: MDS crash when receiving packet with unknown intent.
1832 Details : Do not LBUG in unknown intent case, just return -EFAULT
1834 Severity : enhancement
1835 Bugzilla : 9293, 9385
1836 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1837 Details : Do not serialize getattr (non-intent version) and statfs.
1840 Frequency : occasional, when OST network is overloaded/intermittent
1842 Description: client evicted by OST after bulk IO timeout
1843 Details : If a client sends a bulk IO request (read or write) the OST
1844 may evict the client if it is unresposive to its data GET/PUT
1845 request. This is incorrect if the network is overloaded (takes
1846 too long to transfer the RPC data) or dropped the OST GET/PUT
1847 request. There is no need to evict the client at all, since
1848 the pinger and/or lock callbacks will handle this, and the
1849 client can restart the bulk request.
1852 Frequency : Always when mmapping file with no objects
1854 Description: client crashes when mmapping file with no objects
1855 Details : Check that we actually have objects in a file before doing any
1856 operations on objects in ll_vm_open, ll_vm_close and
1862 Description: Request leak when working with deleted CWD
1863 Details : Introduce advanced request refcount tracking for requests
1864 referenced from lustre intent.
1866 Severity : Enhancement
1868 Description: Cache open file handles on client.
1869 Details : MDS now will return special lock along with openhandle, if
1870 requested and client is allowed to hold openhandle, even if unused,
1871 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1872 closing files for every read/write openration.
1874 Severity : Enhancement
1876 Description: Cache open negative dentries on client when possible.
1877 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1878 negative dentries on lock revocation.
1883 Description: Remounting a client read-only wasn't possible with a zconf mount
1884 Details : It wasn't possible to remount a client read-only with llmount.
1886 Severity : enhancement
1887 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1888 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1889 MPICH is included to add Lustre-specific ADIO interfaces.
1890 This is based closely on the UFS ADIO layer and only differs
1891 in file creation, in order to allow the OST striping to be set.
1892 This is user-contributed code and not supported by CFS.
1897 Description: extended inode attributes (immutable, append-only) work improperly
1898 when 2.4 and 2.6 kernels are used on client/server or vice versa
1899 Details : Introduce kernel-independent values for these flags.
1901 Severity : enhancement
1904 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1905 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1906 granularity. Now, Fractional MB granularity is be supported,
1907 this is very useful for low memory system.
1909 Severity : enhancement
1911 Description: Getattr by fid
1912 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1913 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1914 have locks and avoids getattr rpc requests for it on every stat.
1917 Frequency : Always, for filesystems larger than 2TB
1919 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1920 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1921 filesystems larger than 2TB to be mounted with mballoc enabled.
1924 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1926 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1927 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1928 filesystems larger than 2TB is mounted then the kernel will
1929 silently corrupt the start of the filesystem. CONFIG_LBD is
1930 enabled for all CFS-supported kernels, but the possibility of
1931 this happening with a modified kernel config exists.
1933 Severity : enhancement
1935 Description: add client O_DIRECT support for 2.6 kernels
1936 Details : It is now possible to do O_DIRECT reads and writes to files
1937 in the Lustre client mountpoint on 2.6 kernel clients.
1939 Severity : enhancement
1941 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1942 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1943 parallel, not waiting for response from every OST before sending
1944 a rpc to the next OST.
1949 Description: setattr vs write race when updating file timestamps
1950 Details : Client processes that update a file timestamp into the past
1951 right after writing to the file (e.g. tar) it is possible that
1952 the updated file modification time can be reset to the current
1953 time due to a race between processing the setattr and write RPC.
1955 Severity : enhancement
1957 Description: Bring 'lfs find' closer in line with regular Linux find.
1958 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1959 -print0 options and obtains all the needed info through the lustre
1962 Severity : enhancement
1964 Description: support up to 1024 configured devices on one node
1965 Details : change obd_dev array from statically allocated to dynamically
1966 allocated structs as they are first used to reduce memory usage
1971 Description: Flush dirty partially truncated pages during truncate
1972 Details : Immediatelly flush partially truncated pages in filter_setattr,
1973 this way we completely avoid having any pages in page cache on OST
1974 and can retire ugly workarounds during writes to flush such pages.
1979 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1980 Details : protect inode from truncation within vfs_unlink() context
1981 just take a reference before calling vfs_unlink() and release it
1982 when parent's i_sem is free.
1987 Description: last_id value checked outside lock on OST caused LASSERT failure
1988 Details : If there were multiple MDS->OST object precreate requests in
1989 flight, it was possible that the OST's last object id was checked
1990 outside a lock and incorrectly tripped an assertion. Move checks
1991 inside locks, and discard old precreate requests.
1994 Frequency : always, if extents are used on OSTs
1996 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1997 Details : OSTs using the extents format would not zero the high 16 bits of
1998 the index physical block number. This is not a problem for any
1999 OST filesystems smaller than 16TB, and no kernels support ext3
2000 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2001 new/modified files) and can be fixed for existing filesystems
2002 with e2fsprogs-1.39-cfs1.
2007 Description: import connection selection may be incorrect if timer wraps
2008 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2009 import connections to be ignored if the 32-bit jiffies counter
2010 wraps. Use a 64-bit jiffies counter.
2013 Frequency : very large clusters immediately after boot
2015 Description: LNET request buffers exhausted under heavy short-term load
2016 Details : If a large number of client requests are generated on a service
2017 that has previously never seen so many requests it is possible
2018 that the request buffer growth cannot keep up with the spike in
2019 demand. Instead of dropping incoming requests, they are held in
2020 the LND until the RPC service can accept more requests.
2023 Frequency : Sometimes during replay
2025 Description: Assertion failure in ll_local_open after replay.
2026 Details : If replay happened on an open request reply before we were able
2027 to set replay handler, reply will become not swabbed tripping the
2028 assertion in ll_local_open. Now we set the handler right after
2029 recognising of open request
2032 Frequency : very rare
2034 Description: kernel reports "badness in vsnprintf"
2035 Details : Reading from the "recovery_status" /proc file in small chunks
2036 may cause a negative length in lprocfs_obd_rd_recovery_status()
2037 call to vsnprintf() (which is otherwise harmless). Exit early
2038 if there is no more space in the output buffer.
2040 Severity : enhancement
2042 Description: clear OBD RPC statistics by writing to them
2043 Details : It is now possible to clear the OBD RPC statistics by writing
2044 to the "stats" file.
2049 Description: Client mtime is not the same on different clients after utimes
2050 Details : In some cases, the client was using the utimes() syscall on
2051 a file cached on another node. The clients now validate the
2052 ctime from the MDS + OSTs to determine which one is right.
2057 Description: Inability to activate failout mode
2058 Details : lconf script incorrectly assumed that in pythong string's numeric
2059 value is used in comparisons.
2062 Frequency : always with multiple stripes per file
2064 Description: Inefficient object allocation for mutli-stripe files
2065 Details : When selecting which OSTs to stripe files over, for files with
2066 a stripe count that divides evenly into the number of OSTs,
2067 the MDS is always picking the same starting OST for each file.
2068 Return the OST selection heuristic to the original design.
2073 Description: mount failures may take full timeout to return an error
2074 Details : Under some heavy load conditions it is possible that a
2075 failed mount can wait for the full obd_timeout interval,
2076 possibly several minutes, before reporting an error.
2077 Instead return an error as soon as the status is known.
2079 Frequency : quota enabled and large files being deleted
2081 Description: releasing more than 4GB of quota at once hangs OST
2082 Details : If a user deletes more than 4GB of files on a single OST it
2083 will cause the OST to spin in an infinite loop. Release
2084 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2089 Description: statfs data retrieved from /proc may be stale or zero
2090 Details : When reading per-device statfs data from /proc, in the
2091 {kbytes,files}_{total,free,avail} files, it may appear
2092 as zero or be out of date.
2095 Frequency : systems with MD RAID1 external journal devices
2097 Description: lconf's call to blkid is confused by RAID1 journal devices
2098 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2099 of returning all block devices that match the journal UUID.
2102 Frequency : always, for aggregate stripe size over 4GB
2104 Description: assertion fails when trying to use 4GB stripe size
2105 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2106 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2109 Frequency : always on ppc64
2111 Description: the first write on an ext3 filesystem with mballoc got stuck
2112 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2113 perform endianness conversion.
2115 ------------------------------------------------------------------------------
2117 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2119 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2120 INTEROPERATE with older versions automatically. Please read the
2121 user documentation before upgrading any part of a live system.
2122 * WARNING: Lustre networking configuration changes are required with
2123 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2126 * Support for kernels:
2127 2.6.9-22.0.2.EL (RHEL 4)
2128 2.6.5-7.244 (SLES 9)
2129 2.6.12.6 vanilla (kernel.org)
2132 Severity : enhancement
2133 Bugzilla : 7981/8208
2134 Description: Introduced Lustre Networking (LNET)
2135 Details : LNET is new networking infrastructure for Lustre, it includes
2136 a reorganized network configuration mode (see the user
2137 documentation for full details) as well as support for routing
2138 between different network fabrics. Lustre Networking Devices
2139 (LNDs) for the supported network fabrics have also been
2140 created for this new infrastructure.
2142 Severity : enhancement
2143 Description: Introduced Access control lists
2144 Details : clients can set ACLs on files and directories in order to have
2145 more fine-grained permissions than the standard Unix UGO+RWX.
2146 The MDS must be started with the "-o acl" mount option.
2148 Severity : enhancement
2149 Description: Introduced filesystem quotas
2150 Details : Administrators may now establish per-user quotas on the
2153 Severity : enhancement
2155 Description: Configuration change for the XT3
2156 The PTLLND is now used to run Lustre over Portals on the XT3
2157 The configure option(s) --with-cray-portals are no longer used.
2158 Rather --with-portals=<path-to-portals-includes> is used to
2159 enable building on the XT3. In addition to enable XT3 specific
2160 features the option --enable-cray-xt3 must be used.
2165 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2166 Details : A race between allocating a new llcd and re-getting the llcd_lock
2167 allowed another thread to grab newly-allocated llcd.
2169 Severity : enhancement
2171 Description: 2.6 OST async journal commit and locking fix to improve performance
2172 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2173 2.6 kernels are now async as they already were in 2.4 kernels so
2174 that they can commit concurrently with the network bulk transfer.
2175 For block-allocated files the filter allocation semaphore is held
2176 to avoid filesystem fragmentation during allocation. BKL lock
2177 removed for 2.6 xattr operations where it is no longer needed.
2182 Description: lconf incorrectly determined whether two IP networks could talk
2183 Details : In some more complicated routing and multiple-network
2184 configurations, lconf will avoid trying to make a network
2185 connection to a disjoint part of the IP space. It was doing the
2186 math incorrectly for one set of cases.
2191 Description: Fix for potential infinite loop processing records in an llog.
2192 Details : If an llog record is corrupted/zeroed, it is possible to loop
2193 forever in llog_process(). Validate the llog record length
2194 and skip the remainder of the block on error.
2197 Frequency : occasional (liblustre only)
2199 Description: liblustre could not open files whose last component is a symlink
2200 Details : sysio_path_walk() would incorrectly pass the open intent to
2201 intermediate path components.
2204 Frequency : rare (liblustre only with non-standard tuning)
2205 Bugzilla : 7201 (7350)
2206 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2207 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2208 after it has been released, only do this for referenced locks
2210 Severity : enhancement
2212 Description: specifying an (invalid) directory default stripe_size of -1
2213 would reset the directory default striping
2214 Details : stripe_size -1 was used internally to signal directory stripe
2215 removal, now use "all default" to signal dir stripe removal
2216 as a directory striping of "all default" is not useful
2219 Frequency : common for large clusters running liblustre clients
2221 Description: doing an ls when liblustre clients are running is slow
2222 Details : sending a glimpse AST to a liblustre client waits for every AST
2223 to time out, as liblustre clients will not respond. Since they
2224 cannot cache data we refresh the OST lock LVB from disk instead.
2226 Severity : enhancement
2228 Description: doing an ls at the same time as file IO can be slow
2229 Details : enqueue and other "small" requests can be blocked behind many
2230 large IO requests. Create a new OST IO portal for non-IO
2231 requests so they can be processed faster.
2234 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2236 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2237 created file with mode 000
2238 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2239 allow an ftruncate() as a non-root user to a file with mode 000.
2240 Lustre now respects this flag to disable mode checks when
2241 truncating a file owned by the user
2244 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2247 Description: Revoking locks from clients that went dead or catatonic might take
2249 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2250 cancellation of such locks instant on servers without waiting for
2251 any reply from clients. Clients drops these locks when cancel
2252 notification from server is received without replying.
2255 Frequency : liblustre-only, when liblustre client dies or becomes busy
2257 Description: Doing ls on Linux clients can take a long time with active
2259 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2260 granting such locks to it in the first place if possible. Locks
2261 are taken by proxy on the OST during the read or write and
2262 dropped immediately afterward. Add connect flags handling, do
2263 not grant locks to liblustre clients for glimpse ASTs.
2265 Severity : enhancement
2267 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2268 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2269 sequential reads and behaves badly with applications doing random
2270 reads. Improve it by reading ahead at least read region, and
2271 avoiding excessive large RPC for small reads.
2273 Severity : enhancement
2275 Description: Creating more than 1000 files for a single job may cause a load
2276 imbalance on the OSTs if there are also a large number of OSTs.
2277 Details : qos_prep_create() uses an OST index reseed value that is an
2278 even multiple of the number of available OSTs so that if the
2279 reseed happens in the middle of the object allocation it will
2280 still utilize the OSTs as uniformly as possible.
2285 Description: OST or MDS may oops in ping_evictor_main()
2286 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2287 but doesn't restart at beginning of obd_exports_timed list
2290 Severity : enhancement
2292 Description: improve by-nid export eviction on the MDS and OST
2293 Details : allow multiple exports with the same NID to be evicted at one
2294 time without re-searching the exports list.
2297 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2299 Description: MDS may oops in groups_free()
2300 Details : in rare race conditions a newly allocated group_info struct is
2301 freed again, and this can be NULL. The 2.4 compatibility code
2302 for groups_free() checked for a NULL pointer, but 2.6 did not.
2305 Frequency : common for liblustre clients doing little filesystem IO
2306 Bugzilla : 9352, 7313
2307 Description: server may evict liblustre clients accessing contended locks
2308 Details : if a client is granted a lock or receives a completion AST
2309 with a blocking AST already set it would not reply to the AST
2310 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2313 Frequency : lfs setstripe, only systems with more than 160 OSTs
2315 Description: unable to set striping with a starting offset beyond OST 160
2316 Details : llapi_create_file() incorrectly limited the starting stripe
2317 index to the maximum single-file stripe count.
2320 Frequency : LDAP users only
2322 Description: lconf did not handle in-kernel recovery with LDAP properly
2323 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2325 Severity : enhancement
2327 Description: bind OST threads to NUMA nodes to improve performance
2328 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2329 node and do their allocations there to localize memory access
2331 Severity : enhancement
2333 Description: llmount can determine client NID directly from Myrinet (GM)
2334 Details : the client NID code from gmnalnid was moved directly into
2335 llmount, removing the need to use this or specifying the
2336 client NID explicitly when mounting GM clients with zeroconf
2339 Frequency : if client is started with down MDS
2341 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2342 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2343 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2344 ptlrpc_import_delayed_req() because we didn't check if the
2345 request was interrupted, and we also didn't break out of the
2346 event loop if there was no timeout
2351 Description: data loss during non-page-aligned writes to a single file from
2352 both multiple nodes and multiple threads on one node at same time
2353 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2354 inconsistency led to false short-reads, that were cached and later
2355 used by ->prepare_write() to fill in partially written page,
2356 leading to data loss.
2359 Frequency : always, if lconf --abort_recovery used
2361 Description: lconf --abort_recovery fails with 'Operation not supported'
2362 Details : lconf was attempting to abort recovery on the MDT device and not
2365 Severity : enhancement
2367 Description: remove cleanup logs
2368 Details : replace lconf-generated cleanup logs with lustre internal
2369 cleanup routines. Eliminates the need for client-cleanup and
2372 Severity : enhancement
2374 Description: add support for EAs (user and system) on lustre filesystems
2375 Details : it is now possible to store extended attributes in the Lustre
2376 client filesystem, and with the user_xattr mount option it
2377 is possible to allow users to store EAs on their files also
2379 Severity : enhancement
2381 Description: Add possibility (config option) to show minimal available OST free
2383 Details : When compiled with --enable-mindf configure option, statfs(2)
2384 (and so, df) will return least minimal free space available from
2385 all OSTs as amount of free space on FS, instead of summary of
2386 free spaces of all OSTs.
2388 Severity : enhancement
2390 Description: do not expand extent locks acquired on OST-side
2391 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2392 by server: they are not cached anyway.
2395 Frequency : when mmap is used/binaries executed from Lustre
2397 Description: Unmmap pages before throwing them away from read cache.
2398 Details : llap_shrink cache now attempts to unmap pages before discarding
2399 them (if unmapping failed - do not discard). SLES9 kernel has
2400 extra checks that trigger if this unmapping is not done first.
2405 Description: lconf didn't resolve symlinks before checking to see whether a
2406 given mountpoint was already in use
2409 Frequency : when migrating failover services
2410 Bugzilla : 6395, 9514
2411 Description: When migrating a subset of services from a node (e.g. failback
2412 from a failover service node) the remaining services would
2413 time out and evict clients.
2414 Details : lconf --force (implied by --failover) sets the global obd_timeout
2415 to 5 seconds in order to quickly disconnect, but this caused
2416 other RPCs to time out too quickly. Do not change the global
2417 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2419 Severity : enhancement
2420 Frequency : if MDS is started with down OST
2421 Bugzilla : 9439,5706
2422 Description: Allow startup/shutdown of an MDS without depending on the
2423 availability of the OSTs.
2424 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2425 Add appropriate locking and lov-osc refcounts for safe
2426 cleaning. Add osc abort_inflight calls in case the
2427 synchronize never started.
2430 Frequency : occasional (Cray XT3 only)
2432 Description: root not authorized to access files in CRAY_PORTALS environment
2433 Details : The client process capabilities were not honoured on the MDS in
2434 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2435 been accessed by an authorized user then root was able to access
2436 the file on the local client also. The root user capabilities
2437 are now allowed on the MDS, as this environment has secure UID.
2440 Frequency : occasional
2442 Description: ldiskfs "too long searching" message happens too often
2443 Details : A debugging message (otherwise harmless) prints too often on
2444 the OST console. This has been reduced to only happen when
2445 there are fragmentation problems on the filesystem.
2450 Description: Division by zero in statfs when all OSCs are inactive
2451 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2452 lov_statfs divides by value returned by lov_get_stripecnt().
2456 Bugzilla : 9489, 3273
2457 Description: First write from each client to each OST was only 4kB in size,
2458 to initialize client writeback cache, which caused sub-optimal
2459 RPCs and poor layout on disk for the first writen file.
2460 Details : Clients now request an initial cache grant at (re)connect time
2461 and so that they can start streaming writes to the cache right
2462 away and always do full-sized RPCs if there is enough data.
2463 If the OST is rebooted the client also re-establishes its grant
2464 so that client cached writes will be honoured under the grant.
2469 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2470 Details : Now I/O RPCs go to different portal number and (presumably) fast
2471 lock requests (and glimses) and other RPCs get their own service
2472 threads pool that should be able to service those RPCs
2475 Severity : enhancement
2477 Description: Ability to exchange lustre version between client and servers and
2478 issue warnings at client side if client is too old. Also for
2479 liblustre clients there is ability to refuse connection of too old
2481 Details : New 'version' field is added to connect data structure that is
2482 filled with version info. That info is later checked by server and
2486 Frequency : rare, liblustre only.
2487 Bugzilla : 9296, 9581
2488 Description: Two simultaneous writes from liblustre at offset within same page
2489 might proceed at the same time overwriting eachother with stale
2491 Details : I/O lock withing llu_file_prwv was released too early, before data
2492 actually was hitting the wire. Extended lock-holding time until
2493 server acknowledges receiving data.
2496 Frequency : extremely rare. Never observed in practice.
2498 Description: avoid generating lustre_handle cookie of 0.
2499 Details : class_handle_hash() generates handle cookies by incrementing
2500 global counter, and can hit 0 occasionaly (this is unlikely, but
2501 not impossible, because initial value of cookie counter is
2502 selected randonly). Value of 0 is used as a sentinel meaning
2503 "unassigned handle" --- avoid it. Also coalesce two critical
2504 sections in this function into one.
2506 Severity : enhancement
2508 Description: allow liblustre clients to delegate truncate locking to OST
2509 Details : To avoid overhead of locking, liblustre client instructs OST to
2510 take extent lock in ost_punch() on client's behalf. New connection
2511 flag is added to handle backward compatibility.
2513 Severity : enhancement
2514 Bugzilla : 4928, 7341, 9758
2515 Description: allow number of OST service threads to be specified
2516 Details : a module parameter allows the number of OST service threads
2517 to be specified via "options ost ost_num_threads={N}" in the
2518 OSS's /etc/modules.conf or /etc/modprobe.conf.
2522 Bugzilla : 6146, 9635, 9895
2523 Description: servers crash with bad pointer in target_handle_connect()
2524 Details : In rare cases when a client is reconnecting it was possible that
2525 the connection request was the last reference for that export.
2526 We would temporarily drop the export reference and get a new
2527 one, but this may have been the last reference and the export
2528 was just destroyed. Get new reference before dropping old one.
2530 Severity : enhancement
2531 Frequency : if client is started with failover MDS
2533 Description: Allow multiple MDS hostnames in the mount command
2534 Details : Try to read the configuration from all specified MDS
2535 hostnames during a client mount in case the "primary"
2538 Severity : enhancement
2540 Description: Stop sending data to evicted clients as soon as possible.
2541 Details : Check if the client we are about to send or are sending data to
2542 was evicted already. (Check is done every second of waiting,
2543 for which l_wait_event interface was extended to allow checking
2544 of exit condition at specified intervals).
2547 Frequency : rare, normally only when NFS exporting is done from client
2549 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2551 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2554 Frequency : always, if srand() is called before liblustre initialization
2556 Description: Liblustre uses system PRNG disturbing its usage by user application
2557 Details : Introduce internal to lustre fast and high-quality PRNG for
2558 lustre usage and make liblustre and some other places in generic
2559 lustre code to use it.
2561 Severity : enhancement
2562 Bugzilla : 9477, 9557, 9870
2563 Description: Verify that the MDS configuration logs are updated when xml is
2564 Details : Check if the .xml configuration logs are newer than the config
2565 logs stored on the MDS and report an error if this is the case.
2566 Request --write-conf, or allow starting with --old_conf.
2568 Severity : enhancement
2570 Description: Handle symlinks in the path when checking if Lustre is mounted.
2571 Details : Resolve intermediate symlinks when checking if a client has
2572 mounted a filesystem to avoid duplicate client mounts.
2577 Description: lconf can hit an error exception but still return success.
2578 Details : The lconf command catches the Command error exception at the top
2579 level script context and will exit with the associated exit
2580 status, but doesn't ensure that this exit status is non-zero.
2585 Description: failure of ptlrpc thread startup can cause oops
2586 Details : Starting a ptlrpc service thread can fail if there are a large
2587 number of threads or the server memory is very fragmented.
2588 Handle this without oopsing.
2591 Frequency : always, only if liblustre and non-default acceptor port was used
2593 Description: liblustre cannot connect to servers with non-default acceptor port
2594 Details : tcpnal_set_default_params() was not called and was therefore
2595 ignoring the environment varaible TCPNAL_PORT, as well as other
2596 TCPNAL_ environment variables
2601 Description: two objects could be created on the same OST for a single file
2602 Details : If an OST is down, in some cases it was possible to create two
2603 objects on a single OST for a single file. No problems other
2604 than potential performance impact and spurious error messages.
2608 Bugzilla : 5681, 9562
2609 Description: Client may oops in ll_unhash_aliases
2610 Details : Client dcache may become inconsistent in race condition.
2611 In some cases "getcwd" can fail if the current directory is
2617 Description: Inode refcounting problems in NFS export code
2618 Details : link_raw functions used to call d_instantiate without obtaining
2619 extra inode reference first.
2623 Bugzilla : 9942, 9903
2624 Description: Referencing freed requests leading to crash, memleaks with NFS.
2625 Details : We used to require that call to ll_revalidate_it was always
2626 followed by ll_lookup_it. Also with revalidate_special() it is
2627 possible to call ll_revalidate_it() twice for the same dentry
2628 even if first occurence returned success. This fix changes semantic
2629 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2630 reference on a request referred from the intent.
2631 ll_intent_release() then releases such a request.
2634 Frequency : rare, normally benchmark loads only
2636 Description: unlinked inodes were kept in memory on the client
2637 Details : If a client is repeatedly creating and unlinking files it
2638 can accumulate a lot of stale inodes in the inode slab cache.
2639 If there is no other client load running this can cause the
2640 client node to run out of memory. Instead flush old inodes
2641 from client cache that have the same inode number as a new inode.
2644 Frequency : SLES9 2.6.5 kernel and long filenames only
2645 Bugzilla : 9969, 10379
2646 Description: utime reports stale NFS file handle
2647 Details : SLES9 uses out-of-dentry names in some cases, which confused
2648 the lustre dentry revalidation. Change it to always use the
2652 Frequency : rare, unless heavy write-truncate concurrency is continuous
2653 Bugzilla : 4180, 6984, 7171, 9963, 9331
2654 Description: OST becomes very slow and/or deadlocked during object unlink
2655 Details : filter_destroy() was holding onto the parent directory lock
2656 while truncating+unlinking objects. For very large objects this
2657 may block other threads for a long time and slow overall OST
2658 responsiveness. It may also be possible to get a lock ordering
2659 deadlock in this case, or run out of journal credits because of
2660 the combined truncate+unlink. Solution is to do object truncate
2661 first in one transaction without parent lock, and then do the
2662 final unlink in a new transaction with the parent lock. This
2663 reduces the lock hold time dramatically.
2666 Frequency : rare, 2.4 kernels only
2668 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2669 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2670 before unmounting their filesystem in order to avoid deadlock.
2671 The kernel_locked() function in 2.4 kernels only checks whether
2672 the kernel lock is held, not whether it is this process that is
2673 holding it as 2.6 kernels do.
2678 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2679 Details : The client ptlrpc code may be trying to reconnect to a down
2680 server before a previous connection attempt has timed out.
2681 Increase the reconnect interval to be longer than the connection
2682 timeout interval to avoid sending duplicate connections to
2686 Frequency : echo_client brw_test command
2688 Description: fix echo_client to work with OST preallocated code
2689 Details : OST preallocation code (5137) didn't take echo_client IO path
2690 into account: echo_client calls filter methods outside of any
2691 OST thread and, hence, there is no per-thread preallocated
2692 pages and buffers to use. Solution: hijack pga pages for IO. As
2693 a byproduct, this avoids unnecessary data copying.
2697 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2698 Description: Client can oops in mdc_commit_close() after open replay
2699 Details : It was possible for the MDS to return an open request with no
2700 transaction number in mds_finish_transno() if the client was
2701 evicted, but without actually returning an error. Clients
2702 would later try to replay that open and may trip an assertion
2703 Simplify the client close codepath, and always return an error
2704 from the MDS in case the open is not successful.
2707 Frequency : rare, 2.6 OSTs only
2709 Description: OST may deadlock under high load on fragmented files
2710 Details : If there was a heavy load and highly-fragmented OST filesystems
2711 it was possible to have all the OST threads deadlock waiting on
2712 allocation of biovecs, because the biovecs were not released
2713 until the entire RPC IO was completed. Instead, release biovecs
2714 as soon as they are complete to ensure forward IO progress.
2716 Severity : enhancement
2718 Description: Support for specifying external journal device at mount
2719 Details : If an OST or MDS device is formatted with an external journal
2720 device, this device major/minor is stored in the ext3 superblock
2721 and may not be valid for failover. Allow detecting and
2722 specifying the external journal at mount time.
2727 Description: Mounting an MDS with pending unlinked files may cause oops
2728 Details : target_finish_recovery() calls mds_postrecov() which returned
2729 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2730 considers this an error and immediately begins cleaning up the
2731 lov, just after starting the mds_lov process
2733 Severity : enhancement
2735 Description: Implement 'lfs df' to report actual free space on per-OST basis
2736 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2737 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2738 usage of objects; '-h' to report in human readable format.
2740 ------------------------------------------------------------------------------
2742 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2749 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2750 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2751 at startup during orphan unlinking. Instead, skip buddy update
2752 before setup, as it will be regenerated after recovery is complete.
2757 Description: If an OST is inactive, its locks might reference stale inodes.
2758 Details : lov_change_cbdata() must iterate over all namespaces, even if
2759 they are inactive to clear inode references from the lock.
2761 Severity : enhancement
2762 Frequency : occasional, if non-standard max_dirty_mb used
2764 Description: Client will block write RPCs if not enough grant
2765 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2766 then the client will block writes while waiting for another RPC
2767 to complete instead of consuming its dirty limit. With change
2768 we get improved performance when max_dirty_mb is small.
2770 Severity : enhancement
2771 Bugzilla : 3389, 6253
2772 Description: Add support for supplementary groups on the MDS.
2773 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2774 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2775 lookups for user supplementary groups into a cache.
2779 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2780 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2781 for liblustre O_EXCL is enforced by MDS.
2785 Description: readdir never returns NULL in liblustre.
2786 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2787 dentry in struct dirent.
2791 Frequency : liblustre only, depends on application IO pattern
2792 Description: liblustre clients evicted if not contacting servers
2793 Details : Don't put liblustre clients into the ping_evictor list, so
2794 they will not be evicted by the pinger ever.
2796 Severity : enhancement
2798 Description: Add ability to evict clients by NID from MDS.
2799 Details : By echoing "nid:$NID" string into
2800 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2801 $NID would be instantly evicted from this MDS and from all active
2802 OSTs connected to it.
2806 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2807 Details : lookup_it_finish() used to query file size from OSTs that was not
2812 Description: service threads change working directory to that of init
2813 Details : Starting lustre service threads may pin the working directory
2814 of the parent thread, making that filesystem busy. Threads
2815 now change to the working directory of init to avoid this.
2819 Frequency : during shutdown only
2820 Description: shutdown with a failed MDS or OST can cause unmount to hang
2821 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2826 Frequency : 2.6 only
2827 Description: chmod/chown may include an extra supplementary group
2828 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2829 supplementary group and if none is specified this is used.
2832 Bugzilla : 5479 (6816)
2834 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2835 Details : If lookup is in progress on a file that is unlinked we might try
2836 to revalidate the inode and fail in revalidate after lookup is
2837 complete and ll_file_open() enqueues the open again but
2838 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2841 Frequency : always, if lconf --abort_recovery used
2843 Description: lconf --abort_recovery fails with 'Operation not supported'
2844 Details : lconf was attempting to abort recovery on the MDT device and not
2847 ------------------------------------------------------------------------------
2849 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2854 Frequency : rare (only unsupported configurations with a node running as an
2856 Bugzilla : 6514, 5137
2857 Description: Mounting a Lustre file system on a node running as an OST could
2859 Details : OSTs now preallocates memory needed to write out data at
2860 startup, instead of when needed, to avoid having to
2861 allocate memory in possibly low memory situations.
2862 Specifically, if the file system is mounted on on OST,
2863 memory pressure could force it to try to write out data,
2864 which it needed to allocate memory to do. Due to the low
2865 memory, it would be unable to do so and the node would
2866 become unresponsive.
2868 Severity : enhancement
2870 Description: Addition of lconf --service command line option
2871 Details : lconf now accepts a '--service <arg>' option, which is
2872 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2874 Severity : enhancement
2876 Description: Failover mode is now the default for OSTs.
2877 Details : By default, OSTs will now run in failover mode. To return to
2878 the old behaviour, add '--failout' to the lmc line for OSTs.
2880 Severity : enhancement
2882 Description: Health checks are now provided for MDS and OSTs
2883 Details : Additional detailed health check information on MSD and OSTs
2884 is now provided through the procfs health_check value.
2887 Frequency : occasional, depends on IO load
2889 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2890 after numerous create/delete cycles
2891 Details : The ext3 inode allocation policy would not allocate new inodes
2892 very well on the OSTs because there are no new directories
2893 being created. Instead we look for groups with free space if
2894 the parent directories are nearly full.
2899 Description: Network or server problems during mount may cause partially
2900 mounted clients instead of returning an error.
2901 Details : The config llog parsing code may overwrite the error return
2902 code during mount error handling, returning success instead
2908 Description: MDS can fail to allocate large reply buffers
2909 Details : After long uptimes the MDS can fail to allocate large reply
2910 buffers (e.g. zconf client mount config records) due to memory
2911 fragmentation or consumption by the buffer cache. Preallocate
2912 some large reply buffers so that these replies can be sent even
2913 under memory pressure.
2917 Frequency : rare (liblustre)
2918 Description: fsx running with liblustre complained that using truncate() to
2919 extend the file doesn't work. This patch corrects that issue.
2920 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2921 ATTR_SIZE and lsm use in llu_setattr_raw.
2925 Frequency : rare, only 2.6 kernels
2926 Description: Unusual file access patterns on the MDS may result in inode
2927 data being lost in very rare circumstances.
2928 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2929 code in the 2.6 kernel caused the inode and/or EA data not to
2930 be read from disk, causing single-file corruption.
2934 Frequency : rare, only 2.6 filesystems using extents
2935 Description: Heavy concurrent write and delete load may cause data corruption.
2936 Details : It was possible under high-load situations to have an extent
2937 metadata block in the block device cache from a just-unlinked
2938 file overwrite a newly-allocated data block. We now unmap any
2939 metadata buffers that alias just-allocated data blocks.
2943 Frequency : filesystems with default stripe_count larger than 77
2944 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2945 Details : lconf specifies an inode size of 4096 bytes when the default
2946 stripe_count is larger than 77. This conflicts with the default
2947 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2948 this case to avoid pinning too much memory for large EAs.
2950 ------------------------------------------------------------------------------
2952 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2957 Frequency : rare (extremely heavy IO load with hundreds of clients)
2959 Description: Client is evicted, gets IO error writing to file
2960 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2961 caused clients to be evicted for AST timeouts. The fixes in
2962 bug 5192 mean we no longer need to have such short AST timeouts
2963 so ldlm_timeout has been increased.
2966 Frequency : occasional during --force or --failover shutdown under load
2967 Bugzilla : 5949, 4834
2968 Description: Server oops/LBUG if stopped with --force or --failover under load
2969 Details : a collection of import/export refcount and cleanup ordering
2970 issues fixed for safer force cleanup
2973 Frequency : only filesystems larger than 120 OSTs
2974 Bugzilla : 5990, 6223
2975 Description: lfs getstripe would oops on a very large filesystem
2976 Details : lov_getconfig used kfree on vmalloc'd memory
2979 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2980 Bugzilla : 6242, 6243
2981 Description: reading from files that had been truncated to a non-zero size
2982 but never opened returned no data
2983 Details : ll_file_read() reads zeros from no-object files to EOF
2988 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2989 Details : There's little harm in aborting MDS/OSS recovery and letting it
2990 try again, so I removed the LASSERT and return an error instead.
2992 Severity : enhancement
2994 Description: New debugging infrastructure for tracking down data corruption
2995 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2996 (b) cover more of the client-side code path, and (c) try to narrow
2997 down where problems occurred
3001 Bugzilla : 3819, 4364, 4397, 6313
3002 Description: Racing close and eviction MDS could cause assertion in mds_close
3003 Details : It was possible to get multiple mfd references during close and
3004 client eviction, leading to one thread referencing a freed mfd.
3006 Severity: : enhancement
3007 Bugzilla : 3262, 6359
3008 Description: Attempts to reconnect to servers are now more aggressive.
3009 Details : This builds on the enhanced upcall-less recovery that was added
3010 in 1.4.2. When trying to reconnect to servers, clients will
3011 now try each server in the failover group every 10 seconds. By
3012 default, clients would previously try one server every 25 seconds.
3017 Description: After recovery, certain operations trigger a failed
3018 assertion on a client.
3019 Details : Failing over an mds, using lconf -d --failover, while a
3020 client was doing a readdir() call would cause the client to
3021 LBUG after recovery completed and the readdir() was resent.
3023 Severity : enhancement
3025 Description: Default groups are now added by lconf
3026 Details : You can now run lconf --group <servicename> without having to
3027 manually add groups with lmc.
3030 Frequency : occasional
3032 Description: Nodes with an elan id of 0 trigger a failed assertion
3035 Frequency : always when accessing e.g. tty/console device nodes
3037 Description: tty and some other devices nodes cannot be used on lustre
3038 Details : file's private_data field is used by device data and lustre
3039 values in there got lost. New field was added to struct file to
3040 store fs-specific private data.
3043 Frequency : when exporting Lustre via NFS
3045 Description: NFSD failed occasionally when looking up a path component
3046 Details : NFSD is looking up ".." which was broken in ext3 directories
3047 that had grown large enough to become hashed.
3050 Frequency : Clusters with multiple interfaces not on the same subnet
3052 Description: Nodes will repeatedly try to reconnect to an interface which it
3053 cannot reach and report an error to the log.
3054 Details : Extra peer list entries will be created by lconf with some peers
3055 unreachable. lconf now validates the peer before adding it.
3058 Frequency : Only if a default stripe is set on the filesystem root.
3060 Description: Setting a default stripe on the filesystem root prevented the
3061 filesystem from being remounted.
3062 Details : The client was sending extra request flags in the root getattr
3063 request and did not allocate a reply buffer for the dir EA.
3066 Frequency : occasional, higher if lots of files are accessed by one client
3067 Bugzilla : 6159, 6097
3068 Description: Client trips assertion regarding lsm mismatch/magic
3069 Details : While revalidating inodes the VFS looks up inodes with ifind()
3070 and in rare cases can find an inode that is being freed.
3071 The ll_test_inode() code will free the lsm during ifind()
3072 when it finds an existing inode and then the VFS later attaches
3073 this free lsm to a new inode.
3077 Bugzilla : 6422, 7030
3078 Description: MDS deadlock between mkdir and client eviction
3079 Details : Creating a new file via mkdir or mknod (starting a transaction
3080 and getting the ns lock) can deadlock with client eviction
3081 (gets ns lock and trying to finish a synchronous transaction).
3084 Frequency : occasional
3085 Description: While starting a server, the fsfilt_ext3 module could not be
3087 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3088 kernels. Previously, lconf would still use the ext3 name
3089 when trying to load modules. Now, it will correctly use
3090 ext3 on 2.4 and ldiskfs on 2.6.
3092 Severity : enhancement
3093 Description: The default stripe count has been changed to 1
3094 Details : The interpretation of the default stripe count (0, to lfs
3095 or lmc) has been changed to mean striping across a single
3096 OST, rather than all available. For general usage we have
3097 found a stripe count of 1 or 2 works best.
3099 Severity : enhancement
3100 Description: Add support for compiling against Cray portals.
3101 Details : Conditional compiling for some areas that are different
3105 Frequency : occasional
3106 Bugzilla : 6409, 6834
3107 Description: Creating files with an explicit stripe count may lead to
3108 a failed assertion on the MDS
3109 Details : If some OSTs are full or unavailable, creating files may
3110 trigger a failed assertion on the MDS. Now, Lustre will
3111 try to use other servers or return an error to the
3115 Frequency : occasional
3117 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3118 caused lock timeout during readahead (since 1.4.2).
3119 Details : Processes doing readahead might match a lock that hasn't been
3120 granted yet if there are overlapping and conflicting lock
3121 requests. The readahead process waits on ungranted lock
3122 (original lock is CBPENDING), while OST waits for that process
3123 to cancel CBPENDING read lock and eventually evicts client.
3125 Severity : enhancement
3127 Description: Initial enabling of flock support for clients
3128 Details : Implements fcntl advisory locking and file status functions.
3129 This feature is provided as an optional mount flag (default
3130 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3131 locking are implemented yet, and those that are are not guaranteed
3132 to be completely correct in production environments.
3133 mount -t lustre -o [flock|noflock] ...
3136 Frequency : occasional
3138 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3139 assertion in the ext3 JBD (journaling) layer.
3140 Details : The b_committed_data struct is protected by the big kernel lock
3141 in 2.4 kernels, serializing journal_commit_transaction() and
3142 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3143 access to this struct. In 2.6 kernels there is finer grained
3144 locking to improve SMP performance of the JBD layer.
3148 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3153 Description: merge and cleanup kernel patches.
3156 -----------------------------------------------------------------------------
3158 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3160 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3161 versions of Lustre. Please update all servers and clients to
3162 version 1.4.2 or later at the same time. You must also run
3163 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3164 to update the configuration logs.
3166 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3167 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3168 - swab llog records of type '0' so we get proper header size/idx (5861)
3169 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3170 - fix rename of one directory over another leaking an inode (5953)
3171 - avoid SetPageDirty on 2.6 (5981)
3172 - don't re-add just-being-destroyed locks to the waiting list (5653)
3173 - when creating new directories, inherit the parent's custom
3174 striping settings if present parent (3048)
3175 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3176 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3177 - don't hold client locks on temporary worklist from l_lru (5666)
3178 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3179 - automatically evict dead clients (5921)
3180 - Update file size properly in create+truncate+fstat case (6196)
3181 - Do not unhash mountpoint dentries, do not allow removal of
3183 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3184 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3185 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3186 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3187 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3188 - Correctly handle reads of files with no objects (6243)
3189 - lctl recover will also mark a device active if deactivate used (5933)
3191 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3192 - allow --write-conf on an MDS with different nettype than client (5619)
3193 - don't write config llogs to MDS for mounts not from that MDS (5617)
3194 - lconf should create multiple TCP connections from a client (5201)
3195 - init scripts are now turned off by default; run chkconfig --on
3196 lustre and chkconfig --on lustrefs to use them
3197 - upcalls are no longer needed for clients to recover to failover
3199 - add --abort-recovery option to lconf to abort recovery on device
3201 - add support for an arbitrary number of OSTs (3026)
3202 - Quota support protocol changes.
3203 - forward compatibility changes to wire structs (6007)
3204 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3205 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3206 - improved "lustre status" script
3207 - initialize blocksize for non-regular files (6062)
3208 - added --disable-server and --disable-client configure options (5782)
3209 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3210 - Vanilla 2.4.29 support
3211 - increase maximum number of obd devices to 520 (6242)
3212 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3213 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3214 SLES 9 2.6.5 kernels
3215 - limit stripes per file to 160 (the maximum EA size) (6093)
3217 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3220 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3221 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3222 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3223 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3224 - small ext3 extents cleanups and fixes (5733)
3225 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3226 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3227 - avoid needless client->OST connect, fix handle mismatch (5317)
3228 - fix DLM error path that led to out-of-sync client, long delays (5779)
3229 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3230 - fix several locking issues related to i_size (5492,5624,5654,5672)
3231 - don't move pending lock onto export if it is already evicted (5683)
3232 - fix kernel oops when creating .foo in unlinked directory (5548)
3233 - fix deadlock in obdfilter statistics vs. object create (5811)
3234 - use time_{before,after} to avoid timer jiffies wrap (5882)
3235 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3236 - Do not leak request if server was not able to process it (5154)
3237 - If mds_open unable to find parent dir, make that negative lookup(5154)
3238 - don't create new directories with extent-mapping (5909, 5936)
3240 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3241 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3242 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3243 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3244 - support for RHEL4 kernel on i686 (5773)
3245 - provide error messages when incompatible logs are encountered (5898)
3247 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3248 * version 1.4.0.10 (1.4.1 release candidate 1)
3250 - don't keep a lock reference when lock is not granted (4238)
3251 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3252 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3253 - drop import inflight refcount on signal_completed_replay error (5255)
3254 - unlock page after async write error during send (3677)
3255 - handle missing objects in filter_preprw_read properly (5265)
3256 - no transno return for symlink open, don't save no-trasno open (3440)
3257 - don't try to complete elan receive that already failed (4012)
3258 - free RPC server reply state on error (5406)
3259 - clean up thread from ptlrpc_start_thread() on error (5160)
3260 - readahead could read extra page into cache that wasn't ejected (5388)
3261 - prevent races in class_attach/setup/cleanup/detach (5260)
3262 - don't dereference de->d_inode after l_dput of de (5458)
3263 - use "int" for stripe value returned from lock_to_stripe (5544)
3264 - mballoc allocation and error-checking fixes in 2.6 (5504)
3265 - block device patches to fix I/O request sizes in 2.6 (5482)
3266 - look up hostnames for IB nals (5602)
3267 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3268 - don't start multiple acceptors for the same port (5277)
3269 - fix incorrect LASSERT in mds_getattr_name (5635)
3270 - export a proc file for general "ping" checking (5628)
3271 - fix "lfs check" to not block when the MDS is down (5628)
3273 - service request history (4965)
3274 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3275 - create an "evict_client" /proc entry on OSTs, like the MDS has
3276 - fix mount usage message, return errors per mount(8) (5168)
3277 - change grep [] to grep "[]" in tests so they work in more UMLs
3278 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3279 - remove ancient LOV_MAGIC_V0 EA support (5047)
3280 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3281 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3282 - allow readahead allocations to fail when low on memory (5383)
3283 - mmap locking landed again, after considerable improvement (2828)
3284 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3286 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3289 - send OST transaction number in read/write reply to free req (4966)
3290 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3291 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3292 - the watchdog thread now runs as interruptible (5246)
3293 - flock/lockf fixes (but it's still disabled, pending 5135)
3294 - don't use EXT3 constants in llite code (5094)
3295 - memory shortage at startup could cause assertion (5176)
3297 - reorganization of lov code
3298 - single portals codebase
3300 - add extents/mballoc support (5025)
3301 - direct I/O reads in the obdfilter (4048)
3302 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3304 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3307 - send OST transaction number in read/write reply to free req (4966)
3308 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3309 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3310 - the watchdog thread now runs as interruptible (5246)
3311 - handle missing objects in filter_preprw_read properly (5265)
3312 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3313 - drop import inflight refcount on signal_completed_replay error (5255)
3314 - unlock page after async write error during send (3677)
3315 - return original error code on reconstructed replies (3761)
3316 - no transno return for symlink open, don't save no-trasno open (3440)
3318 - add pid to ldlm debugging output (4922)
3319 - bump the watchdog timeouts -- we can't handle 30sec yet
3320 - extra debugging for orphan dentry/inode bug (5259)
3322 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3325 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3326 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3327 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3328 - fix NULL dereference in /proc/sys/portals/routes (4827)
3329 - allow failed mdc_close() operations to be interrupted (4561)
3330 - stop precreate on OST before MDS would time out on it (4778)
3331 - don't send partial-page writes before EOF from client (4410)
3332 - discard client grant for sub-page writes on large-page clients (4520)
3333 - don't free dentries not owned by NFS code, check generation (4806)
3334 - fix lsm leak if mds_create_objects() fails (4801)
3335 - limit debug_daemon file size, always print CERROR messages (4789)
3336 - use transno after validating reply (3892)
3337 - process timed out requests if import state changes (3754)
3338 - update mtime on OST during writes, return in glimpse (4829)
3339 - add mkfsoptions to LDAP (4679)
3340 - use ->max_readahead method instead of zapping global ra (5039)
3341 - don't interrupt __l_wait_event() during strace
3343 - add software watchdogs to catch hung threads quickly (4941)
3344 - make lustrefs init script start after nfs is mounted
3345 - fix CWARN/ERROR duplication (4930)
3346 - return async write errors to application if possible (2248)
3347 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3348 - print NAL number in %x format (4645)
3349 - update barely-supported suse-2.4.21-171 series (4842)
3350 - support for sles 9 %post scripts
3351 - support for building 2.6 kernel-source packages
3352 - support for sles km_* packages
3354 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3357 - ignore -ENOENT errors in osc_destroy (3639)
3358 - notify osc create thread that OSC is being cleaned up (4600)
3359 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3360 - reconstruct ost_handle() like mds_handle() (4657)
3361 - create a new thread to do import eviction to avoid deadlock (3969)
3362 - let lconf resolve symlinked-to devices (4629)
3363 - don't unlink "objects" from directory with default EA (4554)
3364 - hold socknal file ref over connect in case target is down (4394)
3365 - allow more than 32000 subdirectories in a single directory (3244)
3366 - fix blocks count for O_DIRECT writes (3751)
3367 - OST returns ENOSPC from object create when no space left (4539)
3368 - don't send truncate RPC if file size isn't changing (4410)
3369 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3370 - bind to privileged port in socknal and tcpnal (3689)
3372 - rate limit CERROR/CWARN console message to avoid overload (4519)
3373 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3374 - basic mmap support (3918)
3375 - kernel patch series update from b1_4 (4711)
3377 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3380 - avoid crash during MDS cleanup with OST shut down (2775)
3381 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3382 - don't use bad inodes on the MDS (3744)
3383 - dynamic object preallocation to improve recovery speed (4236)
3384 - don't hold spinlock over lock dumping or change debug flags (4401)
3385 - don't zero obd_dev when it is force cleaned (3651)
3386 - print grants to console if they go negative (4431)
3387 - "lctl deactivate" will stop automatic recovery attempts (3406)
3388 - look for existing locks in ldlm_handle_enqueue() (3764)
3389 - don't resolve lock handle twice in recovery avoiding race (4401)
3390 - revalidate should check working dir is a directory (4134)
3392 - don't always mark "slow" obdfilter messages as errors (4418)
3394 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3397 - don't close LustreDB during write_conf until it is done (3860)
3398 - fix typo in lconf for_each_profile (3821)
3399 - allow dumping logs from multiple threads at one time (3820)
3400 - don't allow multiple threads in OSC recovery (3812)
3401 - fix debug_size parameters (3864)
3402 - fix mds_postrecov to initialize import for llog ctxt (3121)
3403 - replace config semaphore with spinlock (3306)
3404 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3405 - don't allow enqueue to complete on a destroyed export (3822)
3406 - down write_lock before checking llog header bitmap (3825)
3407 - recover from lock replay timeout (3764)
3408 - up llog sem before sending rpc (3652)
3409 - reduce ns lock hold times when setting kms (3267)
3410 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3411 - fix NULL deref and obd_dev leak on setup error (3312)
3412 - replace some LBUG about llog ops with error handling (3841)
3413 - don't match INVALID dentries from d_lookup and spin (3784)
3414 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3415 - fix invalid assertion in ptlrpc_set_wait (3880)
3417 - add libwrap support for the TCP acceptor (3996)
3418 - add /proc/sys/portals/routes for non-root route listing (3994)
3419 - allow setting MDS UUID in .xml (2580)
3420 - print the stack of a process that LBUGs (4228)
3422 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3425 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3426 - make sure to unset replay flag from failed open requests (3440)
3427 - if default stripe count is 0, use OST count for inode size (3636)
3428 - update parent mtime/ctime on client for create/unlink (2611)
3429 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3430 - free recovery state on server during a forced cleanup (3571)
3431 - unregister_reply for resent reqs (3063)
3432 - loop back devices mounting and status check on 2.6 (3563)
3433 - fix resource-creation race that can provoke i_size == 0 (3513)
3434 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3435 - more debugging for page-accounting assertion (3746)
3436 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3437 - avoid deadlock after precreation failure (3758)
3438 - fix race and lock order deadlock in orphan handling (3450, 3750)
3439 - add validity checks when grabbing inodes from l_ast_data (3599)
3441 - add /proc/.../recovery_status to obdfilter (3428)
3442 - lightweight CDEBUG infrastructure, debug daemon (3668)
3443 - change default OSC RPC parameters to be better on small clusters
3444 - turn off OST read cache for files smaller than 32MB
3445 - install man pages and include them in rpms (3100)
3446 - add new init script for (un)mounting lustre filesystems (2593)
3447 - run chkconfig in %post for init scripts (3701)
3448 - drop scimac NAL (unmaintained)
3450 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3453 - clean kiobufs before and after use (3485)
3454 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3455 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3456 - change init script to fail more clearly if not run as root (1528)
3457 - allow clients to reconnect during replay (1742)
3458 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3459 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3460 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3461 - fix cond_resched (3554)
3462 - only evict unfinished clients after recovery (3515)
3463 - allow bulk resend, prevent data loss (3570)
3464 - dynamic ptlrpc request buffer allocation (2102)
3465 - don't allow unlinking open directory if it isn't empty (2904)
3466 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3467 - remove extraneous obd dereference causing LASSERT failure (3334)
3468 - don't use get_cycles() when creating temp. files on the mds (3156)
3469 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3470 - handle EEXIST for set-stripe, set proper directory name (3336)
3472 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3473 - fix ksocknal_fmb_callback() error messages (2918)
3475 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3478 - don't copy lvb into (possibly NULL) reply on error (2983)
3479 - don't deref dentry after dput, don't free lvb on error (2922)
3480 - use the kms to determine writeback rpc length (2947)
3481 - increment oti_logcookies when osc is inactive (2948)
3482 - update client's i_blocks count via lvb messages (2543)
3483 - handle intent open/close of special files properly (1557)
3484 - mount MDS with errors=remount-ro, like obdfilter (2009)
3485 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3486 - don't use cancelling-locks' kms values (2947)
3487 - use highest lock extent for kms, not last one (2925)
3488 - don't dereference ERR_PTR() dentry in error handling path (3107)
3489 - fix thread race in portals_debug_dumplog() (3122)
3490 - create lprocfs device entries at setup instead of at attach (1519)
3491 - common AST error handler, don't evict client on completion race (3145)
3492 - zero nameidata in detach_mnt in 2.6 (3118)
3493 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3494 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3495 - initialize RPC timeout timer earlier for 2.6 (3219)
3496 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3497 - print nal/nid for unknown nid (3258)
3498 - additional checks for oscc recovery before doing precreate (3284)
3499 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3500 - don't crash in mdc_close for bad permissions on open (3285)
3501 - zero i_rdev for non-device files (3147)
3502 - clear page->private before handing to FS, better assertion (3119)
3503 - tune the read pipeline (3236)
3504 - fix incorrect decref of invalidated dentry (2350)
3505 - provide read-ahead stats and refine rpc in flight stats (3328)
3506 - don't hold journal transaction open across create RPC (3313)
3507 - update atime on MDS at close time (3265)
3508 - close LDAP connection when recovering to avoid server load (3315)
3509 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3510 - don't leak open file on MDS after open resend (3325)
3511 - serialize filter_precreate and filter_destroy_precreated (3329)
3512 - loop device shouldn't call sync_dev() for nul device (3092)
3513 - clear page cache after eviction (2766)
3514 - resynchronize MDS->OST in background (2824)
3515 - refuse to mount the same filesystem twice on same mountpoint (3394)
3516 - allow llmount to create routes for mounting behind routers (3320)
3517 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3518 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3519 - fix rare problem with rename on htree directories (3417)
3521 - allow default OST striping configuration per directory (1414)
3522 - fix compilation for qswnal for 2.6 kernels (3125)
3523 - increase maximum number of MDS request buffers for large systems
3524 - change liblustreapi to be useful for external progs like lfsck (3098)
3525 - increase local configuration timeout for slow disks (3353)
3526 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3528 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3531 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3532 - don't overwrite extent policy data in reply if lock was blocked (2901)
3533 - drop filter export grants atomically with removal from device (2663)
3534 - del obd_self_export from work_list in class_disconnect_exports (2908)
3535 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3536 - swab reply message in mdc_close, other PPC fixes (2464)
3537 - fix destroying of named logs (2325)
3538 - overwrite old logs when running lconf --write_conf (2264)
3539 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3540 - fix race in target_handle_connect (2898)
3541 - mds_reint_create() should take same inode create lock (2926)
3542 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3543 - don't close files for self_export to avoid uninitialized obd (2936)
3544 - allow MDS with the same name as client node (2939)
3545 - hold dentry reference for closed log files for unlink (2325)
3546 - reserve space for all logs during transactions (2059)
3547 - don't evict page beyond end of stripe extent (2925)
3548 - don't oops on a deleted current working directory (2399)
3549 - handle hard links to targets without a parent properly (2517)
3550 - don't dereference NULL lock when racing during eviction (2867)
3551 - don't grow lock extents when lots of conflicting locks (2919)
3553 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3556 - account for cache space usage on clients to avoid data loss (974)
3557 - lfsck support in lustre kernel code (2349)
3558 - reduce journal credits needed for BRW writes (2370)
3559 - orphan handling to avoid losing space on client/server crashes
3560 - ptlrpcd can be blocked, stopping ALL progress (2477)
3561 - use lock value blocks to assist in proper KMS, faster stat (1021)
3562 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3563 - recovery for initial connections (2355)
3564 - fixes for mds_cleanup_orphans (1934)
3565 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3566 - block all file creations until orphan recovery completes (1901)
3567 - client remove rq_connection from request struct (2423)
3568 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3569 - recovery timer race (2670)
3570 - mdc_close recovey bug (2532)
3571 - ptlrpc cleanup bug (2710)
3572 - mds timeout on local locks (2588)
3573 - namespace lock held during RPCs (2431)
3574 - handle interrupted sync write properly (2503)
3575 - don't try to handle a message that hasn't been replied to (2699)
3576 - client assert failure during cleanup after abort recovery (2701)
3577 - leak mdc device after failed mount (2712)
3578 - ptlrpc_check_set allows timedout requests to complete (2714)
3579 - wait for inflight reqs when ptlrpcd finishes (2710)
3580 - make sure unregistered services are removed from the srv_list
3581 - reset bulk XID's when resending them (caught by 1138 test)
3582 - unregister_bulk after timeout
3583 - fix lconf error (2694)
3584 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3585 - readahead locks pages, leaves pending causing memory pressure (2673)
3586 - increase OST request buffers to 4096 on large machines (2729)
3587 - fix up permission of existing directories in simple_mkdir (2661)
3588 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3589 - don't assign transno to errored transactions (2742)
3590 - don't delete objects on OST if given a bogus objid from MDS (2751)
3591 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3592 - if rq_no_resend, then timeout request after recovery (2432)
3593 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3594 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3595 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3596 - MDS recovery shouldn't delete objects that it has given out (2730)
3597 - if enqueue arrives after completion, don't clobber LVB (2819)
3598 - don't unlock pages twice when trigger_group_io returns error (2814)
3599 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3600 - don't write pages to disk if there was an error (1450)
3601 - don't ping imports that have recovery disabled (2676)
3602 - take buffered bytes into account when balancing socknal conn (2817)
3603 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3604 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3605 - remove little-endian swabbing of llog records (1987)
3606 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3607 - retry reposting request buffers if they fail (1191)
3608 - grow extent at grant time to avoid granting a revoked lock (2809)
3609 - lock revoke doesn't evict page if covered by a second lock (2765)
3610 - disable VM readahead to avoid reading outside lock extents (2805)
3612 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3613 - updated kernel patches for hp-2.4.20 kernel (2681)
3615 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3618 - fix truncated write corruption (2366)
3619 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3621 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3622 - ensure len(uuid) < 37 in lmc (1171)
3623 - fix ia64 OOPS in llog_test (2255)
3624 - zero end of page at obdfilter for partial page writes (2648)
3625 - don't leave stale dentries around after renames (bug 2428)
3626 - fix timeouts when evicting a client with a single lock held (2642)
3627 - set deadline for the initial HELLO message to drain (2634)
3628 - print out dotted-quad IP addresses in the socknal (2302)
3630 - additional debugging for MDS client eviction problem (2443)
3631 - fix mkfsoptions support for osts (2603, 2604)
3633 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3636 - add series for the vanilla 2.6.0 kernel
3637 - add series for the vanilla 2.4.24 kernel
3638 - add series for a cray x86/64 UL kernel drop
3639 - fix xattr patches for the vanilla 2.4.19 series
3641 - generate true UUIDs in lmc (1171)
3642 - have portals stack dumping break in UML (2466)
3643 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3644 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3645 - wake callback waiting threads on client eviction (2460)
3646 - Add --ptldebug and --subsystem to lmc (1719)
3647 - update assertion to allow safe interrupt allocation
3648 - set rq_no_resend for cancel requests (2432)
3649 - recalculate ptlrpcd timeout after resend (2494)
3650 - call vfs_rmdir when removing pending directories (2368)
3651 - fix renaming a file to itself (2429)
3652 - lmc creates a default one-stripe lov (2454)
3653 - expand procfs space to handle large clusters (2326)
3654 - increase UML stack to avoid overflow
3655 - update lconf's list of debug and subsystem masks
3656 - fix lfs find --obd (2510)
3657 - /proc tunable for disabling filter read caching (2591)
3658 - stop rpm packages from altering slapd.conf (2301)
3659 - disable nagle in the socknal under 0conf (2578)
3660 - choose mds inode size based on stripe count (2572)
3661 - fix kernel-source rpm problems (2516)
3663 - add --disable-doc to avoid pdf generation (2421)
3664 - update documentation, tests, type-os, comments
3665 - avoid format warnings on ia64
3666 - remove the TOE NAL
3667 - tiny code cleanups by removing unused fields
3669 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3672 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3673 - remove the most blatant lies from BUILDING file (2371)
3674 - change default debug level to reasonable production setting
3675 - reduce client side cache size to reduce cache flush time
3676 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3677 - make TCP zerocopy and pinger support enabled by default (2476)
3678 - sync writes completed after process exits caused crashes (2319)
3679 - maintain correct mount count on the MDS (2356)
3680 - backout 1557, because 2316 wasn't really fixed
3681 - better file I/O statistics gathering in /proc
3682 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3683 - another kernel patch to fix zero-copy TCP function export
3684 - don't take duplicate lock when processing re-sent getattr (2420)
3685 - lctl uses obd_self_export instead of creating new conn (2353)
3686 - MDS/OST recovery case which requires object creation asserted (2425)
3687 - move lfs from /usr/sbin to /usr/bin in packages
3688 - fix race between mds_client_add and mds_client_free (2417)
3689 - use kmalloc instead of slabs in portals (2430)
3690 - don't create duplicate records when a failover MDS is present (2442)
3691 - remove unnecessary mount age check (2332)
3692 - don't remove directory inodes from locks prematurely (2451)
3693 - don't break if MDS service name is the same as hostname (2103)
3694 - fix races in client write RPC generation when cache full (2482)
3696 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3699 - remove now-unused request->rq_obd (278)
3700 - if an allocation fails, print out how much memory we've used (1933)
3701 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3702 - add the "configurable stack size" patch to most series files (1256)
3703 - ability to write large log records, for 100+ OST configs (2306)
3704 - fix NULL deref when filter_prep fails (2314)
3705 - fix operator precedence error in filter_sync
3706 - dynamic allocation of socknal TX descriptors (2315)
3707 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3708 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3709 - ensure that CWARN messages in Portals always get to the syslog
3710 - __init/__exit are not for prototype decls (ldlm_init/exit)
3711 - x86-64 compile warning fixes
3712 - fix gateway LMC keyword conflict (2318)
3713 - fix MDS lock inversions in getattr/reint paths (1844)
3714 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3715 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3716 - DLM race condition prevented some lock evictions (2328)
3717 - ENOMEM detection and retry on socknal sends (2230)
3718 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3719 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3720 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3721 - better error messages when a client is rejected during recovery (1505)
3722 - avoid cancelling locks which were never granted, after failure (2330)
3723 - fix i_sem/journal inversion in mds_client_add (2333)
3724 - fix truncate/getattr lock cycle deadlock (2334)
3725 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3726 - fix two rare exit paths which could leak an l_lock() ref (2321)
3727 - fencepost error in MDS/OST orphan recovery (2226)
3728 - make log record alignment 8 bytes (1988)
3729 - lstripe now fails when requested offset > ost_count (2237)
3730 - ensure that all kernel series have a complete list.h (1607)
3731 - fix crashes in special-file operations (2316)
3732 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3734 - allow configurable automake binary, for testing new versions
3735 - small update to the lfs documentation
3737 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3739 * fix negative export reference count in fsfilt_sync (2312)
3741 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3742 * release candidate 0.9.1
3744 - orphans are moved into the PENDING directory for possible recovery
3745 - replayed opens now open by fid for orphan/rename safety (1042)
3746 - last close of an orphan inode generates a transno (683)
3747 - chdir() and mount() now pin the directory entry (1020)
3748 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3749 - discard very old requests without processing them (1502)
3750 - remove some common, well-understood CERRORs (1505)
3751 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3752 - clear "grant" flags in OST replies until OST grant code lands (1644)
3753 - fix read performance by not clobbering i_blksize on client (1598)
3754 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3755 - make LRU size a /proc tunable, clears locks when reduced (707)
3756 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3757 - server locks take references on exports now (1558)
3758 - build fixes for 2.4.20-rh trees (1663)
3759 - return an error from lov_create if all OSCs are inactive (1751)
3760 - fix import levels when a reconnect happens without a timeout (1597)
3761 - exit early from mds_open if we get a lookup error (1749)
3762 - partial page read at EOF wouldn't wait for disk before sending (1642)
3763 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3764 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3765 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3766 - write/truncate lock inversion (1639)
3767 - Don't auto-load obdclass, portals modules during cleanup (1495)
3768 - fix timestamps from jumping to "now" (1763)
3769 - extra journal assertions (1648)
3770 - add an extra multiunlink test (1771)
3771 - fix read_record/write_record API (1776)
3772 - fix leak of offset_extent, possible incorrect i_size later (1772)
3773 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3774 - Debugging for the kqswnal_get_idle_tx problems (1820)
3775 - Allow recovery to be attempted multiple times (1536)
3776 - Write out MDS last_rcvd file after it is first created (1600)
3777 - Fix tx_descriptor leak in failed transmit situations (1827)
3778 - ext3 journaling fixes for assertion failure after IO error (1871)
3779 - class_export_put() on freed export after completion AST error (1896)
3780 - Fix revalidate looping in VFS (1322)
3781 - Don't access a freed export during MDS_REINT timeout (1521)
3782 - Add open-unlink recovery support on the MDS (1673,1764)
3783 - Return an error if no MDS data was read from last_rcvd (1946)
3784 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3785 - Don't setup a disk device that doesn't match exported UUID (317)
3786 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3787 - avoid committing NULL handle in force close
3788 - local.sh is now a one-stripe LOV configuration
3789 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3790 - fix ext3 htree duplicate directory entry corruption (1516)
3791 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3792 - update to new LOV EA format (2097)
3793 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3795 2003-06-15 Phil Schwan <phil@clusterfs.com>
3798 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3799 - per-import/export recovery handling (958, 931, 959)
3800 - multiple last-rcvd slots, for serving multiple FSes (949)
3801 - connections are again shared between multiple imp/exports (963, 964)
3802 - "umount -f" would hang if any requests needed to be sent (393, 978)
3803 - avoid pinning large req buffer by copying for queued messages (989)
3804 - add "uuid" to "lctl device" command to help upcalls (991)
3805 - "open" RPCs with transnos would confuse recovery counters (1037)
3806 - do proper endian conversion of all wire messages (288, 340, 891)
3807 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3808 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3809 - fix ptlrpc_connection leak in target_handle_connect (1174)
3810 - fix import refcounting bug in OST and MDS cleanup (1134)
3811 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3812 - fix very unlikely obd_types race condition (501)
3813 - remove osc_open hack for echo_client (1187)
3814 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3815 - a failure in read_inode2 leads to deadlock (1139)
3816 - cancel ack-locks as soon as transaction is committed (1072)
3817 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3818 - make sure to commitrw after any preprw to avoid deadlock (1162)
3819 - failing to execute a file in a lustre FS would lock inode (1203)
3820 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3821 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3822 - obd_uuid2tgt would walk off the end of the list (1255)
3823 - on IA64 the osc would give portals incorrect bulk size (1258)
3824 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3825 - fix lock inversion caused by new llite matching code (1282)
3826 - limit the number of dirty pages on a client to 10MB (1286)
3827 - timed out locks were not being corrected cancelled (1289)
3828 - fix O_DIRECT above 4GB on IA-32 (1292)
3829 * major user-visible changes
3830 - fail out/fail over policy now controlled by the upcall (993)
3832 - add OBD_PING to check server availability and failure (954)
3833 - lustre messages are now sent in sending host order (288, 340, 891)
3834 - add eadatalen to MDS getattr reply (340)
3835 - OST read replies may contain second buffer, with per-page status (593)
3837 2003-03-11 Phil Schwan <phil@clusterfs.com>
3840 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3841 - failed open()s could cause deadlock; fixed (867, 869)
3842 - stop cancelling OST locks when files are closed (481)
3843 - overlapping XID spaces caused network corruption (851, 853)
3844 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3845 - setattr_raw added, to do single-RPC, server-side setattrs
3846 - lmc/lconf syntax change for OST UUIDs
3847 - fix crashy race condition between ptlrpc_free_req and osc_close
3848 - don't use request in mdc_enqueue if we hit a timeout (889)
3849 - don't set the inode i_size for regular files from the MDS (896)
3850 - handle out of order completion AST (842)
3851 - don't LBUG if a lock request times out after receiving AST (913)
3852 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3853 - if a bad lock AST arrives, send an error instead of dropping entirely
3854 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3855 - fix leak in bulk IO when only partially completed (899, 900, 926)
3856 - fix O_DIRECT for ia64 (55)
3857 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3858 - C-z after timeout could hang a process forever; fixed (977)
3860 - client-side I/O cache (678, 924, 929, 941, 970)
3862 - READPAGE and SETATTRs which don't take server-side locks get
3865 2003-02-11 Phil Schwan <phil@clusterfs.com>
3868 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3869 - Fix fsfilt_extN_readpage() to read a full page of directory
3870 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3871 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3872 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3873 - Most intent operations were removed, in favour of a new RPC mode
3874 that does a single RPC to the server and bypasses most of the VFS
3875 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3876 - Aggressively cancel local locks on DLM servers
3877 - mds_reint_unlink sends EA to the client if it's the last nlink.
3878 client uses that EA to unlink OST objects.
3879 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3880 - recursive symlinks were fixed (439)
3881 - fixed NULL deref in DEBUG_REQ
3882 - filter_update_lastobjid no longer calls sync, which annoyed extN
3883 - fixed multi-client small-writes to a single file problem (445)
3884 - fixed mtime updates during file writes (607)
3885 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3886 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3887 - fixed Linux OST size reporting problem (444, 656)
3888 - OST now updates object mtime with writes or setattr (607, 619)
3889 - client verifies file size before zeroing page past EOF (445)
3890 - OST now writes last allocated objid to disk with allocation (108)
3891 - LOV on echo now works (409)
3893 - mds_reint_unlink sends a new buffer, with the EA included. this
3894 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3895 set if a regular file was being unlinked, and it was the last link
3896 - use PtlGet from the target for bulk writes (315)
3897 - OST now updates object mtime with writes or setattr (607, 619)
3898 - LDLM now has a grant-time callback to revalidate locked items, if
3900 - Many MDS operations were reorganized to combat race conditions
3902 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3903 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3905 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3908 - Fully reactivate OST imports after reconnection (512, others)
3909 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3910 - More graceful error handling for truncating on dead OST (515)
3911 - Don't error out unless we're actually accessing dead stripes (474)
3912 - Fix garbage sizes when stripes are missing (410)
3913 - LRU counters were broken, causing constant lock purge (433, 432)
3914 - garbage on read from stripes with failed OSTs (441)
3915 - mark OSCs as active before reconnecting during recovery (438)
3916 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3917 - lfind did not preserve OST order in output (443)
3918 - symlinks cause hung clients, incorrect data (439)
3919 - stop dereferencing request after dropping refcount (457)
3920 - don't LASSERT(spin_is_locked) on non-SMP (455)
3921 - fixes for many rename() bugs
3922 - fstat didn't correctly synchronize attributes (399)
3923 - server must handle lock cancellation during blocking AST prep (487)
3924 - bulk descriptors were free()d too soon (511)
3925 - fix paths in lconf, which would load incorrect modules (451, 507)
3926 - fix confusing lconf 'host not found' error message (386)
3927 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3928 - fix race condition in mdc_blocking_ast() for inode access (526)
3929 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3930 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3931 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3932 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3933 - fix lstripe in several ways (kernel side) (527)
3934 - fix request leak in ldlm_cli_enqueue (262)
3935 - incorrect OSC was marked inactive after OST failure
3936 - call mds_fs_cleanup before unmounting filesystem (524)
3937 - fix races between taking ns_lock and ldlm_lock_change_resource
3938 - fix races updating LOV export open file list
3939 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3940 - fix recovery NULL deref in ldlm_cli_cancel_unused
3941 - fix some DLM races by using new hash table for lock handles (419)
3942 - permit the client to specify desired inodes, at replay
3943 - duplicate requests when we queue them for replay reintegration
3944 - fix last_rcvd offset calculation
3945 - sync after each recovered transaction, so we always make progress
3946 - never, not always, ERESTART requests without transnos
3947 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3949 - skip replay if the MDS says that the client is already connected
3950 - don't check for a recovery-enabled export to match lctl's UUID
3951 - don't INC_USE_COUNT for phantom exports
3952 - don't crash when cleaning up phantom exports (567)
3953 - don't double-finish or set replay data for errored mdc_open requests
3954 - abort requests when they time out, so we don't get old replies
3955 - send/receive replies for AST messages again
3956 - if the client says that it doesn't have the lock, cancel it on the
3958 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3959 mark it as destroyed, it will all work out when decref is called
3960 - fix module use counts (22, 581)
3962 - ASTs now expect a reply (server cancels lock on error reply)
3964 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3967 - fix many simultaneous client startup (392)
3968 - fix dentry->d_it clobbering
3969 - credentials weren't being shipped for readdir/getattr operations
3970 - remove invalid assertions triggered during some concurrent MD
3972 - proper Lustre versions added (336, 389)
3973 - fix memory leak for create error case (398)
3974 - fix LOV locking bug that would get cli/srv out of sync
3975 - fix echo client over LOV (409)
3976 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3977 - fix double-O_EXCL intent crash (424)
3978 - avoid sending multiple lock CANCELs (352)
3980 - MDS can do multi-client recovery (modulo bugs in new code)
3982 - many updates, edits, cleanups
3984 2002-11-18 Phil Schwan <phil@clusterfs.com>
3987 - fix null d_it dereference (346)
3988 - fix full OST/dbench hang (333)
3989 - fix permission problem with file removal (286)
3990 - fix removal of OSCs from LOV when they fail
3991 - fix NULL deref during bulk timeout (214)
3992 - fix problems related to multiple filesystems on one MDS (241)
3993 - fixed serious subtle metadata locking bugs
3994 - free locks on clients when inodes are removed due to memory
3996 - fix inode pointer in lock data (285)
3997 - partial support for multiple MDS on a single host (241)
3998 - data locks weren't cancelled at clear_inode time (290, 311)
3999 - intent locks could lead to unbounded lock growth (205)
4000 - added a maximum lock count, an LRU list, and a flusher
4001 - fix multiple rename (365)
4002 - properly abstracted the echo client
4003 - OSC locked 1 byte too many; fixed
4004 - rewrote brw callback code:
4005 - fixed recovery bugs related to LOVs (306)
4006 - fixed too-many-pages-in-one-write crash (191)
4007 - fixed (again) crash in sync_io_timeout (214)
4008 - probably fixed callback-related race (385)
4010 - Add capability to MDS protocol
4011 - LDLM cancellations and callbacks on different portals
4013 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4016 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4017 - defer open object destruction to close time (601981, 138)
4018 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4019 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4020 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4021 - don't call obd_setattr in ll_file_release for destroyed objects
4022 * protocol change to lustre_msg: move |version| and add |flags|
4023 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4024 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4025 * added replay of create, unlink, link and rename operations during
4026 MDS failover; recovery should be much more robust now
4027 * remove failed OSCs from LOVs (only lov_create uses this so far)
4028 * the lustre-HOWTO was brought (more) up to date (582544)
4030 2002-10-23 Phil Schwan <phil@clusterfs.com>
4033 - in-use dentries weren't being reused properly (617851)
4034 - prevent multiple LDLM setup (599178)
4035 - fix LOV size calculations for truncate (617853)
4036 - fix client handling of MDS intent errors (POSIX)
4037 - fix permission bug in lovstripe.c test (624321)
4038 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4039 - truncate past end of file could corrupt data
4040 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4041 - a race in recovery could return ETIMEDOUT to apps (623947)
4042 - building outside the source directory was fixed
4043 * the lustre-HOWTO was brought (more) up to date (582544)
4044 * major progress was made on recovery functionality
4046 2002-10-10 Phil Schwan <phil@clusterfs.com>
4049 - recovery deadlock fix
4050 - rm -rf causes LBUG fix (617817)
4051 - file open by multiple tasks fix (618962)
4052 - directory permissions bugs (602707 and 620007)
4053 - journal_stop fixed with locking (611313)
4054 - O_APPEND failures resolved (618273, perhaps 614459)
4055 - lconf PATH fix (619770)
4056 - IA64 build fix (621450)
4057 - RPC buffer sizes scale with amount of memory
4059 2002-10-01 Phil Schwan <phil@clusterfs.com>
4062 - locks would be cancelled without throwing away data pages,
4063 resulting in inconsistent data (605627)
4064 - inode attributes were not always being refreshed (605627, 612449)
4065 - lconf now continues to cleanup after lctl reports an error
4066 - MDS now enforces user permissions (602707)
4067 - lprocfs cleanup fixed, but not yet enabled (614157)
4068 - fixed infinite server hang, should a client not respond to an AST
4069 - avoid going into recovery if user calls readlink() with a buffer
4070 that's too small (613941)
4071 - AST RPCs no longer require replies (614867) -- this may be changed
4072 - don't crash server if client sends an IOV that's too big (611336)
4073 - fixed lock conversion deadlock (611892)
4074 - fixed the following of symlinks (614622)
4075 * recovery: the server can remove locks from a client that dies, other
4076 clients can make progress
4077 * more extN patch fixes
4078 * compile-time configurable ptlrpc buffer allocations
4080 - collaborative read cache document
4081 - Lustre Lite Performance CDR document-in-progress
4083 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4086 - fix typo in patch-2.4.18
4088 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4091 - clear ptlrpc request each time in handle_incoming_request()
4092 - unlink of files now destroys the object on the OST
4094 2002-09-19 Peter Braam <braam@clusterfs.com>
4096 * add hard link support
4097 * change obdfile creation method
4098 * kernel patch changed
4100 2002-09-19 Peter Braam <braam@clusterfs.com>
4103 - stack overflow bug in extN fixed
4105 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4107 * documentation updates
4108 - add man pages for config tools
4109 - update tests/README to describe testing with new config tools
4110 - finish metadata API descriptions
4111 * bug fixes and cleanups
4112 - statfs workaround for 16TB limit
4113 - LOV stripe allocation improved, can stripe on subset of OSTs
4114 - LOV file size/IO offset was wrong for files > 4GB in size
4115 - object EA data was being dropped, caused files to be unreadable
4116 - memory overflow with non-LOV OST caused memory corruption
4117 - fixed regression tests to work with new config tools, obdfilter
4118 - fixed bug when directory size became larger than 1 block
4119 - fixed bug (for single client case) when PWD was deleted
4120 - invalidate local directory pages when doing intent-based ops
4121 - avoid LDLM oops when lock callback contained bad data
4123 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4125 * documentation updates
4126 * bug fixes and cleanups
4127 - configuration tools
4130 - 64-bit compile warnings
4131 - 64-bit internal statfs data
4133 * test_brw on persistent OST devices
4135 * lprocfs (disabled)
4137 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4139 * documentation updates
4140 * bug fixes and cleanups
4141 * configuration tools
4143 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4146 * many small fixes to 0_5_4
4147 * io/network handling
4148 * thinkos in MDS operations
4150 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4153 * crucial basic fixes to 0.5.3
4154 * IOR, Iozone work over Elan
4157 2002-08-07 Phil Schwan <phil@clusterfs.com>
4158 * version 0_5_3, our first alpha
4159 * we use the new Portals iovs
4160 * documentation updates
4161 * bug fixes and cleanups
4162 * small changes in the DLM wire protocol
4164 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4165 * version 0_5_1 with some initial stability,
4166 * locking on MD and file I/O.
4167 * documentation updates
4168 * several bug fixes since 0.5.0
4169 * small changes in wire protocol
4171 2002-07-18 Phil Schwan <phil@clusterfs.com>
4173 * delivered as Lustre Light Alpha
4174 * fixed a crash after handling invalid MDS requests
4175 * fixed directory pages for architectures with non-4k pages sizes
4177 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4178 * release version v0_4_4
4179 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4181 * Fixes a number of interruption problems with OST operations.
4182 * Update documentation for portals header changes
4183 * Move all wire protocol structs/defines to lustre_idl.h
4184 * Fixes symlink length bug.
4185 * Add tcpdump to repository.
4187 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4188 * release version v0_4_3
4189 * Fixes statfs for inodes on extN.
4190 * Fixes bug in runtests which would delete /etc/hosts.
4191 * Use 64-bit object IDs wherever possible (not into VFS though)
4192 Remove ost_get_info, which is unused by lustre, and out of date.
4194 2002-07-03 Peter Braam <braam@clusterfs.com>
4195 * release version v0_4_2 Fixes a lookup error (type not passed)
4196 * move forward to head of Portals
4197 * move forward to latest Lustre kernel
4199 2002-06-25 Peter Braam <braam@clusterfs.com>
4200 * release version v0_4_1. Hopefully stable on single node use.