3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.53-0.16 (SLES 10),
6 2.6.18-8.1.15.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 to RHEL5 kernel 2.6.18-8.1.15.el5.
21 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
25 Description: Update to SLES9 kernel-2.6.5-7.287.3.
29 Description: Update to RHEL4 latest kernel.
33 Description: Remove 2.4 kernel support from lustre 1.8.0
34 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
35 quota_interface.c in quota subsystem.
36 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
38 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
39 lvfs_linux.c upcall_cache.c in lvfs subsystem.
40 Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
41 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
43 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
44 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
46 Remove 2.4 kernel check from lproc_osc.c, osc_request.c in osc
48 Remove 2.4 kernel check from filter.c, lproc_obdfilter.c in
50 Remove 2.4 kernel code from lustre/obdecho/echo_client.c.
51 Remove 2.4 kernel check from lustre/mgs/lproc_mgs.c,
53 Remove 2.4 kernel code from lustre/mgc/mgc_request.c.
54 Remove 2.4 kernel check from lustre/mdt/mdt_lproc.c.
55 Remove 2.4 kernel code from lustre/mdc/mdc_request.c.
56 Remove 2.4 kernel check from lustre/lov/lproc_lov.c.
57 Remove 2.4 kernel check from lustre/lmv/lproc_lmv.c.
58 Remove lock_24kernel from lustre/lvfs/fsfilt_ext3.c.
59 Remove 2.4 kernel code from include/liblustre.h,
60 include/lustre_disk.h, include/lustre_quota.h,
61 include/linux/lprocfs_status.h, include/linux/lvfs.h,
62 include/linux/lustre_lite.h, include/linux/lustre_net.h,
63 include/linux/lustre_user.h, include/linux/lvfs_linux.h,
64 include/linux/obd_class.h, include/linux/lustre_compat25.h.
66 Severity : enhancement
68 Description: Build SLES10 patchless client fails
69 Details : The configure was broken by run ./configure with
70 --with-linux-obj=.... argument for patchless client. When the
71 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
72 can't search header adequately. Use absolute path such as
73 -I($LINUX)/include instead.
77 Description: interrupt oig_wait produce painc on resend.
78 Details : brw_redo_request can be used for resend requests from ptlrpcd and
79 private set, and this produce situation when rq_ptlrpcd_data not
80 copyed to new allocated request and triggered LBUG on assert
81 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
82 ptlrpcd set if request is changed and can be safety changed to use
87 Description: LASSERT_{REQ,REP}SWAB macros are buggy
88 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
89 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
90 these macros and replace them with their difinitions instead.
94 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
95 Details : Update which_patch & target file for SLES10 latest kernel.
98 Description: Remove client patches from SLES 10 kernel.
99 Details : This causes SLES 10 clients to behave as patchless clients
100 even on a Lustre-patched (server) kernel.
102 Severity : enhancement
104 Description: self-adjustable client's lru lists
105 Details : use adaptive algorithm for managing client cached locks lru
106 lists according to current server load, other client's work
107 pattern, memory activities, etc. Both, server and client
108 side namespaces provide number of proc tunables for controlling
111 Severity : enhancement
113 Description: light-weight GSS support
114 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
115 and reduce performance overhead.
117 Severity : enhancement
119 Description: Linux keyring support
120 Details : Support using service of Linux keyring for Lustre GSS internal
121 context refresh/cache mechanism.
125 Description: Fix errors in lfs documentation
126 Details : Fixes man pages
130 Description: don't use GFP_* in generic Lustre code.
131 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
135 Description: obdclass is limited by single OBD_ALLOC(idarray)
136 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
140 Description: warnings with build patchless client with vanila 2.6.19 and up
141 Details : change old ctl_table style and replace ctl_table/ctl_table_header
142 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
146 Frequency : Only for SLES
147 Description: sanity_quota fail test_1
148 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
149 file, which makes TSTID[2] inunique.
153 Frequency : Only for SLES9
154 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
155 Details : Update target/ChangeLog/which_patch .
159 Frequency : Only for test_10 in sanity-quota.sh
160 Description: a bug in quota test code
161 Details : checking a test flag in a wrong place
165 Frequency : Only for test_13 in sanity-quota.sh
166 Description: a bug in quota test script
167 Details : shouldn't check overfull allocated quota
171 Description: jbd statistics
172 Details : Port older jbd statistics patch for sles10
174 Severity : enhancement
176 Description: Add printing inode info into message about error in writepage.
181 Description: reply_lock_interpret crash due to race with it and lock cancel.
182 Details : Do not replay locks that are being cancelled. Do not reference
183 locks by their address during replay, just by their handle.
187 Frequency : When flocks are used.
188 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
189 Details : Flock locks might destroy just granted lock if it could be merged
190 with another existing flock, this is done in completion handler,
191 so teach ldlm_cli_enquque_fini that this is a valid case for
197 Description: Oops in read and write path when failing to allocate lock.
198 Details : Check if lock allocation failed and return error back.
202 Description: lstripe command fails for valid OST index
203 Details : The stripe offset is compared to lov->desc.ld_tgt_count
204 instead of lov->desc.ld_active_tgt_count.
208 Description: sanity.sh failed test 103
209 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
214 Description: df doesn't work properly if diskfs blocksize != 4K
215 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
219 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
220 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
224 Description: obd_health_check_timeout too short
225 Details : set obd_health_check_timeout as 1.5x of obd_timeout
229 Description: llapi_file_create() does not allow some changes
230 Details : add llapi_file_open() that allows specifying the mode and
231 open flags, and also returns an open file handle.
235 Description: merge and cleanup kernel patches.
236 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
240 Description: Add journal checksum support.(Kernel part)
241 Details : The journal checksum feature adds two new flags i.e
242 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
243 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
244 indicates that the commit block contains the checksum for
245 the blocks described by the descriptor blocks. Now commit
246 record can be sent to disk without waiting for descriptor
247 blocks to be written to disk. This behavior is controlled
248 using JBD2_FEATURE_ASYNC_COMMIT flag.
252 Description: OSS needs mutliple precreate threads
253 Details : Add ability to start more than one create thread per OSS.
257 Description: Sanity.sh test_65k failed: setstripe should have succeeded
258 Details : execute lfs setstripe on client
262 Description: mds_obd_create error creating tmp object
263 Details : When the user sets quota on root, llog will be affected and can't
264 create files and write files.
267 Frequency : Always on ia64 patchless client, and possibly others.
269 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
270 Details : This allows the patchless client to be loaded on architectures
275 Description: RedHat Update kernel for RHEL5
276 Details : Add the kernel config file for RHEL5.
280 Description: RedHat Update kernel for RHEL5
281 Details : Modify the kernel config file more closer RHEL5.
285 Description: Build failure against Centos5 (RHEL5)
286 Details : Define PAGE_SIZE when it isn't present.
290 Description: test_7 on sanity-quota.sh can't be used on separate machines
291 Details : add facet to handle it
295 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
296 Details : with NFS, the anon dentry's parent was set to itself in
297 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
298 corresponding dentry other than use rec->ur_name.
300 Severity : enhancement
302 Description: lfs setstripe enhancement
303 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
307 Description: testing performance impact of enabling checksumming
308 Details : enable checksum by default, allow --disable-checksum
309 configure option and "-o nochecksum" mount option
313 Description: lustre support for RHEL5
314 Details : Add support for RHEL5.
318 Description: Client eviction due to failover config
319 Details : after a connection loss, the lustre client should attempt to
320 reconnect to the last active server first before trying the
321 other potential connections.
325 Description: when mds and osts use different quota unit(32bit and 64bit),
326 quota will be released repeatly.
327 Details : void sending multiple quota reqs to mds, which will keep the
328 status between the reqs.
332 Description: osts not allocated evenly to files
333 Details : change the condition to increase offset_idx
336 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
338 Description: Data corruption for OSTs that are formatted larger than 2TB
340 Details : When generating the bio request for lustre file writes the
341 sector number would overflow a temporary variable before being
342 used for the IO. The data reads correctly from Lustre (which
343 will overflow in a similar manner) but other file data or
344 filesystem metadata may be corrupted in some cases.
348 Description: Tune the kernel for good SCSI performance.
349 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
350 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
355 Description: rewrite ext2-derived code in obdclass/uuid.c
356 Details : rewrite inherited code (uuid parsing code from ext2 utils)
357 from scratch preserving functionality.
361 Description: "lfs find -obd UUID" prints directories
362 Details : "lfs find -obd UUID" will return all directory names instead
363 of just file names. It is incorrect because the directories
364 do not reside on the OSTs.
368 Description: use i_size_read and i_size_write in 2.6 port
369 Details : replace inode->i_size access with i_size_read/write()
372 Frequency : when removing large files
374 Description: scheduling issue during removal of large Lustre files
375 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
376 It causes scheduling issues when removing large files (17TB in the
380 Frequency : only with liblustre clients on XT3
382 Description: evictions taking too long
383 Details : allow llrd to evict clients directly on OSTs
386 Frequency : only on ppc
388 Description: /proc/fs/lustre/devices broken on ppc
389 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
390 We should make sure the type of 'index' is loff_t and then cast
391 explicitly as needed below. Do not assign an explicitly cast
395 Frequency : only for rhel5
397 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
398 Details : Modify the target file & which_kernel.
400 Severity : enhancement
402 Description: omit set fsid for export NFS
403 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
404 as FS_REQUIRES_DEV for avoid problems with generate fsid.
408 Frequency : Always, for kernels after 2.6.16
409 Description: Fix warning idr_remove called for id=.. which is not
411 Details : Last kernels save old s_dev before kill super and not allow
412 to restore from callback - restore it before call kill_anon_super.
416 Description: buffer overruns could theoretically occur
417 Details : llapi_semantic_traverse() modifies the "path" argument by
418 appending values to the end of the origin string, and a
419 overrun may occur. Adding buffer overrun check in liblustreapi.
423 Description: Fix error on 'ls .' at the top of the Lustre mount.
424 Details : Don't revalidate dentry if it is a root dentry.
428 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
429 Details : Modify vm-tunables-rhel4.patch.
433 Description: Kernel config for 2.6.18-vanilla.
434 Details : Modify targets/2.6-vanilla.target.in.
435 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
436 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
437 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
438 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
442 Description: improve handling recoverable errors
443 Details : if request processig with error which can be recoverable on server
444 request should be resend, otherwise page released from cache and
449 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
450 Details : Modify target file & which_patch.
451 A flaw was found in the IA32 system call emulation provided
452 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
453 value could be stored in the %RAX register, which could trigger an
454 out-of-bounds system call table access. An untrusted local user
455 could exploit this flaw to run code in the kernel
456 (ie a root privilege escalation). (CVE-2007-4573).
460 Description: O_DIRECT bypasses client statistics.
461 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
462 read_ahead_stats were updated. copied the stats section from
463 osc_send_oap_rpc() into async_internal().
467 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
471 Description: change order of libsysio includes
472 Details : '#include sysio.h' should always come before '#include xtio.h'
476 Description: handle "serious error: objid * already exists" more gracefully
477 Details : If LAST_ID value on disk is smaller than the objects existing in
478 the O/0/d* directories, it indicates disk corruption and causes an
479 LBUG(). If the object is 0-length, then we should use the existing
480 object. This will help to avoid a full fsck in most cases.
482 Severity : enhancement
484 Description: adapt the lustre_config script to support the upgrade case
485 Details : Add "-u" option for lustre_config script to support upgrading 1.4
486 server targets to 1.6 in parallel.
490 Description: To avoid grant space > avaible space when the disk is almost
491 full. Without this patch you might see the error "grant XXXX >
492 available" or some LBUG about grant, when the disk is almost
494 Details : In filter_check_grant, for non_grant cache write, we should
495 check the left space by if (*left > ungranted + bytes), instead
496 of (*left > ungranted), because only we are sure the left space
497 is enough for another "bytes", then the ungrant space should be
498 increase. In client, we should update cl_avail_grant only there
499 is OBD_MD_FLGRANT in the reply.
503 Description: Update RHEL 4 kernel to fix local root privilege escalation.
504 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
505 described in CVE-2007-4573. This problem could allow untrusted
506 local users to gain root access.
510 Frequency : when using O_DIRECT and quotas
512 Description: Incorrect file ownership on O_DIRECT output files
513 Details : block usage reported by 'lfs quota' does not take into account
514 files that have been written with O_DIRECT.
516 --------------------------------------------------------------------------------
518 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
520 * Support for kernels:
521 2.6.5-7.283 (SLES 9),
522 2.6.9-55.EL (RHEL 4),
523 2.6.16.46-0.14 (SLES 10),
524 2.6.18.8 vanilla (kernel.org)
525 * Client support for unpatched kernels:
526 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
527 2.6.16 - 2.6.22 vanilla (kernel.org)
528 * Due to recently discovered recovery problems, we do not recommend
529 using patchless RHEL 4 clients with this or any earlier release.
530 * Recommended e2fsprogs version: 1.39.cfs8
531 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
532 * Starting with this release, the ldiskfs backing filesystem required
533 by Lustre is now in its own package, lustre-ldiskfs. This package
534 should be installed. It is versioned separately from Lustre and
535 may be released separately in future.
539 Description: block reactivating mgc import until all deactivates complete
540 Details : Fix race when failing back MDT/MGS to itself (testing)
543 Frequency : at statup only
545 Description: mds_lov_synchronize race leads to various problems
546 Details : simultaneous MDT->OST connections at startup can cause the
547 sync to abort, leaving the OSC in a bad state.
549 Severity : enhancement
551 Description: add optional extra BUILD_VERSION info
552 Details : add a new environment variable (namely LUSTRE_VERS) which allows
553 to override the lustre version.
556 Frequency : 2.6.18 servers only
558 Description: ll_kern_mount() doesn't release the module reference
559 Details : The ldiskfs module reference count never drops down to 0
560 because ll_kern_mount() doesn't release the module reference.
565 Description: server LBUG when using old ost_num_threads parameter
566 Details : Accept the old ost_num_threads parameter but warn that it
567 is deprecated, and fix an off-by-one error that caused an LBUG.
572 Description: Transient SCSI error results in persistent IO issue
573 Details : iobuf->dr_error is not reinitialized to 0 between two
577 Frequency : sometimes when underlying device returns I/O errors
579 Description: OSTs not going read-only during write failures
580 Details : OSTs are not remounted read-only when the journal commit threads
581 get I/O errors because fsfilt_ext3 calls journal_start/stop()
582 instead of the ext3 wrappers.
586 Description: poor connect scaling with increasing client count
587 Details : Don't run filter_grant_sanity_check for more than 100 exports
588 to improve scaling for large numbers of clients.
591 Frequency : SLES10 only
593 Description: sanity-quota.sh quotacheck failed: rc = -22
594 Details : Quotas cannot be enabled on SLES10.
597 Frequency : liblustre clients only
599 Description: getdirentries does not give error when run on compute nodes
600 Details : getdirentries does not fail when the size specified as an argument
601 is too small to contain at least one entry
603 Severity : enhancement
605 Description: Add LNET router traceability for debug purposes
606 Details : If a checksum failure occurs with a router as part of the
607 IO path, the NID of the last router that forwarded the bulk data
608 is printed so it can be identified.
613 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
614 Details : Due to a race condition, liblustre clients were occasionally
617 Severity : enhancement
619 Description: lfs setstripe use optional parameters instead of postional
622 Severity : enhancement
624 Description: Nanosecond timestamp support for ldiskfs
625 Details : The on-disk ldiskfs filesystem has added support for nanosecond
626 resolution timestamps. There is not yet support for this at
627 the Lustre filesystem level.
630 Frequency : during server recovery
632 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
633 Details : request with rq_no_resend flag not awake l_wait_event if they get a
637 Frequency : nfs export on patchless client
639 Description: connectathon hang when test nfs export over patchless client
640 Details : Disconnected dentry cannot be found with lookup, so we do not need
641 to unhash it or make it invalid
644 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
645 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
646 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
651 Description: open req refcounting wrong on reconnect
652 Details : If reconnect happened between getting open reply from server and
653 call to mdc_set_replay_data in ll_file_open, we will schedule
654 replay for unreferenced request that we are about to free.
655 Subsequent close will crash in variety of ways.
656 Check that request is still eligible for replay in
657 mdc_set_replay_data().
662 Description: disable writes to filesystem when reading health_check file
663 Details : the default for reading the health_check proc file has changed
664 to NOT do a journal transaction and write to disk, because this
665 can cause reads of the /proc file to hang and block HA state
666 checking on a healthy but otherwise heavily loaded system. It
667 is possible to return to the previous behaviour during configure
668 with --enable-health-write.
670 Severity : enhancement
672 Description: 64-bit inode version
673 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
674 to the inode. This will be required for version-based recovery.
679 Description: MDS fails to start if a duplicate client export is detected
680 Details : in some rare cases it was possible for a client to connect to
681 an MDS multiple times. Upon recovery the MDS would detect this
682 and fail during startup. Handle this more gracefully.
684 Severity : enhancement
686 Description: Add -o localflock option to simulate old noflock
688 Details : This will achieve local-only flock/fcntl locks
694 Description: log_commit_thread vs filter_destroy race leads to crash
695 Details : Take import reference before releasing llog record semaphore
700 Description: Wrong request locking in request set processing
701 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
702 imp_delayed_list, in this place should be used imp_lock.
705 Frequency : when reconnection
707 Description: Grant Leak when osc reconnect to OST
708 Details : When osc reconnect ost, OST(filter) should check whether it
709 should grant more space to client by comparing fed_grant and
710 cl_avail_grant, and return the granted space to client instead
711 of "new granted" space, because client will call osc_init_grant
712 to update the client grant space info.
715 Frequency : when client reconnect to OST
717 Description: Grant Leak when osc do resend and replay bulk write
718 Details : When osc reconnect to OST, OST(filter)should clear grant info of
719 bulk write request, because the grant info while be sync between
720 OSC and OST when reconnect, and we should ignore the grant info
721 these of resend/replay write req.
726 Description: Grant space more than avaiable left space sometimes.
727 Details : When then OST is about to be full, if two bulk writing from
728 different clients came to OST. Accord the avaliable space of the
729 OST, the first req should be permitted, and the second one
730 should be denied by ENOSPC. But if the seconde arrived before
731 the first one is commited. The OST might wrongly permit second
732 writing, which will cause grant space > avaiable space.
735 Frequency : when client is evicted
737 Description: Grant might be wrongly erased when osc is evicted by OST
738 Details : when the import is evicted by server, it will fork another
739 thread ptlrpc_invalidate_import_thread to invalidate the
740 import, where the grant will be set to 0. While the original
741 thread will update the grant it got when connecting. So if
742 the former happened latter, the grant will be wrongly errased
743 because of this race.
748 Description: Checking Stale with correct fid
749 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
750 to check whether it is stale, because sometimes, we want the
751 enqueue happened anyway, and op_data.fid2 will not be initialized.
753 Severity : enhancement
755 Description: update patchless client
756 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
759 Frequency : only with 2.4 kernel
761 Description: random memory corruption
762 Details : size of struct ll_inode_info is to big for union inode.u and this
763 can be cause of random memory corruption.
768 Description: Memory leak in recovery
769 Details : Lov_mds_md was not free in an error handler in mds_create_object.
770 It should also check obd_fail before fsfilt_start, otherwise if
771 fsfilt_start return -EROFS,(failover mds during mds recovery).
772 then the req will return with repmsg->transno = 0 and rc = EROFS.
773 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
774 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
775 be freed no matter whether fsfilt_commit success or not.
778 Frequency : only with huge count clients
780 Description: Prevents from taking the superblock lock in llap_from_page for
782 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
783 for llap_from_page call in ll_removepage prevents from taking the
784 superblock lock for a soon died page.
789 Description: Not check open intent error before release open handle
790 Details : in some rare cases, the open intent error is not checked before
791 release open handle, which may cause
792 ASSERTION(open_req->rq_transno != 0), because it tries to release
793 the failed open handle.
798 Description: Set cat log bitmap only after create log success.
799 Details : in some rare cases, the cat log bitmap is set too early. and it
800 should be set only after create log success.
804 Description: Accessing a block bevice can re-enable I/O when Lustre is
805 tearing down a device.
806 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
811 Description: service threads may hog cpus when there are a lot of requests
813 Details : Insert cond_resched to give other threads a chance to use some of
819 Description: the cat log was not initialized in recovery
820 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
821 unlink log on mds will not be initialized until mds post
822 recovery. And also in mds post recovery, the unlink log will
823 initialization will be done asynchronausly, so there will be race
824 between add unlink log and unlink log initialization.
828 Description: brw_stats were being printed incorrectly
829 Details : brw_stats were being printed as log2 but all of them were not
830 recorded as log2. Also remove some code duplication arising from
831 filter_tally_{read,write}.
835 Frequency : rare, only in recovery.
836 Description: ASSERTION(req->rq_type != LI_POISON) failed
837 Details : imp_lock should be held while iterating over imp_sending_list for
838 prevent destroy request after get timeout in ptlrpc_queue_wait.
842 Description: replay-single.sh test 52 fails
843 Details : A lock's skiplist need to be cleanup when it being unlinked
844 from its resource list.
848 Description: Short directio read returns full requested size rather than
850 Details : Direct I/O operations should return actual amount of bytes
851 transferred rather than requested size.
853 Severity : enhancement
855 Description: metadata RPC reduction (e.g. for rm performance)
856 Details : decrease the amount of synchronous RPC between clients and servers
857 by canceling conflicing lock before the operation on the client side
858 and packing thier handles into the main operation RPC to server.
860 Severity : enhancement
862 Description: Async OSC create to avoid the blocking unnecessarily.
863 Details : If a OST has no remain object, system will block on the creating
864 when need to create a new object on this OST. Now, ways use
865 pre-created objects when available, instead of blocking on an
866 empty osc while others are not empty. If we must block, we block
867 for the shortest possible period of time.
869 --------------------------------------------------------------------------------
871 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
876 Frequency : on some architectures
878 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
879 Details : Uninitialized flags sometimes cause configuration commands to
883 Frequency : patchless clients only
885 Description: missing __iget() symbol export
886 Details : The __iget() symbol export is missing. To avoid the need for
887 this on patchless clients the deathrow inode reaper is turned
888 off, and we depend on the VM to clean up old inodes. This
889 dependency was during via the fix for bug 12181.
891 --------------------------------------------------------------------------------
893 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
895 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
896 INTEROPERATE with older versions automatically. In many cases a
897 special upgrade step is needed. Please read the
898 user documentation before upgrading any part of a live system.
899 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
900 version will not interoperate with 1.6 betas before beta5 (1.5.95).
901 * WARNING: Lustre configuration and startup changes are required with
902 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
908 Severity : enhancement
910 Description: MountConf
911 Details : Lustre configuration is now managed via mkfs and mount
912 commands instead of lmc and lconf. New obd types (MGS, MGC)
913 are added for dynamic configuration management. See
914 https://mail.clusterfs.com/wikis/lustre/MountConf for
917 Severity : enhancement
919 Description: dynamic OST addition
920 Details : OSTs can now be added to a live filesystem
922 Severity : enhancement
924 Description: startup order invariance
925 Details : MDTs and OSTs can be started in any order. Clients only
926 require the MDT to complete startup.
928 Severity : enhancement
930 Description: parallel, asynchronous orphan cleanup
931 Details : orphan cleanup is now performed in separate threads for each
932 OST, allowing parallel non-blocking operation.
934 Severity : enhancement
936 Description: optimized stripe assignment
937 Details : stripe assignments are now made based on ost space available,
938 ost previous usage, and OSS previous usage, in order to try
939 to optimize storage space and networking resources.
941 Severity : enhancement
943 Description: Permanently set tunables
944 Details : All writable /proc/fs/lustre tunables can now be permanently
945 set on a per-server basis, at mkfs time or on a live
948 Severity : enhancement
950 Description: Lustre message v2
951 Details : Add lustre message format v2.
953 Severity : enhancement
955 Description: client OST exclusion list
956 Details : Clients can be started with a list of OSTs that should be
957 declared "inactive" for known non-responsive OSTs.
961 Description: SPEC SFS validation failure on NFS v2 over lustre.
962 Details : Changes the blocksize for regular files to be 2x RPC size,
963 and not depend on stripe size.
965 Severity : enhancement
967 Description: Multiple MD RPCs in flight.
968 Details : Further unserialise some read-only MDS RPCs - learn about intents.
969 To avoid overly-overloading MDS, introduce a limit on number of
970 MDS RPCs in flight for a single client and add /proc controls
971 to adjust this limit.
973 Severity : enhancement
975 Description: client read/write statistics
976 Details : Add client read/write call usage stats for performance
977 analysis of user processes.
978 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
979 file access. extents_stats shows chunk size distribution.
980 extents_stats_per_process show chunk size distribution per
983 Severity : enhancement
985 Description: mds statistics
986 Details : Add detailed mds operations statistics in
987 /proc/fs/lustre/mds/*/stats.
991 Description: Failure of copying files with lustre special EAs.
992 Details : Client side always return success for setxattr call for lustre
993 special xattr (currently only "trusted.lov").
998 Description: Refcount LNET uuids
999 Details : The global LNET uuid list grew linearly with every startup;
1000 refcount repeated list entries instead of always adding to
1003 Severity : enhancement
1005 Description: Dynamic service threads
1006 Details : Within a small range, start extra service threads
1007 automatically when the request queue builds up.
1010 Frequency : mixed-endian client/server environments
1012 Description: mixed-endian crashes
1013 Details : The new msg_v2 system had some failures in mixed-endian
1016 Severity : enhancement
1018 Description: Easy OST removal
1019 Details : OSTs can be permanently deactivated with e.g. 'lctl
1020 conf_param lustre-OST0001.osc.active=0'
1022 Severity : enhancement
1024 Description: MGS proc entries
1025 Details : Added basic proc entries for the MGS showing what filesystems
1028 Severity : enhancement
1030 Description: provide MGS failover
1031 Details : Added config lock reacquisition after MGS server failover.
1033 Severity : enhancement
1035 Description: add Linux 2.4 support
1036 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1040 Description: a large application tries to do I/O to the same resource and dies
1041 in the middle of it.
1042 Details : Check the req->rq_arrival time after the call to
1043 ost_brw_lock_get(), but before we do anything about
1044 processing it & sending the BULK transfer request. This
1045 should help move old stale pending locks off the queue as
1046 quickly as obd_timeout.
1049 Frequency : when an incorrect nid is specified during startup
1051 Description: ptlrpc connect to non-existant node causes kernel crash
1052 Details : LNET can't be re-entered from an event callback, which
1053 happened when we expire a message after the export has been
1054 cleaned up. Instead, hand the zombie cleanup off to another
1057 Severity : enhancement
1059 Description: plain/inodebits lock performance improvement
1060 Details : Grouping plain/inodebits in granted list by their request modes
1061 and bits policy, thus improving the performance of search through
1065 Frequency : only if OST filesystem is corrupted
1067 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1068 Details : for a short time RPCs with bulk IO are in the replay list,
1069 but replay of bulk IOs is unimplemented. If the OST filesystem
1070 is corrupted due to disk cache incoherency and then replay is
1071 started it is possible to trip an assertion. Avoid putting
1072 committed RPCs into the replay list at all to avoid this issue.
1075 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1077 Description: System hang on startup
1078 Details : This bug allowed the liblustre (e.g. catamount) client to
1079 return to the app before handling all startup RPCs. This
1080 could leave the node unresponsive to lustre network traffic
1081 and manifested as a server ptllnd timeout.
1083 Severity : enhancement
1085 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1086 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1087 boolean to control whether to print peer debug info when a
1088 client's RPC times out.
1091 Frequency : only for kernels with patches from Lustre below 1.4.3
1093 Description: Remove old rdonly API
1094 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1097 Frequency : only for devices with external journals
1099 Description: Set external device read-only also
1100 Details : During a commanded failover stop, we set the disk device
1101 read-only while the server shuts down. We now also set any
1102 external journal device read-only at the same time.
1105 Frequency : when upgrading from 1.4 while trying to change parameters
1107 Description: The wrong (new) MDC name was used when setting parameters for
1108 upgraded MDT's. Also allows changing of OSC (and MDC)
1109 parameters if --writeconf is specified at tunefs upgrade time.
1112 Frequency : when setting specific ost indicies
1114 Description: QOS code breaks on skipped indicies
1115 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1116 created with --index need not be sequential.
1118 Severity : enhancement
1120 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1121 Details : The uninit_groups feature works in conjunction with the kernel
1122 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1123 up the pass1 processing of e2fsck. This is a read-only feature
1124 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1125 mount filesystems that have had this feature enabled.
1127 Severity : enhancement
1129 Description: Improve multi-block allocation algorithm to avoid fragmentation
1130 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1131 allocation locality and avoid filesystem fragmentation.
1133 ------------------------------------------------------------------------------
1135 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1137 * Support for kernels:
1138 2.4.21-47.0.1.EL (RHEL 3)
1139 2.6.5-7.283 (SLES 9)
1140 2.6.9-42.0.10.EL (RHEL 4)
1141 2.6.12.6 vanilla (kernel.org)
1142 2.6.16.27-0.9 (SLES 10)
1143 * Recommended e2fsprogs version: 1.39.cfs5
1145 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1149 Frequency : occasional, depends on client load and configuration
1150 Bugzilla : 12181, 12203
1151 Description: data loss for recently-modified files
1153 Details : In some cases it is possible that recently written or created
1154 files may not be written to disk in a timely manner (this should
1155 normally be within 30s unless client IO load is very high).
1156 The problem appears as zero-length files or files that are a
1157 multiple of 1MB in size after a client crash or client eviction
1158 that are missing data at the end of the file.
1160 This problem is more likely to be hit on clients where files are
1161 repeatedly created and unlinked in the same directory, clients
1162 have a large amount of RAM, have many CPUs, the filesystem has
1163 many OSTs, the clients are rebooted frequently, and/or the files
1164 are not accessed by other nodes after being written.
1166 The presence of the problem can be detected by looking at
1167 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1168 smaller than the second (nr_unused) then dirty files will not
1169 be flushed automatically to disk. "sync; sleep 10" should be
1170 run several times on the node before unmounting it to update
1171 Lustre (this is also safe to run on nodes without this problem).
1173 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1174 that can cause this same problem, so customers using that kernel
1175 also need to update the kernel in addition to Lustre. In order
1176 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1178 It is normal that files written just before a client crash (less
1179 than 30s) may not yet have been flushed to disk, even for local
1183 Frequency : frequent on thin XT3 nodes
1185 Description: UUID collision on thin XT3 Linux nodes
1186 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1187 randomly, since we relied on an insufficiently-seeded PRNG.
1192 Description: OSS hangs after "All ost request buffers busy"
1193 Details : A deadlock between quota and journal operations caused OSS
1194 hangs after printing "All ost request buffers busy."
1197 Frequency : always on liblustre builds
1199 Description: Cleanup compiler warnings on liblustre
1202 Frequency : always on liblustre builds on XT3
1204 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1208 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1209 > 32000 subdirectories
1210 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1211 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1212 subdirectory count crosses 32000. This will aid e2fsck to
1213 correctly handle more than 32000 subdirectories.
1216 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1218 Description: System hang on startup
1219 Details : This bug allowed the liblustre (e.g. catamount) client to
1220 return to the app before handling all startup RPCs. This
1221 could leave the node unresponsive to lustre network traffic
1222 and manifested as a server ptllnd timeout.
1224 Severity : enhancement
1226 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1227 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1228 boolean to control whether to print peer debug info when a
1229 client's RPC times out.
1234 Description: make O_SYNC working on 2.6 kernels
1235 Details : 2.6 kernels use different method for mark pages for write,
1236 so need add a code to lustre for O_SYNC work.
1241 Description: Failure to close file and release space on NFS
1242 Details : Put inode details into lock acquired in ll_intent_file_open.
1243 Use mdc_intent_lock in ll_intent_open to properly
1244 detect all kind of errors unhandled by mdc_enqueue.
1249 Description: proc file read during shutdown sometimes raced obd removal,
1251 Details : Add lock to prevent obd access after proc file removal.
1254 Frequency : Only for files larger than 4GB on 32-bit clients.
1256 Description: improperly doing page alignment of locks
1257 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1258 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1263 Description: RPCs being resent when they shouldn't be
1264 Details : Some RPCs that should not be resent are being resent. This
1265 can cause inconsistencies in the RPC state machine. Do not
1266 resend such requests.
1269 Frequency : rare, only with NFS export
1271 Description: Crash on NFS re-export node
1272 Details : under very unusual load conditions an assertion is hit in
1273 ll_intent_file_open()
1276 Frequency : only if OST filesystem is corrupted
1278 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1279 Details : for a short time RPCs with bulk IO are in the replay list,
1280 but replay of bulk IOs is unimplemented. If the OST filesystem
1281 is corrupted due to disk cache incoherency and then replay is
1282 started it is possible to trip an assertion. Avoid putting
1283 committed RPCs into the replay list at all to avoid this issue.
1288 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1289 Details : Large single O_DIRECT read and write calls can fail to allocate
1290 a sufficiently large buffer to process the request. In case of
1291 allocation failure the allocation is retried with a smaller
1292 buffer and broken into smaller requests.
1294 Severity : enhancement
1296 Description: Add -o localflock option to simulate old noflock behaviour.
1297 Details : This will achieve local-only flock/fcntl locks coherentness.
1302 Description: versioning check is incomplete
1303 Details : Checking the version difference of client vs. server, report
1304 error if the gap is too big.
1309 Description: add support PG_writeback bit
1310 Details : add support for PG_writeback bit for Lustre, for more carefull
1311 work with page cache in 2.6 kernel. This also fix some deadlocks
1312 and remove hack for work O_SYNC with 2.6 kernel.
1314 Severity : enhancement
1316 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1317 Details : The uninit_groups feature works in conjunction with the kernel
1318 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1319 up the pass1 processing of e2fsck. This is a read-only feature
1320 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1321 mount filesystems that have had this feature enabled.
1323 Severity : enhancement
1325 Description: Improve multi-block allocation algorithm to avoid fragmentation
1326 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1327 allocation locality and avoid filesystem fragmentation.
1329 ------------------------------------------------------------------------------
1331 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1333 * Support for kernels:
1334 2.6.9-42.0.3.EL (RHEL 4)
1335 2.6.5-7.276 (SLES 9)
1336 2.4.21-47.0.1.EL (RHEL 3)
1337 2.6.12.6 vanilla (kernel.org)
1338 2.6.16.21-0.8 (SLES10)
1339 * Recommended e2fsprogs version: 1.39.cfs2-0
1341 * The backwards-compatible /proc/sys/portals symlink has been removed
1342 in this release. Before upgrading, please ensure that you change
1343 any configuration scripts or /etc/sysctl.conf files that access
1344 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1345 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1346 in advance of the upgrade on any system running Lustre 1.4.6 or
1347 newer, since /proc/sys/lnet was added in that version.
1348 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1352 Frequency : only when quota is used
1354 Description: avoid scanning export list for quota master
1355 Details : Change the algorithms to avoid scanning export list in order
1356 to improve the efficiency.
1359 Frequency : MDS failover only, very rarely
1361 Description: "went back in time" messages on mds failover
1362 Details : The greatest transno may be lost when the current operation
1363 finishes with an error (transno==0) and the client's last_rcvd
1364 record is over-written. Save the greatest transno in the
1365 mds_last_transno for this case.
1368 Frequency : always for specific kernels and striping counts
1370 Description: client may get "Matching packet too big" without ACL support
1371 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1372 when trying to access files in certain configurations. The
1373 clients should in fact be denied when mounting because they do
1374 not understand ACLs.
1377 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1379 Description: many clients connecting with IO in progress causes connect timeouts
1380 Details : Avoid synchronous journal commits to avoid delays caused by many
1381 clients connecting/disconnecting when bulk IO is in progress.
1382 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1383 OST_IO_PORTAL to avoid delays behind potentially many pending
1387 Frequency : occasionally with multiple writers to a single file
1389 Description: shared writes to file may result in wrong size reported by stat()
1390 Details : Allow growing of kms when extent lock is cancelled
1393 Frequency : always with random mmap IO to multi-striped file
1395 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1396 Details : Only if the hole is at the end of OST object so that kms is too
1397 small. Fix is to increase kms accordingly in ll_nopage.
1400 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1402 Description: writes to a missing object would leak memory on the OST
1403 Details : If there is an inconsistency between the MDS and OST filesystems,
1404 such that the MDS references an object that doesn't exist, writes
1405 to that object will leak memory due to incorrect cleanup in the
1406 error handling path, eventually running out of memory on the OST.
1411 Description: Creating too long symlink causes lustre errors
1412 Details : Check symlink and name lengths before sending requests to MDS.
1415 Frequency : only if flock is enabled (not on by default)
1417 Description: posix locks not released on fd closure on 2.6.9+
1418 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1419 kernels, this caused such locks not to be released on fd close and
1420 then assertions on fs unmount about still used locks.
1423 Frequency : MDS failover only, very rarely
1425 Description: clients may get ASSERTION(granted_lock != NULL)
1426 Details : When request was taking a long time, and a client was resending
1427 a getattr by name lock request. The were multiple lock requests
1428 with the same client lock handle and
1429 mds_getattr_name->fixup_handle_for_resent_request found one of the
1430 lock handles but later failed with ASSERTION(granted_lock != NULL).
1435 Description: handle->h_buffer_credits > 0, assertion failure
1436 Details : h_buffer_credits is zero after truncate, causing assertion
1437 failure. This patch extends the transaction or creates a new
1441 Frequency : NFS re-export or patchless client
1442 Bugzilla : 11179, 10796
1443 Description: Crash on NFS re-export node (__d_move)
1444 Details : We do not want to hash the dentry if we don't have a lock.
1445 But if this dentry is later used in d_move, we'd hit uninitialised
1446 list head d_hash, so we just do this to init d_hash field but
1447 leave dentry unhashed.
1450 Frequency : NFS re-export or patchless client
1452 Description: NFS exports has problem with symbolic link
1453 Details : lustre client didn't properly install dentry when re-exported
1454 to NFS or running patchless client.
1457 Frequency : NFS re-export or patchless client
1459 Description: Various nfs/patchless fixes.
1460 Details : fixes reuse disconected alias for lookup process - this fixes
1461 warning "find_exported_dentry: npd != pd",
1462 fix permission error with open files at nfs.
1463 fix apply umask when do revalidate.
1466 Frequency : occasional
1468 Description: Crash on NFS re-export node
1469 Details : calling clear_page() on the wrong pointer triggered oops in
1470 generic_mapping_read().
1473 Frequency : rarely, using O_DIRECT IO
1475 Description: unaligned directio crashes client with LASSERT
1476 Details : check for unaligned buffers before trying any requests.
1479 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1481 Description: stale data returned from RAID cache
1482 Details : If only a small amount of IO is done to the RAID device before
1483 reading it again it is possible to get stale data from the RAID
1484 cache instead of reading it from disk.
1487 Frequency : always for sles10 kernel
1489 Description: sles10 support
1490 Details : ll_follow_link: compile fixes and using of nd_set_link
1491 under newer kernels.
1494 Frequency : depends on arch, kernel and compiler version, always on sles10
1497 Description: recursive or deep enough symlinks cause stack overflow
1498 Details : getting rid of large stack-allocated variable in
1502 Frequency : depends on hardware
1504 Description: lustre write performance loss in the SLES10 kernel
1505 Details : the performance loss is caused by using of write barriers in the
1506 ext3 code. The SLES10 kernel turns barrier support on by
1507 default. The fix is to undo that change for ldiskfs.
1509 ------------------------------------------------------------------------------
1511 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1513 * Support for kernels:
1514 2.6.9-42.0.3EL (RHEL 4)
1515 2.6.5-7.276 (SLES 9)
1516 2.4.21-47.0.1.EL (RHEL 3)
1517 2.6.12.6 vanilla (kernel.org)
1521 Frequency : quota enabled and large files being deleted
1523 Description: releasing more than 4GB of quota at once hangs OST
1524 Details : If a user deletes more than 4GB of files on a single OST it
1525 will cause the OST to spin in an infinite loop. Release
1526 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1531 Description: statfs data retrieved from /proc may be stale or zero
1532 Details : When reading per-device statfs data from /proc, in the
1533 {kbytes,files}_{total,free,avail} files, it may appear
1534 as zero or be out of date.
1537 Frequency : systems with MD RAID1 external journal devices
1539 Description: lconf's call to blkid is confused by RAID1 journal devices
1540 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1541 of returning all block devices that match the journal UUID.
1544 Frequency : always, for aggregate stripe size over 4GB
1546 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1547 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1548 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1551 Frequency : always if "lfs find" used on a local file/directory
1553 Description: "lfs find" segfaults if used on a local file/directory
1554 Details : The case where a directory component was not specified wasn't
1555 handled correctly. Handle this properly.
1558 Frequency : always on ppc64
1560 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1561 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1562 perform endianness conversion.
1565 Frequency : rarely (truncate to non-zero file size after write under load)
1566 Bugzilla : 10730, 10687
1567 Description: Files padded with zeros to next 4K multiple
1568 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1569 it is possible that files that are truncated to a non-zero size
1570 immediately after being written are filled with zero bytes beyond
1571 the truncated size. No file data is lost.
1573 Severity : enhancement
1575 Description: Allow recovery/failover for liblustre clients.
1576 Details : liblustre clients were unaware of failover configurations until
1579 Severity : enhancement
1581 Description: user file locks should fail when not mounting with flock option
1582 Details : Set up an error-returning stub in ll_file_operations.lock field
1583 to prevent incorrect behaviour when client is mounted without
1584 flock option. Also, set up properly f_op->flock field for
1588 Frequency : always on ia64
1590 Description: "lfs df" loops on printing out MDS statfs information
1591 Details : The obd_ioctl_data was not initialized and in some systems
1592 this caused a failure during the ioctl that did not return
1593 an error. Initialize the struct and return an error on failure.
1596 Frequency : SLES 9 only
1598 Description: Error of copying files with lustre special EAs as root
1599 Details : Client side always return success for setxattr call for lustre
1600 special xattr (currently only "trusted.lov").
1603 Frequency : rarely on clusters with both ia64+i386 clients
1605 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1606 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1607 clients are doing concurrent writes to the same file it is
1608 possible that the ia64 clients may LASSERT because the OST
1609 extent locks are not PAGE_SIZE aligned. Ensure that grown
1610 locks are always aligned on the request boundary.
1613 Frequency : specific use, occasional
1615 Description: Overwriting in use executable truncates on-disk binary image
1616 Details : If one node attempts to overwrite an executable in use by
1617 another node, we now correctly return ETXTBSY instead of
1618 truncating the file.
1620 Severity : enhancement
1622 Description: Async OSC create to avoid the blocking unnecessarily.
1623 Details : If a OST has no remain object, system will block on the creating
1624 when need to create a new object on this OST. Now, ways use
1625 pre-created objects when available, instead of blocking on an
1626 empty osc while others are not empty. If we must block, we block
1627 for the shortest possible period of time.
1632 Description: chmod on Lustre root is propagated to other clients
1633 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1634 invalid by the follow_mount time.
1637 Frequency : liblustre clients only
1639 Description: Race in 'instant cancel' lock handling could lead to such locks
1640 never to be granted in case of SMP MDS
1641 Details : Do not destroy not yet granted but cbpending locks in
1645 Frequency : replay/resend of open
1647 Description: non null lock assetion failure in mds_intent_policy
1648 Details : Trying to replay/resend lockless open requests resulted in
1649 mds_open() returning 0 with no lock. Now it sets a flag if
1650 a lock is going to be returned.
1652 Severity : enhancement
1654 Description: Checksum enhancements
1655 Details : New checksum enhancements allow for resending RPCs that failed
1658 Severity : enhancement
1660 Description: Tunables on number of dirty pages in cacche
1661 Details : Allow to set limit on number of dirty pages cached.
1666 Description: client crash on unmount - lock still has references
1667 Details : In some error handling cases it was possible to leak a lock
1668 reference on a client while accessing a file. This was not
1669 harmful to the client during operation, but would cause the
1670 client to crash when the filesystem is unmounted.
1673 Frequency : specific case, rare
1675 Description: ETXTBSY on mds though file not in use by client
1676 Details : ETXTBSY is no longer incorrectly returned when attempting to
1677 chmod or chown a directory that the user previously tried to
1678 execute or a currently-executing binary.
1681 Frequency : extremely rare except on liblustre-based clients
1683 Description: Lustre space not freed when files are deleted
1684 Details : Clean up open-unlinked files after client eviction. Previously
1685 the unlink was skipped and the files remained as orphans.
1690 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1691 Details : In some cases it was possible to send a blocking callback to
1692 a client doing a glimpse, even though that client didn't get
1693 a lock granted. When the glimpse lock is cancelled on the OST
1694 the freed lock is left on the waiting list and corrupted the list.
1697 Frequency : all core dumps
1699 Description: Broke core dumps to lustre
1700 Details : Negative dentry may be unhashed if parent does not have UPDATE
1701 lock, but some callers, e.g. do_coredump, expect dentry to be
1702 hashed after successful create, hash it in ll_create_it.
1704 ------------------------------------------------------------------------------
1706 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1708 * Support for kernels:
1709 2.6.9-42.0.2.EL (RHEL 4)
1710 2.6.5-7.276 (SLES 9)
1711 2.4.21-40.EL (RHEL 3)
1712 2.6.12.6 vanilla (kernel.org)
1716 Frequency : always on RHEL 3
1718 Description: Number of open files grows over time
1719 Details : The number of open files grows over time, whether or not
1720 Lustre is started. This was due to a filp leak introduced
1721 by one of our kernel patches.
1723 ------------------------------------------------------------------------------
1725 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1727 * Support for kernels:
1728 2.6.9-42.EL (RHEL 4)
1729 2.6.5-7.276 (SLES 9)
1730 2.4.21-40.EL (RHEL 3)
1731 2.6.12.6 vanilla (kernel.org)
1736 Bugzilla : 5719, 9635, 9792, 9684
1737 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1738 Details : If a server is under heavy load and cannot reply to new
1739 connection requests before the client resends the (re)connect,
1740 the connection handling code can behave badly if two service
1741 threads are concurrently handing separate (re)connections from
1742 the same client. Add better locking to the connection handling
1743 code, and ensure that only a single connection will be processed
1744 for a given client UUID, even if the lock is dropped.
1746 Severity : enhancement
1748 Description: add TCP zero-copy support to kernel
1749 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1750 sends if the hardware supports scatter-gather and checksumming.
1751 This allows socklnd to do client-write and server-read more
1752 efficiently and reduce CPU utilization from skbuf copying.
1755 Frequency : only if NFS exporting from client
1757 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1758 Details : If setting attributes on a file created under NFS that had
1759 never been opened it would be possible to oops the client
1760 if the file had no objects.
1763 Frequency : always for liblustre
1765 Description: liblustre client does MDS+OSTs setattr RPC for each write
1766 Details : When doing a write from a liblustre client, the client
1767 incorrectly issued an RPC to the MDS and each OST the file was
1768 striped over in order to update the timestamps. When writing
1769 with small chunks and many clients this could overwhelm the MDS
1770 with RPCs. In all cases it would slow down the write because
1771 these RPCs are unnecessary.
1773 Severity : enhancement
1775 Description: allow number of MDS service threads to be changed at module load
1776 Details : It is now possible to change the number of MDS service threads
1777 running. Adding "options mds mds_num_threads={N}" to the MDS's
1778 /etc/modprobe.conf will set the number of threads for the next
1779 time Lustre is restarted (assuming the "mds" module is also
1780 reloaded at that time). The default number of threads will
1781 stay the same, 32 for most systems.
1786 Description: OST crash if filesystem is unformatted or corrupt
1787 Details : If an OST is started on a device that has never been formatted
1788 or if the filesystem is corrupt and cannot even mount then the
1789 error handling cleanup routines would dereference a NULL pointer.
1794 Description: NULL pointer deref in llap_from_page.
1795 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1796 incorrect) mapping if the page was truncated/reclaimed while it was
1797 searched for. Check for this condition and skip such pages when
1798 doing readahead. Introduce extra check to llap_from_page() to
1799 verify page->mapping->host is non-NULL (so page is not anonymous).
1802 Frequency : Sometimes when using sys_sendfile
1804 Description: "page not covered by a lock" warnings from ll_readpage
1805 Details : sendfile called ll_readpage without right page locks present.
1806 Now we introduced ll_file_sendfile that does necessary locking
1807 around call to generic_file_sendfile() much like we do in
1811 Frequency : with certain MDS communication failures at client mount time
1813 Description: NULL pointer deref after failed client mount
1814 Details : a client connection request may delayed by the network layer
1815 and not be sent until after the PTLRPC layer has timed out the
1816 request. If the client fails the mount immediately it will try
1817 to clean up before the network times out the request. Add a
1818 reference from the request import to the obd device and delay
1819 the cleanup until the network drops the request.
1822 Frequency : occasionally during client (re)connect
1824 Description: assertion failure during client (re)connect
1825 Details : processing a client connection request may be delayed by the
1826 client or server longer than the client connect timeout. This
1827 causes the client to resend the connection request. If the
1828 original connection request is replied in this interval, the
1829 client may trip an assertion failure in ptlrpc_connect_interpret()
1830 which thought it would be the only running connect process.
1833 Frequency : only with obd_echo servers and clients that are rebooted
1835 Description: kernel BUG accessing uninitialized data structure
1836 Details : When running an obd_echo server it did not start the ping_evictor
1837 thread, and when a client was evicted an uninitialized data
1838 structure was accessed. Start the ping_evictor in the RPC
1839 service startup instead of the OBD startup.
1841 Severity : enhancement
1842 Bugzilla : 10193 (patchless)
1843 Description: Remove dependency on various unexported kernel interfaces.
1844 Details : No longer need reparent_to_init, exit_mm, exit_files,
1845 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1848 Frequency : rare (only users of deprecated and unsupported LDAP config)
1850 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1851 Details : LDAP apparently contains 'lustreName' attributes instead of
1852 'name'. A simple remapping of the name is sufficient.
1855 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1857 Description: waiting_locks_callback trips kernel BUG if client is evicted
1858 Details : Running with the dump_on_timeout debug flag turned on makes
1859 it possible that the waiting_locks_callback() can try to dump
1860 the Lustre kernel debug logs from an interrupt handler. Defer
1861 this log dumping to the expired_lock_main() thread.
1863 Severity : enhancement
1865 Description: Support NFS exporting on 2.6 kernels.
1866 Details : Implement non-rawops metadata methods for NFS server to use without
1867 changing NFS server code.
1870 Frequency : very rare (synthetic metadata workload only)
1872 Description: two racing renames might cause an MDS thread to deadlock
1873 Details : Running the "racer" program may cause one MDS thread to rename
1874 a file from being the source of a rename to being the target of
1875 a rename at exactly the same time that another thread is doing
1876 so, and the second thread has already enqueued these locks after
1877 doing a lookup of the target and is trying to relock them in
1878 order. Ensure that we don't try to re-lock the same resource.
1881 Frequency : only very large systems with liblustre clients
1883 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1884 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1885 all OSTs in parallel. This allows the eviction of stale liblustre
1886 clients to proceed much faster than if they were done in series,
1887 and also offers similar improvements for other set_info RPCs.
1892 Description: excessive CPU usage during initial read phase on client
1893 Details : During the initial read phase on a client, it would agressively
1894 retry readahead on the file, consuming too much CPU and impacting
1895 performance (since 1.4.5.8). Improve the readahead algorithm
1896 to avoid this, and also improve some other common cases (read
1897 of small files in particular, where "small" is files smaller than
1898 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1903 Description: MDS crash when receiving packet with unknown intent.
1904 Details : Do not LBUG in unknown intent case, just return -EFAULT
1906 Severity : enhancement
1907 Bugzilla : 9293, 9385
1908 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1909 Details : Do not serialize getattr (non-intent version) and statfs.
1912 Frequency : occasional, when OST network is overloaded/intermittent
1914 Description: client evicted by OST after bulk IO timeout
1915 Details : If a client sends a bulk IO request (read or write) the OST
1916 may evict the client if it is unresposive to its data GET/PUT
1917 request. This is incorrect if the network is overloaded (takes
1918 too long to transfer the RPC data) or dropped the OST GET/PUT
1919 request. There is no need to evict the client at all, since
1920 the pinger and/or lock callbacks will handle this, and the
1921 client can restart the bulk request.
1924 Frequency : Always when mmapping file with no objects
1926 Description: client crashes when mmapping file with no objects
1927 Details : Check that we actually have objects in a file before doing any
1928 operations on objects in ll_vm_open, ll_vm_close and
1934 Description: Request leak when working with deleted CWD
1935 Details : Introduce advanced request refcount tracking for requests
1936 referenced from lustre intent.
1938 Severity : Enhancement
1940 Description: Cache open file handles on client.
1941 Details : MDS now will return special lock along with openhandle, if
1942 requested and client is allowed to hold openhandle, even if unused,
1943 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1944 closing files for every read/write openration.
1946 Severity : Enhancement
1948 Description: Cache open negative dentries on client when possible.
1949 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1950 negative dentries on lock revocation.
1955 Description: Remounting a client read-only wasn't possible with a zconf mount
1956 Details : It wasn't possible to remount a client read-only with llmount.
1958 Severity : enhancement
1959 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1960 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1961 MPICH is included to add Lustre-specific ADIO interfaces.
1962 This is based closely on the UFS ADIO layer and only differs
1963 in file creation, in order to allow the OST striping to be set.
1964 This is user-contributed code and not supported by CFS.
1969 Description: extended inode attributes (immutable, append-only) work improperly
1970 when 2.4 and 2.6 kernels are used on client/server or vice versa
1971 Details : Introduce kernel-independent values for these flags.
1973 Severity : enhancement
1976 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1977 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1978 granularity. Now, Fractional MB granularity is be supported,
1979 this is very useful for low memory system.
1981 Severity : enhancement
1983 Description: Getattr by fid
1984 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1985 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1986 have locks and avoids getattr rpc requests for it on every stat.
1989 Frequency : Always, for filesystems larger than 2TB
1991 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1992 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1993 filesystems larger than 2TB to be mounted with mballoc enabled.
1996 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1998 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1999 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2000 filesystems larger than 2TB is mounted then the kernel will
2001 silently corrupt the start of the filesystem. CONFIG_LBD is
2002 enabled for all CFS-supported kernels, but the possibility of
2003 this happening with a modified kernel config exists.
2005 Severity : enhancement
2007 Description: add client O_DIRECT support for 2.6 kernels
2008 Details : It is now possible to do O_DIRECT reads and writes to files
2009 in the Lustre client mountpoint on 2.6 kernel clients.
2011 Severity : enhancement
2013 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2014 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2015 parallel, not waiting for response from every OST before sending
2016 a rpc to the next OST.
2021 Description: setattr vs write race when updating file timestamps
2022 Details : Client processes that update a file timestamp into the past
2023 right after writing to the file (e.g. tar) it is possible that
2024 the updated file modification time can be reset to the current
2025 time due to a race between processing the setattr and write RPC.
2027 Severity : enhancement
2029 Description: Bring 'lfs find' closer in line with regular Linux find.
2030 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2031 -print0 options and obtains all the needed info through the lustre
2034 Severity : enhancement
2036 Description: support up to 1024 configured devices on one node
2037 Details : change obd_dev array from statically allocated to dynamically
2038 allocated structs as they are first used to reduce memory usage
2043 Description: Flush dirty partially truncated pages during truncate
2044 Details : Immediatelly flush partially truncated pages in filter_setattr,
2045 this way we completely avoid having any pages in page cache on OST
2046 and can retire ugly workarounds during writes to flush such pages.
2051 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2052 Details : protect inode from truncation within vfs_unlink() context
2053 just take a reference before calling vfs_unlink() and release it
2054 when parent's i_sem is free.
2059 Description: last_id value checked outside lock on OST caused LASSERT failure
2060 Details : If there were multiple MDS->OST object precreate requests in
2061 flight, it was possible that the OST's last object id was checked
2062 outside a lock and incorrectly tripped an assertion. Move checks
2063 inside locks, and discard old precreate requests.
2066 Frequency : always, if extents are used on OSTs
2068 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2069 Details : OSTs using the extents format would not zero the high 16 bits of
2070 the index physical block number. This is not a problem for any
2071 OST filesystems smaller than 16TB, and no kernels support ext3
2072 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2073 new/modified files) and can be fixed for existing filesystems
2074 with e2fsprogs-1.39-cfs1.
2079 Description: import connection selection may be incorrect if timer wraps
2080 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2081 import connections to be ignored if the 32-bit jiffies counter
2082 wraps. Use a 64-bit jiffies counter.
2085 Frequency : very large clusters immediately after boot
2087 Description: LNET request buffers exhausted under heavy short-term load
2088 Details : If a large number of client requests are generated on a service
2089 that has previously never seen so many requests it is possible
2090 that the request buffer growth cannot keep up with the spike in
2091 demand. Instead of dropping incoming requests, they are held in
2092 the LND until the RPC service can accept more requests.
2095 Frequency : Sometimes during replay
2097 Description: Assertion failure in ll_local_open after replay.
2098 Details : If replay happened on an open request reply before we were able
2099 to set replay handler, reply will become not swabbed tripping the
2100 assertion in ll_local_open. Now we set the handler right after
2101 recognising of open request
2104 Frequency : very rare
2106 Description: kernel reports "badness in vsnprintf"
2107 Details : Reading from the "recovery_status" /proc file in small chunks
2108 may cause a negative length in lprocfs_obd_rd_recovery_status()
2109 call to vsnprintf() (which is otherwise harmless). Exit early
2110 if there is no more space in the output buffer.
2112 Severity : enhancement
2114 Description: clear OBD RPC statistics by writing to them
2115 Details : It is now possible to clear the OBD RPC statistics by writing
2116 to the "stats" file.
2121 Description: Client mtime is not the same on different clients after utimes
2122 Details : In some cases, the client was using the utimes() syscall on
2123 a file cached on another node. The clients now validate the
2124 ctime from the MDS + OSTs to determine which one is right.
2129 Description: Inability to activate failout mode
2130 Details : lconf script incorrectly assumed that in pythong string's numeric
2131 value is used in comparisons.
2134 Frequency : always with multiple stripes per file
2136 Description: Inefficient object allocation for mutli-stripe files
2137 Details : When selecting which OSTs to stripe files over, for files with
2138 a stripe count that divides evenly into the number of OSTs,
2139 the MDS is always picking the same starting OST for each file.
2140 Return the OST selection heuristic to the original design.
2145 Description: mount failures may take full timeout to return an error
2146 Details : Under some heavy load conditions it is possible that a
2147 failed mount can wait for the full obd_timeout interval,
2148 possibly several minutes, before reporting an error.
2149 Instead return an error as soon as the status is known.
2151 Frequency : quota enabled and large files being deleted
2153 Description: releasing more than 4GB of quota at once hangs OST
2154 Details : If a user deletes more than 4GB of files on a single OST it
2155 will cause the OST to spin in an infinite loop. Release
2156 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2161 Description: statfs data retrieved from /proc may be stale or zero
2162 Details : When reading per-device statfs data from /proc, in the
2163 {kbytes,files}_{total,free,avail} files, it may appear
2164 as zero or be out of date.
2167 Frequency : systems with MD RAID1 external journal devices
2169 Description: lconf's call to blkid is confused by RAID1 journal devices
2170 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2171 of returning all block devices that match the journal UUID.
2174 Frequency : always, for aggregate stripe size over 4GB
2176 Description: assertion fails when trying to use 4GB stripe size
2177 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2178 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2181 Frequency : always on ppc64
2183 Description: the first write on an ext3 filesystem with mballoc got stuck
2184 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2185 perform endianness conversion.
2187 ------------------------------------------------------------------------------
2189 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2191 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2192 INTEROPERATE with older versions automatically. Please read the
2193 user documentation before upgrading any part of a live system.
2194 * WARNING: Lustre networking configuration changes are required with
2195 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2198 * Support for kernels:
2199 2.6.9-22.0.2.EL (RHEL 4)
2200 2.6.5-7.244 (SLES 9)
2201 2.6.12.6 vanilla (kernel.org)
2204 Severity : enhancement
2205 Bugzilla : 7981/8208
2206 Description: Introduced Lustre Networking (LNET)
2207 Details : LNET is new networking infrastructure for Lustre, it includes
2208 a reorganized network configuration mode (see the user
2209 documentation for full details) as well as support for routing
2210 between different network fabrics. Lustre Networking Devices
2211 (LNDs) for the supported network fabrics have also been
2212 created for this new infrastructure.
2214 Severity : enhancement
2215 Description: Introduced Access control lists
2216 Details : clients can set ACLs on files and directories in order to have
2217 more fine-grained permissions than the standard Unix UGO+RWX.
2218 The MDS must be started with the "-o acl" mount option.
2220 Severity : enhancement
2221 Description: Introduced filesystem quotas
2222 Details : Administrators may now establish per-user quotas on the
2225 Severity : enhancement
2227 Description: Configuration change for the XT3
2228 The PTLLND is now used to run Lustre over Portals on the XT3
2229 The configure option(s) --with-cray-portals are no longer used.
2230 Rather --with-portals=<path-to-portals-includes> is used to
2231 enable building on the XT3. In addition to enable XT3 specific
2232 features the option --enable-cray-xt3 must be used.
2237 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2238 Details : A race between allocating a new llcd and re-getting the llcd_lock
2239 allowed another thread to grab newly-allocated llcd.
2241 Severity : enhancement
2243 Description: 2.6 OST async journal commit and locking fix to improve performance
2244 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2245 2.6 kernels are now async as they already were in 2.4 kernels so
2246 that they can commit concurrently with the network bulk transfer.
2247 For block-allocated files the filter allocation semaphore is held
2248 to avoid filesystem fragmentation during allocation. BKL lock
2249 removed for 2.6 xattr operations where it is no longer needed.
2254 Description: lconf incorrectly determined whether two IP networks could talk
2255 Details : In some more complicated routing and multiple-network
2256 configurations, lconf will avoid trying to make a network
2257 connection to a disjoint part of the IP space. It was doing the
2258 math incorrectly for one set of cases.
2263 Description: Fix for potential infinite loop processing records in an llog.
2264 Details : If an llog record is corrupted/zeroed, it is possible to loop
2265 forever in llog_process(). Validate the llog record length
2266 and skip the remainder of the block on error.
2269 Frequency : occasional (liblustre only)
2271 Description: liblustre could not open files whose last component is a symlink
2272 Details : sysio_path_walk() would incorrectly pass the open intent to
2273 intermediate path components.
2276 Frequency : rare (liblustre only with non-standard tuning)
2277 Bugzilla : 7201 (7350)
2278 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2279 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2280 after it has been released, only do this for referenced locks
2282 Severity : enhancement
2284 Description: specifying an (invalid) directory default stripe_size of -1
2285 would reset the directory default striping
2286 Details : stripe_size -1 was used internally to signal directory stripe
2287 removal, now use "all default" to signal dir stripe removal
2288 as a directory striping of "all default" is not useful
2291 Frequency : common for large clusters running liblustre clients
2293 Description: doing an ls when liblustre clients are running is slow
2294 Details : sending a glimpse AST to a liblustre client waits for every AST
2295 to time out, as liblustre clients will not respond. Since they
2296 cannot cache data we refresh the OST lock LVB from disk instead.
2298 Severity : enhancement
2300 Description: doing an ls at the same time as file IO can be slow
2301 Details : enqueue and other "small" requests can be blocked behind many
2302 large IO requests. Create a new OST IO portal for non-IO
2303 requests so they can be processed faster.
2306 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2308 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2309 created file with mode 000
2310 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2311 allow an ftruncate() as a non-root user to a file with mode 000.
2312 Lustre now respects this flag to disable mode checks when
2313 truncating a file owned by the user
2316 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2319 Description: Revoking locks from clients that went dead or catatonic might take
2321 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2322 cancellation of such locks instant on servers without waiting for
2323 any reply from clients. Clients drops these locks when cancel
2324 notification from server is received without replying.
2327 Frequency : liblustre-only, when liblustre client dies or becomes busy
2329 Description: Doing ls on Linux clients can take a long time with active
2331 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2332 granting such locks to it in the first place if possible. Locks
2333 are taken by proxy on the OST during the read or write and
2334 dropped immediately afterward. Add connect flags handling, do
2335 not grant locks to liblustre clients for glimpse ASTs.
2337 Severity : enhancement
2339 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2340 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2341 sequential reads and behaves badly with applications doing random
2342 reads. Improve it by reading ahead at least read region, and
2343 avoiding excessive large RPC for small reads.
2345 Severity : enhancement
2347 Description: Creating more than 1000 files for a single job may cause a load
2348 imbalance on the OSTs if there are also a large number of OSTs.
2349 Details : qos_prep_create() uses an OST index reseed value that is an
2350 even multiple of the number of available OSTs so that if the
2351 reseed happens in the middle of the object allocation it will
2352 still utilize the OSTs as uniformly as possible.
2357 Description: OST or MDS may oops in ping_evictor_main()
2358 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2359 but doesn't restart at beginning of obd_exports_timed list
2362 Severity : enhancement
2364 Description: improve by-nid export eviction on the MDS and OST
2365 Details : allow multiple exports with the same NID to be evicted at one
2366 time without re-searching the exports list.
2369 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2371 Description: MDS may oops in groups_free()
2372 Details : in rare race conditions a newly allocated group_info struct is
2373 freed again, and this can be NULL. The 2.4 compatibility code
2374 for groups_free() checked for a NULL pointer, but 2.6 did not.
2377 Frequency : common for liblustre clients doing little filesystem IO
2378 Bugzilla : 9352, 7313
2379 Description: server may evict liblustre clients accessing contended locks
2380 Details : if a client is granted a lock or receives a completion AST
2381 with a blocking AST already set it would not reply to the AST
2382 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2385 Frequency : lfs setstripe, only systems with more than 160 OSTs
2387 Description: unable to set striping with a starting offset beyond OST 160
2388 Details : llapi_create_file() incorrectly limited the starting stripe
2389 index to the maximum single-file stripe count.
2392 Frequency : LDAP users only
2394 Description: lconf did not handle in-kernel recovery with LDAP properly
2395 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2397 Severity : enhancement
2399 Description: bind OST threads to NUMA nodes to improve performance
2400 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2401 node and do their allocations there to localize memory access
2403 Severity : enhancement
2405 Description: llmount can determine client NID directly from Myrinet (GM)
2406 Details : the client NID code from gmnalnid was moved directly into
2407 llmount, removing the need to use this or specifying the
2408 client NID explicitly when mounting GM clients with zeroconf
2411 Frequency : if client is started with down MDS
2413 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2414 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2415 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2416 ptlrpc_import_delayed_req() because we didn't check if the
2417 request was interrupted, and we also didn't break out of the
2418 event loop if there was no timeout
2423 Description: data loss during non-page-aligned writes to a single file from
2424 both multiple nodes and multiple threads on one node at same time
2425 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2426 inconsistency led to false short-reads, that were cached and later
2427 used by ->prepare_write() to fill in partially written page,
2428 leading to data loss.
2431 Frequency : always, if lconf --abort_recovery used
2433 Description: lconf --abort_recovery fails with 'Operation not supported'
2434 Details : lconf was attempting to abort recovery on the MDT device and not
2437 Severity : enhancement
2439 Description: remove cleanup logs
2440 Details : replace lconf-generated cleanup logs with lustre internal
2441 cleanup routines. Eliminates the need for client-cleanup and
2444 Severity : enhancement
2446 Description: add support for EAs (user and system) on lustre filesystems
2447 Details : it is now possible to store extended attributes in the Lustre
2448 client filesystem, and with the user_xattr mount option it
2449 is possible to allow users to store EAs on their files also
2451 Severity : enhancement
2453 Description: Add possibility (config option) to show minimal available OST free
2455 Details : When compiled with --enable-mindf configure option, statfs(2)
2456 (and so, df) will return least minimal free space available from
2457 all OSTs as amount of free space on FS, instead of summary of
2458 free spaces of all OSTs.
2460 Severity : enhancement
2462 Description: do not expand extent locks acquired on OST-side
2463 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2464 by server: they are not cached anyway.
2467 Frequency : when mmap is used/binaries executed from Lustre
2469 Description: Unmmap pages before throwing them away from read cache.
2470 Details : llap_shrink cache now attempts to unmap pages before discarding
2471 them (if unmapping failed - do not discard). SLES9 kernel has
2472 extra checks that trigger if this unmapping is not done first.
2477 Description: lconf didn't resolve symlinks before checking to see whether a
2478 given mountpoint was already in use
2481 Frequency : when migrating failover services
2482 Bugzilla : 6395, 9514
2483 Description: When migrating a subset of services from a node (e.g. failback
2484 from a failover service node) the remaining services would
2485 time out and evict clients.
2486 Details : lconf --force (implied by --failover) sets the global obd_timeout
2487 to 5 seconds in order to quickly disconnect, but this caused
2488 other RPCs to time out too quickly. Do not change the global
2489 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2491 Severity : enhancement
2492 Frequency : if MDS is started with down OST
2493 Bugzilla : 9439,5706
2494 Description: Allow startup/shutdown of an MDS without depending on the
2495 availability of the OSTs.
2496 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2497 Add appropriate locking and lov-osc refcounts for safe
2498 cleaning. Add osc abort_inflight calls in case the
2499 synchronize never started.
2502 Frequency : occasional (Cray XT3 only)
2504 Description: root not authorized to access files in CRAY_PORTALS environment
2505 Details : The client process capabilities were not honoured on the MDS in
2506 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2507 been accessed by an authorized user then root was able to access
2508 the file on the local client also. The root user capabilities
2509 are now allowed on the MDS, as this environment has secure UID.
2512 Frequency : occasional
2514 Description: ldiskfs "too long searching" message happens too often
2515 Details : A debugging message (otherwise harmless) prints too often on
2516 the OST console. This has been reduced to only happen when
2517 there are fragmentation problems on the filesystem.
2522 Description: Division by zero in statfs when all OSCs are inactive
2523 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2524 lov_statfs divides by value returned by lov_get_stripecnt().
2528 Bugzilla : 9489, 3273
2529 Description: First write from each client to each OST was only 4kB in size,
2530 to initialize client writeback cache, which caused sub-optimal
2531 RPCs and poor layout on disk for the first writen file.
2532 Details : Clients now request an initial cache grant at (re)connect time
2533 and so that they can start streaming writes to the cache right
2534 away and always do full-sized RPCs if there is enough data.
2535 If the OST is rebooted the client also re-establishes its grant
2536 so that client cached writes will be honoured under the grant.
2541 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2542 Details : Now I/O RPCs go to different portal number and (presumably) fast
2543 lock requests (and glimses) and other RPCs get their own service
2544 threads pool that should be able to service those RPCs
2547 Severity : enhancement
2549 Description: Ability to exchange lustre version between client and servers and
2550 issue warnings at client side if client is too old. Also for
2551 liblustre clients there is ability to refuse connection of too old
2553 Details : New 'version' field is added to connect data structure that is
2554 filled with version info. That info is later checked by server and
2558 Frequency : rare, liblustre only.
2559 Bugzilla : 9296, 9581
2560 Description: Two simultaneous writes from liblustre at offset within same page
2561 might proceed at the same time overwriting eachother with stale
2563 Details : I/O lock withing llu_file_prwv was released too early, before data
2564 actually was hitting the wire. Extended lock-holding time until
2565 server acknowledges receiving data.
2568 Frequency : extremely rare. Never observed in practice.
2570 Description: avoid generating lustre_handle cookie of 0.
2571 Details : class_handle_hash() generates handle cookies by incrementing
2572 global counter, and can hit 0 occasionaly (this is unlikely, but
2573 not impossible, because initial value of cookie counter is
2574 selected randonly). Value of 0 is used as a sentinel meaning
2575 "unassigned handle" --- avoid it. Also coalesce two critical
2576 sections in this function into one.
2578 Severity : enhancement
2580 Description: allow liblustre clients to delegate truncate locking to OST
2581 Details : To avoid overhead of locking, liblustre client instructs OST to
2582 take extent lock in ost_punch() on client's behalf. New connection
2583 flag is added to handle backward compatibility.
2585 Severity : enhancement
2586 Bugzilla : 4928, 7341, 9758
2587 Description: allow number of OST service threads to be specified
2588 Details : a module parameter allows the number of OST service threads
2589 to be specified via "options ost ost_num_threads={N}" in the
2590 OSS's /etc/modules.conf or /etc/modprobe.conf.
2594 Bugzilla : 6146, 9635, 9895
2595 Description: servers crash with bad pointer in target_handle_connect()
2596 Details : In rare cases when a client is reconnecting it was possible that
2597 the connection request was the last reference for that export.
2598 We would temporarily drop the export reference and get a new
2599 one, but this may have been the last reference and the export
2600 was just destroyed. Get new reference before dropping old one.
2602 Severity : enhancement
2603 Frequency : if client is started with failover MDS
2605 Description: Allow multiple MDS hostnames in the mount command
2606 Details : Try to read the configuration from all specified MDS
2607 hostnames during a client mount in case the "primary"
2610 Severity : enhancement
2612 Description: Stop sending data to evicted clients as soon as possible.
2613 Details : Check if the client we are about to send or are sending data to
2614 was evicted already. (Check is done every second of waiting,
2615 for which l_wait_event interface was extended to allow checking
2616 of exit condition at specified intervals).
2619 Frequency : rare, normally only when NFS exporting is done from client
2621 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2623 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2626 Frequency : always, if srand() is called before liblustre initialization
2628 Description: Liblustre uses system PRNG disturbing its usage by user application
2629 Details : Introduce internal to lustre fast and high-quality PRNG for
2630 lustre usage and make liblustre and some other places in generic
2631 lustre code to use it.
2633 Severity : enhancement
2634 Bugzilla : 9477, 9557, 9870
2635 Description: Verify that the MDS configuration logs are updated when xml is
2636 Details : Check if the .xml configuration logs are newer than the config
2637 logs stored on the MDS and report an error if this is the case.
2638 Request --write-conf, or allow starting with --old_conf.
2640 Severity : enhancement
2642 Description: Handle symlinks in the path when checking if Lustre is mounted.
2643 Details : Resolve intermediate symlinks when checking if a client has
2644 mounted a filesystem to avoid duplicate client mounts.
2649 Description: lconf can hit an error exception but still return success.
2650 Details : The lconf command catches the Command error exception at the top
2651 level script context and will exit with the associated exit
2652 status, but doesn't ensure that this exit status is non-zero.
2657 Description: failure of ptlrpc thread startup can cause oops
2658 Details : Starting a ptlrpc service thread can fail if there are a large
2659 number of threads or the server memory is very fragmented.
2660 Handle this without oopsing.
2663 Frequency : always, only if liblustre and non-default acceptor port was used
2665 Description: liblustre cannot connect to servers with non-default acceptor port
2666 Details : tcpnal_set_default_params() was not called and was therefore
2667 ignoring the environment varaible TCPNAL_PORT, as well as other
2668 TCPNAL_ environment variables
2673 Description: two objects could be created on the same OST for a single file
2674 Details : If an OST is down, in some cases it was possible to create two
2675 objects on a single OST for a single file. No problems other
2676 than potential performance impact and spurious error messages.
2680 Bugzilla : 5681, 9562
2681 Description: Client may oops in ll_unhash_aliases
2682 Details : Client dcache may become inconsistent in race condition.
2683 In some cases "getcwd" can fail if the current directory is
2689 Description: Inode refcounting problems in NFS export code
2690 Details : link_raw functions used to call d_instantiate without obtaining
2691 extra inode reference first.
2695 Bugzilla : 9942, 9903
2696 Description: Referencing freed requests leading to crash, memleaks with NFS.
2697 Details : We used to require that call to ll_revalidate_it was always
2698 followed by ll_lookup_it. Also with revalidate_special() it is
2699 possible to call ll_revalidate_it() twice for the same dentry
2700 even if first occurence returned success. This fix changes semantic
2701 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2702 reference on a request referred from the intent.
2703 ll_intent_release() then releases such a request.
2706 Frequency : rare, normally benchmark loads only
2708 Description: unlinked inodes were kept in memory on the client
2709 Details : If a client is repeatedly creating and unlinking files it
2710 can accumulate a lot of stale inodes in the inode slab cache.
2711 If there is no other client load running this can cause the
2712 client node to run out of memory. Instead flush old inodes
2713 from client cache that have the same inode number as a new inode.
2716 Frequency : SLES9 2.6.5 kernel and long filenames only
2717 Bugzilla : 9969, 10379
2718 Description: utime reports stale NFS file handle
2719 Details : SLES9 uses out-of-dentry names in some cases, which confused
2720 the lustre dentry revalidation. Change it to always use the
2724 Frequency : rare, unless heavy write-truncate concurrency is continuous
2725 Bugzilla : 4180, 6984, 7171, 9963, 9331
2726 Description: OST becomes very slow and/or deadlocked during object unlink
2727 Details : filter_destroy() was holding onto the parent directory lock
2728 while truncating+unlinking objects. For very large objects this
2729 may block other threads for a long time and slow overall OST
2730 responsiveness. It may also be possible to get a lock ordering
2731 deadlock in this case, or run out of journal credits because of
2732 the combined truncate+unlink. Solution is to do object truncate
2733 first in one transaction without parent lock, and then do the
2734 final unlink in a new transaction with the parent lock. This
2735 reduces the lock hold time dramatically.
2738 Frequency : rare, 2.4 kernels only
2740 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2741 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2742 before unmounting their filesystem in order to avoid deadlock.
2743 The kernel_locked() function in 2.4 kernels only checks whether
2744 the kernel lock is held, not whether it is this process that is
2745 holding it as 2.6 kernels do.
2750 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2751 Details : The client ptlrpc code may be trying to reconnect to a down
2752 server before a previous connection attempt has timed out.
2753 Increase the reconnect interval to be longer than the connection
2754 timeout interval to avoid sending duplicate connections to
2758 Frequency : echo_client brw_test command
2760 Description: fix echo_client to work with OST preallocated code
2761 Details : OST preallocation code (5137) didn't take echo_client IO path
2762 into account: echo_client calls filter methods outside of any
2763 OST thread and, hence, there is no per-thread preallocated
2764 pages and buffers to use. Solution: hijack pga pages for IO. As
2765 a byproduct, this avoids unnecessary data copying.
2769 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2770 Description: Client can oops in mdc_commit_close() after open replay
2771 Details : It was possible for the MDS to return an open request with no
2772 transaction number in mds_finish_transno() if the client was
2773 evicted, but without actually returning an error. Clients
2774 would later try to replay that open and may trip an assertion
2775 Simplify the client close codepath, and always return an error
2776 from the MDS in case the open is not successful.
2779 Frequency : rare, 2.6 OSTs only
2781 Description: OST may deadlock under high load on fragmented files
2782 Details : If there was a heavy load and highly-fragmented OST filesystems
2783 it was possible to have all the OST threads deadlock waiting on
2784 allocation of biovecs, because the biovecs were not released
2785 until the entire RPC IO was completed. Instead, release biovecs
2786 as soon as they are complete to ensure forward IO progress.
2788 Severity : enhancement
2790 Description: Support for specifying external journal device at mount
2791 Details : If an OST or MDS device is formatted with an external journal
2792 device, this device major/minor is stored in the ext3 superblock
2793 and may not be valid for failover. Allow detecting and
2794 specifying the external journal at mount time.
2799 Description: Mounting an MDS with pending unlinked files may cause oops
2800 Details : target_finish_recovery() calls mds_postrecov() which returned
2801 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2802 considers this an error and immediately begins cleaning up the
2803 lov, just after starting the mds_lov process
2805 Severity : enhancement
2807 Description: Implement 'lfs df' to report actual free space on per-OST basis
2808 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2809 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2810 usage of objects; '-h' to report in human readable format.
2812 ------------------------------------------------------------------------------
2814 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2821 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2822 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2823 at startup during orphan unlinking. Instead, skip buddy update
2824 before setup, as it will be regenerated after recovery is complete.
2829 Description: If an OST is inactive, its locks might reference stale inodes.
2830 Details : lov_change_cbdata() must iterate over all namespaces, even if
2831 they are inactive to clear inode references from the lock.
2833 Severity : enhancement
2834 Frequency : occasional, if non-standard max_dirty_mb used
2836 Description: Client will block write RPCs if not enough grant
2837 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2838 then the client will block writes while waiting for another RPC
2839 to complete instead of consuming its dirty limit. With change
2840 we get improved performance when max_dirty_mb is small.
2842 Severity : enhancement
2843 Bugzilla : 3389, 6253
2844 Description: Add support for supplementary groups on the MDS.
2845 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2846 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2847 lookups for user supplementary groups into a cache.
2851 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2852 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2853 for liblustre O_EXCL is enforced by MDS.
2857 Description: readdir never returns NULL in liblustre.
2858 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2859 dentry in struct dirent.
2863 Frequency : liblustre only, depends on application IO pattern
2864 Description: liblustre clients evicted if not contacting servers
2865 Details : Don't put liblustre clients into the ping_evictor list, so
2866 they will not be evicted by the pinger ever.
2868 Severity : enhancement
2870 Description: Add ability to evict clients by NID from MDS.
2871 Details : By echoing "nid:$NID" string into
2872 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2873 $NID would be instantly evicted from this MDS and from all active
2874 OSTs connected to it.
2878 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2879 Details : lookup_it_finish() used to query file size from OSTs that was not
2884 Description: service threads change working directory to that of init
2885 Details : Starting lustre service threads may pin the working directory
2886 of the parent thread, making that filesystem busy. Threads
2887 now change to the working directory of init to avoid this.
2891 Frequency : during shutdown only
2892 Description: shutdown with a failed MDS or OST can cause unmount to hang
2893 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2898 Frequency : 2.6 only
2899 Description: chmod/chown may include an extra supplementary group
2900 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2901 supplementary group and if none is specified this is used.
2904 Bugzilla : 5479 (6816)
2906 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2907 Details : If lookup is in progress on a file that is unlinked we might try
2908 to revalidate the inode and fail in revalidate after lookup is
2909 complete and ll_file_open() enqueues the open again but
2910 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2913 Frequency : always, if lconf --abort_recovery used
2915 Description: lconf --abort_recovery fails with 'Operation not supported'
2916 Details : lconf was attempting to abort recovery on the MDT device and not
2919 ------------------------------------------------------------------------------
2921 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2926 Frequency : rare (only unsupported configurations with a node running as an
2928 Bugzilla : 6514, 5137
2929 Description: Mounting a Lustre file system on a node running as an OST could
2931 Details : OSTs now preallocates memory needed to write out data at
2932 startup, instead of when needed, to avoid having to
2933 allocate memory in possibly low memory situations.
2934 Specifically, if the file system is mounted on on OST,
2935 memory pressure could force it to try to write out data,
2936 which it needed to allocate memory to do. Due to the low
2937 memory, it would be unable to do so and the node would
2938 become unresponsive.
2940 Severity : enhancement
2942 Description: Addition of lconf --service command line option
2943 Details : lconf now accepts a '--service <arg>' option, which is
2944 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2946 Severity : enhancement
2948 Description: Failover mode is now the default for OSTs.
2949 Details : By default, OSTs will now run in failover mode. To return to
2950 the old behaviour, add '--failout' to the lmc line for OSTs.
2952 Severity : enhancement
2954 Description: Health checks are now provided for MDS and OSTs
2955 Details : Additional detailed health check information on MSD and OSTs
2956 is now provided through the procfs health_check value.
2959 Frequency : occasional, depends on IO load
2961 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2962 after numerous create/delete cycles
2963 Details : The ext3 inode allocation policy would not allocate new inodes
2964 very well on the OSTs because there are no new directories
2965 being created. Instead we look for groups with free space if
2966 the parent directories are nearly full.
2971 Description: Network or server problems during mount may cause partially
2972 mounted clients instead of returning an error.
2973 Details : The config llog parsing code may overwrite the error return
2974 code during mount error handling, returning success instead
2980 Description: MDS can fail to allocate large reply buffers
2981 Details : After long uptimes the MDS can fail to allocate large reply
2982 buffers (e.g. zconf client mount config records) due to memory
2983 fragmentation or consumption by the buffer cache. Preallocate
2984 some large reply buffers so that these replies can be sent even
2985 under memory pressure.
2989 Frequency : rare (liblustre)
2990 Description: fsx running with liblustre complained that using truncate() to
2991 extend the file doesn't work. This patch corrects that issue.
2992 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2993 ATTR_SIZE and lsm use in llu_setattr_raw.
2997 Frequency : rare, only 2.6 kernels
2998 Description: Unusual file access patterns on the MDS may result in inode
2999 data being lost in very rare circumstances.
3000 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3001 code in the 2.6 kernel caused the inode and/or EA data not to
3002 be read from disk, causing single-file corruption.
3006 Frequency : rare, only 2.6 filesystems using extents
3007 Description: Heavy concurrent write and delete load may cause data corruption.
3008 Details : It was possible under high-load situations to have an extent
3009 metadata block in the block device cache from a just-unlinked
3010 file overwrite a newly-allocated data block. We now unmap any
3011 metadata buffers that alias just-allocated data blocks.
3015 Frequency : filesystems with default stripe_count larger than 77
3016 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3017 Details : lconf specifies an inode size of 4096 bytes when the default
3018 stripe_count is larger than 77. This conflicts with the default
3019 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3020 this case to avoid pinning too much memory for large EAs.
3022 ------------------------------------------------------------------------------
3024 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3029 Frequency : rare (extremely heavy IO load with hundreds of clients)
3031 Description: Client is evicted, gets IO error writing to file
3032 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3033 caused clients to be evicted for AST timeouts. The fixes in
3034 bug 5192 mean we no longer need to have such short AST timeouts
3035 so ldlm_timeout has been increased.
3038 Frequency : occasional during --force or --failover shutdown under load
3039 Bugzilla : 5949, 4834
3040 Description: Server oops/LBUG if stopped with --force or --failover under load
3041 Details : a collection of import/export refcount and cleanup ordering
3042 issues fixed for safer force cleanup
3045 Frequency : only filesystems larger than 120 OSTs
3046 Bugzilla : 5990, 6223
3047 Description: lfs getstripe would oops on a very large filesystem
3048 Details : lov_getconfig used kfree on vmalloc'd memory
3051 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3052 Bugzilla : 6242, 6243
3053 Description: reading from files that had been truncated to a non-zero size
3054 but never opened returned no data
3055 Details : ll_file_read() reads zeros from no-object files to EOF
3060 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3061 Details : There's little harm in aborting MDS/OSS recovery and letting it
3062 try again, so I removed the LASSERT and return an error instead.
3064 Severity : enhancement
3066 Description: New debugging infrastructure for tracking down data corruption
3067 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3068 (b) cover more of the client-side code path, and (c) try to narrow
3069 down where problems occurred
3073 Bugzilla : 3819, 4364, 4397, 6313
3074 Description: Racing close and eviction MDS could cause assertion in mds_close
3075 Details : It was possible to get multiple mfd references during close and
3076 client eviction, leading to one thread referencing a freed mfd.
3078 Severity: : enhancement
3079 Bugzilla : 3262, 6359
3080 Description: Attempts to reconnect to servers are now more aggressive.
3081 Details : This builds on the enhanced upcall-less recovery that was added
3082 in 1.4.2. When trying to reconnect to servers, clients will
3083 now try each server in the failover group every 10 seconds. By
3084 default, clients would previously try one server every 25 seconds.
3089 Description: After recovery, certain operations trigger a failed
3090 assertion on a client.
3091 Details : Failing over an mds, using lconf -d --failover, while a
3092 client was doing a readdir() call would cause the client to
3093 LBUG after recovery completed and the readdir() was resent.
3095 Severity : enhancement
3097 Description: Default groups are now added by lconf
3098 Details : You can now run lconf --group <servicename> without having to
3099 manually add groups with lmc.
3102 Frequency : occasional
3104 Description: Nodes with an elan id of 0 trigger a failed assertion
3107 Frequency : always when accessing e.g. tty/console device nodes
3109 Description: tty and some other devices nodes cannot be used on lustre
3110 Details : file's private_data field is used by device data and lustre
3111 values in there got lost. New field was added to struct file to
3112 store fs-specific private data.
3115 Frequency : when exporting Lustre via NFS
3117 Description: NFSD failed occasionally when looking up a path component
3118 Details : NFSD is looking up ".." which was broken in ext3 directories
3119 that had grown large enough to become hashed.
3122 Frequency : Clusters with multiple interfaces not on the same subnet
3124 Description: Nodes will repeatedly try to reconnect to an interface which it
3125 cannot reach and report an error to the log.
3126 Details : Extra peer list entries will be created by lconf with some peers
3127 unreachable. lconf now validates the peer before adding it.
3130 Frequency : Only if a default stripe is set on the filesystem root.
3132 Description: Setting a default stripe on the filesystem root prevented the
3133 filesystem from being remounted.
3134 Details : The client was sending extra request flags in the root getattr
3135 request and did not allocate a reply buffer for the dir EA.
3138 Frequency : occasional, higher if lots of files are accessed by one client
3139 Bugzilla : 6159, 6097
3140 Description: Client trips assertion regarding lsm mismatch/magic
3141 Details : While revalidating inodes the VFS looks up inodes with ifind()
3142 and in rare cases can find an inode that is being freed.
3143 The ll_test_inode() code will free the lsm during ifind()
3144 when it finds an existing inode and then the VFS later attaches
3145 this free lsm to a new inode.
3149 Bugzilla : 6422, 7030
3150 Description: MDS deadlock between mkdir and client eviction
3151 Details : Creating a new file via mkdir or mknod (starting a transaction
3152 and getting the ns lock) can deadlock with client eviction
3153 (gets ns lock and trying to finish a synchronous transaction).
3156 Frequency : occasional
3157 Description: While starting a server, the fsfilt_ext3 module could not be
3159 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3160 kernels. Previously, lconf would still use the ext3 name
3161 when trying to load modules. Now, it will correctly use
3162 ext3 on 2.4 and ldiskfs on 2.6.
3164 Severity : enhancement
3165 Description: The default stripe count has been changed to 1
3166 Details : The interpretation of the default stripe count (0, to lfs
3167 or lmc) has been changed to mean striping across a single
3168 OST, rather than all available. For general usage we have
3169 found a stripe count of 1 or 2 works best.
3171 Severity : enhancement
3172 Description: Add support for compiling against Cray portals.
3173 Details : Conditional compiling for some areas that are different
3177 Frequency : occasional
3178 Bugzilla : 6409, 6834
3179 Description: Creating files with an explicit stripe count may lead to
3180 a failed assertion on the MDS
3181 Details : If some OSTs are full or unavailable, creating files may
3182 trigger a failed assertion on the MDS. Now, Lustre will
3183 try to use other servers or return an error to the
3187 Frequency : occasional
3189 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3190 caused lock timeout during readahead (since 1.4.2).
3191 Details : Processes doing readahead might match a lock that hasn't been
3192 granted yet if there are overlapping and conflicting lock
3193 requests. The readahead process waits on ungranted lock
3194 (original lock is CBPENDING), while OST waits for that process
3195 to cancel CBPENDING read lock and eventually evicts client.
3197 Severity : enhancement
3199 Description: Initial enabling of flock support for clients
3200 Details : Implements fcntl advisory locking and file status functions.
3201 This feature is provided as an optional mount flag (default
3202 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3203 locking are implemented yet, and those that are are not guaranteed
3204 to be completely correct in production environments.
3205 mount -t lustre -o [flock|noflock] ...
3208 Frequency : occasional
3210 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3211 assertion in the ext3 JBD (journaling) layer.
3212 Details : The b_committed_data struct is protected by the big kernel lock
3213 in 2.4 kernels, serializing journal_commit_transaction() and
3214 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3215 access to this struct. In 2.6 kernels there is finer grained
3216 locking to improve SMP performance of the JBD layer.
3220 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3225 Description: merge and cleanup kernel patches.
3228 -----------------------------------------------------------------------------
3230 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3232 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3233 versions of Lustre. Please update all servers and clients to
3234 version 1.4.2 or later at the same time. You must also run
3235 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3236 to update the configuration logs.
3238 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3239 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3240 - swab llog records of type '0' so we get proper header size/idx (5861)
3241 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3242 - fix rename of one directory over another leaking an inode (5953)
3243 - avoid SetPageDirty on 2.6 (5981)
3244 - don't re-add just-being-destroyed locks to the waiting list (5653)
3245 - when creating new directories, inherit the parent's custom
3246 striping settings if present parent (3048)
3247 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3248 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3249 - don't hold client locks on temporary worklist from l_lru (5666)
3250 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3251 - automatically evict dead clients (5921)
3252 - Update file size properly in create+truncate+fstat case (6196)
3253 - Do not unhash mountpoint dentries, do not allow removal of
3255 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3256 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3257 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3258 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3259 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3260 - Correctly handle reads of files with no objects (6243)
3261 - lctl recover will also mark a device active if deactivate used (5933)
3263 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3264 - allow --write-conf on an MDS with different nettype than client (5619)
3265 - don't write config llogs to MDS for mounts not from that MDS (5617)
3266 - lconf should create multiple TCP connections from a client (5201)
3267 - init scripts are now turned off by default; run chkconfig --on
3268 lustre and chkconfig --on lustrefs to use them
3269 - upcalls are no longer needed for clients to recover to failover
3271 - add --abort-recovery option to lconf to abort recovery on device
3273 - add support for an arbitrary number of OSTs (3026)
3274 - Quota support protocol changes.
3275 - forward compatibility changes to wire structs (6007)
3276 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3277 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3278 - improved "lustre status" script
3279 - initialize blocksize for non-regular files (6062)
3280 - added --disable-server and --disable-client configure options (5782)
3281 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3282 - Vanilla 2.4.29 support
3283 - increase maximum number of obd devices to 520 (6242)
3284 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3285 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3286 SLES 9 2.6.5 kernels
3287 - limit stripes per file to 160 (the maximum EA size) (6093)
3289 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3292 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3293 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3294 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3295 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3296 - small ext3 extents cleanups and fixes (5733)
3297 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3298 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3299 - avoid needless client->OST connect, fix handle mismatch (5317)
3300 - fix DLM error path that led to out-of-sync client, long delays (5779)
3301 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3302 - fix several locking issues related to i_size (5492,5624,5654,5672)
3303 - don't move pending lock onto export if it is already evicted (5683)
3304 - fix kernel oops when creating .foo in unlinked directory (5548)
3305 - fix deadlock in obdfilter statistics vs. object create (5811)
3306 - use time_{before,after} to avoid timer jiffies wrap (5882)
3307 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3308 - Do not leak request if server was not able to process it (5154)
3309 - If mds_open unable to find parent dir, make that negative lookup(5154)
3310 - don't create new directories with extent-mapping (5909, 5936)
3312 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3313 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3314 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3315 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3316 - support for RHEL4 kernel on i686 (5773)
3317 - provide error messages when incompatible logs are encountered (5898)
3319 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3320 * version 1.4.0.10 (1.4.1 release candidate 1)
3322 - don't keep a lock reference when lock is not granted (4238)
3323 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3324 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3325 - drop import inflight refcount on signal_completed_replay error (5255)
3326 - unlock page after async write error during send (3677)
3327 - handle missing objects in filter_preprw_read properly (5265)
3328 - no transno return for symlink open, don't save no-trasno open (3440)
3329 - don't try to complete elan receive that already failed (4012)
3330 - free RPC server reply state on error (5406)
3331 - clean up thread from ptlrpc_start_thread() on error (5160)
3332 - readahead could read extra page into cache that wasn't ejected (5388)
3333 - prevent races in class_attach/setup/cleanup/detach (5260)
3334 - don't dereference de->d_inode after l_dput of de (5458)
3335 - use "int" for stripe value returned from lock_to_stripe (5544)
3336 - mballoc allocation and error-checking fixes in 2.6 (5504)
3337 - block device patches to fix I/O request sizes in 2.6 (5482)
3338 - look up hostnames for IB nals (5602)
3339 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3340 - don't start multiple acceptors for the same port (5277)
3341 - fix incorrect LASSERT in mds_getattr_name (5635)
3342 - export a proc file for general "ping" checking (5628)
3343 - fix "lfs check" to not block when the MDS is down (5628)
3345 - service request history (4965)
3346 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3347 - create an "evict_client" /proc entry on OSTs, like the MDS has
3348 - fix mount usage message, return errors per mount(8) (5168)
3349 - change grep [] to grep "[]" in tests so they work in more UMLs
3350 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3351 - remove ancient LOV_MAGIC_V0 EA support (5047)
3352 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3353 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3354 - allow readahead allocations to fail when low on memory (5383)
3355 - mmap locking landed again, after considerable improvement (2828)
3356 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3358 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3361 - send OST transaction number in read/write reply to free req (4966)
3362 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3363 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3364 - the watchdog thread now runs as interruptible (5246)
3365 - flock/lockf fixes (but it's still disabled, pending 5135)
3366 - don't use EXT3 constants in llite code (5094)
3367 - memory shortage at startup could cause assertion (5176)
3369 - reorganization of lov code
3370 - single portals codebase
3372 - add extents/mballoc support (5025)
3373 - direct I/O reads in the obdfilter (4048)
3374 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3376 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3379 - send OST transaction number in read/write reply to free req (4966)
3380 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3381 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3382 - the watchdog thread now runs as interruptible (5246)
3383 - handle missing objects in filter_preprw_read properly (5265)
3384 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3385 - drop import inflight refcount on signal_completed_replay error (5255)
3386 - unlock page after async write error during send (3677)
3387 - return original error code on reconstructed replies (3761)
3388 - no transno return for symlink open, don't save no-trasno open (3440)
3390 - add pid to ldlm debugging output (4922)
3391 - bump the watchdog timeouts -- we can't handle 30sec yet
3392 - extra debugging for orphan dentry/inode bug (5259)
3394 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3397 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3398 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3399 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3400 - fix NULL dereference in /proc/sys/portals/routes (4827)
3401 - allow failed mdc_close() operations to be interrupted (4561)
3402 - stop precreate on OST before MDS would time out on it (4778)
3403 - don't send partial-page writes before EOF from client (4410)
3404 - discard client grant for sub-page writes on large-page clients (4520)
3405 - don't free dentries not owned by NFS code, check generation (4806)
3406 - fix lsm leak if mds_create_objects() fails (4801)
3407 - limit debug_daemon file size, always print CERROR messages (4789)
3408 - use transno after validating reply (3892)
3409 - process timed out requests if import state changes (3754)
3410 - update mtime on OST during writes, return in glimpse (4829)
3411 - add mkfsoptions to LDAP (4679)
3412 - use ->max_readahead method instead of zapping global ra (5039)
3413 - don't interrupt __l_wait_event() during strace
3415 - add software watchdogs to catch hung threads quickly (4941)
3416 - make lustrefs init script start after nfs is mounted
3417 - fix CWARN/ERROR duplication (4930)
3418 - return async write errors to application if possible (2248)
3419 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3420 - print NAL number in %x format (4645)
3421 - update barely-supported suse-2.4.21-171 series (4842)
3422 - support for sles 9 %post scripts
3423 - support for building 2.6 kernel-source packages
3424 - support for sles km_* packages
3426 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3429 - ignore -ENOENT errors in osc_destroy (3639)
3430 - notify osc create thread that OSC is being cleaned up (4600)
3431 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3432 - reconstruct ost_handle() like mds_handle() (4657)
3433 - create a new thread to do import eviction to avoid deadlock (3969)
3434 - let lconf resolve symlinked-to devices (4629)
3435 - don't unlink "objects" from directory with default EA (4554)
3436 - hold socknal file ref over connect in case target is down (4394)
3437 - allow more than 32000 subdirectories in a single directory (3244)
3438 - fix blocks count for O_DIRECT writes (3751)
3439 - OST returns ENOSPC from object create when no space left (4539)
3440 - don't send truncate RPC if file size isn't changing (4410)
3441 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3442 - bind to privileged port in socknal and tcpnal (3689)
3444 - rate limit CERROR/CWARN console message to avoid overload (4519)
3445 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3446 - basic mmap support (3918)
3447 - kernel patch series update from b1_4 (4711)
3449 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3452 - avoid crash during MDS cleanup with OST shut down (2775)
3453 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3454 - don't use bad inodes on the MDS (3744)
3455 - dynamic object preallocation to improve recovery speed (4236)
3456 - don't hold spinlock over lock dumping or change debug flags (4401)
3457 - don't zero obd_dev when it is force cleaned (3651)
3458 - print grants to console if they go negative (4431)
3459 - "lctl deactivate" will stop automatic recovery attempts (3406)
3460 - look for existing locks in ldlm_handle_enqueue() (3764)
3461 - don't resolve lock handle twice in recovery avoiding race (4401)
3462 - revalidate should check working dir is a directory (4134)
3464 - don't always mark "slow" obdfilter messages as errors (4418)
3466 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3469 - don't close LustreDB during write_conf until it is done (3860)
3470 - fix typo in lconf for_each_profile (3821)
3471 - allow dumping logs from multiple threads at one time (3820)
3472 - don't allow multiple threads in OSC recovery (3812)
3473 - fix debug_size parameters (3864)
3474 - fix mds_postrecov to initialize import for llog ctxt (3121)
3475 - replace config semaphore with spinlock (3306)
3476 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3477 - don't allow enqueue to complete on a destroyed export (3822)
3478 - down write_lock before checking llog header bitmap (3825)
3479 - recover from lock replay timeout (3764)
3480 - up llog sem before sending rpc (3652)
3481 - reduce ns lock hold times when setting kms (3267)
3482 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3483 - fix NULL deref and obd_dev leak on setup error (3312)
3484 - replace some LBUG about llog ops with error handling (3841)
3485 - don't match INVALID dentries from d_lookup and spin (3784)
3486 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3487 - fix invalid assertion in ptlrpc_set_wait (3880)
3489 - add libwrap support for the TCP acceptor (3996)
3490 - add /proc/sys/portals/routes for non-root route listing (3994)
3491 - allow setting MDS UUID in .xml (2580)
3492 - print the stack of a process that LBUGs (4228)
3494 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3497 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3498 - make sure to unset replay flag from failed open requests (3440)
3499 - if default stripe count is 0, use OST count for inode size (3636)
3500 - update parent mtime/ctime on client for create/unlink (2611)
3501 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3502 - free recovery state on server during a forced cleanup (3571)
3503 - unregister_reply for resent reqs (3063)
3504 - loop back devices mounting and status check on 2.6 (3563)
3505 - fix resource-creation race that can provoke i_size == 0 (3513)
3506 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3507 - more debugging for page-accounting assertion (3746)
3508 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3509 - avoid deadlock after precreation failure (3758)
3510 - fix race and lock order deadlock in orphan handling (3450, 3750)
3511 - add validity checks when grabbing inodes from l_ast_data (3599)
3513 - add /proc/.../recovery_status to obdfilter (3428)
3514 - lightweight CDEBUG infrastructure, debug daemon (3668)
3515 - change default OSC RPC parameters to be better on small clusters
3516 - turn off OST read cache for files smaller than 32MB
3517 - install man pages and include them in rpms (3100)
3518 - add new init script for (un)mounting lustre filesystems (2593)
3519 - run chkconfig in %post for init scripts (3701)
3520 - drop scimac NAL (unmaintained)
3522 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3525 - clean kiobufs before and after use (3485)
3526 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3527 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3528 - change init script to fail more clearly if not run as root (1528)
3529 - allow clients to reconnect during replay (1742)
3530 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3531 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3532 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3533 - fix cond_resched (3554)
3534 - only evict unfinished clients after recovery (3515)
3535 - allow bulk resend, prevent data loss (3570)
3536 - dynamic ptlrpc request buffer allocation (2102)
3537 - don't allow unlinking open directory if it isn't empty (2904)
3538 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3539 - remove extraneous obd dereference causing LASSERT failure (3334)
3540 - don't use get_cycles() when creating temp. files on the mds (3156)
3541 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3542 - handle EEXIST for set-stripe, set proper directory name (3336)
3544 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3545 - fix ksocknal_fmb_callback() error messages (2918)
3547 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3550 - don't copy lvb into (possibly NULL) reply on error (2983)
3551 - don't deref dentry after dput, don't free lvb on error (2922)
3552 - use the kms to determine writeback rpc length (2947)
3553 - increment oti_logcookies when osc is inactive (2948)
3554 - update client's i_blocks count via lvb messages (2543)
3555 - handle intent open/close of special files properly (1557)
3556 - mount MDS with errors=remount-ro, like obdfilter (2009)
3557 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3558 - don't use cancelling-locks' kms values (2947)
3559 - use highest lock extent for kms, not last one (2925)
3560 - don't dereference ERR_PTR() dentry in error handling path (3107)
3561 - fix thread race in portals_debug_dumplog() (3122)
3562 - create lprocfs device entries at setup instead of at attach (1519)
3563 - common AST error handler, don't evict client on completion race (3145)
3564 - zero nameidata in detach_mnt in 2.6 (3118)
3565 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3566 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3567 - initialize RPC timeout timer earlier for 2.6 (3219)
3568 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3569 - print nal/nid for unknown nid (3258)
3570 - additional checks for oscc recovery before doing precreate (3284)
3571 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3572 - don't crash in mdc_close for bad permissions on open (3285)
3573 - zero i_rdev for non-device files (3147)
3574 - clear page->private before handing to FS, better assertion (3119)
3575 - tune the read pipeline (3236)
3576 - fix incorrect decref of invalidated dentry (2350)
3577 - provide read-ahead stats and refine rpc in flight stats (3328)
3578 - don't hold journal transaction open across create RPC (3313)
3579 - update atime on MDS at close time (3265)
3580 - close LDAP connection when recovering to avoid server load (3315)
3581 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3582 - don't leak open file on MDS after open resend (3325)
3583 - serialize filter_precreate and filter_destroy_precreated (3329)
3584 - loop device shouldn't call sync_dev() for nul device (3092)
3585 - clear page cache after eviction (2766)
3586 - resynchronize MDS->OST in background (2824)
3587 - refuse to mount the same filesystem twice on same mountpoint (3394)
3588 - allow llmount to create routes for mounting behind routers (3320)
3589 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3590 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3591 - fix rare problem with rename on htree directories (3417)
3593 - allow default OST striping configuration per directory (1414)
3594 - fix compilation for qswnal for 2.6 kernels (3125)
3595 - increase maximum number of MDS request buffers for large systems
3596 - change liblustreapi to be useful for external progs like lfsck (3098)
3597 - increase local configuration timeout for slow disks (3353)
3598 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3600 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3603 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3604 - don't overwrite extent policy data in reply if lock was blocked (2901)
3605 - drop filter export grants atomically with removal from device (2663)
3606 - del obd_self_export from work_list in class_disconnect_exports (2908)
3607 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3608 - swab reply message in mdc_close, other PPC fixes (2464)
3609 - fix destroying of named logs (2325)
3610 - overwrite old logs when running lconf --write_conf (2264)
3611 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3612 - fix race in target_handle_connect (2898)
3613 - mds_reint_create() should take same inode create lock (2926)
3614 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3615 - don't close files for self_export to avoid uninitialized obd (2936)
3616 - allow MDS with the same name as client node (2939)
3617 - hold dentry reference for closed log files for unlink (2325)
3618 - reserve space for all logs during transactions (2059)
3619 - don't evict page beyond end of stripe extent (2925)
3620 - don't oops on a deleted current working directory (2399)
3621 - handle hard links to targets without a parent properly (2517)
3622 - don't dereference NULL lock when racing during eviction (2867)
3623 - don't grow lock extents when lots of conflicting locks (2919)
3625 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3628 - account for cache space usage on clients to avoid data loss (974)
3629 - lfsck support in lustre kernel code (2349)
3630 - reduce journal credits needed for BRW writes (2370)
3631 - orphan handling to avoid losing space on client/server crashes
3632 - ptlrpcd can be blocked, stopping ALL progress (2477)
3633 - use lock value blocks to assist in proper KMS, faster stat (1021)
3634 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3635 - recovery for initial connections (2355)
3636 - fixes for mds_cleanup_orphans (1934)
3637 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3638 - block all file creations until orphan recovery completes (1901)
3639 - client remove rq_connection from request struct (2423)
3640 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3641 - recovery timer race (2670)
3642 - mdc_close recovey bug (2532)
3643 - ptlrpc cleanup bug (2710)
3644 - mds timeout on local locks (2588)
3645 - namespace lock held during RPCs (2431)
3646 - handle interrupted sync write properly (2503)
3647 - don't try to handle a message that hasn't been replied to (2699)
3648 - client assert failure during cleanup after abort recovery (2701)
3649 - leak mdc device after failed mount (2712)
3650 - ptlrpc_check_set allows timedout requests to complete (2714)
3651 - wait for inflight reqs when ptlrpcd finishes (2710)
3652 - make sure unregistered services are removed from the srv_list
3653 - reset bulk XID's when resending them (caught by 1138 test)
3654 - unregister_bulk after timeout
3655 - fix lconf error (2694)
3656 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3657 - readahead locks pages, leaves pending causing memory pressure (2673)
3658 - increase OST request buffers to 4096 on large machines (2729)
3659 - fix up permission of existing directories in simple_mkdir (2661)
3660 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3661 - don't assign transno to errored transactions (2742)
3662 - don't delete objects on OST if given a bogus objid from MDS (2751)
3663 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3664 - if rq_no_resend, then timeout request after recovery (2432)
3665 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3666 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3667 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3668 - MDS recovery shouldn't delete objects that it has given out (2730)
3669 - if enqueue arrives after completion, don't clobber LVB (2819)
3670 - don't unlock pages twice when trigger_group_io returns error (2814)
3671 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3672 - don't write pages to disk if there was an error (1450)
3673 - don't ping imports that have recovery disabled (2676)
3674 - take buffered bytes into account when balancing socknal conn (2817)
3675 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3676 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3677 - remove little-endian swabbing of llog records (1987)
3678 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3679 - retry reposting request buffers if they fail (1191)
3680 - grow extent at grant time to avoid granting a revoked lock (2809)
3681 - lock revoke doesn't evict page if covered by a second lock (2765)
3682 - disable VM readahead to avoid reading outside lock extents (2805)
3684 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3685 - updated kernel patches for hp-2.4.20 kernel (2681)
3687 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3690 - fix truncated write corruption (2366)
3691 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3693 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3694 - ensure len(uuid) < 37 in lmc (1171)
3695 - fix ia64 OOPS in llog_test (2255)
3696 - zero end of page at obdfilter for partial page writes (2648)
3697 - don't leave stale dentries around after renames (bug 2428)
3698 - fix timeouts when evicting a client with a single lock held (2642)
3699 - set deadline for the initial HELLO message to drain (2634)
3700 - print out dotted-quad IP addresses in the socknal (2302)
3702 - additional debugging for MDS client eviction problem (2443)
3703 - fix mkfsoptions support for osts (2603, 2604)
3705 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3708 - add series for the vanilla 2.6.0 kernel
3709 - add series for the vanilla 2.4.24 kernel
3710 - add series for a cray x86/64 UL kernel drop
3711 - fix xattr patches for the vanilla 2.4.19 series
3713 - generate true UUIDs in lmc (1171)
3714 - have portals stack dumping break in UML (2466)
3715 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3716 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3717 - wake callback waiting threads on client eviction (2460)
3718 - Add --ptldebug and --subsystem to lmc (1719)
3719 - update assertion to allow safe interrupt allocation
3720 - set rq_no_resend for cancel requests (2432)
3721 - recalculate ptlrpcd timeout after resend (2494)
3722 - call vfs_rmdir when removing pending directories (2368)
3723 - fix renaming a file to itself (2429)
3724 - lmc creates a default one-stripe lov (2454)
3725 - expand procfs space to handle large clusters (2326)
3726 - increase UML stack to avoid overflow
3727 - update lconf's list of debug and subsystem masks
3728 - fix lfs find --obd (2510)
3729 - /proc tunable for disabling filter read caching (2591)
3730 - stop rpm packages from altering slapd.conf (2301)
3731 - disable nagle in the socknal under 0conf (2578)
3732 - choose mds inode size based on stripe count (2572)
3733 - fix kernel-source rpm problems (2516)
3735 - add --disable-doc to avoid pdf generation (2421)
3736 - update documentation, tests, type-os, comments
3737 - avoid format warnings on ia64
3738 - remove the TOE NAL
3739 - tiny code cleanups by removing unused fields
3741 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3744 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3745 - remove the most blatant lies from BUILDING file (2371)
3746 - change default debug level to reasonable production setting
3747 - reduce client side cache size to reduce cache flush time
3748 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3749 - make TCP zerocopy and pinger support enabled by default (2476)
3750 - sync writes completed after process exits caused crashes (2319)
3751 - maintain correct mount count on the MDS (2356)
3752 - backout 1557, because 2316 wasn't really fixed
3753 - better file I/O statistics gathering in /proc
3754 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3755 - another kernel patch to fix zero-copy TCP function export
3756 - don't take duplicate lock when processing re-sent getattr (2420)
3757 - lctl uses obd_self_export instead of creating new conn (2353)
3758 - MDS/OST recovery case which requires object creation asserted (2425)
3759 - move lfs from /usr/sbin to /usr/bin in packages
3760 - fix race between mds_client_add and mds_client_free (2417)
3761 - use kmalloc instead of slabs in portals (2430)
3762 - don't create duplicate records when a failover MDS is present (2442)
3763 - remove unnecessary mount age check (2332)
3764 - don't remove directory inodes from locks prematurely (2451)
3765 - don't break if MDS service name is the same as hostname (2103)
3766 - fix races in client write RPC generation when cache full (2482)
3768 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3771 - remove now-unused request->rq_obd (278)
3772 - if an allocation fails, print out how much memory we've used (1933)
3773 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3774 - add the "configurable stack size" patch to most series files (1256)
3775 - ability to write large log records, for 100+ OST configs (2306)
3776 - fix NULL deref when filter_prep fails (2314)
3777 - fix operator precedence error in filter_sync
3778 - dynamic allocation of socknal TX descriptors (2315)
3779 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3780 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3781 - ensure that CWARN messages in Portals always get to the syslog
3782 - __init/__exit are not for prototype decls (ldlm_init/exit)
3783 - x86-64 compile warning fixes
3784 - fix gateway LMC keyword conflict (2318)
3785 - fix MDS lock inversions in getattr/reint paths (1844)
3786 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3787 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3788 - DLM race condition prevented some lock evictions (2328)
3789 - ENOMEM detection and retry on socknal sends (2230)
3790 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3791 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3792 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3793 - better error messages when a client is rejected during recovery (1505)
3794 - avoid cancelling locks which were never granted, after failure (2330)
3795 - fix i_sem/journal inversion in mds_client_add (2333)
3796 - fix truncate/getattr lock cycle deadlock (2334)
3797 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3798 - fix two rare exit paths which could leak an l_lock() ref (2321)
3799 - fencepost error in MDS/OST orphan recovery (2226)
3800 - make log record alignment 8 bytes (1988)
3801 - lstripe now fails when requested offset > ost_count (2237)
3802 - ensure that all kernel series have a complete list.h (1607)
3803 - fix crashes in special-file operations (2316)
3804 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3806 - allow configurable automake binary, for testing new versions
3807 - small update to the lfs documentation
3809 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3811 * fix negative export reference count in fsfilt_sync (2312)
3813 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3814 * release candidate 0.9.1
3816 - orphans are moved into the PENDING directory for possible recovery
3817 - replayed opens now open by fid for orphan/rename safety (1042)
3818 - last close of an orphan inode generates a transno (683)
3819 - chdir() and mount() now pin the directory entry (1020)
3820 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3821 - discard very old requests without processing them (1502)
3822 - remove some common, well-understood CERRORs (1505)
3823 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3824 - clear "grant" flags in OST replies until OST grant code lands (1644)
3825 - fix read performance by not clobbering i_blksize on client (1598)
3826 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3827 - make LRU size a /proc tunable, clears locks when reduced (707)
3828 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3829 - server locks take references on exports now (1558)
3830 - build fixes for 2.4.20-rh trees (1663)
3831 - return an error from lov_create if all OSCs are inactive (1751)
3832 - fix import levels when a reconnect happens without a timeout (1597)
3833 - exit early from mds_open if we get a lookup error (1749)
3834 - partial page read at EOF wouldn't wait for disk before sending (1642)
3835 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3836 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3837 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3838 - write/truncate lock inversion (1639)
3839 - Don't auto-load obdclass, portals modules during cleanup (1495)
3840 - fix timestamps from jumping to "now" (1763)
3841 - extra journal assertions (1648)
3842 - add an extra multiunlink test (1771)
3843 - fix read_record/write_record API (1776)
3844 - fix leak of offset_extent, possible incorrect i_size later (1772)
3845 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3846 - Debugging for the kqswnal_get_idle_tx problems (1820)
3847 - Allow recovery to be attempted multiple times (1536)
3848 - Write out MDS last_rcvd file after it is first created (1600)
3849 - Fix tx_descriptor leak in failed transmit situations (1827)
3850 - ext3 journaling fixes for assertion failure after IO error (1871)
3851 - class_export_put() on freed export after completion AST error (1896)
3852 - Fix revalidate looping in VFS (1322)
3853 - Don't access a freed export during MDS_REINT timeout (1521)
3854 - Add open-unlink recovery support on the MDS (1673,1764)
3855 - Return an error if no MDS data was read from last_rcvd (1946)
3856 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3857 - Don't setup a disk device that doesn't match exported UUID (317)
3858 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3859 - avoid committing NULL handle in force close
3860 - local.sh is now a one-stripe LOV configuration
3861 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3862 - fix ext3 htree duplicate directory entry corruption (1516)
3863 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3864 - update to new LOV EA format (2097)
3865 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3867 2003-06-15 Phil Schwan <phil@clusterfs.com>
3870 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3871 - per-import/export recovery handling (958, 931, 959)
3872 - multiple last-rcvd slots, for serving multiple FSes (949)
3873 - connections are again shared between multiple imp/exports (963, 964)
3874 - "umount -f" would hang if any requests needed to be sent (393, 978)
3875 - avoid pinning large req buffer by copying for queued messages (989)
3876 - add "uuid" to "lctl device" command to help upcalls (991)
3877 - "open" RPCs with transnos would confuse recovery counters (1037)
3878 - do proper endian conversion of all wire messages (288, 340, 891)
3879 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3880 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3881 - fix ptlrpc_connection leak in target_handle_connect (1174)
3882 - fix import refcounting bug in OST and MDS cleanup (1134)
3883 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3884 - fix very unlikely obd_types race condition (501)
3885 - remove osc_open hack for echo_client (1187)
3886 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3887 - a failure in read_inode2 leads to deadlock (1139)
3888 - cancel ack-locks as soon as transaction is committed (1072)
3889 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3890 - make sure to commitrw after any preprw to avoid deadlock (1162)
3891 - failing to execute a file in a lustre FS would lock inode (1203)
3892 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3893 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3894 - obd_uuid2tgt would walk off the end of the list (1255)
3895 - on IA64 the osc would give portals incorrect bulk size (1258)
3896 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3897 - fix lock inversion caused by new llite matching code (1282)
3898 - limit the number of dirty pages on a client to 10MB (1286)
3899 - timed out locks were not being corrected cancelled (1289)
3900 - fix O_DIRECT above 4GB on IA-32 (1292)
3901 * major user-visible changes
3902 - fail out/fail over policy now controlled by the upcall (993)
3904 - add OBD_PING to check server availability and failure (954)
3905 - lustre messages are now sent in sending host order (288, 340, 891)
3906 - add eadatalen to MDS getattr reply (340)
3907 - OST read replies may contain second buffer, with per-page status (593)
3909 2003-03-11 Phil Schwan <phil@clusterfs.com>
3912 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3913 - failed open()s could cause deadlock; fixed (867, 869)
3914 - stop cancelling OST locks when files are closed (481)
3915 - overlapping XID spaces caused network corruption (851, 853)
3916 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3917 - setattr_raw added, to do single-RPC, server-side setattrs
3918 - lmc/lconf syntax change for OST UUIDs
3919 - fix crashy race condition between ptlrpc_free_req and osc_close
3920 - don't use request in mdc_enqueue if we hit a timeout (889)
3921 - don't set the inode i_size for regular files from the MDS (896)
3922 - handle out of order completion AST (842)
3923 - don't LBUG if a lock request times out after receiving AST (913)
3924 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3925 - if a bad lock AST arrives, send an error instead of dropping entirely
3926 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3927 - fix leak in bulk IO when only partially completed (899, 900, 926)
3928 - fix O_DIRECT for ia64 (55)
3929 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3930 - C-z after timeout could hang a process forever; fixed (977)
3932 - client-side I/O cache (678, 924, 929, 941, 970)
3934 - READPAGE and SETATTRs which don't take server-side locks get
3937 2003-02-11 Phil Schwan <phil@clusterfs.com>
3940 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3941 - Fix fsfilt_extN_readpage() to read a full page of directory
3942 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3943 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3944 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3945 - Most intent operations were removed, in favour of a new RPC mode
3946 that does a single RPC to the server and bypasses most of the VFS
3947 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3948 - Aggressively cancel local locks on DLM servers
3949 - mds_reint_unlink sends EA to the client if it's the last nlink.
3950 client uses that EA to unlink OST objects.
3951 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3952 - recursive symlinks were fixed (439)
3953 - fixed NULL deref in DEBUG_REQ
3954 - filter_update_lastobjid no longer calls sync, which annoyed extN
3955 - fixed multi-client small-writes to a single file problem (445)
3956 - fixed mtime updates during file writes (607)
3957 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3958 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3959 - fixed Linux OST size reporting problem (444, 656)
3960 - OST now updates object mtime with writes or setattr (607, 619)
3961 - client verifies file size before zeroing page past EOF (445)
3962 - OST now writes last allocated objid to disk with allocation (108)
3963 - LOV on echo now works (409)
3965 - mds_reint_unlink sends a new buffer, with the EA included. this
3966 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3967 set if a regular file was being unlinked, and it was the last link
3968 - use PtlGet from the target for bulk writes (315)
3969 - OST now updates object mtime with writes or setattr (607, 619)
3970 - LDLM now has a grant-time callback to revalidate locked items, if
3972 - Many MDS operations were reorganized to combat race conditions
3974 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3975 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3977 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3980 - Fully reactivate OST imports after reconnection (512, others)
3981 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3982 - More graceful error handling for truncating on dead OST (515)
3983 - Don't error out unless we're actually accessing dead stripes (474)
3984 - Fix garbage sizes when stripes are missing (410)
3985 - LRU counters were broken, causing constant lock purge (433, 432)
3986 - garbage on read from stripes with failed OSTs (441)
3987 - mark OSCs as active before reconnecting during recovery (438)
3988 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3989 - lfind did not preserve OST order in output (443)
3990 - symlinks cause hung clients, incorrect data (439)
3991 - stop dereferencing request after dropping refcount (457)
3992 - don't LASSERT(spin_is_locked) on non-SMP (455)
3993 - fixes for many rename() bugs
3994 - fstat didn't correctly synchronize attributes (399)
3995 - server must handle lock cancellation during blocking AST prep (487)
3996 - bulk descriptors were free()d too soon (511)
3997 - fix paths in lconf, which would load incorrect modules (451, 507)
3998 - fix confusing lconf 'host not found' error message (386)
3999 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4000 - fix race condition in mdc_blocking_ast() for inode access (526)
4001 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4002 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4003 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4004 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4005 - fix lstripe in several ways (kernel side) (527)
4006 - fix request leak in ldlm_cli_enqueue (262)
4007 - incorrect OSC was marked inactive after OST failure
4008 - call mds_fs_cleanup before unmounting filesystem (524)
4009 - fix races between taking ns_lock and ldlm_lock_change_resource
4010 - fix races updating LOV export open file list
4011 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4012 - fix recovery NULL deref in ldlm_cli_cancel_unused
4013 - fix some DLM races by using new hash table for lock handles (419)
4014 - permit the client to specify desired inodes, at replay
4015 - duplicate requests when we queue them for replay reintegration
4016 - fix last_rcvd offset calculation
4017 - sync after each recovered transaction, so we always make progress
4018 - never, not always, ERESTART requests without transnos
4019 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4021 - skip replay if the MDS says that the client is already connected
4022 - don't check for a recovery-enabled export to match lctl's UUID
4023 - don't INC_USE_COUNT for phantom exports
4024 - don't crash when cleaning up phantom exports (567)
4025 - don't double-finish or set replay data for errored mdc_open requests
4026 - abort requests when they time out, so we don't get old replies
4027 - send/receive replies for AST messages again
4028 - if the client says that it doesn't have the lock, cancel it on the
4030 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4031 mark it as destroyed, it will all work out when decref is called
4032 - fix module use counts (22, 581)
4034 - ASTs now expect a reply (server cancels lock on error reply)
4036 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4039 - fix many simultaneous client startup (392)
4040 - fix dentry->d_it clobbering
4041 - credentials weren't being shipped for readdir/getattr operations
4042 - remove invalid assertions triggered during some concurrent MD
4044 - proper Lustre versions added (336, 389)
4045 - fix memory leak for create error case (398)
4046 - fix LOV locking bug that would get cli/srv out of sync
4047 - fix echo client over LOV (409)
4048 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4049 - fix double-O_EXCL intent crash (424)
4050 - avoid sending multiple lock CANCELs (352)
4052 - MDS can do multi-client recovery (modulo bugs in new code)
4054 - many updates, edits, cleanups
4056 2002-11-18 Phil Schwan <phil@clusterfs.com>
4059 - fix null d_it dereference (346)
4060 - fix full OST/dbench hang (333)
4061 - fix permission problem with file removal (286)
4062 - fix removal of OSCs from LOV when they fail
4063 - fix NULL deref during bulk timeout (214)
4064 - fix problems related to multiple filesystems on one MDS (241)
4065 - fixed serious subtle metadata locking bugs
4066 - free locks on clients when inodes are removed due to memory
4068 - fix inode pointer in lock data (285)
4069 - partial support for multiple MDS on a single host (241)
4070 - data locks weren't cancelled at clear_inode time (290, 311)
4071 - intent locks could lead to unbounded lock growth (205)
4072 - added a maximum lock count, an LRU list, and a flusher
4073 - fix multiple rename (365)
4074 - properly abstracted the echo client
4075 - OSC locked 1 byte too many; fixed
4076 - rewrote brw callback code:
4077 - fixed recovery bugs related to LOVs (306)
4078 - fixed too-many-pages-in-one-write crash (191)
4079 - fixed (again) crash in sync_io_timeout (214)
4080 - probably fixed callback-related race (385)
4082 - Add capability to MDS protocol
4083 - LDLM cancellations and callbacks on different portals
4085 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4088 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4089 - defer open object destruction to close time (601981, 138)
4090 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4091 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4092 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4093 - don't call obd_setattr in ll_file_release for destroyed objects
4094 * protocol change to lustre_msg: move |version| and add |flags|
4095 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4096 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4097 * added replay of create, unlink, link and rename operations during
4098 MDS failover; recovery should be much more robust now
4099 * remove failed OSCs from LOVs (only lov_create uses this so far)
4100 * the lustre-HOWTO was brought (more) up to date (582544)
4102 2002-10-23 Phil Schwan <phil@clusterfs.com>
4105 - in-use dentries weren't being reused properly (617851)
4106 - prevent multiple LDLM setup (599178)
4107 - fix LOV size calculations for truncate (617853)
4108 - fix client handling of MDS intent errors (POSIX)
4109 - fix permission bug in lovstripe.c test (624321)
4110 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4111 - truncate past end of file could corrupt data
4112 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4113 - a race in recovery could return ETIMEDOUT to apps (623947)
4114 - building outside the source directory was fixed
4115 * the lustre-HOWTO was brought (more) up to date (582544)
4116 * major progress was made on recovery functionality
4118 2002-10-10 Phil Schwan <phil@clusterfs.com>
4121 - recovery deadlock fix
4122 - rm -rf causes LBUG fix (617817)
4123 - file open by multiple tasks fix (618962)
4124 - directory permissions bugs (602707 and 620007)
4125 - journal_stop fixed with locking (611313)
4126 - O_APPEND failures resolved (618273, perhaps 614459)
4127 - lconf PATH fix (619770)
4128 - IA64 build fix (621450)
4129 - RPC buffer sizes scale with amount of memory
4131 2002-10-01 Phil Schwan <phil@clusterfs.com>
4134 - locks would be cancelled without throwing away data pages,
4135 resulting in inconsistent data (605627)
4136 - inode attributes were not always being refreshed (605627, 612449)
4137 - lconf now continues to cleanup after lctl reports an error
4138 - MDS now enforces user permissions (602707)
4139 - lprocfs cleanup fixed, but not yet enabled (614157)
4140 - fixed infinite server hang, should a client not respond to an AST
4141 - avoid going into recovery if user calls readlink() with a buffer
4142 that's too small (613941)
4143 - AST RPCs no longer require replies (614867) -- this may be changed
4144 - don't crash server if client sends an IOV that's too big (611336)
4145 - fixed lock conversion deadlock (611892)
4146 - fixed the following of symlinks (614622)
4147 * recovery: the server can remove locks from a client that dies, other
4148 clients can make progress
4149 * more extN patch fixes
4150 * compile-time configurable ptlrpc buffer allocations
4152 - collaborative read cache document
4153 - Lustre Lite Performance CDR document-in-progress
4155 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4158 - fix typo in patch-2.4.18
4160 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4163 - clear ptlrpc request each time in handle_incoming_request()
4164 - unlink of files now destroys the object on the OST
4166 2002-09-19 Peter Braam <braam@clusterfs.com>
4168 * add hard link support
4169 * change obdfile creation method
4170 * kernel patch changed
4172 2002-09-19 Peter Braam <braam@clusterfs.com>
4175 - stack overflow bug in extN fixed
4177 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4179 * documentation updates
4180 - add man pages for config tools
4181 - update tests/README to describe testing with new config tools
4182 - finish metadata API descriptions
4183 * bug fixes and cleanups
4184 - statfs workaround for 16TB limit
4185 - LOV stripe allocation improved, can stripe on subset of OSTs
4186 - LOV file size/IO offset was wrong for files > 4GB in size
4187 - object EA data was being dropped, caused files to be unreadable
4188 - memory overflow with non-LOV OST caused memory corruption
4189 - fixed regression tests to work with new config tools, obdfilter
4190 - fixed bug when directory size became larger than 1 block
4191 - fixed bug (for single client case) when PWD was deleted
4192 - invalidate local directory pages when doing intent-based ops
4193 - avoid LDLM oops when lock callback contained bad data
4195 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4197 * documentation updates
4198 * bug fixes and cleanups
4199 - configuration tools
4202 - 64-bit compile warnings
4203 - 64-bit internal statfs data
4205 * test_brw on persistent OST devices
4207 * lprocfs (disabled)
4209 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4211 * documentation updates
4212 * bug fixes and cleanups
4213 * configuration tools
4215 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4218 * many small fixes to 0_5_4
4219 * io/network handling
4220 * thinkos in MDS operations
4222 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4225 * crucial basic fixes to 0.5.3
4226 * IOR, Iozone work over Elan
4229 2002-08-07 Phil Schwan <phil@clusterfs.com>
4230 * version 0_5_3, our first alpha
4231 * we use the new Portals iovs
4232 * documentation updates
4233 * bug fixes and cleanups
4234 * small changes in the DLM wire protocol
4236 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4237 * version 0_5_1 with some initial stability,
4238 * locking on MD and file I/O.
4239 * documentation updates
4240 * several bug fixes since 0.5.0
4241 * small changes in wire protocol
4243 2002-07-18 Phil Schwan <phil@clusterfs.com>
4245 * delivered as Lustre Light Alpha
4246 * fixed a crash after handling invalid MDS requests
4247 * fixed directory pages for architectures with non-4k pages sizes
4249 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4250 * release version v0_4_4
4251 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4253 * Fixes a number of interruption problems with OST operations.
4254 * Update documentation for portals header changes
4255 * Move all wire protocol structs/defines to lustre_idl.h
4256 * Fixes symlink length bug.
4257 * Add tcpdump to repository.
4259 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4260 * release version v0_4_3
4261 * Fixes statfs for inodes on extN.
4262 * Fixes bug in runtests which would delete /etc/hosts.
4263 * Use 64-bit object IDs wherever possible (not into VFS though)
4264 Remove ost_get_info, which is unused by lustre, and out of date.
4266 2002-07-03 Peter Braam <braam@clusterfs.com>
4267 * release version v0_4_2 Fixes a lookup error (type not passed)
4268 * move forward to head of Portals
4269 * move forward to latest Lustre kernel
4271 2002-06-25 Peter Braam <braam@clusterfs.com>
4272 * release version v0_4_1. Hopefully stable on single node use.