4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.53-0.8 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: Update to RHEL4 latest kernel.
21 Description: Remove 2.4 kernel support from lustre 1.8.0
22 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
23 quota_interface.c in quota subsystem.
24 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
26 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
27 lvfs_linux.c upcall_cache.c in lvfs subsystem.
28 Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
29 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
31 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
32 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
34 Remove 2.4 kernel check from lproc_osc.c, osc_request.c in osc
36 Remove 2.4 kernel check from filter.c, lproc_obdfilter.c in
38 Remove 2.4 kernel code from lustre/obdecho/echo_client.c.
39 Remove 2.4 kernel check from lustre/mgs/lproc_mgs.c,
41 Remove 2.4 kernel code from lustre/mgc/mgc_request.c.
42 Remove 2.4 kernel check from lustre/mdt/mdt_lproc.c.
43 Remove 2.4 kernel code from lustre/mdc/mdc_request.c.
44 Remove 2.4 kernel check from lustre/lov/lproc_lov.c.
45 Remove 2.4 kernel check from lustre/lmv/lproc_lmv.c.
46 Remove lock_24kernel from lustre/lvfs/fsfilt_ext3.c.
47 Remove 2.4 kernel code from include/liblustre.h,
48 include/lustre_disk.h, include/lustre_quota.h,
49 include/linux/lprocfs_status.h, include/linux/lvfs.h,
50 include/linux/lustre_lite.h, include/linux/lustre_net.h,
51 include/linux/lustre_user.h, include/linux/lvfs_linux.h,
52 include/linux/obd_class.h, include/linux/lustre_compat25.h.
54 Severity : enhancement
56 Description: Build SLES10 patchless client fails
57 Details : The configure was broken by run ./configure with
58 --with-linux-obj=.... argument for patchless client. When the
59 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
60 can't search header adequately. Use absolute path such as
61 -I($LINUX)/include instead.
65 Description: interrupt oig_wait produce painc on resend.
66 Details : brw_redo_request can be used for resend requests from ptlrpcd and
67 private set, and this produce situation when rq_ptlrpcd_data not
68 copyed to new allocated request and triggered LBUG on assert
69 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
70 ptlrpcd set if request is changed and can be safety changed to use
75 Description: LASSERT_{REQ,REP}SWAB macros are buggy
76 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
77 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
78 these macros and replace them with their difinitions instead.
82 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
83 Details : Update which_patch & target file for SLES10 latest kernel.
86 Description: Remove client patches from SLES 10 kernel.
87 Details : This causes SLES 10 clients to behave as patchless clients
88 even on a Lustre-patched (server) kernel.
90 Severity : enhancement
92 Description: self-adjustable client's lru lists
93 Details : use adaptive algorithm for managing client cached locks lru
94 lists according to current server load, other client's work
95 pattern, memory activities, etc. Both, server and client
96 side namespaces provide number of proc tunables for controlling
99 Severity : enhancement
101 Description: light-weight GSS support
102 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
103 and reduce performance overhead.
105 Severity : enhancement
107 Description: Linux keyring support
108 Details : Support using service of Linux keyring for Lustre GSS internal
109 context refresh/cache mechanism.
113 Description: Fix errors in lfs documentation
114 Details : Fixes man pages
118 Description: don't use GFP_* in generic Lustre code.
119 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
123 Description: obdclass is limited by single OBD_ALLOC(idarray)
124 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
128 Description: warnings with build patchless client with vanila 2.6.19 and up
129 Details : change old ctl_table style and replace ctl_table/ctl_table_header
130 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
134 Frequency : Only for SLES
135 Description: sanity_quota fail test_1
136 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
137 file, which makes TSTID[2] inunique.
141 Frequency : Only for SLES9
142 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
143 Details : Update target/ChangeLog/which_patch .
147 Frequency : Only for test_10 in sanity-quota.sh
148 Description: a bug in quota test code
149 Details : checking a test flag in a wrong place
153 Frequency : Only for test_13 in sanity-quota.sh
154 Description: a bug in quota test script
155 Details : shouldn't check overfull allocated quota
159 Description: jbd statistics
160 Details : Port older jbd statistics patch for sles10
162 Severity : enhancement
164 Description: Add printing inode info into message about error in writepage.
169 Description: reply_lock_interpret crash due to race with it and lock cancel.
170 Details : Do not replay locks that are being cancelled. Do not reference
171 locks by their address during replay, just by their handle.
175 Frequency : When flocks are used.
176 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
177 Details : Flock locks might destroy just granted lock if it could be merged
178 with another existing flock, this is done in completion handler,
179 so teach ldlm_cli_enquque_fini that this is a valid case for
185 Description: Oops in read and write path when failing to allocate lock.
186 Details : Check if lock allocation failed and return error back.
190 Description: lstripe command fails for valid OST index
191 Details : The stripe offset is compared to lov->desc.ld_tgt_count
192 instead of lov->desc.ld_active_tgt_count.
196 Description: sanity.sh failed test 103
197 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
202 Description: df doesn't work properly if diskfs blocksize != 4K
203 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
207 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
208 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
212 Description: obd_health_check_timeout too short
213 Details : set obd_health_check_timeout as 1.5x of obd_timeout
217 Description: llapi_file_create() does not allow some changes
218 Details : add llapi_file_open() that allows specifying the mode and
219 open flags, and also returns an open file handle.
223 Description: merge and cleanup kernel patches.
224 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
228 Description: Add journal checksum support.(Kernel part)
229 Details : The journal checksum feature adds two new flags i.e
230 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
231 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
232 indicates that the commit block contains the checksum for
233 the blocks described by the descriptor blocks. Now commit
234 record can be sent to disk without waiting for descriptor
235 blocks to be written to disk. This behavior is controlled
236 using JBD2_FEATURE_ASYNC_COMMIT flag.
240 Description: OSS needs mutliple precreate threads
241 Details : Add ability to start more than one create thread per OSS.
245 Description: Sanity.sh test_65k failed: setstripe should have succeeded
246 Details : execute lfs setstripe on client
250 Description: mds_obd_create error creating tmp object
251 Details : When the user sets quota on root, llog will be affected and can't
252 create files and write files.
255 Frequency : Always on ia64 patchless client, and possibly others.
257 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
258 Details : This allows the patchless client to be loaded on architectures
263 Description: RedHat Update kernel for RHEL5
264 Details : Add the kernel config file for RHEL5.
268 Description: RedHat Update kernel for RHEL5
269 Details : Modify the kernel config file more closer RHEL5.
273 Description: Build failure against Centos5 (RHEL5)
274 Details : Define PAGE_SIZE when it isn't present.
278 Description: test_7 on sanity-quota.sh can't be used on separate machines
279 Details : add facet to handle it
283 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
284 Details : with NFS, the anon dentry's parent was set to itself in
285 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
286 corresponding dentry other than use rec->ur_name.
288 Severity : enhancement
290 Description: lfs setstripe enhancement
291 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
295 Description: testing performance impact of enabling checksumming
296 Details : enable checksum by default, allow --disable-checksum
297 configure option and "-o nochecksum" mount option
301 Description: lustre support for RHEL5
302 Details : Add support for RHEL5.
306 Description: Client eviction due to failover config
307 Details : after a connection loss, the lustre client should attempt to
308 reconnect to the last active server first before trying the
309 other potential connections.
313 Description: when mds and osts use different quota unit(32bit and 64bit),
314 quota will be released repeatly.
315 Details : void sending multiple quota reqs to mds, which will keep the
316 status between the reqs.
320 Description: osts not allocated evenly to files
321 Details : change the condition to increase offset_idx
324 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
326 Description: Data corruption for OSTs that are formatted larger than 2TB
328 Details : When generating the bio request for lustre file writes the
329 sector number would overflow a temporary variable before being
330 used for the IO. The data reads correctly from Lustre (which
331 will overflow in a similar manner) but other file data or
332 filesystem metadata may be corrupted in some cases.
336 Description: Tune the kernel for good SCSI performance.
337 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
338 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
343 Description: rewrite ext2-derived code in obdclass/uuid.c
344 Details : rewrite inherited code (uuid parsing code from ext2 utils)
345 from scratch preserving functionality.
349 Description: "lfs find -obd UUID" prints directories
350 Details : "lfs find -obd UUID" will return all directory names instead
351 of just file names. It is incorrect because the directories
352 do not reside on the OSTs.
356 Description: use i_size_read and i_size_write in 2.6 port
357 Details : replace inode->i_size access with i_size_read/write()
360 Frequency : when removing large files
362 Description: scheduling issue during removal of large Lustre files
363 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
364 It causes scheduling issues when removing large files (17TB in the
368 Frequency : only with liblustre clients on XT3
370 Description: evictions taking too long
371 Details : allow llrd to evict clients directly on OSTs
374 Frequency : only on ppc
376 Description: /proc/fs/lustre/devices broken on ppc
377 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
378 We should make sure the type of 'index' is loff_t and then cast
379 explicitly as needed below. Do not assign an explicitly cast
383 Frequency : only for rhel5
385 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
386 Details : Modify the target file & which_kernel.
388 Severity : enhancement
390 Description: omit set fsid for export NFS
391 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
392 as FS_REQUIRES_DEV for avoid problems with generate fsid.
396 Frequency : Always, for kernels after 2.6.16
397 Description: Fix warning idr_remove called for id=.. which is not
399 Details : Last kernels save old s_dev before kill super and not allow
400 to restore from callback - restore it before call kill_anon_super.
404 Description: buffer overruns could theoretically occur
405 Details : llapi_semantic_traverse() modifies the "path" argument by
406 appending values to the end of the origin string, and a
407 overrun may occur. Adding buffer overrun check in liblustreapi.
411 Description: Fix error on 'ls .' at the top of the Lustre mount.
412 Details : Don't revalidate dentry if it is a root dentry.
416 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
417 Details : Modify vm-tunables-rhel4.patch.
421 Description: Kernel config for 2.6.18-vanilla.
422 Details : Modify targets/2.6-vanilla.target.in.
423 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
424 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
425 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
426 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
430 Description: improve handling recoverable errors
431 Details : if request processig with error which can be recoverable on server
432 request should be resend, otherwise page released from cache and
437 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
438 Details : Modify target file & which_patch.
439 A flaw was found in the IA32 system call emulation provided
440 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
441 value could be stored in the %RAX register, which could trigger an
442 out-of-bounds system call table access. An untrusted local user
443 could exploit this flaw to run code in the kernel
444 (ie a root privilege escalation). (CVE-2007-4573).
448 Description: O_DIRECT bypasses client statistics.
449 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
450 read_ahead_stats were updated. copied the stats section from
451 osc_send_oap_rpc() into async_internal().
455 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
459 Description: change order of libsysio includes
460 Details : '#include sysio.h' should always come before '#include xtio.h'
464 Description: handle "serious error: objid * already exists" more gracefully
465 Details : If LAST_ID value on disk is smaller than the objects existing in
466 the O/0/d* directories, it indicates disk corruption and causes an
467 LBUG(). If the object is 0-length, then we should use the existing
468 object. This will help to avoid a full fsck in most cases.
470 Severity : enhancement
472 Description: adapt the lustre_config script to support the upgrade case
473 Details : Add "-u" option for lustre_config script to support upgrading 1.4
474 server targets to 1.6 in parallel.
478 Description: To avoid grant space > avaible space when the disk is almost
479 full. Without this patch you might see the error "grant XXXX >
480 available" or some LBUG about grant, when the disk is almost
482 Details : In filter_check_grant, for non_grant cache write, we should
483 check the left space by if (*left > ungranted + bytes), instead
484 of (*left > ungranted), because only we are sure the left space
485 is enough for another "bytes", then the ungrant space should be
486 increase. In client, we should update cl_avail_grant only there
487 is OBD_MD_FLGRANT in the reply.
491 Description: Update RHEL 4 kernel to fix local root privilege escalation.
492 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
493 described in CVE-2007-4573. This problem could allow untrusted
494 local users to gain root access.
498 Frequency : when using O_DIRECT and quotas
500 Description: Incorrect file ownership on O_DIRECT output files
501 Details : block usage reported by 'lfs quota' does not take into account
502 files that have been written with O_DIRECT.
504 --------------------------------------------------------------------------------
506 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
508 * Support for kernels:
509 2.6.5-7.283 (SLES 9),
510 2.6.9-55.EL (RHEL 4),
511 2.6.16.46-0.14 (SLES 10),
512 2.6.18.8 vanilla (kernel.org)
513 * Client support for unpatched kernels:
514 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
515 2.6.16 - 2.6.22 vanilla (kernel.org)
516 * Due to recently discovered recovery problems, we do not recommend
517 using patchless RHEL 4 clients with this or any earlier release.
518 * Recommended e2fsprogs version: 1.39.cfs8
519 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
520 * Starting with this release, the ldiskfs backing filesystem required
521 by Lustre is now in its own package, lustre-ldiskfs. This package
522 should be installed. It is versioned separately from Lustre and
523 may be released separately in future.
527 Description: block reactivating mgc import until all deactivates complete
528 Details : Fix race when failing back MDT/MGS to itself (testing)
531 Frequency : at statup only
533 Description: mds_lov_synchronize race leads to various problems
534 Details : simultaneous MDT->OST connections at startup can cause the
535 sync to abort, leaving the OSC in a bad state.
537 Severity : enhancement
539 Description: add optional extra BUILD_VERSION info
540 Details : add a new environment variable (namely LUSTRE_VERS) which allows
541 to override the lustre version.
544 Frequency : 2.6.18 servers only
546 Description: ll_kern_mount() doesn't release the module reference
547 Details : The ldiskfs module reference count never drops down to 0
548 because ll_kern_mount() doesn't release the module reference.
553 Description: server LBUG when using old ost_num_threads parameter
554 Details : Accept the old ost_num_threads parameter but warn that it
555 is deprecated, and fix an off-by-one error that caused an LBUG.
560 Description: Transient SCSI error results in persistent IO issue
561 Details : iobuf->dr_error is not reinitialized to 0 between two
565 Frequency : sometimes when underlying device returns I/O errors
567 Description: OSTs not going read-only during write failures
568 Details : OSTs are not remounted read-only when the journal commit threads
569 get I/O errors because fsfilt_ext3 calls journal_start/stop()
570 instead of the ext3 wrappers.
574 Description: poor connect scaling with increasing client count
575 Details : Don't run filter_grant_sanity_check for more than 100 exports
576 to improve scaling for large numbers of clients.
579 Frequency : SLES10 only
581 Description: sanity-quota.sh quotacheck failed: rc = -22
582 Details : Quotas cannot be enabled on SLES10.
585 Frequency : liblustre clients only
587 Description: getdirentries does not give error when run on compute nodes
588 Details : getdirentries does not fail when the size specified as an argument
589 is too small to contain at least one entry
591 Severity : enhancement
593 Description: Add LNET router traceability for debug purposes
594 Details : If a checksum failure occurs with a router as part of the
595 IO path, the NID of the last router that forwarded the bulk data
596 is printed so it can be identified.
601 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
602 Details : Due to a race condition, liblustre clients were occasionally
605 Severity : enhancement
607 Description: lfs setstripe use optional parameters instead of postional
610 Severity : enhancement
612 Description: Nanosecond timestamp support for ldiskfs
613 Details : The on-disk ldiskfs filesystem has added support for nanosecond
614 resolution timestamps. There is not yet support for this at
615 the Lustre filesystem level.
618 Frequency : during server recovery
620 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
621 Details : request with rq_no_resend flag not awake l_wait_event if they get a
625 Frequency : nfs export on patchless client
627 Description: connectathon hang when test nfs export over patchless client
628 Details : Disconnected dentry cannot be found with lookup, so we do not need
629 to unhash it or make it invalid
632 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
633 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
634 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
639 Description: open req refcounting wrong on reconnect
640 Details : If reconnect happened between getting open reply from server and
641 call to mdc_set_replay_data in ll_file_open, we will schedule
642 replay for unreferenced request that we are about to free.
643 Subsequent close will crash in variety of ways.
644 Check that request is still eligible for replay in
645 mdc_set_replay_data().
650 Description: disable writes to filesystem when reading health_check file
651 Details : the default for reading the health_check proc file has changed
652 to NOT do a journal transaction and write to disk, because this
653 can cause reads of the /proc file to hang and block HA state
654 checking on a healthy but otherwise heavily loaded system. It
655 is possible to return to the previous behaviour during configure
656 with --enable-health-write.
658 Severity : enhancement
660 Description: 64-bit inode version
661 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
662 to the inode. This will be required for version-based recovery.
667 Description: MDS fails to start if a duplicate client export is detected
668 Details : in some rare cases it was possible for a client to connect to
669 an MDS multiple times. Upon recovery the MDS would detect this
670 and fail during startup. Handle this more gracefully.
672 Severity : enhancement
674 Description: Add -o localflock option to simulate old noflock
676 Details : This will achieve local-only flock/fcntl locks
682 Description: log_commit_thread vs filter_destroy race leads to crash
683 Details : Take import reference before releasing llog record semaphore
688 Description: Wrong request locking in request set processing
689 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
690 imp_delayed_list, in this place should be used imp_lock.
693 Frequency : when reconnection
695 Description: Grant Leak when osc reconnect to OST
696 Details : When osc reconnect ost, OST(filter) should check whether it
697 should grant more space to client by comparing fed_grant and
698 cl_avail_grant, and return the granted space to client instead
699 of "new granted" space, because client will call osc_init_grant
700 to update the client grant space info.
703 Frequency : when client reconnect to OST
705 Description: Grant Leak when osc do resend and replay bulk write
706 Details : When osc reconnect to OST, OST(filter)should clear grant info of
707 bulk write request, because the grant info while be sync between
708 OSC and OST when reconnect, and we should ignore the grant info
709 these of resend/replay write req.
714 Description: Grant space more than avaiable left space sometimes.
715 Details : When then OST is about to be full, if two bulk writing from
716 different clients came to OST. Accord the avaliable space of the
717 OST, the first req should be permitted, and the second one
718 should be denied by ENOSPC. But if the seconde arrived before
719 the first one is commited. The OST might wrongly permit second
720 writing, which will cause grant space > avaiable space.
723 Frequency : when client is evicted
725 Description: Grant might be wrongly erased when osc is evicted by OST
726 Details : when the import is evicted by server, it will fork another
727 thread ptlrpc_invalidate_import_thread to invalidate the
728 import, where the grant will be set to 0. While the original
729 thread will update the grant it got when connecting. So if
730 the former happened latter, the grant will be wrongly errased
731 because of this race.
736 Description: Checking Stale with correct fid
737 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
738 to check whether it is stale, because sometimes, we want the
739 enqueue happened anyway, and op_data.fid2 will not be initialized.
741 Severity : enhancement
743 Description: update patchless client
744 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
747 Frequency : only with 2.4 kernel
749 Description: random memory corruption
750 Details : size of struct ll_inode_info is to big for union inode.u and this
751 can be cause of random memory corruption.
756 Description: Memory leak in recovery
757 Details : Lov_mds_md was not free in an error handler in mds_create_object.
758 It should also check obd_fail before fsfilt_start, otherwise if
759 fsfilt_start return -EROFS,(failover mds during mds recovery).
760 then the req will return with repmsg->transno = 0 and rc = EROFS.
761 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
762 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
763 be freed no matter whether fsfilt_commit success or not.
766 Frequency : only with huge count clients
768 Description: Prevents from taking the superblock lock in llap_from_page for
770 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
771 for llap_from_page call in ll_removepage prevents from taking the
772 superblock lock for a soon died page.
777 Description: Not check open intent error before release open handle
778 Details : in some rare cases, the open intent error is not checked before
779 release open handle, which may cause
780 ASSERTION(open_req->rq_transno != 0), because it tries to release
781 the failed open handle.
786 Description: Set cat log bitmap only after create log success.
787 Details : in some rare cases, the cat log bitmap is set too early. and it
788 should be set only after create log success.
792 Description: Accessing a block bevice can re-enable I/O when Lustre is
793 tearing down a device.
794 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
799 Description: service threads may hog cpus when there are a lot of requests
801 Details : Insert cond_resched to give other threads a chance to use some of
807 Description: the cat log was not initialized in recovery
808 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
809 unlink log on mds will not be initialized until mds post
810 recovery. And also in mds post recovery, the unlink log will
811 initialization will be done asynchronausly, so there will be race
812 between add unlink log and unlink log initialization.
816 Description: brw_stats were being printed incorrectly
817 Details : brw_stats were being printed as log2 but all of them were not
818 recorded as log2. Also remove some code duplication arising from
819 filter_tally_{read,write}.
823 Frequency : rare, only in recovery.
824 Description: ASSERTION(req->rq_type != LI_POISON) failed
825 Details : imp_lock should be held while iterating over imp_sending_list for
826 prevent destroy request after get timeout in ptlrpc_queue_wait.
830 Description: replay-single.sh test 52 fails
831 Details : A lock's skiplist need to be cleanup when it being unlinked
832 from its resource list.
836 Description: Short directio read returns full requested size rather than
838 Details : Direct I/O operations should return actual amount of bytes
839 transferred rather than requested size.
841 Severity : enhancement
843 Description: metadata RPC reduction (e.g. for rm performance)
844 Details : decrease the amount of synchronous RPC between clients and servers
845 by canceling conflicing lock before the operation on the client side
846 and packing thier handles into the main operation RPC to server.
848 Severity : enhancement
850 Description: Async OSC create to avoid the blocking unnecessarily.
851 Details : If a OST has no remain object, system will block on the creating
852 when need to create a new object on this OST. Now, ways use
853 pre-created objects when available, instead of blocking on an
854 empty osc while others are not empty. If we must block, we block
855 for the shortest possible period of time.
857 --------------------------------------------------------------------------------
859 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
864 Frequency : on some architectures
866 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
867 Details : Uninitialized flags sometimes cause configuration commands to
871 Frequency : patchless clients only
873 Description: missing __iget() symbol export
874 Details : The __iget() symbol export is missing. To avoid the need for
875 this on patchless clients the deathrow inode reaper is turned
876 off, and we depend on the VM to clean up old inodes. This
877 dependency was during via the fix for bug 12181.
879 --------------------------------------------------------------------------------
881 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
883 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
884 INTEROPERATE with older versions automatically. In many cases a
885 special upgrade step is needed. Please read the
886 user documentation before upgrading any part of a live system.
887 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
888 version will not interoperate with 1.6 betas before beta5 (1.5.95).
889 * WARNING: Lustre configuration and startup changes are required with
890 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
896 Severity : enhancement
898 Description: MountConf
899 Details : Lustre configuration is now managed via mkfs and mount
900 commands instead of lmc and lconf. New obd types (MGS, MGC)
901 are added for dynamic configuration management. See
902 https://mail.clusterfs.com/wikis/lustre/MountConf for
905 Severity : enhancement
907 Description: dynamic OST addition
908 Details : OSTs can now be added to a live filesystem
910 Severity : enhancement
912 Description: startup order invariance
913 Details : MDTs and OSTs can be started in any order. Clients only
914 require the MDT to complete startup.
916 Severity : enhancement
918 Description: parallel, asynchronous orphan cleanup
919 Details : orphan cleanup is now performed in separate threads for each
920 OST, allowing parallel non-blocking operation.
922 Severity : enhancement
924 Description: optimized stripe assignment
925 Details : stripe assignments are now made based on ost space available,
926 ost previous usage, and OSS previous usage, in order to try
927 to optimize storage space and networking resources.
929 Severity : enhancement
931 Description: Permanently set tunables
932 Details : All writable /proc/fs/lustre tunables can now be permanently
933 set on a per-server basis, at mkfs time or on a live
936 Severity : enhancement
938 Description: Lustre message v2
939 Details : Add lustre message format v2.
941 Severity : enhancement
943 Description: client OST exclusion list
944 Details : Clients can be started with a list of OSTs that should be
945 declared "inactive" for known non-responsive OSTs.
949 Description: SPEC SFS validation failure on NFS v2 over lustre.
950 Details : Changes the blocksize for regular files to be 2x RPC size,
951 and not depend on stripe size.
953 Severity : enhancement
955 Description: Multiple MD RPCs in flight.
956 Details : Further unserialise some read-only MDS RPCs - learn about intents.
957 To avoid overly-overloading MDS, introduce a limit on number of
958 MDS RPCs in flight for a single client and add /proc controls
959 to adjust this limit.
961 Severity : enhancement
963 Description: client read/write statistics
964 Details : Add client read/write call usage stats for performance
965 analysis of user processes.
966 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
967 file access. extents_stats shows chunk size distribution.
968 extents_stats_per_process show chunk size distribution per
971 Severity : enhancement
973 Description: mds statistics
974 Details : Add detailed mds operations statistics in
975 /proc/fs/lustre/mds/*/stats.
979 Description: Failure of copying files with lustre special EAs.
980 Details : Client side always return success for setxattr call for lustre
981 special xattr (currently only "trusted.lov").
986 Description: Refcount LNET uuids
987 Details : The global LNET uuid list grew linearly with every startup;
988 refcount repeated list entries instead of always adding to
991 Severity : enhancement
993 Description: Dynamic service threads
994 Details : Within a small range, start extra service threads
995 automatically when the request queue builds up.
998 Frequency : mixed-endian client/server environments
1000 Description: mixed-endian crashes
1001 Details : The new msg_v2 system had some failures in mixed-endian
1004 Severity : enhancement
1006 Description: Easy OST removal
1007 Details : OSTs can be permanently deactivated with e.g. 'lctl
1008 conf_param lustre-OST0001.osc.active=0'
1010 Severity : enhancement
1012 Description: MGS proc entries
1013 Details : Added basic proc entries for the MGS showing what filesystems
1016 Severity : enhancement
1018 Description: provide MGS failover
1019 Details : Added config lock reacquisition after MGS server failover.
1021 Severity : enhancement
1023 Description: add Linux 2.4 support
1024 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1028 Description: a large application tries to do I/O to the same resource and dies
1029 in the middle of it.
1030 Details : Check the req->rq_arrival time after the call to
1031 ost_brw_lock_get(), but before we do anything about
1032 processing it & sending the BULK transfer request. This
1033 should help move old stale pending locks off the queue as
1034 quickly as obd_timeout.
1037 Frequency : when an incorrect nid is specified during startup
1039 Description: ptlrpc connect to non-existant node causes kernel crash
1040 Details : LNET can't be re-entered from an event callback, which
1041 happened when we expire a message after the export has been
1042 cleaned up. Instead, hand the zombie cleanup off to another
1045 Severity : enhancement
1047 Description: plain/inodebits lock performance improvement
1048 Details : Grouping plain/inodebits in granted list by their request modes
1049 and bits policy, thus improving the performance of search through
1053 Frequency : only if OST filesystem is corrupted
1055 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1056 Details : for a short time RPCs with bulk IO are in the replay list,
1057 but replay of bulk IOs is unimplemented. If the OST filesystem
1058 is corrupted due to disk cache incoherency and then replay is
1059 started it is possible to trip an assertion. Avoid putting
1060 committed RPCs into the replay list at all to avoid this issue.
1063 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1065 Description: System hang on startup
1066 Details : This bug allowed the liblustre (e.g. catamount) client to
1067 return to the app before handling all startup RPCs. This
1068 could leave the node unresponsive to lustre network traffic
1069 and manifested as a server ptllnd timeout.
1071 Severity : enhancement
1073 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1074 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1075 boolean to control whether to print peer debug info when a
1076 client's RPC times out.
1079 Frequency : only for kernels with patches from Lustre below 1.4.3
1081 Description: Remove old rdonly API
1082 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1085 Frequency : only for devices with external journals
1087 Description: Set external device read-only also
1088 Details : During a commanded failover stop, we set the disk device
1089 read-only while the server shuts down. We now also set any
1090 external journal device read-only at the same time.
1093 Frequency : when upgrading from 1.4 while trying to change parameters
1095 Description: The wrong (new) MDC name was used when setting parameters for
1096 upgraded MDT's. Also allows changing of OSC (and MDC)
1097 parameters if --writeconf is specified at tunefs upgrade time.
1100 Frequency : when setting specific ost indicies
1102 Description: QOS code breaks on skipped indicies
1103 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1104 created with --index need not be sequential.
1106 Severity : enhancement
1108 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1109 Details : The uninit_groups feature works in conjunction with the kernel
1110 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1111 up the pass1 processing of e2fsck. This is a read-only feature
1112 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1113 mount filesystems that have had this feature enabled.
1115 Severity : enhancement
1117 Description: Improve multi-block allocation algorithm to avoid fragmentation
1118 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1119 allocation locality and avoid filesystem fragmentation.
1121 ------------------------------------------------------------------------------
1123 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1125 * Support for kernels:
1126 2.4.21-47.0.1.EL (RHEL 3)
1127 2.6.5-7.283 (SLES 9)
1128 2.6.9-42.0.10.EL (RHEL 4)
1129 2.6.12.6 vanilla (kernel.org)
1130 2.6.16.27-0.9 (SLES 10)
1131 * Recommended e2fsprogs version: 1.39.cfs5
1133 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1137 Frequency : occasional, depends on client load and configuration
1138 Bugzilla : 12181, 12203
1139 Description: data loss for recently-modified files
1141 Details : In some cases it is possible that recently written or created
1142 files may not be written to disk in a timely manner (this should
1143 normally be within 30s unless client IO load is very high).
1144 The problem appears as zero-length files or files that are a
1145 multiple of 1MB in size after a client crash or client eviction
1146 that are missing data at the end of the file.
1148 This problem is more likely to be hit on clients where files are
1149 repeatedly created and unlinked in the same directory, clients
1150 have a large amount of RAM, have many CPUs, the filesystem has
1151 many OSTs, the clients are rebooted frequently, and/or the files
1152 are not accessed by other nodes after being written.
1154 The presence of the problem can be detected by looking at
1155 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1156 smaller than the second (nr_unused) then dirty files will not
1157 be flushed automatically to disk. "sync; sleep 10" should be
1158 run several times on the node before unmounting it to update
1159 Lustre (this is also safe to run on nodes without this problem).
1161 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1162 that can cause this same problem, so customers using that kernel
1163 also need to update the kernel in addition to Lustre. In order
1164 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1166 It is normal that files written just before a client crash (less
1167 than 30s) may not yet have been flushed to disk, even for local
1171 Frequency : frequent on thin XT3 nodes
1173 Description: UUID collision on thin XT3 Linux nodes
1174 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1175 randomly, since we relied on an insufficiently-seeded PRNG.
1180 Description: OSS hangs after "All ost request buffers busy"
1181 Details : A deadlock between quota and journal operations caused OSS
1182 hangs after printing "All ost request buffers busy."
1185 Frequency : always on liblustre builds
1187 Description: Cleanup compiler warnings on liblustre
1190 Frequency : always on liblustre builds on XT3
1192 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1196 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1197 > 32000 subdirectories
1198 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1199 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1200 subdirectory count crosses 32000. This will aid e2fsck to
1201 correctly handle more than 32000 subdirectories.
1204 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1206 Description: System hang on startup
1207 Details : This bug allowed the liblustre (e.g. catamount) client to
1208 return to the app before handling all startup RPCs. This
1209 could leave the node unresponsive to lustre network traffic
1210 and manifested as a server ptllnd timeout.
1212 Severity : enhancement
1214 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1215 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1216 boolean to control whether to print peer debug info when a
1217 client's RPC times out.
1222 Description: make O_SYNC working on 2.6 kernels
1223 Details : 2.6 kernels use different method for mark pages for write,
1224 so need add a code to lustre for O_SYNC work.
1229 Description: Failure to close file and release space on NFS
1230 Details : Put inode details into lock acquired in ll_intent_file_open.
1231 Use mdc_intent_lock in ll_intent_open to properly
1232 detect all kind of errors unhandled by mdc_enqueue.
1237 Description: proc file read during shutdown sometimes raced obd removal,
1239 Details : Add lock to prevent obd access after proc file removal.
1242 Frequency : Only for files larger than 4GB on 32-bit clients.
1244 Description: improperly doing page alignment of locks
1245 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1246 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1251 Description: RPCs being resent when they shouldn't be
1252 Details : Some RPCs that should not be resent are being resent. This
1253 can cause inconsistencies in the RPC state machine. Do not
1254 resend such requests.
1257 Frequency : rare, only with NFS export
1259 Description: Crash on NFS re-export node
1260 Details : under very unusual load conditions an assertion is hit in
1261 ll_intent_file_open()
1264 Frequency : only if OST filesystem is corrupted
1266 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1267 Details : for a short time RPCs with bulk IO are in the replay list,
1268 but replay of bulk IOs is unimplemented. If the OST filesystem
1269 is corrupted due to disk cache incoherency and then replay is
1270 started it is possible to trip an assertion. Avoid putting
1271 committed RPCs into the replay list at all to avoid this issue.
1276 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1277 Details : Large single O_DIRECT read and write calls can fail to allocate
1278 a sufficiently large buffer to process the request. In case of
1279 allocation failure the allocation is retried with a smaller
1280 buffer and broken into smaller requests.
1282 Severity : enhancement
1284 Description: Add -o localflock option to simulate old noflock behaviour.
1285 Details : This will achieve local-only flock/fcntl locks coherentness.
1290 Description: versioning check is incomplete
1291 Details : Checking the version difference of client vs. server, report
1292 error if the gap is too big.
1297 Description: add support PG_writeback bit
1298 Details : add support for PG_writeback bit for Lustre, for more carefull
1299 work with page cache in 2.6 kernel. This also fix some deadlocks
1300 and remove hack for work O_SYNC with 2.6 kernel.
1302 Severity : enhancement
1304 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1305 Details : The uninit_groups feature works in conjunction with the kernel
1306 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1307 up the pass1 processing of e2fsck. This is a read-only feature
1308 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1309 mount filesystems that have had this feature enabled.
1311 Severity : enhancement
1313 Description: Improve multi-block allocation algorithm to avoid fragmentation
1314 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1315 allocation locality and avoid filesystem fragmentation.
1317 ------------------------------------------------------------------------------
1319 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1321 * Support for kernels:
1322 2.6.9-42.0.3.EL (RHEL 4)
1323 2.6.5-7.276 (SLES 9)
1324 2.4.21-47.0.1.EL (RHEL 3)
1325 2.6.12.6 vanilla (kernel.org)
1326 2.6.16.21-0.8 (SLES10)
1327 * Recommended e2fsprogs version: 1.39.cfs2-0
1329 * The backwards-compatible /proc/sys/portals symlink has been removed
1330 in this release. Before upgrading, please ensure that you change
1331 any configuration scripts or /etc/sysctl.conf files that access
1332 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1333 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1334 in advance of the upgrade on any system running Lustre 1.4.6 or
1335 newer, since /proc/sys/lnet was added in that version.
1336 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1340 Frequency : only when quota is used
1342 Description: avoid scanning export list for quota master
1343 Details : Change the algorithms to avoid scanning export list in order
1344 to improve the efficiency.
1347 Frequency : MDS failover only, very rarely
1349 Description: "went back in time" messages on mds failover
1350 Details : The greatest transno may be lost when the current operation
1351 finishes with an error (transno==0) and the client's last_rcvd
1352 record is over-written. Save the greatest transno in the
1353 mds_last_transno for this case.
1356 Frequency : always for specific kernels and striping counts
1358 Description: client may get "Matching packet too big" without ACL support
1359 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1360 when trying to access files in certain configurations. The
1361 clients should in fact be denied when mounting because they do
1362 not understand ACLs.
1365 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1367 Description: many clients connecting with IO in progress causes connect timeouts
1368 Details : Avoid synchronous journal commits to avoid delays caused by many
1369 clients connecting/disconnecting when bulk IO is in progress.
1370 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1371 OST_IO_PORTAL to avoid delays behind potentially many pending
1375 Frequency : occasionally with multiple writers to a single file
1377 Description: shared writes to file may result in wrong size reported by stat()
1378 Details : Allow growing of kms when extent lock is cancelled
1381 Frequency : always with random mmap IO to multi-striped file
1383 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1384 Details : Only if the hole is at the end of OST object so that kms is too
1385 small. Fix is to increase kms accordingly in ll_nopage.
1388 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1390 Description: writes to a missing object would leak memory on the OST
1391 Details : If there is an inconsistency between the MDS and OST filesystems,
1392 such that the MDS references an object that doesn't exist, writes
1393 to that object will leak memory due to incorrect cleanup in the
1394 error handling path, eventually running out of memory on the OST.
1399 Description: Creating too long symlink causes lustre errors
1400 Details : Check symlink and name lengths before sending requests to MDS.
1403 Frequency : only if flock is enabled (not on by default)
1405 Description: posix locks not released on fd closure on 2.6.9+
1406 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1407 kernels, this caused such locks not to be released on fd close and
1408 then assertions on fs unmount about still used locks.
1411 Frequency : MDS failover only, very rarely
1413 Description: clients may get ASSERTION(granted_lock != NULL)
1414 Details : When request was taking a long time, and a client was resending
1415 a getattr by name lock request. The were multiple lock requests
1416 with the same client lock handle and
1417 mds_getattr_name->fixup_handle_for_resent_request found one of the
1418 lock handles but later failed with ASSERTION(granted_lock != NULL).
1423 Description: handle->h_buffer_credits > 0, assertion failure
1424 Details : h_buffer_credits is zero after truncate, causing assertion
1425 failure. This patch extends the transaction or creates a new
1429 Frequency : NFS re-export or patchless client
1430 Bugzilla : 11179, 10796
1431 Description: Crash on NFS re-export node (__d_move)
1432 Details : We do not want to hash the dentry if we don't have a lock.
1433 But if this dentry is later used in d_move, we'd hit uninitialised
1434 list head d_hash, so we just do this to init d_hash field but
1435 leave dentry unhashed.
1438 Frequency : NFS re-export or patchless client
1440 Description: NFS exports has problem with symbolic link
1441 Details : lustre client didn't properly install dentry when re-exported
1442 to NFS or running patchless client.
1445 Frequency : NFS re-export or patchless client
1447 Description: Various nfs/patchless fixes.
1448 Details : fixes reuse disconected alias for lookup process - this fixes
1449 warning "find_exported_dentry: npd != pd",
1450 fix permission error with open files at nfs.
1451 fix apply umask when do revalidate.
1454 Frequency : occasional
1456 Description: Crash on NFS re-export node
1457 Details : calling clear_page() on the wrong pointer triggered oops in
1458 generic_mapping_read().
1461 Frequency : rarely, using O_DIRECT IO
1463 Description: unaligned directio crashes client with LASSERT
1464 Details : check for unaligned buffers before trying any requests.
1467 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1469 Description: stale data returned from RAID cache
1470 Details : If only a small amount of IO is done to the RAID device before
1471 reading it again it is possible to get stale data from the RAID
1472 cache instead of reading it from disk.
1475 Frequency : always for sles10 kernel
1477 Description: sles10 support
1478 Details : ll_follow_link: compile fixes and using of nd_set_link
1479 under newer kernels.
1482 Frequency : depends on arch, kernel and compiler version, always on sles10
1485 Description: recursive or deep enough symlinks cause stack overflow
1486 Details : getting rid of large stack-allocated variable in
1490 Frequency : depends on hardware
1492 Description: lustre write performance loss in the SLES10 kernel
1493 Details : the performance loss is caused by using of write barriers in the
1494 ext3 code. The SLES10 kernel turns barrier support on by
1495 default. The fix is to undo that change for ldiskfs.
1497 ------------------------------------------------------------------------------
1499 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1501 * Support for kernels:
1502 2.6.9-42.0.3EL (RHEL 4)
1503 2.6.5-7.276 (SLES 9)
1504 2.4.21-47.0.1.EL (RHEL 3)
1505 2.6.12.6 vanilla (kernel.org)
1509 Frequency : quota enabled and large files being deleted
1511 Description: releasing more than 4GB of quota at once hangs OST
1512 Details : If a user deletes more than 4GB of files on a single OST it
1513 will cause the OST to spin in an infinite loop. Release
1514 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1519 Description: statfs data retrieved from /proc may be stale or zero
1520 Details : When reading per-device statfs data from /proc, in the
1521 {kbytes,files}_{total,free,avail} files, it may appear
1522 as zero or be out of date.
1525 Frequency : systems with MD RAID1 external journal devices
1527 Description: lconf's call to blkid is confused by RAID1 journal devices
1528 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1529 of returning all block devices that match the journal UUID.
1532 Frequency : always, for aggregate stripe size over 4GB
1534 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1535 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1536 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1539 Frequency : always if "lfs find" used on a local file/directory
1541 Description: "lfs find" segfaults if used on a local file/directory
1542 Details : The case where a directory component was not specified wasn't
1543 handled correctly. Handle this properly.
1546 Frequency : always on ppc64
1548 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1549 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1550 perform endianness conversion.
1553 Frequency : rarely (truncate to non-zero file size after write under load)
1554 Bugzilla : 10730, 10687
1555 Description: Files padded with zeros to next 4K multiple
1556 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1557 it is possible that files that are truncated to a non-zero size
1558 immediately after being written are filled with zero bytes beyond
1559 the truncated size. No file data is lost.
1561 Severity : enhancement
1563 Description: Allow recovery/failover for liblustre clients.
1564 Details : liblustre clients were unaware of failover configurations until
1567 Severity : enhancement
1569 Description: user file locks should fail when not mounting with flock option
1570 Details : Set up an error-returning stub in ll_file_operations.lock field
1571 to prevent incorrect behaviour when client is mounted without
1572 flock option. Also, set up properly f_op->flock field for
1576 Frequency : always on ia64
1578 Description: "lfs df" loops on printing out MDS statfs information
1579 Details : The obd_ioctl_data was not initialized and in some systems
1580 this caused a failure during the ioctl that did not return
1581 an error. Initialize the struct and return an error on failure.
1584 Frequency : SLES 9 only
1586 Description: Error of copying files with lustre special EAs as root
1587 Details : Client side always return success for setxattr call for lustre
1588 special xattr (currently only "trusted.lov").
1591 Frequency : rarely on clusters with both ia64+i386 clients
1593 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1594 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1595 clients are doing concurrent writes to the same file it is
1596 possible that the ia64 clients may LASSERT because the OST
1597 extent locks are not PAGE_SIZE aligned. Ensure that grown
1598 locks are always aligned on the request boundary.
1601 Frequency : specific use, occasional
1603 Description: Overwriting in use executable truncates on-disk binary image
1604 Details : If one node attempts to overwrite an executable in use by
1605 another node, we now correctly return ETXTBSY instead of
1606 truncating the file.
1608 Severity : enhancement
1610 Description: Async OSC create to avoid the blocking unnecessarily.
1611 Details : If a OST has no remain object, system will block on the creating
1612 when need to create a new object on this OST. Now, ways use
1613 pre-created objects when available, instead of blocking on an
1614 empty osc while others are not empty. If we must block, we block
1615 for the shortest possible period of time.
1620 Description: chmod on Lustre root is propagated to other clients
1621 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1622 invalid by the follow_mount time.
1625 Frequency : liblustre clients only
1627 Description: Race in 'instant cancel' lock handling could lead to such locks
1628 never to be granted in case of SMP MDS
1629 Details : Do not destroy not yet granted but cbpending locks in
1633 Frequency : replay/resend of open
1635 Description: non null lock assetion failure in mds_intent_policy
1636 Details : Trying to replay/resend lockless open requests resulted in
1637 mds_open() returning 0 with no lock. Now it sets a flag if
1638 a lock is going to be returned.
1640 Severity : enhancement
1642 Description: Checksum enhancements
1643 Details : New checksum enhancements allow for resending RPCs that failed
1646 Severity : enhancement
1648 Description: Tunables on number of dirty pages in cacche
1649 Details : Allow to set limit on number of dirty pages cached.
1654 Description: client crash on unmount - lock still has references
1655 Details : In some error handling cases it was possible to leak a lock
1656 reference on a client while accessing a file. This was not
1657 harmful to the client during operation, but would cause the
1658 client to crash when the filesystem is unmounted.
1661 Frequency : specific case, rare
1663 Description: ETXTBSY on mds though file not in use by client
1664 Details : ETXTBSY is no longer incorrectly returned when attempting to
1665 chmod or chown a directory that the user previously tried to
1666 execute or a currently-executing binary.
1669 Frequency : extremely rare except on liblustre-based clients
1671 Description: Lustre space not freed when files are deleted
1672 Details : Clean up open-unlinked files after client eviction. Previously
1673 the unlink was skipped and the files remained as orphans.
1678 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1679 Details : In some cases it was possible to send a blocking callback to
1680 a client doing a glimpse, even though that client didn't get
1681 a lock granted. When the glimpse lock is cancelled on the OST
1682 the freed lock is left on the waiting list and corrupted the list.
1685 Frequency : all core dumps
1687 Description: Broke core dumps to lustre
1688 Details : Negative dentry may be unhashed if parent does not have UPDATE
1689 lock, but some callers, e.g. do_coredump, expect dentry to be
1690 hashed after successful create, hash it in ll_create_it.
1692 ------------------------------------------------------------------------------
1694 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1696 * Support for kernels:
1697 2.6.9-42.0.2.EL (RHEL 4)
1698 2.6.5-7.276 (SLES 9)
1699 2.4.21-40.EL (RHEL 3)
1700 2.6.12.6 vanilla (kernel.org)
1704 Frequency : always on RHEL 3
1706 Description: Number of open files grows over time
1707 Details : The number of open files grows over time, whether or not
1708 Lustre is started. This was due to a filp leak introduced
1709 by one of our kernel patches.
1711 ------------------------------------------------------------------------------
1713 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1715 * Support for kernels:
1716 2.6.9-42.EL (RHEL 4)
1717 2.6.5-7.276 (SLES 9)
1718 2.4.21-40.EL (RHEL 3)
1719 2.6.12.6 vanilla (kernel.org)
1724 Bugzilla : 5719, 9635, 9792, 9684
1725 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1726 Details : If a server is under heavy load and cannot reply to new
1727 connection requests before the client resends the (re)connect,
1728 the connection handling code can behave badly if two service
1729 threads are concurrently handing separate (re)connections from
1730 the same client. Add better locking to the connection handling
1731 code, and ensure that only a single connection will be processed
1732 for a given client UUID, even if the lock is dropped.
1734 Severity : enhancement
1736 Description: add TCP zero-copy support to kernel
1737 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1738 sends if the hardware supports scatter-gather and checksumming.
1739 This allows socklnd to do client-write and server-read more
1740 efficiently and reduce CPU utilization from skbuf copying.
1743 Frequency : only if NFS exporting from client
1745 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1746 Details : If setting attributes on a file created under NFS that had
1747 never been opened it would be possible to oops the client
1748 if the file had no objects.
1751 Frequency : always for liblustre
1753 Description: liblustre client does MDS+OSTs setattr RPC for each write
1754 Details : When doing a write from a liblustre client, the client
1755 incorrectly issued an RPC to the MDS and each OST the file was
1756 striped over in order to update the timestamps. When writing
1757 with small chunks and many clients this could overwhelm the MDS
1758 with RPCs. In all cases it would slow down the write because
1759 these RPCs are unnecessary.
1761 Severity : enhancement
1763 Description: allow number of MDS service threads to be changed at module load
1764 Details : It is now possible to change the number of MDS service threads
1765 running. Adding "options mds mds_num_threads={N}" to the MDS's
1766 /etc/modprobe.conf will set the number of threads for the next
1767 time Lustre is restarted (assuming the "mds" module is also
1768 reloaded at that time). The default number of threads will
1769 stay the same, 32 for most systems.
1774 Description: OST crash if filesystem is unformatted or corrupt
1775 Details : If an OST is started on a device that has never been formatted
1776 or if the filesystem is corrupt and cannot even mount then the
1777 error handling cleanup routines would dereference a NULL pointer.
1782 Description: NULL pointer deref in llap_from_page.
1783 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1784 incorrect) mapping if the page was truncated/reclaimed while it was
1785 searched for. Check for this condition and skip such pages when
1786 doing readahead. Introduce extra check to llap_from_page() to
1787 verify page->mapping->host is non-NULL (so page is not anonymous).
1790 Frequency : Sometimes when using sys_sendfile
1792 Description: "page not covered by a lock" warnings from ll_readpage
1793 Details : sendfile called ll_readpage without right page locks present.
1794 Now we introduced ll_file_sendfile that does necessary locking
1795 around call to generic_file_sendfile() much like we do in
1799 Frequency : with certain MDS communication failures at client mount time
1801 Description: NULL pointer deref after failed client mount
1802 Details : a client connection request may delayed by the network layer
1803 and not be sent until after the PTLRPC layer has timed out the
1804 request. If the client fails the mount immediately it will try
1805 to clean up before the network times out the request. Add a
1806 reference from the request import to the obd device and delay
1807 the cleanup until the network drops the request.
1810 Frequency : occasionally during client (re)connect
1812 Description: assertion failure during client (re)connect
1813 Details : processing a client connection request may be delayed by the
1814 client or server longer than the client connect timeout. This
1815 causes the client to resend the connection request. If the
1816 original connection request is replied in this interval, the
1817 client may trip an assertion failure in ptlrpc_connect_interpret()
1818 which thought it would be the only running connect process.
1821 Frequency : only with obd_echo servers and clients that are rebooted
1823 Description: kernel BUG accessing uninitialized data structure
1824 Details : When running an obd_echo server it did not start the ping_evictor
1825 thread, and when a client was evicted an uninitialized data
1826 structure was accessed. Start the ping_evictor in the RPC
1827 service startup instead of the OBD startup.
1829 Severity : enhancement
1830 Bugzilla : 10193 (patchless)
1831 Description: Remove dependency on various unexported kernel interfaces.
1832 Details : No longer need reparent_to_init, exit_mm, exit_files,
1833 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1836 Frequency : rare (only users of deprecated and unsupported LDAP config)
1838 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1839 Details : LDAP apparently contains 'lustreName' attributes instead of
1840 'name'. A simple remapping of the name is sufficient.
1843 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1845 Description: waiting_locks_callback trips kernel BUG if client is evicted
1846 Details : Running with the dump_on_timeout debug flag turned on makes
1847 it possible that the waiting_locks_callback() can try to dump
1848 the Lustre kernel debug logs from an interrupt handler. Defer
1849 this log dumping to the expired_lock_main() thread.
1851 Severity : enhancement
1853 Description: Support NFS exporting on 2.6 kernels.
1854 Details : Implement non-rawops metadata methods for NFS server to use without
1855 changing NFS server code.
1858 Frequency : very rare (synthetic metadata workload only)
1860 Description: two racing renames might cause an MDS thread to deadlock
1861 Details : Running the "racer" program may cause one MDS thread to rename
1862 a file from being the source of a rename to being the target of
1863 a rename at exactly the same time that another thread is doing
1864 so, and the second thread has already enqueued these locks after
1865 doing a lookup of the target and is trying to relock them in
1866 order. Ensure that we don't try to re-lock the same resource.
1869 Frequency : only very large systems with liblustre clients
1871 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1872 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1873 all OSTs in parallel. This allows the eviction of stale liblustre
1874 clients to proceed much faster than if they were done in series,
1875 and also offers similar improvements for other set_info RPCs.
1880 Description: excessive CPU usage during initial read phase on client
1881 Details : During the initial read phase on a client, it would agressively
1882 retry readahead on the file, consuming too much CPU and impacting
1883 performance (since 1.4.5.8). Improve the readahead algorithm
1884 to avoid this, and also improve some other common cases (read
1885 of small files in particular, where "small" is files smaller than
1886 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1891 Description: MDS crash when receiving packet with unknown intent.
1892 Details : Do not LBUG in unknown intent case, just return -EFAULT
1894 Severity : enhancement
1895 Bugzilla : 9293, 9385
1896 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1897 Details : Do not serialize getattr (non-intent version) and statfs.
1900 Frequency : occasional, when OST network is overloaded/intermittent
1902 Description: client evicted by OST after bulk IO timeout
1903 Details : If a client sends a bulk IO request (read or write) the OST
1904 may evict the client if it is unresposive to its data GET/PUT
1905 request. This is incorrect if the network is overloaded (takes
1906 too long to transfer the RPC data) or dropped the OST GET/PUT
1907 request. There is no need to evict the client at all, since
1908 the pinger and/or lock callbacks will handle this, and the
1909 client can restart the bulk request.
1912 Frequency : Always when mmapping file with no objects
1914 Description: client crashes when mmapping file with no objects
1915 Details : Check that we actually have objects in a file before doing any
1916 operations on objects in ll_vm_open, ll_vm_close and
1922 Description: Request leak when working with deleted CWD
1923 Details : Introduce advanced request refcount tracking for requests
1924 referenced from lustre intent.
1926 Severity : Enhancement
1928 Description: Cache open file handles on client.
1929 Details : MDS now will return special lock along with openhandle, if
1930 requested and client is allowed to hold openhandle, even if unused,
1931 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1932 closing files for every read/write openration.
1934 Severity : Enhancement
1936 Description: Cache open negative dentries on client when possible.
1937 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1938 negative dentries on lock revocation.
1943 Description: Remounting a client read-only wasn't possible with a zconf mount
1944 Details : It wasn't possible to remount a client read-only with llmount.
1946 Severity : enhancement
1947 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1948 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1949 MPICH is included to add Lustre-specific ADIO interfaces.
1950 This is based closely on the UFS ADIO layer and only differs
1951 in file creation, in order to allow the OST striping to be set.
1952 This is user-contributed code and not supported by CFS.
1957 Description: extended inode attributes (immutable, append-only) work improperly
1958 when 2.4 and 2.6 kernels are used on client/server or vice versa
1959 Details : Introduce kernel-independent values for these flags.
1961 Severity : enhancement
1964 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1965 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1966 granularity. Now, Fractional MB granularity is be supported,
1967 this is very useful for low memory system.
1969 Severity : enhancement
1971 Description: Getattr by fid
1972 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1973 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1974 have locks and avoids getattr rpc requests for it on every stat.
1977 Frequency : Always, for filesystems larger than 2TB
1979 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1980 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1981 filesystems larger than 2TB to be mounted with mballoc enabled.
1984 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1986 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1987 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1988 filesystems larger than 2TB is mounted then the kernel will
1989 silently corrupt the start of the filesystem. CONFIG_LBD is
1990 enabled for all CFS-supported kernels, but the possibility of
1991 this happening with a modified kernel config exists.
1993 Severity : enhancement
1995 Description: add client O_DIRECT support for 2.6 kernels
1996 Details : It is now possible to do O_DIRECT reads and writes to files
1997 in the Lustre client mountpoint on 2.6 kernel clients.
1999 Severity : enhancement
2001 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2002 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2003 parallel, not waiting for response from every OST before sending
2004 a rpc to the next OST.
2009 Description: setattr vs write race when updating file timestamps
2010 Details : Client processes that update a file timestamp into the past
2011 right after writing to the file (e.g. tar) it is possible that
2012 the updated file modification time can be reset to the current
2013 time due to a race between processing the setattr and write RPC.
2015 Severity : enhancement
2017 Description: Bring 'lfs find' closer in line with regular Linux find.
2018 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2019 -print0 options and obtains all the needed info through the lustre
2022 Severity : enhancement
2024 Description: support up to 1024 configured devices on one node
2025 Details : change obd_dev array from statically allocated to dynamically
2026 allocated structs as they are first used to reduce memory usage
2031 Description: Flush dirty partially truncated pages during truncate
2032 Details : Immediatelly flush partially truncated pages in filter_setattr,
2033 this way we completely avoid having any pages in page cache on OST
2034 and can retire ugly workarounds during writes to flush such pages.
2039 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2040 Details : protect inode from truncation within vfs_unlink() context
2041 just take a reference before calling vfs_unlink() and release it
2042 when parent's i_sem is free.
2047 Description: last_id value checked outside lock on OST caused LASSERT failure
2048 Details : If there were multiple MDS->OST object precreate requests in
2049 flight, it was possible that the OST's last object id was checked
2050 outside a lock and incorrectly tripped an assertion. Move checks
2051 inside locks, and discard old precreate requests.
2054 Frequency : always, if extents are used on OSTs
2056 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2057 Details : OSTs using the extents format would not zero the high 16 bits of
2058 the index physical block number. This is not a problem for any
2059 OST filesystems smaller than 16TB, and no kernels support ext3
2060 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2061 new/modified files) and can be fixed for existing filesystems
2062 with e2fsprogs-1.39-cfs1.
2067 Description: import connection selection may be incorrect if timer wraps
2068 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2069 import connections to be ignored if the 32-bit jiffies counter
2070 wraps. Use a 64-bit jiffies counter.
2073 Frequency : very large clusters immediately after boot
2075 Description: LNET request buffers exhausted under heavy short-term load
2076 Details : If a large number of client requests are generated on a service
2077 that has previously never seen so many requests it is possible
2078 that the request buffer growth cannot keep up with the spike in
2079 demand. Instead of dropping incoming requests, they are held in
2080 the LND until the RPC service can accept more requests.
2083 Frequency : Sometimes during replay
2085 Description: Assertion failure in ll_local_open after replay.
2086 Details : If replay happened on an open request reply before we were able
2087 to set replay handler, reply will become not swabbed tripping the
2088 assertion in ll_local_open. Now we set the handler right after
2089 recognising of open request
2092 Frequency : very rare
2094 Description: kernel reports "badness in vsnprintf"
2095 Details : Reading from the "recovery_status" /proc file in small chunks
2096 may cause a negative length in lprocfs_obd_rd_recovery_status()
2097 call to vsnprintf() (which is otherwise harmless). Exit early
2098 if there is no more space in the output buffer.
2100 Severity : enhancement
2102 Description: clear OBD RPC statistics by writing to them
2103 Details : It is now possible to clear the OBD RPC statistics by writing
2104 to the "stats" file.
2109 Description: Client mtime is not the same on different clients after utimes
2110 Details : In some cases, the client was using the utimes() syscall on
2111 a file cached on another node. The clients now validate the
2112 ctime from the MDS + OSTs to determine which one is right.
2117 Description: Inability to activate failout mode
2118 Details : lconf script incorrectly assumed that in pythong string's numeric
2119 value is used in comparisons.
2122 Frequency : always with multiple stripes per file
2124 Description: Inefficient object allocation for mutli-stripe files
2125 Details : When selecting which OSTs to stripe files over, for files with
2126 a stripe count that divides evenly into the number of OSTs,
2127 the MDS is always picking the same starting OST for each file.
2128 Return the OST selection heuristic to the original design.
2133 Description: mount failures may take full timeout to return an error
2134 Details : Under some heavy load conditions it is possible that a
2135 failed mount can wait for the full obd_timeout interval,
2136 possibly several minutes, before reporting an error.
2137 Instead return an error as soon as the status is known.
2139 Frequency : quota enabled and large files being deleted
2141 Description: releasing more than 4GB of quota at once hangs OST
2142 Details : If a user deletes more than 4GB of files on a single OST it
2143 will cause the OST to spin in an infinite loop. Release
2144 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2149 Description: statfs data retrieved from /proc may be stale or zero
2150 Details : When reading per-device statfs data from /proc, in the
2151 {kbytes,files}_{total,free,avail} files, it may appear
2152 as zero or be out of date.
2155 Frequency : systems with MD RAID1 external journal devices
2157 Description: lconf's call to blkid is confused by RAID1 journal devices
2158 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2159 of returning all block devices that match the journal UUID.
2162 Frequency : always, for aggregate stripe size over 4GB
2164 Description: assertion fails when trying to use 4GB stripe size
2165 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2166 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2169 Frequency : always on ppc64
2171 Description: the first write on an ext3 filesystem with mballoc got stuck
2172 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2173 perform endianness conversion.
2175 ------------------------------------------------------------------------------
2177 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2179 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2180 INTEROPERATE with older versions automatically. Please read the
2181 user documentation before upgrading any part of a live system.
2182 * WARNING: Lustre networking configuration changes are required with
2183 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2186 * Support for kernels:
2187 2.6.9-22.0.2.EL (RHEL 4)
2188 2.6.5-7.244 (SLES 9)
2189 2.6.12.6 vanilla (kernel.org)
2192 Severity : enhancement
2193 Bugzilla : 7981/8208
2194 Description: Introduced Lustre Networking (LNET)
2195 Details : LNET is new networking infrastructure for Lustre, it includes
2196 a reorganized network configuration mode (see the user
2197 documentation for full details) as well as support for routing
2198 between different network fabrics. Lustre Networking Devices
2199 (LNDs) for the supported network fabrics have also been
2200 created for this new infrastructure.
2202 Severity : enhancement
2203 Description: Introduced Access control lists
2204 Details : clients can set ACLs on files and directories in order to have
2205 more fine-grained permissions than the standard Unix UGO+RWX.
2206 The MDS must be started with the "-o acl" mount option.
2208 Severity : enhancement
2209 Description: Introduced filesystem quotas
2210 Details : Administrators may now establish per-user quotas on the
2213 Severity : enhancement
2215 Description: Configuration change for the XT3
2216 The PTLLND is now used to run Lustre over Portals on the XT3
2217 The configure option(s) --with-cray-portals are no longer used.
2218 Rather --with-portals=<path-to-portals-includes> is used to
2219 enable building on the XT3. In addition to enable XT3 specific
2220 features the option --enable-cray-xt3 must be used.
2225 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2226 Details : A race between allocating a new llcd and re-getting the llcd_lock
2227 allowed another thread to grab newly-allocated llcd.
2229 Severity : enhancement
2231 Description: 2.6 OST async journal commit and locking fix to improve performance
2232 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2233 2.6 kernels are now async as they already were in 2.4 kernels so
2234 that they can commit concurrently with the network bulk transfer.
2235 For block-allocated files the filter allocation semaphore is held
2236 to avoid filesystem fragmentation during allocation. BKL lock
2237 removed for 2.6 xattr operations where it is no longer needed.
2242 Description: lconf incorrectly determined whether two IP networks could talk
2243 Details : In some more complicated routing and multiple-network
2244 configurations, lconf will avoid trying to make a network
2245 connection to a disjoint part of the IP space. It was doing the
2246 math incorrectly for one set of cases.
2251 Description: Fix for potential infinite loop processing records in an llog.
2252 Details : If an llog record is corrupted/zeroed, it is possible to loop
2253 forever in llog_process(). Validate the llog record length
2254 and skip the remainder of the block on error.
2257 Frequency : occasional (liblustre only)
2259 Description: liblustre could not open files whose last component is a symlink
2260 Details : sysio_path_walk() would incorrectly pass the open intent to
2261 intermediate path components.
2264 Frequency : rare (liblustre only with non-standard tuning)
2265 Bugzilla : 7201 (7350)
2266 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2267 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2268 after it has been released, only do this for referenced locks
2270 Severity : enhancement
2272 Description: specifying an (invalid) directory default stripe_size of -1
2273 would reset the directory default striping
2274 Details : stripe_size -1 was used internally to signal directory stripe
2275 removal, now use "all default" to signal dir stripe removal
2276 as a directory striping of "all default" is not useful
2279 Frequency : common for large clusters running liblustre clients
2281 Description: doing an ls when liblustre clients are running is slow
2282 Details : sending a glimpse AST to a liblustre client waits for every AST
2283 to time out, as liblustre clients will not respond. Since they
2284 cannot cache data we refresh the OST lock LVB from disk instead.
2286 Severity : enhancement
2288 Description: doing an ls at the same time as file IO can be slow
2289 Details : enqueue and other "small" requests can be blocked behind many
2290 large IO requests. Create a new OST IO portal for non-IO
2291 requests so they can be processed faster.
2294 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2296 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2297 created file with mode 000
2298 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2299 allow an ftruncate() as a non-root user to a file with mode 000.
2300 Lustre now respects this flag to disable mode checks when
2301 truncating a file owned by the user
2304 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2307 Description: Revoking locks from clients that went dead or catatonic might take
2309 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2310 cancellation of such locks instant on servers without waiting for
2311 any reply from clients. Clients drops these locks when cancel
2312 notification from server is received without replying.
2315 Frequency : liblustre-only, when liblustre client dies or becomes busy
2317 Description: Doing ls on Linux clients can take a long time with active
2319 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2320 granting such locks to it in the first place if possible. Locks
2321 are taken by proxy on the OST during the read or write and
2322 dropped immediately afterward. Add connect flags handling, do
2323 not grant locks to liblustre clients for glimpse ASTs.
2325 Severity : enhancement
2327 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2328 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2329 sequential reads and behaves badly with applications doing random
2330 reads. Improve it by reading ahead at least read region, and
2331 avoiding excessive large RPC for small reads.
2333 Severity : enhancement
2335 Description: Creating more than 1000 files for a single job may cause a load
2336 imbalance on the OSTs if there are also a large number of OSTs.
2337 Details : qos_prep_create() uses an OST index reseed value that is an
2338 even multiple of the number of available OSTs so that if the
2339 reseed happens in the middle of the object allocation it will
2340 still utilize the OSTs as uniformly as possible.
2345 Description: OST or MDS may oops in ping_evictor_main()
2346 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2347 but doesn't restart at beginning of obd_exports_timed list
2350 Severity : enhancement
2352 Description: improve by-nid export eviction on the MDS and OST
2353 Details : allow multiple exports with the same NID to be evicted at one
2354 time without re-searching the exports list.
2357 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2359 Description: MDS may oops in groups_free()
2360 Details : in rare race conditions a newly allocated group_info struct is
2361 freed again, and this can be NULL. The 2.4 compatibility code
2362 for groups_free() checked for a NULL pointer, but 2.6 did not.
2365 Frequency : common for liblustre clients doing little filesystem IO
2366 Bugzilla : 9352, 7313
2367 Description: server may evict liblustre clients accessing contended locks
2368 Details : if a client is granted a lock or receives a completion AST
2369 with a blocking AST already set it would not reply to the AST
2370 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2373 Frequency : lfs setstripe, only systems with more than 160 OSTs
2375 Description: unable to set striping with a starting offset beyond OST 160
2376 Details : llapi_create_file() incorrectly limited the starting stripe
2377 index to the maximum single-file stripe count.
2380 Frequency : LDAP users only
2382 Description: lconf did not handle in-kernel recovery with LDAP properly
2383 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2385 Severity : enhancement
2387 Description: bind OST threads to NUMA nodes to improve performance
2388 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2389 node and do their allocations there to localize memory access
2391 Severity : enhancement
2393 Description: llmount can determine client NID directly from Myrinet (GM)
2394 Details : the client NID code from gmnalnid was moved directly into
2395 llmount, removing the need to use this or specifying the
2396 client NID explicitly when mounting GM clients with zeroconf
2399 Frequency : if client is started with down MDS
2401 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2402 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2403 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2404 ptlrpc_import_delayed_req() because we didn't check if the
2405 request was interrupted, and we also didn't break out of the
2406 event loop if there was no timeout
2411 Description: data loss during non-page-aligned writes to a single file from
2412 both multiple nodes and multiple threads on one node at same time
2413 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2414 inconsistency led to false short-reads, that were cached and later
2415 used by ->prepare_write() to fill in partially written page,
2416 leading to data loss.
2419 Frequency : always, if lconf --abort_recovery used
2421 Description: lconf --abort_recovery fails with 'Operation not supported'
2422 Details : lconf was attempting to abort recovery on the MDT device and not
2425 Severity : enhancement
2427 Description: remove cleanup logs
2428 Details : replace lconf-generated cleanup logs with lustre internal
2429 cleanup routines. Eliminates the need for client-cleanup and
2432 Severity : enhancement
2434 Description: add support for EAs (user and system) on lustre filesystems
2435 Details : it is now possible to store extended attributes in the Lustre
2436 client filesystem, and with the user_xattr mount option it
2437 is possible to allow users to store EAs on their files also
2439 Severity : enhancement
2441 Description: Add possibility (config option) to show minimal available OST free
2443 Details : When compiled with --enable-mindf configure option, statfs(2)
2444 (and so, df) will return least minimal free space available from
2445 all OSTs as amount of free space on FS, instead of summary of
2446 free spaces of all OSTs.
2448 Severity : enhancement
2450 Description: do not expand extent locks acquired on OST-side
2451 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2452 by server: they are not cached anyway.
2455 Frequency : when mmap is used/binaries executed from Lustre
2457 Description: Unmmap pages before throwing them away from read cache.
2458 Details : llap_shrink cache now attempts to unmap pages before discarding
2459 them (if unmapping failed - do not discard). SLES9 kernel has
2460 extra checks that trigger if this unmapping is not done first.
2465 Description: lconf didn't resolve symlinks before checking to see whether a
2466 given mountpoint was already in use
2469 Frequency : when migrating failover services
2470 Bugzilla : 6395, 9514
2471 Description: When migrating a subset of services from a node (e.g. failback
2472 from a failover service node) the remaining services would
2473 time out and evict clients.
2474 Details : lconf --force (implied by --failover) sets the global obd_timeout
2475 to 5 seconds in order to quickly disconnect, but this caused
2476 other RPCs to time out too quickly. Do not change the global
2477 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2479 Severity : enhancement
2480 Frequency : if MDS is started with down OST
2481 Bugzilla : 9439,5706
2482 Description: Allow startup/shutdown of an MDS without depending on the
2483 availability of the OSTs.
2484 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2485 Add appropriate locking and lov-osc refcounts for safe
2486 cleaning. Add osc abort_inflight calls in case the
2487 synchronize never started.
2490 Frequency : occasional (Cray XT3 only)
2492 Description: root not authorized to access files in CRAY_PORTALS environment
2493 Details : The client process capabilities were not honoured on the MDS in
2494 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2495 been accessed by an authorized user then root was able to access
2496 the file on the local client also. The root user capabilities
2497 are now allowed on the MDS, as this environment has secure UID.
2500 Frequency : occasional
2502 Description: ldiskfs "too long searching" message happens too often
2503 Details : A debugging message (otherwise harmless) prints too often on
2504 the OST console. This has been reduced to only happen when
2505 there are fragmentation problems on the filesystem.
2510 Description: Division by zero in statfs when all OSCs are inactive
2511 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2512 lov_statfs divides by value returned by lov_get_stripecnt().
2516 Bugzilla : 9489, 3273
2517 Description: First write from each client to each OST was only 4kB in size,
2518 to initialize client writeback cache, which caused sub-optimal
2519 RPCs and poor layout on disk for the first writen file.
2520 Details : Clients now request an initial cache grant at (re)connect time
2521 and so that they can start streaming writes to the cache right
2522 away and always do full-sized RPCs if there is enough data.
2523 If the OST is rebooted the client also re-establishes its grant
2524 so that client cached writes will be honoured under the grant.
2529 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2530 Details : Now I/O RPCs go to different portal number and (presumably) fast
2531 lock requests (and glimses) and other RPCs get their own service
2532 threads pool that should be able to service those RPCs
2535 Severity : enhancement
2537 Description: Ability to exchange lustre version between client and servers and
2538 issue warnings at client side if client is too old. Also for
2539 liblustre clients there is ability to refuse connection of too old
2541 Details : New 'version' field is added to connect data structure that is
2542 filled with version info. That info is later checked by server and
2546 Frequency : rare, liblustre only.
2547 Bugzilla : 9296, 9581
2548 Description: Two simultaneous writes from liblustre at offset within same page
2549 might proceed at the same time overwriting eachother with stale
2551 Details : I/O lock withing llu_file_prwv was released too early, before data
2552 actually was hitting the wire. Extended lock-holding time until
2553 server acknowledges receiving data.
2556 Frequency : extremely rare. Never observed in practice.
2558 Description: avoid generating lustre_handle cookie of 0.
2559 Details : class_handle_hash() generates handle cookies by incrementing
2560 global counter, and can hit 0 occasionaly (this is unlikely, but
2561 not impossible, because initial value of cookie counter is
2562 selected randonly). Value of 0 is used as a sentinel meaning
2563 "unassigned handle" --- avoid it. Also coalesce two critical
2564 sections in this function into one.
2566 Severity : enhancement
2568 Description: allow liblustre clients to delegate truncate locking to OST
2569 Details : To avoid overhead of locking, liblustre client instructs OST to
2570 take extent lock in ost_punch() on client's behalf. New connection
2571 flag is added to handle backward compatibility.
2573 Severity : enhancement
2574 Bugzilla : 4928, 7341, 9758
2575 Description: allow number of OST service threads to be specified
2576 Details : a module parameter allows the number of OST service threads
2577 to be specified via "options ost ost_num_threads={N}" in the
2578 OSS's /etc/modules.conf or /etc/modprobe.conf.
2582 Bugzilla : 6146, 9635, 9895
2583 Description: servers crash with bad pointer in target_handle_connect()
2584 Details : In rare cases when a client is reconnecting it was possible that
2585 the connection request was the last reference for that export.
2586 We would temporarily drop the export reference and get a new
2587 one, but this may have been the last reference and the export
2588 was just destroyed. Get new reference before dropping old one.
2590 Severity : enhancement
2591 Frequency : if client is started with failover MDS
2593 Description: Allow multiple MDS hostnames in the mount command
2594 Details : Try to read the configuration from all specified MDS
2595 hostnames during a client mount in case the "primary"
2598 Severity : enhancement
2600 Description: Stop sending data to evicted clients as soon as possible.
2601 Details : Check if the client we are about to send or are sending data to
2602 was evicted already. (Check is done every second of waiting,
2603 for which l_wait_event interface was extended to allow checking
2604 of exit condition at specified intervals).
2607 Frequency : rare, normally only when NFS exporting is done from client
2609 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2611 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2614 Frequency : always, if srand() is called before liblustre initialization
2616 Description: Liblustre uses system PRNG disturbing its usage by user application
2617 Details : Introduce internal to lustre fast and high-quality PRNG for
2618 lustre usage and make liblustre and some other places in generic
2619 lustre code to use it.
2621 Severity : enhancement
2622 Bugzilla : 9477, 9557, 9870
2623 Description: Verify that the MDS configuration logs are updated when xml is
2624 Details : Check if the .xml configuration logs are newer than the config
2625 logs stored on the MDS and report an error if this is the case.
2626 Request --write-conf, or allow starting with --old_conf.
2628 Severity : enhancement
2630 Description: Handle symlinks in the path when checking if Lustre is mounted.
2631 Details : Resolve intermediate symlinks when checking if a client has
2632 mounted a filesystem to avoid duplicate client mounts.
2637 Description: lconf can hit an error exception but still return success.
2638 Details : The lconf command catches the Command error exception at the top
2639 level script context and will exit with the associated exit
2640 status, but doesn't ensure that this exit status is non-zero.
2645 Description: failure of ptlrpc thread startup can cause oops
2646 Details : Starting a ptlrpc service thread can fail if there are a large
2647 number of threads or the server memory is very fragmented.
2648 Handle this without oopsing.
2651 Frequency : always, only if liblustre and non-default acceptor port was used
2653 Description: liblustre cannot connect to servers with non-default acceptor port
2654 Details : tcpnal_set_default_params() was not called and was therefore
2655 ignoring the environment varaible TCPNAL_PORT, as well as other
2656 TCPNAL_ environment variables
2661 Description: two objects could be created on the same OST for a single file
2662 Details : If an OST is down, in some cases it was possible to create two
2663 objects on a single OST for a single file. No problems other
2664 than potential performance impact and spurious error messages.
2668 Bugzilla : 5681, 9562
2669 Description: Client may oops in ll_unhash_aliases
2670 Details : Client dcache may become inconsistent in race condition.
2671 In some cases "getcwd" can fail if the current directory is
2677 Description: Inode refcounting problems in NFS export code
2678 Details : link_raw functions used to call d_instantiate without obtaining
2679 extra inode reference first.
2683 Bugzilla : 9942, 9903
2684 Description: Referencing freed requests leading to crash, memleaks with NFS.
2685 Details : We used to require that call to ll_revalidate_it was always
2686 followed by ll_lookup_it. Also with revalidate_special() it is
2687 possible to call ll_revalidate_it() twice for the same dentry
2688 even if first occurence returned success. This fix changes semantic
2689 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2690 reference on a request referred from the intent.
2691 ll_intent_release() then releases such a request.
2694 Frequency : rare, normally benchmark loads only
2696 Description: unlinked inodes were kept in memory on the client
2697 Details : If a client is repeatedly creating and unlinking files it
2698 can accumulate a lot of stale inodes in the inode slab cache.
2699 If there is no other client load running this can cause the
2700 client node to run out of memory. Instead flush old inodes
2701 from client cache that have the same inode number as a new inode.
2704 Frequency : SLES9 2.6.5 kernel and long filenames only
2705 Bugzilla : 9969, 10379
2706 Description: utime reports stale NFS file handle
2707 Details : SLES9 uses out-of-dentry names in some cases, which confused
2708 the lustre dentry revalidation. Change it to always use the
2712 Frequency : rare, unless heavy write-truncate concurrency is continuous
2713 Bugzilla : 4180, 6984, 7171, 9963, 9331
2714 Description: OST becomes very slow and/or deadlocked during object unlink
2715 Details : filter_destroy() was holding onto the parent directory lock
2716 while truncating+unlinking objects. For very large objects this
2717 may block other threads for a long time and slow overall OST
2718 responsiveness. It may also be possible to get a lock ordering
2719 deadlock in this case, or run out of journal credits because of
2720 the combined truncate+unlink. Solution is to do object truncate
2721 first in one transaction without parent lock, and then do the
2722 final unlink in a new transaction with the parent lock. This
2723 reduces the lock hold time dramatically.
2726 Frequency : rare, 2.4 kernels only
2728 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2729 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2730 before unmounting their filesystem in order to avoid deadlock.
2731 The kernel_locked() function in 2.4 kernels only checks whether
2732 the kernel lock is held, not whether it is this process that is
2733 holding it as 2.6 kernels do.
2738 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2739 Details : The client ptlrpc code may be trying to reconnect to a down
2740 server before a previous connection attempt has timed out.
2741 Increase the reconnect interval to be longer than the connection
2742 timeout interval to avoid sending duplicate connections to
2746 Frequency : echo_client brw_test command
2748 Description: fix echo_client to work with OST preallocated code
2749 Details : OST preallocation code (5137) didn't take echo_client IO path
2750 into account: echo_client calls filter methods outside of any
2751 OST thread and, hence, there is no per-thread preallocated
2752 pages and buffers to use. Solution: hijack pga pages for IO. As
2753 a byproduct, this avoids unnecessary data copying.
2757 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2758 Description: Client can oops in mdc_commit_close() after open replay
2759 Details : It was possible for the MDS to return an open request with no
2760 transaction number in mds_finish_transno() if the client was
2761 evicted, but without actually returning an error. Clients
2762 would later try to replay that open and may trip an assertion
2763 Simplify the client close codepath, and always return an error
2764 from the MDS in case the open is not successful.
2767 Frequency : rare, 2.6 OSTs only
2769 Description: OST may deadlock under high load on fragmented files
2770 Details : If there was a heavy load and highly-fragmented OST filesystems
2771 it was possible to have all the OST threads deadlock waiting on
2772 allocation of biovecs, because the biovecs were not released
2773 until the entire RPC IO was completed. Instead, release biovecs
2774 as soon as they are complete to ensure forward IO progress.
2776 Severity : enhancement
2778 Description: Support for specifying external journal device at mount
2779 Details : If an OST or MDS device is formatted with an external journal
2780 device, this device major/minor is stored in the ext3 superblock
2781 and may not be valid for failover. Allow detecting and
2782 specifying the external journal at mount time.
2787 Description: Mounting an MDS with pending unlinked files may cause oops
2788 Details : target_finish_recovery() calls mds_postrecov() which returned
2789 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2790 considers this an error and immediately begins cleaning up the
2791 lov, just after starting the mds_lov process
2793 Severity : enhancement
2795 Description: Implement 'lfs df' to report actual free space on per-OST basis
2796 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2797 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2798 usage of objects; '-h' to report in human readable format.
2800 ------------------------------------------------------------------------------
2802 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2809 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2810 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2811 at startup during orphan unlinking. Instead, skip buddy update
2812 before setup, as it will be regenerated after recovery is complete.
2817 Description: If an OST is inactive, its locks might reference stale inodes.
2818 Details : lov_change_cbdata() must iterate over all namespaces, even if
2819 they are inactive to clear inode references from the lock.
2821 Severity : enhancement
2822 Frequency : occasional, if non-standard max_dirty_mb used
2824 Description: Client will block write RPCs if not enough grant
2825 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2826 then the client will block writes while waiting for another RPC
2827 to complete instead of consuming its dirty limit. With change
2828 we get improved performance when max_dirty_mb is small.
2830 Severity : enhancement
2831 Bugzilla : 3389, 6253
2832 Description: Add support for supplementary groups on the MDS.
2833 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2834 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2835 lookups for user supplementary groups into a cache.
2839 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2840 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2841 for liblustre O_EXCL is enforced by MDS.
2845 Description: readdir never returns NULL in liblustre.
2846 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2847 dentry in struct dirent.
2851 Frequency : liblustre only, depends on application IO pattern
2852 Description: liblustre clients evicted if not contacting servers
2853 Details : Don't put liblustre clients into the ping_evictor list, so
2854 they will not be evicted by the pinger ever.
2856 Severity : enhancement
2858 Description: Add ability to evict clients by NID from MDS.
2859 Details : By echoing "nid:$NID" string into
2860 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2861 $NID would be instantly evicted from this MDS and from all active
2862 OSTs connected to it.
2866 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2867 Details : lookup_it_finish() used to query file size from OSTs that was not
2872 Description: service threads change working directory to that of init
2873 Details : Starting lustre service threads may pin the working directory
2874 of the parent thread, making that filesystem busy. Threads
2875 now change to the working directory of init to avoid this.
2879 Frequency : during shutdown only
2880 Description: shutdown with a failed MDS or OST can cause unmount to hang
2881 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2886 Frequency : 2.6 only
2887 Description: chmod/chown may include an extra supplementary group
2888 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2889 supplementary group and if none is specified this is used.
2892 Bugzilla : 5479 (6816)
2894 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2895 Details : If lookup is in progress on a file that is unlinked we might try
2896 to revalidate the inode and fail in revalidate after lookup is
2897 complete and ll_file_open() enqueues the open again but
2898 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2901 Frequency : always, if lconf --abort_recovery used
2903 Description: lconf --abort_recovery fails with 'Operation not supported'
2904 Details : lconf was attempting to abort recovery on the MDT device and not
2907 ------------------------------------------------------------------------------
2909 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2914 Frequency : rare (only unsupported configurations with a node running as an
2916 Bugzilla : 6514, 5137
2917 Description: Mounting a Lustre file system on a node running as an OST could
2919 Details : OSTs now preallocates memory needed to write out data at
2920 startup, instead of when needed, to avoid having to
2921 allocate memory in possibly low memory situations.
2922 Specifically, if the file system is mounted on on OST,
2923 memory pressure could force it to try to write out data,
2924 which it needed to allocate memory to do. Due to the low
2925 memory, it would be unable to do so and the node would
2926 become unresponsive.
2928 Severity : enhancement
2930 Description: Addition of lconf --service command line option
2931 Details : lconf now accepts a '--service <arg>' option, which is
2932 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2934 Severity : enhancement
2936 Description: Failover mode is now the default for OSTs.
2937 Details : By default, OSTs will now run in failover mode. To return to
2938 the old behaviour, add '--failout' to the lmc line for OSTs.
2940 Severity : enhancement
2942 Description: Health checks are now provided for MDS and OSTs
2943 Details : Additional detailed health check information on MSD and OSTs
2944 is now provided through the procfs health_check value.
2947 Frequency : occasional, depends on IO load
2949 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2950 after numerous create/delete cycles
2951 Details : The ext3 inode allocation policy would not allocate new inodes
2952 very well on the OSTs because there are no new directories
2953 being created. Instead we look for groups with free space if
2954 the parent directories are nearly full.
2959 Description: Network or server problems during mount may cause partially
2960 mounted clients instead of returning an error.
2961 Details : The config llog parsing code may overwrite the error return
2962 code during mount error handling, returning success instead
2968 Description: MDS can fail to allocate large reply buffers
2969 Details : After long uptimes the MDS can fail to allocate large reply
2970 buffers (e.g. zconf client mount config records) due to memory
2971 fragmentation or consumption by the buffer cache. Preallocate
2972 some large reply buffers so that these replies can be sent even
2973 under memory pressure.
2977 Frequency : rare (liblustre)
2978 Description: fsx running with liblustre complained that using truncate() to
2979 extend the file doesn't work. This patch corrects that issue.
2980 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2981 ATTR_SIZE and lsm use in llu_setattr_raw.
2985 Frequency : rare, only 2.6 kernels
2986 Description: Unusual file access patterns on the MDS may result in inode
2987 data being lost in very rare circumstances.
2988 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2989 code in the 2.6 kernel caused the inode and/or EA data not to
2990 be read from disk, causing single-file corruption.
2994 Frequency : rare, only 2.6 filesystems using extents
2995 Description: Heavy concurrent write and delete load may cause data corruption.
2996 Details : It was possible under high-load situations to have an extent
2997 metadata block in the block device cache from a just-unlinked
2998 file overwrite a newly-allocated data block. We now unmap any
2999 metadata buffers that alias just-allocated data blocks.
3003 Frequency : filesystems with default stripe_count larger than 77
3004 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3005 Details : lconf specifies an inode size of 4096 bytes when the default
3006 stripe_count is larger than 77. This conflicts with the default
3007 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3008 this case to avoid pinning too much memory for large EAs.
3010 ------------------------------------------------------------------------------
3012 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3017 Frequency : rare (extremely heavy IO load with hundreds of clients)
3019 Description: Client is evicted, gets IO error writing to file
3020 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3021 caused clients to be evicted for AST timeouts. The fixes in
3022 bug 5192 mean we no longer need to have such short AST timeouts
3023 so ldlm_timeout has been increased.
3026 Frequency : occasional during --force or --failover shutdown under load
3027 Bugzilla : 5949, 4834
3028 Description: Server oops/LBUG if stopped with --force or --failover under load
3029 Details : a collection of import/export refcount and cleanup ordering
3030 issues fixed for safer force cleanup
3033 Frequency : only filesystems larger than 120 OSTs
3034 Bugzilla : 5990, 6223
3035 Description: lfs getstripe would oops on a very large filesystem
3036 Details : lov_getconfig used kfree on vmalloc'd memory
3039 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3040 Bugzilla : 6242, 6243
3041 Description: reading from files that had been truncated to a non-zero size
3042 but never opened returned no data
3043 Details : ll_file_read() reads zeros from no-object files to EOF
3048 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3049 Details : There's little harm in aborting MDS/OSS recovery and letting it
3050 try again, so I removed the LASSERT and return an error instead.
3052 Severity : enhancement
3054 Description: New debugging infrastructure for tracking down data corruption
3055 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3056 (b) cover more of the client-side code path, and (c) try to narrow
3057 down where problems occurred
3061 Bugzilla : 3819, 4364, 4397, 6313
3062 Description: Racing close and eviction MDS could cause assertion in mds_close
3063 Details : It was possible to get multiple mfd references during close and
3064 client eviction, leading to one thread referencing a freed mfd.
3066 Severity: : enhancement
3067 Bugzilla : 3262, 6359
3068 Description: Attempts to reconnect to servers are now more aggressive.
3069 Details : This builds on the enhanced upcall-less recovery that was added
3070 in 1.4.2. When trying to reconnect to servers, clients will
3071 now try each server in the failover group every 10 seconds. By
3072 default, clients would previously try one server every 25 seconds.
3077 Description: After recovery, certain operations trigger a failed
3078 assertion on a client.
3079 Details : Failing over an mds, using lconf -d --failover, while a
3080 client was doing a readdir() call would cause the client to
3081 LBUG after recovery completed and the readdir() was resent.
3083 Severity : enhancement
3085 Description: Default groups are now added by lconf
3086 Details : You can now run lconf --group <servicename> without having to
3087 manually add groups with lmc.
3090 Frequency : occasional
3092 Description: Nodes with an elan id of 0 trigger a failed assertion
3095 Frequency : always when accessing e.g. tty/console device nodes
3097 Description: tty and some other devices nodes cannot be used on lustre
3098 Details : file's private_data field is used by device data and lustre
3099 values in there got lost. New field was added to struct file to
3100 store fs-specific private data.
3103 Frequency : when exporting Lustre via NFS
3105 Description: NFSD failed occasionally when looking up a path component
3106 Details : NFSD is looking up ".." which was broken in ext3 directories
3107 that had grown large enough to become hashed.
3110 Frequency : Clusters with multiple interfaces not on the same subnet
3112 Description: Nodes will repeatedly try to reconnect to an interface which it
3113 cannot reach and report an error to the log.
3114 Details : Extra peer list entries will be created by lconf with some peers
3115 unreachable. lconf now validates the peer before adding it.
3118 Frequency : Only if a default stripe is set on the filesystem root.
3120 Description: Setting a default stripe on the filesystem root prevented the
3121 filesystem from being remounted.
3122 Details : The client was sending extra request flags in the root getattr
3123 request and did not allocate a reply buffer for the dir EA.
3126 Frequency : occasional, higher if lots of files are accessed by one client
3127 Bugzilla : 6159, 6097
3128 Description: Client trips assertion regarding lsm mismatch/magic
3129 Details : While revalidating inodes the VFS looks up inodes with ifind()
3130 and in rare cases can find an inode that is being freed.
3131 The ll_test_inode() code will free the lsm during ifind()
3132 when it finds an existing inode and then the VFS later attaches
3133 this free lsm to a new inode.
3137 Bugzilla : 6422, 7030
3138 Description: MDS deadlock between mkdir and client eviction
3139 Details : Creating a new file via mkdir or mknod (starting a transaction
3140 and getting the ns lock) can deadlock with client eviction
3141 (gets ns lock and trying to finish a synchronous transaction).
3144 Frequency : occasional
3145 Description: While starting a server, the fsfilt_ext3 module could not be
3147 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3148 kernels. Previously, lconf would still use the ext3 name
3149 when trying to load modules. Now, it will correctly use
3150 ext3 on 2.4 and ldiskfs on 2.6.
3152 Severity : enhancement
3153 Description: The default stripe count has been changed to 1
3154 Details : The interpretation of the default stripe count (0, to lfs
3155 or lmc) has been changed to mean striping across a single
3156 OST, rather than all available. For general usage we have
3157 found a stripe count of 1 or 2 works best.
3159 Severity : enhancement
3160 Description: Add support for compiling against Cray portals.
3161 Details : Conditional compiling for some areas that are different
3165 Frequency : occasional
3166 Bugzilla : 6409, 6834
3167 Description: Creating files with an explicit stripe count may lead to
3168 a failed assertion on the MDS
3169 Details : If some OSTs are full or unavailable, creating files may
3170 trigger a failed assertion on the MDS. Now, Lustre will
3171 try to use other servers or return an error to the
3175 Frequency : occasional
3177 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3178 caused lock timeout during readahead (since 1.4.2).
3179 Details : Processes doing readahead might match a lock that hasn't been
3180 granted yet if there are overlapping and conflicting lock
3181 requests. The readahead process waits on ungranted lock
3182 (original lock is CBPENDING), while OST waits for that process
3183 to cancel CBPENDING read lock and eventually evicts client.
3185 Severity : enhancement
3187 Description: Initial enabling of flock support for clients
3188 Details : Implements fcntl advisory locking and file status functions.
3189 This feature is provided as an optional mount flag (default
3190 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3191 locking are implemented yet, and those that are are not guaranteed
3192 to be completely correct in production environments.
3193 mount -t lustre -o [flock|noflock] ...
3196 Frequency : occasional
3198 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3199 assertion in the ext3 JBD (journaling) layer.
3200 Details : The b_committed_data struct is protected by the big kernel lock
3201 in 2.4 kernels, serializing journal_commit_transaction() and
3202 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3203 access to this struct. In 2.6 kernels there is finer grained
3204 locking to improve SMP performance of the JBD layer.
3208 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3213 Description: merge and cleanup kernel patches.
3216 -----------------------------------------------------------------------------
3218 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3220 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3221 versions of Lustre. Please update all servers and clients to
3222 version 1.4.2 or later at the same time. You must also run
3223 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3224 to update the configuration logs.
3226 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3227 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3228 - swab llog records of type '0' so we get proper header size/idx (5861)
3229 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3230 - fix rename of one directory over another leaking an inode (5953)
3231 - avoid SetPageDirty on 2.6 (5981)
3232 - don't re-add just-being-destroyed locks to the waiting list (5653)
3233 - when creating new directories, inherit the parent's custom
3234 striping settings if present parent (3048)
3235 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3236 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3237 - don't hold client locks on temporary worklist from l_lru (5666)
3238 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3239 - automatically evict dead clients (5921)
3240 - Update file size properly in create+truncate+fstat case (6196)
3241 - Do not unhash mountpoint dentries, do not allow removal of
3243 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3244 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3245 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3246 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3247 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3248 - Correctly handle reads of files with no objects (6243)
3249 - lctl recover will also mark a device active if deactivate used (5933)
3251 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3252 - allow --write-conf on an MDS with different nettype than client (5619)
3253 - don't write config llogs to MDS for mounts not from that MDS (5617)
3254 - lconf should create multiple TCP connections from a client (5201)
3255 - init scripts are now turned off by default; run chkconfig --on
3256 lustre and chkconfig --on lustrefs to use them
3257 - upcalls are no longer needed for clients to recover to failover
3259 - add --abort-recovery option to lconf to abort recovery on device
3261 - add support for an arbitrary number of OSTs (3026)
3262 - Quota support protocol changes.
3263 - forward compatibility changes to wire structs (6007)
3264 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3265 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3266 - improved "lustre status" script
3267 - initialize blocksize for non-regular files (6062)
3268 - added --disable-server and --disable-client configure options (5782)
3269 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3270 - Vanilla 2.4.29 support
3271 - increase maximum number of obd devices to 520 (6242)
3272 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3273 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3274 SLES 9 2.6.5 kernels
3275 - limit stripes per file to 160 (the maximum EA size) (6093)
3277 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3280 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3281 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3282 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3283 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3284 - small ext3 extents cleanups and fixes (5733)
3285 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3286 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3287 - avoid needless client->OST connect, fix handle mismatch (5317)
3288 - fix DLM error path that led to out-of-sync client, long delays (5779)
3289 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3290 - fix several locking issues related to i_size (5492,5624,5654,5672)
3291 - don't move pending lock onto export if it is already evicted (5683)
3292 - fix kernel oops when creating .foo in unlinked directory (5548)
3293 - fix deadlock in obdfilter statistics vs. object create (5811)
3294 - use time_{before,after} to avoid timer jiffies wrap (5882)
3295 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3296 - Do not leak request if server was not able to process it (5154)
3297 - If mds_open unable to find parent dir, make that negative lookup(5154)
3298 - don't create new directories with extent-mapping (5909, 5936)
3300 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3301 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3302 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3303 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3304 - support for RHEL4 kernel on i686 (5773)
3305 - provide error messages when incompatible logs are encountered (5898)
3307 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3308 * version 1.4.0.10 (1.4.1 release candidate 1)
3310 - don't keep a lock reference when lock is not granted (4238)
3311 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3312 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3313 - drop import inflight refcount on signal_completed_replay error (5255)
3314 - unlock page after async write error during send (3677)
3315 - handle missing objects in filter_preprw_read properly (5265)
3316 - no transno return for symlink open, don't save no-trasno open (3440)
3317 - don't try to complete elan receive that already failed (4012)
3318 - free RPC server reply state on error (5406)
3319 - clean up thread from ptlrpc_start_thread() on error (5160)
3320 - readahead could read extra page into cache that wasn't ejected (5388)
3321 - prevent races in class_attach/setup/cleanup/detach (5260)
3322 - don't dereference de->d_inode after l_dput of de (5458)
3323 - use "int" for stripe value returned from lock_to_stripe (5544)
3324 - mballoc allocation and error-checking fixes in 2.6 (5504)
3325 - block device patches to fix I/O request sizes in 2.6 (5482)
3326 - look up hostnames for IB nals (5602)
3327 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3328 - don't start multiple acceptors for the same port (5277)
3329 - fix incorrect LASSERT in mds_getattr_name (5635)
3330 - export a proc file for general "ping" checking (5628)
3331 - fix "lfs check" to not block when the MDS is down (5628)
3333 - service request history (4965)
3334 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3335 - create an "evict_client" /proc entry on OSTs, like the MDS has
3336 - fix mount usage message, return errors per mount(8) (5168)
3337 - change grep [] to grep "[]" in tests so they work in more UMLs
3338 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3339 - remove ancient LOV_MAGIC_V0 EA support (5047)
3340 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3341 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3342 - allow readahead allocations to fail when low on memory (5383)
3343 - mmap locking landed again, after considerable improvement (2828)
3344 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3346 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3349 - send OST transaction number in read/write reply to free req (4966)
3350 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3351 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3352 - the watchdog thread now runs as interruptible (5246)
3353 - flock/lockf fixes (but it's still disabled, pending 5135)
3354 - don't use EXT3 constants in llite code (5094)
3355 - memory shortage at startup could cause assertion (5176)
3357 - reorganization of lov code
3358 - single portals codebase
3360 - add extents/mballoc support (5025)
3361 - direct I/O reads in the obdfilter (4048)
3362 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3364 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3367 - send OST transaction number in read/write reply to free req (4966)
3368 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3369 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3370 - the watchdog thread now runs as interruptible (5246)
3371 - handle missing objects in filter_preprw_read properly (5265)
3372 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3373 - drop import inflight refcount on signal_completed_replay error (5255)
3374 - unlock page after async write error during send (3677)
3375 - return original error code on reconstructed replies (3761)
3376 - no transno return for symlink open, don't save no-trasno open (3440)
3378 - add pid to ldlm debugging output (4922)
3379 - bump the watchdog timeouts -- we can't handle 30sec yet
3380 - extra debugging for orphan dentry/inode bug (5259)
3382 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3385 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3386 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3387 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3388 - fix NULL dereference in /proc/sys/portals/routes (4827)
3389 - allow failed mdc_close() operations to be interrupted (4561)
3390 - stop precreate on OST before MDS would time out on it (4778)
3391 - don't send partial-page writes before EOF from client (4410)
3392 - discard client grant for sub-page writes on large-page clients (4520)
3393 - don't free dentries not owned by NFS code, check generation (4806)
3394 - fix lsm leak if mds_create_objects() fails (4801)
3395 - limit debug_daemon file size, always print CERROR messages (4789)
3396 - use transno after validating reply (3892)
3397 - process timed out requests if import state changes (3754)
3398 - update mtime on OST during writes, return in glimpse (4829)
3399 - add mkfsoptions to LDAP (4679)
3400 - use ->max_readahead method instead of zapping global ra (5039)
3401 - don't interrupt __l_wait_event() during strace
3403 - add software watchdogs to catch hung threads quickly (4941)
3404 - make lustrefs init script start after nfs is mounted
3405 - fix CWARN/ERROR duplication (4930)
3406 - return async write errors to application if possible (2248)
3407 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3408 - print NAL number in %x format (4645)
3409 - update barely-supported suse-2.4.21-171 series (4842)
3410 - support for sles 9 %post scripts
3411 - support for building 2.6 kernel-source packages
3412 - support for sles km_* packages
3414 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3417 - ignore -ENOENT errors in osc_destroy (3639)
3418 - notify osc create thread that OSC is being cleaned up (4600)
3419 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3420 - reconstruct ost_handle() like mds_handle() (4657)
3421 - create a new thread to do import eviction to avoid deadlock (3969)
3422 - let lconf resolve symlinked-to devices (4629)
3423 - don't unlink "objects" from directory with default EA (4554)
3424 - hold socknal file ref over connect in case target is down (4394)
3425 - allow more than 32000 subdirectories in a single directory (3244)
3426 - fix blocks count for O_DIRECT writes (3751)
3427 - OST returns ENOSPC from object create when no space left (4539)
3428 - don't send truncate RPC if file size isn't changing (4410)
3429 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3430 - bind to privileged port in socknal and tcpnal (3689)
3432 - rate limit CERROR/CWARN console message to avoid overload (4519)
3433 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3434 - basic mmap support (3918)
3435 - kernel patch series update from b1_4 (4711)
3437 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3440 - avoid crash during MDS cleanup with OST shut down (2775)
3441 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3442 - don't use bad inodes on the MDS (3744)
3443 - dynamic object preallocation to improve recovery speed (4236)
3444 - don't hold spinlock over lock dumping or change debug flags (4401)
3445 - don't zero obd_dev when it is force cleaned (3651)
3446 - print grants to console if they go negative (4431)
3447 - "lctl deactivate" will stop automatic recovery attempts (3406)
3448 - look for existing locks in ldlm_handle_enqueue() (3764)
3449 - don't resolve lock handle twice in recovery avoiding race (4401)
3450 - revalidate should check working dir is a directory (4134)
3452 - don't always mark "slow" obdfilter messages as errors (4418)
3454 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3457 - don't close LustreDB during write_conf until it is done (3860)
3458 - fix typo in lconf for_each_profile (3821)
3459 - allow dumping logs from multiple threads at one time (3820)
3460 - don't allow multiple threads in OSC recovery (3812)
3461 - fix debug_size parameters (3864)
3462 - fix mds_postrecov to initialize import for llog ctxt (3121)
3463 - replace config semaphore with spinlock (3306)
3464 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3465 - don't allow enqueue to complete on a destroyed export (3822)
3466 - down write_lock before checking llog header bitmap (3825)
3467 - recover from lock replay timeout (3764)
3468 - up llog sem before sending rpc (3652)
3469 - reduce ns lock hold times when setting kms (3267)
3470 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3471 - fix NULL deref and obd_dev leak on setup error (3312)
3472 - replace some LBUG about llog ops with error handling (3841)
3473 - don't match INVALID dentries from d_lookup and spin (3784)
3474 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3475 - fix invalid assertion in ptlrpc_set_wait (3880)
3477 - add libwrap support for the TCP acceptor (3996)
3478 - add /proc/sys/portals/routes for non-root route listing (3994)
3479 - allow setting MDS UUID in .xml (2580)
3480 - print the stack of a process that LBUGs (4228)
3482 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3485 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3486 - make sure to unset replay flag from failed open requests (3440)
3487 - if default stripe count is 0, use OST count for inode size (3636)
3488 - update parent mtime/ctime on client for create/unlink (2611)
3489 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3490 - free recovery state on server during a forced cleanup (3571)
3491 - unregister_reply for resent reqs (3063)
3492 - loop back devices mounting and status check on 2.6 (3563)
3493 - fix resource-creation race that can provoke i_size == 0 (3513)
3494 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3495 - more debugging for page-accounting assertion (3746)
3496 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3497 - avoid deadlock after precreation failure (3758)
3498 - fix race and lock order deadlock in orphan handling (3450, 3750)
3499 - add validity checks when grabbing inodes from l_ast_data (3599)
3501 - add /proc/.../recovery_status to obdfilter (3428)
3502 - lightweight CDEBUG infrastructure, debug daemon (3668)
3503 - change default OSC RPC parameters to be better on small clusters
3504 - turn off OST read cache for files smaller than 32MB
3505 - install man pages and include them in rpms (3100)
3506 - add new init script for (un)mounting lustre filesystems (2593)
3507 - run chkconfig in %post for init scripts (3701)
3508 - drop scimac NAL (unmaintained)
3510 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3513 - clean kiobufs before and after use (3485)
3514 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3515 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3516 - change init script to fail more clearly if not run as root (1528)
3517 - allow clients to reconnect during replay (1742)
3518 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3519 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3520 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3521 - fix cond_resched (3554)
3522 - only evict unfinished clients after recovery (3515)
3523 - allow bulk resend, prevent data loss (3570)
3524 - dynamic ptlrpc request buffer allocation (2102)
3525 - don't allow unlinking open directory if it isn't empty (2904)
3526 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3527 - remove extraneous obd dereference causing LASSERT failure (3334)
3528 - don't use get_cycles() when creating temp. files on the mds (3156)
3529 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3530 - handle EEXIST for set-stripe, set proper directory name (3336)
3532 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3533 - fix ksocknal_fmb_callback() error messages (2918)
3535 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3538 - don't copy lvb into (possibly NULL) reply on error (2983)
3539 - don't deref dentry after dput, don't free lvb on error (2922)
3540 - use the kms to determine writeback rpc length (2947)
3541 - increment oti_logcookies when osc is inactive (2948)
3542 - update client's i_blocks count via lvb messages (2543)
3543 - handle intent open/close of special files properly (1557)
3544 - mount MDS with errors=remount-ro, like obdfilter (2009)
3545 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3546 - don't use cancelling-locks' kms values (2947)
3547 - use highest lock extent for kms, not last one (2925)
3548 - don't dereference ERR_PTR() dentry in error handling path (3107)
3549 - fix thread race in portals_debug_dumplog() (3122)
3550 - create lprocfs device entries at setup instead of at attach (1519)
3551 - common AST error handler, don't evict client on completion race (3145)
3552 - zero nameidata in detach_mnt in 2.6 (3118)
3553 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3554 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3555 - initialize RPC timeout timer earlier for 2.6 (3219)
3556 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3557 - print nal/nid for unknown nid (3258)
3558 - additional checks for oscc recovery before doing precreate (3284)
3559 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3560 - don't crash in mdc_close for bad permissions on open (3285)
3561 - zero i_rdev for non-device files (3147)
3562 - clear page->private before handing to FS, better assertion (3119)
3563 - tune the read pipeline (3236)
3564 - fix incorrect decref of invalidated dentry (2350)
3565 - provide read-ahead stats and refine rpc in flight stats (3328)
3566 - don't hold journal transaction open across create RPC (3313)
3567 - update atime on MDS at close time (3265)
3568 - close LDAP connection when recovering to avoid server load (3315)
3569 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3570 - don't leak open file on MDS after open resend (3325)
3571 - serialize filter_precreate and filter_destroy_precreated (3329)
3572 - loop device shouldn't call sync_dev() for nul device (3092)
3573 - clear page cache after eviction (2766)
3574 - resynchronize MDS->OST in background (2824)
3575 - refuse to mount the same filesystem twice on same mountpoint (3394)
3576 - allow llmount to create routes for mounting behind routers (3320)
3577 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3578 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3579 - fix rare problem with rename on htree directories (3417)
3581 - allow default OST striping configuration per directory (1414)
3582 - fix compilation for qswnal for 2.6 kernels (3125)
3583 - increase maximum number of MDS request buffers for large systems
3584 - change liblustreapi to be useful for external progs like lfsck (3098)
3585 - increase local configuration timeout for slow disks (3353)
3586 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3588 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3591 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3592 - don't overwrite extent policy data in reply if lock was blocked (2901)
3593 - drop filter export grants atomically with removal from device (2663)
3594 - del obd_self_export from work_list in class_disconnect_exports (2908)
3595 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3596 - swab reply message in mdc_close, other PPC fixes (2464)
3597 - fix destroying of named logs (2325)
3598 - overwrite old logs when running lconf --write_conf (2264)
3599 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3600 - fix race in target_handle_connect (2898)
3601 - mds_reint_create() should take same inode create lock (2926)
3602 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3603 - don't close files for self_export to avoid uninitialized obd (2936)
3604 - allow MDS with the same name as client node (2939)
3605 - hold dentry reference for closed log files for unlink (2325)
3606 - reserve space for all logs during transactions (2059)
3607 - don't evict page beyond end of stripe extent (2925)
3608 - don't oops on a deleted current working directory (2399)
3609 - handle hard links to targets without a parent properly (2517)
3610 - don't dereference NULL lock when racing during eviction (2867)
3611 - don't grow lock extents when lots of conflicting locks (2919)
3613 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3616 - account for cache space usage on clients to avoid data loss (974)
3617 - lfsck support in lustre kernel code (2349)
3618 - reduce journal credits needed for BRW writes (2370)
3619 - orphan handling to avoid losing space on client/server crashes
3620 - ptlrpcd can be blocked, stopping ALL progress (2477)
3621 - use lock value blocks to assist in proper KMS, faster stat (1021)
3622 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3623 - recovery for initial connections (2355)
3624 - fixes for mds_cleanup_orphans (1934)
3625 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3626 - block all file creations until orphan recovery completes (1901)
3627 - client remove rq_connection from request struct (2423)
3628 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3629 - recovery timer race (2670)
3630 - mdc_close recovey bug (2532)
3631 - ptlrpc cleanup bug (2710)
3632 - mds timeout on local locks (2588)
3633 - namespace lock held during RPCs (2431)
3634 - handle interrupted sync write properly (2503)
3635 - don't try to handle a message that hasn't been replied to (2699)
3636 - client assert failure during cleanup after abort recovery (2701)
3637 - leak mdc device after failed mount (2712)
3638 - ptlrpc_check_set allows timedout requests to complete (2714)
3639 - wait for inflight reqs when ptlrpcd finishes (2710)
3640 - make sure unregistered services are removed from the srv_list
3641 - reset bulk XID's when resending them (caught by 1138 test)
3642 - unregister_bulk after timeout
3643 - fix lconf error (2694)
3644 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3645 - readahead locks pages, leaves pending causing memory pressure (2673)
3646 - increase OST request buffers to 4096 on large machines (2729)
3647 - fix up permission of existing directories in simple_mkdir (2661)
3648 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3649 - don't assign transno to errored transactions (2742)
3650 - don't delete objects on OST if given a bogus objid from MDS (2751)
3651 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3652 - if rq_no_resend, then timeout request after recovery (2432)
3653 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3654 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3655 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3656 - MDS recovery shouldn't delete objects that it has given out (2730)
3657 - if enqueue arrives after completion, don't clobber LVB (2819)
3658 - don't unlock pages twice when trigger_group_io returns error (2814)
3659 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3660 - don't write pages to disk if there was an error (1450)
3661 - don't ping imports that have recovery disabled (2676)
3662 - take buffered bytes into account when balancing socknal conn (2817)
3663 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3664 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3665 - remove little-endian swabbing of llog records (1987)
3666 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3667 - retry reposting request buffers if they fail (1191)
3668 - grow extent at grant time to avoid granting a revoked lock (2809)
3669 - lock revoke doesn't evict page if covered by a second lock (2765)
3670 - disable VM readahead to avoid reading outside lock extents (2805)
3672 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3673 - updated kernel patches for hp-2.4.20 kernel (2681)
3675 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3678 - fix truncated write corruption (2366)
3679 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3681 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3682 - ensure len(uuid) < 37 in lmc (1171)
3683 - fix ia64 OOPS in llog_test (2255)
3684 - zero end of page at obdfilter for partial page writes (2648)
3685 - don't leave stale dentries around after renames (bug 2428)
3686 - fix timeouts when evicting a client with a single lock held (2642)
3687 - set deadline for the initial HELLO message to drain (2634)
3688 - print out dotted-quad IP addresses in the socknal (2302)
3690 - additional debugging for MDS client eviction problem (2443)
3691 - fix mkfsoptions support for osts (2603, 2604)
3693 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3696 - add series for the vanilla 2.6.0 kernel
3697 - add series for the vanilla 2.4.24 kernel
3698 - add series for a cray x86/64 UL kernel drop
3699 - fix xattr patches for the vanilla 2.4.19 series
3701 - generate true UUIDs in lmc (1171)
3702 - have portals stack dumping break in UML (2466)
3703 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3704 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3705 - wake callback waiting threads on client eviction (2460)
3706 - Add --ptldebug and --subsystem to lmc (1719)
3707 - update assertion to allow safe interrupt allocation
3708 - set rq_no_resend for cancel requests (2432)
3709 - recalculate ptlrpcd timeout after resend (2494)
3710 - call vfs_rmdir when removing pending directories (2368)
3711 - fix renaming a file to itself (2429)
3712 - lmc creates a default one-stripe lov (2454)
3713 - expand procfs space to handle large clusters (2326)
3714 - increase UML stack to avoid overflow
3715 - update lconf's list of debug and subsystem masks
3716 - fix lfs find --obd (2510)
3717 - /proc tunable for disabling filter read caching (2591)
3718 - stop rpm packages from altering slapd.conf (2301)
3719 - disable nagle in the socknal under 0conf (2578)
3720 - choose mds inode size based on stripe count (2572)
3721 - fix kernel-source rpm problems (2516)
3723 - add --disable-doc to avoid pdf generation (2421)
3724 - update documentation, tests, type-os, comments
3725 - avoid format warnings on ia64
3726 - remove the TOE NAL
3727 - tiny code cleanups by removing unused fields
3729 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3732 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3733 - remove the most blatant lies from BUILDING file (2371)
3734 - change default debug level to reasonable production setting
3735 - reduce client side cache size to reduce cache flush time
3736 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3737 - make TCP zerocopy and pinger support enabled by default (2476)
3738 - sync writes completed after process exits caused crashes (2319)
3739 - maintain correct mount count on the MDS (2356)
3740 - backout 1557, because 2316 wasn't really fixed
3741 - better file I/O statistics gathering in /proc
3742 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3743 - another kernel patch to fix zero-copy TCP function export
3744 - don't take duplicate lock when processing re-sent getattr (2420)
3745 - lctl uses obd_self_export instead of creating new conn (2353)
3746 - MDS/OST recovery case which requires object creation asserted (2425)
3747 - move lfs from /usr/sbin to /usr/bin in packages
3748 - fix race between mds_client_add and mds_client_free (2417)
3749 - use kmalloc instead of slabs in portals (2430)
3750 - don't create duplicate records when a failover MDS is present (2442)
3751 - remove unnecessary mount age check (2332)
3752 - don't remove directory inodes from locks prematurely (2451)
3753 - don't break if MDS service name is the same as hostname (2103)
3754 - fix races in client write RPC generation when cache full (2482)
3756 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3759 - remove now-unused request->rq_obd (278)
3760 - if an allocation fails, print out how much memory we've used (1933)
3761 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3762 - add the "configurable stack size" patch to most series files (1256)
3763 - ability to write large log records, for 100+ OST configs (2306)
3764 - fix NULL deref when filter_prep fails (2314)
3765 - fix operator precedence error in filter_sync
3766 - dynamic allocation of socknal TX descriptors (2315)
3767 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3768 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3769 - ensure that CWARN messages in Portals always get to the syslog
3770 - __init/__exit are not for prototype decls (ldlm_init/exit)
3771 - x86-64 compile warning fixes
3772 - fix gateway LMC keyword conflict (2318)
3773 - fix MDS lock inversions in getattr/reint paths (1844)
3774 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3775 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3776 - DLM race condition prevented some lock evictions (2328)
3777 - ENOMEM detection and retry on socknal sends (2230)
3778 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3779 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3780 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3781 - better error messages when a client is rejected during recovery (1505)
3782 - avoid cancelling locks which were never granted, after failure (2330)
3783 - fix i_sem/journal inversion in mds_client_add (2333)
3784 - fix truncate/getattr lock cycle deadlock (2334)
3785 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3786 - fix two rare exit paths which could leak an l_lock() ref (2321)
3787 - fencepost error in MDS/OST orphan recovery (2226)
3788 - make log record alignment 8 bytes (1988)
3789 - lstripe now fails when requested offset > ost_count (2237)
3790 - ensure that all kernel series have a complete list.h (1607)
3791 - fix crashes in special-file operations (2316)
3792 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3794 - allow configurable automake binary, for testing new versions
3795 - small update to the lfs documentation
3797 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3799 * fix negative export reference count in fsfilt_sync (2312)
3801 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3802 * release candidate 0.9.1
3804 - orphans are moved into the PENDING directory for possible recovery
3805 - replayed opens now open by fid for orphan/rename safety (1042)
3806 - last close of an orphan inode generates a transno (683)
3807 - chdir() and mount() now pin the directory entry (1020)
3808 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3809 - discard very old requests without processing them (1502)
3810 - remove some common, well-understood CERRORs (1505)
3811 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3812 - clear "grant" flags in OST replies until OST grant code lands (1644)
3813 - fix read performance by not clobbering i_blksize on client (1598)
3814 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3815 - make LRU size a /proc tunable, clears locks when reduced (707)
3816 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3817 - server locks take references on exports now (1558)
3818 - build fixes for 2.4.20-rh trees (1663)
3819 - return an error from lov_create if all OSCs are inactive (1751)
3820 - fix import levels when a reconnect happens without a timeout (1597)
3821 - exit early from mds_open if we get a lookup error (1749)
3822 - partial page read at EOF wouldn't wait for disk before sending (1642)
3823 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3824 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3825 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3826 - write/truncate lock inversion (1639)
3827 - Don't auto-load obdclass, portals modules during cleanup (1495)
3828 - fix timestamps from jumping to "now" (1763)
3829 - extra journal assertions (1648)
3830 - add an extra multiunlink test (1771)
3831 - fix read_record/write_record API (1776)
3832 - fix leak of offset_extent, possible incorrect i_size later (1772)
3833 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3834 - Debugging for the kqswnal_get_idle_tx problems (1820)
3835 - Allow recovery to be attempted multiple times (1536)
3836 - Write out MDS last_rcvd file after it is first created (1600)
3837 - Fix tx_descriptor leak in failed transmit situations (1827)
3838 - ext3 journaling fixes for assertion failure after IO error (1871)
3839 - class_export_put() on freed export after completion AST error (1896)
3840 - Fix revalidate looping in VFS (1322)
3841 - Don't access a freed export during MDS_REINT timeout (1521)
3842 - Add open-unlink recovery support on the MDS (1673,1764)
3843 - Return an error if no MDS data was read from last_rcvd (1946)
3844 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3845 - Don't setup a disk device that doesn't match exported UUID (317)
3846 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3847 - avoid committing NULL handle in force close
3848 - local.sh is now a one-stripe LOV configuration
3849 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3850 - fix ext3 htree duplicate directory entry corruption (1516)
3851 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3852 - update to new LOV EA format (2097)
3853 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3855 2003-06-15 Phil Schwan <phil@clusterfs.com>
3858 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3859 - per-import/export recovery handling (958, 931, 959)
3860 - multiple last-rcvd slots, for serving multiple FSes (949)
3861 - connections are again shared between multiple imp/exports (963, 964)
3862 - "umount -f" would hang if any requests needed to be sent (393, 978)
3863 - avoid pinning large req buffer by copying for queued messages (989)
3864 - add "uuid" to "lctl device" command to help upcalls (991)
3865 - "open" RPCs with transnos would confuse recovery counters (1037)
3866 - do proper endian conversion of all wire messages (288, 340, 891)
3867 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3868 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3869 - fix ptlrpc_connection leak in target_handle_connect (1174)
3870 - fix import refcounting bug in OST and MDS cleanup (1134)
3871 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3872 - fix very unlikely obd_types race condition (501)
3873 - remove osc_open hack for echo_client (1187)
3874 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3875 - a failure in read_inode2 leads to deadlock (1139)
3876 - cancel ack-locks as soon as transaction is committed (1072)
3877 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3878 - make sure to commitrw after any preprw to avoid deadlock (1162)
3879 - failing to execute a file in a lustre FS would lock inode (1203)
3880 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3881 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3882 - obd_uuid2tgt would walk off the end of the list (1255)
3883 - on IA64 the osc would give portals incorrect bulk size (1258)
3884 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3885 - fix lock inversion caused by new llite matching code (1282)
3886 - limit the number of dirty pages on a client to 10MB (1286)
3887 - timed out locks were not being corrected cancelled (1289)
3888 - fix O_DIRECT above 4GB on IA-32 (1292)
3889 * major user-visible changes
3890 - fail out/fail over policy now controlled by the upcall (993)
3892 - add OBD_PING to check server availability and failure (954)
3893 - lustre messages are now sent in sending host order (288, 340, 891)
3894 - add eadatalen to MDS getattr reply (340)
3895 - OST read replies may contain second buffer, with per-page status (593)
3897 2003-03-11 Phil Schwan <phil@clusterfs.com>
3900 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3901 - failed open()s could cause deadlock; fixed (867, 869)
3902 - stop cancelling OST locks when files are closed (481)
3903 - overlapping XID spaces caused network corruption (851, 853)
3904 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3905 - setattr_raw added, to do single-RPC, server-side setattrs
3906 - lmc/lconf syntax change for OST UUIDs
3907 - fix crashy race condition between ptlrpc_free_req and osc_close
3908 - don't use request in mdc_enqueue if we hit a timeout (889)
3909 - don't set the inode i_size for regular files from the MDS (896)
3910 - handle out of order completion AST (842)
3911 - don't LBUG if a lock request times out after receiving AST (913)
3912 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3913 - if a bad lock AST arrives, send an error instead of dropping entirely
3914 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3915 - fix leak in bulk IO when only partially completed (899, 900, 926)
3916 - fix O_DIRECT for ia64 (55)
3917 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3918 - C-z after timeout could hang a process forever; fixed (977)
3920 - client-side I/O cache (678, 924, 929, 941, 970)
3922 - READPAGE and SETATTRs which don't take server-side locks get
3925 2003-02-11 Phil Schwan <phil@clusterfs.com>
3928 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3929 - Fix fsfilt_extN_readpage() to read a full page of directory
3930 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3931 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3932 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3933 - Most intent operations were removed, in favour of a new RPC mode
3934 that does a single RPC to the server and bypasses most of the VFS
3935 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3936 - Aggressively cancel local locks on DLM servers
3937 - mds_reint_unlink sends EA to the client if it's the last nlink.
3938 client uses that EA to unlink OST objects.
3939 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3940 - recursive symlinks were fixed (439)
3941 - fixed NULL deref in DEBUG_REQ
3942 - filter_update_lastobjid no longer calls sync, which annoyed extN
3943 - fixed multi-client small-writes to a single file problem (445)
3944 - fixed mtime updates during file writes (607)
3945 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3946 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3947 - fixed Linux OST size reporting problem (444, 656)
3948 - OST now updates object mtime with writes or setattr (607, 619)
3949 - client verifies file size before zeroing page past EOF (445)
3950 - OST now writes last allocated objid to disk with allocation (108)
3951 - LOV on echo now works (409)
3953 - mds_reint_unlink sends a new buffer, with the EA included. this
3954 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3955 set if a regular file was being unlinked, and it was the last link
3956 - use PtlGet from the target for bulk writes (315)
3957 - OST now updates object mtime with writes or setattr (607, 619)
3958 - LDLM now has a grant-time callback to revalidate locked items, if
3960 - Many MDS operations were reorganized to combat race conditions
3962 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3963 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3965 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3968 - Fully reactivate OST imports after reconnection (512, others)
3969 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3970 - More graceful error handling for truncating on dead OST (515)
3971 - Don't error out unless we're actually accessing dead stripes (474)
3972 - Fix garbage sizes when stripes are missing (410)
3973 - LRU counters were broken, causing constant lock purge (433, 432)
3974 - garbage on read from stripes with failed OSTs (441)
3975 - mark OSCs as active before reconnecting during recovery (438)
3976 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3977 - lfind did not preserve OST order in output (443)
3978 - symlinks cause hung clients, incorrect data (439)
3979 - stop dereferencing request after dropping refcount (457)
3980 - don't LASSERT(spin_is_locked) on non-SMP (455)
3981 - fixes for many rename() bugs
3982 - fstat didn't correctly synchronize attributes (399)
3983 - server must handle lock cancellation during blocking AST prep (487)
3984 - bulk descriptors were free()d too soon (511)
3985 - fix paths in lconf, which would load incorrect modules (451, 507)
3986 - fix confusing lconf 'host not found' error message (386)
3987 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3988 - fix race condition in mdc_blocking_ast() for inode access (526)
3989 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3990 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3991 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3992 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3993 - fix lstripe in several ways (kernel side) (527)
3994 - fix request leak in ldlm_cli_enqueue (262)
3995 - incorrect OSC was marked inactive after OST failure
3996 - call mds_fs_cleanup before unmounting filesystem (524)
3997 - fix races between taking ns_lock and ldlm_lock_change_resource
3998 - fix races updating LOV export open file list
3999 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4000 - fix recovery NULL deref in ldlm_cli_cancel_unused
4001 - fix some DLM races by using new hash table for lock handles (419)
4002 - permit the client to specify desired inodes, at replay
4003 - duplicate requests when we queue them for replay reintegration
4004 - fix last_rcvd offset calculation
4005 - sync after each recovered transaction, so we always make progress
4006 - never, not always, ERESTART requests without transnos
4007 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4009 - skip replay if the MDS says that the client is already connected
4010 - don't check for a recovery-enabled export to match lctl's UUID
4011 - don't INC_USE_COUNT for phantom exports
4012 - don't crash when cleaning up phantom exports (567)
4013 - don't double-finish or set replay data for errored mdc_open requests
4014 - abort requests when they time out, so we don't get old replies
4015 - send/receive replies for AST messages again
4016 - if the client says that it doesn't have the lock, cancel it on the
4018 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4019 mark it as destroyed, it will all work out when decref is called
4020 - fix module use counts (22, 581)
4022 - ASTs now expect a reply (server cancels lock on error reply)
4024 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4027 - fix many simultaneous client startup (392)
4028 - fix dentry->d_it clobbering
4029 - credentials weren't being shipped for readdir/getattr operations
4030 - remove invalid assertions triggered during some concurrent MD
4032 - proper Lustre versions added (336, 389)
4033 - fix memory leak for create error case (398)
4034 - fix LOV locking bug that would get cli/srv out of sync
4035 - fix echo client over LOV (409)
4036 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4037 - fix double-O_EXCL intent crash (424)
4038 - avoid sending multiple lock CANCELs (352)
4040 - MDS can do multi-client recovery (modulo bugs in new code)
4042 - many updates, edits, cleanups
4044 2002-11-18 Phil Schwan <phil@clusterfs.com>
4047 - fix null d_it dereference (346)
4048 - fix full OST/dbench hang (333)
4049 - fix permission problem with file removal (286)
4050 - fix removal of OSCs from LOV when they fail
4051 - fix NULL deref during bulk timeout (214)
4052 - fix problems related to multiple filesystems on one MDS (241)
4053 - fixed serious subtle metadata locking bugs
4054 - free locks on clients when inodes are removed due to memory
4056 - fix inode pointer in lock data (285)
4057 - partial support for multiple MDS on a single host (241)
4058 - data locks weren't cancelled at clear_inode time (290, 311)
4059 - intent locks could lead to unbounded lock growth (205)
4060 - added a maximum lock count, an LRU list, and a flusher
4061 - fix multiple rename (365)
4062 - properly abstracted the echo client
4063 - OSC locked 1 byte too many; fixed
4064 - rewrote brw callback code:
4065 - fixed recovery bugs related to LOVs (306)
4066 - fixed too-many-pages-in-one-write crash (191)
4067 - fixed (again) crash in sync_io_timeout (214)
4068 - probably fixed callback-related race (385)
4070 - Add capability to MDS protocol
4071 - LDLM cancellations and callbacks on different portals
4073 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4076 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4077 - defer open object destruction to close time (601981, 138)
4078 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4079 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4080 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4081 - don't call obd_setattr in ll_file_release for destroyed objects
4082 * protocol change to lustre_msg: move |version| and add |flags|
4083 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4084 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4085 * added replay of create, unlink, link and rename operations during
4086 MDS failover; recovery should be much more robust now
4087 * remove failed OSCs from LOVs (only lov_create uses this so far)
4088 * the lustre-HOWTO was brought (more) up to date (582544)
4090 2002-10-23 Phil Schwan <phil@clusterfs.com>
4093 - in-use dentries weren't being reused properly (617851)
4094 - prevent multiple LDLM setup (599178)
4095 - fix LOV size calculations for truncate (617853)
4096 - fix client handling of MDS intent errors (POSIX)
4097 - fix permission bug in lovstripe.c test (624321)
4098 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4099 - truncate past end of file could corrupt data
4100 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4101 - a race in recovery could return ETIMEDOUT to apps (623947)
4102 - building outside the source directory was fixed
4103 * the lustre-HOWTO was brought (more) up to date (582544)
4104 * major progress was made on recovery functionality
4106 2002-10-10 Phil Schwan <phil@clusterfs.com>
4109 - recovery deadlock fix
4110 - rm -rf causes LBUG fix (617817)
4111 - file open by multiple tasks fix (618962)
4112 - directory permissions bugs (602707 and 620007)
4113 - journal_stop fixed with locking (611313)
4114 - O_APPEND failures resolved (618273, perhaps 614459)
4115 - lconf PATH fix (619770)
4116 - IA64 build fix (621450)
4117 - RPC buffer sizes scale with amount of memory
4119 2002-10-01 Phil Schwan <phil@clusterfs.com>
4122 - locks would be cancelled without throwing away data pages,
4123 resulting in inconsistent data (605627)
4124 - inode attributes were not always being refreshed (605627, 612449)
4125 - lconf now continues to cleanup after lctl reports an error
4126 - MDS now enforces user permissions (602707)
4127 - lprocfs cleanup fixed, but not yet enabled (614157)
4128 - fixed infinite server hang, should a client not respond to an AST
4129 - avoid going into recovery if user calls readlink() with a buffer
4130 that's too small (613941)
4131 - AST RPCs no longer require replies (614867) -- this may be changed
4132 - don't crash server if client sends an IOV that's too big (611336)
4133 - fixed lock conversion deadlock (611892)
4134 - fixed the following of symlinks (614622)
4135 * recovery: the server can remove locks from a client that dies, other
4136 clients can make progress
4137 * more extN patch fixes
4138 * compile-time configurable ptlrpc buffer allocations
4140 - collaborative read cache document
4141 - Lustre Lite Performance CDR document-in-progress
4143 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4146 - fix typo in patch-2.4.18
4148 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4151 - clear ptlrpc request each time in handle_incoming_request()
4152 - unlink of files now destroys the object on the OST
4154 2002-09-19 Peter Braam <braam@clusterfs.com>
4156 * add hard link support
4157 * change obdfile creation method
4158 * kernel patch changed
4160 2002-09-19 Peter Braam <braam@clusterfs.com>
4163 - stack overflow bug in extN fixed
4165 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4167 * documentation updates
4168 - add man pages for config tools
4169 - update tests/README to describe testing with new config tools
4170 - finish metadata API descriptions
4171 * bug fixes and cleanups
4172 - statfs workaround for 16TB limit
4173 - LOV stripe allocation improved, can stripe on subset of OSTs
4174 - LOV file size/IO offset was wrong for files > 4GB in size
4175 - object EA data was being dropped, caused files to be unreadable
4176 - memory overflow with non-LOV OST caused memory corruption
4177 - fixed regression tests to work with new config tools, obdfilter
4178 - fixed bug when directory size became larger than 1 block
4179 - fixed bug (for single client case) when PWD was deleted
4180 - invalidate local directory pages when doing intent-based ops
4181 - avoid LDLM oops when lock callback contained bad data
4183 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4185 * documentation updates
4186 * bug fixes and cleanups
4187 - configuration tools
4190 - 64-bit compile warnings
4191 - 64-bit internal statfs data
4193 * test_brw on persistent OST devices
4195 * lprocfs (disabled)
4197 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4199 * documentation updates
4200 * bug fixes and cleanups
4201 * configuration tools
4203 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4206 * many small fixes to 0_5_4
4207 * io/network handling
4208 * thinkos in MDS operations
4210 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4213 * crucial basic fixes to 0.5.3
4214 * IOR, Iozone work over Elan
4217 2002-08-07 Phil Schwan <phil@clusterfs.com>
4218 * version 0_5_3, our first alpha
4219 * we use the new Portals iovs
4220 * documentation updates
4221 * bug fixes and cleanups
4222 * small changes in the DLM wire protocol
4224 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4225 * version 0_5_1 with some initial stability,
4226 * locking on MD and file I/O.
4227 * documentation updates
4228 * several bug fixes since 0.5.0
4229 * small changes in wire protocol
4231 2002-07-18 Phil Schwan <phil@clusterfs.com>
4233 * delivered as Lustre Light Alpha
4234 * fixed a crash after handling invalid MDS requests
4235 * fixed directory pages for architectures with non-4k pages sizes
4237 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4238 * release version v0_4_4
4239 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4241 * Fixes a number of interruption problems with OST operations.
4242 * Update documentation for portals header changes
4243 * Move all wire protocol structs/defines to lustre_idl.h
4244 * Fixes symlink length bug.
4245 * Add tcpdump to repository.
4247 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4248 * release version v0_4_3
4249 * Fixes statfs for inodes on extN.
4250 * Fixes bug in runtests which would delete /etc/hosts.
4251 * Use 64-bit object IDs wherever possible (not into VFS though)
4252 Remove ost_get_info, which is unused by lustre, and out of date.
4254 2002-07-03 Peter Braam <braam@clusterfs.com>
4255 * release version v0_4_2 Fixes a lookup error (type not passed)
4256 * move forward to head of Portals
4257 * move forward to latest Lustre kernel
4259 2002-06-25 Peter Braam <braam@clusterfs.com>
4260 * release version v0_4_1. Hopefully stable on single node use.