4 2.6.9-55.0.9.EL (RHEL 4),
5 2.6.16.53-0.8 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: Remove 2.4 kernel support from lustre 1.8.0
18 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
19 quota_interface.c in quota subsystem.
20 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
22 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
23 lvfs_linux.c upcall_cache.c in lvfs subsystem.
24 Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
25 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
27 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
28 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
30 Remove 2.4 kernel check from lproc_osc.c, osc_request.c in osc
32 Remove 2.4 kernel check from filter.c, lproc_obdfilter.c in
34 Remove 2.4 kernel code from lustre/obdecho/echo_client.c.
35 Remove 2.4 kernel check from lustre/mgs/lproc_mgs.c,
38 Severity : enhancement
40 Description: Build SLES10 patchless client fails
41 Details : The configure was broken by run ./configure with
42 --with-linux-obj=.... argument for patchless client. When the
43 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
44 can't search header adequately. Use absolute path such as
45 -I($LINUX)/include instead.
49 Description: interrupt oig_wait produce painc on resend.
50 Details : brw_redo_request can be used for resend requests from ptlrpcd and
51 private set, and this produce situation when rq_ptlrpcd_data not
52 copyed to new allocated request and triggered LBUG on assert
53 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
54 ptlrpcd set if request is changed and can be safety changed to use
59 Description: LASSERT_{REQ,REP}SWAB macros are buggy
60 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
61 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
62 these macros and replace them with their difinitions instead.
66 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
67 Details : Update which_patch & target file for SLES10 latest kernel.
70 Description: Remove client patches from SLES 10 kernel.
71 Details : This causes SLES 10 clients to behave as patchless clients
72 even on a Lustre-patched (server) kernel.
74 Severity : enhancement
76 Description: self-adjustable client's lru lists
77 Details : use adaptive algorithm for managing client cached locks lru
78 lists according to current server load, other client's work
79 pattern, memory activities, etc. Both, server and client
80 side namespaces provide number of proc tunables for controlling
83 Severity : enhancement
85 Description: light-weight GSS support
86 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
87 and reduce performance overhead.
89 Severity : enhancement
91 Description: Linux keyring support
92 Details : Support using service of Linux keyring for Lustre GSS internal
93 context refresh/cache mechanism.
97 Description: Fix errors in lfs documentation
98 Details : Fixes man pages
102 Description: don't use GFP_* in generic Lustre code.
103 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
107 Description: obdclass is limited by single OBD_ALLOC(idarray)
108 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
112 Description: warnings with build patchless client with vanila 2.6.19 and up
113 Details : change old ctl_table style and replace ctl_table/ctl_table_header
114 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
118 Frequency : Only for SLES
119 Description: sanity_quota fail test_1
120 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
121 file, which makes TSTID[2] inunique.
125 Frequency : Only for SLES9
126 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
127 Details : Update target/ChangeLog/which_patch .
131 Frequency : Only for test_10 in sanity-quota.sh
132 Description: a bug in quota test code
133 Details : checking a test flag in a wrong place
137 Frequency : Only for test_13 in sanity-quota.sh
138 Description: a bug in quota test script
139 Details : shouldn't check overfull allocated quota
143 Description: jbd statistics
144 Details : Port older jbd statistics patch for sles10
146 Severity : enhancement
148 Description: Add printing inode info into message about error in writepage.
153 Description: reply_lock_interpret crash due to race with it and lock cancel.
154 Details : Do not replay locks that are being cancelled. Do not reference
155 locks by their address during replay, just by their handle.
159 Frequency : When flocks are used.
160 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
161 Details : Flock locks might destroy just granted lock if it could be merged
162 with another existing flock, this is done in completion handler,
163 so teach ldlm_cli_enquque_fini that this is a valid case for
169 Description: Oops in read and write path when failing to allocate lock.
170 Details : Check if lock allocation failed and return error back.
174 Description: lstripe command fails for valid OST index
175 Details : The stripe offset is compared to lov->desc.ld_tgt_count
176 instead of lov->desc.ld_active_tgt_count.
180 Description: sanity.sh failed test 103
181 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
186 Description: df doesn't work properly if diskfs blocksize != 4K
187 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
191 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
192 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
196 Description: obd_health_check_timeout too short
197 Details : set obd_health_check_timeout as 1.5x of obd_timeout
201 Description: llapi_file_create() does not allow some changes
202 Details : add llapi_file_open() that allows specifying the mode and
203 open flags, and also returns an open file handle.
207 Description: merge and cleanup kernel patches.
208 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
212 Description: Add journal checksum support.(Kernel part)
213 Details : The journal checksum feature adds two new flags i.e
214 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
215 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
216 indicates that the commit block contains the checksum for
217 the blocks described by the descriptor blocks. Now commit
218 record can be sent to disk without waiting for descriptor
219 blocks to be written to disk. This behavior is controlled
220 using JBD2_FEATURE_ASYNC_COMMIT flag.
224 Description: OSS needs mutliple precreate threads
225 Details : Add ability to start more than one create thread per OSS.
229 Description: Sanity.sh test_65k failed: setstripe should have succeeded
230 Details : execute lfs setstripe on client
234 Description: mds_obd_create error creating tmp object
235 Details : When the user sets quota on root, llog will be affected and can't
236 create files and write files.
239 Frequency : Always on ia64 patchless client, and possibly others.
241 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
242 Details : This allows the patchless client to be loaded on architectures
247 Description: RedHat Update kernel for RHEL5
248 Details : Add the kernel config file for RHEL5.
252 Description: RedHat Update kernel for RHEL5
253 Details : Modify the kernel config file more closer RHEL5.
257 Description: Build failure against Centos5 (RHEL5)
258 Details : Define PAGE_SIZE when it isn't present.
262 Description: test_7 on sanity-quota.sh can't be used on separate machines
263 Details : add facet to handle it
267 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
268 Details : with NFS, the anon dentry's parent was set to itself in
269 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
270 corresponding dentry other than use rec->ur_name.
272 Severity : enhancement
274 Description: lfs setstripe enhancement
275 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
279 Description: testing performance impact of enabling checksumming
280 Details : enable checksum by default, allow --disable-checksum
281 configure option and "-o nochecksum" mount option
285 Description: lustre support for RHEL5
286 Details : Add support for RHEL5.
290 Description: Client eviction due to failover config
291 Details : after a connection loss, the lustre client should attempt to
292 reconnect to the last active server first before trying the
293 other potential connections.
297 Description: when mds and osts use different quota unit(32bit and 64bit),
298 quota will be released repeatly.
299 Details : void sending multiple quota reqs to mds, which will keep the
300 status between the reqs.
304 Description: osts not allocated evenly to files
305 Details : change the condition to increase offset_idx
308 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
310 Description: Data corruption for OSTs that are formatted larger than 2TB
312 Details : When generating the bio request for lustre file writes the
313 sector number would overflow a temporary variable before being
314 used for the IO. The data reads correctly from Lustre (which
315 will overflow in a similar manner) but other file data or
316 filesystem metadata may be corrupted in some cases.
320 Description: Tune the kernel for good SCSI performance.
321 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
322 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
327 Description: rewrite ext2-derived code in obdclass/uuid.c
328 Details : rewrite inherited code (uuid parsing code from ext2 utils)
329 from scratch preserving functionality.
333 Description: "lfs find -obd UUID" prints directories
334 Details : "lfs find -obd UUID" will return all directory names instead
335 of just file names. It is incorrect because the directories
336 do not reside on the OSTs.
340 Description: use i_size_read and i_size_write in 2.6 port
341 Details : replace inode->i_size access with i_size_read/write()
344 Frequency : when removing large files
346 Description: scheduling issue during removal of large Lustre files
347 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
348 It causes scheduling issues when removing large files (17TB in the
352 Frequency : only with liblustre clients on XT3
354 Description: evictions taking too long
355 Details : allow llrd to evict clients directly on OSTs
358 Frequency : only on ppc
360 Description: /proc/fs/lustre/devices broken on ppc
361 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
362 We should make sure the type of 'index' is loff_t and then cast
363 explicitly as needed below. Do not assign an explicitly cast
367 Frequency : only for rhel5
369 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
370 Details : Modify the target file & which_kernel.
372 Severity : enhancement
374 Description: omit set fsid for export NFS
375 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
376 as FS_REQUIRES_DEV for avoid problems with generate fsid.
380 Frequency : Always, for kernels after 2.6.16
381 Description: Fix warning idr_remove called for id=.. which is not
383 Details : Last kernels save old s_dev before kill super and not allow
384 to restore from callback - restore it before call kill_anon_super.
388 Description: buffer overruns could theoretically occur
389 Details : llapi_semantic_traverse() modifies the "path" argument by
390 appending values to the end of the origin string, and a
391 overrun may occur. Adding buffer overrun check in liblustreapi.
395 Description: Fix error on 'ls .' at the top of the Lustre mount.
396 Details : Don't revalidate dentry if it is a root dentry.
400 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
401 Details : Modify vm-tunables-rhel4.patch.
405 Description: Kernel config for 2.6.18-vanilla.
406 Details : Modify targets/2.6-vanilla.target.in.
407 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
408 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
409 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
410 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
414 Description: improve handling recoverable errors
415 Details : if request processig with error which can be recoverable on server
416 request should be resend, otherwise page released from cache and
421 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
422 Details : Modify target file & which_patch.
423 A flaw was found in the IA32 system call emulation provided
424 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
425 value could be stored in the %RAX register, which could trigger an
426 out-of-bounds system call table access. An untrusted local user
427 could exploit this flaw to run code in the kernel
428 (ie a root privilege escalation). (CVE-2007-4573).
432 Description: O_DIRECT bypasses client statistics.
433 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
434 read_ahead_stats were updated. copied the stats section from
435 osc_send_oap_rpc() into async_internal().
439 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
443 Description: change order of libsysio includes
444 Details : '#include sysio.h' should always come before '#include xtio.h'
448 Description: handle "serious error: objid * already exists" more gracefully
449 Details : If LAST_ID value on disk is smaller than the objects existing in
450 the O/0/d* directories, it indicates disk corruption and causes an
451 LBUG(). If the object is 0-length, then we should use the existing
452 object. This will help to avoid a full fsck in most cases.
454 Severity : enhancement
456 Description: adapt the lustre_config script to support the upgrade case
457 Details : Add "-u" option for lustre_config script to support upgrading 1.4
458 server targets to 1.6 in parallel.
462 Description: To avoid grant space > avaible space when the disk is almost
463 full. Without this patch you might see the error "grant XXXX >
464 available" or some LBUG about grant, when the disk is almost
466 Details : In filter_check_grant, for non_grant cache write, we should
467 check the left space by if (*left > ungranted + bytes), instead
468 of (*left > ungranted), because only we are sure the left space
469 is enough for another "bytes", then the ungrant space should be
470 increase. In client, we should update cl_avail_grant only there
471 is OBD_MD_FLGRANT in the reply.
475 Description: Update RHEL 4 kernel to fix local root privilege escalation.
476 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
477 described in CVE-2007-4573. This problem could allow untrusted
478 local users to gain root access.
482 Frequency : when using O_DIRECT and quotas
484 Description: Incorrect file ownership on O_DIRECT output files
485 Details : block usage reported by 'lfs quota' does not take into account
486 files that have been written with O_DIRECT.
488 --------------------------------------------------------------------------------
490 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
492 * Support for kernels:
493 2.6.5-7.283 (SLES 9),
494 2.6.9-55.EL (RHEL 4),
495 2.6.16.46-0.14 (SLES 10),
496 2.6.18.8 vanilla (kernel.org)
497 * Client support for unpatched kernels:
498 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
499 2.6.16 - 2.6.22 vanilla (kernel.org)
500 * Due to recently discovered recovery problems, we do not recommend
501 using patchless RHEL 4 clients with this or any earlier release.
502 * Recommended e2fsprogs version: 1.39.cfs8
503 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
504 * Starting with this release, the ldiskfs backing filesystem required
505 by Lustre is now in its own package, lustre-ldiskfs. This package
506 should be installed. It is versioned separately from Lustre and
507 may be released separately in future.
511 Description: block reactivating mgc import until all deactivates complete
512 Details : Fix race when failing back MDT/MGS to itself (testing)
515 Frequency : at statup only
517 Description: mds_lov_synchronize race leads to various problems
518 Details : simultaneous MDT->OST connections at startup can cause the
519 sync to abort, leaving the OSC in a bad state.
521 Severity : enhancement
523 Description: add optional extra BUILD_VERSION info
524 Details : add a new environment variable (namely LUSTRE_VERS) which allows
525 to override the lustre version.
528 Frequency : 2.6.18 servers only
530 Description: ll_kern_mount() doesn't release the module reference
531 Details : The ldiskfs module reference count never drops down to 0
532 because ll_kern_mount() doesn't release the module reference.
537 Description: server LBUG when using old ost_num_threads parameter
538 Details : Accept the old ost_num_threads parameter but warn that it
539 is deprecated, and fix an off-by-one error that caused an LBUG.
544 Description: Transient SCSI error results in persistent IO issue
545 Details : iobuf->dr_error is not reinitialized to 0 between two
549 Frequency : sometimes when underlying device returns I/O errors
551 Description: OSTs not going read-only during write failures
552 Details : OSTs are not remounted read-only when the journal commit threads
553 get I/O errors because fsfilt_ext3 calls journal_start/stop()
554 instead of the ext3 wrappers.
558 Description: poor connect scaling with increasing client count
559 Details : Don't run filter_grant_sanity_check for more than 100 exports
560 to improve scaling for large numbers of clients.
563 Frequency : SLES10 only
565 Description: sanity-quota.sh quotacheck failed: rc = -22
566 Details : Quotas cannot be enabled on SLES10.
569 Frequency : liblustre clients only
571 Description: getdirentries does not give error when run on compute nodes
572 Details : getdirentries does not fail when the size specified as an argument
573 is too small to contain at least one entry
575 Severity : enhancement
577 Description: Add LNET router traceability for debug purposes
578 Details : If a checksum failure occurs with a router as part of the
579 IO path, the NID of the last router that forwarded the bulk data
580 is printed so it can be identified.
585 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
586 Details : Due to a race condition, liblustre clients were occasionally
589 Severity : enhancement
591 Description: lfs setstripe use optional parameters instead of postional
594 Severity : enhancement
596 Description: Nanosecond timestamp support for ldiskfs
597 Details : The on-disk ldiskfs filesystem has added support for nanosecond
598 resolution timestamps. There is not yet support for this at
599 the Lustre filesystem level.
602 Frequency : during server recovery
604 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
605 Details : request with rq_no_resend flag not awake l_wait_event if they get a
609 Frequency : nfs export on patchless client
611 Description: connectathon hang when test nfs export over patchless client
612 Details : Disconnected dentry cannot be found with lookup, so we do not need
613 to unhash it or make it invalid
616 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
617 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
618 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
623 Description: open req refcounting wrong on reconnect
624 Details : If reconnect happened between getting open reply from server and
625 call to mdc_set_replay_data in ll_file_open, we will schedule
626 replay for unreferenced request that we are about to free.
627 Subsequent close will crash in variety of ways.
628 Check that request is still eligible for replay in
629 mdc_set_replay_data().
634 Description: disable writes to filesystem when reading health_check file
635 Details : the default for reading the health_check proc file has changed
636 to NOT do a journal transaction and write to disk, because this
637 can cause reads of the /proc file to hang and block HA state
638 checking on a healthy but otherwise heavily loaded system. It
639 is possible to return to the previous behaviour during configure
640 with --enable-health-write.
642 Severity : enhancement
644 Description: 64-bit inode version
645 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
646 to the inode. This will be required for version-based recovery.
651 Description: MDS fails to start if a duplicate client export is detected
652 Details : in some rare cases it was possible for a client to connect to
653 an MDS multiple times. Upon recovery the MDS would detect this
654 and fail during startup. Handle this more gracefully.
656 Severity : enhancement
658 Description: Add -o localflock option to simulate old noflock
660 Details : This will achieve local-only flock/fcntl locks
666 Description: log_commit_thread vs filter_destroy race leads to crash
667 Details : Take import reference before releasing llog record semaphore
672 Description: Wrong request locking in request set processing
673 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
674 imp_delayed_list, in this place should be used imp_lock.
677 Frequency : when reconnection
679 Description: Grant Leak when osc reconnect to OST
680 Details : When osc reconnect ost, OST(filter) should check whether it
681 should grant more space to client by comparing fed_grant and
682 cl_avail_grant, and return the granted space to client instead
683 of "new granted" space, because client will call osc_init_grant
684 to update the client grant space info.
687 Frequency : when client reconnect to OST
689 Description: Grant Leak when osc do resend and replay bulk write
690 Details : When osc reconnect to OST, OST(filter)should clear grant info of
691 bulk write request, because the grant info while be sync between
692 OSC and OST when reconnect, and we should ignore the grant info
693 these of resend/replay write req.
698 Description: Grant space more than avaiable left space sometimes.
699 Details : When then OST is about to be full, if two bulk writing from
700 different clients came to OST. Accord the avaliable space of the
701 OST, the first req should be permitted, and the second one
702 should be denied by ENOSPC. But if the seconde arrived before
703 the first one is commited. The OST might wrongly permit second
704 writing, which will cause grant space > avaiable space.
707 Frequency : when client is evicted
709 Description: Grant might be wrongly erased when osc is evicted by OST
710 Details : when the import is evicted by server, it will fork another
711 thread ptlrpc_invalidate_import_thread to invalidate the
712 import, where the grant will be set to 0. While the original
713 thread will update the grant it got when connecting. So if
714 the former happened latter, the grant will be wrongly errased
715 because of this race.
720 Description: Checking Stale with correct fid
721 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
722 to check whether it is stale, because sometimes, we want the
723 enqueue happened anyway, and op_data.fid2 will not be initialized.
725 Severity : enhancement
727 Description: update patchless client
728 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
731 Frequency : only with 2.4 kernel
733 Description: random memory corruption
734 Details : size of struct ll_inode_info is to big for union inode.u and this
735 can be cause of random memory corruption.
740 Description: Memory leak in recovery
741 Details : Lov_mds_md was not free in an error handler in mds_create_object.
742 It should also check obd_fail before fsfilt_start, otherwise if
743 fsfilt_start return -EROFS,(failover mds during mds recovery).
744 then the req will return with repmsg->transno = 0 and rc = EROFS.
745 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
746 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
747 be freed no matter whether fsfilt_commit success or not.
750 Frequency : only with huge count clients
752 Description: Prevents from taking the superblock lock in llap_from_page for
754 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
755 for llap_from_page call in ll_removepage prevents from taking the
756 superblock lock for a soon died page.
761 Description: Not check open intent error before release open handle
762 Details : in some rare cases, the open intent error is not checked before
763 release open handle, which may cause
764 ASSERTION(open_req->rq_transno != 0), because it tries to release
765 the failed open handle.
770 Description: Set cat log bitmap only after create log success.
771 Details : in some rare cases, the cat log bitmap is set too early. and it
772 should be set only after create log success.
776 Description: Accessing a block bevice can re-enable I/O when Lustre is
777 tearing down a device.
778 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
783 Description: service threads may hog cpus when there are a lot of requests
785 Details : Insert cond_resched to give other threads a chance to use some of
791 Description: the cat log was not initialized in recovery
792 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
793 unlink log on mds will not be initialized until mds post
794 recovery. And also in mds post recovery, the unlink log will
795 initialization will be done asynchronausly, so there will be race
796 between add unlink log and unlink log initialization.
800 Description: brw_stats were being printed incorrectly
801 Details : brw_stats were being printed as log2 but all of them were not
802 recorded as log2. Also remove some code duplication arising from
803 filter_tally_{read,write}.
807 Frequency : rare, only in recovery.
808 Description: ASSERTION(req->rq_type != LI_POISON) failed
809 Details : imp_lock should be held while iterating over imp_sending_list for
810 prevent destroy request after get timeout in ptlrpc_queue_wait.
814 Description: replay-single.sh test 52 fails
815 Details : A lock's skiplist need to be cleanup when it being unlinked
816 from its resource list.
820 Description: Short directio read returns full requested size rather than
822 Details : Direct I/O operations should return actual amount of bytes
823 transferred rather than requested size.
825 Severity : enhancement
827 Description: metadata RPC reduction (e.g. for rm performance)
828 Details : decrease the amount of synchronous RPC between clients and servers
829 by canceling conflicing lock before the operation on the client side
830 and packing thier handles into the main operation RPC to server.
832 Severity : enhancement
834 Description: Async OSC create to avoid the blocking unnecessarily.
835 Details : If a OST has no remain object, system will block on the creating
836 when need to create a new object on this OST. Now, ways use
837 pre-created objects when available, instead of blocking on an
838 empty osc while others are not empty. If we must block, we block
839 for the shortest possible period of time.
841 --------------------------------------------------------------------------------
843 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
848 Frequency : on some architectures
850 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
851 Details : Uninitialized flags sometimes cause configuration commands to
855 Frequency : patchless clients only
857 Description: missing __iget() symbol export
858 Details : The __iget() symbol export is missing. To avoid the need for
859 this on patchless clients the deathrow inode reaper is turned
860 off, and we depend on the VM to clean up old inodes. This
861 dependency was during via the fix for bug 12181.
863 --------------------------------------------------------------------------------
865 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
867 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
868 INTEROPERATE with older versions automatically. In many cases a
869 special upgrade step is needed. Please read the
870 user documentation before upgrading any part of a live system.
871 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
872 version will not interoperate with 1.6 betas before beta5 (1.5.95).
873 * WARNING: Lustre configuration and startup changes are required with
874 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
880 Severity : enhancement
882 Description: MountConf
883 Details : Lustre configuration is now managed via mkfs and mount
884 commands instead of lmc and lconf. New obd types (MGS, MGC)
885 are added for dynamic configuration management. See
886 https://mail.clusterfs.com/wikis/lustre/MountConf for
889 Severity : enhancement
891 Description: dynamic OST addition
892 Details : OSTs can now be added to a live filesystem
894 Severity : enhancement
896 Description: startup order invariance
897 Details : MDTs and OSTs can be started in any order. Clients only
898 require the MDT to complete startup.
900 Severity : enhancement
902 Description: parallel, asynchronous orphan cleanup
903 Details : orphan cleanup is now performed in separate threads for each
904 OST, allowing parallel non-blocking operation.
906 Severity : enhancement
908 Description: optimized stripe assignment
909 Details : stripe assignments are now made based on ost space available,
910 ost previous usage, and OSS previous usage, in order to try
911 to optimize storage space and networking resources.
913 Severity : enhancement
915 Description: Permanently set tunables
916 Details : All writable /proc/fs/lustre tunables can now be permanently
917 set on a per-server basis, at mkfs time or on a live
920 Severity : enhancement
922 Description: Lustre message v2
923 Details : Add lustre message format v2.
925 Severity : enhancement
927 Description: client OST exclusion list
928 Details : Clients can be started with a list of OSTs that should be
929 declared "inactive" for known non-responsive OSTs.
933 Description: SPEC SFS validation failure on NFS v2 over lustre.
934 Details : Changes the blocksize for regular files to be 2x RPC size,
935 and not depend on stripe size.
937 Severity : enhancement
939 Description: Multiple MD RPCs in flight.
940 Details : Further unserialise some read-only MDS RPCs - learn about intents.
941 To avoid overly-overloading MDS, introduce a limit on number of
942 MDS RPCs in flight for a single client and add /proc controls
943 to adjust this limit.
945 Severity : enhancement
947 Description: client read/write statistics
948 Details : Add client read/write call usage stats for performance
949 analysis of user processes.
950 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
951 file access. extents_stats shows chunk size distribution.
952 extents_stats_per_process show chunk size distribution per
955 Severity : enhancement
957 Description: mds statistics
958 Details : Add detailed mds operations statistics in
959 /proc/fs/lustre/mds/*/stats.
963 Description: Failure of copying files with lustre special EAs.
964 Details : Client side always return success for setxattr call for lustre
965 special xattr (currently only "trusted.lov").
970 Description: Refcount LNET uuids
971 Details : The global LNET uuid list grew linearly with every startup;
972 refcount repeated list entries instead of always adding to
975 Severity : enhancement
977 Description: Dynamic service threads
978 Details : Within a small range, start extra service threads
979 automatically when the request queue builds up.
982 Frequency : mixed-endian client/server environments
984 Description: mixed-endian crashes
985 Details : The new msg_v2 system had some failures in mixed-endian
988 Severity : enhancement
990 Description: Easy OST removal
991 Details : OSTs can be permanently deactivated with e.g. 'lctl
992 conf_param lustre-OST0001.osc.active=0'
994 Severity : enhancement
996 Description: MGS proc entries
997 Details : Added basic proc entries for the MGS showing what filesystems
1000 Severity : enhancement
1002 Description: provide MGS failover
1003 Details : Added config lock reacquisition after MGS server failover.
1005 Severity : enhancement
1007 Description: add Linux 2.4 support
1008 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1012 Description: a large application tries to do I/O to the same resource and dies
1013 in the middle of it.
1014 Details : Check the req->rq_arrival time after the call to
1015 ost_brw_lock_get(), but before we do anything about
1016 processing it & sending the BULK transfer request. This
1017 should help move old stale pending locks off the queue as
1018 quickly as obd_timeout.
1021 Frequency : when an incorrect nid is specified during startup
1023 Description: ptlrpc connect to non-existant node causes kernel crash
1024 Details : LNET can't be re-entered from an event callback, which
1025 happened when we expire a message after the export has been
1026 cleaned up. Instead, hand the zombie cleanup off to another
1029 Severity : enhancement
1031 Description: plain/inodebits lock performance improvement
1032 Details : Grouping plain/inodebits in granted list by their request modes
1033 and bits policy, thus improving the performance of search through
1037 Frequency : only if OST filesystem is corrupted
1039 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1040 Details : for a short time RPCs with bulk IO are in the replay list,
1041 but replay of bulk IOs is unimplemented. If the OST filesystem
1042 is corrupted due to disk cache incoherency and then replay is
1043 started it is possible to trip an assertion. Avoid putting
1044 committed RPCs into the replay list at all to avoid this issue.
1047 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1049 Description: System hang on startup
1050 Details : This bug allowed the liblustre (e.g. catamount) client to
1051 return to the app before handling all startup RPCs. This
1052 could leave the node unresponsive to lustre network traffic
1053 and manifested as a server ptllnd timeout.
1055 Severity : enhancement
1057 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1058 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1059 boolean to control whether to print peer debug info when a
1060 client's RPC times out.
1063 Frequency : only for kernels with patches from Lustre below 1.4.3
1065 Description: Remove old rdonly API
1066 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1069 Frequency : only for devices with external journals
1071 Description: Set external device read-only also
1072 Details : During a commanded failover stop, we set the disk device
1073 read-only while the server shuts down. We now also set any
1074 external journal device read-only at the same time.
1077 Frequency : when upgrading from 1.4 while trying to change parameters
1079 Description: The wrong (new) MDC name was used when setting parameters for
1080 upgraded MDT's. Also allows changing of OSC (and MDC)
1081 parameters if --writeconf is specified at tunefs upgrade time.
1084 Frequency : when setting specific ost indicies
1086 Description: QOS code breaks on skipped indicies
1087 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1088 created with --index need not be sequential.
1090 Severity : enhancement
1092 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1093 Details : The uninit_groups feature works in conjunction with the kernel
1094 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1095 up the pass1 processing of e2fsck. This is a read-only feature
1096 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1097 mount filesystems that have had this feature enabled.
1099 Severity : enhancement
1101 Description: Improve multi-block allocation algorithm to avoid fragmentation
1102 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1103 allocation locality and avoid filesystem fragmentation.
1105 ------------------------------------------------------------------------------
1107 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1109 * Support for kernels:
1110 2.4.21-47.0.1.EL (RHEL 3)
1111 2.6.5-7.283 (SLES 9)
1112 2.6.9-42.0.10.EL (RHEL 4)
1113 2.6.12.6 vanilla (kernel.org)
1114 2.6.16.27-0.9 (SLES 10)
1115 * Recommended e2fsprogs version: 1.39.cfs5
1117 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1121 Frequency : occasional, depends on client load and configuration
1122 Bugzilla : 12181, 12203
1123 Description: data loss for recently-modified files
1125 Details : In some cases it is possible that recently written or created
1126 files may not be written to disk in a timely manner (this should
1127 normally be within 30s unless client IO load is very high).
1128 The problem appears as zero-length files or files that are a
1129 multiple of 1MB in size after a client crash or client eviction
1130 that are missing data at the end of the file.
1132 This problem is more likely to be hit on clients where files are
1133 repeatedly created and unlinked in the same directory, clients
1134 have a large amount of RAM, have many CPUs, the filesystem has
1135 many OSTs, the clients are rebooted frequently, and/or the files
1136 are not accessed by other nodes after being written.
1138 The presence of the problem can be detected by looking at
1139 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1140 smaller than the second (nr_unused) then dirty files will not
1141 be flushed automatically to disk. "sync; sleep 10" should be
1142 run several times on the node before unmounting it to update
1143 Lustre (this is also safe to run on nodes without this problem).
1145 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1146 that can cause this same problem, so customers using that kernel
1147 also need to update the kernel in addition to Lustre. In order
1148 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1150 It is normal that files written just before a client crash (less
1151 than 30s) may not yet have been flushed to disk, even for local
1155 Frequency : frequent on thin XT3 nodes
1157 Description: UUID collision on thin XT3 Linux nodes
1158 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1159 randomly, since we relied on an insufficiently-seeded PRNG.
1164 Description: OSS hangs after "All ost request buffers busy"
1165 Details : A deadlock between quota and journal operations caused OSS
1166 hangs after printing "All ost request buffers busy."
1169 Frequency : always on liblustre builds
1171 Description: Cleanup compiler warnings on liblustre
1174 Frequency : always on liblustre builds on XT3
1176 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1180 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1181 > 32000 subdirectories
1182 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1183 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1184 subdirectory count crosses 32000. This will aid e2fsck to
1185 correctly handle more than 32000 subdirectories.
1188 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1190 Description: System hang on startup
1191 Details : This bug allowed the liblustre (e.g. catamount) client to
1192 return to the app before handling all startup RPCs. This
1193 could leave the node unresponsive to lustre network traffic
1194 and manifested as a server ptllnd timeout.
1196 Severity : enhancement
1198 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1199 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1200 boolean to control whether to print peer debug info when a
1201 client's RPC times out.
1206 Description: make O_SYNC working on 2.6 kernels
1207 Details : 2.6 kernels use different method for mark pages for write,
1208 so need add a code to lustre for O_SYNC work.
1213 Description: Failure to close file and release space on NFS
1214 Details : Put inode details into lock acquired in ll_intent_file_open.
1215 Use mdc_intent_lock in ll_intent_open to properly
1216 detect all kind of errors unhandled by mdc_enqueue.
1221 Description: proc file read during shutdown sometimes raced obd removal,
1223 Details : Add lock to prevent obd access after proc file removal.
1226 Frequency : Only for files larger than 4GB on 32-bit clients.
1228 Description: improperly doing page alignment of locks
1229 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1230 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1235 Description: RPCs being resent when they shouldn't be
1236 Details : Some RPCs that should not be resent are being resent. This
1237 can cause inconsistencies in the RPC state machine. Do not
1238 resend such requests.
1241 Frequency : rare, only with NFS export
1243 Description: Crash on NFS re-export node
1244 Details : under very unusual load conditions an assertion is hit in
1245 ll_intent_file_open()
1248 Frequency : only if OST filesystem is corrupted
1250 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1251 Details : for a short time RPCs with bulk IO are in the replay list,
1252 but replay of bulk IOs is unimplemented. If the OST filesystem
1253 is corrupted due to disk cache incoherency and then replay is
1254 started it is possible to trip an assertion. Avoid putting
1255 committed RPCs into the replay list at all to avoid this issue.
1260 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1261 Details : Large single O_DIRECT read and write calls can fail to allocate
1262 a sufficiently large buffer to process the request. In case of
1263 allocation failure the allocation is retried with a smaller
1264 buffer and broken into smaller requests.
1266 Severity : enhancement
1268 Description: Add -o localflock option to simulate old noflock behaviour.
1269 Details : This will achieve local-only flock/fcntl locks coherentness.
1274 Description: versioning check is incomplete
1275 Details : Checking the version difference of client vs. server, report
1276 error if the gap is too big.
1281 Description: add support PG_writeback bit
1282 Details : add support for PG_writeback bit for Lustre, for more carefull
1283 work with page cache in 2.6 kernel. This also fix some deadlocks
1284 and remove hack for work O_SYNC with 2.6 kernel.
1286 Severity : enhancement
1288 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1289 Details : The uninit_groups feature works in conjunction with the kernel
1290 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1291 up the pass1 processing of e2fsck. This is a read-only feature
1292 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1293 mount filesystems that have had this feature enabled.
1295 Severity : enhancement
1297 Description: Improve multi-block allocation algorithm to avoid fragmentation
1298 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1299 allocation locality and avoid filesystem fragmentation.
1301 ------------------------------------------------------------------------------
1303 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1305 * Support for kernels:
1306 2.6.9-42.0.3.EL (RHEL 4)
1307 2.6.5-7.276 (SLES 9)
1308 2.4.21-47.0.1.EL (RHEL 3)
1309 2.6.12.6 vanilla (kernel.org)
1310 2.6.16.21-0.8 (SLES10)
1311 * Recommended e2fsprogs version: 1.39.cfs2-0
1313 * The backwards-compatible /proc/sys/portals symlink has been removed
1314 in this release. Before upgrading, please ensure that you change
1315 any configuration scripts or /etc/sysctl.conf files that access
1316 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1317 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1318 in advance of the upgrade on any system running Lustre 1.4.6 or
1319 newer, since /proc/sys/lnet was added in that version.
1320 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1324 Frequency : only when quota is used
1326 Description: avoid scanning export list for quota master
1327 Details : Change the algorithms to avoid scanning export list in order
1328 to improve the efficiency.
1331 Frequency : MDS failover only, very rarely
1333 Description: "went back in time" messages on mds failover
1334 Details : The greatest transno may be lost when the current operation
1335 finishes with an error (transno==0) and the client's last_rcvd
1336 record is over-written. Save the greatest transno in the
1337 mds_last_transno for this case.
1340 Frequency : always for specific kernels and striping counts
1342 Description: client may get "Matching packet too big" without ACL support
1343 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1344 when trying to access files in certain configurations. The
1345 clients should in fact be denied when mounting because they do
1346 not understand ACLs.
1349 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1351 Description: many clients connecting with IO in progress causes connect timeouts
1352 Details : Avoid synchronous journal commits to avoid delays caused by many
1353 clients connecting/disconnecting when bulk IO is in progress.
1354 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1355 OST_IO_PORTAL to avoid delays behind potentially many pending
1359 Frequency : occasionally with multiple writers to a single file
1361 Description: shared writes to file may result in wrong size reported by stat()
1362 Details : Allow growing of kms when extent lock is cancelled
1365 Frequency : always with random mmap IO to multi-striped file
1367 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1368 Details : Only if the hole is at the end of OST object so that kms is too
1369 small. Fix is to increase kms accordingly in ll_nopage.
1372 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1374 Description: writes to a missing object would leak memory on the OST
1375 Details : If there is an inconsistency between the MDS and OST filesystems,
1376 such that the MDS references an object that doesn't exist, writes
1377 to that object will leak memory due to incorrect cleanup in the
1378 error handling path, eventually running out of memory on the OST.
1383 Description: Creating too long symlink causes lustre errors
1384 Details : Check symlink and name lengths before sending requests to MDS.
1387 Frequency : only if flock is enabled (not on by default)
1389 Description: posix locks not released on fd closure on 2.6.9+
1390 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1391 kernels, this caused such locks not to be released on fd close and
1392 then assertions on fs unmount about still used locks.
1395 Frequency : MDS failover only, very rarely
1397 Description: clients may get ASSERTION(granted_lock != NULL)
1398 Details : When request was taking a long time, and a client was resending
1399 a getattr by name lock request. The were multiple lock requests
1400 with the same client lock handle and
1401 mds_getattr_name->fixup_handle_for_resent_request found one of the
1402 lock handles but later failed with ASSERTION(granted_lock != NULL).
1407 Description: handle->h_buffer_credits > 0, assertion failure
1408 Details : h_buffer_credits is zero after truncate, causing assertion
1409 failure. This patch extends the transaction or creates a new
1413 Frequency : NFS re-export or patchless client
1414 Bugzilla : 11179, 10796
1415 Description: Crash on NFS re-export node (__d_move)
1416 Details : We do not want to hash the dentry if we don't have a lock.
1417 But if this dentry is later used in d_move, we'd hit uninitialised
1418 list head d_hash, so we just do this to init d_hash field but
1419 leave dentry unhashed.
1422 Frequency : NFS re-export or patchless client
1424 Description: NFS exports has problem with symbolic link
1425 Details : lustre client didn't properly install dentry when re-exported
1426 to NFS or running patchless client.
1429 Frequency : NFS re-export or patchless client
1431 Description: Various nfs/patchless fixes.
1432 Details : fixes reuse disconected alias for lookup process - this fixes
1433 warning "find_exported_dentry: npd != pd",
1434 fix permission error with open files at nfs.
1435 fix apply umask when do revalidate.
1438 Frequency : occasional
1440 Description: Crash on NFS re-export node
1441 Details : calling clear_page() on the wrong pointer triggered oops in
1442 generic_mapping_read().
1445 Frequency : rarely, using O_DIRECT IO
1447 Description: unaligned directio crashes client with LASSERT
1448 Details : check for unaligned buffers before trying any requests.
1451 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1453 Description: stale data returned from RAID cache
1454 Details : If only a small amount of IO is done to the RAID device before
1455 reading it again it is possible to get stale data from the RAID
1456 cache instead of reading it from disk.
1459 Frequency : always for sles10 kernel
1461 Description: sles10 support
1462 Details : ll_follow_link: compile fixes and using of nd_set_link
1463 under newer kernels.
1466 Frequency : depends on arch, kernel and compiler version, always on sles10
1469 Description: recursive or deep enough symlinks cause stack overflow
1470 Details : getting rid of large stack-allocated variable in
1474 Frequency : depends on hardware
1476 Description: lustre write performance loss in the SLES10 kernel
1477 Details : the performance loss is caused by using of write barriers in the
1478 ext3 code. The SLES10 kernel turns barrier support on by
1479 default. The fix is to undo that change for ldiskfs.
1481 ------------------------------------------------------------------------------
1483 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1485 * Support for kernels:
1486 2.6.9-42.0.3EL (RHEL 4)
1487 2.6.5-7.276 (SLES 9)
1488 2.4.21-47.0.1.EL (RHEL 3)
1489 2.6.12.6 vanilla (kernel.org)
1493 Frequency : quota enabled and large files being deleted
1495 Description: releasing more than 4GB of quota at once hangs OST
1496 Details : If a user deletes more than 4GB of files on a single OST it
1497 will cause the OST to spin in an infinite loop. Release
1498 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1503 Description: statfs data retrieved from /proc may be stale or zero
1504 Details : When reading per-device statfs data from /proc, in the
1505 {kbytes,files}_{total,free,avail} files, it may appear
1506 as zero or be out of date.
1509 Frequency : systems with MD RAID1 external journal devices
1511 Description: lconf's call to blkid is confused by RAID1 journal devices
1512 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1513 of returning all block devices that match the journal UUID.
1516 Frequency : always, for aggregate stripe size over 4GB
1518 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1519 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1520 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1523 Frequency : always if "lfs find" used on a local file/directory
1525 Description: "lfs find" segfaults if used on a local file/directory
1526 Details : The case where a directory component was not specified wasn't
1527 handled correctly. Handle this properly.
1530 Frequency : always on ppc64
1532 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1533 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1534 perform endianness conversion.
1537 Frequency : rarely (truncate to non-zero file size after write under load)
1538 Bugzilla : 10730, 10687
1539 Description: Files padded with zeros to next 4K multiple
1540 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1541 it is possible that files that are truncated to a non-zero size
1542 immediately after being written are filled with zero bytes beyond
1543 the truncated size. No file data is lost.
1545 Severity : enhancement
1547 Description: Allow recovery/failover for liblustre clients.
1548 Details : liblustre clients were unaware of failover configurations until
1551 Severity : enhancement
1553 Description: user file locks should fail when not mounting with flock option
1554 Details : Set up an error-returning stub in ll_file_operations.lock field
1555 to prevent incorrect behaviour when client is mounted without
1556 flock option. Also, set up properly f_op->flock field for
1560 Frequency : always on ia64
1562 Description: "lfs df" loops on printing out MDS statfs information
1563 Details : The obd_ioctl_data was not initialized and in some systems
1564 this caused a failure during the ioctl that did not return
1565 an error. Initialize the struct and return an error on failure.
1568 Frequency : SLES 9 only
1570 Description: Error of copying files with lustre special EAs as root
1571 Details : Client side always return success for setxattr call for lustre
1572 special xattr (currently only "trusted.lov").
1575 Frequency : rarely on clusters with both ia64+i386 clients
1577 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1578 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1579 clients are doing concurrent writes to the same file it is
1580 possible that the ia64 clients may LASSERT because the OST
1581 extent locks are not PAGE_SIZE aligned. Ensure that grown
1582 locks are always aligned on the request boundary.
1585 Frequency : specific use, occasional
1587 Description: Overwriting in use executable truncates on-disk binary image
1588 Details : If one node attempts to overwrite an executable in use by
1589 another node, we now correctly return ETXTBSY instead of
1590 truncating the file.
1592 Severity : enhancement
1594 Description: Async OSC create to avoid the blocking unnecessarily.
1595 Details : If a OST has no remain object, system will block on the creating
1596 when need to create a new object on this OST. Now, ways use
1597 pre-created objects when available, instead of blocking on an
1598 empty osc while others are not empty. If we must block, we block
1599 for the shortest possible period of time.
1604 Description: chmod on Lustre root is propagated to other clients
1605 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1606 invalid by the follow_mount time.
1609 Frequency : liblustre clients only
1611 Description: Race in 'instant cancel' lock handling could lead to such locks
1612 never to be granted in case of SMP MDS
1613 Details : Do not destroy not yet granted but cbpending locks in
1617 Frequency : replay/resend of open
1619 Description: non null lock assetion failure in mds_intent_policy
1620 Details : Trying to replay/resend lockless open requests resulted in
1621 mds_open() returning 0 with no lock. Now it sets a flag if
1622 a lock is going to be returned.
1624 Severity : enhancement
1626 Description: Checksum enhancements
1627 Details : New checksum enhancements allow for resending RPCs that failed
1630 Severity : enhancement
1632 Description: Tunables on number of dirty pages in cacche
1633 Details : Allow to set limit on number of dirty pages cached.
1638 Description: client crash on unmount - lock still has references
1639 Details : In some error handling cases it was possible to leak a lock
1640 reference on a client while accessing a file. This was not
1641 harmful to the client during operation, but would cause the
1642 client to crash when the filesystem is unmounted.
1645 Frequency : specific case, rare
1647 Description: ETXTBSY on mds though file not in use by client
1648 Details : ETXTBSY is no longer incorrectly returned when attempting to
1649 chmod or chown a directory that the user previously tried to
1650 execute or a currently-executing binary.
1653 Frequency : extremely rare except on liblustre-based clients
1655 Description: Lustre space not freed when files are deleted
1656 Details : Clean up open-unlinked files after client eviction. Previously
1657 the unlink was skipped and the files remained as orphans.
1662 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1663 Details : In some cases it was possible to send a blocking callback to
1664 a client doing a glimpse, even though that client didn't get
1665 a lock granted. When the glimpse lock is cancelled on the OST
1666 the freed lock is left on the waiting list and corrupted the list.
1669 Frequency : all core dumps
1671 Description: Broke core dumps to lustre
1672 Details : Negative dentry may be unhashed if parent does not have UPDATE
1673 lock, but some callers, e.g. do_coredump, expect dentry to be
1674 hashed after successful create, hash it in ll_create_it.
1676 ------------------------------------------------------------------------------
1678 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1680 * Support for kernels:
1681 2.6.9-42.0.2.EL (RHEL 4)
1682 2.6.5-7.276 (SLES 9)
1683 2.4.21-40.EL (RHEL 3)
1684 2.6.12.6 vanilla (kernel.org)
1688 Frequency : always on RHEL 3
1690 Description: Number of open files grows over time
1691 Details : The number of open files grows over time, whether or not
1692 Lustre is started. This was due to a filp leak introduced
1693 by one of our kernel patches.
1695 ------------------------------------------------------------------------------
1697 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1699 * Support for kernels:
1700 2.6.9-42.EL (RHEL 4)
1701 2.6.5-7.276 (SLES 9)
1702 2.4.21-40.EL (RHEL 3)
1703 2.6.12.6 vanilla (kernel.org)
1708 Bugzilla : 5719, 9635, 9792, 9684
1709 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1710 Details : If a server is under heavy load and cannot reply to new
1711 connection requests before the client resends the (re)connect,
1712 the connection handling code can behave badly if two service
1713 threads are concurrently handing separate (re)connections from
1714 the same client. Add better locking to the connection handling
1715 code, and ensure that only a single connection will be processed
1716 for a given client UUID, even if the lock is dropped.
1718 Severity : enhancement
1720 Description: add TCP zero-copy support to kernel
1721 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1722 sends if the hardware supports scatter-gather and checksumming.
1723 This allows socklnd to do client-write and server-read more
1724 efficiently and reduce CPU utilization from skbuf copying.
1727 Frequency : only if NFS exporting from client
1729 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1730 Details : If setting attributes on a file created under NFS that had
1731 never been opened it would be possible to oops the client
1732 if the file had no objects.
1735 Frequency : always for liblustre
1737 Description: liblustre client does MDS+OSTs setattr RPC for each write
1738 Details : When doing a write from a liblustre client, the client
1739 incorrectly issued an RPC to the MDS and each OST the file was
1740 striped over in order to update the timestamps. When writing
1741 with small chunks and many clients this could overwhelm the MDS
1742 with RPCs. In all cases it would slow down the write because
1743 these RPCs are unnecessary.
1745 Severity : enhancement
1747 Description: allow number of MDS service threads to be changed at module load
1748 Details : It is now possible to change the number of MDS service threads
1749 running. Adding "options mds mds_num_threads={N}" to the MDS's
1750 /etc/modprobe.conf will set the number of threads for the next
1751 time Lustre is restarted (assuming the "mds" module is also
1752 reloaded at that time). The default number of threads will
1753 stay the same, 32 for most systems.
1758 Description: OST crash if filesystem is unformatted or corrupt
1759 Details : If an OST is started on a device that has never been formatted
1760 or if the filesystem is corrupt and cannot even mount then the
1761 error handling cleanup routines would dereference a NULL pointer.
1766 Description: NULL pointer deref in llap_from_page.
1767 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1768 incorrect) mapping if the page was truncated/reclaimed while it was
1769 searched for. Check for this condition and skip such pages when
1770 doing readahead. Introduce extra check to llap_from_page() to
1771 verify page->mapping->host is non-NULL (so page is not anonymous).
1774 Frequency : Sometimes when using sys_sendfile
1776 Description: "page not covered by a lock" warnings from ll_readpage
1777 Details : sendfile called ll_readpage without right page locks present.
1778 Now we introduced ll_file_sendfile that does necessary locking
1779 around call to generic_file_sendfile() much like we do in
1783 Frequency : with certain MDS communication failures at client mount time
1785 Description: NULL pointer deref after failed client mount
1786 Details : a client connection request may delayed by the network layer
1787 and not be sent until after the PTLRPC layer has timed out the
1788 request. If the client fails the mount immediately it will try
1789 to clean up before the network times out the request. Add a
1790 reference from the request import to the obd device and delay
1791 the cleanup until the network drops the request.
1794 Frequency : occasionally during client (re)connect
1796 Description: assertion failure during client (re)connect
1797 Details : processing a client connection request may be delayed by the
1798 client or server longer than the client connect timeout. This
1799 causes the client to resend the connection request. If the
1800 original connection request is replied in this interval, the
1801 client may trip an assertion failure in ptlrpc_connect_interpret()
1802 which thought it would be the only running connect process.
1805 Frequency : only with obd_echo servers and clients that are rebooted
1807 Description: kernel BUG accessing uninitialized data structure
1808 Details : When running an obd_echo server it did not start the ping_evictor
1809 thread, and when a client was evicted an uninitialized data
1810 structure was accessed. Start the ping_evictor in the RPC
1811 service startup instead of the OBD startup.
1813 Severity : enhancement
1814 Bugzilla : 10193 (patchless)
1815 Description: Remove dependency on various unexported kernel interfaces.
1816 Details : No longer need reparent_to_init, exit_mm, exit_files,
1817 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1820 Frequency : rare (only users of deprecated and unsupported LDAP config)
1822 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1823 Details : LDAP apparently contains 'lustreName' attributes instead of
1824 'name'. A simple remapping of the name is sufficient.
1827 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1829 Description: waiting_locks_callback trips kernel BUG if client is evicted
1830 Details : Running with the dump_on_timeout debug flag turned on makes
1831 it possible that the waiting_locks_callback() can try to dump
1832 the Lustre kernel debug logs from an interrupt handler. Defer
1833 this log dumping to the expired_lock_main() thread.
1835 Severity : enhancement
1837 Description: Support NFS exporting on 2.6 kernels.
1838 Details : Implement non-rawops metadata methods for NFS server to use without
1839 changing NFS server code.
1842 Frequency : very rare (synthetic metadata workload only)
1844 Description: two racing renames might cause an MDS thread to deadlock
1845 Details : Running the "racer" program may cause one MDS thread to rename
1846 a file from being the source of a rename to being the target of
1847 a rename at exactly the same time that another thread is doing
1848 so, and the second thread has already enqueued these locks after
1849 doing a lookup of the target and is trying to relock them in
1850 order. Ensure that we don't try to re-lock the same resource.
1853 Frequency : only very large systems with liblustre clients
1855 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1856 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1857 all OSTs in parallel. This allows the eviction of stale liblustre
1858 clients to proceed much faster than if they were done in series,
1859 and also offers similar improvements for other set_info RPCs.
1864 Description: excessive CPU usage during initial read phase on client
1865 Details : During the initial read phase on a client, it would agressively
1866 retry readahead on the file, consuming too much CPU and impacting
1867 performance (since 1.4.5.8). Improve the readahead algorithm
1868 to avoid this, and also improve some other common cases (read
1869 of small files in particular, where "small" is files smaller than
1870 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1875 Description: MDS crash when receiving packet with unknown intent.
1876 Details : Do not LBUG in unknown intent case, just return -EFAULT
1878 Severity : enhancement
1879 Bugzilla : 9293, 9385
1880 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1881 Details : Do not serialize getattr (non-intent version) and statfs.
1884 Frequency : occasional, when OST network is overloaded/intermittent
1886 Description: client evicted by OST after bulk IO timeout
1887 Details : If a client sends a bulk IO request (read or write) the OST
1888 may evict the client if it is unresposive to its data GET/PUT
1889 request. This is incorrect if the network is overloaded (takes
1890 too long to transfer the RPC data) or dropped the OST GET/PUT
1891 request. There is no need to evict the client at all, since
1892 the pinger and/or lock callbacks will handle this, and the
1893 client can restart the bulk request.
1896 Frequency : Always when mmapping file with no objects
1898 Description: client crashes when mmapping file with no objects
1899 Details : Check that we actually have objects in a file before doing any
1900 operations on objects in ll_vm_open, ll_vm_close and
1906 Description: Request leak when working with deleted CWD
1907 Details : Introduce advanced request refcount tracking for requests
1908 referenced from lustre intent.
1910 Severity : Enhancement
1912 Description: Cache open file handles on client.
1913 Details : MDS now will return special lock along with openhandle, if
1914 requested and client is allowed to hold openhandle, even if unused,
1915 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1916 closing files for every read/write openration.
1918 Severity : Enhancement
1920 Description: Cache open negative dentries on client when possible.
1921 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1922 negative dentries on lock revocation.
1927 Description: Remounting a client read-only wasn't possible with a zconf mount
1928 Details : It wasn't possible to remount a client read-only with llmount.
1930 Severity : enhancement
1931 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1932 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1933 MPICH is included to add Lustre-specific ADIO interfaces.
1934 This is based closely on the UFS ADIO layer and only differs
1935 in file creation, in order to allow the OST striping to be set.
1936 This is user-contributed code and not supported by CFS.
1941 Description: extended inode attributes (immutable, append-only) work improperly
1942 when 2.4 and 2.6 kernels are used on client/server or vice versa
1943 Details : Introduce kernel-independent values for these flags.
1945 Severity : enhancement
1948 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1949 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1950 granularity. Now, Fractional MB granularity is be supported,
1951 this is very useful for low memory system.
1953 Severity : enhancement
1955 Description: Getattr by fid
1956 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1957 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1958 have locks and avoids getattr rpc requests for it on every stat.
1961 Frequency : Always, for filesystems larger than 2TB
1963 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1964 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1965 filesystems larger than 2TB to be mounted with mballoc enabled.
1968 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1970 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1971 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1972 filesystems larger than 2TB is mounted then the kernel will
1973 silently corrupt the start of the filesystem. CONFIG_LBD is
1974 enabled for all CFS-supported kernels, but the possibility of
1975 this happening with a modified kernel config exists.
1977 Severity : enhancement
1979 Description: add client O_DIRECT support for 2.6 kernels
1980 Details : It is now possible to do O_DIRECT reads and writes to files
1981 in the Lustre client mountpoint on 2.6 kernel clients.
1983 Severity : enhancement
1985 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1986 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1987 parallel, not waiting for response from every OST before sending
1988 a rpc to the next OST.
1993 Description: setattr vs write race when updating file timestamps
1994 Details : Client processes that update a file timestamp into the past
1995 right after writing to the file (e.g. tar) it is possible that
1996 the updated file modification time can be reset to the current
1997 time due to a race between processing the setattr and write RPC.
1999 Severity : enhancement
2001 Description: Bring 'lfs find' closer in line with regular Linux find.
2002 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2003 -print0 options and obtains all the needed info through the lustre
2006 Severity : enhancement
2008 Description: support up to 1024 configured devices on one node
2009 Details : change obd_dev array from statically allocated to dynamically
2010 allocated structs as they are first used to reduce memory usage
2015 Description: Flush dirty partially truncated pages during truncate
2016 Details : Immediatelly flush partially truncated pages in filter_setattr,
2017 this way we completely avoid having any pages in page cache on OST
2018 and can retire ugly workarounds during writes to flush such pages.
2023 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2024 Details : protect inode from truncation within vfs_unlink() context
2025 just take a reference before calling vfs_unlink() and release it
2026 when parent's i_sem is free.
2031 Description: last_id value checked outside lock on OST caused LASSERT failure
2032 Details : If there were multiple MDS->OST object precreate requests in
2033 flight, it was possible that the OST's last object id was checked
2034 outside a lock and incorrectly tripped an assertion. Move checks
2035 inside locks, and discard old precreate requests.
2038 Frequency : always, if extents are used on OSTs
2040 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2041 Details : OSTs using the extents format would not zero the high 16 bits of
2042 the index physical block number. This is not a problem for any
2043 OST filesystems smaller than 16TB, and no kernels support ext3
2044 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2045 new/modified files) and can be fixed for existing filesystems
2046 with e2fsprogs-1.39-cfs1.
2051 Description: import connection selection may be incorrect if timer wraps
2052 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2053 import connections to be ignored if the 32-bit jiffies counter
2054 wraps. Use a 64-bit jiffies counter.
2057 Frequency : very large clusters immediately after boot
2059 Description: LNET request buffers exhausted under heavy short-term load
2060 Details : If a large number of client requests are generated on a service
2061 that has previously never seen so many requests it is possible
2062 that the request buffer growth cannot keep up with the spike in
2063 demand. Instead of dropping incoming requests, they are held in
2064 the LND until the RPC service can accept more requests.
2067 Frequency : Sometimes during replay
2069 Description: Assertion failure in ll_local_open after replay.
2070 Details : If replay happened on an open request reply before we were able
2071 to set replay handler, reply will become not swabbed tripping the
2072 assertion in ll_local_open. Now we set the handler right after
2073 recognising of open request
2076 Frequency : very rare
2078 Description: kernel reports "badness in vsnprintf"
2079 Details : Reading from the "recovery_status" /proc file in small chunks
2080 may cause a negative length in lprocfs_obd_rd_recovery_status()
2081 call to vsnprintf() (which is otherwise harmless). Exit early
2082 if there is no more space in the output buffer.
2084 Severity : enhancement
2086 Description: clear OBD RPC statistics by writing to them
2087 Details : It is now possible to clear the OBD RPC statistics by writing
2088 to the "stats" file.
2093 Description: Client mtime is not the same on different clients after utimes
2094 Details : In some cases, the client was using the utimes() syscall on
2095 a file cached on another node. The clients now validate the
2096 ctime from the MDS + OSTs to determine which one is right.
2101 Description: Inability to activate failout mode
2102 Details : lconf script incorrectly assumed that in pythong string's numeric
2103 value is used in comparisons.
2106 Frequency : always with multiple stripes per file
2108 Description: Inefficient object allocation for mutli-stripe files
2109 Details : When selecting which OSTs to stripe files over, for files with
2110 a stripe count that divides evenly into the number of OSTs,
2111 the MDS is always picking the same starting OST for each file.
2112 Return the OST selection heuristic to the original design.
2117 Description: mount failures may take full timeout to return an error
2118 Details : Under some heavy load conditions it is possible that a
2119 failed mount can wait for the full obd_timeout interval,
2120 possibly several minutes, before reporting an error.
2121 Instead return an error as soon as the status is known.
2123 Frequency : quota enabled and large files being deleted
2125 Description: releasing more than 4GB of quota at once hangs OST
2126 Details : If a user deletes more than 4GB of files on a single OST it
2127 will cause the OST to spin in an infinite loop. Release
2128 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2133 Description: statfs data retrieved from /proc may be stale or zero
2134 Details : When reading per-device statfs data from /proc, in the
2135 {kbytes,files}_{total,free,avail} files, it may appear
2136 as zero or be out of date.
2139 Frequency : systems with MD RAID1 external journal devices
2141 Description: lconf's call to blkid is confused by RAID1 journal devices
2142 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2143 of returning all block devices that match the journal UUID.
2146 Frequency : always, for aggregate stripe size over 4GB
2148 Description: assertion fails when trying to use 4GB stripe size
2149 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2150 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2153 Frequency : always on ppc64
2155 Description: the first write on an ext3 filesystem with mballoc got stuck
2156 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2157 perform endianness conversion.
2159 ------------------------------------------------------------------------------
2161 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2163 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2164 INTEROPERATE with older versions automatically. Please read the
2165 user documentation before upgrading any part of a live system.
2166 * WARNING: Lustre networking configuration changes are required with
2167 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2170 * Support for kernels:
2171 2.6.9-22.0.2.EL (RHEL 4)
2172 2.6.5-7.244 (SLES 9)
2173 2.6.12.6 vanilla (kernel.org)
2176 Severity : enhancement
2177 Bugzilla : 7981/8208
2178 Description: Introduced Lustre Networking (LNET)
2179 Details : LNET is new networking infrastructure for Lustre, it includes
2180 a reorganized network configuration mode (see the user
2181 documentation for full details) as well as support for routing
2182 between different network fabrics. Lustre Networking Devices
2183 (LNDs) for the supported network fabrics have also been
2184 created for this new infrastructure.
2186 Severity : enhancement
2187 Description: Introduced Access control lists
2188 Details : clients can set ACLs on files and directories in order to have
2189 more fine-grained permissions than the standard Unix UGO+RWX.
2190 The MDS must be started with the "-o acl" mount option.
2192 Severity : enhancement
2193 Description: Introduced filesystem quotas
2194 Details : Administrators may now establish per-user quotas on the
2197 Severity : enhancement
2199 Description: Configuration change for the XT3
2200 The PTLLND is now used to run Lustre over Portals on the XT3
2201 The configure option(s) --with-cray-portals are no longer used.
2202 Rather --with-portals=<path-to-portals-includes> is used to
2203 enable building on the XT3. In addition to enable XT3 specific
2204 features the option --enable-cray-xt3 must be used.
2209 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2210 Details : A race between allocating a new llcd and re-getting the llcd_lock
2211 allowed another thread to grab newly-allocated llcd.
2213 Severity : enhancement
2215 Description: 2.6 OST async journal commit and locking fix to improve performance
2216 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2217 2.6 kernels are now async as they already were in 2.4 kernels so
2218 that they can commit concurrently with the network bulk transfer.
2219 For block-allocated files the filter allocation semaphore is held
2220 to avoid filesystem fragmentation during allocation. BKL lock
2221 removed for 2.6 xattr operations where it is no longer needed.
2226 Description: lconf incorrectly determined whether two IP networks could talk
2227 Details : In some more complicated routing and multiple-network
2228 configurations, lconf will avoid trying to make a network
2229 connection to a disjoint part of the IP space. It was doing the
2230 math incorrectly for one set of cases.
2235 Description: Fix for potential infinite loop processing records in an llog.
2236 Details : If an llog record is corrupted/zeroed, it is possible to loop
2237 forever in llog_process(). Validate the llog record length
2238 and skip the remainder of the block on error.
2241 Frequency : occasional (liblustre only)
2243 Description: liblustre could not open files whose last component is a symlink
2244 Details : sysio_path_walk() would incorrectly pass the open intent to
2245 intermediate path components.
2248 Frequency : rare (liblustre only with non-standard tuning)
2249 Bugzilla : 7201 (7350)
2250 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2251 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2252 after it has been released, only do this for referenced locks
2254 Severity : enhancement
2256 Description: specifying an (invalid) directory default stripe_size of -1
2257 would reset the directory default striping
2258 Details : stripe_size -1 was used internally to signal directory stripe
2259 removal, now use "all default" to signal dir stripe removal
2260 as a directory striping of "all default" is not useful
2263 Frequency : common for large clusters running liblustre clients
2265 Description: doing an ls when liblustre clients are running is slow
2266 Details : sending a glimpse AST to a liblustre client waits for every AST
2267 to time out, as liblustre clients will not respond. Since they
2268 cannot cache data we refresh the OST lock LVB from disk instead.
2270 Severity : enhancement
2272 Description: doing an ls at the same time as file IO can be slow
2273 Details : enqueue and other "small" requests can be blocked behind many
2274 large IO requests. Create a new OST IO portal for non-IO
2275 requests so they can be processed faster.
2278 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2280 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2281 created file with mode 000
2282 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2283 allow an ftruncate() as a non-root user to a file with mode 000.
2284 Lustre now respects this flag to disable mode checks when
2285 truncating a file owned by the user
2288 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2291 Description: Revoking locks from clients that went dead or catatonic might take
2293 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2294 cancellation of such locks instant on servers without waiting for
2295 any reply from clients. Clients drops these locks when cancel
2296 notification from server is received without replying.
2299 Frequency : liblustre-only, when liblustre client dies or becomes busy
2301 Description: Doing ls on Linux clients can take a long time with active
2303 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2304 granting such locks to it in the first place if possible. Locks
2305 are taken by proxy on the OST during the read or write and
2306 dropped immediately afterward. Add connect flags handling, do
2307 not grant locks to liblustre clients for glimpse ASTs.
2309 Severity : enhancement
2311 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2312 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2313 sequential reads and behaves badly with applications doing random
2314 reads. Improve it by reading ahead at least read region, and
2315 avoiding excessive large RPC for small reads.
2317 Severity : enhancement
2319 Description: Creating more than 1000 files for a single job may cause a load
2320 imbalance on the OSTs if there are also a large number of OSTs.
2321 Details : qos_prep_create() uses an OST index reseed value that is an
2322 even multiple of the number of available OSTs so that if the
2323 reseed happens in the middle of the object allocation it will
2324 still utilize the OSTs as uniformly as possible.
2329 Description: OST or MDS may oops in ping_evictor_main()
2330 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2331 but doesn't restart at beginning of obd_exports_timed list
2334 Severity : enhancement
2336 Description: improve by-nid export eviction on the MDS and OST
2337 Details : allow multiple exports with the same NID to be evicted at one
2338 time without re-searching the exports list.
2341 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2343 Description: MDS may oops in groups_free()
2344 Details : in rare race conditions a newly allocated group_info struct is
2345 freed again, and this can be NULL. The 2.4 compatibility code
2346 for groups_free() checked for a NULL pointer, but 2.6 did not.
2349 Frequency : common for liblustre clients doing little filesystem IO
2350 Bugzilla : 9352, 7313
2351 Description: server may evict liblustre clients accessing contended locks
2352 Details : if a client is granted a lock or receives a completion AST
2353 with a blocking AST already set it would not reply to the AST
2354 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2357 Frequency : lfs setstripe, only systems with more than 160 OSTs
2359 Description: unable to set striping with a starting offset beyond OST 160
2360 Details : llapi_create_file() incorrectly limited the starting stripe
2361 index to the maximum single-file stripe count.
2364 Frequency : LDAP users only
2366 Description: lconf did not handle in-kernel recovery with LDAP properly
2367 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2369 Severity : enhancement
2371 Description: bind OST threads to NUMA nodes to improve performance
2372 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2373 node and do their allocations there to localize memory access
2375 Severity : enhancement
2377 Description: llmount can determine client NID directly from Myrinet (GM)
2378 Details : the client NID code from gmnalnid was moved directly into
2379 llmount, removing the need to use this or specifying the
2380 client NID explicitly when mounting GM clients with zeroconf
2383 Frequency : if client is started with down MDS
2385 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2386 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2387 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2388 ptlrpc_import_delayed_req() because we didn't check if the
2389 request was interrupted, and we also didn't break out of the
2390 event loop if there was no timeout
2395 Description: data loss during non-page-aligned writes to a single file from
2396 both multiple nodes and multiple threads on one node at same time
2397 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2398 inconsistency led to false short-reads, that were cached and later
2399 used by ->prepare_write() to fill in partially written page,
2400 leading to data loss.
2403 Frequency : always, if lconf --abort_recovery used
2405 Description: lconf --abort_recovery fails with 'Operation not supported'
2406 Details : lconf was attempting to abort recovery on the MDT device and not
2409 Severity : enhancement
2411 Description: remove cleanup logs
2412 Details : replace lconf-generated cleanup logs with lustre internal
2413 cleanup routines. Eliminates the need for client-cleanup and
2416 Severity : enhancement
2418 Description: add support for EAs (user and system) on lustre filesystems
2419 Details : it is now possible to store extended attributes in the Lustre
2420 client filesystem, and with the user_xattr mount option it
2421 is possible to allow users to store EAs on their files also
2423 Severity : enhancement
2425 Description: Add possibility (config option) to show minimal available OST free
2427 Details : When compiled with --enable-mindf configure option, statfs(2)
2428 (and so, df) will return least minimal free space available from
2429 all OSTs as amount of free space on FS, instead of summary of
2430 free spaces of all OSTs.
2432 Severity : enhancement
2434 Description: do not expand extent locks acquired on OST-side
2435 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2436 by server: they are not cached anyway.
2439 Frequency : when mmap is used/binaries executed from Lustre
2441 Description: Unmmap pages before throwing them away from read cache.
2442 Details : llap_shrink cache now attempts to unmap pages before discarding
2443 them (if unmapping failed - do not discard). SLES9 kernel has
2444 extra checks that trigger if this unmapping is not done first.
2449 Description: lconf didn't resolve symlinks before checking to see whether a
2450 given mountpoint was already in use
2453 Frequency : when migrating failover services
2454 Bugzilla : 6395, 9514
2455 Description: When migrating a subset of services from a node (e.g. failback
2456 from a failover service node) the remaining services would
2457 time out and evict clients.
2458 Details : lconf --force (implied by --failover) sets the global obd_timeout
2459 to 5 seconds in order to quickly disconnect, but this caused
2460 other RPCs to time out too quickly. Do not change the global
2461 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2463 Severity : enhancement
2464 Frequency : if MDS is started with down OST
2465 Bugzilla : 9439,5706
2466 Description: Allow startup/shutdown of an MDS without depending on the
2467 availability of the OSTs.
2468 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2469 Add appropriate locking and lov-osc refcounts for safe
2470 cleaning. Add osc abort_inflight calls in case the
2471 synchronize never started.
2474 Frequency : occasional (Cray XT3 only)
2476 Description: root not authorized to access files in CRAY_PORTALS environment
2477 Details : The client process capabilities were not honoured on the MDS in
2478 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2479 been accessed by an authorized user then root was able to access
2480 the file on the local client also. The root user capabilities
2481 are now allowed on the MDS, as this environment has secure UID.
2484 Frequency : occasional
2486 Description: ldiskfs "too long searching" message happens too often
2487 Details : A debugging message (otherwise harmless) prints too often on
2488 the OST console. This has been reduced to only happen when
2489 there are fragmentation problems on the filesystem.
2494 Description: Division by zero in statfs when all OSCs are inactive
2495 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2496 lov_statfs divides by value returned by lov_get_stripecnt().
2500 Bugzilla : 9489, 3273
2501 Description: First write from each client to each OST was only 4kB in size,
2502 to initialize client writeback cache, which caused sub-optimal
2503 RPCs and poor layout on disk for the first writen file.
2504 Details : Clients now request an initial cache grant at (re)connect time
2505 and so that they can start streaming writes to the cache right
2506 away and always do full-sized RPCs if there is enough data.
2507 If the OST is rebooted the client also re-establishes its grant
2508 so that client cached writes will be honoured under the grant.
2513 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2514 Details : Now I/O RPCs go to different portal number and (presumably) fast
2515 lock requests (and glimses) and other RPCs get their own service
2516 threads pool that should be able to service those RPCs
2519 Severity : enhancement
2521 Description: Ability to exchange lustre version between client and servers and
2522 issue warnings at client side if client is too old. Also for
2523 liblustre clients there is ability to refuse connection of too old
2525 Details : New 'version' field is added to connect data structure that is
2526 filled with version info. That info is later checked by server and
2530 Frequency : rare, liblustre only.
2531 Bugzilla : 9296, 9581
2532 Description: Two simultaneous writes from liblustre at offset within same page
2533 might proceed at the same time overwriting eachother with stale
2535 Details : I/O lock withing llu_file_prwv was released too early, before data
2536 actually was hitting the wire. Extended lock-holding time until
2537 server acknowledges receiving data.
2540 Frequency : extremely rare. Never observed in practice.
2542 Description: avoid generating lustre_handle cookie of 0.
2543 Details : class_handle_hash() generates handle cookies by incrementing
2544 global counter, and can hit 0 occasionaly (this is unlikely, but
2545 not impossible, because initial value of cookie counter is
2546 selected randonly). Value of 0 is used as a sentinel meaning
2547 "unassigned handle" --- avoid it. Also coalesce two critical
2548 sections in this function into one.
2550 Severity : enhancement
2552 Description: allow liblustre clients to delegate truncate locking to OST
2553 Details : To avoid overhead of locking, liblustre client instructs OST to
2554 take extent lock in ost_punch() on client's behalf. New connection
2555 flag is added to handle backward compatibility.
2557 Severity : enhancement
2558 Bugzilla : 4928, 7341, 9758
2559 Description: allow number of OST service threads to be specified
2560 Details : a module parameter allows the number of OST service threads
2561 to be specified via "options ost ost_num_threads={N}" in the
2562 OSS's /etc/modules.conf or /etc/modprobe.conf.
2566 Bugzilla : 6146, 9635, 9895
2567 Description: servers crash with bad pointer in target_handle_connect()
2568 Details : In rare cases when a client is reconnecting it was possible that
2569 the connection request was the last reference for that export.
2570 We would temporarily drop the export reference and get a new
2571 one, but this may have been the last reference and the export
2572 was just destroyed. Get new reference before dropping old one.
2574 Severity : enhancement
2575 Frequency : if client is started with failover MDS
2577 Description: Allow multiple MDS hostnames in the mount command
2578 Details : Try to read the configuration from all specified MDS
2579 hostnames during a client mount in case the "primary"
2582 Severity : enhancement
2584 Description: Stop sending data to evicted clients as soon as possible.
2585 Details : Check if the client we are about to send or are sending data to
2586 was evicted already. (Check is done every second of waiting,
2587 for which l_wait_event interface was extended to allow checking
2588 of exit condition at specified intervals).
2591 Frequency : rare, normally only when NFS exporting is done from client
2593 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2595 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2598 Frequency : always, if srand() is called before liblustre initialization
2600 Description: Liblustre uses system PRNG disturbing its usage by user application
2601 Details : Introduce internal to lustre fast and high-quality PRNG for
2602 lustre usage and make liblustre and some other places in generic
2603 lustre code to use it.
2605 Severity : enhancement
2606 Bugzilla : 9477, 9557, 9870
2607 Description: Verify that the MDS configuration logs are updated when xml is
2608 Details : Check if the .xml configuration logs are newer than the config
2609 logs stored on the MDS and report an error if this is the case.
2610 Request --write-conf, or allow starting with --old_conf.
2612 Severity : enhancement
2614 Description: Handle symlinks in the path when checking if Lustre is mounted.
2615 Details : Resolve intermediate symlinks when checking if a client has
2616 mounted a filesystem to avoid duplicate client mounts.
2621 Description: lconf can hit an error exception but still return success.
2622 Details : The lconf command catches the Command error exception at the top
2623 level script context and will exit with the associated exit
2624 status, but doesn't ensure that this exit status is non-zero.
2629 Description: failure of ptlrpc thread startup can cause oops
2630 Details : Starting a ptlrpc service thread can fail if there are a large
2631 number of threads or the server memory is very fragmented.
2632 Handle this without oopsing.
2635 Frequency : always, only if liblustre and non-default acceptor port was used
2637 Description: liblustre cannot connect to servers with non-default acceptor port
2638 Details : tcpnal_set_default_params() was not called and was therefore
2639 ignoring the environment varaible TCPNAL_PORT, as well as other
2640 TCPNAL_ environment variables
2645 Description: two objects could be created on the same OST for a single file
2646 Details : If an OST is down, in some cases it was possible to create two
2647 objects on a single OST for a single file. No problems other
2648 than potential performance impact and spurious error messages.
2652 Bugzilla : 5681, 9562
2653 Description: Client may oops in ll_unhash_aliases
2654 Details : Client dcache may become inconsistent in race condition.
2655 In some cases "getcwd" can fail if the current directory is
2661 Description: Inode refcounting problems in NFS export code
2662 Details : link_raw functions used to call d_instantiate without obtaining
2663 extra inode reference first.
2667 Bugzilla : 9942, 9903
2668 Description: Referencing freed requests leading to crash, memleaks with NFS.
2669 Details : We used to require that call to ll_revalidate_it was always
2670 followed by ll_lookup_it. Also with revalidate_special() it is
2671 possible to call ll_revalidate_it() twice for the same dentry
2672 even if first occurence returned success. This fix changes semantic
2673 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2674 reference on a request referred from the intent.
2675 ll_intent_release() then releases such a request.
2678 Frequency : rare, normally benchmark loads only
2680 Description: unlinked inodes were kept in memory on the client
2681 Details : If a client is repeatedly creating and unlinking files it
2682 can accumulate a lot of stale inodes in the inode slab cache.
2683 If there is no other client load running this can cause the
2684 client node to run out of memory. Instead flush old inodes
2685 from client cache that have the same inode number as a new inode.
2688 Frequency : SLES9 2.6.5 kernel and long filenames only
2689 Bugzilla : 9969, 10379
2690 Description: utime reports stale NFS file handle
2691 Details : SLES9 uses out-of-dentry names in some cases, which confused
2692 the lustre dentry revalidation. Change it to always use the
2696 Frequency : rare, unless heavy write-truncate concurrency is continuous
2697 Bugzilla : 4180, 6984, 7171, 9963, 9331
2698 Description: OST becomes very slow and/or deadlocked during object unlink
2699 Details : filter_destroy() was holding onto the parent directory lock
2700 while truncating+unlinking objects. For very large objects this
2701 may block other threads for a long time and slow overall OST
2702 responsiveness. It may also be possible to get a lock ordering
2703 deadlock in this case, or run out of journal credits because of
2704 the combined truncate+unlink. Solution is to do object truncate
2705 first in one transaction without parent lock, and then do the
2706 final unlink in a new transaction with the parent lock. This
2707 reduces the lock hold time dramatically.
2710 Frequency : rare, 2.4 kernels only
2712 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2713 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2714 before unmounting their filesystem in order to avoid deadlock.
2715 The kernel_locked() function in 2.4 kernels only checks whether
2716 the kernel lock is held, not whether it is this process that is
2717 holding it as 2.6 kernels do.
2722 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2723 Details : The client ptlrpc code may be trying to reconnect to a down
2724 server before a previous connection attempt has timed out.
2725 Increase the reconnect interval to be longer than the connection
2726 timeout interval to avoid sending duplicate connections to
2730 Frequency : echo_client brw_test command
2732 Description: fix echo_client to work with OST preallocated code
2733 Details : OST preallocation code (5137) didn't take echo_client IO path
2734 into account: echo_client calls filter methods outside of any
2735 OST thread and, hence, there is no per-thread preallocated
2736 pages and buffers to use. Solution: hijack pga pages for IO. As
2737 a byproduct, this avoids unnecessary data copying.
2741 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2742 Description: Client can oops in mdc_commit_close() after open replay
2743 Details : It was possible for the MDS to return an open request with no
2744 transaction number in mds_finish_transno() if the client was
2745 evicted, but without actually returning an error. Clients
2746 would later try to replay that open and may trip an assertion
2747 Simplify the client close codepath, and always return an error
2748 from the MDS in case the open is not successful.
2751 Frequency : rare, 2.6 OSTs only
2753 Description: OST may deadlock under high load on fragmented files
2754 Details : If there was a heavy load and highly-fragmented OST filesystems
2755 it was possible to have all the OST threads deadlock waiting on
2756 allocation of biovecs, because the biovecs were not released
2757 until the entire RPC IO was completed. Instead, release biovecs
2758 as soon as they are complete to ensure forward IO progress.
2760 Severity : enhancement
2762 Description: Support for specifying external journal device at mount
2763 Details : If an OST or MDS device is formatted with an external journal
2764 device, this device major/minor is stored in the ext3 superblock
2765 and may not be valid for failover. Allow detecting and
2766 specifying the external journal at mount time.
2771 Description: Mounting an MDS with pending unlinked files may cause oops
2772 Details : target_finish_recovery() calls mds_postrecov() which returned
2773 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2774 considers this an error and immediately begins cleaning up the
2775 lov, just after starting the mds_lov process
2777 Severity : enhancement
2779 Description: Implement 'lfs df' to report actual free space on per-OST basis
2780 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2781 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2782 usage of objects; '-h' to report in human readable format.
2784 ------------------------------------------------------------------------------
2786 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2793 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2794 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2795 at startup during orphan unlinking. Instead, skip buddy update
2796 before setup, as it will be regenerated after recovery is complete.
2801 Description: If an OST is inactive, its locks might reference stale inodes.
2802 Details : lov_change_cbdata() must iterate over all namespaces, even if
2803 they are inactive to clear inode references from the lock.
2805 Severity : enhancement
2806 Frequency : occasional, if non-standard max_dirty_mb used
2808 Description: Client will block write RPCs if not enough grant
2809 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2810 then the client will block writes while waiting for another RPC
2811 to complete instead of consuming its dirty limit. With change
2812 we get improved performance when max_dirty_mb is small.
2814 Severity : enhancement
2815 Bugzilla : 3389, 6253
2816 Description: Add support for supplementary groups on the MDS.
2817 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2818 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2819 lookups for user supplementary groups into a cache.
2823 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2824 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2825 for liblustre O_EXCL is enforced by MDS.
2829 Description: readdir never returns NULL in liblustre.
2830 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2831 dentry in struct dirent.
2835 Frequency : liblustre only, depends on application IO pattern
2836 Description: liblustre clients evicted if not contacting servers
2837 Details : Don't put liblustre clients into the ping_evictor list, so
2838 they will not be evicted by the pinger ever.
2840 Severity : enhancement
2842 Description: Add ability to evict clients by NID from MDS.
2843 Details : By echoing "nid:$NID" string into
2844 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2845 $NID would be instantly evicted from this MDS and from all active
2846 OSTs connected to it.
2850 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2851 Details : lookup_it_finish() used to query file size from OSTs that was not
2856 Description: service threads change working directory to that of init
2857 Details : Starting lustre service threads may pin the working directory
2858 of the parent thread, making that filesystem busy. Threads
2859 now change to the working directory of init to avoid this.
2863 Frequency : during shutdown only
2864 Description: shutdown with a failed MDS or OST can cause unmount to hang
2865 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2870 Frequency : 2.6 only
2871 Description: chmod/chown may include an extra supplementary group
2872 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2873 supplementary group and if none is specified this is used.
2876 Bugzilla : 5479 (6816)
2878 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2879 Details : If lookup is in progress on a file that is unlinked we might try
2880 to revalidate the inode and fail in revalidate after lookup is
2881 complete and ll_file_open() enqueues the open again but
2882 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2885 Frequency : always, if lconf --abort_recovery used
2887 Description: lconf --abort_recovery fails with 'Operation not supported'
2888 Details : lconf was attempting to abort recovery on the MDT device and not
2891 ------------------------------------------------------------------------------
2893 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2898 Frequency : rare (only unsupported configurations with a node running as an
2900 Bugzilla : 6514, 5137
2901 Description: Mounting a Lustre file system on a node running as an OST could
2903 Details : OSTs now preallocates memory needed to write out data at
2904 startup, instead of when needed, to avoid having to
2905 allocate memory in possibly low memory situations.
2906 Specifically, if the file system is mounted on on OST,
2907 memory pressure could force it to try to write out data,
2908 which it needed to allocate memory to do. Due to the low
2909 memory, it would be unable to do so and the node would
2910 become unresponsive.
2912 Severity : enhancement
2914 Description: Addition of lconf --service command line option
2915 Details : lconf now accepts a '--service <arg>' option, which is
2916 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2918 Severity : enhancement
2920 Description: Failover mode is now the default for OSTs.
2921 Details : By default, OSTs will now run in failover mode. To return to
2922 the old behaviour, add '--failout' to the lmc line for OSTs.
2924 Severity : enhancement
2926 Description: Health checks are now provided for MDS and OSTs
2927 Details : Additional detailed health check information on MSD and OSTs
2928 is now provided through the procfs health_check value.
2931 Frequency : occasional, depends on IO load
2933 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2934 after numerous create/delete cycles
2935 Details : The ext3 inode allocation policy would not allocate new inodes
2936 very well on the OSTs because there are no new directories
2937 being created. Instead we look for groups with free space if
2938 the parent directories are nearly full.
2943 Description: Network or server problems during mount may cause partially
2944 mounted clients instead of returning an error.
2945 Details : The config llog parsing code may overwrite the error return
2946 code during mount error handling, returning success instead
2952 Description: MDS can fail to allocate large reply buffers
2953 Details : After long uptimes the MDS can fail to allocate large reply
2954 buffers (e.g. zconf client mount config records) due to memory
2955 fragmentation or consumption by the buffer cache. Preallocate
2956 some large reply buffers so that these replies can be sent even
2957 under memory pressure.
2961 Frequency : rare (liblustre)
2962 Description: fsx running with liblustre complained that using truncate() to
2963 extend the file doesn't work. This patch corrects that issue.
2964 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2965 ATTR_SIZE and lsm use in llu_setattr_raw.
2969 Frequency : rare, only 2.6 kernels
2970 Description: Unusual file access patterns on the MDS may result in inode
2971 data being lost in very rare circumstances.
2972 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2973 code in the 2.6 kernel caused the inode and/or EA data not to
2974 be read from disk, causing single-file corruption.
2978 Frequency : rare, only 2.6 filesystems using extents
2979 Description: Heavy concurrent write and delete load may cause data corruption.
2980 Details : It was possible under high-load situations to have an extent
2981 metadata block in the block device cache from a just-unlinked
2982 file overwrite a newly-allocated data block. We now unmap any
2983 metadata buffers that alias just-allocated data blocks.
2987 Frequency : filesystems with default stripe_count larger than 77
2988 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2989 Details : lconf specifies an inode size of 4096 bytes when the default
2990 stripe_count is larger than 77. This conflicts with the default
2991 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2992 this case to avoid pinning too much memory for large EAs.
2994 ------------------------------------------------------------------------------
2996 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3001 Frequency : rare (extremely heavy IO load with hundreds of clients)
3003 Description: Client is evicted, gets IO error writing to file
3004 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3005 caused clients to be evicted for AST timeouts. The fixes in
3006 bug 5192 mean we no longer need to have such short AST timeouts
3007 so ldlm_timeout has been increased.
3010 Frequency : occasional during --force or --failover shutdown under load
3011 Bugzilla : 5949, 4834
3012 Description: Server oops/LBUG if stopped with --force or --failover under load
3013 Details : a collection of import/export refcount and cleanup ordering
3014 issues fixed for safer force cleanup
3017 Frequency : only filesystems larger than 120 OSTs
3018 Bugzilla : 5990, 6223
3019 Description: lfs getstripe would oops on a very large filesystem
3020 Details : lov_getconfig used kfree on vmalloc'd memory
3023 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3024 Bugzilla : 6242, 6243
3025 Description: reading from files that had been truncated to a non-zero size
3026 but never opened returned no data
3027 Details : ll_file_read() reads zeros from no-object files to EOF
3032 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3033 Details : There's little harm in aborting MDS/OSS recovery and letting it
3034 try again, so I removed the LASSERT and return an error instead.
3036 Severity : enhancement
3038 Description: New debugging infrastructure for tracking down data corruption
3039 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3040 (b) cover more of the client-side code path, and (c) try to narrow
3041 down where problems occurred
3045 Bugzilla : 3819, 4364, 4397, 6313
3046 Description: Racing close and eviction MDS could cause assertion in mds_close
3047 Details : It was possible to get multiple mfd references during close and
3048 client eviction, leading to one thread referencing a freed mfd.
3050 Severity: : enhancement
3051 Bugzilla : 3262, 6359
3052 Description: Attempts to reconnect to servers are now more aggressive.
3053 Details : This builds on the enhanced upcall-less recovery that was added
3054 in 1.4.2. When trying to reconnect to servers, clients will
3055 now try each server in the failover group every 10 seconds. By
3056 default, clients would previously try one server every 25 seconds.
3061 Description: After recovery, certain operations trigger a failed
3062 assertion on a client.
3063 Details : Failing over an mds, using lconf -d --failover, while a
3064 client was doing a readdir() call would cause the client to
3065 LBUG after recovery completed and the readdir() was resent.
3067 Severity : enhancement
3069 Description: Default groups are now added by lconf
3070 Details : You can now run lconf --group <servicename> without having to
3071 manually add groups with lmc.
3074 Frequency : occasional
3076 Description: Nodes with an elan id of 0 trigger a failed assertion
3079 Frequency : always when accessing e.g. tty/console device nodes
3081 Description: tty and some other devices nodes cannot be used on lustre
3082 Details : file's private_data field is used by device data and lustre
3083 values in there got lost. New field was added to struct file to
3084 store fs-specific private data.
3087 Frequency : when exporting Lustre via NFS
3089 Description: NFSD failed occasionally when looking up a path component
3090 Details : NFSD is looking up ".." which was broken in ext3 directories
3091 that had grown large enough to become hashed.
3094 Frequency : Clusters with multiple interfaces not on the same subnet
3096 Description: Nodes will repeatedly try to reconnect to an interface which it
3097 cannot reach and report an error to the log.
3098 Details : Extra peer list entries will be created by lconf with some peers
3099 unreachable. lconf now validates the peer before adding it.
3102 Frequency : Only if a default stripe is set on the filesystem root.
3104 Description: Setting a default stripe on the filesystem root prevented the
3105 filesystem from being remounted.
3106 Details : The client was sending extra request flags in the root getattr
3107 request and did not allocate a reply buffer for the dir EA.
3110 Frequency : occasional, higher if lots of files are accessed by one client
3111 Bugzilla : 6159, 6097
3112 Description: Client trips assertion regarding lsm mismatch/magic
3113 Details : While revalidating inodes the VFS looks up inodes with ifind()
3114 and in rare cases can find an inode that is being freed.
3115 The ll_test_inode() code will free the lsm during ifind()
3116 when it finds an existing inode and then the VFS later attaches
3117 this free lsm to a new inode.
3121 Bugzilla : 6422, 7030
3122 Description: MDS deadlock between mkdir and client eviction
3123 Details : Creating a new file via mkdir or mknod (starting a transaction
3124 and getting the ns lock) can deadlock with client eviction
3125 (gets ns lock and trying to finish a synchronous transaction).
3128 Frequency : occasional
3129 Description: While starting a server, the fsfilt_ext3 module could not be
3131 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3132 kernels. Previously, lconf would still use the ext3 name
3133 when trying to load modules. Now, it will correctly use
3134 ext3 on 2.4 and ldiskfs on 2.6.
3136 Severity : enhancement
3137 Description: The default stripe count has been changed to 1
3138 Details : The interpretation of the default stripe count (0, to lfs
3139 or lmc) has been changed to mean striping across a single
3140 OST, rather than all available. For general usage we have
3141 found a stripe count of 1 or 2 works best.
3143 Severity : enhancement
3144 Description: Add support for compiling against Cray portals.
3145 Details : Conditional compiling for some areas that are different
3149 Frequency : occasional
3150 Bugzilla : 6409, 6834
3151 Description: Creating files with an explicit stripe count may lead to
3152 a failed assertion on the MDS
3153 Details : If some OSTs are full or unavailable, creating files may
3154 trigger a failed assertion on the MDS. Now, Lustre will
3155 try to use other servers or return an error to the
3159 Frequency : occasional
3161 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3162 caused lock timeout during readahead (since 1.4.2).
3163 Details : Processes doing readahead might match a lock that hasn't been
3164 granted yet if there are overlapping and conflicting lock
3165 requests. The readahead process waits on ungranted lock
3166 (original lock is CBPENDING), while OST waits for that process
3167 to cancel CBPENDING read lock and eventually evicts client.
3169 Severity : enhancement
3171 Description: Initial enabling of flock support for clients
3172 Details : Implements fcntl advisory locking and file status functions.
3173 This feature is provided as an optional mount flag (default
3174 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3175 locking are implemented yet, and those that are are not guaranteed
3176 to be completely correct in production environments.
3177 mount -t lustre -o [flock|noflock] ...
3180 Frequency : occasional
3182 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3183 assertion in the ext3 JBD (journaling) layer.
3184 Details : The b_committed_data struct is protected by the big kernel lock
3185 in 2.4 kernels, serializing journal_commit_transaction() and
3186 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3187 access to this struct. In 2.6 kernels there is finer grained
3188 locking to improve SMP performance of the JBD layer.
3192 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3197 Description: merge and cleanup kernel patches.
3200 -----------------------------------------------------------------------------
3202 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3204 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3205 versions of Lustre. Please update all servers and clients to
3206 version 1.4.2 or later at the same time. You must also run
3207 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3208 to update the configuration logs.
3210 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3211 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3212 - swab llog records of type '0' so we get proper header size/idx (5861)
3213 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3214 - fix rename of one directory over another leaking an inode (5953)
3215 - avoid SetPageDirty on 2.6 (5981)
3216 - don't re-add just-being-destroyed locks to the waiting list (5653)
3217 - when creating new directories, inherit the parent's custom
3218 striping settings if present parent (3048)
3219 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3220 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3221 - don't hold client locks on temporary worklist from l_lru (5666)
3222 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3223 - automatically evict dead clients (5921)
3224 - Update file size properly in create+truncate+fstat case (6196)
3225 - Do not unhash mountpoint dentries, do not allow removal of
3227 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3228 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3229 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3230 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3231 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3232 - Correctly handle reads of files with no objects (6243)
3233 - lctl recover will also mark a device active if deactivate used (5933)
3235 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3236 - allow --write-conf on an MDS with different nettype than client (5619)
3237 - don't write config llogs to MDS for mounts not from that MDS (5617)
3238 - lconf should create multiple TCP connections from a client (5201)
3239 - init scripts are now turned off by default; run chkconfig --on
3240 lustre and chkconfig --on lustrefs to use them
3241 - upcalls are no longer needed for clients to recover to failover
3243 - add --abort-recovery option to lconf to abort recovery on device
3245 - add support for an arbitrary number of OSTs (3026)
3246 - Quota support protocol changes.
3247 - forward compatibility changes to wire structs (6007)
3248 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3249 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3250 - improved "lustre status" script
3251 - initialize blocksize for non-regular files (6062)
3252 - added --disable-server and --disable-client configure options (5782)
3253 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3254 - Vanilla 2.4.29 support
3255 - increase maximum number of obd devices to 520 (6242)
3256 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3257 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3258 SLES 9 2.6.5 kernels
3259 - limit stripes per file to 160 (the maximum EA size) (6093)
3261 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3264 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3265 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3266 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3267 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3268 - small ext3 extents cleanups and fixes (5733)
3269 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3270 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3271 - avoid needless client->OST connect, fix handle mismatch (5317)
3272 - fix DLM error path that led to out-of-sync client, long delays (5779)
3273 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3274 - fix several locking issues related to i_size (5492,5624,5654,5672)
3275 - don't move pending lock onto export if it is already evicted (5683)
3276 - fix kernel oops when creating .foo in unlinked directory (5548)
3277 - fix deadlock in obdfilter statistics vs. object create (5811)
3278 - use time_{before,after} to avoid timer jiffies wrap (5882)
3279 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3280 - Do not leak request if server was not able to process it (5154)
3281 - If mds_open unable to find parent dir, make that negative lookup(5154)
3282 - don't create new directories with extent-mapping (5909, 5936)
3284 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3285 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3286 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3287 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3288 - support for RHEL4 kernel on i686 (5773)
3289 - provide error messages when incompatible logs are encountered (5898)
3291 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3292 * version 1.4.0.10 (1.4.1 release candidate 1)
3294 - don't keep a lock reference when lock is not granted (4238)
3295 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3296 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3297 - drop import inflight refcount on signal_completed_replay error (5255)
3298 - unlock page after async write error during send (3677)
3299 - handle missing objects in filter_preprw_read properly (5265)
3300 - no transno return for symlink open, don't save no-trasno open (3440)
3301 - don't try to complete elan receive that already failed (4012)
3302 - free RPC server reply state on error (5406)
3303 - clean up thread from ptlrpc_start_thread() on error (5160)
3304 - readahead could read extra page into cache that wasn't ejected (5388)
3305 - prevent races in class_attach/setup/cleanup/detach (5260)
3306 - don't dereference de->d_inode after l_dput of de (5458)
3307 - use "int" for stripe value returned from lock_to_stripe (5544)
3308 - mballoc allocation and error-checking fixes in 2.6 (5504)
3309 - block device patches to fix I/O request sizes in 2.6 (5482)
3310 - look up hostnames for IB nals (5602)
3311 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3312 - don't start multiple acceptors for the same port (5277)
3313 - fix incorrect LASSERT in mds_getattr_name (5635)
3314 - export a proc file for general "ping" checking (5628)
3315 - fix "lfs check" to not block when the MDS is down (5628)
3317 - service request history (4965)
3318 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3319 - create an "evict_client" /proc entry on OSTs, like the MDS has
3320 - fix mount usage message, return errors per mount(8) (5168)
3321 - change grep [] to grep "[]" in tests so they work in more UMLs
3322 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3323 - remove ancient LOV_MAGIC_V0 EA support (5047)
3324 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3325 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3326 - allow readahead allocations to fail when low on memory (5383)
3327 - mmap locking landed again, after considerable improvement (2828)
3328 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3330 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3333 - send OST transaction number in read/write reply to free req (4966)
3334 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3335 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3336 - the watchdog thread now runs as interruptible (5246)
3337 - flock/lockf fixes (but it's still disabled, pending 5135)
3338 - don't use EXT3 constants in llite code (5094)
3339 - memory shortage at startup could cause assertion (5176)
3341 - reorganization of lov code
3342 - single portals codebase
3344 - add extents/mballoc support (5025)
3345 - direct I/O reads in the obdfilter (4048)
3346 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3348 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3351 - send OST transaction number in read/write reply to free req (4966)
3352 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3353 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3354 - the watchdog thread now runs as interruptible (5246)
3355 - handle missing objects in filter_preprw_read properly (5265)
3356 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3357 - drop import inflight refcount on signal_completed_replay error (5255)
3358 - unlock page after async write error during send (3677)
3359 - return original error code on reconstructed replies (3761)
3360 - no transno return for symlink open, don't save no-trasno open (3440)
3362 - add pid to ldlm debugging output (4922)
3363 - bump the watchdog timeouts -- we can't handle 30sec yet
3364 - extra debugging for orphan dentry/inode bug (5259)
3366 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3369 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3370 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3371 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3372 - fix NULL dereference in /proc/sys/portals/routes (4827)
3373 - allow failed mdc_close() operations to be interrupted (4561)
3374 - stop precreate on OST before MDS would time out on it (4778)
3375 - don't send partial-page writes before EOF from client (4410)
3376 - discard client grant for sub-page writes on large-page clients (4520)
3377 - don't free dentries not owned by NFS code, check generation (4806)
3378 - fix lsm leak if mds_create_objects() fails (4801)
3379 - limit debug_daemon file size, always print CERROR messages (4789)
3380 - use transno after validating reply (3892)
3381 - process timed out requests if import state changes (3754)
3382 - update mtime on OST during writes, return in glimpse (4829)
3383 - add mkfsoptions to LDAP (4679)
3384 - use ->max_readahead method instead of zapping global ra (5039)
3385 - don't interrupt __l_wait_event() during strace
3387 - add software watchdogs to catch hung threads quickly (4941)
3388 - make lustrefs init script start after nfs is mounted
3389 - fix CWARN/ERROR duplication (4930)
3390 - return async write errors to application if possible (2248)
3391 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3392 - print NAL number in %x format (4645)
3393 - update barely-supported suse-2.4.21-171 series (4842)
3394 - support for sles 9 %post scripts
3395 - support for building 2.6 kernel-source packages
3396 - support for sles km_* packages
3398 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3401 - ignore -ENOENT errors in osc_destroy (3639)
3402 - notify osc create thread that OSC is being cleaned up (4600)
3403 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3404 - reconstruct ost_handle() like mds_handle() (4657)
3405 - create a new thread to do import eviction to avoid deadlock (3969)
3406 - let lconf resolve symlinked-to devices (4629)
3407 - don't unlink "objects" from directory with default EA (4554)
3408 - hold socknal file ref over connect in case target is down (4394)
3409 - allow more than 32000 subdirectories in a single directory (3244)
3410 - fix blocks count for O_DIRECT writes (3751)
3411 - OST returns ENOSPC from object create when no space left (4539)
3412 - don't send truncate RPC if file size isn't changing (4410)
3413 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3414 - bind to privileged port in socknal and tcpnal (3689)
3416 - rate limit CERROR/CWARN console message to avoid overload (4519)
3417 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3418 - basic mmap support (3918)
3419 - kernel patch series update from b1_4 (4711)
3421 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3424 - avoid crash during MDS cleanup with OST shut down (2775)
3425 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3426 - don't use bad inodes on the MDS (3744)
3427 - dynamic object preallocation to improve recovery speed (4236)
3428 - don't hold spinlock over lock dumping or change debug flags (4401)
3429 - don't zero obd_dev when it is force cleaned (3651)
3430 - print grants to console if they go negative (4431)
3431 - "lctl deactivate" will stop automatic recovery attempts (3406)
3432 - look for existing locks in ldlm_handle_enqueue() (3764)
3433 - don't resolve lock handle twice in recovery avoiding race (4401)
3434 - revalidate should check working dir is a directory (4134)
3436 - don't always mark "slow" obdfilter messages as errors (4418)
3438 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3441 - don't close LustreDB during write_conf until it is done (3860)
3442 - fix typo in lconf for_each_profile (3821)
3443 - allow dumping logs from multiple threads at one time (3820)
3444 - don't allow multiple threads in OSC recovery (3812)
3445 - fix debug_size parameters (3864)
3446 - fix mds_postrecov to initialize import for llog ctxt (3121)
3447 - replace config semaphore with spinlock (3306)
3448 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3449 - don't allow enqueue to complete on a destroyed export (3822)
3450 - down write_lock before checking llog header bitmap (3825)
3451 - recover from lock replay timeout (3764)
3452 - up llog sem before sending rpc (3652)
3453 - reduce ns lock hold times when setting kms (3267)
3454 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3455 - fix NULL deref and obd_dev leak on setup error (3312)
3456 - replace some LBUG about llog ops with error handling (3841)
3457 - don't match INVALID dentries from d_lookup and spin (3784)
3458 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3459 - fix invalid assertion in ptlrpc_set_wait (3880)
3461 - add libwrap support for the TCP acceptor (3996)
3462 - add /proc/sys/portals/routes for non-root route listing (3994)
3463 - allow setting MDS UUID in .xml (2580)
3464 - print the stack of a process that LBUGs (4228)
3466 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3469 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3470 - make sure to unset replay flag from failed open requests (3440)
3471 - if default stripe count is 0, use OST count for inode size (3636)
3472 - update parent mtime/ctime on client for create/unlink (2611)
3473 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3474 - free recovery state on server during a forced cleanup (3571)
3475 - unregister_reply for resent reqs (3063)
3476 - loop back devices mounting and status check on 2.6 (3563)
3477 - fix resource-creation race that can provoke i_size == 0 (3513)
3478 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3479 - more debugging for page-accounting assertion (3746)
3480 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3481 - avoid deadlock after precreation failure (3758)
3482 - fix race and lock order deadlock in orphan handling (3450, 3750)
3483 - add validity checks when grabbing inodes from l_ast_data (3599)
3485 - add /proc/.../recovery_status to obdfilter (3428)
3486 - lightweight CDEBUG infrastructure, debug daemon (3668)
3487 - change default OSC RPC parameters to be better on small clusters
3488 - turn off OST read cache for files smaller than 32MB
3489 - install man pages and include them in rpms (3100)
3490 - add new init script for (un)mounting lustre filesystems (2593)
3491 - run chkconfig in %post for init scripts (3701)
3492 - drop scimac NAL (unmaintained)
3494 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3497 - clean kiobufs before and after use (3485)
3498 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3499 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3500 - change init script to fail more clearly if not run as root (1528)
3501 - allow clients to reconnect during replay (1742)
3502 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3503 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3504 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3505 - fix cond_resched (3554)
3506 - only evict unfinished clients after recovery (3515)
3507 - allow bulk resend, prevent data loss (3570)
3508 - dynamic ptlrpc request buffer allocation (2102)
3509 - don't allow unlinking open directory if it isn't empty (2904)
3510 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3511 - remove extraneous obd dereference causing LASSERT failure (3334)
3512 - don't use get_cycles() when creating temp. files on the mds (3156)
3513 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3514 - handle EEXIST for set-stripe, set proper directory name (3336)
3516 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3517 - fix ksocknal_fmb_callback() error messages (2918)
3519 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3522 - don't copy lvb into (possibly NULL) reply on error (2983)
3523 - don't deref dentry after dput, don't free lvb on error (2922)
3524 - use the kms to determine writeback rpc length (2947)
3525 - increment oti_logcookies when osc is inactive (2948)
3526 - update client's i_blocks count via lvb messages (2543)
3527 - handle intent open/close of special files properly (1557)
3528 - mount MDS with errors=remount-ro, like obdfilter (2009)
3529 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3530 - don't use cancelling-locks' kms values (2947)
3531 - use highest lock extent for kms, not last one (2925)
3532 - don't dereference ERR_PTR() dentry in error handling path (3107)
3533 - fix thread race in portals_debug_dumplog() (3122)
3534 - create lprocfs device entries at setup instead of at attach (1519)
3535 - common AST error handler, don't evict client on completion race (3145)
3536 - zero nameidata in detach_mnt in 2.6 (3118)
3537 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3538 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3539 - initialize RPC timeout timer earlier for 2.6 (3219)
3540 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3541 - print nal/nid for unknown nid (3258)
3542 - additional checks for oscc recovery before doing precreate (3284)
3543 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3544 - don't crash in mdc_close for bad permissions on open (3285)
3545 - zero i_rdev for non-device files (3147)
3546 - clear page->private before handing to FS, better assertion (3119)
3547 - tune the read pipeline (3236)
3548 - fix incorrect decref of invalidated dentry (2350)
3549 - provide read-ahead stats and refine rpc in flight stats (3328)
3550 - don't hold journal transaction open across create RPC (3313)
3551 - update atime on MDS at close time (3265)
3552 - close LDAP connection when recovering to avoid server load (3315)
3553 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3554 - don't leak open file on MDS after open resend (3325)
3555 - serialize filter_precreate and filter_destroy_precreated (3329)
3556 - loop device shouldn't call sync_dev() for nul device (3092)
3557 - clear page cache after eviction (2766)
3558 - resynchronize MDS->OST in background (2824)
3559 - refuse to mount the same filesystem twice on same mountpoint (3394)
3560 - allow llmount to create routes for mounting behind routers (3320)
3561 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3562 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3563 - fix rare problem with rename on htree directories (3417)
3565 - allow default OST striping configuration per directory (1414)
3566 - fix compilation for qswnal for 2.6 kernels (3125)
3567 - increase maximum number of MDS request buffers for large systems
3568 - change liblustreapi to be useful for external progs like lfsck (3098)
3569 - increase local configuration timeout for slow disks (3353)
3570 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3572 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3575 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3576 - don't overwrite extent policy data in reply if lock was blocked (2901)
3577 - drop filter export grants atomically with removal from device (2663)
3578 - del obd_self_export from work_list in class_disconnect_exports (2908)
3579 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3580 - swab reply message in mdc_close, other PPC fixes (2464)
3581 - fix destroying of named logs (2325)
3582 - overwrite old logs when running lconf --write_conf (2264)
3583 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3584 - fix race in target_handle_connect (2898)
3585 - mds_reint_create() should take same inode create lock (2926)
3586 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3587 - don't close files for self_export to avoid uninitialized obd (2936)
3588 - allow MDS with the same name as client node (2939)
3589 - hold dentry reference for closed log files for unlink (2325)
3590 - reserve space for all logs during transactions (2059)
3591 - don't evict page beyond end of stripe extent (2925)
3592 - don't oops on a deleted current working directory (2399)
3593 - handle hard links to targets without a parent properly (2517)
3594 - don't dereference NULL lock when racing during eviction (2867)
3595 - don't grow lock extents when lots of conflicting locks (2919)
3597 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3600 - account for cache space usage on clients to avoid data loss (974)
3601 - lfsck support in lustre kernel code (2349)
3602 - reduce journal credits needed for BRW writes (2370)
3603 - orphan handling to avoid losing space on client/server crashes
3604 - ptlrpcd can be blocked, stopping ALL progress (2477)
3605 - use lock value blocks to assist in proper KMS, faster stat (1021)
3606 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3607 - recovery for initial connections (2355)
3608 - fixes for mds_cleanup_orphans (1934)
3609 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3610 - block all file creations until orphan recovery completes (1901)
3611 - client remove rq_connection from request struct (2423)
3612 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3613 - recovery timer race (2670)
3614 - mdc_close recovey bug (2532)
3615 - ptlrpc cleanup bug (2710)
3616 - mds timeout on local locks (2588)
3617 - namespace lock held during RPCs (2431)
3618 - handle interrupted sync write properly (2503)
3619 - don't try to handle a message that hasn't been replied to (2699)
3620 - client assert failure during cleanup after abort recovery (2701)
3621 - leak mdc device after failed mount (2712)
3622 - ptlrpc_check_set allows timedout requests to complete (2714)
3623 - wait for inflight reqs when ptlrpcd finishes (2710)
3624 - make sure unregistered services are removed from the srv_list
3625 - reset bulk XID's when resending them (caught by 1138 test)
3626 - unregister_bulk after timeout
3627 - fix lconf error (2694)
3628 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3629 - readahead locks pages, leaves pending causing memory pressure (2673)
3630 - increase OST request buffers to 4096 on large machines (2729)
3631 - fix up permission of existing directories in simple_mkdir (2661)
3632 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3633 - don't assign transno to errored transactions (2742)
3634 - don't delete objects on OST if given a bogus objid from MDS (2751)
3635 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3636 - if rq_no_resend, then timeout request after recovery (2432)
3637 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3638 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3639 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3640 - MDS recovery shouldn't delete objects that it has given out (2730)
3641 - if enqueue arrives after completion, don't clobber LVB (2819)
3642 - don't unlock pages twice when trigger_group_io returns error (2814)
3643 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3644 - don't write pages to disk if there was an error (1450)
3645 - don't ping imports that have recovery disabled (2676)
3646 - take buffered bytes into account when balancing socknal conn (2817)
3647 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3648 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3649 - remove little-endian swabbing of llog records (1987)
3650 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3651 - retry reposting request buffers if they fail (1191)
3652 - grow extent at grant time to avoid granting a revoked lock (2809)
3653 - lock revoke doesn't evict page if covered by a second lock (2765)
3654 - disable VM readahead to avoid reading outside lock extents (2805)
3656 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3657 - updated kernel patches for hp-2.4.20 kernel (2681)
3659 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3662 - fix truncated write corruption (2366)
3663 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3665 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3666 - ensure len(uuid) < 37 in lmc (1171)
3667 - fix ia64 OOPS in llog_test (2255)
3668 - zero end of page at obdfilter for partial page writes (2648)
3669 - don't leave stale dentries around after renames (bug 2428)
3670 - fix timeouts when evicting a client with a single lock held (2642)
3671 - set deadline for the initial HELLO message to drain (2634)
3672 - print out dotted-quad IP addresses in the socknal (2302)
3674 - additional debugging for MDS client eviction problem (2443)
3675 - fix mkfsoptions support for osts (2603, 2604)
3677 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3680 - add series for the vanilla 2.6.0 kernel
3681 - add series for the vanilla 2.4.24 kernel
3682 - add series for a cray x86/64 UL kernel drop
3683 - fix xattr patches for the vanilla 2.4.19 series
3685 - generate true UUIDs in lmc (1171)
3686 - have portals stack dumping break in UML (2466)
3687 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3688 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3689 - wake callback waiting threads on client eviction (2460)
3690 - Add --ptldebug and --subsystem to lmc (1719)
3691 - update assertion to allow safe interrupt allocation
3692 - set rq_no_resend for cancel requests (2432)
3693 - recalculate ptlrpcd timeout after resend (2494)
3694 - call vfs_rmdir when removing pending directories (2368)
3695 - fix renaming a file to itself (2429)
3696 - lmc creates a default one-stripe lov (2454)
3697 - expand procfs space to handle large clusters (2326)
3698 - increase UML stack to avoid overflow
3699 - update lconf's list of debug and subsystem masks
3700 - fix lfs find --obd (2510)
3701 - /proc tunable for disabling filter read caching (2591)
3702 - stop rpm packages from altering slapd.conf (2301)
3703 - disable nagle in the socknal under 0conf (2578)
3704 - choose mds inode size based on stripe count (2572)
3705 - fix kernel-source rpm problems (2516)
3707 - add --disable-doc to avoid pdf generation (2421)
3708 - update documentation, tests, type-os, comments
3709 - avoid format warnings on ia64
3710 - remove the TOE NAL
3711 - tiny code cleanups by removing unused fields
3713 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3716 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3717 - remove the most blatant lies from BUILDING file (2371)
3718 - change default debug level to reasonable production setting
3719 - reduce client side cache size to reduce cache flush time
3720 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3721 - make TCP zerocopy and pinger support enabled by default (2476)
3722 - sync writes completed after process exits caused crashes (2319)
3723 - maintain correct mount count on the MDS (2356)
3724 - backout 1557, because 2316 wasn't really fixed
3725 - better file I/O statistics gathering in /proc
3726 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3727 - another kernel patch to fix zero-copy TCP function export
3728 - don't take duplicate lock when processing re-sent getattr (2420)
3729 - lctl uses obd_self_export instead of creating new conn (2353)
3730 - MDS/OST recovery case which requires object creation asserted (2425)
3731 - move lfs from /usr/sbin to /usr/bin in packages
3732 - fix race between mds_client_add and mds_client_free (2417)
3733 - use kmalloc instead of slabs in portals (2430)
3734 - don't create duplicate records when a failover MDS is present (2442)
3735 - remove unnecessary mount age check (2332)
3736 - don't remove directory inodes from locks prematurely (2451)
3737 - don't break if MDS service name is the same as hostname (2103)
3738 - fix races in client write RPC generation when cache full (2482)
3740 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3743 - remove now-unused request->rq_obd (278)
3744 - if an allocation fails, print out how much memory we've used (1933)
3745 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3746 - add the "configurable stack size" patch to most series files (1256)
3747 - ability to write large log records, for 100+ OST configs (2306)
3748 - fix NULL deref when filter_prep fails (2314)
3749 - fix operator precedence error in filter_sync
3750 - dynamic allocation of socknal TX descriptors (2315)
3751 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3752 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3753 - ensure that CWARN messages in Portals always get to the syslog
3754 - __init/__exit are not for prototype decls (ldlm_init/exit)
3755 - x86-64 compile warning fixes
3756 - fix gateway LMC keyword conflict (2318)
3757 - fix MDS lock inversions in getattr/reint paths (1844)
3758 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3759 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3760 - DLM race condition prevented some lock evictions (2328)
3761 - ENOMEM detection and retry on socknal sends (2230)
3762 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3763 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3764 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3765 - better error messages when a client is rejected during recovery (1505)
3766 - avoid cancelling locks which were never granted, after failure (2330)
3767 - fix i_sem/journal inversion in mds_client_add (2333)
3768 - fix truncate/getattr lock cycle deadlock (2334)
3769 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3770 - fix two rare exit paths which could leak an l_lock() ref (2321)
3771 - fencepost error in MDS/OST orphan recovery (2226)
3772 - make log record alignment 8 bytes (1988)
3773 - lstripe now fails when requested offset > ost_count (2237)
3774 - ensure that all kernel series have a complete list.h (1607)
3775 - fix crashes in special-file operations (2316)
3776 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3778 - allow configurable automake binary, for testing new versions
3779 - small update to the lfs documentation
3781 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3783 * fix negative export reference count in fsfilt_sync (2312)
3785 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3786 * release candidate 0.9.1
3788 - orphans are moved into the PENDING directory for possible recovery
3789 - replayed opens now open by fid for orphan/rename safety (1042)
3790 - last close of an orphan inode generates a transno (683)
3791 - chdir() and mount() now pin the directory entry (1020)
3792 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3793 - discard very old requests without processing them (1502)
3794 - remove some common, well-understood CERRORs (1505)
3795 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3796 - clear "grant" flags in OST replies until OST grant code lands (1644)
3797 - fix read performance by not clobbering i_blksize on client (1598)
3798 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3799 - make LRU size a /proc tunable, clears locks when reduced (707)
3800 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3801 - server locks take references on exports now (1558)
3802 - build fixes for 2.4.20-rh trees (1663)
3803 - return an error from lov_create if all OSCs are inactive (1751)
3804 - fix import levels when a reconnect happens without a timeout (1597)
3805 - exit early from mds_open if we get a lookup error (1749)
3806 - partial page read at EOF wouldn't wait for disk before sending (1642)
3807 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3808 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3809 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3810 - write/truncate lock inversion (1639)
3811 - Don't auto-load obdclass, portals modules during cleanup (1495)
3812 - fix timestamps from jumping to "now" (1763)
3813 - extra journal assertions (1648)
3814 - add an extra multiunlink test (1771)
3815 - fix read_record/write_record API (1776)
3816 - fix leak of offset_extent, possible incorrect i_size later (1772)
3817 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3818 - Debugging for the kqswnal_get_idle_tx problems (1820)
3819 - Allow recovery to be attempted multiple times (1536)
3820 - Write out MDS last_rcvd file after it is first created (1600)
3821 - Fix tx_descriptor leak in failed transmit situations (1827)
3822 - ext3 journaling fixes for assertion failure after IO error (1871)
3823 - class_export_put() on freed export after completion AST error (1896)
3824 - Fix revalidate looping in VFS (1322)
3825 - Don't access a freed export during MDS_REINT timeout (1521)
3826 - Add open-unlink recovery support on the MDS (1673,1764)
3827 - Return an error if no MDS data was read from last_rcvd (1946)
3828 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3829 - Don't setup a disk device that doesn't match exported UUID (317)
3830 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3831 - avoid committing NULL handle in force close
3832 - local.sh is now a one-stripe LOV configuration
3833 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3834 - fix ext3 htree duplicate directory entry corruption (1516)
3835 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3836 - update to new LOV EA format (2097)
3837 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3839 2003-06-15 Phil Schwan <phil@clusterfs.com>
3842 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3843 - per-import/export recovery handling (958, 931, 959)
3844 - multiple last-rcvd slots, for serving multiple FSes (949)
3845 - connections are again shared between multiple imp/exports (963, 964)
3846 - "umount -f" would hang if any requests needed to be sent (393, 978)
3847 - avoid pinning large req buffer by copying for queued messages (989)
3848 - add "uuid" to "lctl device" command to help upcalls (991)
3849 - "open" RPCs with transnos would confuse recovery counters (1037)
3850 - do proper endian conversion of all wire messages (288, 340, 891)
3851 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3852 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3853 - fix ptlrpc_connection leak in target_handle_connect (1174)
3854 - fix import refcounting bug in OST and MDS cleanup (1134)
3855 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3856 - fix very unlikely obd_types race condition (501)
3857 - remove osc_open hack for echo_client (1187)
3858 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3859 - a failure in read_inode2 leads to deadlock (1139)
3860 - cancel ack-locks as soon as transaction is committed (1072)
3861 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3862 - make sure to commitrw after any preprw to avoid deadlock (1162)
3863 - failing to execute a file in a lustre FS would lock inode (1203)
3864 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3865 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3866 - obd_uuid2tgt would walk off the end of the list (1255)
3867 - on IA64 the osc would give portals incorrect bulk size (1258)
3868 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3869 - fix lock inversion caused by new llite matching code (1282)
3870 - limit the number of dirty pages on a client to 10MB (1286)
3871 - timed out locks were not being corrected cancelled (1289)
3872 - fix O_DIRECT above 4GB on IA-32 (1292)
3873 * major user-visible changes
3874 - fail out/fail over policy now controlled by the upcall (993)
3876 - add OBD_PING to check server availability and failure (954)
3877 - lustre messages are now sent in sending host order (288, 340, 891)
3878 - add eadatalen to MDS getattr reply (340)
3879 - OST read replies may contain second buffer, with per-page status (593)
3881 2003-03-11 Phil Schwan <phil@clusterfs.com>
3884 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3885 - failed open()s could cause deadlock; fixed (867, 869)
3886 - stop cancelling OST locks when files are closed (481)
3887 - overlapping XID spaces caused network corruption (851, 853)
3888 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3889 - setattr_raw added, to do single-RPC, server-side setattrs
3890 - lmc/lconf syntax change for OST UUIDs
3891 - fix crashy race condition between ptlrpc_free_req and osc_close
3892 - don't use request in mdc_enqueue if we hit a timeout (889)
3893 - don't set the inode i_size for regular files from the MDS (896)
3894 - handle out of order completion AST (842)
3895 - don't LBUG if a lock request times out after receiving AST (913)
3896 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3897 - if a bad lock AST arrives, send an error instead of dropping entirely
3898 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3899 - fix leak in bulk IO when only partially completed (899, 900, 926)
3900 - fix O_DIRECT for ia64 (55)
3901 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3902 - C-z after timeout could hang a process forever; fixed (977)
3904 - client-side I/O cache (678, 924, 929, 941, 970)
3906 - READPAGE and SETATTRs which don't take server-side locks get
3909 2003-02-11 Phil Schwan <phil@clusterfs.com>
3912 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3913 - Fix fsfilt_extN_readpage() to read a full page of directory
3914 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3915 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3916 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3917 - Most intent operations were removed, in favour of a new RPC mode
3918 that does a single RPC to the server and bypasses most of the VFS
3919 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3920 - Aggressively cancel local locks on DLM servers
3921 - mds_reint_unlink sends EA to the client if it's the last nlink.
3922 client uses that EA to unlink OST objects.
3923 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3924 - recursive symlinks were fixed (439)
3925 - fixed NULL deref in DEBUG_REQ
3926 - filter_update_lastobjid no longer calls sync, which annoyed extN
3927 - fixed multi-client small-writes to a single file problem (445)
3928 - fixed mtime updates during file writes (607)
3929 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3930 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3931 - fixed Linux OST size reporting problem (444, 656)
3932 - OST now updates object mtime with writes or setattr (607, 619)
3933 - client verifies file size before zeroing page past EOF (445)
3934 - OST now writes last allocated objid to disk with allocation (108)
3935 - LOV on echo now works (409)
3937 - mds_reint_unlink sends a new buffer, with the EA included. this
3938 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3939 set if a regular file was being unlinked, and it was the last link
3940 - use PtlGet from the target for bulk writes (315)
3941 - OST now updates object mtime with writes or setattr (607, 619)
3942 - LDLM now has a grant-time callback to revalidate locked items, if
3944 - Many MDS operations were reorganized to combat race conditions
3946 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3947 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3949 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3952 - Fully reactivate OST imports after reconnection (512, others)
3953 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3954 - More graceful error handling for truncating on dead OST (515)
3955 - Don't error out unless we're actually accessing dead stripes (474)
3956 - Fix garbage sizes when stripes are missing (410)
3957 - LRU counters were broken, causing constant lock purge (433, 432)
3958 - garbage on read from stripes with failed OSTs (441)
3959 - mark OSCs as active before reconnecting during recovery (438)
3960 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3961 - lfind did not preserve OST order in output (443)
3962 - symlinks cause hung clients, incorrect data (439)
3963 - stop dereferencing request after dropping refcount (457)
3964 - don't LASSERT(spin_is_locked) on non-SMP (455)
3965 - fixes for many rename() bugs
3966 - fstat didn't correctly synchronize attributes (399)
3967 - server must handle lock cancellation during blocking AST prep (487)
3968 - bulk descriptors were free()d too soon (511)
3969 - fix paths in lconf, which would load incorrect modules (451, 507)
3970 - fix confusing lconf 'host not found' error message (386)
3971 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3972 - fix race condition in mdc_blocking_ast() for inode access (526)
3973 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3974 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3975 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3976 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3977 - fix lstripe in several ways (kernel side) (527)
3978 - fix request leak in ldlm_cli_enqueue (262)
3979 - incorrect OSC was marked inactive after OST failure
3980 - call mds_fs_cleanup before unmounting filesystem (524)
3981 - fix races between taking ns_lock and ldlm_lock_change_resource
3982 - fix races updating LOV export open file list
3983 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3984 - fix recovery NULL deref in ldlm_cli_cancel_unused
3985 - fix some DLM races by using new hash table for lock handles (419)
3986 - permit the client to specify desired inodes, at replay
3987 - duplicate requests when we queue them for replay reintegration
3988 - fix last_rcvd offset calculation
3989 - sync after each recovered transaction, so we always make progress
3990 - never, not always, ERESTART requests without transnos
3991 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3993 - skip replay if the MDS says that the client is already connected
3994 - don't check for a recovery-enabled export to match lctl's UUID
3995 - don't INC_USE_COUNT for phantom exports
3996 - don't crash when cleaning up phantom exports (567)
3997 - don't double-finish or set replay data for errored mdc_open requests
3998 - abort requests when they time out, so we don't get old replies
3999 - send/receive replies for AST messages again
4000 - if the client says that it doesn't have the lock, cancel it on the
4002 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4003 mark it as destroyed, it will all work out when decref is called
4004 - fix module use counts (22, 581)
4006 - ASTs now expect a reply (server cancels lock on error reply)
4008 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4011 - fix many simultaneous client startup (392)
4012 - fix dentry->d_it clobbering
4013 - credentials weren't being shipped for readdir/getattr operations
4014 - remove invalid assertions triggered during some concurrent MD
4016 - proper Lustre versions added (336, 389)
4017 - fix memory leak for create error case (398)
4018 - fix LOV locking bug that would get cli/srv out of sync
4019 - fix echo client over LOV (409)
4020 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4021 - fix double-O_EXCL intent crash (424)
4022 - avoid sending multiple lock CANCELs (352)
4024 - MDS can do multi-client recovery (modulo bugs in new code)
4026 - many updates, edits, cleanups
4028 2002-11-18 Phil Schwan <phil@clusterfs.com>
4031 - fix null d_it dereference (346)
4032 - fix full OST/dbench hang (333)
4033 - fix permission problem with file removal (286)
4034 - fix removal of OSCs from LOV when they fail
4035 - fix NULL deref during bulk timeout (214)
4036 - fix problems related to multiple filesystems on one MDS (241)
4037 - fixed serious subtle metadata locking bugs
4038 - free locks on clients when inodes are removed due to memory
4040 - fix inode pointer in lock data (285)
4041 - partial support for multiple MDS on a single host (241)
4042 - data locks weren't cancelled at clear_inode time (290, 311)
4043 - intent locks could lead to unbounded lock growth (205)
4044 - added a maximum lock count, an LRU list, and a flusher
4045 - fix multiple rename (365)
4046 - properly abstracted the echo client
4047 - OSC locked 1 byte too many; fixed
4048 - rewrote brw callback code:
4049 - fixed recovery bugs related to LOVs (306)
4050 - fixed too-many-pages-in-one-write crash (191)
4051 - fixed (again) crash in sync_io_timeout (214)
4052 - probably fixed callback-related race (385)
4054 - Add capability to MDS protocol
4055 - LDLM cancellations and callbacks on different portals
4057 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4060 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4061 - defer open object destruction to close time (601981, 138)
4062 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4063 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4064 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4065 - don't call obd_setattr in ll_file_release for destroyed objects
4066 * protocol change to lustre_msg: move |version| and add |flags|
4067 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4068 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4069 * added replay of create, unlink, link and rename operations during
4070 MDS failover; recovery should be much more robust now
4071 * remove failed OSCs from LOVs (only lov_create uses this so far)
4072 * the lustre-HOWTO was brought (more) up to date (582544)
4074 2002-10-23 Phil Schwan <phil@clusterfs.com>
4077 - in-use dentries weren't being reused properly (617851)
4078 - prevent multiple LDLM setup (599178)
4079 - fix LOV size calculations for truncate (617853)
4080 - fix client handling of MDS intent errors (POSIX)
4081 - fix permission bug in lovstripe.c test (624321)
4082 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4083 - truncate past end of file could corrupt data
4084 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4085 - a race in recovery could return ETIMEDOUT to apps (623947)
4086 - building outside the source directory was fixed
4087 * the lustre-HOWTO was brought (more) up to date (582544)
4088 * major progress was made on recovery functionality
4090 2002-10-10 Phil Schwan <phil@clusterfs.com>
4093 - recovery deadlock fix
4094 - rm -rf causes LBUG fix (617817)
4095 - file open by multiple tasks fix (618962)
4096 - directory permissions bugs (602707 and 620007)
4097 - journal_stop fixed with locking (611313)
4098 - O_APPEND failures resolved (618273, perhaps 614459)
4099 - lconf PATH fix (619770)
4100 - IA64 build fix (621450)
4101 - RPC buffer sizes scale with amount of memory
4103 2002-10-01 Phil Schwan <phil@clusterfs.com>
4106 - locks would be cancelled without throwing away data pages,
4107 resulting in inconsistent data (605627)
4108 - inode attributes were not always being refreshed (605627, 612449)
4109 - lconf now continues to cleanup after lctl reports an error
4110 - MDS now enforces user permissions (602707)
4111 - lprocfs cleanup fixed, but not yet enabled (614157)
4112 - fixed infinite server hang, should a client not respond to an AST
4113 - avoid going into recovery if user calls readlink() with a buffer
4114 that's too small (613941)
4115 - AST RPCs no longer require replies (614867) -- this may be changed
4116 - don't crash server if client sends an IOV that's too big (611336)
4117 - fixed lock conversion deadlock (611892)
4118 - fixed the following of symlinks (614622)
4119 * recovery: the server can remove locks from a client that dies, other
4120 clients can make progress
4121 * more extN patch fixes
4122 * compile-time configurable ptlrpc buffer allocations
4124 - collaborative read cache document
4125 - Lustre Lite Performance CDR document-in-progress
4127 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4130 - fix typo in patch-2.4.18
4132 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4135 - clear ptlrpc request each time in handle_incoming_request()
4136 - unlink of files now destroys the object on the OST
4138 2002-09-19 Peter Braam <braam@clusterfs.com>
4140 * add hard link support
4141 * change obdfile creation method
4142 * kernel patch changed
4144 2002-09-19 Peter Braam <braam@clusterfs.com>
4147 - stack overflow bug in extN fixed
4149 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4151 * documentation updates
4152 - add man pages for config tools
4153 - update tests/README to describe testing with new config tools
4154 - finish metadata API descriptions
4155 * bug fixes and cleanups
4156 - statfs workaround for 16TB limit
4157 - LOV stripe allocation improved, can stripe on subset of OSTs
4158 - LOV file size/IO offset was wrong for files > 4GB in size
4159 - object EA data was being dropped, caused files to be unreadable
4160 - memory overflow with non-LOV OST caused memory corruption
4161 - fixed regression tests to work with new config tools, obdfilter
4162 - fixed bug when directory size became larger than 1 block
4163 - fixed bug (for single client case) when PWD was deleted
4164 - invalidate local directory pages when doing intent-based ops
4165 - avoid LDLM oops when lock callback contained bad data
4167 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4169 * documentation updates
4170 * bug fixes and cleanups
4171 - configuration tools
4174 - 64-bit compile warnings
4175 - 64-bit internal statfs data
4177 * test_brw on persistent OST devices
4179 * lprocfs (disabled)
4181 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4183 * documentation updates
4184 * bug fixes and cleanups
4185 * configuration tools
4187 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4190 * many small fixes to 0_5_4
4191 * io/network handling
4192 * thinkos in MDS operations
4194 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4197 * crucial basic fixes to 0.5.3
4198 * IOR, Iozone work over Elan
4201 2002-08-07 Phil Schwan <phil@clusterfs.com>
4202 * version 0_5_3, our first alpha
4203 * we use the new Portals iovs
4204 * documentation updates
4205 * bug fixes and cleanups
4206 * small changes in the DLM wire protocol
4208 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4209 * version 0_5_1 with some initial stability,
4210 * locking on MD and file I/O.
4211 * documentation updates
4212 * several bug fixes since 0.5.0
4213 * small changes in wire protocol
4215 2002-07-18 Phil Schwan <phil@clusterfs.com>
4217 * delivered as Lustre Light Alpha
4218 * fixed a crash after handling invalid MDS requests
4219 * fixed directory pages for architectures with non-4k pages sizes
4221 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4222 * release version v0_4_4
4223 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4225 * Fixes a number of interruption problems with OST operations.
4226 * Update documentation for portals header changes
4227 * Move all wire protocol structs/defines to lustre_idl.h
4228 * Fixes symlink length bug.
4229 * Add tcpdump to repository.
4231 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4232 * release version v0_4_3
4233 * Fixes statfs for inodes on extN.
4234 * Fixes bug in runtests which would delete /etc/hosts.
4235 * Use 64-bit object IDs wherever possible (not into VFS though)
4236 Remove ost_get_info, which is unused by lustre, and out of date.
4238 2002-07-03 Peter Braam <braam@clusterfs.com>
4239 * release version v0_4_2 Fixes a lookup error (type not passed)
4240 * move forward to head of Portals
4241 * move forward to latest Lustre kernel
4243 2002-06-25 Peter Braam <braam@clusterfs.com>
4244 * release version v0_4_1. Hopefully stable on single node use.